Book Fall2011
Book Fall2011
Book Fall2011
2
20 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
so the relationship between capture area and antenna gain is
A =
2
G
4
(1.1)
This relationship allows us to write the equation that governs power transfer in a radio link
in terms of gains or capture area only, i.e.
P
avr
= P
t
G
t
G
r
2
16
2
R
2
(1.2)
or
P
avr
= P
t
A
t
A
r
2
R
2
(1.3)
The last version is known as the Friis transmission formula [4] who liked this form of the link
equation because it does not contain any numerical coecients, a feature that he claimed
would make it easier to remember.
1.3.1 Link example
The Friis transmission formula can be combined with knowledge of the operating system
temperature (dened and discussed in Chapter 10) and required SNR to predict the perfor-
mance of a communications link. The signal to noise ratio in an antenna-receiver system can
be written as the ratio of available signal and noise powers at the antenna terminals, after
referring all receiver noise to the antenna by assigning an operating system temperature
T
op
= T
A
+T
e
to the antenna terminals. Here, T
A
is the eective antenna temperature and
T
e
is the eective input noise temperature of the receiver:
SNR =
P
avr
kT
op
B
n
= P
t
G
t
G
r
2
16
2
R
2
1
kT
op
B
n
If the minimum SNR required for communications is denoted by SNR
min
, then the maxi-
mum distance between the antennas is:
R
max
=
P
t
G
t
G
r
2
SNR
min
(16
2
kT
op
B
n
)
For example, suppose that detection and/or demodulation of a particular signal requires
an SNR of 2 (3 dB). Well assume that the receiver lters dene a noise bandwidth B
n
equal
to 10 kHz, that the transmitter power is (P
t
) is 1 Watt, and the receiver and transmitting
antennas are both lossless dipole antennas with length of one half wavelength, with the
dipoles oriented in space so as to maximize the received signal power. The maximum gain
of the lossless (100% ecient) half-wave dipole antenna is G
t
= G
r
= 1.64, or 10 log(1.64) =
2.14 dBi.
Suppose that the link operates at a frequency of 300 MHz ( = 1 m) and that the
operating system temperature T
op
= 1000 K (typical for an operating frequency in the VHF
part of the spectrum). The maximum distance between the the two antennas in this case
is 2.4 10
7
m, or 24, 000 km. This tells us that the power required to communicate over a
distance of 24,000 km is comparable to that required to power a small ashlight!
Notice that R
max
so that increasing (decreasing) the frequency will decrease (in-
crease) R
max
. At the same time, the size of both the transmitting and receiving antennas
1.4. REVIEW OF FOURIER TRANSFORMS AND SPECTRA 21
would decrease (increase) since they are assumed to be half-wavelength dipoles in this exam-
ple. It is worthwhile to think about what would happen if, instead of keeping the antenna
length xed in terms of wavelengths, the absolute size of the transmitting and receiving an-
tennas was held constant while the frequency is varied. This is left as a thought experiment
to be carried out by the reader!
1.4 Review of Fourier Transforms and Spectra
Let f(t) be a real-valued time-signal. The Fourier transform of f(t) will be denoted by F(!)
where
F(!) =
1
1
f(t)e
j!t
dt (1.4)
The time-signal can be recovered from its Fourier transform using
f(t) =
1
2
1
1
F(!)e
+j!t
d! (1.5)
The reader is assumed to be familiar with the general properties of Fourier transforms
and transform pairs, e.g., the scaling and shifting properties. Time signals of interest are
assumed to be absolutely integrable or to be periodic functions. In either case, the Fourier
transform and its inverse will exist.
The magnitude of the Fourier transform, |F(!)| is called the amplitude spectrum. For
signals with nite energy, the square of the magnitude, |F(!)|
2
is called the energy spectrum
since, by Parsevals theorem, the total energy, E, in the signal is
E =
1
1
|f(t)|
2
dt =
1
2
1
1
|F(!)|
2
d!. (1.6)
When the time-domain signal is real, the negative frequency part of the Fourier transform
is related to the positive frequency part through
F(!) = F
(!) (1.7)
where the
denotes complex conjugate. Thus if the positive part of the frequency spectrum
is known, the negative frequency part can be obtained, since the amplitude of the spectrum
is always symmetric about the frequency origin (|F(!)| = |F(!)|) and the phase of the
spectrum is anti-symmetric (arg[F(!)] = arg[F(!)]). For real signals, it is sometimes
convenient to plot only the positive frequency part of the amplitude or energy spectrum.
1.5 Message signals
We will call the signal that is to be transmitted over the communications channel the message
signal , denoted by m(t). In an analog communications system the message signal could be
the audio signal from a microphone. For example, Figure 1.5 (top) shows 0.8 seconds of
audio from a radio talk show. Figure 1.5 (bottom) shows the average energy spectrum
calculated using an audio sample of duration 15 seconds. This audio sample contained male
and female speakers. Notice that the energy is contained within the frequency range 200-
3500 Hz. We shall use the symbol W to refer to the bandwidth of the message signal. The
22 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
10 log
10
(2
<|M(f)|
2
>
E
)
-0.5
0
0.5
1
0.4 0.6 0.8 1.0 1.2
t (seconds)
m(t)
80
70
60
50
40
30
20
10
0
0 1000 2000 3000 4000 5000
f (Hz)
-1
Figure 1.5: Top: message signal consisting of 0.8 seconds of audio from a radio talk show.
Bottom: normalized average energy spectrum (in dB) of a 15-second segment of audio
from the same talk show. The average energy spectrum was calculated by segmenting the
original time series, sampled at 44.1 ksamples/s, into 512 point segments. The discrete
Fourier transform (DFT) was calculated for each segment, and the squared magnitude of
each DFT was averaged.
1.5. MESSAGE SIGNALS 23
message signal bandwidth is dened such that the frequency interval [W, W] contains all
(or, essentially all) of the energy in the signal. For the signal shown in Figure 1.5, W is
approximately 3.5 kHz.
In a digital communication system the message signal will typically be formed from a
superposition of pulses with amplitudes chosen to represent a sequence of information bits.
In this case m(t) can be written
m(t) =
n
a
n
p(t nT) (1.8)
where p(t) is a pulse function chosen to control the shape of the amplitude spectrum of
m(t), {a
n
} represents the sequence of information bits and is a sequence of numbers chosen
from a nite set of possible values, and T
1
is the signaling rate the rate at which pulses
are transmitted. In the case of binary data, the pulse amplitudes would typically be chosen
from the antipodal set {1, 1}, so that a
n
= 1. In this case each information bit results
in one transmitted pulse so that 1 bit of information is transmitted each T seconds. More
generally, a
n
could take on any of M discrete values, in which case we have an M ary
data sequence and log
2
M bits are transmitted each T seconds. In general, the number of
information bits transmitted per second will be the signaling rate (T
1
) times the number
of bits represented by each pulse that is transmitted (log
2
M).
Pulse-shapes that are commonly used to generate the message signal include the rectan-
gular pulse (Figure 1.6a):
p(t) =
1 ,
1
2
t
T
<
1
2
0 , elsewhere
,
with Fourier transform
P(f) =
sin fT
fT
.
The half-cosine pulse (Figure 1.6b):
p(t) =
cos
t
T
,
1
2
t
T
<
1
2
0 , elsewhere
,
with Fourier transform
P(f) =
sin T(
1
2
+f)
T(
1
2
+f)
+
sin T(
1
2
f)
T(
1
2
f)
has a Fourier spectrum with a wider central lobe than that of the rectangular pulse, but
the sidelobes are signicantly smaller than those of the rectangular pulse. The rectangular
and half-cosine pulses have innite bandwidth, however the half-cosine pulse has a more
compact spectrum because the amplitude of the half-cosine pulses spectrum falls o much
more rapidly with increasing frequency.
A very useful family of pulses with nite bandwidth is obtained by dening the pulse such
that its Fourier transform transitions from a constant (at) central region to zero through a
smooth transition region having a raised cosine shape. These pulses are called raised-cosine
24 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
(a)
p(t)
1.0
5 4 3 2 1 0 1 2 3 4 5
t/T
rectangular pulse
0.0
rectangular pulse spectrum
0.5
1.0
5 4 3 2 1 0 1 2 3 4 5
fT
|P(f)|
0.0
(b)
p(t)
1.0
5 4 3 2 1 0 1 2 3 4 5
t/T
half-cosine pulse
0.0
half-cosine pulse spectrum
0.5
1.0
5 4 3 2 1 0 1 2 3 4 5
fT
|P(f)|
0.0
(c)
p(t)
1.0
5 4 3 2 1 0 1 2 3 4 5
t/T
= 0.1, raised cosine pulse
0.0
= 0.1, raised cosine spectrum
0.5
1.0
2 1 0 1 2
fT
|P(f)|
0.0
(d)
p(t)
1.0
5 4 3 2 1 0 1 2 3 4 5
t/T
= 0.5, raised cosine pulse
0.0
= 0.5, raised cosine spectrum
0.5
1.0
2 1 0 1 2
fT
|P(f)|
0.0
(e)
p(t)
1.0
5 4 3 2 1 0 1 2 3 4 5
t/T
= 0.9, raised cosine pulse
0.0
= 0.9, raised cosine spectrum
0.5
1.0
2 1 0 1 2
fT
|P(f)|
0.0
Figure 1.6: Pulse shapes suitable for use in digital communications systems. The time-
domain pulse waveform is shown on the left and the magnitude of the Fourier Transform of
the pulse is shown to the right. (a) rectangular pulse, (b) half-cosine pulse, (c)-(e) pulses
with raised-cosine spectrum with excess bandwidths of (c) 10% ( = 0.1), (d) 50% ( = 0.5)
and (e) 90% ( = 0.9).
1.5. MESSAGE SIGNALS 25
pulses (Figure 1.6(c-e)). The Fourier transform of the raised-cosine pulse is
P(f) =
T |fT|
1
2
(1 )
T
2
{1 + cos[
(|fT|
1
2
(1 )]}
1
2
(1 ) < |fT|
1
2
(1 + )
0 |fT| >
1
2
(1 + )
, (1.9)
where 0 1. The dimensionless parameter controls the width of the raised-cosine
transition region. The time-domain pulse shape of the raised-cosine pulse is
p(t) =
sin(t/T)
t/T
cos(t/T)
1 (2t/T)
2
. (1.10)
The bandwidth of these pulses is
W =
1
2T
(1 + ). (1.11)
The dimensionless parameter is called the fractional excess bandwidth and it is often
expressed as a percentage. The minimum possible bandwidth results when = 0 which
results in W =
1
2T
. In this case, the spectrum becomes rectangular and the pulse shape is
a sinc function with relatively large sidelobes that extend over many signaling intervals on
either side of the center of the pulse. For 0 < 1, the amplitude spectrum exhibits a
gradual transition to zero. As increases, the sidelobes are increasingly damped causing the
time-domain pulses sidelobes to have signicant amplitudes over fewer signaling intervals.
It is important to point out that the spectrum of the chosen pulse will determine the
shape of the spectrum of the message signal m(t) and hence the bandwidth occupied by the
message signal. Let us assume that the summation in equation 1.8 is nite and consists of
N terms. The resulting m(t) has nite energy. The Fourier transform of m(t) is
M(f) = F[m(t)] =
N
n=1
a
n
F[p(t nT)] = P(f)
N
n=1
a
n
e
j2fnT
. (1.12)
The energy spectrum is
|M(f)|
2
= |P(f)|
2
|
N
n=1
a
n
e
j2nfT
|
2
. (1.13)
For a specic data sequence {a
n
} the second term on the right-hand side will vary in an
irregular manner as a function of frequency. Nevertheless, the second term has a well-dened
average value. If a
n
= 1 the average value is equal to the length of the data sequence, N.
Therefore, if the energy spectra of many message signals resulting from dierent random
data sequences of length N are averaged, the average of the second term will be close to N.
Using brackets <> to denote an average over a large number of message signals,
< |M(f)|
2
>' N|P(f)|
2
, (1.14)
which means that the energy spectrum of the pulse determines the average energy spectrum
of the message signal. This is illustrated in Figure 1.7 which shows a portion of a message
26 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
2
<|M(f)|
2
>
E
1.0
0.0
1.0
2.0
0 40 80 120 160 200
t/T
m(t)
0.0
1.0
2.0
3.0
0.00 0.25 0.50 0.75 1.00
fT
2.0
Figure 1.7: Top: A message signal produced using a random data sequence and raised-
cosine pulses with = 0.5. Bottom: Normalized average energy spectrum of m(t) computed
numerically using the following procedure: a message signal of length 32768T was generated
using a random sequence of bits. The long message signal was divided into short segments of
length approximately 100T. The energy spectrum of each short segment was computed, and
all spectra were averaged. The average spectrum was normalized so that the area under the
spectrum is 1.0. Note that the shape of the energy spectrum exhibits the raised-cosine shape
of the pulses. Since = 0.5, the bandwidth of the spectrum is WT =
1
2
(1 + 0.5) = 0.75.
1.6. LINEAR MODULATION 27
signal generated using raised-cosine pulses with = 0.5 and the average energy spectrum
calculated by dividing the long message signal into shorter segments and averaging the
energy spectra of each segment. Note that the shape of the average energy spectrum is
determined by the raised-cosine spectrum of the individual pulses.
The raised-cosine pulses are Nyquist pulses, having the property that they are zero at
non-zero multiples of the signaling interval T. Because of this property, such pulses can be
superposed as in equation 1.8 without causing any inter-pulse (or intersymbol) interference
at times corresponding to integer multiples of the signaling interval (t = nT) because only
the pulse centered at t = nT contributes to m(t) at those times. Refer ahead to the upper
plot in Figure 1.8 to see a sample message signal generated from a binary pulse amplitude
sequence (a
n
= 1) and raised-cosine pulses with = 0.1. The binary message can be read
o of the upper plot by taking samples at integer values of t/T; in this case, the message
was the sequence
{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}.
In the following sections the basic types of analog modulation will be discussed. They
fall into two general categories: (1) Linear and (2) Nonlinear (or angle) modulation schemes.
Each category can be subdivided into special cases:
1. Linear modulation:
(a) DSB/SC - Double Sideband-Suppressed Carrier
(b) DSB/with carrier
(c) SSB - Single Sideband
(d) VSB - Vestigial Sideband
2. Angle modulation (nonlinear modulation):
(a) FM - Frequency Modulation
(b) PM - Phase Modulation
1.6 Linear Modulation
Throughout all of our discussions we will assume that the message signal, m(t), is a real
function of time and that the time-average value of m(t) is zero (m(t) has no DC component).
The linear modulation schemes can be represented by
s(t) = A(t) cos(!
c
t + ) (1.15)
where s(t) is called the modulated carrier signal, A(t) = Am(t) + B, !
c
is the carrier
frequency, and is a constant phase angle. The magnitude of the instantaneous amplitude,
|A(t)| is called the envelope of s(t).
The various types of linear modulation come about from dierent choices for B (e.g.
B = 0 or B > max |m(t)|) or, in the case of single-sideband (SSB), from superposing two
or more linearly-modulated signals, e.g., A
1
(t) cos(!
c
t +
1
) + A
2
(t) cos(!
c
t +
2
) where
A
1
(t) and A
2
(t) are are obtained by passing m(t) through dierent lters. The linear
modulation schemes are linear in the sense that if m
1
(t) !s
1
(t) and m
2
(t) !s
2
(t), then
m
1
(t) +m
2
(t) !s
1
(t) +s
2
(t)
1
.
1
Strictly speaking, this linearity property applies only when B = 0.
28 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
1.6.1 Modulation Theorem
Before considering the various linear modulation schemes it will be useful to derive a fun-
damental relationship between the Fourier transforms of s(t) and A(t). Denote the Fourier
transforms of s(t) and A(t) by S(!) and A(!), respectively; i.e.,
S(!) =
1
1
s(t)e
j!t
dt (1.16)
A(!) =
1
1
A(t)e
j!t
dt. (1.17)
Substituting Equation 1.15 into Equation 1.16 we obtain:
S(!) =
1
1
A(t) cos(!
c
t + )e
j!t
dt. (1.18)
Using the identity cos() =
1
2
[e
j
+e
j
], Equation 1.18 becomes
S(!) =
1
1
A(t)
1
2
[e
j(!ct+)
+e
j(!ct+)
]e
j!t
dt (1.19)
=
1
2
e
j
1
1
A(t)e
j(!!c)t
dt +
1
2
e
j
1
1
A(t)e
j(!+!c)t
dt
which leads to the nal result:
S(!) =
1
2
e
j
A(! !
c
) +
1
2
e
j
A(! + !
c
). (1.20)
This important relationship between the spectrum of A(t) and the spectrum of the modu-
lated carrier s(t) is known as the modulation theorem. It states that the spectrum of s(t)
can be obtained by superposing two copies of the spectrum of A(t) that have been displaced
by +!
c
and !
c
on the frequency axis.
1.6.2 Double-sideband suppressed-carrier (DSB/SC) Modulation
and Demodulation
DSB/SC represents the simplest possible mapping between the message signal and A(t).
Here we take A(t) = Am(t) where A is a constant, i.e.,
s(t) = Am(t) cos(!
c
t + ) (1.21)
This corresponds to multiplying the carrier by the message signal as illustrated in Figure
1.8, where the message signal is shown in the upper plot and the modulated carrier is shown
below. For this example, the message signal is generated from a binary pulse amplitude
sequence {a
n
} using raised-cosine pulses with = 0.1. The binary message can be read o
of the upper plot by taking samples at integer values of t/T; in this case, the message was
the sequence
{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}.
1.6. LINEAR MODULATION 29
DSB
1.0
0.0
1.0
2.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
m(t)
3.0
2.0
1.0
0.0
1.0
2.0
3.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
s(t)
2.0
Figure 1.8: Top: message signal representing 20 information bits generated using raised-
cosine pulses with = 0.1. Bottom: DSB signal, s(t), normalized such that < s
2
(t) >= 1.
The DSB/SC signal, s(t), shown in the lower panel has been normalized to have a mean-
square value of 1.0. Thus, if this signal represents the voltage developed across a 1 resistor,
the average power dissipated in the resistor would be 1 Watt.
Using the modulation theorem, it is a simple matter to relate the Fourier spectra of m(t)
and s(t). Suppose that the two-sided spectrum of m(t) is as shown in Figure 1.9 where
m(t) is assumed to be band-limited to W Hz. For simplicity, we plot only the magnitude
of the spectrum. Then, if the carrier frequency is larger than W(f
c
> W), the spectrum
f
|M(f)|
W W
|M|
max
Figure 1.9: Two-sided spectrum of m(t).
of s(t) will look like Figure 1.10. If f
c
< W, the picture would be dierent, since the two
components of the spectrum would overlap in a region centered on f = 0. To avoid such
overlap the carrier frequency is chosen so that f
c
> W. It should be apparent that the DSB
signal bandwidth is 2W (Hz) and therefore takes up twice as much of the spectrum as the
original message signal.
Schematically DSB modulation can be represented as in Figure 1.11 where the modulated
30 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
f
|S(f)|
f
c
+ W
2W
1
2
|M|
max
f
c
f
c
W
2W
f
c
f
c
W f
c
+ W
Figure 1.10: Two-sided spectrum of s(t).
signal is shown being fed to an antenna for transmission.
m(t)
s(t) = Am(t) cos !
c
t
cos !
c
t
A
Figure 1.11: DSB modulation. The triangle represents a linear amplier which scales the
signal by the constant gain, A.
At the receiver, DSB can be demodulated using a process called coherent demodulation
(or synchronous demodulation), which is illustrated in Figure 1.12. In this Figure, we have
accounted for the fact that propagation between a transmitter and receiver will cause the
carrier component of the signal to be phase-shifted with respect to the signal that was
actually transmitted. We have ignored the fact that the received signals amplitude will be
reduced and that the message signal will be delayed.
The synchronous demodulation process requires the receiver to have a carrier recovery
system which somehow recovers the phase-shifted carrier, cos(!
c
t +), from the DSB signal.
Assuming that the carrier recovery system does its job perfectly, then it is easy to understand
how the coherent demodulation process works. The signal after the multiplier, s
0
(t), is the
1.6. LINEAR MODULATION 31
s
0
(t)
s(t) = Am(t) cos(!
c
t + )
cos(!
c
t + )
s
out
(t)
LPF
Carrier
Recovery
Figure 1.12: Coherent detection/demodulation.
product of s(t) and the recovered carrier:
s
0
(t) = s(t) cos(!
c
t + ) (1.22)
= Am(t) cos
2
(!
c
t + )
= Am(t)
1
2
[1 + cos(2!
c
t + 2)]
=
1
2
Am(t) +
1
2
Am(t) cos(2!
c
t + 2)
The spectrum of s
0
(t) is easily found using the modulation theorem and is sketched in
Figure 1.13.
f
|S
0
(f)|
2W
1
2
A|M|
max
2f
c
W
1
4
A|M|
max
Lowpass lter response
Figure 1.13: Spectrum of s
0
(t). The ideal lowpass lters cuto frequency can be anywhere
in the range W < f
cutoff
< 2f
c
W.
The purpose of the lowpass lter is to reject the term
1
2
Am(t) cos(2!
c
t + 2) which
corresponds to the part of the spectrum centered on 2f
c
. The output from the lowpass lter
is
s
out
(t) =
1
2
Am(t) (1.23)
32 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
Note that for the purpose of this discussion we assumed that f
c
> W, so that the spectra
of the two terms in s
0
(t) (after the multiplier) did not overlap. If the terms overlap, then
it is not possible to separate the undesired component from the desired one with a lowpass
lter and the demodulation scheme will not work. We also assumed that the receivers
carrier-recovery circuit was able to exactly recover the frequency and phase of the received
carrier signal. Since these quantities must be estimated from a noisy received signal, we
should consider the possibility that the carriers frequency and phase might be imperfectly
recovered, as in Figure 1.14. In this case
s
0
(t)
s(t) = Am(t) cos(!
c
t + )
cos((!
c
+ !)t + + )
s
out
(t)
LPF
Carrier
Recovery
Figure 1.14: Recovered carrier slightly o-frequency and out-of-phase.
s
0
(t) = Am(t) cos(!
c
t + ) cos((!
c
+ !)t + + ) (1.24)
=
1
2
Am(t)[cos(!t + ) + cos((2!
c
+ !)t + 2 + )]
After the lowpass lter
s
out
(t) =
1
2
Am(t) cos(!t + ) (1.25)
Comparing Equation 1.25 with Equation 1.23 we see that the eect of the frequency and
phase error is to multiply the desired signal by cos(!t + ). This is unsatisfactory and
would result in a distorted output. As an example, suppose that the message signal is a
pure tone, i.e., m(t) = cos(!
m
t). Then the output from a demodulator with frequency and
phase error would be
s
out
(t) =
1
2
Acos(!
m
t) cos(!t + ) (1.26)
=
1
4
A[cos((!
m
+ !)t + ) + cos((!
m
!)t )]
i.e., the output would consist of two tones separated in frequency by twice the frequency
error. Clearly, a speech waveform that consists of the superposition of many tones would
be seriously distorted unless the frequency error is extremely small. Thus it is important
that ! = 0, i.e., it is necessary for the receivers oscillator frequency to be exactly the same
as that of the transmitters carrier oscillator. Suppose ! = 0, then:
s
out
(t) =
1
2
Am(t) cos (1.27)
1.6. LINEAR MODULATION 33
This is acceptable if is constant and 6= /2. It is not satisfactory if varies with
time or if is close to /2. The conclusion is that it is necessary for the receiver and
transmitter oscillators to be frequency-synchronized and phase-locked to within a constant
phase oset that is less than /2. Carrier synchronization can be relatively easy to achieve
if some synchronization information is sent by the transmitter.
An example of a familiar system that employs DSB/SC and provides separate carrier
synchronization information is the FM stereo multiplex message signal where the dierence
between the left and right channel audio signals (L(t) R(t)) DSB/SC modulates a 38 kHz
subcarrier. This DSB/SC is summed with the monophonic signal (the sum of the left and
right channels) along with a 19 kHz pilot tone derived by dividing the 38 kHz subcarrier
frequency by two. The pilot tone is used by the receiver to reconstruct a 38 kHz carrier
with the proper phase for demodulating the stereo information.
The analog color television signal also includes a component provided explicitly for the
purpose of helping the receiver to achieve proper phase synchronization for coherent demod-
ulation. In this case, coherent demodulation is necessary to extract the color information
from the received video signal.
It is possible to design a carrier synchronization system that recovers the carrier from
the received signal on its own without the aid of any extra synchronization information. It
turns out that the missing carrier can be regenerated if the signal is rst passed through
a nonlinear device. For example, if the signal is passed through a square-law device, then
the output will contain a component at twice the carrier frequency. This component can
be extracted and its frequency divided by two to give the necessary carrier reference. This
operation can be carried out using a phase-locked loop (PLL) in a circuit called a squaring
loop or using a so-called Costas loop. (See Chapter 13.)
1.6.3 DSB with carrier
One way to provide the carrier synchronization information that is required to demodulate a
DSB signal is to include an unmodulated carrier component in the transmitted signal. When
a carrier component is included the the modulation is called DSB w/carrier. The dierence
between DSB/SC and DSB w/carrier is simply the addition of a carrier component to the
spectrum in the latter case. The signal is represented mathematically by:
s(t) = (Am(t) +B) cos(!
c
t) (1.28)
The signal can be written as the superposition of a DSB/SC signal and a carrier component:
s(t) = Am(t) cos(!
c
t) +Bcos(!
c
t) (1.29)
The spectrum of s(t) is shown in Figure 1.15. The bandwidth occupied by DSB w/carrier
is the same as DSB/SC and is 2W, where W is the highest frequency in the message signal,
m(t). The coherent demodulator shown in Figure 1.12 can be used to demodulate DSB
w/carrier. The presence of the carrier simplies the implementation of the carrier recovery
module. When a carrier component is present, the output of a coherent demodulator will
contain a DC oset that is proportional to the strength of the received carrier component.
The magnitude of the DC term can be used as a relative indication of received signal
strength. It can also be used to control the gain of ampliers preceeding the demodulator in
order to keep the amplitude of the signal at the demodulator input relatively constant. Such
a feedback system for controlling gain is called an automatic gain control (AGC) system.
34 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
f
|S(f)|
2W
f
c
Figure 1.15: Spectrum of s(t) for DSB w/carrier.
1.6.3.1 DSB with full carrier
There is an important distinction between the cases where Am(t) +B is allowed to become
negative and where Am(t)+B is constrained to always stay positive. If Am(t)+B is always
> 0 the envelope of the modulated signal (dened by a line that connects positive peaks of
s(t)) reproduces the message signal, m(t). When Am(t) +B > 0 for all t the signal is called
DSB with full carrier. In this case it is possible to use a particularly simple demodulation
circuit to recover m(t) at the receiver. This circuit is known as an envelope detector and in
its simplest form it consists of a half- or full-wave rectier followed by a lowpass lter. Figure
1.16 (top) shows the same message signal that was given in Figure 1.8 and the bottom plot
shows the corresponding DSB signal with full carrier. The signal in the bottom panel was
produced from m(t) in the upper plot by rst forming the signal
s
0
(t) = (
m(t)
max(|m(t)|)
+ 1) cos !
c
t.
For this example the carrier frequency was chosen to be 8 times the signaling rate, i.e.
f
c
=
!c
2
=
8
T
. Dividing m(t) by max(|m(t)|) ensures that the envelope (
m(t)
max(|m(t)|)
+ 1) > 0
for all t. The signal s
0
(t) was then normalized to have a mean-square of 1.0. The normalized
signal, s(t), is shown in the bottom plot of Figure 1.16.
Figure 1.17 illustrates how the envelope detector demodulates a full-carrier DSB signal.
The upper plot shows the result of passing s(t) from Figure 1.16 through a full-wave rectier,
which produces |s(t)|. The lower plot shows the result of applying a low-pass lter to |s(t)|.
Compare the lower plot of Figure 1.17 with the original message signal, m(t), as shown
in the upper plot of Figure 1.16 to verify that, aside from a constant oset and a scaling
factor, the rectied and low-pass ltered signal is the same as m(t).
2
It should be clear that
if the envelope detector is used in a situation where Am(t) + B becomes < 0 (DSB with
partial carrier), then the recovered envelope will only follow m(t) during those times where
Am(t) +B > 0. In such cases an envelope detector will recover a distorted version of m(t).
The circuit of Figure 1.18 is commonly used to approximate the operation of an ideal
rectier/low-pass lter envelope detector. Assuming an ideal diode, the capacitor voltage
will charge to the peak of the input signal s(t) on a positive excursion of s(t), and then
decay exponentially with time constant = RC thereafter until the input voltage rises and
2
The recovered signal is also slightly delayed in time due to the group-delay of the lowpass lter.
1.6. LINEAR MODULATION 35
DSB, full carrier
1.0
0.0
1.0
2.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
m(t)
3.0
2.0
1.0
0.0
1.0
2.0
3.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
s(t)
2.0
Figure 1.16: Top: message signal representing 20 information bits generated using raised-
cosine pulses with = 0.1. Bottom: DSB with full-carrier signal, s(t), normalized such that
< s
2
(t) >= 1. Compare the bottom plot with the corresponding plot in Figure 1.8.
after low-pass lter
1.0
2.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
full-wave rectied s(t)
0.0
1.0
2.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
0.0
Figure 1.17: Upper plot shows |s(t)|, which is the result of passing s(t) through a full-wave
rectier. The lower plot was obtained by passing the signal shown in the upper plot through
a low-pass lter with cuto frequency larger than the bandwidth of m(t) (W) and smaller
than f
c
W. The transient at the beginning of the lower plot is the start-up transient of
the digital lowpass lter used to produce the simulated signal.
36 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
R C
s(t)
+
-
V
o
' A(1 + m(t))
Figure 1.18: Practical envelope detector circuit.
exceeds the capacitor voltage, at which time the capacitor voltage again follows the input
signal to its peak. The resulting output voltage is an approximation to the signal envelope,
i.e. V
o
' Am(t) + B. In order for the output to be a faithful reproduction of m(t), it is
necessary for the cut-o frequency of the lowpass RC lter to be larger than the bandwidth
of m(t) but smaller than the carrier frequency, i.e., W
1
2RC
f
c
W.
It is also possible to use a square-law device instead of the rectier, i.e., the rectier is
replaced with a square-law device as shown in Figure 1.19. Here the squaring operation is
drawn to emphasize the similarity to a coherent detector as shown in Figure 1.12. Recall
that the coherent demodulator can provide perfect recovery of the message signal. The
square-law detector can be thought of as an approximation to the coherent detector, where
the input signal is used as the local carrier reference.
s
0
(t)
s(t) = A(1 + m(t)) cos(!
c
t + )
s
out
(t)
LPF
Figure 1.19: A square-law detector.
After the squarer the signal is
s
0
(t) = A
2
(1 + 2m(t) +m
2
(t)) cos
2
(!
c
t + ) (1.30)
=
A
2
2
(1 + 2m(t) +m
2
(t)) [1 + cos(2!
c
t + 2)]
The cos 2 !
c
t term is rejected by the lowpass lter. What remains is
s
out
(t) =
1
2
A
2
(1 + 2m(t) + m
2
(t)) (1.31)
The dc term can be removed with a coupling capacitor, leaving
A
2
m(t) +
1
2
m
2
(t)
(1.32)
The rst term is the desired output and the second term is an unwanted distortion term
which will be small compared to the rst term when |m(t)| 1.
1.6. LINEAR MODULATION 37
The circuit shown in Figure 1.18 can function as an envelope detector or square-law
detector, depending on the magnitude of the input signal. If the peak input signal voltage is
relatively small (i.e. less than approx. 26 mV peak), then the diodes exponential current-
voltage characteristic can be expanded in Taylor series, in which case the leading nonlinear
term will be a square-law term. For large input signals, with peak values exceeding the diode
turn-on voltage, the circuit behaves more like an envelope detector based on a rectier and
low-pass lter.
1.6.4 Power eciency for DSB signals
Let us assume that the DSB signal (with carrier) is applied across a 1 resistor and dene
the time-average power in the signal to be the average of the instantaneous power over a
time interval that is long compared to all time scales of the message signal:
P
avg
=
1
T
T/2
T/2
s
2
(t)dt.
For a DSB signal
P
avg
=
1
T
T/2
T/2
(Am(t) +B)
2
cos
2
(!
c
t)dt.
Using brackets (<>) to denote the averaging operation, we have
P
avg
= < (Am(t) +B)
2
cos
2
(!
c
t) >
= <
1
2
(A
2
m
2
(t) + 2ABm(t) +B
2
)(1 + cos(2!
c
t)) >
'
1
2
(A
2
< m
2
(t) > +B
2
)
where we have used the fact that the long-time average value of the terms m(t), m
2
(t) cos(2!
c
t)
and m(t) cos(2!
c
t) are negligibly small. The term
1
2
A
2
< m
2
(t) > is the contribution to
average power from the message-signal modulation - this power is contained in the upper
and lower sidebands of the DSB signal. The term
1
2
B
2
is the carriers contribution to the
average power. Dene the modulation power eciency of a DSB-with-carrier signal to be the
fraction of the total transmitted power that resides in the information-bearing sidebands.
P
sidebands
P
avg
=
A
2
< m
2
>
A
2
< m
2
> +B
2
(1.33)
In the above equations < m
2
> represents the mean square value of the modulating signal,
m(t). Let us assume that m(t) varies symmetrically around zero. To ensure that Am(t) +
B > 0 a full-carrier signal will have B Amax |m(t)|. The largest power eciency will
occur when B is chosen to be as small as possible so let us assume that B = Amax |m(t)|.
The power eciency in this case is
P
sidebands
P
avg
<m
2
>
max |m(t)|
2
<m
2
>
max |m(t)|
2
+ 1
38 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
Since
<m
2
>
max |m(t)|
2
1, we conclude that the power eciency of a full-carrier signal can be
no larger than 50%. For example, suppose that m(t) is a waveform consisting of contiguous
rectangular pulses with amplitudes equal to 1. This corresponds to the digital modulation
scheme known as on-o keying; when m(t) = 1, then s(t) = 0 and the transmitter is
o. In this case,
<m
2
>
max |m(t)|
2
= 1 and the modulation eciency is 50%. Now suppose
that m(t) is something more like a typical analog message signal, i.e. suppose that m(t)
is a single sinusoidal tone, e.g., m(t) = cos(!
m
t). In this case
<m
2
>
max |m(t)|
2
= 0.5 and the
modulation eciency is 33%. A realistic value for analog voice or music signals might be
<m
2
>
max |m(t)|
2
< 0.25, which results in modulation eciency smaller than 20%.
In summary, in DSB with full carrier more than 50% of the average power is used to
send the carrier which carries no message essential information it is sent only to simplify
the demodulation process. With realistic signals, as much as 80% of the transmitted power
might be used to send the carrier. The remaining power is split between the two information-
bearing sidebands. This seems like a signicant waste of power. However, the advantage
gained by using this scheme is that the demodulator in the receiver can be a very simple
circuit consisting of as few as 3 passive components. If we attempt to make the scheme more
ecient by reducing the carrier power such that the modulation index becomes greater than
1, then it is necessary to use coherent demodulation (as described for DSB/SC) to recover
m(t) and demodulator complexity rises dramatically.
1.6.5 Single-Sideband (SSB)
As we have seen in sections 1.6.1 and 1.6.3, DSB signals (with or without carrier) occupy
a bandwidth B = 2W, twice the bandwidth of the message signal, m(t). Since all of the
information contained in m(t) is in one sideband, the other sideband is redundant. This
should be intuitively clear if we recall that the lower sideband in DSB results from translation
of the negative-frequency part of the spectrum of m(t) up into the positive frequency range.
We also know that the negative-frequency part of the spectrum of m(t) is related to the
positive-frequency part through M(!) = M
cos !
c
t
sin !
c
t
90
phase
shift
m(t)
+
-
Figure 1.22: Phasing method for generation of an SSB signal.
where sgn(!) is the signum function, dened by
sgn(!) =
1, ! > 0
0, ! = 0
1, ! < 0
The single-sideband modulator forms the modulated carrier signal
s(t) = m(t) cos !
c
t m(t) sin !
c
t,
as illustrated in Figure 1.22. The upper sign corresponds to a lower-sideband (LSB) signal,
and the bottom sign corresponds to an upper-sideband (USB) signal. To prove this, we can
calculate the Fourier transform of s(t). It is necessary to remember the following facts:
F[cos(!
c
t)] = [(! !
c
) + (! + !
c
)]
F[sin(!
c
t)] = j[(! + !
c
) (! !
c
)]
F[f(t)g(t)] =
1
2
F(!) G(!)
where denotes convolution. Now we can calculate the Fourier transform of s(t):
S(!) = F[s(t)]
=
1
2
[M(!) [(! !
c
] + (! + !
c
)] jsgn(!)M(!) j[(! + !
c
) (! !
c
)]]
=
1
2
[M(! !
c
) +M(! + !
c
) [sgn(! + !
c
)M(! + !
c
) sgn(! !
c
)M(! !
c
)]]
= M(! !
c
)
1
2
[1 sgn(! !
c
)] +M(! + !
c
)
1
2
[1 sgn(! + !
c
)]
Consider the upper sign for now, and examine the rst term, which represents the positive-
frequency part of S(!). The term
1
2
[1 sgn(! !
c
)] =
1, ! < !
c
1
2
, ! = !
c
0, ! > !
c
1.6. LINEAR MODULATION 41
is equal to zero above the carrier frequency, and hence removes the upper sideband of
M(! !
c
). If the lower sign is chosen, then the lower sideband will be multiplied by zero.
Hence, when the upper sign is chosen, the part of the spectrum above the carrier frequency
will be zeroed out, producing LSB. If the lower sign is chosen, the part of the spectrum
below the carrier frequency will be zeroed out, producing USB. Exact cancellation of the
unwanted sideband does not occur in practice, because of imperfections in the lters that
produce the /2 phase shift and because of amplitude imbalance between the two terms.
1.6.6 SSB Demodulation
Single sideband signals can be demodulated using coherent demodulation. The input signal
is either a USB or LSB signal:
s(t) = A[m(t) cos !
c
t m(t) sin !
c
t] (1.34)
With perfect carrier recovery, the output of the coherent demodulator is
s
out
(t) = LPF[s(t) cos(!
c
t)] =
1
2
Am(t).
If the recovered carrier has a phase error, i.e., if the recovered carrier is cos(!
c
t + ) then
it is easy to show that
s
out
(t) = LPF[s(t) cos(!
c
t + )] =
1
2
Am(t) cos
1
2
A m(t) sin ,
where LPF[] represents the lowpass lter operation. The rst term is the desired signal
which has amplitude
1
2
Acos , and the second term represents undesired crosstalk from
the Hilbert transform signal m(t). If m(t) is an analog voice signal, the crosstalk contribution
from m(t) would not sound signicantly dierent from the desired signal, m(t). Intelligibility
of voice does not depend critically on the phase relationship between the composite frequency
components. For digital data transmission, however, the crosstalk term represents distortion
and will degrade the performance of the system. It is necessary for the phase error to
be nearly zero in order for the desired signal to have an amplitude that is much larger than
the crosstalk term. In a digital data transmission system some form of automatic carrier
synchronization would be necessary.
When the message signal is an analog voice waveform, there is an important distinction
between coherent demodulation as applied to demodulation of DSB and SSB. As we learned
earlier, when demodulating DSB it is necessary to have frequency perfect synchronization
in order to recover an intelligible signal. When demodulating an analog SSB voice signal,
however, it turns out that some frequency error can be tolerated. The eect of a frequency
error, !, is to change the pitch of the received signal. For example, suppose that the
modulating signal is a tone, i.e., m(t) = cos !
m
t. Then s
USB
(t) = A[cos(!
m
t) cos(!
c
t)
sin(!
m
t) sin(!
c
t)]. Lets assume that at the receiver this signal is multiplied by cos[(!
c
+
!)t]. Then, after the multiplier we have the following:
s
0
(t) = A{cos(!
m
t) cos(!
c
t) sin(!
m
t) sin(!
c
t)} cos[(!
c
+ !)t] (1.35)
= Acos[(!
c
+ !
m
)t] cos[(!
c
+ !)t]
=
1
2
A{cos[(2!
c
+ !
m
+ !)t] + cos[(!
m
!)t]}
42 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
We assume that both !
m
and ! are !
c
so that the lowpass lter will reject the rst
term. The output of the lter is then
s
out
(t) =
A
2
cos[(!
m
!)t] (1.36)
Note that the eect of the frequency error is to cause the demodulated tone to be shifted
in frequency by an amount equal to the frequency error. Contrast this to the DSB case
where a frequency error resulted in two tones at the output for a single input tone. In
SSB communication receivers used for analog voice reception the local oscillator can be a
free-running oscillator (often called the beat frequency oscillator, or BFO). In practice the
operator of the receiver would tune the BFO until the recovered speech signal is intelligible.
The tuning is fairly critical, but frequency errors of several 10s of Hz can be tolerated. The
audible eect is to shift the pitch of the recovered speech up or down by the frequency error,
f.
Notice that coherent detection can be used to demodulate all of the forms of linear
modulation that have been discussed so far - DSB/SC, DSB/with carrier, and SSB. Only in
the latter case is it permissible for the receiver oscillator to have a frequency error, and this
applies only when the message signal is an analog voice signal. For DSB/SC and DSB/with
carrier it is necessary to have perfect frequency synchronization between the transmitter
and receiver.
While SSB makes the most ecient use of the radio-frequency spectrum, this type of
modulation places relatively severe demands on the power amplier used to amplify the
signal for transmission, because the amplier needs to be capable of delivering peak power
that is signicantly larger than the average power contained in the signal. This is illustrated
in homework problem 10. In many cases, a better scheme for obtaining the same bandwidth
eciency as SSB with more reasonable peak to average power requirements is the quadrature
multiplexing scheme discussed in the following section.
1.6.7 Quadrature Multiplexing
Quadrature multiplexing is not really a new modulation scheme, rather it is a method for
modulating two baseband signals, each having bandwidth W, onto one carrier such that the
total occupied bandwidth is 2W, i.e. the bandwidth occupied by the modulated carrier is
equal to the total bandwidth occupied by the original message signals. This doubles the
bandwidth eciency of single-channel DSB, wherein a baseband signal with bandwidth W is
associated with an occupied bandwidth of 2W. Of course, we already have encountered one
modulation scheme (SSB) which produces a modulated carrier with the same bandwidth as
the original message signal. In fact, using SSB, it is possible to transmit the upper sideband
of one signal, m
1
(t), and the lower sideband of another signal, m
2
(t), thereby transmitting
two signals using one carrier, and using the same total bandwidth as that occupied by the
original baseband signals. This is sometimes done and the technique is called independent
sideband or ISB modulation. Quadrature multiplexing oers a simpler way to accomplish
the same thing while avoiding the complexities (lters and phasing networks) involved with
generating the SSB signal, and that also results in a signal with more favorable peak to
average power characteristics, The idea behind quadrature multiplexing it is to use two
carriers with the same frequency but which dier in phase by 90 degrees. If the two message
signals are m
1
(t) and m
2
(t), then the modulated carrier signal resulting from quadrature
multiplexing is:
s(t) = m
1
(t) cos !
c
t +m
2
(t) sin !
c
t (1.37)
1.6. LINEAR MODULATION 43
Figure 1.23 is a schematic of a quadrature multiplexer/modulator that implements this
scheme. The quadrature-multiplexed signal simply consists of two DSB/SC components,
m
1
(t)
s(t) = m
1
(t) cos !
c
t + m
2
(t) sin!
c
t
cos !
c
t
m
2
(t)
90
cos !
c
t
sin !
c
t
Figure 1.23: Quadrature multiplexer/modulator.
and the spectra of the two components will overlap, as in Figure 1.24. Because the spectra
f
|S(f)|
2W
f
c
Figure 1.24: Two DSB/SC components with overlapping spectra.
overlap it may look as if it would be impossible to recover the individual signals, m
1
(t)
and m
2
(t), however it can be done if the signals are coherently demodulated as shown
in Figure 1.25, which shows a quadrature de-multiplexer. This system consists of two
coherent demodulators, one (the upper branch) operates with an in-phase local oscillator,
and produces an output, I(t), that is called the in-phase signal component. The lower branch
operates with a quadrature local oscillator, and produces an output, Q(t), that is called the
quadrature signal component. Provided that the local oscillator is properly synchronized
with the carrier of the incoming signal, the in-phase output will be proportional to m
1
(t) and
the quadrature output is proportional to m
2
(t). The demonstration that this demodulation
scheme will work is left as an exercise.
An application of quadrature multiplexing is the analog color television signal. The color,
or chrominance, information is contained in two signals, I(t) and Q(t), and is transmitted
by quadrature multiplexing these signals onto a subcarrier with a frequency of approximately
3.58 MHz. In order to detect the color information at the receiver, it is necessary to have
44 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
I(t) =
1
2
m
1
(t)
s(t) = m
1
(t) cos !
c
t + m
2
(t) sin!
c
t
cos !
c
t
Q(t) =
1
2
m
2
(t)
90
cos !
c
t
sin !
c
t
Figure 1.25: De-multiplexing of quadrature-multiplexed signals.
synchronization information for the receiver oscillator. This information is provided in the
television signal by adding a short burst of the transmitters 3.58 MHz signal into the
composite video signal at the end of each horizontal scan line. The signal is known as
the color burst. Although with this scheme the carrier information is present for only a
small fraction of the time, it is still possible to use its information to lock an oscillator in
the receiver to the transmitters frequency and phase. This is usually done by employing a
phase-locked loop which generates a continuous carrier and uses the periodic synchronization
information to adjust the frequency and phase of the carrier.
1.6.8 Vestigial Sideband Modulation - VSB
Recall that SSB is the most bandwidth-ecient modulation, but it is dicult to get good
low-frequency response in a SSB system because of the obstacles involved in the ltering
or phasing methods of SSB generation. Low-frequency response is necessary in an analog
television system so that video images with large areas of the same color and brightness can
be transmitted. In a digital communication system, low frequency response is necessary in
order to transmit pulses with non-zero mean. The hardware requirements are considerably
relaxed if we allow a part (vestige) of the unneeded sideband to be transmitted. If, in
addition, a carrier component is transmitted, demodulation is simplied signicantly. This
vestigial sideband scheme is used for all commercial television transmissions, wherein the
upper sideband and a vestige of the lower sideband are transmitted. Vestigial sideband is
used for transmission of the video portion of analog television signals using the National
Television Systems Committee (NTSC) system (see Figure 1.26a) and for transmittion of
digital television (DTV) signals using the Advanced Television Systems Committee (ATSC)
standard (Figure 1.26b). The ATSC modulation is called 8VSB.
ATSC message signals are constructed using pulses whose Fourier transform is the
square-root of the raised-cosine pulses discussed earlier. In the receiver, the signal is l-
tered using a square-root raised-cosine response so that the received pulses will have a
raised-cosine spectrum. A pulse can take on 8 possible amplitudes, so each pulse repre-
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 45
309.441 kHz 309.441 kHz
Carrier
6.0 MHz
1
1
p
2
f
r
/2 = 5.381119 MHz
Figure 1.26: The spectrum of a television signal as transmitted for digital television (DTV)
using the ATSC (Advanced Television Systems Committee) 8VSB system.
sents 3 bits of information. The fractional excess bandwidth of the pulses is = 0.0575
and the signaling rate is f
r
= T
1
= 10.762238 Mpulses/sec. Each pulse represents 3 bits
of information so channel bit-rate is 3 10.762 = 32.286 Mbits/s. Redundant bits (cod-
ing) and overhead reduce the eective data rate to 19.3 Mbits/s. The signaling interval is
T = 1/10.762 10
6
' 93 ns. The one-sided bandwidth of the pulses is (1 +)f
r
/2 = 5.6906
MHz. A DSB signal generated from a message signal constructed from these pulses would
have bandwidth (1 + )f
r
= 11.381 MHz which exceeds the 6 MHz width of a single TV
channel. As shown in Figure 1.26b, most of the lower sideband is removed to form the
ATSC 8VSB signal spectrum. This is accomplished with a root raised-cosine rollo that
begins at approx. 309 kHz above the carrier frequency and falls to zero 309 kHz below the
carrier frequency. Hence, the bandwidth occupied by the VSB signal is equal to the width
of the upper sideband (5.6912 MHz) plus the width of the vestige of the lower sideband
(309.4 kHz). The 8VSB signal is usually described as having an excess bandwidth of 11.5%
(although the message signal pulses have 5.75% excess bandwidth) because the vestige of
the lower sideband doubles the excess bandwidth.
1.7 Angle Modulation (Nonlinear Modulation)
We turn now the the other class of modulation schemes known as angle modulation, or
sometimes nonlinear modulation. In this case the transmitted signal has the form
s(t) = A
c
cos(!
c
t + (t)) (1.38)
where A
c
is a constant. Thus the peak amplitude of the modulated signal is constant, but
the phase angle is varied in response to the modulating signal. One important advantage
of angle-modulation systems is their inherent insensitivity to amplitude uctuations that
may be present on the the received signal due to fading or noise. Since the angle-modulated
signal has a constant amplitude envelope, the received signal can be passed through a hard
limiter to remove amplitude uctuations due to noise and fading; the modulation will not be
distorted by this operation. The constant amplitude feature of these signals is also desirable
because the peak to average power ratio is equal to one, allowing for ecient amplication
46 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
of such signals to high power levels. Angle modulation is nonlinear in the sense that if m
1
(t)
angle modulates a carrier to produce s
1
(t) (i.e. m
1
!s
1
) and m
2
!s
2
, then m
1
+m
2
does
not result in the modulated signal s
1
+s
2
. As a result, the spectrum of an angle-modulated
signal is not related to the spectrum of the modulating signal in a straightforward way.
There are two basic types of angle modulation used for analog signals. These are distin-
guished by the way in which m(t) is mapped onto the angle (t).
1. Phase Modulation (PM):
The phase angle is a scaled version of m(t), i.e.,
(t) = k
p
m(t) (1.39)
where k
p
is called the phase deviation constant.
2. Frequency Modulation (FM):
The derivative of the phase angle is proportional to m(t), i.e.,
d
dt
= k
f
m(t) (1.40)
where k
f
is called the frequency deviation constant.
Note that in either case (PM or FM) the instantaneous frequency of the signal is given by
!
inst
=
d
dt
(!
c
t + (t)) = !
c
+
d
dt
(1.41)
In the case of FM we have
!
inst
= !
c
+k
f
m(t) = !
c
+!(t) (1.42)
Thus, frequency modulation is generated simply by varying the frequency of the carrier
signal in direct response to m(t). For a general m(t), then, the mathematical forms of
phase- and frequency-modulated signals are as given in Equations 1.43 and 1.44:
Phase modulated : s(t) = A
c
cos(!
c
t +k
p
m(t)) (1.43)
Frequency modulated : s(t) = A
c
cos(!
c
t +k
f
t
1
m(t
0
)dt
0
) (1.44)
1.7.1 Spectrum of Angle-modulated Signals
For nonlinear modulation there is no simple modulation theorem that provides a rela-
tionship between M(f) and S(f). Well consider a simple case to gain some insight, and
then well give a general rule of thumb for estimating the bandwidth of an angle-modulated
signal.
1.7.1.1 Sinusoidal Modulation
Suppose that the message signal is a sine wave, i.e.,
m(t) = A
m
sin !
m
t (1.45)
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 47
Then
s
PM
(t) = A
c
cos(!
c
t + A
m
k
p
sin !
m
t) (1.46)
s
FM
(t) = A
c
cos(!
c
t
A
m
k
f
!
m
cos !
m
t) (1.47)
This example illustrates that except for a phase shift the PM and FM signals have the
same functional form for the sinusoidal modulation case. For the purpose of deriving the
spectrum of this type of signal it is sucient to consider a function of the form
s(t) = A
c
cos(!
c
t + sin !
m
t) (1.48)
where is called the modulation index, which is simply the maximum phase deviation for
either FM or PM. The modulation index can be related to the phase, or frequency, deviation
constants through
= A
m
k
p
(PM) (1.49)
=
A
m
k
f
!
m
(FM) (1.50)
Another interpretation for the modulation index comes from noting that the maximum
instantaneous frequency deviation of the phase- or frequency-modulated signal is !
max
=
!
m
, so
=
!
max
!
m
=
f
max
f
m
(1.51)
Now the spectrum of s(t) can be derived. The Fourier transform of s(t) can be written as
follows:
S(!) =
1
1
s(t)e
j!t
dt (1.52)
= A
c
1
1
e
j!t
cos(!
c
t + sin !
m
t)dt (1.53)
This integral can be found in an integral table or table of Fourier transforms and the result
is
S(!) = A
c
1
n=1
J
n
()(! !
c
n!
m
) +J
n
()(! + !
c
+n!
m
) (1.54)
This result shows that the signal s(t) has a spectrum which consists of a set of impulses
that are located at the discrete frequencies ! = (!
c
n!
m
). The spectrum consists of
an innite number of sidebands that are separated from the carrier frequency by integer
multiples of the frequency of the modulating tone, !
m
. The modulation index is just a
constant that describes the peak phase deviation of the signal. The function J
n
(x) is called
the Bessel function of rst kind of order n. These functions are tabulated, and for a given
value of n, tables of J
n
(x) can be found in many mathematics reference books. The Bessel
functions of negative order (n < 0) are related to Bessel functions of positive order through
J
n
(x) = (1)
n
J
n
(x) (1.55)
48 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
The terms J
n
() that appear in the expression for the spectrum can be thought of as
coecients that determine the strength of each of the impulses. Plots of the Bessel functions
for n = 0, 1, 2, 3, 4, and 5 are shown in Figure 1.27.
Taking the inverse transform of the S(!) gives an alternative form of the time-signal
that may yield some insight. Note that each pair of delta functions at frequency !
n
will
yield a term of the form 2 cos !
n
t when the inverse transform is applied. Then
s(t) =
1
2
1
1
S(!)e
j!t
d! = A
c
1
n=1
J
n
() cos(!
c
t + n!
m
t) (1.56)
Thus, the angle-modulated signal (with sinusoidal modulation) can be represented as an
innite superposition of cosinusoidal components having frequencies |!
c
+n!
m
|.
2. 4. 6. 8. 10.
-0.4
-0.2
0.2
0.4
0.6
0.8
1.
Figure 1.27: Bessel functions of order n for n=0,1,2,3,4,5.
1.7.1.2 Example - sinusoidally modulated signal
Consider an angle-modulated signal of the form:
s(t) = A
c
cos(220t + sin(22t)) (1.57)
The carrier frequency for this signal is 20 Hz and the frequency of the modulating tone
is 2 Hz. The peak phase deviation associated with the signal () is 1 radian, and the
peak frequency deviation is 2 Hz. According to Equation 1.56 this signal can be written as
s(t) = A
c
1
n=1
J
n
(1) cos(2(20 + 2n)t). The spectrum of this signal is easily determined
by inspection of Equation 1.57. Each cosinusoidal component will contribute a pair of
delta functions located at f = (20 + 2n) with amplitude given by the value of J
n
(1). The
resulting line spectrum is shown in Figure 1.28. Note that for this example the amplitudes
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 49
of the sidebands are smaller than that of the carrier. Figure 1.28 should be compared with
Figure 1.27. The amplitude of each line in the spectrum can be determined by inspecting
the values of J
n
() at = 1 shown in Figure 1.27 by the dashed line. The value of J
0
(1)
is the amplitude of the carrier component; the value of J
1
(1) gives the amplitude of the
rst sidebands, etc. If the peak phase deviation of an angle-modulated signal happens to
10. 20. 30. 40.
-0.8
-0.6
-0.4
-0.2
0.2
0.4
0.6
0.8
Figure 1.28: Line spectrum of angle-modulated signal with 20 Hz carrier and 2 Hz modu-
lating tone.
correspond to a zero of one of the Bessel functions, then the corresponding component will
not be present in the spectrum. A few of the zeros of the Bessel functions J
0
(x) and J
1
(x)
are given in Table 1.5.
Table 1.5: Zeros of Bessel functions.
J
0
(x) has zeros at: x= 2.4048 5.5201 8.6537 11.7915
J
1
(x) has zeros at: x= 0.0000 3.8317 7.0156 10.1735
If, for example, the peak phase deviation is 5.5201 radians, the Fourier spectrum of the
(sinusoidally modulated) angle-modulated signal will not have a carrier component. This
phenomenon is useful for precise adjustment of the frequency deviation of FM transmitters.
In practice the peak frequency deviation (f
max
) allowed for an FM signal is xed at some
value which depends on the type of signal that is being generated. For example, a commercial
FM broadcast signal is limited to a peak frequency deviation of 75 kHz. With sinusoidal
modulation the peak phase deviation () of such a signal will depend on the frequency of
the modulating tone through
=
f
max
f
m
(1.58)
Suppose an engineer wants to adjust the peak frequency deviation of a transmitter to some
value, f
max
. A precise adjustment can easily be made by using sinusoidal modulation and
50 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
adjusting the frequency of the modulating tone to a value f
m
such that, when f
max
is set
properly, the peak phase deviation will take on a value corresponding to a zero of J
0
(x).
The spectrum of the transmitted signal is then monitored on a spectrum analyzer, and the
frequency deviation is adjusted until the carrier component vanishes. Similar adjustments
could be made by using any of the sidebands instead of the carrier component.
1.7.2 Bandwidth of Angle-modulated Signals
Figures 1.29 through 1.32 show the magnitude of the Fourier Transform of a carrier that
is angle modulated by a sinusoidal tone with = 1, 2.4, 7, and 20. The spectrum actually
consists of delta functions. The plots show the location and the relative amplitudes (absolute
value) of the delta functions. Notice that in all cases the most signicant sidebands are
contained within the interval f
c
( + 1)f
m
. Outside of this interval the amplitude of the
sidebands is relatively small and decreases rapidly with increasing separation from the carrier
frequency. So a good approximation for the bandwidth of a carrier that is angle-modulated
by a sinusoidal message signal is
BW ' 2( + 1)f
m
.
Since
=
f
max
f
m
,
the bandwidth can be written as
BW ' 2(f
max
+f
m
). (1.59)
This is known as Carsons rule. It can be shown that the bandwidth given by equation 1.59
will contain at least 98% of the power associated with the angle modulated signal.
When the message signal is non-sinusoidal (as in almost all practical situations) and
has bandwidth W, a reasonable approximation to the bandwidth results if Carsons rule is
modied by replacing the tone frequency by the bandwidth of the message signal. Thus, for
a non-sinusoidal message signal
BW ' 2(f
max
+W).
When the modulating signal has nite bandwidth W can usually be taken to be the highest
frequency that is contained in the signal. If the modulating signal has innite bandwidth
then W is chosen so that it contains most of the power in the signal.
Notice that when the peak frequency deviation is much smaller than the highest fre-
quency in m(t), i.e., if f
max
W, then Carsons Rule reduces to
BW ' 2W (1.60)
When this approximation is valid, the modulation is called narrowband angle modulation,
and the bandwidth is essentially the same as for AM or DSB. On the other hand, if f
max
W, the modulation is called wideband angle modulation and Carsons rule becomes
BW ' 2f
max
(1.61)
which states that the bandwidth is approximately twice the peak frequency deviation.
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 51
0.0
0.2
0.4
0.6
0.8
1.0
15 10 5 0 5 10 15
|S(f)|
ffc
fm
= 1.0
Figure 1.29: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 1. Note that the frequency axis is the normalized frequency dierence from the carrier
frequency, i.e.
ffc
fm
.
0.0
0.2
0.4
0.6
0.8
1.0
15 10 5 0 5 10 15
|S(f)|
ffc
fm
= 2.4
Figure 1.30: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 2.4.
52 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
0.0
0.2
0.4
0.6
0.8
1.0
15 10 5 0 5 10 15
|S(f)|
ffc
fm
= 7.0
Figure 1.31: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 7.
0.0
0.2
0.4
0.6
0.8
1.0
25 20 15 10 5 0 5 10 15 20 25
|S(f)|
ffc
fm
= 20.0
Figure 1.32: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 20.
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 53
1.7.3 Demodulation of FM
Most schemes for demodulating FM make use of the concept illustrated in the idealized
demodulator shown in Figure 1.33. The input signal is of the form:
d
dt
Envelope
Detector
DC Block
s(t) s
1
(t) s
2
(t) s
out
(t)
Figure 1.33: Idealized demodulator for FM signals.
s(t) = A
c
cos(!
c
t +k
f
t
1
m(t
0
)dt
0
) (1.62)
After dierentiation,
s
1
(t) =
d
dt
s(t) = A
c
[!
c
+k
f
m(t)] sin(!
c
t +k
f
t
1
m(t
0
)dt
0
) (1.63)
If !
c
+k
f
m(t) 0, this looks very much like the AM signal. Recall:
S
AM
(t) = A
c
(1 +m(t)) cos(!
c
t + ) (1.64)
After the envelope detector, approximately,
s
2
(t) = A
c
(!
c
+k
f
m(t)) (1.65)
After dc blocking,
s
out
(t) = A
c
k
f
m(t) (1.66)
The dierentiation operation can be approximated using a delay/dierence scheme as
shown in Figure 1.34. For the dierence approximation to be a good approximation to the
Envelope
Detector
DC Block
s(t) s
1
(t) s
2
(t) s
out
(t)
Delay,
Figure 1.34: Approximation to the idealized FM demodulator.
derivative, the delay must be small compared to the period of the carrier oscillation, i.e.
1/f
c
. This approach to demodulating FM yields a very small output signal, because
the envelope of the signal after the dierentiator is proportional to !
c
+k
f
m(t). The second
54 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
term is simply the instantaneous frequency deviation, so if the maximum deviation from the
carrier frequency is f
max
, the ratio of peak to minimum envelope amplitude is
1 +
fmax
fc
1
fmax
fc
.
Consider FM Broadcast, where f
max
= 75 kHz and f
c
100 MHz. The maximum to min-
imum envelope amplitude ratio is then 1.0015. Thus, the output from the envelope detector
will be very small. For this reason, the dierentiator/envelope detector is practical only for
signals with relatively large deviations. When some distortion can be tolerated, much larger
recovered signal can be obtained by replacing the dierentiator with a single-resonator band-
pass lter whose center frequency is deliberately oset from the carrier frequency. The idea
is to place the carrier frequency on the skirts of the bandpass response, so that frequency
deviations are translated into envelope variation. By employing a lter with fairly steep
skirts (e.g. an LC lter with high Q), relatively large envelope variation can be obtained,
even for small frequency deviations. This approach to demodulating FM is called slope
detection.
An even better method for demodulating FM can be implemented by converting the
instantaneous frequency deviation to relative phase deviations, and then using a phase-
detector to convert the phase deviation to a voltage. A simple LC resonator can be used
to convert the frequency deviation of the incoming signal into (relative) phase deviation. A
circuit that accomplishes this is described in detail in Section 4.5.
1.8 Quadrature Modulation/Demodulation
In section 1.6.7 we pointed out that quadrature multiplexing can be used to send two
message signals, m
1
(t) and m
2
(t), on one carrier, thereby doubling the bandwidth eciency
relative to single-channel DSB modulation. This property makes the quadrature multiplexer
essentially a universal modulator, since it can be used to produce any type of modulation by
properly choosing the two message signals. Consider a carrier signal that is both amplitude
and angle modulated,
s(t) = A(t) cos(!
c
t + (t)). (1.67)
Straightforward application of the identity cos(a+b) = cos a cos bsin a sin b allows equation
1.67 to be written as
s(t) = A(t) cos (t) cos !
c
t A(t) sin (t) sin !
c
t
or
s(t) = m
1
(t) cos !
c
t +m
2
(t) sin !
c
t
with
m
1
(t) = A(t) cos (t) (1.68)
m
2
(t) = A(t) sin (t). (1.69)
Therefore, to create a signal that is amplitude and/or angle modulated using a quadra-
ture modulator, it is necessary to create the appropriate m
1
(t) and m
2
(t) signals according
to 1.68 and 1.69, and apply these signals to a quadrature multiplexer, as shown in Figure
1.23.
1.8. QUADRATURE MODULATION/DEMODULATION 55
Similarly, a quadrature de-multiplexer, combined with appropriate signal processing can
serve as a universal demodulator whether or not a modulated signal was produced using
a quadrature modulator. Refer to Figure 1.25, and note that when the input signal is an
amplitude and/or angle modulated signal, the components m
1
(t) and m
2
(t) are given by
equations 1.68 and 1.69. Thus, the output from the in-phase channel will be I(t) =
1
2
m
1
(t) =
1
2
A(t) cos (t), and the output from the quadrature channel will be Q(t) =
1
2
A(t) sin (t).
The I(t) and Q(t) outputs can be provided as inputs to a signal processor, which extracts
the envelope, A(t), and/or the phase modulation, (t). The envelope is obtained by noting
that
A(t) = 2
I(t)
2
+Q(t)
2
. (1.70)
The phase modulation is obtained from
(t) = tan
1
(
Q(t)
I(t)
). (1.71)
If the input signal is frequency modulated, then the demodulator must produce a signal
that is proportional to
d
dt
. Dierentiating equation 1.71 gives the instantaneous frequency
of the signal in terms of the I and Q outputs of the quadrature demultiplexer:
!(t) =
d
dt
=
1
I
2
+Q
2
(I
dQ
dt
Q
dI
dt
). (1.72)
If the signal envelope is constant, i.e. if the signal is purely angle modulated, then the
term I
2
+Q
2
will be constant. This will be the case if the signal has been passed through a
limiter before application to the quadrature demodulator. In that case, it is only necessary to
compute the quantity I
dQ
dt
Q
dI
dt
, which will be proportional to the instantaneous frequency
deviation of the input signal, where the deviation is measured with respect to the local carrier
reference.
In many applications, frequency demodulation is performed after the signal has been
digitized. A discrete-time version of equation 1.72 can be written in terms of sampled-data
sequences {I
n
} and {Q
n
} by replacing the time-derivatives with rst-dierence approxima-
tions, e.g.
dQ
dt
!
1
t
(Q
n
Q
n1
). The result simplies nicely and is easily implemented in
a digital signal processor:
!
n
=
(I
n
Q
n1
Q
n
I
n1
)
t(I
2
n
+Q
2
n
)
. (1.73)
It can be shown (see homework problem 14) that if the quadrature demodulator is to be
used only for implementing envelope or instantaneous frequency recovery (as in equations
1.70 and 1.72), then it is not necessary for the local oscillator to be synchronized with the
carrier of the incoming signal. In the case of envelope recovery, frequency or phase osets
have no eect, provided that the bandwidth of the LPFs is large enough to accomodate
any frequency shifts in the I and Q branches. In the case of FM demodulation according
to equation 1.72, frequency error in the local oscillator causes a DC oset in the output of
the demodulator. The DC oset can be used as a tuning aid, or as an error signal in an
automatic frequency control (AFC) loop, which would tune the local oscillator to minimize
the frequency error.
1.8.1 Carrier Frequency and Phase Synchronization
The I and Q outputs from a quadrature demodulator can be used to form a control signal
that corrects the instantaneous frequency of the local oscillator, for the purpose of syn-
56 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
chronizing the local oscillator to that of the incoming signals carrier. First, suppose that
the local oscillator is perfectly synchronized with the incoming carrer, and that the input
signal is a simple DSB signal so that s(t) = m(t) cos !
c
t. In this case, the I branch of the
quadrature demodulator output contains the desired message signal, I(t) =
1
2
m(t), and the
Q branch output is zero. Notice that the product of the I and Q outputs is zero in this case.
If the local oscillator is not synchronized to the incoming signals carrier, then, in general,
the local oscillator has both frequency and phase errors, i.e. the local oscillator is given by
cos((!
c
+ !)t + ).
The quadrature local oscillator will then be
sin((!
c
+ !)t + ).
If the input signal is DSB, then s(t) = m(t) cos(!
c
t). The output from the I channel is:
I(t) =
1
2
m(t) cos(!t + ).
The output from the Q channel is:
Q(t) =
1
2
m(t) sin(!t + ).
The product of the I and Q channel outputs is
I(t)Q(t) =
1
8
m
2
(t) sin(2!t + 2).
Notice that the IQ product is non-zero in the presence of frequency and/or phase errors.
It turns out that if this IQ product is used to control the instantaneous frequency of the
local oscillator, the resulting closed-loop feedback system will quickly drive the IQ product
to a small value, and will reach an equilibrium where the frequency error ! = 0, and the
phase error is small. Such a feedback control system is called a Costas Loop, and is
shown in Figure 1.35, where the IQ product is formed to generate a control voltage, V
c
(t).
The local oscillator must be implemented as a Voltage Controlled Oscillator, so that the
voltage V
c
controls the instantaneous frequency of the oscillator.
In Figure 1.35, the feedback loop is opened by a switch in the control voltage path.
It can be shown that, if the initial frequency error is small enough, when the switch is
closed, the frequency error will be driven to zero, so that the control voltage becomes
V
c
(t) =
1
8
m
2
(t) sin(2). The DC component of this control signal is V
c
=
1
8
m
2
(t) sin 2.
The loop adjusts itself so that this DC voltage is just large enough to tune the local VCO
to the carrier frequency of the incoming signal. The adjustment is achieved by allowing a
nite, static phase error, in the loop. The phase error will be just enough to provide the
DC control voltage necessary to tune the VCO to the frequency of the incoming signal. If
the VCO output frequency is a sensitive function of the control voltage, the required static
phase oset will be very small, and the associated phase error will also be small. Practical
implementations of the Costas Loop usually include a lowpass lter between V
c
and the VCO
to smooth the control voltage and control the dynamic response of the loop (the response
to transient changes in the incoming signals phase).
1.9. REFERENCES 57
I(t)
s(t) = m(t) cos !
c
t
cos((!
c
+ !t) + )
Q(t)
90
V
c
(t)
Figure 1.35: Costas Loop for carrier synchronization. The IQ product is used to control the
instantaneous frequency of the local oscillator. When the switch is closed, the loop drives
the IQ product to a small value.
1.9 References
1. Jordan, Edward C. and Keith G. Balmain, Electromagnetic Waves and Radiating
Systems, Prentice Hall, 1968.
2. Proakis, J. and M. Salehi, Communication Systems Engineering, 2nd ed., Prentice-
Hall, 2002.
3. Haykin, Simon, Communication Systems, 3rd ed., John Wiley and Sons, Inc., 1994.
4. Friis, Harald T., A Note on a Simple Transmission Formula, Proc. IRE, pp. 254-256,
May, 1946.
58 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
1.10 Homework Problems
1. One signal applied to the input of an ideal multiplier is
s
1
(t) = cos(1000 t) + cos(1800 t) (1.74)
The signal applied to the other input is
s
2
(t) = cos(7000 t) + cos(20, 000 t) (1.75)
The output from the multiplier is applied to an ideal lter that passes only frequencies
between 3.5 kHz and 10.0 kHz. List all frequencies that are present at the output of
the lter.
2. An AM signal with full carrier can be represented as follows:
s(t) = A(1 +m(t)) cos(!
c
t), |m(t)| < 1 for all t.
An envelope detector can be used to demodulate full-carrier AM. An ideal envelope
detector is realized with a rectier and a lowpass lter, as shown in Figure 1.36. The
s(t)
s
0
(t)
s
out
(t)
Full-wave
Rectier
B
Figure 1.36: An ideal envelope detector, consisting of ideal full-wave rectier and lowpass
lter.
output of the full-wave rectier can be written as s
0
(t) = |s(t)| = s(t)p(t) where p(t)
is a square wave with the following properties:
p(t) is an even function,
max{p(t)} = 1, min{p(t)} = 1,
p(t) has period T = 2/!
c
.
(a) Express p(t) in a Fourier series.
(b) Sketch the amplitude spectrum of s
0
(t). Assume that m(t) has the amplitude
spectrum shown in Figure 1.37. The bandwidth of m(t) is assumed to be
1
|M(f)|
f W W
Figure 1.37: Amplitude spectrum of m(t).
small compared to the carrier frequency, i.e. W f
c
.
1.10. HOMEWORK PROBLEMS 59
(c) Specify the minimum and maximum possible cuto frequency, B, for the
low-pass lter that will allow m(t) to be recovered at the output of the lter.
3. The message signal m(t) = A sin (!
m
t) modulates a carrier signal given by cos(!
c
t)
where we assume !
m
!
c
.
(a) If the modulation is lower sideband (LSB), show that the modulated signal can
be represented by
f(t) =
A
2
[sin(!
m
t) cos(!
c
t) cos(!
m
t) sin(!
c
t)] (1.76)
(Hint: Write down an expression for the DSB signal and manipulate it so that
you have two terms one for the upper sideband and one for the lower sideband.
Then subtract the upper sideband component.)
(b) Suppose that we attempt to demodulate this signal using a square-law detector.
The output of a square-law detector is the square of the input signal. Assume that
the detector is followed by a low-pass lter that removes all frequency components
with frequencies larger than !
c
. Find the output of the square-law detector/low-
pass lter. Can the modulating signal (A sin(!
m
t)) be recovered?
(c) Consider the same problem as in 3b, except assume that the carrier is rst rein-
serted, i.e., we add a term K cos(!
c
t) to f(t) to give:
f
0
(t) =
A
2
[sin(!
m
t) cos(!
c
t) cos(!
m
t) sin(!
c
t)] + K cos(!
c
t) (1.77)
Assume that this signal is applied to the square-law detector/low-pass lter. Can
the modulating signal be recovered? How will the output of the demodulator
depend on the amplitude of the reinjected carrier? Note: In practice the oscillator
that provides the reinjected carrier for SSB demodulation is called the beat-
frequency oscillator (BFO).
4. Find the instantaneous frequency as a function of time for the following signal:
s(t) = 10 cos[2(10
8
t 10
4
t
2
) 4]
5. Consider an angle-modulated signal of the form
s(t) = A cos(!
c
t + 5 cos !
m
t + 15 cos 3 !
m
t) (1.78)
where !
c
= (2) 10
7
s
1
, and !
m
= (4) 10
3
s
1
.
(a) What is the peak frequency deviation of this angle-modulated signal? Give your
result in kHz. Be careful, you need to nd the absolute maximum value of a
function with several sub-maxima.
(b) Estimate the bandwidth of this signal using Carsons rule. You will need to
determine the bandwidth of the baseband signal (W) for this calculation. Choose
the smallest bandwidth that contains all of the power in the baseband signal. Give
your result in kHz.
60 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
(c) Suppose that the given s(t) is the result of frequency modulation by a baseband
signal denoted by m(t). Specify the function m(t). (Note: You may assume that
the frequency deviation constant, k
f
, is equal to 1.0.)
6. Consider a signal m(t) with Fourier transform M(!) where
M(!) =
1
1
m(t)e
j!t
dt and m(t) =
1
2
1
1
M(!)e
j!t
dw (1.79)
(a) Consider a particular M(!) which is sketched in Figure 1.38. You may assume
that M(!) is a real function for this problem. Sketch the Fourier transform
2 1 1 2
1
t+c/2
tc/2
s(t
0
)
2
dt
0
,
c
=
2
!
c
62 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
It is usually a good approximation to assume that the envelope of the signal is eec-
tively constant over the duration of a carrier cycle, so that the envelope function may
be pulled out of the integral. The peak envelope power (PEP) is the peak value of
P
env
(t).
The average power is dened as the average of the instantaneous power over a time
interval that is long compared to all time scales of the message signal:
P
avg
= lim
T!1
1
T
T/2
T/2
s(t
0
)
2
dt
0
Consider the message signal given below:
m(t) =
4
n=1,3,5...
sin(n!
m
t)
n
This series is the truncated Fourier series for a square wave.
(a) Write down the series expression for the Hilbert transform, m(t). The Hilbert
transform of m(t) ( m(t)) is obtained by delaying every frequency component of
m(t) by 90
t
1
m(t
0
)dt
0
).
The quadrature demodulator produces two output signals, I(t) and Q(t), by multi-
plying s(t) with in-phase and quadrature local oscillator signals, respectively. Each
multiplier is followed by a low-pass lter that rejects double-frequency terms. A signal
processor takes the outputs of the quadrature demodulator and produces the output
signal s
out
(t) = I(t)
dQ(t)
dt
Q(t)
dI(t)
dt
. In general, the local oscillator signals employed
within the quadrature demodulator have frequency and phase error, i.e. the in-phase
and quadrature local oscillator signals are cos((!
c
+!)t+) and sin((!
c
+!)t+),
respectively.
(a) If ! = 0 and = 0, show how s
out
(t) depends on the message signal m(t); i.e.,
derive an expression for s
out
(t) that is in terms of m(t).
(b) Now allow the frequency and phase errors to be nite. How do frequency and
phase errors aect s
out
(t)?
15. Two methods for generating SSB (and VSB) signals were introduced in this chapter
the lter method and the phasing method. There is a third method, called the
Weaver method which is illustrated in the Figure. The bandwidth of the message
64 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
s(t)
m(t)
cos(!
1
t)
90
cos(!
2
t)
W
90
cos(!
1
t)
sin(!
1
t) sin(!
2
t)
cos(!
2
t)
a b c
a
0
b
0
c
0
H(!)
H(!)
W
!
W W
!
1
1
signal is assumed to be W Hz (2W s
1
). In the frequency domain, the signal at
point a can be written as
S
a
(!) =
1
2
[M(! !
1
) +M(! + !
1
)].
At point b, the signal becomes
S
b
(!) = S
a
(!)H(!) =
1
2
[M(! !
1
) +M(! + !
1
)]H(!).
And at point c
S
c
(!) =
1
2
S
b
(!) [(! !
2
) + (! + !
2
)]
=
1
4
[M(! !
2
!
1
) +M(! !
2
+ !
1
)]H(! !
2
)+
1
4
[M(! + !
2
!
1
) +M(! + !
2
+ !
1
)]H(! + !
2
).
(a) Find the corresponding expressions for the frequency-domain signals at points a,
b, and c.
(b) If the upper (+) sign is chosen for the combiner, determine the frequency-domain
signal at the output of the combiner. Simplify as much as possible - some terms
should cancel. If !
2
!
1
, specify !
1
in terms of W if the output is to be a
SSB signal. Determine whether the modulator yields an upper sideband (USB)
or lower sideband (LSB) signal. Hint: the term H(! !
2
) is 1 for !
2
W
! !
2
+ W and is zero elsewhere.
(c) Now assume that the lower (-) sign is chosen. Determine !
1
such that the mod-
ulator yields a SSB signal. Is the output USB or LSB?
(d) If the message signal has bandwidth W = 5.0 MHz, and you want to generate
a USB signal with carrier frequency at 1250 MHz and upper cuto frequency at
1255 MHz, specify the correct sign (+ or -) to use in the combiner and the values
of f
1
= !
1
/(2) and f
2
= !
2
/(2). Give your results in MHz.
1.10. HOMEWORK PROBLEMS 65
(e) For the same message signal as in part (d), specify the sign to use in the combiner
and f
1
and f
2
if it is desired to generate LSB with the carrier frequency at 1250
MHz and the lower cuto frequency at 1245 MHz.
16. DSB-SC is used to transmit a message signal, m(t) =
a
n
p(t nT), where T = 1 s
and p(t) is a raised-cosine pulse with = 0.5. Calculate the bandwidth occupied by
the transmitted signal.
17. Consider the following signal:
s(t) = 10 sin[2(10
9
)t + 2 cos(310
5
t)].
(a) Find the instantaneous frequency (in Hz) as a function of time for this signal.
(b) What is the peak instantaneous frequency deviation of s(t)? Express your result
in kHz.
(c) Estimate the bandwidth of s(t). Express your result in kHz.
(d) The fractional bandwidth of s(t) is dened to be the signal bandwidth divided
by the carrier frequency. What is the fractional bandwidth of s(t)?
(e) Suppose that s(t) is the result of frequency modulation by a message signal
denoted by m(t). Specify the function m(t). You may assume that the frequency
deviation constant, k
f
, is equal to 1.0.
(f) In general, the Fourier spectrum of s(t) will contain delta functions with compo-
nents at the carrier frequency, f
c
, and at frequencies separated from the carrier
frequency by multiples of some frequency f
m
. What is f
m
? Give your answer in
kHz.
(g) Refer to Figure 1.27 to answer this question. In the Fourier amplitude spectrum
of s(t), (|S(f)|), denote the strength of the delta function at f
c
by a
0
, and the
strength of the delta functions at f
c
nf
m
by a
n
. Sort the list {a
0
, a
1
, a
2
, a
3
, a
4
}
in order of decreasing amplitude.
66 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
Chapter 2
Receivers
2.1 Introduction and Historical Progression
The history of radio communication begins at the end of the 19th century, when Italian
inventor Guglielmo Marconi developed the rst practical radio communication system using
spark-gap radio transmitters and a nonlinear circuit element, called a coherer, as a detec-
tor at the receiving end of the link. His most famous accomplishment, was demonstration
of transatlantic wireless communications, in 1902, when the Morse-code letter S (dot dot
dot) was transmitted from England, and received in Newfoundland. Wireless telegraphy
transmission using spark-gap transmitters used the broadband damped oscillations gener-
ated when the DC current in an LC circuit is suddenly interrupted. Spark-gap telegraphy
was the primary wireless communications technology in use throughout World War I. Mean-
while, beginning in 1906, the technology required for the next phase of radio communication
development was put into place when Lee De Forest patented the audion, a device created
by inserting a third electrode into the vacuum diode, creating a triode vacuum tube. This
3-terminal device could be used to amplify signals, and could be made to oscillate, thus
serving as a very stable source of continuous-wave (CW) high-frequency signals. By 1915,
the American Telegraph and Telephone Company (AT&T) had developed a high-power
amplitude-modulated (AM) radio transmitter based on the triode, and in that year, for the
rst time in history, voice transmissions from Arlingon, VA were heard in Paris, France,
and in Honolulu, HI. In 1920, the rst commercial AM radio stations began operating. For
much more information on the fascinating early history of radio communication, the book
The Science of Radio, by Paul J. Nahin, is highly recommended.
Our purpose in this chapter is, rst, to give a brief outline of the historical development
of radio receivers, beginning with the simple, passive, tuned detector, and second, to present
a fairly detailed discussion of the modern superheterodyne receiver.
2.1.1 Tuned Detector/Demodulator
The purpose of a receiver is to select, amplify, and demodulate a modulated-carrier signal,
while ignoring the many other signals which will be picked up by the antenna, but are
not of interest. The earliest receivers employed a single tuned circuit as a lter in front of
an envelope detector implemented with a crystal detector, a point-contact type of diode.
Voice transmissions were amplitude modulated with full-carrier, so an envelope detector was
67
68 CHAPTER 2. RECEIVERS
sucient to demodulate the signal. The block diagram of such a tuned-detector receiver is
shown in Figure 2.1. The tuned-detector is a passive circuit, and the audio power available
s
o
(t) = A(1 + m(t))+noise
Demodulator
s
i
(t) = A(1 + m(t)) cos(!
c
t + )+noise
Figure 2.1: Top: block diagram of a tuned-detector receiver. Bottom: typical implementa-
tion. This type of receiver is also known as a crystal set, because early implementations
of the diode junction were implemented with the point-contact junction between a wire and
crystal such as galena (lead sulde) or carborundum (silicon carbide).
from the detector output must be provided by the energy collected by the antenna. Large
antennas are necessary to collect enough signal, even from relatively close transmitters, to
provide an output that is audible even with sensitive headphones. Selection of the desired
signal, and rejection of unwanted signals, is the responsibility of the resonant circuit, which
acts as a lter, and which must be tuned to pass the desired carrier frequency of interest.
It is easy to improve the ability of a tuned-detector to receive weak signals by adding an
amplier to the system, but this does not solve the other fundamental problem with this
receiver, which is the fundamental lower limit to the bandwidth that can be achieved with
simple lters based on inductor/capacitor resonators. This limitation will be discussed in
more detail in section 2.2. For the time being, it suces to say that an LC lter, whose center
frequency must be adjustable, is limited to a bandwidth no smaller than approximately 1% of
its center frequency. While tunable LC lters with smaller bandwidths can be designed, the
attenuation (loss) of such lters soars as the design bandwidth is decreased below about 1%.
In the early days of radio, carrier frequencies were low, so fractional bandwidths of signals
were relatively high, and the 1% lower limit on bandwidth was not a serious problem. As
shorter wavelengths came into use, the fractional bandwidth occupied by signals became
smaller than 1%, and some means of obtaining narrower bandwidth was needed.
2.1.2 Tuned Radio Frequency (TRF) Receiver
One very popular receiver architecture that was employed in the 1920s for AM broadcast
receivers was sort of a brute-force approach to obtaining high gain and narrow bandwidth.
The tuned-radio-frequency (TRF) receiver utilized a cascade of several tuned ampliers,
2.1. INTRODUCTION AND HISTORICAL PROGRESSION 69
each having a bandpass response, as shown in Figure 2.2, where each tuned amplier is
represented by separate lter and amplier blocks. If the lter/amplier stages are identical,
and if the ampliers provide an eective buer between the lters, then the overall transfer
function of the cascade is equal to the transfer function of a single lter/amplier stage
raised to the nth power, where n is the number of stages. This architecture provides high
gain and narrower bandwidth than can be obtained with a single lter. The TRF receivers
were fairly dicult to tune properly, as each of the tuned circuits had to be tuned to the
carrier frequency of interest. If only one of the lters was mis-tuned, the overall gain of
the receiver would drop to a very low value. Another problem with the TRF receiver
was the tendency for the system to oscillate. This problem is a manifestation of the too
much gain in one box syndrome, which occurs whenever reverse isolation (the attenuation
between the output and the input) of an amplier chain does not exceed the forward gain
by a comfortable margin. Leakage from the output back to the input provides a feedback
path, which can lead to oscillation. In early receivers, the internal feedback within the triode
vacuum tubes used as ampliers provided enough reverse coupling to cause oscillation unless
the forward gain was kept small. A technique called neutralization was developed to cancel
the internal feedback of the active devices, and the TRF receiver employing neutralization
became known as the neutrodyne receiver. The triode-based neutrodyne was followed by a
second generation TRF receiver based on the tetrode vacuum tube, which had almost no
internal feedback and did not require neutralization. The tetrode-based TRF receiver was
available through the early 1930s.
Demodulator
f
c f
c
f
c
Figure 2.2: Tuned-radio-frequency (TRF) receiver architecture.
2.1.3 Regenerative Receiver
The TRF receiver became a practical commercial device only when the cost, and availability,
of triode vacuum tubes made these devices available to the general public. Even before the
dawn of AM broadcasting, and the subsequent popularity of the TRF receiver, a particularly
ingenious concept was developed by Edwin Howard Armstrong, in 1912, which allowed him
to develop a receiver with high gain, and narrow bandwidth, using only a single triode.
Armstrong, an undergraduate at Columbia University at the time, used positive feedback
to increase the gain, and narrow the bandwidth of a single-stage tuned amplier. A block
diagram of the system is shown in Figure 2.3. The transfer function of the system between
the antenna terminals and the demodulator input is
s
o
(!)
s
i
(!)
=
GF(!)
1 AGF(!)
. (2.1)
Without the positive feedback, the transfer function would be equal to the numerator of
equation 2.1. The gain of the lter/amplier would be GF(!
o
), and the fractional bandwidth
70 CHAPTER 2. RECEIVERS
Demodulator
f
c
G
A
F(!)
s
i
(!)
s
o
(!)
s
o
(!)
s
i
(!)
=
GF(!)
1AGF(!)
Figure 2.3: A regenerative receiver employing a regenerative amplier in front of a demod-
ulator.
of the system would equal to that of the lter. With feedback, the gain at the center
frequency is GF(!
o
)/(1 AGF(!
o
)), which can be made arbitrarily large if AGF(!
o
) is
allowed to approach 1 (from below!). The positive feedback that is built into the regenerative
amplier is precisely what is necessary to build an oscillator. Note that if the quantity
AGF(!
o
) is allowed to become equal to one, the transfer function becomes innite, which
may be interpreted as a condition where nite output signal occurs with zero input, in
other words, the system oscillates and becomes a radio-frequency signal source, rather than
an amplier. Armstrongs regenerative receiver operates the feedback loop just below the
threshold of oscillation, where the denominator of equation 2.1 is very small. Hence, the
gain of the system is very large. It turns out that if the lter is a simple LC resonator, then
the gain-bandwidth product of the system is independent of the value of AGF(!
o
). Hence,
when the gain is very large, the bandwidth of the system can be very small. This provides a
means to obtain extremely high gain and extremely small fractional bandwidths, using only
a single active device.
The drawback to the regenerative receiver is the fact that the desired operating point is
where AGF(!
o
) is close to one. If something happens to cause AGF(!
o
) to become equal
to, or greater than, one, then the system becomes unstable, and oscillates. When the system
is oscillating, it no longer acts as an extremely high gain and narrow bandwidth amplier
for small signals delivered by the antenna. In the oscillating mode, the regenerative receiver
essentially becomes a self-oscillating mixer, which does not provide any narrow bandwidth
ltering action. The diculty associated with adjusting the lter and feedback to achieve
high gain and narrow bandwidth made this receiver relatively dicult to use. It is also
inherently very sensitive to changes in component values and circuit layout, so that it is
rarely used in modern systems. Nevertheless, in the 1920s, there was an intense competition
between Westinghouse, who had acquired the rights to Armstrongs regenerative patent, and
a group of radio manufacturers that owned the rights to the neutrodyne patent.
An interesting, and even more exotic, variant of the regenerative receiver concept can be
found in the superregenerative receiver. Keeping in mind that the regenerative receiver is
most sensitive (in the sense that it has the highest gain) when it is operated at the threshold
between stability and instability (oscillation), the superregenerative receiver is designed so
that the quantity AGF(!
o
) is actually larger than 1. Thus, oscillation starts and builds
2.2. CHARACTERISTICS OF PRACTICAL FILTERS 71
up immediately after power is applied to the system. Superregenerative receivers have a
built-in circuit that senses when oscillation starts (this is easily detected by changes in the
bias point of the active device) and, when oscillation is detected, the gain of the system is
momentarily reduced, so that the oscillation is quenched. After the oscillation is quenched,
the gain is restored, and oscillation builds up again. This cycle is repeated with a period
that is short compared to the shortest time scale associated with the modulation of the
signal that is being received. Thus, the system operates at extremely high gain for short
periods of time (while oscillation is building up) during each cycle, eectively providing
samples of the desired signal at a sampling rate that is adjusted to be high enough to allow
the detector to recover the modulation. Superregenerative receivers can still be found today
in devices such as garage door openers and radio controlled toys, where the primary design
requirement is absolute minimum cost.
2.1.4 Genesis of the Superheterodyne Receiver
Although the regenerative and TRF receivers dominated the consumer radio market into the
early 1930s, their replacement, and the architecture that is dominant today, was patented
by Edwin Armstrong in 1917, and is called the superheterodyne receiver. After his stint
at Columbia, Armstrong became a member of the U.S. Army Signal Corps during World
War I. Involved with eorts to nd a way to detect enemy airplanes from a distance, he
knew that it might be possible to detect the electromagnetic emissions from the spark plugs
in the engine. The problem was that the emissions were strongest at the (then) unusually
high frequencies above a few MHz. Triodes of the day had very little gain at such high
frequencies, so Armstrong hit on the idea of employing the heterodyne principle to shift the
high-frequency signals to a lower frequency, where they could be more eciently amplied
and ltered. Once he decided to incorporate the heterodyne concept into his receiver, it
became possible to heterodyne any signal of interest, regardless of its frequency, to a xed
intermediate frequency (IF). Selective ltering, and high-gain amplication could be done
at the xed IF.
In contrast to the TRF and regenerative receivers, the new heterodyne-based receiver
did not require a tunable, high-gain, narrow-bandwidth, lter/amplier. Instead, the high
gain, narrow bandwidth lter/amplier only had to operate at the xed IF which was a
relatively low frequency. This made it much easier to optimize the IF lters and ampliers
for high gain and narrow bandwidth. A further advantage was obtained because converting
the desired radio frequency (RF) signal to a lower IF does not change the absolute band-
width of the signal. For example, a signal that occupies a bandwidth of 10 kHz at a carrier
frequency of 10 MHz occupies a fractional bandwidth of 0.01/10=0.001, or 0.1%. When
that same signal is heterodyned to an IF of, say 400 kHz, it retains its 10 kHz bandwidth,
so its fractional bandwidth becomes 10/400=0.025, or 2.5%, a comfortably large fractional
bandwidth. Receivers based on Armstrongs superheterodyne concept were originally pro-
duced by Radio Corporation of America (RCA) who, as a result, dominated the commercial
radio market by 1930. Before proceeding with a detailed overview of the superheterodyne
receiver, we shall rst discuss the properties of practical lters.
2.2 Characteristics of Practical Filters
Losses in the components used to implement lter networks set a lower limit on the band-
width that can be achieved in a bandpass lter. For a bandpass lter with center frequency
72 CHAPTER 2. RECEIVERS
f
o
, a common measure of the bandwidth of the lter is the frequency spacing between the
3dB points on the lter transfer function, denoted by f
3dB
in Figure 2.4. The dimen-
f
o
F(!)
F
o
/
p
2
F
o
f
3dB
Figure 2.4: Denition of -3dB bandwidth.
sionless fractional bandwidth, f
3dB
/f
o
, of a lter implemented with a single LC resonant
circuit is 1/Q , where Q is the so-called loaded Q of the resonant circuit. In many cases,
the resonant circuit Q is limited by the Q of the inductor. Capacitor losses are usually
small compared to the losses in inductors. For miniature inductors, Qs may be limited to
values in the range 10-100, whereas inductors implemented using spirals of metallization
in integrated circuits typically have very low Q, often <10. The fractional bandwidth of
a lter implemented using an LC resonator with Q of 100 is 0.01, which means that the
minimum bandwidth of an LC lter is about 1% of the center frequency. If physically large
inductors can be tolerated, it is possible to achieve Qs as large as several hundred, even
approaching 1000, and fractional bandwidths signicantly smaller than 0.01. Such high-
Q inductors are rarely practical for use in modern systems where small size, light weight,
and low cost are primary considerations. Generally speaking, however, LC lters with frac-
tional bandwidths smaller than around 0.01 tend to become impractical, even when they
can be realized, as mechanical and thermal stability of the components becomes an issue at
such small bandwidths, and such lters are easily de-tuned by changes in temperature, or
mechanical stresses.
Filters implemented using a single resonant circuit (resonator) do not have a very well-
dened passband. The transfer function is not constant within the -3dB bandwidth of the
lter, and it decays slowly outside of the passband. Thus, the bandwidth of the lter at
the -30dB points is much wider than the bandwidth between the -3dB points. A measure
of how closely a bandpass lters passband approaches that of an ideal, rectangular lter, is
the so-called shape factor of the lter response. Shape factor is dened as the ratio of the
bandwidth at some large attenuation (say 30dB) and the bandwidth at the 3dB attenuation
points. As a lters transfer function approaches the ideal, rectangular shape, its shape
factor approaches 1.0. A single-resonator LC lter is far from this ideal - for example, such
a lter with a Q of 50 has a fractional bandwidth of 1/50, or 0.02, and a shape factor (using
the -30 dB and -3 dB bandwidths) of 31.6. Thus, the -30 dB bandwidth is more than 30
times as wide as the -3 dB bandwidth. Obviously, if a single resonator lter response is
adjusted so that the signal of interest lls the -3dB bandwidth, an adjacent channel will not
2.2. CHARACTERISTICS OF PRACTICAL FILTERS 73
be attenuated by very much. Thus, the single resonator lter is not appropriate for providing
so-called adjacent-channel rejection. The shape of a lters transfer function can be made
to approach the idealized rectangular shape by using more (sometimes, many more) than a
single resonator. For example, using only two resonators, a lter with fractional bandwidth
of 0.02 and shape factor of 10.1 is easily obtained this is more than a 3:1 improvement
over the shape factor of a single-resonator lter. A comparison between the attenuation of
a single-resonator lter and a 2-resonator lter, each having fractional bandwidth of 0.02,
is shown in Figure 2.5.
Normalized Frequency (f/f
c
)
0.97 0.98 0.99 1.01 1.02 1.03
2
4
6
8
10
Attenuation in dB
2 critically coupled
resonators, Q=62
single resonator, Q=50
Figure 2.5: Attenuation vs normalized frequency for single and double resonator lters with
the same fractional bandwidth. The single resonator lter has a shape factor of approxi-
mately 32, whereas the 2-resonator lter has a shape factor of 10.
2.2.1 Transmission-line and cavity resonator lters
For microwave applications it becomes feasible to utilize resonators based on distributed
elements such as transmission lines, or waveguide cavities. Generally speaking, in their
optimum implementations these resonators allow higher Qs than lumped LC resonators,
and provide smaller minimum fractional bandwidths than provided by lumped LC lters.
They are not easily tunable, however, restricting their application to xed center frequencies
and bandwidths. Filters based on distributed elements are most suitable for use as RF front-
end lters in microwave receivers.
2.2.2 Filters based on piezoelectric devices - Quartz-Crystal Filter,
Ceramic Filter
When very small fractional bandwidths and shape factors approaching 1.0 are required,
it is necessary to turn to electro-mechanical resonators implemented using piezoelectric
materials such as quartz crystals or ceramics, or electro-acoustic resonators implemented
using surface-acoustic-wave (SAW) devices. These devices have Q-factors several orders of
magnitude larger than those available from LC resonators, and they have much smaller
thermal coecients, allowing extremely narrowband lters to be realized. For example, a
quartz crystal may have a Q of 100,000, allowing fractional bandwidths as small as 10
5
to be
implemented. At a center frequency of 10 MHz, a fractional bandwidth of 10
5
corresponds
to -3 dB bandwidth of 100 Hz. Unfortunately, the electro-mechanical resonators are useful
74 CHAPTER 2. RECEIVERS
only at relatively low center frequencies (typically, < 30 MHz). Quartz-crystal and ceramic
lters are xed-frequency devices, and cannot be tuned over any appreciable range.
2.2.3 SAW lters
At frequencies in the range 30 MHz2.5 GHz, lters based on surface-acoustic wave (SAW)
devices provide a convenient means to realize lters with fractional bandwidths as small
as 0.1% (f/f
o
= 0.001) with excellent shape factors in a small package. These lters
eectively t many resonators within a small volume by converting an electrical signal into
an acoustic signal, where the wavelength is 5 orders of magnitude smaller because of the
slow speed of sound relative to the speed of electromagnetic wave propagation. By ltering
the acoustic signal, it is possible to build lters which, eectively, contain 10s or hundreds
of resonators within an extremely small package. This makes it possible to realize lters
with excellent shape factors. SAW lters are fabricated for a xed center frequency, and are
not tunable.
2.2.4 Filter limitations dictate carrier-frequency conversion
Consider the fractional bandwidth required to make a single PCS-band CDMA (code-
division-multiple-access) cellphone channel ll the -3dB bandwidth of a lter. Why ll
the bandwidth of the lter with the signal of interest? Generally speaking, when a signal
is accompanied by noise and interference, maximum signal-to-noise ratio will occur at the
output of a lter when the lter shape is matched to the spectrum of the signal. The band-
width of CDMA channel is 1.25 MHz, and the center frequency is approximately 1900 MHz.
A fractional bandwidth of 1.25/1900 .00066 is required, with shape factor as close to 1 as
possible, to select one channel while providing rejection of adjacent channels.
It is not possible to implement anything close to this small fractional bandwidth using
practical lters based on any type of conventional electronic resonator (LC, transmission line,
cavity resonators, etc). Quartz crystal lters can achieve the required fractional bandwidth,
but not at a center frequency above 100 MHz or so. The solution is to move the carrier
frequency of the desired channel to a lower frequency, without changing the bandwidth of
the signal or distorting the modulation in any way. If the carrier frequency is lowered to, say
45 MHz, then the fractional bandwidth of the signal becomes 1.25/45 0.028. This center
frequency and fractional bandwidth are well within the range where SAW lters provide
excellent performance. The concept of using frequency conversion to downconvert high
frequency signal to a lower frequency where ltering and amplication is easier to achieve
is embodied in the superheterodyne receiver.
2.3 The Superheterodyne Receiver
A block diagram of a single-conversion receiver is shown in Figure 2.6. The superheterodyne
receiver operates by converting the input frequency of interest (f
c
) to a xed intermediate
frequency (IF). This frequency conversion is performed by the mixer and local oscillator
(LO). The intermediate frequency may be higher or lower than the carrier frequency of
interest, but it is important to note that the IF is xed (constant). Conversion to a xed
frequency takes advantage of the fact that it is much easier to realize narrow-band lters
and stable, high-gain ampliers if the frequency of operation doesnt change.
2.3. THE SUPERHETERODYNE RECEIVER 75
Demodulator
f
c
f
IF
Intermediate Frequency (IF)
section
Radio Frequency (RF)
section
IF amplier IF lter
Preselector
lter RF amplier
Local Oscillator (LO)
f
LO
Mixer
Figure 2.6: Superheterodyne receiver
A perfect multiplier can be used to model the operation of the mixer for preliminary
discussions of the basic operation of the receiver, in which case the output of the mixer will
consist of two signals with carrier frequencies equal to the sum and dierence of the input
and LO frequencies. Consider a modulated input signal with carrier frequency f
c
which
may be both angle- and amplitude-modulated in general:
s(t) = A(t) cos[!
c
t + (t)] (2.2)
The local oscillator signal will be an unmodulated carrier with frequency f
LO
, i.e.,
cos(!
LO
t). The output signal from the mixer will be
s
out
(t) = A(t) cos[!
c
t + (t)] cos(!
LO
t) (2.3)
=
1
2
{A(t) cos [(!
c
!
LO
)t + (t)] +A(t) cos [(!
c
+ !
LO
)t + (t)]}
Note that the output from the mixer/LO consists of two signals with carrier frequencies
f
c
f
LO
and f
c
+f
LO
. Note also that the amplitude and angle modulation that was present
on the input signal has been transferred to the two output signals without any distortion.
The IF lter picks out one of the two signals and rejects the other. If a range of
input frequencies is to be covered, then the LO will be tunable. Since the IF frequency
is xed, the LO frequency is adjusted, in practice, in order to make f
IF
= f
c
+ f
LO
or
f
IF
= |f
c
f
LO
|. The choice of whether the sum or dierence frequency is picked out
by the IF lter is determined in the design stage. If f
IF
< f
c
the conguration is called
down-conversion, since the carrier frequency has been converted down to f
IF
. If f
IF
> f
c
,
the receiver is said to employ up-conversion.
2.3.1 Image Frequencies
Just as there are two output signals from the mixer for each input frequency, there are two
input frequencies to the mixer that will give an output at the IF frequency. In practice
only one of these will be the desired frequency, and the other, undesired, frequency is called
76 CHAPTER 2. RECEIVERS
the image frequency. If an undesired signal happens to have a carrier frequency that is the
same as the image frequency, then that signal would be mixed into the IF lters passband
along with the desired signal. In the superhet (see Figure 2.6) the primary function of the
preselector lter is to pass the desired signal and to reject signals at the image frequency.
In a well designed receiver the image frequency is separated from the carrier frequency by
a relatively large interval. Thus the preselector lter does not need to have a very narrow
bandwidth. This is desirable, since the preselector lter often needs to be tunable. It is
dicult to build a narrowband lter that has a tunable center frequency.
The preselector lter does not provide adjacent channel rejection, i.e., it does not reject
signals that have carrier frequencies immediately adjacent to that of the desired signal.
This is the function of the intermediate frequency (IF) lter which will determine the signal
bandwidth of the receiver and, therefore, provide the rejection of adjacent channels. The
bandwidth of the IF lter will normally be chosen to be just wide enough to accept the
spectrum of the desired signal. Of course, interfering signals may still occur within the
spectrum of the desired signal. In this case a notch lter can sometimes be used in the
IF path to reject a very narrow slice of the signal spectrum which contains the interfering
signal without signicantly aecting the desired signal. The adjacent channel selectivity of
a receiver is a function of the steepness at which the IF lter response rolls o outside of
the passband.
2.3.2 Operation of the Mixer/LO Stage - Useful Relationships
For given IF and input frequencies, there are two possible LO frequencies that will cause
the input frequency to be converted to the IF frequency:
f
LO
= f
IF
+f
C
or f
LO
= |f
IF
f
C
| (2.4)
For given IF and LO frequencies, there are two possible carrier frequencies that will give
an output at the IF frequency:
f
c1
= f
LO
+f
IF
and f
c2
= |f
LO
f
IF
| (2.5)
One of these will be the desired carrier frequency, f
c
, and the other will be an undesired
image frequency, f
IM
. The choice that was made for the LO frequency will determine which
of these equations gives the image and/or desired frequency. The primary purpose of the
preselector is to reject the undesired image frequency.
There are four possible generic congurations for a single-conversion superheterodyne
receiver. Dene f
cmin
and f
cmax
to be the lower and upper limits, respectively, of the
input frequency range that the receiver is to cover. In practice the IF frequency will either
be smaller than f
cmin
or larger than f
cmax
. These two cases can be further subdivided,
because there are two possible choices for the LO tuning range for each choice of IF. The
image frequency and the separation between the desired and image frequencies (|f
IM
f
c
|)
are summarized in Table 2.1 for each of the four cases.
2.3.3 Example - AM Broadcast Receiver.
A typical AM broadcast receiver covers the frequency range 540 - 1700 kHz. AM broadcast
stations in the U.S. are assigned frequencies that are integer multiples of 10 kHz, therefore
the adjacent channel separation is 10 kHz. The IF frequency is very often chosen to be 455
2.3. THE SUPERHETERODYNE RECEIVER 77
Table 2.1: Generic Congurations for a Single-conversion Superheterodyne Receiver
(1) Up - conversion: f
IF
> f
cmax
> f
cmin
(1a) f
LO
= f
IF
+f
c
f
IM
= f
c
+ 2f
IF
|f
IM
f
c
| = 2f
IF
(1b) f
LO
= f
IF
f
c
f
IM
= 2f
IF
f
c
|f
IM
f
c
| = 2f
LO
(2) Down - conversion: f
IF
< f
cmin
< f
cmax
(2a) f
LO
= f
IF
+f
c
f
IM
= f
c
+ 2f
IF
|f
IM
f
c
| = 2f
IF
(2b) f
LO
= f
c
f
IF
f
IM
= |f
c
2f
IF
| |f
IM
f
c
| = 2f
IF
if f
c
> 2f
IF
|f
IM
f
c
| = 2f
LO
if f
c
< 2f
IF
kHz. This is an example of down-conversion. From (2a) and (2b), there are two choices
for the local oscillator tuning range:
1. 995 kHz to 2155 kHz
2. 85 kHz to 1245 kHz
The tuning ratio (R = f
LOmax
/f
LOmin
) is an important factor in determining the cost
and stability of an oscillator. It is easier to build a stable oscillator if the tuning ratio is
small. Note that the tuning ratio for choices (1) and (2) are 2.16 and 14.6, respectively.
For this reason the rst choice given above is almost always used. In fact, in most (but not
all) cases the higher of the two possible LO frequencies will be used (i.e., (1a) or (2a) from
Table 2.1), because this leads to the smaller tuning ratio. In this particular example the
second choice has another disadvantage, since it would require the local oscillator to tune
through the IF frequency, i.e., to receive a signal at 910 kHz; the second choice requires the
local oscillator to be tuned to 455 kHz. It is very likely that some of the local oscillator
signal would leak into the IF stage of the receiver. The result would be interference with
the desired signal, as well as possible overload of the IF and succeeding stages.
Assuming the LO tunes from 995 to 2155 kHz, then the image frequency will always be
(from (2a)) given by f
IM
= f
c
+2f
IF
. For example, if we wish to tune the receiver to WILL
at 580 kHz, the preselector lter would be tuned to 580 kHz and the local oscillator would
be tuned to 1035 kHz. The bandwidth of the preselector needs to be narrow enough to
reject any incoming signals at the image frequency f
IM
= 1490 kHz. The IF lter needs
to be 10 kHz wide in order to provide adjacent channel selectivity.
As the radio is tuned across the band, it is necessary to tune both the LO and the
preselector simultaneously. This is why the tuning knob on mechanically tuned AM radios
is connected to a two-section, or ganged, variable capacitor. One section tunes the LO,
and the other tunes the preselector lter. One or more adjustments is usually provided to
insure that the two sections track each other.
Note carefully that for carrier frequencies less than 1245 kHz, the local oscillator fre-
quency will be less than 1700 kHz and therefore inside the AM broadcast band. It is
sometimes possible to hear this LO signal on another nearby AM radio. This phenomenon
is undesirable and is caused by the local oscillator signal being fed back through the mixer
and out the antenna, or by direct radiation from the LO circuitry. Commercial and military
receivers are carefully designed to minimize such radiation. This is done by using balanced
mixers that have very low direct output-to-input coupling coecients and by carefully shield-
ing the LO circuitry to avoid direct radiation. If the LO frequency is well removed from the
carrier frequency, a notch lter can be put before the mixer to reduce the LO signal that is
78 CHAPTER 2. RECEIVERS
coupled to the antenna.
The conguration of a standard FM broadcast receiver is described in the next example.
2.3.4 Example - FM Broadcast Receiver
The FM broadcast band covers 88 - 108 MHz. The channels are separated by 200 kHz and
are assigned to odd multiples of 100 kHz. In almost all cases the IF frequency is chosen to
be 10.7 MHz. As in the previous example high-side LO is often used, i.e., the LO tunes
from 98.7 to 118.7 MHz.
Notice that the IF used for FM broadcast receivers is substantially higher than that
used for AM receivers. This choice is motivated by the fact that the image frequency is
separated from the desired carrier frequency by 2f
IF
. Generally speaking, in order to make
the preselector lter relatively easy to build and tune, the IF frequency must be raised as
the carrier frequency increases.
2.3.5 Up-conversion versus Down-conversion
The AM and FM broadcast receivers considered above both use down-conversion and the
higher of the two possible LO frequencies. The image frequency was 2f
IF
above the desired
carrier frequency. In the AM receiver case the width of the carrier frequency band of interest
is larger than 2f
IF
. This means that the image frequency can fall within the band of interest,
thus necessitating the use of a tunable preselector lter that has a bandpass characteristic.
In receivers designed to cover very wide frequency bands the use of a tunable preselector
becomes impractical, because it is dicult to build lters whose center frequency must be
tuned over a wide frequency range. In this case it is advantageous to use up-conversion,
because with up-conversion the image frequency will always be larger than f
cmax
. Thus,
the image frequency will always fall outside (above) the band of interest, and a preselector
lter with a lowpass characteristic is sucient. A more practical choice, however, would be
a x-tuned bandpass lter that passes the entire carrier frequency band of interest. The
bandpass lter is usually the best option, since it minimizes the possibility of interference
and receiver overload from strong signals outside of the frequency range of interest. The
up-conversion scheme can be a signicant advantage, especially in a receiver that is designed
to cover a wide frequency range, since only the LO needs to be tuned.
2.3.6 Single- versus Double-conversion
There is a lower limit to the fractional bandwidth (f/f
o
) that can be realized with practi-
cal lters. This sets an upper limit on the intermediate frequency for a given signal spectrum
bandwidth. However, because the separation of the image frequency from the desired fre-
quency is usually twice the intermediate frequency, a large intermediate frequency is desired.
A large separation is desirable because it makes rejection of the image easier. In some cases
these conicting considerations make it impossible to select an intermediate frequency that
is low enough to achieve adequate adjacent channel selectivity but large enough to make it
possible to reject the image frequency with a simple preselector lter. Then it becomes nec-
essary to use a double-conversion scheme (Figure 2.7) where the carrier frequency of interest
is rst converted to a relatively high IF (allowing rejection of the image frequency with a
simple preselector lter) and then converted again to a lower IF, in order to obtain the
2.3. THE SUPERHETERODYNE RECEIVER 79
f
c
f
IF1
IF amplier First IF lter
Preselector
lter RF amplier
f
LO1
Mixer
f
IF2
IF amplier 2nd IF lter
Mixer
f
LO2
to
demod
Figure 2.7: Double-conversion superheterodyne receiver. The rst conversion (f
c
! f
IF1
)
can be either up- or down-conversion. The second conversion (f
IF1
! f
IF2
) is always a
downconversion.
required bandwidth and adjacent channel selectivity. In practice, the rst conversion can be
either an up- or down-conversion, but the second conversion is always a down-conversion.
In certain cases there are advantages to using more than two conversions, i.e., triple-
conversion, although problems with spurious responses multiply rapidly as the number of
frequency conversions increases. The image frequency causes one type of spurious response
that is commonly observed in inexpensive AM and FM broadcast receivers where a certain
radio station is received when the receiver is tuned to a frequency other than the true carrier
frequency of the signal. There are several other origins for spurious responses involving both
external signals and signals generated within the receiver. These will be discussed in some
detail in Chapter 12. An important step in the design procedure is to identify and evaluate
the impact of spurious responses on the overall operation of the receiver.
In a multiple-conversion receiver the bandwidth of the last IF stage sets the overall
bandwidth of the receiver. In a double-conversion receiver the bandwidth of the rst IF
lter would be chosen to reject secondary image frequencies within the rst IF bandwidth.
A secondary image would be an undesired frequency within the passband of the rst IF lter
that could be mixed into the second IF lters passband. The second IF lters bandwidth
would usually be just wide enough to pass the entire spectrum of the desired signal.
2.3.7 The 1/2-IF response
Nonlinearity in the RF, LO, and mixer sections of the receiver causes mixing products
involving harmonics of the RF and LO signals to be present at the mixer output. This
causes spurious responses, i.e. the receiver responds to input frequencies other than the
desired frequency. One example of this type of spurious response is the so-called 1/2-IF
response. The 1/2-IF response results from the second harmonic of an input signal mixing
with the second harmonic of the LO signal to produce an output at f
IF
. The mechanism is
illustrated here for the case of high-LO and down-conversion. Suppose that the receiver is
tuned to receive a desired signal with carrier frequency f
C
. The LO will be tuned to
f
LO
= f
C
+f
IF
(2.6)
Suppose a signal with frequency f
S
> f
C
is also present at the receiver input. If the
second harmonic of this signal mixes with the second harmonic of the LO, the mixer output
will have a component at |2f
S
2f
LO
|. This signal will interfere with the desired signal if
80 CHAPTER 2. RECEIVERS
|2f
S
2f
LO
| = f
IF
, or if
f
S
= f
C
+f
IF
f
IF
/2.
Thus a spurious response occurs when f
S
= f
C
+(3/2)f
IF
or f
S
= f
C
+f
IF
/2. Notice that
both of these responses are closer to the desired signal than is the image frequency. The
closest spurious response is separated from the desired signal by 1/2 of the IF frequency.
The 1/2-IF response plays a role in the choice of f
IF
. Suppose that the signals of
interest are located within a carrier frequency range denoted by R. It is desirable to use
a x-tuned preselector that passes the entire carrier frequency band of interest. In order
for the preselector to reject potential spurious responses it is necessary for all signicant
spurious responses to fall outside of the carrier frequency range of interest. If only the
image response is considered, then this constraint is satised if f
IF
is chosen such that
2f
IF
> R, or f
IF
> R/2. When the 1/2-IF response is considered, then spurious responses
will fall outside of the desired range if f
IF
/2 > R, or f
IF
> 2R. Notice that when the
1/2-IF response is considered the minimum value of f
IF
is 4 times as large as when only
the image response is considered.
2.4 Zero-IF receiver
In many applications, it is highly desirable to be able to implement an entire receiver in an
integrated circuit, without the need for bulky and expensive IF lters, which require that
signals be routed o-chip to the lter. For such applications the zero-IF (ZIF) (also called
direct-conversion or homodyne) approach has gained popularity. There are various ways
of explaining the evolution of this concept, e.g. it is sometimes described as the limit in
which the intermediate frequency of a superhet approaches zero. Perhaps it is more accurate
to say that a ZIF receiver amounts to using a quadrature demodulator as the receiver. A
ZIF receiver is shown in Figure 2.8. The dierence between this and a basic quadrature
cos !
c
t
Q(t)
90
cos !
c
t
sin !
c
t
I(t)
f
c
Figure 2.8: Direct-conversion receiver. Depending on the application, the local carrier
oscillator may need to be derived from a carrier recovery circuit.
demodulator is the addition of an RF lter and low-noise amplier (LNA) in front of the
demodulator, and addition of baseband ampliers after the lowpass lters in each arm of
the demodulator. In a ZIF receiver, the lters responsible for selecting a desired channel,
2.5. SOFTWARE DEFINED RADIO 81
and rejecting undesired channels, are the lowpass lters at the output of each multiplier
(mixer). This is attractive, since active lowpass lters can be applied here. A very exible
receiver architecture results if the I and Q outputs of the ZIF demodulator are sampled with
an analog-to-digital converter so that digital signal processing techniques can be applied to
implement very good channel selection lters and to perform the signal processing required
for demodulation. This relaxes the requirement on the analog LPFs to the simpler task of
providing the anti-aliasing function.
The ZIF concept looks deceptively simple. In practice, a number of issues combine to
make it a challenge to obtain good performance using this architecture. In a ZIF receiver,
most of the gain will be provided by the baseband ampliers after the LPFs. Hence, desired
signals at the output of the multipliers are generally much smaller than they would be if
the quadrature demodulator was used after the IF stage of a conventional superhet receiver,
where signal levels are relatively high by virtue of the high IF gain. The small desired signals
are easily corrupted by noise and DC osets contributed by the mixers, local oscillator,
and the DC-coupled baseband ampliers. The local oscillator contributes to the DC oset
problem because self-mixing of the local oscillator with itself results in a DC component. In
addition, the self-mixing brings the phase noise sidebands of the local oscillator down into
the baseband spectrum, especially in the region near DC. The downconverted LO phase
noise adds to 1/f noise (icker noise) contributed by the mixer and ampliers, and can
signicantly degrade the signal to noise ratio for signal components located near the carrier
frequency, which are converted to very low frequencies near DC after the mixers.
Development of ZIF receivers is an active research area, with eorts focused on mini-
mizing mixer and LNA noise, development of extremely linear LNAs, and development of
low phase noise oscillators.
Some of the advantages of the zero-IF conguration can be obtained while avoiding the
problems associated with DC osets and 1/f noise near 0 Hz by downconverting the signal
to a low-IF instead of all the way to 0 Hz. Figure 2.9 compares the architectures of a
conventional superhet receiver, the low-IF, and the zero-IF congurations.
2.5 Software Dened Radio
The term Software Dened Radio (SDR) has become popular to describe transmitter and
receiver systems that utilize digital signal processing techniques to minimize the amount of
specialized analog hardware required for a specic application, and to allow reconguring
the communications system with changes in software alone. This is especially important
in view of the large number of standards in place, and under development, for the various
communications applications; cellular telephony is a good example - as of today, there are at
least 5 popular standards for modulation format in use, implemented in two widely separated
frequency bands.
Considering the receiver portion of a SDR-based transceiver, the ideal SDR would
employ an A/D converter to sample the voltage across the antenna terminals, and then
perform all ltering and demodulation in software. This is not practical at the present time,
mainly due to the limitations on A/D converter dynamic range (number of useful bits of
resolution) and sampling speed. Practical implementations of SDRs, in order of increasing
amount of software control and decreasing amount of hardware required, include:
1. sample the I and Q outputs of a quadrature demodulator, allowing ne-tuning, carrier
synchronization, demodulation, to be carried out entirely in software. The analog
82 CHAPTER 2. RECEIVERS
(a)
0
90
A/D
A/D
LPF
BW/2
LPF
BW/2
DSP-Demodulator
Superheterodyne
fIF
(b)
0
90
A/D
LPF
BW/2
LPF
BW/2
Demodulator
Low IF
Digital Signal Processor
fIF
Image-reject mixer
(c)
0
90
A/D
A/D
LPF
BW/2
LPF
BW/2
DSP-Demodulator
Zero-IF
Figure 2.9: Comparison of (a) superheterodyne, (b) low-IF superhet, and (c) zero-IF receiver
congurations with analog-to-digital (A/D) conversion and digital signal processing stages.
In (a) and (b) the IF lter passbands have been drawn to illustrate the fact that f
IF
is
typically large compared to the signal bandwidth in the case of a standard superheterodyne
receiver, whereas f
IF
is relatively small (low) in the case of the low-IF receiver. The rela-
tively low IF in the latter case makes it possible to digitize the IF output with high delity,
allowing the quadrature downconversion and demodulation stages to be implemented in a
DSP. The need for only one A/D converter is an advantage for the low-IF superhet receiver.
On the other hand, the preselector in a low-IF receiver will not be able to provide much
image rejection because the image response will be located close to the desired frequency.
An image-reject mixer can be used to provide additional image rejection (see homework
problem 18). The advantage of having only a single A/D converter comes at the cost of the
additional complexity of the image-reject mixer.
2.5. SOFTWARE DEFINED RADIO 83
frontend must implement some ltering, amplication, quadrature LO generation,
and multiplication/mixing of the RF and LO signals.
2. when very high speed A/D converters are available, a wide-bandwidth slice of the
ltered/amplied antenna output, or a wideband IF output, can be sampled at high
speed and digitally downconverted to produce sampled I and Q components with
narrower bandwidth and lower sampling frequency using a high speed dedicated signal
processor. Several specialized digital downconverter chips are available to perform this
function. Typically, the digital downverter samples at high rate and then applies a
rudimentary digital lter to the samples. The main purpose of the digital lter is to
place nulls at frequencies that will be folded into the desired slice of the spectrum after
decimation is performed. After the lter, the high-speed samples are decimated to a
rate that is at least twice the bandwidth required for subsequent processing. The lower-
rate samples are then processed by a conventional programmable DSP or computer.
The analog frontend provides ltering, amplication, and possible downconversion to
an IF; generation of the quadrature LO, mixing to I/Q channels, and demodulation is
performed in the digital downconverter and in software.
For those interested in learning more about SDR techniques through experimentation, there
are two excellent open-source programs available for *NIX platforms:
1. Linrad (www.nitehawk.com/sm5bsz/linuxdsp/linroot.htm) is a software receiver for
*NIX plaftorms that uses a stereo sound card to sample the I and Q outputs of a
quadrature demodulator. The program has AM/FM/SSB demodulators, and provides
a realtime waterfall display. This program is intended for demodulation of weak,
narrowband signals. A great deal of work has been put into implementing state-of-
the-art noise blanking, making this the program of choice for amateur radio operators
experimenting with communications using signals bounced o of the moon; called
earth-moon-earth, or eme, communications.
2. GnuRadio is the GNU software radio (www.gnu.org/software/gnuradio/index.html).
According to their web site: GNU Radio is a collection of software that when combined
with minimal hardware, allows the construction of radios where the actual waveforms
transmitted and received are dened by software. What this means is that it turns
the digital modulation schemes used in todays high performance wireless devices into
software problems.. GnuRadio software runs on *NIX, and can use various input
devices, including sound cards and high-speed A/D cards. Many state-of-the-art DSP
techniques applicable to SDR are available as modules. Receiving systems can be
easily put together and recongured using a Python interface.
84 CHAPTER 2. RECEIVERS
2.6 References
1. Communications Receivers: Principles and Design, Ulrich L. Rohde, Jerry Whitaker,
and T. T. N. Bucher, 2nd Edition, McGraw Hill, New York, 1997.
2. Single-Sideband Systems and Circuits, Edited by William E. Sabin and Edgar O.
Schoenike, McGraw Hill, New York, 1987.
3. The Science of Radio, Paul J. Nahim, 2nd Edition, Springer Verlag, New York, 2001.
4. The Design of CMOS Radio-Frequency Integrated Circuits, Thomas H. Lee, Cambridge
University Press, 1998.
5. RF Microelectronics, Behzad Razavi, Prentice Hall, 1998.
2.7. HOMEWORK PROBLEMS 85
2.7 Homework Problems
1. The transfer function for a regenerative amplier is found to be
T(!) =
GF(!)
1 AGF(!)
(2.7)
where A and G are assumed to be positive real constants with AG < 1. Suppose that
the lter response is approximated by a Gaussian function, i.e.,
F(!) = exp[(! !
c
)
2
/B
2
] (2.8)
This lter has a 3 dB bandwidth W = 1.177 B.
(a) Find an expression for the 3 dB bandwidth of the regenerative ampliers transfer
function, T(!).
(b) Evaluate the 3 dB bandwidth for the cases AG = 0.9 and AG = 0.99.
2. The transfer function for a regenerative amplier is:
T(!) =
GF(!)
1 AGF(!)
where A and G are positive, real constants with AG < 1. Suppose that the lter is
implemented with a series RLC circuit, in which case the transfer function F(!) will
have the form:
F(!) =
1
1 +jQ(
!
!o
!o
!
)
,
where the factor Q is a dimensionless constant.
(a) Find an expression for the -3 dB bandwidth of F(!). Note that |F(!)| will be a
bandpass function with a peak at ! = !
o
. The -3 dB bandwidth is dened as the
separation between the two frequencies !
1
, and !
2
which satisfy !
1
> 0, !
2
> 0
and |F(!
1
)| = |F(!
2
)| = |F(!
o
)|/
p
2.
(b) Find an expression for the -3 dB bandwidth of T(!). Discuss how the magnitude
of the loop gain (AG) aects the peak gain of T(!) and the bandwidth of T(!).
(c) Refer to your results from part b, and comment on how the gain-bandwidth
product associated with T(!) depends on the feedback gain parameter, A.
3. Consider a bandpass lter consisting of a number, N, of cascaded lters which are
isolated from each other by buer ampliers. Assume that each lter has the transfer
function given by F(!) in problem 2. The overall transfer function of the system will
be F(!)
N
. Derive an expression for the -3 dB bandwidth of the cascade of N lters.
Express the bandwidth as a product of two terms: (i) the bandwidth of a single lter
(this was found in problem 2, part a), and (ii) a bandwidth reduction factor which
depends only on N. Calculate the bandwidth reduction factor for N = 2, 3, 4, 5, 6.
4. Suppose we want to design a superhet receiver to cover the frequency range 1 - 30 MHz.
Two possible intermediate frequencies (IFs) for the receiver are:
86 CHAPTER 2. RECEIVERS
(i) 500 kHz
(ii) 60 MHz
For these IFs:
(a) Find the possible local oscillator frequencies (f
LO
) and image frequencies (f
IM
)
for reception of signals with carrier frequencies of 1, 5, 15, and 25 MHz. Make a
table.
(b) What are the LO tuning ratios required to cover the entire 1-30 MHz range for
each possible choice of LO frequency?
(c) From the standpoint of the easiest and least expensive preselector and LO design,
what is the optimum conguration? Assume that the required IF bandwidth is
easy to obtain, whatever the choice of the LO frequency. For the optimum con-
guration, describe the required preselector frequency response. Is it necessary
to tune the preselector?
5. Consider a single conversion superhet receiver that uses low-LO and downconversion.
(a) If the signals of interest have carrier frequencies in the range 140-170 MHz and
f
IF
= 21.4 MHz, nd the tuning range for the local oscillator. Specify the
minimum and maximum local oscillator frequencies.
(b) If the receiver is tuned to receive a desired signal with carrier frequency f
c
= 150
MHz, nd the image frequency.
(c) Assume that the preselector will have an ideal rectangular frequency response
function. Does the preselector for this receiver have to be tunable? (Yes or no).
Explain your answer. Do not consider the 1/2-IF response for this question.
(d) When the receiver is tuned to receive a desired signal with carrier frequency
f
c
= 150 MHz, denote the frequency of a signal that could cause interference if
the second harmonic of the interferer mixes with the second harmonic of the LO
by f
i
. Find all possible values of f
i
.
6. Consider the double-conversion receiver shown in Figure 2.10.
f
c
f
IF1
IF amplier First IF lter
Preselector
lter RF amplier
f
LO1
Mixer
f
IF2
IF amplier 2nd IF lter
Mixer
f
LO2
to
demod
Figure 2.10: Double-conversion receiver
Note that a double-conversion receiver requires two IF lters. Assume that the receiver
is to cover the 2-to-30 MHz range. The center frequency of the rst IF lter is 50 MHz
and that of the second is 500 kHz.
2.7. HOMEWORK PROBLEMS 87
(a) Specify the possible frequencies (f
LO1
and f
LO2
) for a receiver covering 2 to 30
MHz. You must specify a range of frequencies for f
LO1
, and a single frequency
for f
LO2
. Be sure to list both possibilities for each.
(b) Assume the rst IF lter is a bandpass lter with ideal (rectangular) frequency
response. What is the maximum bandwidth of the lter if secondary images are
to be avoided? A secondary image is an undesired frequency within the rst IF
lters passband that could be mixed into the second IF lters passband. Assume
that the second IF lter has a rectangular response with bandwidth 10 kHz.
7. We will design a receiver that can receive lower sideband signals. A block diagram
of the receiver is shown in Figure 2.11. Note that the SSB signal is demodulated
using a mixer and beat-frequency oscillator (BFO). The BFO and second mixer can
be thought of as a second frequency conversion stage where the second IF is at 0 Hz
(DC). The purpose of this stage is to translate the signal spectrum down to base-band,
i.e., to translate the carrier frequency to DC.
f
c
f
IF1
IF amplier First IF lter
Preselector
lter RF amplier
f
LO1
Mixer
3 kHz
Low-pass lter
Mixer
Beat Frequency
Oscillator (BFO)
Audio out
9 MHz
BW
1
Figure 2.11: Receiver to receive lower sidebands
Suppose the input signal is an LSB signal with carrier frequency f
c
=14.3 MHz and
bandwidth = 3 kHz as shown in Figure 2.12. You may assume that the IF lter has
f
c
3 kHz
LSB
f
c
Figure 2.12: LSB signal
an ideal rectangular bandpass characteristic with center frequency of 9.0 MHz and
bandwidth BW
1
= 3 kHz. Thus, the IF lter bandwidth is just wide enough to pass
the LSB signals spectrum.
(a) Give two choices for the frequency of the rst local oscillator, f
LO1
. Be careful:
make sure that your choice of local oscillator frequency will cause the entire
spectrum of the desired signal to fall within the IF lters passband.
88 CHAPTER 2. RECEIVERS
(b) For each choice given in part 7a, sketch the spectrum of the signal as it would
appear just after the IF lter.
(c) For each choice given in part 7a, give the band of frequencies that correspond to
the image response of the receiver.
(d) For each choice given in part 7a, give the frequency of the beat-frequency oscillator
such that the signal is properly demodulated.
8. Consider the triple-conversion receiver shown in Figure 2.13. Suppose that f
LO1
=
1300 MHz, f
LO2
= 1410 MHz, and f
LO3
= 179.3 MHz. Note that the only lter is the
nal IF lter which has an ideal rectangular frequency response function centered at
f
IF
=10.7 MHz:
f
LO1
f
IF
f
LO3
f
LO2
Figure 2.13: Triple-conversion receiver
(a) Find and list all input frequencies that will give an output from the nal IF
lter. You may ignore the nite bandwidth of the IF lter, i.e., list only the
input frequencies that will give an output at exactly 10.7 MHz.
(b) How many input frequencies would be on your list if there were 4 conversions
instead of 3?
9. Consider the design of the receiver portion of a portable cellular telephone. The
cellular telephone receives frequency-modulated that are transmitted from a base-
station. The received signals have a bandwidth of 30 kHz and a carrier frequency, f
c
,
somewhere within the frequency range 869-894 MHz. While receiving, the telephone
must simultaneously transmit frequency-modulated signals back to the base-station
at a carrier frequency f
c
-45 MHz, i.e., the telephone transmits at a frequency 45 MHz
below the frequency at which it receives.
Sketch a block diagram of a double-conversion superhet receiver that could be used for
receiving cellular phone signals. Your design should have a tunable rst local oscillator
that will serve a dual purpose. This oscillator will also serve as the carrier oscillator
for the signals transmitted by the telephone. Assume that the second IF lter has a
center frequency of 5.5 MHz.
Specify the following and state the reasoning that led to your choice.
(a) The center frequency of the rst IF lter.
2.7. HOMEWORK PROBLEMS 89
(b) The maximum bandwidth of the rst IF lter. Assume that the lter response is
symmetric around the center frequency. Indicate what considerations led to your
choice for the maximum bandwidth.
(c) The bandwidth of the second IF lter.
(d) Sketch an appropriate transfer function for the preselector.
(e) Specify the tuning range for the rst LO, i.e., specify the minimum and maximum
rst LO frequencies.
(f) Specify the image frequency when the phone is set up to receive signals at f
c
=875
MHz.
(g) As noted in the problem description, the bandwidth of the cellular phone signals
is 30 kHz. If the highest frequency present in the voice signal (m(t)) is 3 kHz,
estimate the peak frequency deviation of the cellular phone transmitter.
10. Design a double-conversion receiver (refer to Fig. 2.10) that will receive frequencies in
the range 144-148 MHz. The signal bandwidth is 10 kHz. The rst IF frequency is to
be 10.7 MHz. The second IF frequency is 455 kHz. You also know that the receiver is
going to be operated in a region where very strong local stations exist in the frequency
range 122-125 MHz.
Specify the following:
(a) The bandwidths of the rst and second IF lters, BW
1
and BW
2
. Explain your
choices.
(b) The frequency (or frequency range) for each local oscillator, f
LO1
and f
LO2
.
Explain your choices. Tuning ratio is not a signicant consideration in this
receiver, so do not invoke smaller tuning ratio as a reason for preferring one
choice for f
LO1
over another one.
(c) The preselector lter would normally be a bandpass lter in this application. Is
it necessary to use a tunable (variable) preselector lter? Why or why not?
11. Consider a single-conversion superhet receiver with f
IF
= 260 MHz. The receiver
must be able to tune to Sirius and XM satellite radio signals with carrier frequencies
in the range 2320 MHz f
RF
2345 MHz.
(a) Specify the two possible tuning ranges for the local oscillator.
(b) For each answer in part (a.), specify the image frequency with the receiver is
tuned such that a signal at frequency f
RF
= 2335 MHz will be centered in the
IF lters passband.
(c) Suppose that the receiver is implemented using low-LO. Give the frequency
of a signal that would cause interference as a result of the 1/2 IF response
mechanism when the receiver is tuned to receive a desired signal with carrier
frequency f
RF
= 2335 MHz.
(d) Does the preselector for this receiver have to be tunable? Explain your answer.
12. Consider the double-conversion superhet receiver in Figure 2.14:
This receiver is to be designed to cover two dierent carrier-frequency bands (144-194
MHz and 420-470 MHz) with a single tunable rst local oscillator and rst IF Filter.
90 CHAPTER 2. RECEIVERS
f
IF1
First IF lter
Preselector 1
144-194 MHz
f
LO1
f
IF2
2nd IF lter
f
LO2
to
demod
BW
1
BW
2
Preselector 2
420-470 MHz
Figure 2.14: Dual band superhet receiver
To change the carrier-frequency band, only the preselector lter will be switched as
shown in Figure 2.14. You may assume that the signals of interest have a bandwidth
of 15 kHz.
(a) Specify the rst intermediate frequency (f
IF1
). There are two possibilities - nd
both of them.
(b) For each possible IF found in part (a), specify the 50 MHz-wide range of frequen-
cies covered by the rst local oscillator (f
LO1
).
(c) Specify the smallest and the largest bandwidth (BW
1
) that could be used for the
rst IF lter. Explain what factors determined your choices. You may assume
that the second IF lter has an ideal rectangular frequency response function
with f
IF2
= 10.7 MHz and BW
2
= 15 kHz.
13. The double conversion receiver shown in Figure 2.15 uses a special system to generate
the rst local oscillator signal. This is called the Wadley Loop drift cancelling system.
It is used to make the receiver tuning insensitive to relatively small changes in the
second local oscillator frequency that arise because of frequency drift. This makes it
possible to use a relatively inexpensive oscillator for the second LO.
Suppose the input signal of interest has carrier frequency f
RF
=200 MHz. The rst IF
lter is centered at f
IF1
=790 MHz and the second IF lter is centered at f
IF2
=70 MHz.
The local oscillator lter is centered on 590 MHz. The RF signal has a bandwidth
of 50 kHz. Assume that the IF lters have ideal, symmetrical rectangular passbands
with bandwidth BW
1
and BW
2
as shown in Figure 2.15.
(a) Specify the bandwidth of the second IF lter, BW
2
. State the reason for your
choice.
(b) Give the maximum allowable bandwidth of the rst IF lter (BW
1
) so that
secondary images will be rejected.
(c) Determine f
0
LO
such that an input signal with carrier frequency of 200 MHz will
be centered in the passband of both the rst and second IF lters when f
LO2
=860
MHz and such that the frequency of the desired signal at the output of the second
2.7. HOMEWORK PROBLEMS 91
Preselector
Filter
First IF Filter Second IF Filter
Local Oscillator
Filter
Mixer Mixer
f
IF2
= 70 MHz
IF1
f = 790 MHz
f
RF
= 200 MHz
f
LO
f
LO1
=590 MHz
f
LO2
= 860 MHz
BW
1
BW
2
Figure 2.15: Wadley Loop drift canceling system
mixer is independent of the exact value of f
LO2
. Hint: Write f
LO2
as f
LO2
=860
+ and nd the choice for f
0
LO
that makes the frequency after the second mixer
independent of . There is only one correct choice for f
0
LO
.
(d) If you have done part 13c correctly, then the frequency of the desired signal after
the second mixer will not change if f
LO2
drifts away from its nominal value of 860
MHz. This does not mean that f
LO2
can take on any value, however. Suppose
that represents the frequency drift, i.e., suppose that f
LO2
=860 + . What
factor(s) limit the allowable size of ?
14. Consider a single conversion superhet receiver with f
IF
= 70 MHz. The receiver must
be able to tune to carrier frequencies in the range 1930 MHz f
RF
1975 MHz.
(a) Specify the two possible frequency ranges for the rst local oscillator.
(b) For each answer in part (a.), specify the image frequency when the receiver is
tuned to receive a desired signal with carrier frequency f
RF
= 1940MHz.
(c) In one or two sentences, explain how the 1/2 IF response of a receiver comes
about.
(d) Suppose that the receiver described in part (a.) is implemented using high-LO.
Give the frequency of a signal that would cause interference as a result of the
1/2 IF response mechanism when the receiver is tuned to receive a desired signal
with carrier frequency f
RF
= 1935 MHz.
15. You are designing a superhet receiver for a PCS-band cellular phone handset to receive
carrier frequencies in the range 1930 MHz f
RF
1975 MHz. Assume that the
receiver will employ high-LO.
(a) Suppose that the preselector has a xed rectangular bandpass response that
passes 1930-1975 MHz. Determine the smallest value for the IF that will allow
the preselector to reject signal(s) at the image frequency.
92 CHAPTER 2. RECEIVERS
(b) Suppose that the preselector is the same as described in part a. Determine the
smallest value for the IF that will allow the preselector to reject signal(s) whose
second harmonic could mix with the second harmonic of the LO to produce an
output at the IF.
16. Consider a double conversion superhet receiver with rst IF and last IF denoted by
f
IF1
and f
IF2
, and bandwidth of the rst and last IF lters denoted by f
1
and f
2
,
respectively. You may assume that both IF lters have ideal rectangular response
functions that are symmetrical about the center frequencies and that f
IF2
f
IF1
.
(a) Give an expression for the maximum bandwidth of the rst IF lter, f
1,max
,
such that secondary images will be rejected. You do not need to consider the
half-IF response.
(b) Suppose, for the purpose of this problem, that it is impractical to build or acquire
IF lters with fractional bandwidth f/f
IF
smaller than 2%. For a receiver using
a last IF of 455kHz, specify the largest practical value for the rst IF. For this
calculation, you may assume that the bandwidth of the second IF lter (f
2
) is
small enough to ignore.
17. Consider a single conversion superhet receiver that uses downconversion to an IF
denoted by f
IF
and high-LO. We showed that when such a receiver is tuned to
receive a desired signal with carrier frequency f
c
, an undesired signal with frequency
f
i
= f
c
+
f
IF
2
could cause interference if the second harmonic of the undesired signal
mixes with the second harmonic of the LO. This is called the half-IF response because
the potential interferer is oset from the desired carrier frequency by
f
IF
2
. Find the
oset from the desired carrier frequency, f
c
, for a signal at frequency f
i
that could
cause interference when the third harmonic (3f
i
) mixes with the third harmonic of
the local oscillator (3f
LO
). There are two answers. Find them both and express your
results in terms of f
IF
.
18. (Image-reject mixer) When a multiplier is used to perform frequency conversion it is
necessary to include a pre-selector lter in front of the multiplier to prevent signals at
the image frequency from being converted to the intermediate frequency (IF) at the
mixer output. The lter becomes hard to realize when the IF is small, since the image
frequency is very close to the desired frequency. When the IF is small, a better method
for rejecting images is the phasing method, whereby the response from the image is
cancelled at the mixer output. One implementation of a phasing-type image-reject
mixer is shown in Figure 2.16.
(a) For the RF input signal Acos !
RF
t, nd the time-domain IF output signal if the
+ sign is used in the signal combiner. You will need to consider two cases: (i)
!
LO
> !
RF
and (ii) !
LO
< !
RF
. Be careful - if you have a signal of the form
sin[(!
1
!
2
)t] and you want to delay it by 90 degrees, then the result will be
cos[(!
1
!
2
)t] where the upper sign applies if !
1
> !
2
and the lower sign
applies if !
1
< !
2
.
(b) Repeat part a for the case where the sign is used in the signal combiner.
(c) Use your results from parts (a) and (b) to answer this part. Suppose this mixer
is to be used to convert a desired signal at frequency !
c
(i.e. RF input signal is
2.7. HOMEWORK PROBLEMS 93
RF
cos(!
LO
t)
90
cos(!
LO
t)
sin(!
LO
t)
+
-
90
phase
delay
IF
a
b
c
Figure 2.16: Image-reject mixer
Acos !
c
t) to an IF denoted by !
IF
using high LO (!
LO
= !
c
+!
IF
). The image
frequency is !
IM
= !
c
+ 2!
IF
. Which sign should be used in the combiner in
order for the system to produce the desired frequency conversion?
(d) For the sign that you chose in part (c), write the IF output signal when the
input signals frequency is !
IM
= !
c
+ 2!
IF
, i.e. when the RF input signal is
Acos !
IM
t. You should nd that this input signal produces a nite output from
the mixer. Why, then, is this called an image-reject mixer?
19. We have seen several systems that use cancellation to eliminate unwanted signal com-
ponents. In practice, perfect cancellation is dicult to achieve because of inevitable
amplitude and phase dierences between the two signals that are being combined. In
this problem, we analyze the impact of amplitude and phase imbalance on the sup-
pression of unwanted terms. Consider the superposition of two signals which dier
slightly in amplitude and phase, i.e.
s
. (3.1)
At higher frequencies, where the skin depth is small compared to the wire diameter, an
accurate approximation for the resistance of the wire is obtained by assuming that the
current ows with uniform density within one skin-depth, , from the wire surface, and with
zero density elsewhere. The AC resistance is then obtained by replacing A in equation (3.1)
with the area contained within one skin depth of the surface, i.e. d. The AC resistance
can then be written as
R = R
DC
(d/2)
2
d
= R
DC
d
4
, d (3.2)
where R
DC
is the DC resistance of the wire. Since the skin depth decreases with increasing
frequency, the resistance will increase as f
1/2
. Obviously, the wire resistance can be much
greater than the DC resistance, especially at high frequencies. For example, consider AWG
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 97
22 wire, which has DC resistance 0.0053 /cm. At 10 MHz the resistance is 0.04 /cm,
and at 1 GHz the resistance is 0.4 /cm.
The approximation given in equation (3.2) is based on the assumption that the wire is
far from any other conductors, so that the time-varying magnetic eld within the wire is
only due to the current within the wire. Suppose that two parallel wires, carrying the same
current, are brought into close proximity. In this case, the magnetic eld within each wire
will include a contribution from the current in the neighboring wire, i.e. the wires will be
inductively coupled. In this case, the current density within the wire will not be uniformly
distributed around the outer shell dened by the skin-depth. The electric eld induced
within the wires will cause the current density to be decreased on the side of the wire that
is closest to the neighboring wire. This proximity eect reduces the eective cross-sectional
area even further, and increases the AC resistance of the wire. The proximity eect can be
very important in inductors where multiple wires are in close proximity.
3.1.1.2 Inductance of wires
The inductance, per unit length, of a wire with length l, diameter d, and distance h from a
ground plane is (when d l and h l):
L =
0
2
cosh
1
2h
d
. (3.3)
If the wire axis is located at least one wire diameter above the ground plane (h/d > 1), then
the following approximation is useful
L '
0
2
ln
4h
d
. (3.4)
For values of h/d in the range 1 to 100, equation (3.4) predicts that L ranges from 2.8
nH/cm to 12 nH/cm. Notice that the inductance per unit length is relatively insensitive to
the exact value of h/d.
A useful number to remember is that AWG 22 copper wire placed directly on the top
surface of a printed circuit board with dielectric thickness of .062 inches =1.57 mm (a
standard thickness), and with a ground plane on the bottom of the board, will have an
inductance of approximately 4.8 nH/cm. At 100 MHz the inductive reactance of such a wire,
X
L
= !L, will be approximately 3 /cm, and at 1 GHz the gure is 30 /cm. For short
wires, the resistance is often small enough to be ignored, however the inductance of the wire
is often signicant. In a circuit where impedances are relatively low, the series impedance
of even a short connecting lead may have a signicant impact on circuit performance. This
leads to a fundamental rule of RF circuit design - at high frequencies it is important to
keep the length of interconnecting wires and circuit-board traces short in order to minimize
lead inductance. When components are separated by signicant distances, interconnections
must be treated as distributed circuit elements, and transmission line models are used to
model the conductors that interconnect components.
3.1.2 Resistors
Several types of resistors are used in RF circuits, including wire-wound, carbon composition,
thick lm, and thin lm units. Wire-wound resistors consist of a length of lossy wire that is
coiled up to t into a small package. This type of resistor is seldom used at RF because they
98 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
have relatively large inductance. Carbon composition resistors consist of a lossy dielectric
material sandwiched between two conducting electrodes. Thick or thin-lm resistors consist
of a lm of conducting material deposited on an insulating substrate. The lm is in contact
with two conductive electrodes to provide a means for connection to external circuitry.
Film resistors are available in cylindrical packages with attached connecting leads and also
as surface mount devices (SMDs). Thick or thin lm resistors in a surface mount package (
aka, chip resistors) are the most common type for RF applications.
Over a fairly wide frequency range, resistors can be modeled using the equivalent circuit
shown in Figure 3.2.
R L
C
Figure 3.2: High frequency equivalent circuit for a resistor without external connecting
leads.
The inductor in the model shown in Figure 3.2 represents the inductance associated with
the current path through the resistor, and the capacitor represents the capacitance between
the two electrodes used to connect the resistor to external circuitry. The unavoidable induc-
tance and capacitance associated with the resistors (and other components) are sometimes
termed parasitic inductance and capacitance. Both the inductance and capacitance of a re-
sistor depend on the geometry and dimensions of the resistor. The inductance is primarily
determined by the length of the current path within the element, and the capacitance is
determined by the size and separation of the contact electrodes, as well as the dielectric
permittivity of the material between the electrodes. Generally, the inductance and capaci-
tance associated with a miniature surface mount resistor package are on the order of 1 nH
and 1 pF, respectively. Signicantly higher inductance would be associated with a part in
an axial package with wire leads. Even with very short leads, such a package would have a
typical inductance value on the order of 10 nH. If the component has external connecting
leads attached to the package, then the inductance of the wire leads and the capacitance
between the leads may need to be added to the model, as shown in Figure 3.3 .
C
w
L
w
/2 L
w
/2
C
R L
Figure 3.3: Model for a resistor with connecting leads. The total inductance of the leads is
L
w
and the capacitance between the leads is C
w
.
For many purposes, the simpler model of Figure 3.2 can be used even for resistors with
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 99
external connecting leads if the inductance is taken to be the sum of the resistor inductance
and the lead inductance, and the shunt capacitance is taken to be the sum of the package
capacitance and the capacitance between the leads. The following discussion will be based
on the simpler model shown in Figure 3.2.
When the resistance is small ( 100 ), and the frequency is not too high, the series
LR branch of the model has a much lower impedance than the capacitance that shunts it.
The capacitive reactance can then be neglected, leading to the simplied series RL model
in Figure 3.4.
R L
Figure 3.4: Equivalent circuit for a small resistance
For example, consider a 50 resistor with lead inductance L = 10 nH, shunt capacitance
C = 1 pF. Figure 3.5 shows the magnitude of the impedance versus frequency up through
1 GHz calculated using the full model. In this case the series inductance acts to increase
the impedance at high frequencies.
0.0
20.0
40.0
60.0
80.0
100.0
120.0
140.0
0.0 200.0 400.0 600.0 800.0 1000.0
|Z|,
Frequency (MHz)
Figure 3.5: Impedance versus frequency of a 50 resistor. The dotted line shows the
impedance of an ideal resistor.
When the resistance is large (100 ) and the frequency is not too high the inductive
reactance will be small compared to R and can be neglected; Figure 3.6 shows the equivalent
circuit in this case.
For example, consider a 10 k resistor with the same inductance and shunt capacitance
as before (L=10 nH, C=1 pF). Figure 3.7 shows the impedance versus frequency calculated
using the full model. Here the shunt capacitance is dominant and tends to short out the
resistance, resulting in a dramatic reduction in the impedance at high frequencies.
For moderate resistance values the parasitic elements tend to have a smaller eect on
the impedance. Figure 3.8 shows the magnitude of the impedance for a 200 resistor with
L = 10 nH, C = 1 pF. In this case neither of the parasitic reactances is signicant compared
to 200 and the impedance variation is relatively small up through 1000 MHz.
100 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
R
C
Figure 3.6: Equivalent circuit for a resistor with high resistance.
0.0
2000.0
4000.0
6000.0
8000.0
10000.0
12000.0
0.0 200.0 400.0 600.0 800.0 1000.0
|Z|,
Frequency (MHz)
Figure 3.7: Impedance versus frequency of a 10 k resistor. The dotted line shows the
impedance for an ideal resistor.
0.0
50.0
100.0
150.0
200.0
250.0
300.0
350.0
0.0 200.0 400.0 600.0 800.0 1000.0
|Z|,
Frequency (MHz)
Figure 3.8: Impedance versus frequency of a 200 resistor
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 101
Notice that the impedance of a resistor can be much larger or smaller than the DC resis-
tance, depending on the resistance value and frequency. Until some intuition is developed,
it is a good idea to measure the impedance of any resistor that is being considered for use
in an RF circuit.
3.1.3 Capacitors
Capacitors are constructed by separating two conducting electrodes by an insulating medium
such as air, or a low-loss dielectric material. Loss in the dielectric is modeled as a resistance
(R
p
) in parallel with the intrinsic capacitance. The inductance associated with the current
path through the electrodes and any connected leads appears in series as shown in Figure
3.9. A resistance in series with the inductor (R
s
) models losses in the electrodes and leads.
C
R
p
L R
s
Figure 3.9: Capacitor model including lead inductance (L), dielectric loss resistance (R
p
),
and conductor loss (R
s
).
The model can be transformed into a series RLC model using a parallel to series trans-
formation, as shown in Figure 3.10. where it is assumed that the dielectric loss resistance
is large compared to the capacitive reactance (R
p
1/(!C)). The sum of the ohmic resis-
tance and the transformed dielectric loss resistance is termed the equivalent series resistance
(ESR), denoted here by R
esr
, i.e.
R
esr
= R
s
+ 1/(R
p
!
2
C
2
).
The ESR of a capacitor is dominated by dielectric loss at suciently low frequencies, whereas
the resistance of the electrodes and leads will dominate at high frequencies.
C
1
R
p
(!C)
2
L R
s
Figure 3.10: Equivalent series RLC circuit of a capacitor, valid where R
p
1/(!C), derived
from the model shown in Figure 3.9.
The Q of a capacitor at any frequency is the ratio of the reactance and the ESR
Q =
|X|
R
esr
, (3.5)
where the reactance X = !L 1/(!C). The dissipation factor, d, is the inverse of the Q,
i.e
d =
1
Q
=
R
esr
|X|
. (3.6)
102 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
The dissipation factor is also called the loss tangent because it is the complement of the
tangent of the phase angle associated with the capacitor impedance.
tan =
R
esr
|X|
= !CR
esr
At frequencies well below the series resonant frequency of a capacitor, the reactance reduces
to X ' 1/(!C) and Q, d, and tan , can be written as follows:
Q =
1
!CR
eq
d = tan = !CR
eq
.
All capacitors will have a series resonant frequency, f
s
= 1/(2
p
LC). Above this frequency,
the inductive reactance dominates, and the net reactance is positive. Hence, capacitors are
inductive at frequencies above f
s
! For a given package type the inductance will be roughly
independent of the capacitance value, hence the series resonant frequency will be lower for
larger values of capacitance.
As an example, consider a 0.01 F capacitor with total lead length of 1 cm. We know,
from previous discussion, that a typical value of the lead inductance is 10 nH/cm, which gives
a total inductance of 10 nH and a series resonant frequency of 15.9 MHz The magnitude of
the impedance and the reactance versus frequency are shown in Figure 3.11. The impedance
goes to zero at the series resonant frequency in this lossless model. A real capacitors
impedance will fall to a minimum value equal to R
esr
at the series resonant frequency, f
s
.
Circuit designers make explicit use of the fact that capacitor impedance is smallest at and
near the series resonant frequency. When a capacitor is used as a DC bias circuit decoupling
element or as an interstage DC-blocking coupling element, it is sometimes possible to choose
the capacitor so that the intended frequency of operation falls near the series resonant
frequency of the capacitor, where the impedance is smallest.
3.1.4 Inductors
3.1.4.1 Air Core Inductors
An approximate formula for the inductance of a close-wound single-layer coil with nonmag-
netic core (e.g., air) is
L =
(rN)
2
9r + 10l
(3.7)
where
L = inductance in H
N = number of turns
r = radius of coil (inches)
l = length of coil (inches)
Formula 3.7 is accurate to within 1% if l > 0.8r, i.e., if the coil is not too short.
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 103
(a)
0.1
1.0
10.0
100.0
1.0 10.0 100.0
|Z|,
Frequency (MHz)
(b)
4.0
2.0
0.0
2.0
4.0
10.0 20.0 30.0 40.0 50.0
X,
Frequency (MHz)
Figure 3.11: (a) Log-magnitude of impedance and (b) reactance for a 0.01 F capacitor with
series inductance of 10 nH. The dotted line shows the result for an ideal 0.01 F capacitor.
104 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
3.1.4.2 Toroidal Inductors
Toroidal inductors are formed by winding a coil of wire around a donut-shaped (toroidal)
core. An important advantage of this type of inductor is its self-shielding property. Typ-
ically, the core material will have high relative magnetic permeability (
r
1), and the
magnetic eld will be essentially conned to the core. This means that the coil inductance
will be unaected by its physical orientation and also that negligible mutual coupling will
exist between toroidal coils in close proximity. Circuit design can be greatly simplied if
mutual coupling eects can be neglected.
The core material for a toroidal inductor is usually a magnetic material, that is, a material
with magnetic permeability larger than the permeability of free space. Inductor cores for RF
applications are either manufactured from iron powder or ferrite material. The powdered
iron cores consist of small iron particles suspended in an insulating compound. The mixture
is compressed and baked at high temperature to produce a rigid structure. Dierent mixtures
exhibit dierent magnetic permeabilities. Ferrites are ceramic materials with iron oxide
as the dominant constituent. The iron oxide is mixed with nickel, manganese, zinc, or
magnesium. Again, dierent mixtures exhibit dierent magnetic permeabilities. Generally,
ferrite cores oer higher magnetic permeability than iron-powder cores, which results in
fewer turns required to realize a given inductance value. Iron-powder cores generally have
higher saturation ux densities. Saturation ux density refers to the largest ux density
for which the linear relationship B = H holds. A general rule of thumb for high power
RF applications is that the power handling capability of a ferrite core is limited by ux
saturation, while the limiting factor for iron powder is temperature rise.
The important parameters of the core material are its relative magnetic permeability,
cross-sectional area and diameter, and volume resistivity (which determines core loss). Other
considerations for high power applications are the saturation ux density, which determines
the largest magnetic eld that the core can support, and the temperature rise resulting
from power dissipation. For small signal applications these need not be considered, but the
temperature stability of the core may need to be considered.
An approximate formula for the inductance of a toroidal winding having cross sectional
area A and eective length l
eff
is:
L =
A
l
eff
N
2
(3.8)
= A
0
L
N
2
where
=
r
o
(3.9)
A
0
L
=
A
l
eff
(3.10)
The numerical value of the parameter A
0
L
is the inductance for a single-turn winding.
The core manufacturer will provide the A
0
L
values for each type of core. The inductance of
an N-turn winding is found from equation (3.8). Often, core manufacturers will characterize
their cores by specifying the inductance of a winding with some particular number of turns.
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 105
For example, one manufacturer species the inductance per 100 turns for iron powder cores.
If this constant is denoted by A
L
, it is related to A
0
L
in equation (3.10) by
A
L
= A
0
L
(100)
2
. (3.11)
On data sheets, the units of A
L
would typically be given as as H/(100 turns).
3.1.4.3 Equivalent circuit model for an inductor
A reasonably accurate equivalent circuit for an inductor consisting of a coil of wire wound
as a solenoid, around a torus, or in a plane includes a series resistance to model the ohmic
loss in the wire and a shunt capacitance to model the distributed capacitance between the
turns of the coil. The inductance of the wire results in voltage dierences between the
dierent parts of the coil. This voltage dierence sets up an electric eld in the air and in
any dielectric material near the coil. The eect of this stored electric energy can be modeled
with a capacitance shunted across the terminals of the coil. This eective capacitance is
called the distributed capacity of the coil. The distributed capacity of a coil is dependent on
the geometry of the coil and the number of turns. In general, very closely spaced windings
will have a larger distributed capacity. The equivalent circuit for an inductor is shown in
Figure 3.12. If the inductor is wound on a core with nite conductivity (ferrite or iron-
powder are often used), then it may be necessary to account for core losses by augmenting
the model shown in Figure 3.12 with a resistor in parallel with the capacitor.
R L
C
Figure 3.12: Equivalent circuit for an inductor
A lower-bound on the coil resistance may be estimated from the AC resistance of an
isolated wire having length equal to the total length of wire in the coil, i.e.
R = r
DC
ndD
4
(3.12)
where r
DC
is the DC resistance per unit length of the wire, d is the wire diameter, D is the
coil diameter, n is the number of turns in the coil, and is the skin depth in the wire. A
more accurate estimate of the coil resistance can be obtained by accounting for the proximity
eect. The importance of the proximity eect will depend on the geometry of the coil. The
proximity eect may cause the actual resistance to be signicantly larger than the estimate
given in equation 3.12.
The impedance and reactance versus frequency are shown in Figures 3.13 and 3.14 for
a 10 H inductor with (constant) series resistance of 15 and distributed capacity of 20
pF. Note the parallel resonant frequency f
p
' 1/(2
p
LC) at which the impedance of the
inductor has a relatively large value.
Circuit designers often make use of the fact that an inductor has a very large impedance
near the parallel resonant frequency, and will deliberately use the device at or just below
106 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
0.0
5000.0
10000.0
15000.0
20000.0
25000.0
30000.0
5.0 10.0 15.0 20.0 25.0
|Z|,
Frequency (MHz)
Figure 3.13: Impedance versus frequency for a 10 H inductor with series resistance of 15
and distributed capacity of 20 pF. The dotted line shows the impedance for an ideal 10
H inductor.
15000.0
10000.0
5000.0
0.0
5000.0
10000.0
15000.0
5.0 10.0 15.0 20.0 25.0
X,
Frequency (MHz)
Figure 3.14: Reactance versus frequency for a 10 H inductor with series resistance of 15
and distributed capacity of 20 pF. The dotted line shows the reactance for an ideal 10 H
inductor.
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 107
this frequency. For example, in many cases an inductor is used to provide a DC bias signal
to a circuit, but it is necessary to isolate the bias supply from the circuit at the operating
frequency. In this case, an inductor with parallel resonant frequency near the operating
frequency may be employed to provide a low impedance to DC and large impedance to
RF signals. In such an application the inductor element is referred to as an RF choke. It
should be noted that the parallel resonant frequency of an inductor depends critically on
its construction. It is dicult to accurately estimate the parallel resonant frequency of an
inductor. In practice it is usually necessary to measure the parallel resonant frequency as
well as the series resistance.
When considering a particular inductor for use in a circuit, the designer needs to be aware
of the parallel resonant frequency as well as the Quality Factor, or Q, of the inductor. The
Q of an inductor is dened to be the ratio of inductive reactance and resistance associated
with the component, i.e.,
Q =
|X
s
|
R
s
(3.13)
where the impedance of the inductor is Z = R
s
+ jX
s
. The higher the Q, the better
the inductor approximates an ideal component. The Q is an important parameter if the
inductor is to be used in a resonant circuit, lter, or matching network.
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
50.0
0.0 5.0 10.0 15.0 20.0 25.0
Q
L
Frequency (MHz)
Figure 3.15: The quality factor (Q
L
= |X
S
|/R
S
) for the inductor whose impedance charac-
teristics are plotted in Figures 3.13 and 3.14. The component is inductive only at frequencies
below its self-resonant frequency (approx. 11.2 MHz in this case), so the Q is not plotted
at frequencies above the self-resonant frequency. The optimum frequency of operation for
an inductor is at or near the peak of the Q vs. frequency curve.
Finally, it should be noted that substantial mutual inductance may exist between induc-
tors that are located in close proximity. This coupling can cause serious problems if it has
not been accounted for in the design of the system, especially when it occurs between the
input and output circuits of an amplier, since unwanted oscillation may result. Mutual
coupling eects can be minimized by using self-shielding construction (e.g., toroidal induc-
tors) or, in the case of air-wound coils, by orienting the axes of the coils so that they are
perpendicular. A model and theoretical analysis for coupled inductors will be presented in
a later chapter.
108 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
3.2 References
1. Bowick, Chris and Howard W. Sams, RF Circuit Design, Indianapolis, Indiana, 1982.
2. DeMaw, M. F., Ferromagnetic Core Design and Application Handbook, Prentice Hall,
Inc, 1981.
3. Ludwig, Reinhold and Pavel Bretchko, RF Circuit Design - Theory and Applications,
Prentice Hall, 2000.
4. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw Hill, 1943.
3.3. HOMEWORK PROBLEMS 109
3.3 Homework Problems
1. You are given a black box with two terminals. Suppose that you know that the
box contains a passive circuit that is constructed from 3 elements: a resistor (R),
lossless capacitor (C), and lossless inductor (L). Your task is to gure out how the
elements are connected, and what their values are. You make some measurements of
the impedance of the box Z(f) = R(f) + jX(f). The results of the measurements
are shown in Figure 3.16. Sketch the circuit that is inside of the box. Estimate the
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
2 4 6 8 10 12 14 16 18 20 22 24
R,
Frequency (MHz)
400
200
0
200
400
600
800
1000
1200
2 4 6 8 10 12 14 16 18 20 22 24
X,
Frequency (MHz)
Figure 3.16: Measured resistance (top) and reactance (bottom).
110 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
element values. You may assume that the elements (R, L, C) are ideal, i.e., assume
that distributed capacitance across the inductor and the lead inductance of the resistor
and capacitor can be neglected.
2. Same as problem 1, except the measured data is shown in Figure 3.17. Sketch the
0
25
50
75
100
125
150
175
200
225
250
275
300
325
350
375
400
2 4 6 8 10 12 14 16 18 20 22 24
R,
Frequency (MHz)
250
225
200
175
150
125
100
75
50
25
0
25
50
75
100
125
150
175
200
225
250
2 4 6 8 10 12 14 16 18 20 22 24
X,
Frequency (MHz)
Figure 3.17: Measured resistance (top) and reactance (bottom).
circuit that is inside of the box. Estimate the element values. You may assume that
the elements (R, L, C) are ideal, i.e., assume that distributed capacitance across the
inductor and the lead inductance of the resistor and capacitor can be neglected.
3.3. HOMEWORK PROBLEMS 111
3. Same as problem 1, except the measured data is shown in Figure 3.18. Sketch the
0
25
50
75
100
125
150
175
200
225
250
2 4 6 8 10 12 14 16 18 20 22 24
R,
Frequency (MHz)
200
150
100
50
0
50
100
150
200
2 4 6 8 10 12 14 16 18 20 22 24
X,
Frequency (MHz)
Figure 3.18: Measured resistance (top) and reactance (bottom).
circuit that is inside of the box. Estimate the element values. You may assume that
the elements (R, L, C) are ideal, i.e., assume that distributed capacitance across the
inductor and the lead inductance of the resistor and capacitor can be neglected.
4. The circuit shown in Figure 3.19 is used as a model for a realistic resistor or inductor.
It can also be used to model a realistic parallel resonant circuit.
(a) Find an expression for the impedance of this circuit.
112 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
L
C
R
Figure 3.19: Model for realistic inductor and for parallel resonant circuit
(b) The frequency at which the impedance is purely resistive is called the resonant
frequency. Under what conditions will there be a frequency (> 0) at which
the impedance is purely resistive? Show that if this condition is satised, and if
CR
2
/L 1, then the resonant frequency is given approximately by:
!
o
=
1
p
LC
(1
CR
2
2L
) (3.14)
Note that this frequency is the parallel resonant frequency of the circuit and that
the eect of the resistance, R, is to make the parallel resonant frequency slightly
smaller than that of the lossless (R=0) case.
(c) Continuing from part 4b, assume that the term CR
2
/L can be ignored and nd
the magnitude of the impedance at the parallel resonant frequency. How does it
depend on R?
(d) For parts 4d and 4e you should start with the exact expression for the impedance.
Find an approximate expression for the impedance valid when !
1
p
LC
and
L
R
2
C
1. Draw a simplied equivalent circuit that is valid under these conditions.
(e) Find an approximate expression for the impedance valid when !
1
p
LC
and
L
R
2
C
1. Draw a simplied equivalent circuit that is valid under these conditions.
5. Dene the eective capacitance, C
e
, of a realistic capacitor having nite lead in-
ductance to be the capacitance of the ideal capacitor that has the same reactance as
the real capacitor. C
e
will depend on frequency.
(a) Find an expression for the eective capacitance of a realistic capacitor. Express
your result in terms of the capacitance, C, and the resonant frequency, f
o
of the
real capacitor.
(b) What is the eective capacitance of a 300 pF capacitor with 20 nH of lead induc-
tance at 10 MHz and at 40 MHz?
6. Figure 3.12 shows the equivalent circuit for an inductor.
(a) Derive an exact expression for the self-resonant frequency of the inductor. The
self resonant frequency is dened to be the frequency (> 0) where the impedance
of the inductor is purely resistive. Express your result in terms of R, L, and C.
When is
1
2
p
LC
a good approximation to f
p
?
(b) Consider a 500 nH inductor with series resistance R = 1 and shunt capacitance
C = 1 pF. Find the self-resonant frequency of the inductor, f
p
. Express your
result in MHz.
3.3. HOMEWORK PROBLEMS 113
(c) For the inductor specied in part (b), nd the impedance of the inductor at f
p
.
(d) For the inductor specied in part (b), nd the eective inductance, L
e
, of the
inductor at 0.75 f
p
. L
e
is dened to be X
L
(!)/!, where X
L
(!) is the reactance
of the inductor at frequency !. Express your result in nH.
114 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
Chapter 4
RLC Networks, Resonance, and Q
4.1 Series RLC Network
Consider the series RLC circuit in a lter conguration where the output voltage is taken
across the resistor, as shown in Figure 4.1. The voltage transfer function is
L
R
C
V
in
V
out
+
-
Figure 4.1: Series RLC circuit as a lter.
H(s) =
V
out
(s)
V
in
(s)
=
R
R +sL +
1
sC
(4.1)
We will consider sinusoidal excitation under steady-state conditions, in which case we are
interested in the frequency response, H(j!):
H(j!) =
R
R +j!L
1
1
!
2
LC
(4.2)
When ! = 1/
p
LC, the phase shift of the transfer function is zero; this is called the resonant
frequency, !
o
, of the network and is the frequency at which the inductive and capacitive
reactances are exactly equal in magnitude and, consequently, cancel each other:
!
o
=
1
p
LC
(4.3)
The transfer function depends on R, L, and C, but only two parameters are necessary to
specify the characteristics of the function. Dene another quantity Q
s
where:
Q
s
=
!
o
L
R
=
1
R
L
C
(4.4)
115
116 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
The frequency response function can be rewritten in terms of only !
o
and Q
s
:
H(j!) =
1
1 +jQ
s
!
!o
!o
!
(4.5)
The parameter Q
s
is referred to as the series resonant circuit Q. In a subsequent section
it will be shown that the inverse of this quantity tells us what fraction of the total energy
stored in the RLC circuit is dissipated in one complete cycle of the resonant frequency.
The magnitude and phase of the voltage transfer function (Equation 4.5) are plotted as
a function of !/!
o
in Figure 4.2(a) and (b) for Q=2 and Q=10. The same data is plotted
with a logarithmic frequency axis in Figure 4.3(a) and (b).
(a)
0.0
0.2
0.4
0.6
0.8
1.0
1 3 5 7 9
|H(j!)|
!/!
o
Q=2
Q=10
(b)
90
70
50
30
10
10
30
50
70
90
1 3 5 7 9
arg[H(j!)]
!/!
o
Q=2
Q=10
Figure 4.2: (a) Magnitude and (b) phase of the voltage frequency response for Q=2 (solid
line) and Q=10 (dotted line).
4.1. SERIES RLC NETWORK 117
(a)
0.0
0.5
1.0
0.1 1.0 10.0
|H(j!)|
!/!
o
Q=2
Q=10
(b)
90
70
50
30
10
10
30
50
70
90
0.1 1.0 10.0
arg[H(j!)]
!/!
o
Q=2
Q=10
Figure 4.3: (a) Magnitude and (b) phase of the voltage frequency response for Q=2 (solid)
and Q=10 (dotted) vs. !/!
o
with a logarithmic frequency axis.
118 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
The RLC lter has a bandpass characteristic. The separation between the half-power
(-3 dB) frequencies is often used to specify the bandwidth of a lter. The 3 dB bandwidth
of the lter can be found by determining the dierence between the frequencies where
|H(j!)| = 0.707. Denote the lower and upper -3 dB frequencies by !
1
and !
2
as shown in
Figure 4.4. Then ! = (!
2
!
1
) is referred to as the 3 dB bandwidth of the lter. It is
left as an exercise to show that
! = (!
2
!
1
) =
!
o
Q
s
(4.6)
The bandwidth of a series RLC lter is inversely proportional to the Q
s
of the circuit.
1.0
0.707
1 0 2
Figure 4.4: 3dB bandwidth of lter
4.1.1 Example - Series RLC circuit as a lter.
Use a series RLC circuit to couple a voltage source with negligible source resistance to a 50
load as shown in Figure 4.5. The circuit should have a center frequency of 5 MHz and a
3 dB bandwidth of 100 kHz. The bandwidth and center frequency determine Q
s
:
L
50
C
Figure 4.5: Circuit with 5 MHz center frequency and a 3 dB bandwidth of 100 kHz
Q
s
=
f
o
f
=
5 MHz
100 kHz
= 50 =
!
o
L
R
=
2(5 10
6
)L
50
(4.7)
so
L =
50 50
2(5 10
6
)
= 79.6 H
C =
1
!
2
o
L
= 12.7 pF
4.2. PARALLEL RLC 119
L R
p C I
in
V
out
+
-
Figure 4.6: Parallel RLC as a lter
4.2 Parallel RLC
A parallel RLC circuit being driven by an ideal current source is shown in Figure 4.6. In
this application the input current and output voltage are related by an impedance function,
i.e.,
V
out
(s)
I
in
(s)
= Z(s) (impedance) (4.8)
Z(s) =
1
R
p
+
1
sL
+sC
1
For sinusoidal steady-state excitation
Z(j!) =
R
p
1 +jQ
p
!
!o
!o
!
(4.9)
where
!
o
=
1
p
LC
(4.10)
Q
p
=
R
p
!
0
L
(4.11)
=
C
L
R
p
This transfer function has exactly the same form as that of the series RLC circuit except
for the scaling factor, R
p
. Note, however, that the Q is dened dierently for the parallel
RLC. As before, the 3dB bandwidth is
! =
!
o
Q
p
(4.12)
4.2.1 Unloaded vs Loaded Q of RLC circuits
If the source has a non-negligible impedance as shown in Figure 4.7 then
V
out
I
in
=
R
p
kR
S
1 +jQ
0
p
(
!
!o
!o
!
)
(4.13)
120 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
L R
p C I
in
V
out
+
-
R
S
Figure 4.7: Parallel RLC lter driven by a source with nite source impedance.
where
Q
0
p
=
R
p
kR
S
!
o
L
(4.14)
Compared to the case with innite source impedance, the nite source impedance causes the
Q to be reduced and, hence, the bandwidth to be increased. It is common practice to call
the Q of the resonant circuit alone (either series or parallel RLC) the unloaded Q, and the
Q of the composite circuit, which includes the source resistance and any other resistances
that are external to the LC resonator, the loaded Q. The loaded Q is always smaller than
the unloaded Q.
In the case of a series resonant circuit with resistance R driven by a source with nite
impedance R
S
, the loaded Q becomes
Q
0
s
=
!
o
L
R
S
+R
.
Again, the loaded Q is smaller than the unloaded Q.
4.3 More on Q
The previously dened Qs (Q
s
, Q
p
) describe the frequency selectivity of the simplest type
of resonant RLC networks. The term Q has a more general interpretation which can be
applied to any type of system that contains energy storage elements and dissipation. The
general denition of Q for a system is
Q = 2
Maximum instantaneous stored energy
Energy dissipated per cycle
(4.15)
= 2f
Maximum instantaneous stored energy
Time average power dissipated
This denition can be applied to resonant and nonresonant circuits. If this energy-
based denition is applied to resonant second-order RLC circuits the result is compatible
with the (Q
s
, Q
p
) dened in the previous section. In principle, the denition could be
applied to higher-order circuits, e.g, circuits with more than one inductor and capacitor,
however this is usually not very useful. On the other hand, higher order circuits are often
constructed by combining second-order circuits. In such cases, it is useful to characterize
each of the constituent second-order resonant circuits by a Q. This is common practice
in lter-design, where multiple LC resonators are coupled to form a more complex lter.
The energy denition is also commonly applied to characterize lossy inductors or capacitors
which, by themselves, are non-resonant.
4.3. MORE ON Q 121
We shall rst show that the energy denition is consistent with the resonant-circuit
Q
s
that has already been dened for series RLC circuits. Consider the series RLC with
sinusoidal excitation as shown in Figure 4.8.
L
R
C
V
in
V
out
+
-
I
in
Figure 4.8: Series RLC circuit for calculation of Q
At resonance,
V
out
= V
in
(4.16)
I
in
= V
in
/R (4.17)
The fact that the voltage across the capacitor is 90 degrees out of phase with the current
through the inductor means that the current maximizes at the time when the capacitor
voltage is zero. At that instant in time, all of the stored energy resides in the inductor and
the magnitude of the current phasor (which is the peak current magnitude) can be used to
calculate the total stored energy. The stored energy is:
E
max
=
1
2
L|I
in
|
2
=
1
2
L
|V
in
|
2
R
2
(4.18)
Alternatively, at the time instant when the capacitor voltage is maximum then the current
in the system is zero and all of the stored energy resides in the capacitor. The magnitude of
the capacitor voltage phasor can then be used to calculate the stored energy at that time:
E
max
=
1
2
C|V
cap
|
2
=
1
2
C|
I
in
!
o
C
|
2
=
1
2
L|I
in
|
2
(4.19)
The stored energy comes out the same either way. Actually, it can be shown that the
total stored energy in this driven resonant RLC circuit is a constant, so that the maximum
instantaneous stored energy is equal to the energy stored at any instant of time. The
time-averaged power delivered to (and dissipated in) the network is:
P
avg
=
1
2
Re[V
in
I
in
] (4.20)
=
1
2
|V
in
|
2
/R
Then using the energy denition of Q:
Q
s
= 2f
o
E
max
P
avg
=
!
o
L
R
(4.21)
This result is the same as the denition for the series resonant Q that was given earlier.
122 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
L
R V
in
I
in
Figure 4.9: Nonresonant RL circuit.
It is also common to apply the denition to a nonresonant circuit, e.g., consider an RL
circuit as in Figure 4.9.
I
in
=
V
in
R +j!L
(4.22)
In this case, the inductor is the only energy storage element. Because the current through the
inductor varies sinusoidally, the stored energy will oscillate between zero and the maximum
value given by:
E
max
=
1
2
L|I
in
|
2
(4.23)
=
1
2
L|V
in
|
2
1
r
2
s
+ !
2
L
2
The time-averaged power delivered is
P
avg
=
1
2
Re[V
in
I
in
] (4.24)
=
1
2
|V
in
|
2
Re
1
r
s
j!L
=
1
2
|V
in
|
2
r
s
r
2
s
+ !
2
L
2
Therefore
Q
L
= 2f
E
max
P
avg
=
!L
r
s
(4.25)
In this case ! can be any frequency. This type of Q will be called the component Q. In this
example, the RL circuit could represent a model for a lossy inductor. The component Q of
an inductor evaluated at some frequency ! can be thought of as the resonant Q
s
or Q
p
that
results if a lossless capacitor is added to form a resonant circuit at the frequency !.
The concept of component Q is often used to describe the properties of arbitrary circuit
elements at a particular frequency. For example, if an arbitrary circuit element can be
represented by a series impedance Z = R
s
+ jX
s
at some frequency, then applying the
denition of Q to that component yields Q = |X
s
|/R
s
as illustrated in Figure 4.10(a). For
a parallel representation of a circuit branch the component Q is as shown in Figure 4.10(b).
4.4. SERIES-TO-PARALLEL TRANSFORMATIONS 123
(a)
R
S
Q =
|XS|
RS
jX
S
(b)
R
P
Q =
RP
|XP |
jX
P
Figure 4.10: Denition of Q applied to a branch represented in (a) series form and (b)
parallel form.
4.4 Series-to-Parallel Transformations
Any circuit element has both a series and a parallel representation. Since the energy storage
and dissipation properties of the element do not depend on how we represent it, the Q is
the same for either representation. The component Q concept is useful for series-to-parallel
impedance transformations. For example, suppose the series impedance representation for
a circuit element is known at a particular frequency as shown in Figure 4.11(a).
(a)
R
S
Q =
|XS|
RS
jX
S
(b)
R
P
jX
P
Z = R
S
+jX
S
Figure 4.11: (a) Series impedance representation for a circuit at a particular frequency
and (b) equivalent parallel representation having the same impedance as the original series
branch.
The circuit Q is given by
Q =
|X
s
|
R
s
(4.26)
The equivalent parallel representation for the circuit element is shown in Figure 4.11(b). The
equivalent parallel resistance and reactance are easily found after equating the impedances
of the two models:
R
p
= R
s
(1 + Q
2
) (4.27)
X
p
= X
s
1 +
1
Q
2
(c)
50
1+
50
2
|X|
2
j100
jX
1+
|X|
2
50
2
50
1+
50
2
|X|
2
jX
1+
|X|
2
50
2
jX
C
(d)
50
50
j100
j16.7 j16.7
j70
V
s
V
o
+
Figure 4.19: (a) With 50 source and load resistances and inductor reactance 100 , Q
s
= 1,
and the bandwidth cannot be specied independently. (b) Shunt capacitors with reactance
X (X < 0) have been added to transform the source and load resistances to a smaller value
as shown in (c). The values given in (d) set the Q
s
of the equivalent series RLC circuit
(shown in (c)) to Q
s
= 10.
Figure 4.19(b), shunt capacitors with impedance jX (X < 0) have been added to transform
the source and load resistance to smaller values, as shown in Figure 4.19(c). The network
shown in Figure 4.19(c) is similar to a series RLC lter and will have a similar frequency
128 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
response function. The resonant Q of this network can be approximated by Q
s
= 100/(2R
0
),
where R
0
= 50/(1 + (50/|X|)
2
). Hence, X can be chosen to set Q
s
to any desired value.
Suppose that the desired fractional bandwidth of the lter is f/f
o
= 0.1 (10%). Then the
required loaded Q
s
=
1
0.1
= 10 and the total series resistance should be set to 10 , i.e.
10 =
2(50)
1 +
50
2
|X|
2
,
which has the solution |X| = 50/3 = 16.7 . Therefore, the impedance of the shunt coupling
capacitors should be set to j16.7 . The shunt capacitors are transformed to series reac-
tances of j
16.7
1+(16.7/50)
2
= j15 . At the resonant frequency, !
o
, the total series capacitive
reactance must be equal to 100 in order to resonate with the inductor, so the capacitive
reactance X
C
= 70 . The nal design is shown in Figure 4.19(d).
If the desired center frequency is f
o
= 500 MHz, then the values of the inductor, shunt
capacitors, and series capacitor are L = 31.8 nH, C
shunt
= 19.1 pF, and C = 4.55 pF,
respectively. The frequency response of the network with these values is shown by the solid
line in Figure 4.20. For comparison, the dashed line shows the frequency response of a
series RLC lter (i.e., the topology shown in 4.19(a)) with Q
s
= 10. The series RLC lter
would need an inductor and capacitor each having reactance 1000 to produce a lter with
Q
s
= 10. Notice that the two curves correspond closely near the peak of the response,
illustrating the validity of approximating the network shown in Figure 4.19(c) as a simple
series RLC network near the resonant frequency of the network.
0 200 400 600 800 1000
0.1
0.2
0.3
0.4
0.5
Frequency (MHz)
|
V
o
V
s
|
Figure 4.20: Solid line shows the frequency response of the lter shown in Figure 4.19(d),
designed for a center frequency of 500 MHz. The dashed line is a plot of the function
|
0.5
1+jQs(!/!o!o/!)
|. The dashed line is the frequency response for a series RLC lter, as
shown in Figure 4.19(a), but with the inductor and capacitor impedances equal to j1000
at f
o
= 500 MHz. Close correspondence between the solid and dashed lines near the resonant
frequency illustrates the validity of approximating the network shown in Figure 4.19(b)-(d)
as a series RLC network near f
o
.
4.5. APPLICATION EXAMPLE - QUADRATURE DEMODULATOR FOR FM 129
4.5 Application Example - Quadrature demodulator for
FM
A common application of the parallel RLC resonant circuit is found in the so-called quadra-
ture demodulator for frequency modulated signals. This type of detector is implemented
in a number of receiver-on-a-chip integrated circuits. The block diagram of a quadrature
demodulator is shown in Figure 4.21. The parts within the dashed box are implemented
on-chip. Typically, the multiplier is implemented with a Gilbert cell and the capacitance of
C
0
is a relatively small value, which is easily implemented in an integrated circuit. The res-
onant RLC network is implemented o-chip, and is often sold under the name quadrature
coil.
C
L C R
v
in
v
out
v
Q
p
=
o
RC
o
=(LC)
-1/2
2f
o
Low-pass filter
Figure 4.21: Quadrature demodulator for frequency modulated signals. The Q
p
of the
parallel resonant circuit determines the sensitivity and the bandwidth of the detector.
The principle of operation behind the quadrature demodulator is as follows. The fre-
quency modulated input signal is typically provided by the output of the last IF stage of
a superhet receiver. Usually, the signal will have passed through a limiter, so that the
amplitude is constant. The input voltage can be written as:
v
in
(t) = Acos[(!
o
+!)t] (4.32)
where ! represents the instantaneous frequency deviation from the carrier frequency, !
o
.
The deviation will be a function of time, in general, but since time variations of ! are very
slow compared to the period of !
o
, we will employ a quasi-static analysis and treat ! as
a (slowly varying) constant. In most applications, the carrier frequency will be equal to the
last IF. The RLC circuit is tuned to resonance at !
o
. Thus, at !
o
the impedance of the RLC
circuit is equal to R. If the on-chip capacitance, C
0
, is small so that
1
!oC
0
R, then the
voltage v
0
will be shifted in phase by /2 relative to v
in
. Thus v
0
and v
in
are in quadrature
when the input frequency is equal to !
o
. When the frequency of the input signal deviates
from !
o
, the RLC circuit looks capacitive for positive deviations, and inductive for negative
deviations. This upsets the nominal quadrature relationship between v
in
and v
0
and, for
small frequency deviations, causes the phase dierence between v
in
and v
0
to vary around
the nominal value of /2 in proportion to the instantaneous frequency deviation, !. The
multiplier/low-pass lter functions as a phase-detector and provides an output voltage that
is proportional to the o-quadrature phase dierence between v
in
and v
0
. Therefore, for
small frequency deviations, the output voltage is proportional to the frequency deviation.
130 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
For analytical analysis of the quadrature demodulator circuit, it is convenient to assume
that the input impedance of the multiplier circuit is high, so that loading of the RLC circuit
can be neglected. (Alternatively, one can assume that the nite input resistance of the
multiplier has been lumped into R.) Then, using upper-case symbols to denote phasors, the
voltage V
0
can be written in terms of the phasor input voltage, V
in
as:
V
0
=
Z
p
Z
p
+
1
j!C
0
V
in
(4.33)
where
Z
p
=
R
1 +jQ
p
(
!
!o
!o
!
)
(4.34)
So
V
0
=
R
R +
1
j!C
0
(1 +jQ
p
(
!
!o
!o
!
))
V
in
(4.35)
If C
0
is small enough so that R
1
!C
0
then
V
0
'
j!RC
0
1 +jQ
p
(
!
!o
!o
!
)
V
in
(4.36)
The transfer function can be re-written in terms of magnitude and phase as
V
0
= B(!)e
j(!)
V
in
where
B(!) =
!RC
0
1 +Q
2
p
(
!
!o
!o
!
)
2
(4.37)
and
(!) = /2 tan
1
[Q
p
(
!
!
o
!
o
!
)] (4.38)
Thus, the time-domain voltage, v
0
(t) is given by (with ! = !
o
+!):
v
0
(t) = AB(!) cos[!t + (!)] (4.39)
The multiplier forms the product v
in
v
0
, so the signal at the output of the multiplier is:
v
in
(t)v
0
(t) = A
2
B(!) cos[!t] cos[!t + (!)] (4.40)
Taking the nominal phase-shift, /2, in the expression for (!) into account, we can write
v
in
(t)v
0
(t) = A
2
B(!) cos[!t] sin[!t tan
1
[Q
p
(
!
!
o
!
o
!
)] (4.41)
The sine-cosine product yields a double frequency term, and a dierence frequency term.
The lowpass lter removes the double frequency term, so the output of the system can be
written:
v
o
= LPF[v
in
(t)v
0
(t)] =
1
2
A
2
B(!) sin[tan
1
(Q
p
(
!
!
o
!
o
!
))] (4.42)
4.5. APPLICATION EXAMPLE - QUADRATURE DEMODULATOR FOR FM 131
Employing the trigonometric identity sin[tan
1
x] =
x
p
1+x
2
the output voltage can be writ-
ten as
v
o
=
1
2
A
2
B(!)
Q
p
(
!
!o
!o
!
)
1 +Q
2
p
(
!
!o
!o
!
)
2
. (4.43)
Inserting the expression for B(!), we nd:
v
o
=
1
2
A
2
!RC
0
Q
p
(
!
!o
!o
!
)
1 +Q
2
p
(
!
!o
!o
!
)
2
. (4.44)
Now, with ! = !
o
+!
!
!
o
!
o
!
=
!
o
+!
!
o
!
o
!
o
+!
=
2!
!
o
+!
+
!
2
!
o
(!
o
+!)
. (4.45)
For small frequency deviations such that ! !
o
:
!
!
o
!
o
!
'
2!
!
o
. (4.46)
So, for small frequency deviations we have
v
o
'
1
2
A
2
!
o
RC
0
Q
p
2!
!o
1 +Q
2
p
4!
2
!
2
o
=
1
2
A
2
Q
p
C
0
C
Q
p
2!
!o
1 +Q
2
p
4!
2
!
2
o
. (4.47)
This expression has the form:
v
o
x
1 +x
2
where x = 2Q
p
f/f
o
. This function is plotted in Figure 4.22.
Some quadrature detector implementations use a multiplier with limiting inputs, i.e.
one or both of the inputs is driven into saturation so that the multiplier output is insensitive
to changes in the amplitude of the input signals. If the v
0
input is saturated, then the output
becomes independent of the amplitude of v
0
. In this case, equation 4.40 must be modied
by replacing the term A
2
B(!) with a constant. In this case, the output can be written:
v
o
= K
Q
p
(
!
!o
!o
!
)
1 +Q
2
p
(
!
!o
!o
!
)
2
. (4.48)
Using the same approximations that were used before, we obtain:
v
o
' K
Q
p
2!
!o
1 +Q
2
p
4!
2
!
2
o
(4.49)
which has the form:
v
o
x
p
1 +x
2
with x = 2Q
p
f/f
o
. This function is plotted in Figure 4.23.
132 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
1.0
0.8
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
1.0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
x
1+x
2
x =
2Qpf
fo
Figure 4.22: Detector output as a function of frequency deviation. This curve is referred to
as the demodulators S-curve. Notice that the region where the output is approximately
proportional to the frequency deviation is limited |2Q
p
f/f
o
| 1. The dotted line shows
an ideal linear response with the same slope as the actual response at x = 0.
1.0
0.8
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
1.0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
x
p
1+x
2
x =
2Qpf
fo
Figure 4.23: Output voltage verses normalized frequency deviation for a quadrature demod-
ulator employing a saturated input for v
0
. The dotted line shows an ideal linear response
with the same slope as the actual response at x = 0.
4.6. REFERENCES 133
In either case (ideal multiplier, or multiplier with saturated inputs), if the frequency
deviation is small enough, i.e. if 2Q
p
!/!
o
1, then the output voltage is proportional
to the instantaneous frequency deviation:
v
o
Q
p
f
o
f, f f
o
and 2Q
p
f f
o
(4.50)
The coecient in front of f determines the sensitivity of the demodulator. It is apparent
that to obtain larger output voltage for a given deviation one would like to choose a higher
Q
p
for the resonant RLC circuit. On the other hand, the maximum Q
p
that can be employed
is limited by the requirement that 2Q
p
f
max
/f
o
1 (where f
max
is the peak frequency
deviation of the modulated input signal). Recall that the -3 dB bandwidth of an RLC lter
can be written in terms of the Q, i.e. BW = f
o
/Q
p
, so the constraint on Q
p
can be rewritten
in terms of a constraint on the bandwidth of the RLC circuit: BW 2f
max
. In other
words, the -3 dB bandwidth of the RLC circuit should be large compared to the deviation
of the input signal in order for the demodulator to have a linear response.
A useful practical constraint for design purposes is to require that the demodulators
response characteristic should not deviate from perfect linearity by more than 1% over
the whole range of deviation, i.e. for f
max
< f < f
max
. If the demodulator is
implemented with an ideal multiplier, this requirement leads to x < 0.100. The largest Q
p
that will satisfy this constraint is:
Q
p
= 0.1
f
o
2f
max
(4.51)
If a multiplier with saturated inputs is used to implement the detector, then the departure
from linearity will be held to the 1% limit if x < 0.142. The largest Q
p
that satises this
constraint is:
Q
p
= 0.142
f
o
2f
max
(4.52)
The extent to which the demodulators characteristic departs from linearity will determine
the amount of distortion in the demodulated signals waveform. For some applications,
such as demodulation of frequency-shift-keyed (FSK) data signals, low distortion is not
particularly important. In such cases a larger deviation from linearity can be tolerated in
return for larger output voltage from the detector.
4.6 References
1. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw Hill, 1943.
134 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
4.7 Homework Problems
1. The circuit shown in Figure 4.24 is usually a good model for a parallel LC circuit
implemented with real, lossy components. The element values are C = 800 pF, L =
15 H, r = 1 , R = 10 k. Use series to parallel transformations to transform this
circuit into an equivalent parallel RLC circuit and nd the approximate resonant
frequency and Q
p
of the equivalent circuit.
r
L
C R
Figure 4.24: Resonant circuit formed from lossy capacitor (modeled by R and C) and a
lossy inductor (modeled by r and L).
2. Two series RLC circuits are connected in series to form a new resonant circuit. Denote
the elements of the individual resonant circuits by (R
1
, L
1
, C
1
) and (R
2
, L
2
, C
2
),
respectively. The series resonant frequency of each of the two circuits is the same and
is denoted by !
o
. The Qs of the two circuits are Q
1
and Q
2
.
(a) What is the resonant frequency of the series combination of the two circuits?
(b) Find an expression for the Q of the overall circuit. Express your result in terms
of Q
1
, Q
2
, R
1
and R
2
only.
(c) Is the statement below TRUE or FALSE? You must justify your answer to receive
full credit.
min(Q
1
, Q
2
) Q max(Q
1
, Q
2
) (4.53)
3. One property of a parallel resonant circuit is that the current in the reactive compo-
nents can be much larger than the applied current. This must be considered when
choosing the current ratings of components such as capacitors and inductors (espe-
cially in high power devices such as transmitters). Show that the peak current through
the inductor or capacitor in a parallel resonant circuit is Q
p
I
i
at resonance, where I
i
is the current supplied to the entire resonant circuit
4. One property of a series resonant circuit is that the voltage across the reactive com-
ponents can be much larger than the applied voltage. This must be considered when
choosing the voltage ratings of components such as capacitors and inductors (espe-
cially in high power devices such as transmitters). Show that the peak voltage across
the capacitor in a series resonant circuit is Q
s
V
i
at resonance where V
i
is the voltage
across the entire resonant circuit. Note that the peak voltage across the inductor will
be the same.
4.7. HOMEWORK PROBLEMS 135
5. Consider the circuit shown in Figure 4.25. The current source has constant amplitude
and frequency f
c
, and it drives a bandpass lter consisting of a lossy inductor in par-
allel with a variable capacitor and a resistor. You may assume that any capacitance
associated with the inductor has been incorporated into the variable capacitor indi-
cated in the schematic. The variable capacitor C can be set to any value in the range
36 365 pF, and r = 10 , R = 100 k. Suppose that the frequency of the current
r
L
C R
I
o
cos 2f
c
t
V
out
Figure 4.25: LC lter with variable center frequency.
source can be adjusted to any frequency in the range 540-1700 kHz. For a given value
of the source frequency, f
c
, the variable capacitor will be tuned to maximize the out-
put voltage. Approximate the lter as a parallel RLC circuit to answer the following
questions:
(a) Specify a single value of L that would allow the variable capacitor to tune the
lter (i.e. maximize the output voltage) to any frequency in the range 540-1700
kHz. Give your result in H.
(b) With the value of L that was determined in part (a), determine the approximate
3dB bandwidth of the lter when f
c
= 540 kHz and C is adjusted to maximize
the output voltage.
(c) Same as part (b) but for f
c
= 1700 kHz.
6. The circuit in Figure 4.26 is a model for the operation of a ferrite loopstick antenna
that is commonly employed in AM broadcast band radios. The antenna is tuned to
resonance by a capacitor C which is usually adjustable to allow the circuit to cover
the entire broadcast band. This circuit also performs the function of the preselector.
The voltage source V
s
represents the emf induced in the coil as a result of an incident
electromagnetic wave with frequency !. The resistance r represents the losses in the
coil and R represents the input impedance of the following stage.
(a) Find an exact expression in terms of r, R, L, and C for the frequency where the
output voltage is a maximum (the resonance frequency).
(b) Suppose that V
s
= (1 mV ) cos[ 2 f
c
t ], L = 100H, R = 100k , and r = 6 .
Find the range of values that C must cover in order for the circuit to tune the AM
broadcast band (540-1700 kHz). Note: For this purpose you need an expression
for the value of C that maximizes the voltage response at a given frequency. An
approximate analysis is acceptable, but be sure to carefully state and justify your
assumptions.
(c) Now suppose that the source frequency, f
s
, is swept from 540 to 1700 kHz. The
zero-to-peak value of the source voltage is held constant at 1 mV as the frequency
136 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
r
L
C R
V
s
V
out
Figure 4.26: Ferrite loopstick antenna model.
is swept. Also suppose that the circuit is tuned to follow the frequency of the
source so that the output voltage is always maximized. Plot the output voltage
as a function of frequency.
7. Consider a parallel resonant circuit that is used as the preselector of an AM broadcast
band receiver. The inductor used in the resonant circuit is near the peak of its Q-
versus-frequency curve in the band of frequencies from 550-1600 kHz, and hence
the Q
L
of the inductor can be assumed to be constant over the band. Assume that
Q
L
= 100 and that the capacitor is lossless.
(a) What will the 3 dB bandwidth of the preselector be when the preselector is tuned
to 550 kHz? How about 1600 kHz?
(b) Now consider a dierent situation where the inductor is chosen such that the skin
eect and distributed capacitance can be neglected for the frequencies of interest,
i.e., the inductor is operated at frequencies well below the peak of its Q-versus-
frequency curve. Suppose the inductor has Q
L
= 50 at 550 kHz. What will the
3 dB bandwidth of the preselector be at 550 and 1600 kHz?
8. In a series circuit that is resonant at 1150 kHz it is found that when the frequency
diers from resonance by 15 kHz, the current drops to 0.53 of the current at resonance,
for the same applied voltage. Determine the Q
S
of the circuit.
9. Consider the circuit shown in Figure 4.27 where R
S
= R
L
= 50 , L = 48 H, C = 10
pF:
+
-
V
s
V
o
L C
R
L
R
S
Figure 4.27: Series resonant circuit
(a) Find the resonant frequency where maximum power would be delivered to the 50
load resistor. Express your result in MHz.
4.7. HOMEWORK PROBLEMS 137
(b) Suppose that the time-averaged power delivered to the 50 load resistor at
resonance is 1000 W. Find the peak value of the voltage across the capacitor.
(c) Approximately how far o of resonance would the source frequency need to be
moved in order to cause the average power delivered to the load resistor to drop
by 6 dB? Express your result in kHz.
Note: For small frequency shifts, !, around the resonant frequency, !
o
, the following
approximation is useful:
!
o
+ !
!
o
!
o
!
o
+ !
2
!
!
o
(4.54)
10. Consider the circuit shown in Figure 4.28 and its Thevenin equivalent circuit.
V
s
L C
r
R V
th
R
th
Figure 4.28: Series resonant circuit with Thevenin equivalent circuit.
(a) If V
s
= 2 V, R = 1000 , C = 1000 pF, L = 10 H, r = 1 , nd the (non-zero)
frequency, !
o
, where the indicated Thevenin equivalent circuit would be valid. In
other words, nd the frequency where the Thevenin impedance is purely resistive.
Clearly state any assumptions or approximations that you make.
(b) Find R
th
and V
th
at the frequency found in part (a). For V
th
, specify magnitude
and phase (in degrees).
11. Shown in Figure 4.29 are two dierent ways of coupling a voltage source to a load
resistor using a resonant circuit for the purpose of providing a bandpass output voltage
response. The values are: V
s
= 10V (peak), R
s
= 500 , R
L
= 500 , L = 1 H, C =
500 pF.
Compute the following for each circuit:
(a) The output voltage at resonance.
(b) The 3dB bandwidth of the transfer function V
o
/V
s
. Express your result in MHz.
(c) The upper and lower 3 dB frequencies, i.e., give the actual frequencies (in MHz).
The dierence between these frequencies will be the 3dB bandwidth that you
found in part 11b.
12. Consider the capacitive transformer drawn in Figure 4.30.
Under certain conditions this circuit will approximately transform the resistance R to
a new resistance n
2
R in parallel with an eective capacitance, C
p
, where n = 1 +
C1
C2
.
138 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
(a)
+
-
V
s
V
o
L C R
L
R
S
(b)
+
-
V
s
V
o
L C
R
L
R
S
Figure 4.29: Voltage source coupled to load resistor with (a) parallel LC circuit and (b)
series resonant circuit to provide a bandpass voltage transfer function V
o
/V
s
.
R
Z
C
2
C
1
Figure 4.30: Capacitive transformer
4.7. HOMEWORK PROBLEMS 139
(a) Under what condition(s) will this occur?
(b) Under this condition, give an expression for C
p
.
13. The circuit shown in Figure 4.31 is a capacitive transformer with resonating inductance
L. Suppose that
R = 50
C
1
= 3183 pF
C
2
= 3183 pF
R
Z
C
2
C
1
L
Figure 4.31: Resonant capacitive transformer
Use parallel-series and series-parallel transformations, with appropriate approxima-
tions, and nd:
(a) The inductance, L, required to resonate the circuit at 10 MHz.
(b) The input impedance, Z at 10 MHz.
(c) The Q of the circuit can be approximated by the Q
p
of the equivalent parallel
RLC circuit (at 10 MHz). Find the approximate Q.
To save time, note that the reactances of C
1
and C
2
have magnitude 5 at 10 MHz.
14. Consider in Figure 4.32 the small signal model for a tuned-output amplier:
+
-
V
in
g
m
V
in
C L R
+
-
V
out
Figure 4.32: Small signal model for tuned-output amplier.
Suppose that N of these stages are cascaded. Show that the 3dB bandwidth of the
cascade will be given by
140 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
BW = BW
1
2
1/N
1 (Hz) (4.55)
where BW
1
is the 3 dB bandwidth of a single stage
BW
1
=
f
0
Q
(4.56)
=
1
2RC
15. Consider the circuit shown in Figure 4.33. Dene the following quantities:
C
0
C L R V
out
V
in
Figure 4.33: Phase shifter based on parallel RLC circuit.
!
o
=
1
p
LC
and Q
p
= !
o
RC (4.57)
(a) Find an expression for the phase shift between V
out
and V
in
, i.e., nd an expres-
sion for the phase angle of the transfer function T(!) = V
out
(!)/V
in
(!).
(b) Show that if the reactance of the capacitor C
0
is much larger than R, then the
expression for the phase shift reduces to
= /2 tan
1
[Q
p
(
!
!
o
!
o
!
)] (4.58)
(c) Show that the phase shift is approximately a linear function of frequency for
frequencies near the resonant frequency of the tuned circuit.
16. A 100 pF capacitor and 2.53 H inductor are used to form a series resonant circuit
at 10 MHz. The capacitor is lossy and has a component Q of 100 at 10 MHz. The
inductor is also lossy and has a component Q of 80 at 10 MHz.
(a) Find the Q
s
of the resonant circuit formed by the series combination of the lossy
capacitor and inductor.
(b) Find the impedance of the series resonant circuit at 10 MHz.
(c) Suppose that it is desired to couple a 10 source to a 10 load through a
bandpass lter formed either by:
i. a series resonant circuit or
ii. a parallel resonant circuit.
4.7. HOMEWORK PROBLEMS 141
In either case, the resonant circuit would be formed by the lossy L and C con-
sidered in parts 16a and 16b. Sketch a schematic showing how the source should
be coupled to the load if the smallest possible bandwidth is desired.
(d) Estimate the bandwidth (in MHz) of the lter that you sketched in part 16c.
17. A series LC circuit consisting of an ideal lossless inductor L and ideal lossless capacitor
C is used as a bandpass lter to couple a source with source impedance R
S
= 50
to a load R
L
= 50 . The resonant frequency of the lter is f
o
= 100 MHz, and the
values are selected such that the -3 dB bandwidth of the lter is 10 MHz.
(a) Sketch the system, including the source, the LC lter, and the load and indicate
the values for L and C in nH and pF, respectively.
(b) Now, replace the inductor and capacitor with lossy components that have the
same inductance and capacitance as the original components but with component
Qs that are equal to 40 at the resonant frequency, i.e. Q
C
= Q
L
= 40 at 100
MHz. Calculate the -3 dB bandwidth of the lter implemented using the lossy
components.
18. The circuit shown below is used to transform a load resistance, R
L
, to a new value,
R
in
, at the resonant frequency of the network. Away from resonance, the circuit
provides a bandpass lter response.
jX
L1
jX
L2
R
L
jX
C
Z
in
(a) The load resistance R
L
= 50 . At the resonant frequency assume that X
L1
=
X
L2
= 5 . Determine the value of X
C
required to resonate the circuit (i.e. to
make the input impedance Z
in
purely resistive).
(b) For the value of X
C
found in part a. determine Z
in
.
(c) The Q of the resonant circuit can be approximated by Q
p
of the equivalent parallel
RLC circuit obtained by applying parallel<=>series transformations. Find the
approximate Q of the network.
19. A 50 source is coupled to a 50 load through a bandpass lter consisting of a series
LC circuit which is resonant at 20 MHz.
(a) Sketch the system, including the source, the lter, and the load.
(b) Design the lter (i.e., nd L and C) so that the -3 dB bandwidth of the lter is
5 MHz.
142 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
(c) Now suppose the inductor is lossy and has nite component Q, denoted by Q
L
.
(You may assume that the inductance is the same as determined in part b.) What
is the minimum acceptable Q
L
if the attenuation caused by the lossy L at 20 MHz
is to be smaller than 2 dB?
20. Design an LC circuit consisting of an ideal lossless inductor L and ideal lossless ca-
pacitor C to be used as a bandpass lter to couple a source with source impedance
R
S
= 100 to a load R
L
= 100 . Assume that a limited range of inductance values
is available such that 10 nH L 100 nH.
(a) Determine the topology of the LC lter and the values for L and C if the lter
is to have center frequency f
0
= 100 MHz and the smallest possible bandwidth.
Sketch the system, including the source, the LC lter, and the load. Specify the
values for L and C in nH and pF, respectively.
(b) Repeat part a for the center frequency f
0
= 1 GHz.
21. An LC circuit consisting of an ideal lossless inductor L and ideal lossless capacitor C
is to be used as a bandpass lter to couple a source with source impedance R
S
to a
load R
L
. Assume that R
S
= R
L
= R. Denote the resonant frequency of the lter by
!
o
, and denote the reactance of the inductor and capacitor at !
o
by X
o
.
(a) If X
o
> R, sketch the system (source, load, and LC lter) that will provide the
smallest possible bandwidth.
(b) Consider the system that you sketched in part a., and denote the loaded Q of
the system by Q
o
. The -3 dB bandwidth of the system is then BW
o
=
!o
Qo
. Now,
suppose that the system is modied by replacing the lossless inductor with a
lossy inductor. Denote the component Q of the lossy inductor at !
o
by Q
L
. The
inductance of the lossy inductor is the same as the inductance of the original
lossless component. Denote the bandwidth of the modied circuit by BW
0
. Find
an expression for BW
0
that involves only BW
o
, Q
o
, and Q
L
.
(c) If Q
o
= 10 and Q
L
= 40, by what percentage is the bandwidth of the lter
increased because of the loss in the inductor?
(d) Let P
L
denote the power delivered to the load at !
o
with the lossless inductor
in the system, and let P
0
L
denote the power delivered to the load at !
o
with the
lossy inductor in the system. Find an expression for the power ratio, P
0
L
/P
L
.
Express your result only in terms of Q
o
and Q
L
.
(e) If Q
o
= 10 and Q
L
= 40, determine the power loss, in dB, of the lter at the
resonant frequency.
22. An LC circuit consisting of an ideal lossless inductor L and ideal lossless capacitor C
is to be used as a bandpass lter to couple a source with source impedance R
S
to a
load R
L
. Assume that R
S
= R
L
= R. Denote the resonant frequency of the lter by
!
o
, and denote the reactance of the inductor and capacitor at !
o
by X
o
.
(a) If X
o
< R, sketch the system (source, load, and LC lter) that will provide the
smallest possible bandwidth.
4.7. HOMEWORK PROBLEMS 143
(b) Consider the system sketched in part a., and denote the loaded Q of the system by
Q
o
. Now, suppose that the system is modied by replacing the lossless inductor
with a lossy inductor and the lossless capacitor with a lossy capacitor. Denote
the component Qs of the lossy inductor and capacitor at !
o
by Q
L
and Q
C
,
respectively. The inductance (capacitance) of the lossy inductor (capacitor) are
the same as the inductance and capacitance of the original lossless components.
Denote the new loaded Q of the system by Q
0
o
. Find an expression for Q
0
o
that
involves only Q
o
, Q
L
, and Q
C
. You may assume that Q
L
1 and Q
C
1.
23. Consider the single-resonator bandpass lter shown in Figure 4.34.
50
50
C
c
C
c
C L
Figure 4.34: Single-resonator bandpass lter.
(a) Suppose that the inductor has reactance X
L
= 50 at the center frequency of
the lter. Find the reactances of the coupling capacitors, X
Cc
= 1/(!
o
C
c
), and
the reactance of the resonator capacitor, X
C
= 1/(!
o
C), such that the lter has
a fractional -3 dB bandwidth of 10%. Hint: use series to parallel transformations
to convert the circuit to a parallel RLC circuit.
(b) Find L, C, C
c
that will place the center frequency of the lter at f
o
= 500 MHz.
(c) Plot the magnitude of the voltage frequency response V
o
/V
s
for the lter designed
in parts a. and b. Find the actual -3 dB bandwidth.
24. The system shown in Figure 4.35 consists of a resistive source and load coupled with a
bandpass lter. Design the lter to have fractional bandwidth
f
3dB
fo
= 0.1. The lter
V
s
100
jX
o
j50 jX
C
jX
o 100
Figure 4.35: Series-resonator bandpass lter.
uses a single inductor with reactance j50 at f
o
. The reactances of the capacitors
at f
o
are denoted by X
o
and X
C
. Find numerical values for X
o
and X
C
. Note that
X
C
< 0 and X
o
< 0. Plot the voltage frequency response.
144 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q
Chapter 5
Oscillators
5.1 Introduction
This chapter will describe the methods used to analyze and design oscillators for applicatons
in transmitters and receivers. We will employ small-signal linear analysis techniques to
determine whether or not a particular circuit will oscillate, and a simplied nonlinear model
will be used to estimate the amplitude of the oscillations in a circuit based on a BJT.
Numerical simulations will be used to illustrate the characteristics of realistic oscillator
circuits, and the results will be compared to the analytical predictions.
For small signal analysis it is convenient to think of oscillators as unstable feedback
systems. An unstable system is one in which an initially small excitation or disturbance
produces an output that grows in time due to constructive, or positive, feedback. A block
diagram of a simple feedback system is shown in Figure 5.1.
A
B
V
i
V
o
Figure 5.1: A simple feedback system
The voltage transfer function for this system is
V
o
V
i
=
A(j!)
1 A(j!)B(j!)
(5.1)
where the quantity A
lo
(!) = A(j!)B(j!) is called the open loop gain of the system
sometimes shortened to loop gain. The subscript o is used to indicate that the loop gain is
computed assuming small-signal operation of the active devices. Note carefully that the loop
gain is the gain obtained by opening the feedback loop and taking the output at the point
145
146 CHAPTER 5. OSCILLATORS
A
B
V
i
V
o
V
f
Figure 5.2: Loop gain at summing junction
where the loop was opened. For example, if the loop is opened at the summing junction,
then the loop gain is V
f
/V
i
as shown in Figure 5.2.
Suppose that the loop gain is equal to 1 at some frequency !
o
, i.e., A
lo
(!
o
) = 1. Then
the voltage transfer function in Figure 5.1 is singular (innite), which can be interpreted
as nite output for zero input. In other words, the circuit is a potential source of radio
frequency energy at the frequency where the loop gain is 1, even in the absence of any input
excitation V
i
.
The condition for steady-state oscillation (A
lo
(!
o
) = 1) is intuitively satisfying - when
the loop gain is 1, a sinusoidal excitation presented to the input of the circuit traverses the
feedback loop and appears back at the input with the same amplitude and phase that it
started with. This re-circulation of the disturbance proceeds indenitely, with the circuit
oscillating in a steady state. In practice the small-signal loop gain is set to a value
somewhat larger than 1. This means that the disturbance is amplied after each pass
through the loop, and the output grows as the disturbance passes repeatedly through the
loop. In most radio frequency oscillators the loop gain is equal to 1 (or some real number
larger than 1) at one particular frequency. At other frequencies the amplitude may be less
than or greater than 1, but the phase angle is non-zero. This means that only one frequency
component can travel around the loop with no phase shift. Only this frequency component
will be amplied and grow to become a steady-state oscillation.
We shall see later in this chapter that even when the loop gain is 1 at only one
frequency, it is possible to have a non-sinusoidal output. However, this is due to nonlinear
eects in the amplifying devices. As noted above, practical oscillator circuits are designed
so that the small-signal loop gain is larger than one at the desired frequency of oscillation,
!
o
. This ensures that even thermal-noise-level signals will provide enough excitation to
cause oscillation to start and grow at !
o
. Some mechanism must be built into the circuit
to limit the amplitude of the oscillation to a nite value. The loop gain must be decreased
as the oscillation amplitude grows so that the loop gain can eventually settle down to the
steady-state value of 1.0. In so-called self-limiting oscillators, the amplitude of the oscillation
eventually becomes large enough to begin to saturate the active device, eectively reducing
the gain of the device, and also reducing the loop gain. Oscillation amplitude stabilizes
at the amplitude where the gain of the active device is reduced just enough to set the
loop gain to 1.0. Since self-limiting oscillators rely on driving the active device to levels
where nonlinearity is important, they often produce non-sinusoidal outputs with signicant
harmonic content.
In some cases, it is desired to limit the amplitude of the oscillation at a level smaller than
that required to saturate the active device. One mechanism for doing this was implemented
5.2. OSCILLATOR ANALYSIS USING LOOP GAIN 147
by Hewlett and Packard in their rst commercial product, an audio oscillator. They used
a small light bulb as a resistor to set the voltage gain of the amplier in the oscillator.
As oscillation amplitude grows, the light bulb lament heats up, and the resistance of the
lament increases, reducing the gain of the amplier. The thermal time constant of the
lament was much longer than the period of the oscillation, so that the lament acts as a
linear resistor, with resistance proportional to the oscillation amplitude. This mechanism
allowed Hewlett and Packard to create an audio oscillator with a nearly perfect sinewave
output. The oscillator exhibited an extremely low level of harmonic distortion in the output
because the amplitude of oscillation was limited at a level well within the linear range of
the amplier.
In any case, some initial disturbance is necessary to start the oscillations. As long as
the small-signal loop gain is set equal to a value greater than 1 at the potential frequency
of oscillation, the thermal noise voltage that is always present in electrical circuits, or the
turn-on transient which results when power is applied to the circuit, will suce to excite
the oscillation.
5.2 Oscillator Analysis using Loop Gain
We will now proceed to analyze one type of oscillator circuit using the loop gain approach.
The analysis proceeds as follows. First the feedback loop must be identied and the loop
gain computed. Then the condition for oscillation is applied to the loop gain. Oscillation
occurs if there is some frequency where the loop gain has magnitude 1 and phase angle equal
to zero. The combination of these two constraints is called the Barkhausen Criterion for
oscillation. The two conditions can be written as
arg[A
lo
(!
o
)] = 0 (5.2)
|A
lo
(!)|
!=!o
1 (5.3)
In practice we are usually able to apply condition (5.2) to solve for the potential frequency
of oscillation, !
o
. Then applying condition (5.3) will determine how much gain is necessary
in order to make the loop gain larger than 1 at !
o
.
Circuits with the topology shown in Figure 5.3a are commonly employed as oscillators.
The active device could be a BJT or an FET. This circuit can be analyzed as a feedback
loop. The circuit is redrawn in Figure 5.3b to explicitly show that the feedback from output
to input is through the element Z
3
. The loop gain is easily computed for circuits of the type
represented by Figure 5.3. For small-signal analysis we can model the transistor as shown
in Figure 5.4. Note that this is a simplied version of the hybrid-pi model for the transistor
(see Appendix A). The passive elements of the model (e.g., r
, C
, C
g
m
v
be
r
C
o
C
2
L
R
1
||R
2
C
R
e
r
Figure 5.10: Small-signal equivalent circuit for common-collector Colpitts. The resistance,
r, represents the series resistance of the inductor.
hybrid-pi model, and the nite Q of the inductor is modeled with a series resistance, r.
The small-signal equivalent circuit can be further simplied by combining some elements as
shown in Figure 5.11 , where
+
-
v
be
Z
3
g
m
v
be Z
2
Z
1
Z
in
= Z
1
+ Z
2
+ g
m
Z
1
Z
2
Figure 5.11: Simplied small-signal equivalent circuit of common-collector Colpitts oscilla-
tor.
Z
1
=
1
j!(C
1
+C
)
||r
(5.12)
Z
2
=
1
j!(C
2
+C
o
)
||R
e
Z
3
= R
1
||R
2
||(r +j!L)||
1
j!C
, C
2
C
o
. This ensures that the exter-
nal components swamp the internal capacitances of the transistor, thereby minimizing the
circuits dependence on variations in the internal transistor capacitances. Such variations
could be caused by dierences between individual transistors, or by variations in junction
capacitances as a function of the voltage across the junctions. It is also useful to choose C
1
154 CHAPTER 5. OSCILLATORS
and C
2
to be large enough so that 1/(!C
1
) r
, 1/!C
2
R
e
. This causes Z
1
and Z
2
to be
dominated by the capacitances external to the transistor, thereby minimizing dependence
on r
and R
e
. In view
of these considerations, in the following analysis we shall make the following replacements:
C
1
+ C
! C
1
and C
2
+ C
o
! C
2
. In other words, C
1
should be interpreted as the to-
tal capacitance across the base-emitter junction, including the internal capacitance of the
transistor and any external capacitance. A similar interpretation holds for C
2
.
It is now useful to make some approximations in order to simplify the analysis. We shall
assume that the impedances of the capacitances C
1
and C
2
are small compared to r
and
R
e
, respectively. In other words, dene Q
1
= !C
1
r
and Q
2
= !C
2
R
e
; we assume that
Q
1
1 and Q
2
1. Then Z
1
and Z
2
can be transformed using a high-Q parallel to series
transformation, i.e.
Z
1
'
r
Q
2
1
+
1
j!C
1
=
1
!
2
C
2
1
r
+
1
j!C
1
, Q
1
1 (5.13)
Z
2
'
R
e
Q
2
2
+
1
j!C
2
=
1
!
2
C
2
2
R
e
+
1
j!C
2
, Q
2
1. (5.14)
We shall also assume that the impedance of the Z
3
branch is well approximated by the
impedance of the inductor, i.e.
Z
3
' r +j!L. (5.15)
If this is not the case, then r and L should be interpreted as the eective series resistance,
and inductance, of Z
3
.
The circuit has been split into two parts by the dashed line in Figure 5.11. Denote the
impedance looking into the part of the circuit to the right of the dashed line by Z
in
. Then
Z
in
= Z
1
+Z
2
+g
m
Z
1
Z
2
.
The condition for steady-state oscillation is Z
in
+Z
3
= 0, or,
Z
1
+Z
2
+Z
3
+g
m
Z
1
Z
2
= 0. (5.16)
The requirements for steady-state oscillation are obtained using equations 5.13-5.15 in 5.16.
The real part of this equation is
1
!
2
C
2
1
r
+
1
!
2
C
2
2
R
e
+r +g
m
(
1
!
4
C
2
1
C
2
2
r
R
e
1
!
2
C
1
C
2
) = 0. (5.17)
It is now necessary to remember that r
depends on g
m
, i.e. r
= /g
m
, so equation 5.17
is really quadratic in g
m
, and the solution is somewhat messy. However, at suciently high
frequencies, the term involving !
4
may be neglected. More precisely, the term involving
!
4
may be neglected provided that !
2
C
1
C
2
r
R
e
1, or if Q
1
Q
2
1. Since we have
already assumed that Q
1
1 and Q
2
1, neglect of this term is justied within the
context of our original assumptions. Hence, neglecting the !
4
term, and using r
= /g
m
in equation 5.17, the net resistance becomes
g
m
!
2
C
2
1
+
1
!
2
C
2
2
R
e
+r
g
m
!
2
C
1
C
2
= 0. (5.18)
This equation can be solved for the transconductance needed to set the net resistance to
zero at any particular frequency. This transconductance will be denoted g
m.ss
because
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 155
it is the transconductance needed to support steady-state oscillations. The steady-state
transconductance can be written as follows:
g
m,ss
=
!
2
C
1
C
2
r +
C1
C2Re
1
C2
C1
. (5.19)
Remember that this result was derived under the assumption that Q
1
1, and Q
2
1. The
rst assumption requires that g
m
!C
1
, and should be checked after g
m,ss
is calculated
from equation 5.19, however in most practical applications this inequality will be satised.
If R
e
is allowed to approach innity (so that Z
2
becomes a pure reactance), and if
C2
C1
, then g
m,ss
reduces to g
m,ss
= !
2
C
1
C
2
r, the same as the result given in equation
5.7, which was derived by assuming that Z
1
and Z
2
were pure reactances. In the next
section, it will be shown that it is desirable to make C
1
> C
2
. Thus, in practical cases with
1, the second term in the denominator of 5.19 can be neglected, in which case
g
m,ss
' !
2
C
1
C
2
r +
C
1
C
2
1
R
e
. (5.20)
It is worth noting that equation 5.20 can be written in terms of the inductor Q
L
. In
terms of the inductor reactance, X
L
, and the inductor Q
L
(both are assumed to be known,
or specied, at the frequency of oscillation) r = X
L
/Q
L
. As long as Q
1
1 and Q
2
1, the
frequency of oscillation will be the frequency where the inductor reactance resonates with
the series combination of C
1
and C
2
, hence X
L
= !
o
L '
1
!o
C
1
C
2
C
1
+C
2
. Thus, r = X
L
/Q
L
'
(C
1
+C
2
)/(!
o
C
1
C
2
Q
L
) - if this substitution is made in equation 5.20, the result is:
g
m,ss
'
!
o
(C
1
+C
2
)
Q
L
+
C
1
C
2
1
R
e
. (5.21)
In practical circuits additional reactances or resistances may be placed in parallel with
the inductor. To implement a tunable oscillator it is common practice to use a variable
capacitor in series or in parallel with the inductor. Since power must be extracted from the
oscillator to drive other stages, a resistance representing an external load may be placed in
shunt with the inductor if the output is taken across the inductor. The existing analysis is
easily modied to handle such cases. In the original model, the resistance, r, represented
the series resistance associated with the inductor. If the additional components are in
parallel with the inductor, it is simply necessary to determine the series representation of
the resulting Z
3
branch (at !
o
) and use the real part of this impedance in place of r in
equation 5.20. Alternatively, evaluate the component Q of the Z
3
branch at !
o
, and use
this value in place of Q
L
in equation 5.21. In some circuits the output is taken from the
emitter - in this case, the external load resistance will appear in parallel with R
e
, and the
value used for R
e
can be changed to include the contribution from the load.
As mentioned previously, the oscillator circuit will be designed such that the small-signal
loop gain is larger than 1. The loop gain is equal to the ratio g
m
/g
m,ss
. As the oscillation
amplitude builds up, the base-emitter voltage swing will increase. Transistor operation
moves out of the small-signal linear regime, nonlinear eects become important, and the
eective transconductance is reduced. Oscillation amplitude will grow until the eective
large-signal transconductance is reduced to g
m,ss
, which is the value required to sustain
steady-state oscillations. The loop gain is the ratio of the initial transconductance to g
m,ss
(A
lo
=g
m
/g
m,ss
), and is an important parameter that determines how deeply the transistor
156 CHAPTER 5. OSCILLATORS
must be driven into the nonlinear regime before the steady-state condition is reached. In
the next section we will see that under certain realistic assumptions, the ratio g
m
/g
m,ss
determines the steady-state amplitude of the base-emitter voltage swing. This means that
it is possible to predict, at least approximately, the amplitude of the voltage swing in the
oscillating circuit.
5.4.2 Numerical Simulation
In order to test the results derived in the preceeding section, a common collector Colpitts
circuit was simulated using the Agilent ADS simulation program. The frequency of oscilla-
tion was chosen to be f
o
= 50 MHz. The circuit was simulated using a 2N5179 transistor
model. To compare the simulated results with analytical predictions, the parameters of the
hybrid-pi model are needed. The transistor model was probed using simulations to deter-
mine that = 62, C
' 10 pF, C
, which also appears across the inductor, is small and was ignored. To start
the oscillations, the circuit was articially excited by pulsing the supply voltage from 12.0
V to 12.1 V and back to 12.0 V within a period of 10 ns. This is necessary in a numerical
simulation in order to provide some initial disturbance that can then grow into a steady-
state oscillation. In practice, omnipresent thermal noise (or the step excitation provided by
connecting the supply voltage) would play this role.
Six dierent {C
1
, C
2
} pairs were considered (see Table 5.2). For each case, the transcon-
ductance required to sustain steady-state oscillation (g
m,ss
), and the small-signal loop gain
(g
m
/g
m,ss
) are given. The values of C
1
and C
2
were chosen to satisfy 2(50 MHz) =
L
C1C2
C1+C2
1
, so that the oscillation frequency is xed at approximately 50 MHz in all cases.
Notice that if a {C
1
, C
2
} pair is found that resonates with L at the desired !
o
, the values
can be reversed without signicantly changing the resonant frequency. Compare cases 4 and
6 for an example. As we shall see, however, the case with C
1
> C
2
will usually give better
performance.
Simulation results are summarized in Figures 5.12-5.17. In each Figure, the upper left
plot shows the voltage across the inductor for an interval of 1 s after the initial transient.
The other plots show the emitter current, base-emitter voltage, and emitter voltage for
1
A common base conguration would be better in this respect, as the bias resistors do not load the
inductor in that circuit.
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 157
Table 5.2: Parameters used for oscillator simulations.
Case C
1
C
2
g
m,ss
g
m
/g
m,ss
1 2500 pF 20.4 pF 45.3 mS 0.88
2 1000 pF 20.6 pF 18.1 mS 2.2
3 500 pF 21.1 pF 9.42 mS 4.4
4 200 pF 22.5 pF 3.77 mS 11.0
5 100 pF 25.4 pF 1.83 mS 21.8
6 22.5 pF 200 pF 2.23 mS 17.9
the period 0.9-1.0 s after the initial transient. For the cases where oscillation occurs, the
expanded plots show the current and voltage waveforms when the circuit is undergoing
steady-state oscillation.
Figure 5.12: Case 1 - loop gain is 0.88, which is less than one, so oscillation does not start.
The loop gain (g
m
/g
m,ss
, column 4 in Table 5.2) is smaller than 1 for Case 1. Therefore
the initial disturbance excites damped oscillations at a frequency approximately equal to f
o
.
This case illustrates the fact that sustained oscillations cannot develop if the small-signal
loop gain is smaller than 1.
In cases 2 through 6 the loop gain is larger than 1, and the initial transient excites
growing oscillations, as shown in the plot of the voltage across the inductor (upper left plot
in each Figure). As the loop gain increases from 2.2 to 21.8 in cases 2 through 5 the initial
transient builds up to steady-state condition faster, and the amplitude of the steady-state
voltage swing increases. Notice that increases in the base-emitter voltage swing result in the
158 CHAPTER 5. OSCILLATORS
Figure 5.13: Case 2 - g
m
/g
m,ss
= 2.2.
Figure 5.14: Case 3 - g
m
/g
m,ss
= 4.4.
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 159
Figure 5.15: Case 4 - g
m
/g
m,ss
= 11.0.
Figure 5.16: Case 5 - g
m
/g
m,ss
= 21.8.
160 CHAPTER 5. OSCILLATORS
Figure 5.17: Case 6 - g
m
/g
m,ss
= 17.9.
collector current (lower, left) waveform exhibiting increasingly narrow and increasingly large
current spikes. For the larger loop gains, the transistor is essentially cut o for most of the
oscillation cycle. The transistor injects a short current pulse into the resonator once each
cycle, near the positive peak of the voltage swing. Thus, for most of the oscillation period,
the circuit is un-excited and oscillation is maintained by the ywheel eect of the high-Q
resonant circuit composed of L, C
1
, and C
2
. The emitter current ows into an impedance
that has large magnitude at !
o
, and very small magnitude at harmonics of !
o
- hence, the
emitter voltage is nearly sinusoidal. In cases 4 and 5 where the loop gain is largest, and the
current spikes are narrowest, some distortion of the emitter voltage waveform can be seen
near the voltage peaks. In general, larger loop gains are associated with more distortion in
the output waveform.
Case 6 was chosen to illustrate what happens when C
1
< C
2
. Notice that the loop gain
in case 6 is not very dierent from the loop gain in case 5 (17.9 vs. 21.8), yet the voltage
swing across the inductor and the emitter voltage swing are quite small. Furthermore, the
emitter voltage is clearly non-sinusoidal, indicative of relatively high harmonic content. This
reects the fact that C
1
is not large enough to swamp r
results in a lower overall Q for the LC tank circuit than in case 4. As a result, the harmonics
contained in the spiky emitter current waveform are more prominent in the emitter voltage
waveform. The output from the oscillator would be taken from the top of the inductor, or
from the emitter - so comparison of cases 4 and 6 shows that choosing C
1
> C
2
(as in case
4) yields a larger output swing, and less distortion in the output.
Cases 4 and 6, which have comparable loop gains but very dierent amplitudes of oscilla-
tion, motivate the need to understand the factors which govern the amplitude of oscillations.
Consider the large-signal equivalent circuit for the oscillator circuit in Figure 5.18.
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 161
+
-
V
be
Z
3
G
m
V
be
Z
2
Z
1
+
-
V
e
Figure 5.18: A model for the Colpitts oscillator when oscillating in steady-state. The small-
signal transconductance has been replaced by a large signal transconductance, G
m
, which
is a function of the base-emitter voltage swing, |V
be
|.
The small-signal transconductance has been replaced by an eective (large-signal) transcon-
ductance (G
m
) in Figure 5.18. If the circuit is undergoing steady-state oscillations, the large
signal transconductance will be equal to the transconductance required to sustain oscilla-
tions, i.e. G
m
= g
m,ss
. In Appendix A, it is shown that when a BJT is biased for constant
DC emitter current, and when the base-emitter voltage swing is sinusoidal at frequency !
o
,
a large-signal transconductance can be dened, which relates the amplitude of the emitter
current component at !
o
to the amplitude of V
be
. The large signal transconductance can be
written as
G
m
(x) = g
m
2
x
I
1
(x)
I
0
(x)
, (5.22)
where g
m
is the small-signal transconductance, and x = |V
be
|/(25 mV) is the normalized,
peak base-emitter voltage amplitude. Figure 5.19 shows how the ratio G
m
/g
m
decreases
as |V
be
| increases. In steady-state, |V
be
| settles at the value required to set the large signal
2 4 6 8 10
0.2
0.4
0.6
0.8
1
G
m
(x)
g
m
x
Figure 5.19: Ratio of large signal to small-signal transconductance. The parameter x =
|V
be
|
25 mV
.
transconductance equal to the minimum value required to sustain oscillations, i.e. in steady
state G
m
(|V
be
|/(25 mV)) = g
m,ss
. For example, consider case 2, where the small-signal loop
gain is g
m
/g
m,ss
= 2.2. To reach the steady state, |V
be
| must increase until G
m
/g
m
=
1/2.2 = 0.45. From Figure 5.19 it is apparent that this occurs when |V
be
|/(25 mV) is
approximately equal to 3.8. A more exact value can be obtained using equation 5.22;
162 CHAPTER 5. OSCILLATORS
numerically solving the equation 0.45 =
2
x
I
1
(x)/I
o
(x) yields x ' 3.75. The predicted steady-
state amplitude of |V
be
| is then (3.75)(25 mV) = 94 mV, which compares favorably with what
is observed in the simulations. Table 5.3 compares the base-emitter swings observed in the
simulations with values predicted in this way. In cases 2-6, where oscillation occurs, the
Table 5.3: Comparison between predicted and simulated |V
be
| and |V
e
|/|V
be
| ratio.
Case g
m
/g
m,ss
|V
be
| pred. |V
be
| sim. |V
e
| sim. C
1
/C
2
|V
e
|/|V
be
|
1 0.88 - - - 122.5 -
2 2.2 94 mV 71 mV 3.27 V 48.5 46.1
3 4.4 207 mV 185 mV 4.30 V 23.7 23.2
4 11.0 537 mV 554 mV 4.70 V 8.9 8.5
5 21.8 1.08 V 1.29 V 5.40 V 3.9 4.2
6 17.9 882 mV 483 mV * 0.11 *
prediction yields a reasonable approximation to the simulated values.
Once |V
be
| is known, or has been predicted, the amplitude of the emitter voltage, |V
e
|,
can be estimated. A node equation at the junction between the current source, Z
1
, and Z
2
in Figure 5.22 yields
V
e
V
be
=
Z
2
Z
1
+Z
2
g
m,ss
(5.23)
=
C
1
C
2
+Z
2
g
m,ss
. (5.24)
The second term will be small compared to the rst in practical cases. This can be veried
for the cases considered here by inserting numerical values, or by noting that the simulation
results show that V
e
and V
be
are nearly in-phase. Hence, the ratio V
e
/V
be
is real - conrming
that the second term is negligible in all cases considered here. Thus, the emitter voltage
swing is determined by the capacitive transformer consisting of C
1
and C
2
and can be
written as:
|V
e
| = |V
be
|
C
1
C
2
. (5.25)
Table 5.3 provides data showing how the ratio |V
e
|/|V
be
| determined from simulation com-
pares to the C
1
/C
2
ratio. In cases 2-5 the agreement is excellent. In case 6, the emitter
voltage waveform was non-sinusoidal, and the amplitude of the fundamental component is
not easily extracted from the time-domain waveform, so the ratio was not calculated. These
results show that making the ratio C
1
/C
2
> 1 will cause the emitter (output) voltage to
be larger than the base-emitter voltage swing. When C
1
/C
2
> 1, increasing C
1
/C
2
yields
smaller loop gains, so oscillators with large C
1
/C
2
tend to have more sinusoidal output
waveforms because the base-emitter swing is relatively small, and the emitter current is
more sinusoidal.
5.5 Example: Voltage Controlled Oscillator (VCO)
Figure 5.20 is an example of an oscillator circuit used in a commercial television receiver.
The oscillator is part of the VHF tuner that performs the function of converting the VHF
5.5. EXAMPLE: VOLTAGE CONTROLLED OSCILLATOR (VCO) 163
channels to the IF frequency near 45 MHz, this local oscillator is voltage-tuned using a
varactor diode. Figure 5.20 also illustrates how dierent inductances can be switched in and
out of the circuit using diode switches. Although this circuit looks more complicated than
the one we have been studying, it is essentially the same, except for the fact that the base
of the transistor is at RF ground (common-base conguration).
+12 V
10 pF
C1
2.2 pF
C2
Vout
10 k
33 k
5 F
5 k 680
L2 L1
Tuning
Voltage
Band-switch
2 k
Tuning
Diode, Cv
Figure 5.20: Example: VHF oscillator for TV tuner.
A simplied schematic of Figure 5.20 is shown in Figure 5.21.
Cv
L
C1
C2
Figure 5.21: Simplied schematic of VHF oscillator circuit.
Based on earlier analysis, it should be clear the the approximate frequency of oscillation
for this circuit will be the frequency where C
1
, C
2
and the L-C
v
combination are resonant.
We also know that the frequency of oscillation will be such that the L-C
v
combination looks
inductive.
164 CHAPTER 5. OSCILLATORS
Thus !
o
is the solution to
1
j!
o
C
S
+
j!
o
L
1
j!oCv
j!
o
L +
1
j!oCv
= 0 (5.26)
where
C
S
=
C
1
C
2
C
1
+ C
2
(5.27)
Therefore
!
o
=
1
L(C
S
+C
v
)
(5.28)
Equation 5.28 can be used to nd the range of values over which C
v
must vary in order
to cover a certain range of oscillation frequencies.
Voltage controlled oscillators (VCOs) nd extensive application in phase-locked loops
(PLLs) which in turn are used for a majority of frequency synthesizer applications. For
PLL or synthesizer design, an important parameter is the so-called VCO gain constant,
which is simply the incremental change in oscillation frequency for an incremental change
in varactor bias voltage. Typically, the varactor would be an abrupt transition type, so
that
C
v
=
C
(1)
p
V
D
where V
D
is the varactor bias voltage and C
(1)
is the junction capacitance at 1 volt of bias.
The gain constant is
K
o
=
@!
o
@V
D
!o = oscillation frequency
(5.29)
=
@!
o
@C
v
@C
v
@V
D
!o = oscillation frequency
Generally, K
o
will be a nonlinear function of V
D
. This implies that the gain constant
will dier at various varactor bias settings, a factor which must be considered in the design
of a PLL.
5.6 Oscillator Phase Noise
The most signicant departure from ideal behavior in oscillators is the presence of phase
noise, represented by a random variation in the oscillators phase angle, i.e.
V (t) = V
o
cos[2f
o
t + (t)] (5.30)
where (t) is a random noise process that has various physical processes. Amplitude noise
may also be present, but it is easy to remove this noise component by passing the signal
through a limiter. Figure 5.22 shows measurements of the output spectrum of a synthesized
signal generator operating at 10.240 MHz. The width of the narrow central spike is deter-
mined by the resolution bandwidth of the spectrum analyzer which was set to 100 Hz for this
5.6. OSCILLATOR PHASE NOISE 165
Figure 5.22: Spectrum analyzer display showing phase noise on the output signal from
an HP E4432B synthesized signal generator. The signal generators output frequency was
10.240 MHz and the output level was 0 dBm. The triangular marker is oset 1000 Hz
from the carrier frequency, and the phase noise at this oset is 94.2 dBc/Hz (see text for
explanation).
166 CHAPTER 5. OSCILLATORS
measurement. Residual phase modulation phase noise on the signal contributes the
broad bandwidth pedestal underneath the spike. Phase noise spectral density is specied as
a function of frequency separation from the carrier in terms of the noise power level within
a 1 Hz bandwidth. The noise power level is given in dB referenced to the carrier power
level. For example, in Figure 5.22 the triangular marker has been placed 1000 Hz above the
carrier frequency, and the display indicates that the noise pedestal is at 74.2 dBc, where
dBc means relative to the carrier. Assuming that the noise spectral density is constant
within the resolution bandwidth the measured noise power level in 100 Hz bandwidth can
be scaled to 1 Hz bandwidth by dividing by 100 (or by subtracting 10 log
10
(100) = 20 dB).
Thus, the phase noise level at 1000 Hz oset would be reported as 74.2 20 = 94.2
dBc/Hz.
Assuming that the random phase process is dierentiable, the instantaneous frequency
of the noisy oscillator can be written as
f
inst
= f
o
+
1
2
d(t)
dt
. (5.31)
The instantaneous frequency deviation is therefore
f(t) =
1
2
d(t)
dt
. (5.32)
The fractional deviation in instantaneous frequency can be written:
y(t) =
f(t)
f
o
=
1
2f
o
d(t)
dt
. (5.33)
Measures of phase (or frequency) stability may be dened in the frequency domain in
terms of a power spectral density, or in the time domain in terms of a phase, or frequency,
or fractional-frequency variance dened over some time-interval.
If the random phase process has spectral density S
(f). (5.34)
The spectral density S
y
(f) will be invariant to multiplication or division of the frequency
using ideal frequency multipliers or dividers because frequency multiplication or division
scales the reference frequency, f
o
, and the frequency deviation, f, by the same factor.
In the time-domain, oscillator stability is usually characterized by the expected value
of the sample variance of fractional frequency uctuations. Dene the average fractional
frequency deviation over an interval , beginning at time t
k
by
y
k
=
1
t
k
+
t
k
y(t)dt
=
(t
k
+)(t
k
)
2fo
.
(5.35)
When N measurements of y
k
are available, each taken over interval and with starting
times separated by interval T, the sample variance of the measurements is:
2
y
(N, T, ) =
1
N 1
N
n=1
(y
n
1
N
N
k=1
y
k
)
2
(5.36)
5.6. OSCILLATOR PHASE NOISE 167
The expected value of the quantity
2
y
is known as the Allan variance of the oscillator.
For most purposes, oscillator stability is specied in terms of the Allan variance for N=2
and T = , i.e. the quantity that is specied most often is the variance of the dierence
between 2 successive measurements with no dead time between the measurement intervals.
In this case, the Allan variance reduces to:
<
2
y
() >=
< (y
2
y
1
)
2
>
2
(5.37)
Thus, the Allan variance is the variance of the dierence between two successive measure-
ments of the fractional frequency deviation, measured over a time interval . The square-root
of the Allan variance is called the Allan Deviation, and is commonly supplied on oscillator
data sheets where it is often plotted verses the measurement interval .
168 CHAPTER 5. OSCILLATORS
5.7 References
1. Clarke, Kenneth K. and Donald T. Hess, Communication Circuits: Analysis and De-
sign, Addison-Wesley, 1978.
2. Frerking, Marvin E., Crystal Oscillator Design and Temperature Compensation, Van
Nostrand Reinhold, New York, 1978.
3. Krauss, H. L., C. W. Bostian, and F. H. Raab, Solid State Radio Engineering, John
Wiley & Sons, New York, 1980.
4. Matthys, Robert J., Crystal Oscillator Circuits, John Wiley & Sons, New York, 1984.
5. Parzen, Benjamin, Design of Crystal and other Harmonic Oscillators, John Wiley &
Sons, New York, 1983.
6. Smith, Jack, Modern Communications Circuits, Second Edition, McGraw Hill, 1998.
5.8. HOMEWORK PROBLEMS 169
5.8 Homework Problems
1. Consider the voltage amplier in Figure 5.23, with
V
cc
=12 V R
1
= 10 k R
2
= 30 k R
e
= 1 k
R
C
= 1 R
L
= 1 k L = 2 H C = 50 pF
Capacitors that are not labeled are assumed to be short circuits over the frequency
range of interest. The transistors is large enough so that the bias point does not
explicitly depend on its value. You may neglect the transistor parameters r
x
, r
, C
,
r
o
, and C
o
in your analysis for parts 1b, 1c, and 1d.
C
R
1
R
2
R
e
V
cc
R
L
R
C
L
v
i
v
o
Figure 5.23: Common-emitter amplier with tuned output.
(a) Find the quiescent collector current, I
CQ
. Express your result in mA.
(b) Find the resonant frequency of the amplier. The voltage gain will be largest at
this frequency. Express your result in MHz. Do not make any 2 errors!
(c) Find the voltage gain at resonance.
(d) Find the 3 dB bandwidth of the amplier. Express your result in MHz.
2. The hybrid-pi parameters for a transistor are:
r
=
0.025
I
CQ
(5.38)
C
= 25 pF
C
= 8 pF
r
o
= 100 k
= 100
(a) The transistor, with parameters given above, is used in a common emitter ampli-
er, as shown in Figure 5.24. Find an expression for the voltage gain of amplier.
Express your result in terms of R
C
, R
L
, C
, and g
m
. You may assume that r
o
is much larger than R
C
kR
L
and can be neglected. Assume that r
x
and C
o
are
small and may also be neglected.
(b) If the quiescent collector current is 1mA, R
C
= R
L
= 5k , and all coupling and
bypass capacitors are assumed to be perfect short circuits, sketch the magnitude
and phase of the voltage gain as a function of frequency. What are the magnitude
and phase of the transfer function at 15 MHz?
170 CHAPTER 5. OSCILLATORS
R
1
R
2
R
e
V
cc
R
L
R
C
v
i
v
o
Figure 5.24: Common emitter amplier
(c) Suppose that we attempt to measure the voltage gain of this amplier using a
10X scope probe. The shunt capacitance of this type of probe is 12pF. Compute
the gain (at 15 MHz) that would be measured. Compare to the result from 2b.
3. Consider the voltage amplier shown in Figure 5.23. Capacitors that are not labeled
are assumed to be short circuits over the frequency range of interest. The other circuit
parameters are
V
CC
=12 V R
1
= 5 k R
2
= 10 k R
E
= 3.3 k
R
C
= 5 k R
L
= 1 k L = 20 H C = 90 pF
Use the simplied hybrid-pi model shown in Figure 5.25 for the transistor:
+
-
v
be
g
m
v
be
B
E
C
r
C
L
C1C2
C1+C2
(5.41)
A measure of the relative stability of the oscillator is the slope of the loop-gain phase
characteristic evaluated at the oscillation frequency, i.e.,
d
d!
|
!=!o
(5.42)
where is the phase of the loop gain.
(a) Show that
d
d!
|
!=!o
=
2Q
!o
where Q is the inductor Q evaluated at the frequency
of oscillation.
(b) Suppose that an oscillator is built at 5MHz with an inductor having a Q = 50.
Estimate the shift in oscillation frequency, if the overall phase of the loop gain is
perturbed by 5 degrees due to temperature drift of the capacitors.
(c) Now assume that the inductor is replaced with a crystal having an eective
Q = 20, 000. Estimate the frequency shift.
5.8. HOMEWORK PROBLEMS 175
C R
1
L R
2
Figure 5.35: Two-stage oscillator circuit.
11. Consider the oscillator circuit in Figure 5.35.
Assume that the transistors are identical and can be represented by the simplied
equivalent circuit in Figure 5.36.
+
-
v
be
g
m
v
be
B
E
C
Figure 5.36: Simplied equivalent circuit.
(a) Find an expression for the (non-zero) frequency of oscillation (if it occurs).
(b) Find the minimum value of g
m
required for oscillation to occur.
12. Consider the oscillator circuit in Figure 5.37 where a voltage amplier with 50
input and output impedance and open-circuit voltage gain A is used with a lossy
section of transmission line having characteristic impedance Z
o
= 50 , length L, and
attenuation and phase shift per unit length denoted by and , respectively. Since
the line is terminated in its characteristic impedance, the voltage at the output of the
line will be related to the voltage at the input of the line by the following equation:
V
out
= V
in
e
(+j)L
(5.43)
where
=
2f
v
p
(5.44)
and v
p
is the phase velocity for propagation on the line.
(a) Find an expression for the potential frequencies of oscillation. Assume that the
voltage gain, A, is real (zero phase shift) and that the input and/or output of
the voltage amplier is AC-coupled with a coupling capacitor (not shown) that
will prevent the circuit from oscillating at zero frequency. The capacitor may be
assumed to have no eect on the circuit performance at nite frequencies.
176 CHAPTER 5. OSCILLATORS
Av
i
v
i 50
Z
o
, ,
Lossy T-line
50
+
-
+
-
L
Figure 5.37: Oscillator circuit.
(b) Suppose that the T-line loss per unit length, , is proportional to frequency, i.e.,
= Kf. Dene the minimum and maximum values for the voltage gain, A, such
that the circuit will oscillate only at the smallest (non-zero) frequency found in
part 12a.
13. Suppose a capacitor is placed in parallel with a quartz crystal.
(a) Describe how the series and parallel resonant frequencies of the crystal and ca-
pacitor combination will dier from those of the crystal.
(b) Suppose the crystal is used in an oscillator where it operates in the parallel
resonant mode, i.e., the frequency of oscillation is a frequency where the crystal
impedance is inductive. How will the frequency of oscillation change if a capacitor
is added in parallel with the crystal? Explain your reasoning and any underlying
assumptions.
14. The crystal oscillator circuit shown in Figure 5.38 is called a series mode oscillator,
because it will oscillate very close to the series resonant frequency of the crystal.
Here the crystal grounds the base of the transistor at its series resonant frequency.
C
1
C
2
L
R
1
R
2
R
e
V
cc
Z
x
R
L
Figure 5.38: Series mode xtal oscillator derived from common-base Colpitts conguration.
The circuit will also oscillate if the crystal is replaced by an AC short circuit (e.g.,
a bypass capacitor). Use the negative resistance approach to study this oscillator.
You may assume that the transistor can be modeled using the equivalent circuit in
5.8. HOMEWORK PROBLEMS 177
+
-
v
be
g
m
v
be
B
E
C
r
+
Z
1
(Z
L
+ Z
2
) + g
m
r
Z
1
Z
2
Z
1
+ Z
2
+ Z
L
(5.45)
where
Z
1
=
1
j! C
1
(5.46)
Z
2
=
1
j! C
2
Z
L
=
j!LR
L
j!L + R
L
(b) Solve for the frequency at which the circuit will oscillate (!
o
), if an AC short is
connected from the base of the transistor to ground (instead of the crystal). You
can assume that R
L
!L at the frequency of oscillation.
(c) What condition must be satised in order to guarantee that oscillations will start?
15. Consider the oscillator circuit in Figure 5.40. The active device is modeled as a
+
-
1
2
3
R
2
L C
Figure 5.40: Oscillator circuit.
transconductance amplier with input resistance R
1
as shown in Figure 5.41. (Note
178 CHAPTER 5. OSCILLATORS
+
-
v
g
m
v
R
1
1
2
3
Figure 5.41: Transconductance amplier with input resistance R
1
the important dierence between this equivalent circuit and the simplied hybrid-pi
model!)
(a) Find an expression for the potential frequency of oscillation, !
o
.
(b) Find the threshold value, g
m
min
, which the transconductance must exceed in
order for oscillation to occur.
16. Consider the oscillator circuit shown in Figure 5.42. In this problem, V
cc
= 3 V, R
b
=
33 k, R
e
= 100 k, C
1
= 10 pF, C
2
= 22 pF. Unlabeled capacitors are either coupling
or bypass elements and have negligibly small impedance at the frequencies of interest.
You may assume that the load resistor is disconnected from the circuit for parts (a)-(d).
C
1
C
2
R
b
R
e
V
cc
R
L
L
Z
in
not connected for (a)-(d)
Figure 5.42: Load resistor is not connected for parts (a)-(d).
(a) Find an approximate value for the quiescent collector current, I
CQ
. You may
assume that = 100.
(b) Find the transconductance, g
m
, for the transistor in this circuit.
(c) Assume that the reactances of C
1
and C
2
are small enough such that r
, R
e
,
and R
b
can be ignored for small-signal analysis. In this case, if the inductor is
removed the impedance Z
in
looking in to the circuit at the point shown in the
Figure will be
Z
in
=
g
m
!
2
C
1
C
2
+
1
j!
C1C2
C1+C2
.
Ignore the load resistance, R
L
, (i.e. assume that the load is disconnected) and
calculate the inductance, L, required to set the potential frequency of oscillation
in this circuit to 50 MHz.
5.8. HOMEWORK PROBLEMS 179
(d) Suppose a lossy inductor is available with inductance equal to the value that you
calculated in part (c). Calculate the minimum inductor Q (i.e., Q
L
) required for
oscillations to be sustained in this circuit.
(e) Now suppose that the actual inductor Q
L
is 50. A load is placed on the oscillator
by connecting R
L
to the inductor through a coupling capacitor as shown by the
dotted line in the gure. Calculate the smallest value of load resistance for which
the circuit will still oscillate.
17. Design a tunable oscillator that covers the frequency range 10 20 MHz using the
Colpitts conguration shown in Figure 5.43 . The coil inductance L = 2H and
C
1
L
R
C
2
Figure 5.43: Colpitts oscillator with bias circuitry omitted.
the resistance R = 4. For simplicity, assume that the immittances associated with
the transistor and the bias circuitry can be ignored. Furthermore, assume that the
oscillator will be tuned with a single variable capacitor. For this problem, assume that
C
2
is the variable capacitor and is adjustable within the limits [C
2,min
, C
2,max
]. As
discussed in the supplementary notes, it is good practice to ensure that C
1
C
2
to
allow large collector voltage swing with relatively small base-emitter voltage swing.
This allows the transistor to operate without being driven deep into saturation. For
your design, use C
1
= 10C
2,max
in order to satisfy this constraint. Design your circuit
so that the minimum value of the loop gain over the range 10-20 MHz is 2. For full
credit you must specify the value of C
1
, the minimum and maximum values for the
variable capacitor (C
2min
and C
2max
), the transconductance, g
m
, of the transistor,
and the quiescent collector current I
CQ
.
18. A crystal has packaging capacitance C
o
= 5 pF and series and parallel resonant fre-
quencies f
s
= 10 MHz and f
p
= 10.002 MHz.
(a) Determine the inductance, L
1
, and capacitance, C
1
, of the elements in the mo-
tional arm of the equivalent circuit for the crystal.
(b) The crystal is used in a parallel-resonance oscillator in which the crystal resonates
with an external load capacitance of 30 pF at the frequency of oscillation. The
frequency of oscillation can be written as f
o
= 10.000 MHz + f. Specify f to
an accuracy of +/- 5 Hz.
19. Consider the crystal oscillator shown in Figure 5.44. Without the network consisting
of the components labeled L and C, this circuit would oscillate near the fundamental
(lowest) resonant frequency of the crystal, denoted by f
o
, where f
s
< f
o
< f
p
, and f
s
,
180 CHAPTER 5. OSCILLATORS
C
1
C
2
C
L
Figure 5.44: Crystal oscillator circuit for operation on 3rd overtone.
f
p
are the lowest series and parallel resonant frequencies of the crystal. The purpose
of the series LC network is to make it impossible for the circuit to oscillate at the
fundamental frequency, and to allow the circuit to oscillate at the 3rd overtone, 3f
o
.
If f
o
= 50 MHz, C
1
= 10 pF, C
2
= 30 pF, C = 500 pF, nd the range of values for
L that will prevent the circuit from oscillating at f
o
, and allow the circuit to oscillate
at 3f
o
. For this problem, consider only the fact that the combination of C
2
, C, and L
must be capacitive in order for the circuit to oscillate at a frequency where the crystal
is inductive. You do not need to ensure that the magnitude of the loop gain will be
sucient to actually cause oscillations to start and grow.
20. The circuit shown in Figure 5.45 is often used as an overtone XTAL oscillator. In this
C
1
C L
Figure 5.45:
problem, we consider how to design the circuit so that it oscillates at the 5th overtone
as a parallel-mode oscillator. Note that bias resistors and coupling/bypass capacitors
are not shown and may be ignored. You may also ignore the internal immittances of
the transistor.
(a) Consider only the XTAL in this part. Denote the series and parallel resonant
frequencies associated with the 5th overtone by f
s,5
and f
p,5
, respectively. It
is known that f
s,5
= 49.990 MHz and f
p,5
= 50.030 MHz and the packaging
capacitance, C
o
, is 5 pF. Find the external load capacitance, C
L
, required to
form a parallel resonant circuit with the XTAL at exactly 50.000 MHz.
(b) Write down a constraint that must be satised by the resonant frequency of the
parallel LC branch in the circuit to allow the circuit to oscillate at the fth
overtone, and not at the fundamental or third overtone. Denote the desired
frequency of oscillation by f
o
. The XTALs fundamental resonances are near
f
o
/5 and the third overtone resonances near 3f
o
/5.
5.8. HOMEWORK PROBLEMS 181
(c) Write an equation in terms of C
L
, C
1
, L, and C which must be satised (along
with the constraint derived in part b.) to set the lowest potential frequency of
oscillation to exactly 50.000 MHz ( f
o
= 50.000 MHz). You do not need to solve
the equation.
21. Consider the Colpitts oscillator shown in Figure 5.46.
L C
V
C
1
C
2
Figure 5.46: Voltage-controlled oscillator (VCO).
(a) Find an expression for the frequency of oscillation, !
o
. You may assume that the
loop gain of the circuit is large enough to cause oscillation to start. You may
neglect the transistor immittances, so that your result will be in terms of only
the four parameters L, C
1
, C
2
, C
V
.
(b) Denote the minimum and maximum tuning-diode capacitances by C
V,min
and
C
V,max
and the capacitor tuning ratio by r
C
=
C
V,max
C
V,min
. Denote the minimum
and maximum frequencies of oscillation by !
o,max
and !
o,min
and the oscillator
tuning ratio by r
o
=
!o,max
!o,min
. Finally, denote the series combination of C
1
and C
2
by C
0
, i.e. C
0
=
C1C2
C1+C2
. Find an expression for r
o
in terms of r
C
, C
0
, and C
V,max
only.
(c) Find a numerical value for the tuning ratio r
o
if C
0
= 2C
V,max
and r
C
= 4.
182 CHAPTER 5. OSCILLATORS
Chapter 6
Impedance Matching Networks
6.1 Impedance Matching for Maximum Power Transfer
In this section we review the motivation for impedance matching and introduce important
concepts which will be used in later chapters. Let us rst illustrate the basic principles of
impedance matching for maximum power transfer. In Figure 6.1 a source with impedance
Z
S
= R
S
+ j X
S
is connected to a load Z
L
= R
L
+ j X
L
. The peak voltage of the source
(assumed to be sinusoidal) is V
S
:
Z
S
Z
L V
S
+
-
V
L
Figure 6.1: A voltage source driving an arbitrary load
The time-averaged real power delivered to the load can be written as
P
L
=
1
2
Re{V
L
I
L
} (6.1)
where V
L
and I
L
are the voltage across and current through the load impedance, respectively.
Note that the factor
1
2
is present because V
L
and I
L
are peak phasors,i.e., phasors whose
magnitude is the peak value of the sinusoidal time function. The current through the load
is
I
L
=
V
L
Z
L
(6.2)
so the time-averaged real power is
P
L
=
1
2
|V
L
|
2
Re{
1
Z
L
} (6.3)
=
1
2
|V
L
|
2
R
L
|Z
L
|
2
183
184 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
This can be written in terms of the source voltage V
S
as follows:
P
L
=
1
2
|V
S
|
2
R
L
|Z
L
+Z
S
|
2
(6.4)
=
1
2
|V
S
|
2
R
L
(R
L
+R
S
)
2
+ (X
L
+X
S
)
2
By studying Equation 6.4 we can determine what conditions are required to maximize
the power delivered to the load. First we will assume that the source voltage and source
impedance are set at the outset, i.e., they are not under our control. This will usually be
the case. The problem is then to choose R
L
and X
L
to maximize P
L
. Of course, we should
restrict the solution to values of R
L
that are greater than or equal to 0; values of R
L
that
are < 0 would require that the load contain an active device, whereas we are considering
only passive load terminations. The solution to this simple exercise is
Z
L
= Z
S
(6.5)
This familiar result states that the load impedance should be equal to the complex conjugate
of the source impedance, if maximum real power is to be delivered to the load. Using
Equation 6.5 in Equation 6.4, we can compute the maximum power:
P
L,max
=
1
2
|V
S
|
2
1
4R
S
(6.6)
=
|V
S
|
2
8R
S
Remember that this result was obtained by adjusting the load impedance to maximize
the power delivered by the source. The power given in Equation 6.6 is referred to as the
power available from the source or P
avs
. This concept of available power is often used in
the radio frequency literature and to describe RF signal generators. The output level of
these generators is usually calibrated in terms of power specied in dBm, i.e., decibels
referred to 1 mW. A power level of 6 dBm indicates a power 6 dB higher than 1 mW, or
approximately 4 mW. It is important to realize that this power level is the power available
from the generator, P
avs
, which means that it is the power that the generator will deliver to
a conjugately matched load. Signal generators commonly have source impedances of 50
(sometimes 75 ), i.e., the generator will deliver the rated power only to a 50 load. With
a dierent load impedance the power delivered to the load will be less than the rated value.
6.1.1 Mismatch Factor
The degree to which the actual power delivered to an arbitrary load is smaller than the
available power can be quantied in terms of a mismatch factor, a quantity that depends
on the degree of impedance mismatch between the source and load. For an arbitrary load
impedance Z
L
the mismatch factor will be dened as the ratio of actual delivered power to
available power:
Mismatch Factor =
P
L
P
avs
(6.7)
=
4R
S
R
L
(R
S
+R
L
)
2
+ (X
S
+X
L
)
2
6.1. IMPEDANCE MATCHING FOR MAXIMUM POWER TRANSFER 185
This important result tells us how much could be gained by modifying the load impedance
to obtain a conjugate match.
6.1.1.1 Example - Mismatch Factor
Suppose a 50 signal generator has available power of 1 mW. The generator is to drive a
load impedance of 250 +j100 . What is the power delivered to the load?
The problem could be solved by computing the power delivered to the load using Equa-
tion 6.1. Another approach is to compute the mismatch factor from Equation 6.7. This
gives a mismatch factor of 0.5, or -3 dB. The power available from the source is 1 mW, or 0
dBm. Thus, the actual power delivered to the load is 0 dBm - 3 dB or -3 dBm (0.5 mW).
6.1.2 Properties of Lossless Impedance Matching Networks
We have illustrated how impedance matching (or mis-matching) inuences the transfer of
power from a source to a load. In many applications it is desirable to maximize the transfer
of power from the source to the load. This can be achieved by using a lossless 2-port
network inserted between the source and the load. The purpose of the matching network is
to transform the load impedance, Z
L
, into Z
S
at the input terminals, thereby permitting
the source to deliver all of its available power to the network. If a matching network is
lossless, then all of the power that is delivered to the network must be delivered to the load.
This simple concept has implications that may not be obvious at rst glance.
Consider a source and load which are connected by a lossless matching network as shown
in Figure 6.2. As already noted, the matching network transforms the load impedance Z
L
Z
S
Z
L V
S
Lossless
Matching
Network
Z
S
Z
L
P
avs
P
avs
P
L
= P
avs
Figure 6.2: Source and load coupled by a lossless matching network
into Z
S
at the input terminals of the network. Since the source looks into the conjugate of
its impedance, it delivers all of its available power P
avs
to the network. Because no power
is dissipated in a lossless network, all of this power is delivered to the load. Now the power
available at the output of the lossless matching network must be the same as the power
available from the source, and we have already stated that all of this available power is
delivered to the load. Therefore the load must be conjugately matched to the output of
the matching network, which means that the impedance at the output of the network (as
seen by the load) is Z
L
. Thus a lossless matching network has the property of providing a
conjugate match at both the input and output ports.
Another way to look at this is to note that the original source plus the matching network
can be viewed as a new source with available power P
avs
and source impedance Z
L
. Adding
a lossless network to the output of a source does not change the available power from the
186 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
source; it only changes the source impedance. Thus, we can think of the matching network
as (i) a network that transforms the load impedance into Z
S
or (ii) a network that transforms
the source impedance into Z
L
without changing the available power of the original source.
The network can be designed either way.
Finally, it is interesting to point out that most lumped-element matching networks are
versions of ladder networks as shown in Figure 6.3.
Z
S
Z
L V
S
Z
S
Z
L
P
avs
P
avs
P
L
= P
avs
Z
Z
Figure 6.3: Ladder-type matching network
The ladder-type matching network in Figure 6.3 is assumed to be lossless; the blank
boxes represent purely reactive circuit elements. Now suppose that the network is broken
at the dashed line. The part of the network to the left of the line can be thought of as a
source. It will have the same available power as that of the original source, P
avs
, and some
source impedance, Z. The part of the network to the right of the dashed line is a load
for this source and must receive all of the available power (since that power is ultimately
transferred to Z
L
). Thus, a conjugate match must exist at the junction dened by the
dashed line. Clearly, the dashed line could have been drawn anywhere within the lossless
network and the same argument would hold. The conclusion is: A circuit consisting of
a source connected to a load through a lossless matching network can be broken
at any point between the source and the load and a conjugate match must exist
between the two sides of the circuit.
6.2 Impedance Matching with Lossless L-networks
6.2.1 Resistive Terminations
Figure 6.4 shows two resistances to be matched with a lossless L-network. The goal is to
transform R
1
to R
2
at one frequency. The unknown reactances X
s
and X
p
are easily found
with the use of a parallel-to-series transformation as in Figure 6.5.
Z
IN
= R
2 R
1
jX
s
jX
p
Figure 6.4: L-network which transforms R
1
into R
2
6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 187
Z
IN
= R
2
R
1
X
2
p
R
2
1
+X
2
p
jX
s
j
X
p
R
2
1
R
2
1
+X
2
p
Figure 6.5: Parallel-to-series transformation
To make the input impedance equal to R
2
, we choose
X
s
=
X
p
R
2
1
R
2
1
+ X
2
p
(6.8)
R
2
=
R
1
X
2
p
R
2
1
+ X
2
p
Solving for X
p
and X
s
gives
X
p
= R
1
R
2
R
1
R
2
(6.9)
X
s
=
R
2
R
1
R
2
2
(6.10)
The solutions yield real values for X
p
and X
s
(and hence purely reactive L-network com-
ponents) only if R
1
> R
2
. This leads to a rule for using a lossless L-network to match two
resistances:
The series arm of the L-network is connected to the smaller of the two resistances.
Also note that there are two possible solutions for the resistive matching problem corre-
sponding to the upper and lower signs in Equation 6.9 and Equation 6.10. These solutions
are referred to as low-pass and high-pass solutions. The circuit congurations for the
two solutions are in shown Figures 6.6 and 6.7.
R
1
R
2 R
1
> R
2
Figure 6.6: High-pass L-network
The high-pass and low-pass designations refer to the behavior of the network transfer
function at frequencies away from the design frequency.
188 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
R
1
R
2 R
1
> R
2
Figure 6.7: Low-pass L-network
6.2.2 Q of an L-network
Referring to Figure 6.5, it is apparent that the transformed L-network looks like a series-
resonant circuit. At the design frequency the two reactances have equal magnitudes and
opposite signs, so that the net series reactance is zero. Treating this circuit in the same
manner as a resonant RLC network, the Q of the network is the magnitude of one of the
series reactances divided by the series resistance:
Q =
X
p
R
2
1
/(R
2
1
+X
2
p
)
R
1
X
2
p
/(R
2
1
+X
2)
p
(6.11)
(6.12)
=
R
1
X
p
(6.13)
(6.14)
=
R
1
R
2
1 (6.15)
This result shows that the Q of the L-network is determined by the terminating resistances
R
1
and R
2
, and, therefore, cannot be chosen independently. This suggests that the band-
width of the matching network depends only on the ratio R
1
/R
2
.
A word of caution is in order. We have dened the Q of the L-network by analogy with
the series RLC network. This analogy works well only when R
1
|X
p
| or, equivalently,
when Q 1. In such cases the Q can be used to predict the 3 dB bandwidth of the networks
voltage transfer function. For moderate or small values of Q the expression
Q =
R
1
|X
p
|
=
R
1
R
2
1
is still valid, but a simple relationship between Q and bandwidth does not exist, since
the series reactance that results from the parallel-to-series transformation has a dierent
frequency dependence from that of a simple inductor or capacitor. Thus an L-network
does not behave exactly like a series RLC circuit. Only when the Q is very large will the
equivalent series reactance behave approximately like a capacitor or inductor.
6.2.3 Summary: L-network design equations
The design of an L-network can be summarized as follows. If the terminating resistances
are denoted by R
big
and R
small
(where R
big
> R
small
), then the design equations can be
6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 189
written in terms of the network Q,
Q =
R
big
R
small
1, (6.16)
as
X
p
= R
big
/Q X
s
= R
small
Q. (6.17)
When the upper signs are chosen, the resulting network is of highpass type. The lower signs
give a lowpass network. The L-network will be oriented such that the series arm connects
to R
small
, and the parallel arm is in shunt with R
big
.
6.2.3.1 Example - Matching resistive source and load with a low-pass L-network
Match a 100 source to a 25 load with a lossless L-network having a low-pass topology.
Since the series arm of the L-network connects to the smaller of the two resistances, the L-
25
100
Figure 6.8: Example for L-net matching
network will be oriented as shown in Figure 6.9. A capacitor and inductor have been chosen
C
L
Figure 6.9: L-network orientation
for the shunt and series elements, respectively, because the problem statement specied a
low-pass network.
Q =
R
big
R
small
1 =
100
25
1 =
p
3
X
p
= X
C
=
R
big
Q
=
100
p
3
= 57.7
X
s
= X
L
= QR
small
=
p
3 25 = 43.3
The completed network is shown in Figure 6.10. It is interesting to compare the power
delivered to the load with and without the matching network in place. Suppose the peak
voltage of the source is 1 Volt. Then the power delivered to the load without the matching
network would be
P
out
=
1
2
|V
out
|
2
25
=
1
50
1
5
2
= 0.8 mW (6.18)
190 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
100
25
j43.3
j57.7
Figure 6.10: Completed lossless L-network with low-pass topology
With the matching network the source sees 100 ; therefore, the power delivered to the
matching network is
P
in
=
1
2
(1/2)
2
100
=
1
800
= 1.25 mW (6.19)
Since the matching network is lossless, all of this power will be delivered to the load. Thus,
with the matching network, P
out
= 1.25 mW. The improvement gained is
10 log(1.25/.8) ' 2 dB (6.20)
6.2.4 Matching Complex Loads with a Lossless L-network
So far, only purely resistive source and load terminations have been considered. When
complex source and loads are involved, there are two basic conceptual approaches that can
be used:
1. Absorption - absorb the source or load reactance into the matching network.
2. Resonance - series or parallel resonate the source or load reactance at the frequency
of interest.
These approaches will be illustrated by example in the following sections.
6.2.4.1 Example - Absorption
Absorption will be illustrated with an example. Suppose that it is necessary to match the
source and load shown in Figure 6.11 at 100 MHz with a lossless L-network.
100
750
j126
Lossless
Matching
Network
j250
Figure 6.11: L-network with complex source and load
Absorption is applied by lumping the source and load reactances into the series and
parallel reactances of the matching network, as shown in Figure 6.12. The lumped reactances
6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 191
100
750
j126
j250
jX
0
p
jX
0
s
jX
p
jX
s
Figure 6.12: Absorbing the complex part of the load impedance into the network
X
0
s
and X
0
p
can be found by using the design equations for matching between two resistive
terminations:
Q =
750/100 1 = 2.55
X
0
s
= 100 (2.55) = 255
X
0
p
= 750/2.55 = 294.1
(6.21)
The lumped reactances can be written in terms of the reactances associated with the source
and load, and the L-network reactances:
X
0
s
= X
s
+ 126, (6.22)
X
0
p
=
250 (X
p
)
250 + X
p
. (6.23)
Thus, values of X
s
and X
p
can be obtained:
X
s
= X
0
s
126 = 255 126 =
129
381
X
p
=
250X
0
p
250+X
0
p
=
1667.2
135.1
(6.24)
The two solutions found so far are shown in Figure 6.13.
j1667.2
2.65H
j129
0.205H
j135.1
0.215H
j381
4.18pF
Figure 6.13: Two L-network solutions.
192 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
The element values (in H and pF) were obtained from the calculated reactances using
the design frequency of 100 MHz. It should be noted that the two solutions found so far
are not the only possibilities for this particular source and load. This becomes apparent if
series-to-parallel transformations are applied to the source and load. After transformation,
the source and load look like Figure 6.14.
258.8 75 j205.4
j225
Figure 6.14: Source and load after application of series-to-parallel transformation
For simplicity, the Norton equivalent current source has been omitted from the source
representation. Note that the smaller of the two resistances is now on the load side and
hence it is possible to match the source and load using L-networks oriented as in Figure
6.15.
jX
s
jX
p
Figure 6.15: L-network reoriented for smaller resistance on load side
The values of X
s
and X
p
could be found using absorption. Alternatively, the reso-
nance concept can be used. For illustration, the resonance concept will be used to nd the
remaining L-network solutions.
6.2.4.2 Example - Resonance
Continuing with the example from the previous section, the resonance concept will be em-
ployed to nd two more L-network solutions for the source and load shown in Figure 6.11.
After transforming the source from series to parallel form, and the load from parallel to
series form, the source and load are represented as in Figure 6.14. To apply the resonance
concept, the source and load are augmented with reactances that resonate with the source
and load reactances as shown in Figure 6.16.
After resonating the source and load in this manner, the new source and load impedances
are purely real (resistive) at the design frequency. An L-network is then designed to match
these two resistances, i.e.,
6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 193
258.8 75
j205.4
j225
j205.4
j225
Figure 6.16: Transformed source and load augmented with resonating reactances
Q =
258.76
75
1 = 1.57
X
0
s
= 75 (1.57) = 117.8
X
0
p
= 258.76 /1.57 = 164.8
(6.25)
Now the circuit can be drawn as shown in Figure 6.17. To complete the design, the resonating
258.8 75
j205.4
j225
j205.4
j225
j164.8
j117.8
Figure 6.17: L-net with resonating reactances
reactances must be incorporated into the matching network. For example, when the upper
signs are chosen, the net parallel reactance will be j205.4|| j164.8 = j91.4. The
net series reactance will be j117.8 + j225 = j342.8. The nal solutions are shown in
Figure 6.18.
j833.7
j107.2
j91.4
j342.8
Figure 6.18: Resonating reactances incorporated into matching network
Thus, we have found four possible solutions that can be obtained using a lossless L-
network.
The example considered here allowed 4 possible L-network solutions because transform-
ing the source and load caused R
big
and R
small
to swap positions. This will not always be
the case. Thus, for some source and load combinations, there will be only two L-network
solutions, and in other cases there will be four solutions.
194 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
6.3 Harmonic Attenuation in Lossless Matching Networks
Using Traps
In certain applications it is necessary to provide a match at some frequency !
o
and to
attenuate one or more other frequencies. Typically the undesired frequencies are harmonics
of !
o
, i.e., 2!
o
, 3!
o
, etc.; however, they could also be subharmonic or even unrelated to !
o
.
We will concentrate on the case where the undesired frequencies are harmonics. General-
ization to other cases is straightforward. The basic idea is to rst design a lossless matching
network to provide a match at the desired frequency (!
o
). This matching network might
consist of an L-, T-, or Pi-network. Next, one or more series elements of the network are
replaced with parallel L-C-networks. Alternatively, or in addition, the shunt elements of
the network are replaced with series L-C-networks. The replacement (series or parallel L-C)
elements are designed to have the same reactance as the original elements at !
o
; however,
they are designed to be resonant at the undesired harmonic frequency. As an example,
consider Figure 6.19 which shows a matching network with a match at f
o
= 10
7
/2 Hz.
jX
C
= j134.2
jX
L
= j111.8
50 300
Figure 6.19: L = 11.18 H, C = 745.2 pF
Suppose it is desired to trap the second harmonic 2f
o
in the shunt arm. The shunt
arm would be replaced with a series L-C that has reactance 134.2 at f
o
and is series
resonant (looks like a short circuit) at 2f
o
. The design equations are
2 !
o
= 1/
p
LC (6.26)
!
o
L
1
!
o
C
= 134.2 (6.27)
The solution is
L = 4.47 H (6.28)
C = 558.9 pF
The new network looks like Figure 6.20.
The series trap acts to shunt the component at frequency 2f
o
to ground.
The third (or any other) harmonic could be trapped in the series arm. The inductor
would be replaced with a parallel LC whose element values are found from
3!
o
=
1
p
LC
(6.29)
1
!
o
L
+ !
o
C =
1
111.8
(6.30)
6.3. HARMONIC ATTENUATIONINLOSSLESS MATCHINGNETWORKS USINGTRAPS195
558.9pF
11.18H
300
4.47H
!
0
2!
0
Figure 6.20: Network with trapped second harmonic
The solution is
C = 111.8 pF (6.31)
L = 9.94 H
Figure 6.21 shows the network with both traps installed. The network will look like an open
circuit to the 3f
o
component.
558.9pF
300
4.47H
!
0
2!
0
111.8pF
9.94H
3!
0
Figure 6.21: Network with second and third harmonics trapped
It should be noted that this approach will work only when applied to capacitive shunt
elements and inductive series elements if frequencies higher than !
o
are to be trapped. Thus,
the original network must have a low-pass topology. The student should convince himself
or herself that such is the case. On the other hand, if frequencies smaller than !
o
are to
be trapped, then the approach can be employed only with inductive shunt elements and
capacitive series elements.
The trapping approach is also useful for reducing the feed-through of nonharmonic fre-
quency components. If the frequencies of the undesired components are very close to the
desired frequency, it may not be possible to build an eective trap, because the Q of the
series or parallel L-C circuits will be too high to be realized.
196 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
6.4 Three-element Matching Networks
The L-network does not give the designer freedom to choose the Q (bandwidth) or phase
shift of the matching network. The addition of a third matching element makes it possible
to design for a match and a specied phase shift or Q. The following section describes a
procedure for designing 3 element matching networks with specied Q. Then, a general
procedure allowing for specied attenuation and phase shift will be discussed.
6.4.1 Design of Pi- and T-networks for Specied Bandwidth (Q)
First consider the Pi-network with a low-pass topology. In addition, we assume that ab-
sorption has been used so that the shunt reactance of the source or load impedance is
included in the Pi-network. The problem is then reduced to matching between two resistive
terminations as shown in Figure 6.22 (where it is assumed that R
2
> R
1
).
jX
C1
jX
C2
jX
L
R
1
R
2
Figure 6.22: Matching two resistive terminations where R
2
> R
1
The Pi-network can be thought of as two back-to-back L-networks that act to match
both R
1
and R
2
to a virtual resistance R
v
as shown in Figure 6.23.
jX
C1
jX
C2
jX
L2
R
v
R
2
jX
L1
R
1
R
v
Figure 6.23: The Pi-network as 2 back-to-back L-networks
Because the series arms of both L-networks are connected to R
v
, it is clear that R
v
is
smaller than R
1
and R
2
. Dene the Qs of the two L-networks to be Q
1
and Q
2
where
Q
1
=
R
1
R
v
1 , and Q
2
=
R
2
R
v
1 (6.32)
We are assuming that R
2
> R
1
, and therefore Q
2
will be larger than Q
1
. For most practical
purposes the Q of the Pi-network can be approximated by Q
2
. This is especially true if
R
2
R
1
. If R
2
is only slightly larger than R
1
, then the overall Q of the network will be
somewhat larger than Q
2
. The design procedure follows:
6.4. THREE-ELEMENT MATCHING NETWORKS 197
1. Determine the required Q of the matching network by considering the required band-
width, BW, and center frequency, f
o
(Q = f
o
/BW). This Q is taken to be equal
to Q
2
, and thus the virtual resistance, R
v
, is determined. Note that R
v
must be
smaller than R
1
, and therefore the Pi-network can only be used to obtain
a larger Q than would have been provided by the simpler L-network. Also
note that the relationship BW = f
o
/Q is only exactly true for a simple parallel or se-
ries RLC circuit. Thus, the actual bandwidth of your circuit may be dierent from the
specied design value. If a particular design requires that the bandwidth be precisely
determined, it is a good idea to simulate the performance of the matching network
using a computer-aided design program in order to verify that the performance will
be satisfactory.
2. Once R
v
is found, the values of X
C1
, X
L1
, X
C2
, and X
L2
can be calculated using the
previously derived formulas for L-network matching.
The design procedure can be summarized by Equations 6.33, where Q
2
is determined by the
desired bandwidth. You should verify that these equations result from steps (1) and (2):
X
C2
=
R
2
Q
2
(6.33)
X
C1
=
R
1
R
2
(Q
2
2
+ 1)
R2
R1
(6.34)
(6.35)
(6.36)
X
L
=
R
2
Q
2
+R
2
R1
R2
(Q
2
2
+ 1) 1
Q
2
2
+ 1
(6.37)
The Pi-network is most useful for matching when the values of R
1
and R
2
are not too small.
If R
1
and R
2
are small, the virtual resistance will be even smaller, and the capacitor values
will turn out to be impractically large. If either terminating resistance is signicantly less
than 50 , the T-network will usually be a more practical choice. One possible T-network
is the band-pass case shown in Figure 6.24.
jX
C1
jX
L2
R
2
R
1 jX
L3
Figure 6.24: Band-pass T-network
As before, we can think of this network as two back-to-back L-networks as shown in
Figure 6.25.
The Qs of the two L-networks are
Q
1
=
R
v
R
1
1 and Q
2
=
R
v
R
2
1 (6.38)
198 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
jX
C1
jX
0
C
jX
L2
R
v
R
2
R
1
R
v
jX
0
L
Figure 6.25: Band-pass T-network as 2 back-to-back L-networks
Note that since R
v
> R
1
and R
v
> R
2
, this network will have a larger Q than a single
L-network that matches R
1
to R
2
. The overall Q of the network is set by Q
1
since we
assume R
2
> R
1
. The design formulas are
X
C1
= R
1
Q
1
(6.39)
X
L2
= R
2
R
1
R
2
(Q
2
1
+ 1) 1 (6.40)
(6.41)
X
L3
=
R
1
(Q
2
1
+ 1)
Q
1
R1
R2
(Q
2
1
+ 1) 1
(6.42)
Note carefully that the two elements X
0
L
and X
0
C
that appear in the back-to-back L-networks
are combined into a single element, X
L3
, in the T-network. In practice this is not necessary
and, in fact, may not be desirable. Whether or not the elements are combined makes no
dierence at the design frequency, but it may make a signicant dierence at frequencies
well removed from the design frequency. The dierent possibilities are best examined using
a CAD program.
You should be able to derive similar formulas for the other possible topologies, e.g.,
band-pass Pi, low-pass T, etc.
6.4.2 Matching Two Resistive Terminations with Specied Atten-
uation and Phase Shift
In this section we will consider a more general type of matching network than was considered
in section 6.2. Specically, we allow the matching element impedances to be complex in the
initial development of our solution. This will allow for a solutions with specied attenua-
tion and phase shift. Then we will specialize to the cases where the elements are purely
reactive (lossless networks) and purely resistive (lossy networks or attenuators). Note that
the terminating impedances are assumed to be real (resistive).
6.4.2.1 Pi-network with specied attenuation and phase shift
Referring to Figure 6.26 we make the following assumptions:
Y
1
, Y
2
are assumed to be real.
6.4. THREE-ELEMENT MATCHING NETWORKS 199
Y
A
, Y
B
, Y
C
may be complex.
Y
1
Y
A
Y
2
Y
B
Y
C
I
IN
I
OUT
V
OUT
+
-
V
IN
+
-
Y
1
Y
2
Figure 6.26: Matching two resistive terminations with a Pi-network
The conditions required for a match can be summarized as follows:
1. Must see Y
1
looking in from the left when terminated with Y
2
on the right. Thus,
Y
1
= Y
A
+
(Y
2
+ Y
B
)Y
C
Y
2
+ Y
B
+ Y
C
(6.43)
2. Must see Y
2
looking in from the right when terminated with Y
1
on the left. Thus,
Y
2
= Y
B
+
(Y
1
+ Y
A
)Y
C
Y
1
+ Y
A
+ Y
C
(6.44)
So far, we have 2 equations but 3 unknowns. The third equation can be used to specify either
the Q of the network (and therefore, the bandwidth) or the phase shift and attenuation.
For now, we consider the latter. The third equation can be written in the form:
e
+j
=
V
IN
I
IN
V
OUT
I
OUT
(6.45)
Equation 6.45 can be interpreted as follows. Since Y
1
and Y
2
are assumed to be real,
the input voltage and current will be in phase and so will the output voltage and current.
Denoting the phase of the input voltage (current) by
IN
and the phase of the output voltage
(current) by
OUT
e
+j
=
|V
IN
| |I
IN
|
|V
OUT
| |I
OUT
|
e
j(
IN
OUT
)
(6.46)
=
P
IN
P
OUT
e
j(
IN
OUT
)
Thus
=
IN
OUT
(6.47)
and
=
1
2
ln
P
IN
P
OUT
(6.48)
200 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
Therefore, is the phase shift of the network and is the power attenuation expressed in
nepers.
Equations 6.43, 6.44 and 6.45 must be solved for the unknowns Y
A
, Y
B
and Y
C
. Note that
Equation 6.45 can be written in terms of the unknowns as shown below. Dene = + j,
then
e
V
IN
I
IN
V
OUT
I
OUT
=
V
IN
V
OUT
I
IN
/V
IN
I
OUT/V
OUT
(6.49)
(6.50)
=
V
IN
V
OUT
Y
1
Y
2
(6.51)
(6.52)
e
= (1 +Y
1
C
(Y
2
+Y
B
))
Y
1
Y
2
(6.53)
Equations 6.43, 6.44 and 6.45 can now be solved. The result is
Y
C
=
p
Y
1
Y
2
sinh
(6.54)
Y
B
=
Y
2
tanh
Y
C
(6.55)
Y
A
=
Y
1
tanh
Y
C
(6.56)
6.4.2.2 T-network with specied attenuation and phase shift
Similar considerations apply to using a T-network to match two resistive terminations as
shown in Figure 6.27.
Z
1
Z
B
Z
C
I
IN
I
OUT
V
OUT
+
-
V
IN
+
-
Z
A
Z
2
Z
2
Z
1
Figure 6.27: Matching two resistive terminations with a T-network
The equations that ensure a simultaneous conjugate match at both ports are
Z
1
= Z
A
+
Z
C
(Z
B
+ Z
2
)
Z
C
+ Z
B
+ Z
2
(6.57)
Z
2
= Z
B
+
Z
C
(Z
A
+ Z
1
)
Z
C
+ Z
A
+ Z
1
(6.58)
6.4. THREE-ELEMENT MATCHING NETWORKS 201
The third equation is the same as before,
e
+j
= e
V
IN
I
IN
V
OUT
I
OUT
(6.59)
which can be written
e
=
I
IN
I
OUT
Z
1
Z
2
(6.60)
The solutions for Z
A
, Z
B
and Z
C
are
Z
C
=
p
Z
1
Z
2
sinh
(6.61)
Z
B
=
Z
2
tanh
Z
C
(6.62)
Z
A
=
Z
1
tanh
Z
C
(6.63)
The interpretation of = +j is the same for the T-network as it was for the Pi-network,
that is,
=
IN
OUT
) voltage (current) phase shift in radians
=
1
2
ln
P
IN
P
OUT
) power attenuation in nepers
(6.64)
Remember that these discussions have assumed that Y
1
and Y
2
(or Z
1
and Z
2
) are real
(resistive). Complex loads are handled by incorporating the reactive part of the termination
into the matching network using either resonance or absorption as discussed in section 6.2.
6.4.3 Design of Lossless Pi- and T- Matching Networks with Spec-
ied Phase Shift
The solutions found so far allow the designer to specify both phase shift and attenuation of
the network. In practice one is usually concerned with one of the special cases: (i) Y
A
, Y
B
,
Y
C
or Z
A
, Z
B
, Z
C
are purely reactive this corresponds to the lossless matching network;
(ii) Y
A
, Y
B
, Y
C
orZ
A
, Z
B
, Z
C
are purely resistive this corresponds to a lossy network.
Networks of type (ii) are often used to provide specied amounts of attenuation and/or
isolation between circuits. In this section we will consider the lossless matching networks.
When Y
A
, Y
B
, Y
C
or Z
A
, Z
B
, Z
C
are purely reactive, then the network is lossless and
= 0. Hence = j and the design equations reduce to
Pi:
Y
C
=
p
Y
1
Y
2
j sin
(6.65)
Y
B
=
Y
2
j tan
Y
C
(6.66)
Y
A
=
Y
1
j tan
Y
C
(6.67)
202 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
T:
Z
C
=
p
Z
1
Z
2
j sin
(6.68)
Z
A
=
Z
1
j tan
Z
C
(6.69)
Z
B
=
Z
2
j tan
Z
C
(6.70)
A word of caution on interpreting when source and/or load admittances are complex.
We have seen that when the source and load are resistive, can be interpreted as the voltage
or current phase shift, since voltage and current are in phase at both the input and output.
If the source and/or load are complex, then the voltage and current are not in phase. In
this situation has a dierent interpretation for the Pi- and T-networks. This can be seen
by considering Figure 6.28.
jX
B
jX
C
I
IN
I
OUT
V
OUT
+
-
V
IN
+
-
jX
A
jX
2
R
2
R
1
jX
1
Figure 6.28: T-network with complex source and load
The network would be designed by incorporating X
1
and X
2
into the matching network
as shown in Figure 6.29.
j(X
B
+ X
2
)
jX
C
I
IN
I
OUT
V
0
OUT
+
-
V
0
IN
+
-
j(X
1
+ X
A
)
R
2
R
1
Figure 6.29: Network designed with X
1
and X
2
incorporated into the matching network
The currents I
IN
and I
OUT
in Figure 6.29 are the same as those shown in Figure 6.28,
but the voltages are not. In fact, V
0
IN
and V
0
OUT
are related to the V
IN
and V
OUT
of Figure
6.28 by
V
0
OUT
= V
OUT
R
2
R
2
+ j X
2
(6.71)
V
IN
= V
0
IN
R
1
jX
1
R
1
(6.72)
6.4. THREE-ELEMENT MATCHING NETWORKS 203
Now, is the phase shift between I
IN
and I
OUT
or, equivalently, V
0
IN
and V
0
OUT
. It is
not the phase shift between V
IN
and V
OUT
, however. The voltage phase shift (
V
IN
V
OUT
)
can be determined as follows:
V
IN
V
OUT
=
V
0
IN
V
0
OUT
R
1
jX
1
R
2
+ jX
2
R
2
R
1
(6.73)
Dene
Z1
= tan
1
X
1
R
1
(6.74)
Z2
= tan
1
X
2
R
2
(6.75)
Then
V
IN
V
OUT
=
0
V
IN
0
V
OUT
Z1
Z2
=
Z1
Z2
(6.76)
Thus for the lossless T-network gives the current phase shift, and the voltage phase shift
can be found from Equation 6.76.
Similar considerations lead to the conclusion that gives the voltage phase shift for the
lossless Pi-network. In this case Equation 6.77 yields the current phase shift:
I
IN
I
OUT
=
Y1
Y2
(6.77)
6.4.3.1 Example - Design of a lossless Pi-network with specied phase shift
Design a lossless Pi-network to match a load of 300 to a 50 source with V
OUT
leading
V
IN
by 45
at ! = 10
7
rad/s as in Figure 6.30. We want
V
OUT
V
IN
= 45
. Hence
V
OUT
V
IN
45
.
204 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
= 45
Y
1
=
1
50
= 20 mS
Y
2
=
1
300
= 3.3 mS
Y
C
=
p
20(3.33)
j sin(/4)
= +j 11.54 mS
Y
B
=
3.33 mS
j tan(/4)
j1154 mS = j 8.21 mS
Y
A
=
20 mS
j tan(/4)
j1154 mS = +j 8.46 mS
(6.78)
The nal solution is shown in Figure 6.31. The network will have a band-pass transfer
846pF 12.2H
1154pF
50 300
Figure 6.31: Pi-network solution
function characteristic, as the series capacitance and shunt inductor guarantee that the
network transfer function will have zero response at DC, whereas the shunt capacitor will
provide high frequency attenuation which increases at 6 dB per octave for frequencies well
above the design frequency.
Complex source or load impedances can be handled by incorporating the load reactances
into the network, as illustrated in the following example.
6.4.3.2 Example - Matching complex load with a specied current phase shift.
At ! = 10
7
rad/s design a lossless matching network to match a load impedance of 150 - j
75 to a generator having an internal impedance of 50 . The output current is to be in
phase with the input current as in Figure 6.32. Start by trying to nd a T-network solution:
j75
150
50
I
IN
I
OUT
Figure 6.32: Example
=
I
IN
I
OUT
= 0 (6.79)
sin = 0
tan = 0
6.4. THREE-ELEMENT MATCHING NETWORKS 205
Thus, Z
A
, Z
B
, Z
C
) 1. This illustrates that 0 current phase shift cannot be obtained
with a T-network!
Let us consider a Pi-network. For the Pi-network, is the voltage phase shift. As noted
earlier, can be written in terms of the current phase shift and the phase angles of the
terminating impedances:
=
I
IN
I
OUT
Z1
Z2
= 0 0 0
Z2
=
Z2
Z2
= tan
1
75
150
= 26.6
(6.80)
Thus = +26.6
= j 23.07 mS
Y
A
=
1/50
j tan 26.6
+j 23.07 mS = j 16.9 mS
(6.81)
1
j 375
+Y
B
=
1/187.5
j tan 26.6
+j 23.07 mS
Y
B
= j 9.73 mS (6.82)
The nal result is shown in Figure 6.34.
5.92H 973pF
4.34H
Figure 6.34: Final Result.
This example illustrated how to properly account for phase shifts in complex load
impedance. A similar approach would be necessary if the source impedance was complex.
We now turn from the purely reactive 3-element matching networks to a discussion of
purely resistive networks.
206 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
6.4.4 Resistive Three-element Matching Networks
The purely resistive matching networks are also known as attenuators, attenuating pads,
or just pads. Here we assume Y
A
, Y
B
, Y
C
(or Z
A
, Z
B
, and Z
C
) are real (resistances). The
resistive matching network can be employed to provide a broadband match between two
resistive terminations. Of course, some power loss must be accepted. This type of network is
used to build attenuators with specied attenuation. In addition, as we will see, the resistive
network can be used to ensure that a particular device sees a well-dened impedance, even
when the input of the following device has a variable or unknown impedance.
In this application Y
A
, Y
B
, Y
C
are real (resistive). The terminating impedances are also
assumed to be resistive. There will be no phase shift and hence
= (real) nepers (6.83)
As noted previously
2 = ln
P
IN
P
OUT
(6.84)
Thus, determines the attenuation of the network. It is useful to relate to the attenuation
in dB:
Attenuation in dB = 10 log
P
IN
P
OUT
(6.85)
= 20 log e
= 8.686
If a network that provides 10 dB of attenuation is required, would be 10/8.686 = 1.1513
nepers.
6.4.4.1 Example - Design a 10 dB Pi-type resistive attenuator
Design a 10 dB Pi-type attenuator for use in a system with 75 impedance (R
1
= R
2
=
75 ) as in Figure 6.35.
= 10/8.686 = 1.1513
Y
1
= Y
2
=
1
75
= 13.33 mS
Y
C
=
p
Y1Y2
sinh
=
13.33 mS
1.423
= 9.370 mS
Y
A
= Y
B
=
13.33 mS
tanh
Y
C
= 6.922 mS
Z
A
= Z
B
= 144.5 ' 145
Z
C
= 106.7 ' 107
The resistive matching networks/attenuators have the useful characteristic of isolating
the impedance found at the input from the impedance that terminates the output. This can
6.4. THREE-ELEMENT MATCHING NETWORKS 207
145 145
107
75 75
Figure 6.35: 10 dB Pi-type attenuator with 75 impedance
be seen by computing the input impedance for the extreme cases of shorted and open-circuit
output terminations in the example considered above:
shorted output : Z
IN
= 61.6
open output : Z
IN
= 92.0
Clearly, for any resistive termination, the input impedance will be in the range 61.6 to 92.0
. This property can be used to advantage when the input impedance of a particular stage
is not well known or is subject to variation. If this stage follows a stage that requires a
certain load impedance in order to operate correctly, an attenuator can sometimes be an
eective isolation stage. For example, for proper operation of certain types of frequency
mixers, it is necessary that all of the frequency components at the output of the mixer see
a well-dened impedance (usually 50 ). Typically the frequencies at the output of a mixer
may span a very broad range, while the following stage is often a narrow-band IF amplier.
The IF amplier would often have a well dened input impedance only within its passband.
A resistive matching network is sometimes employed between the mixer and the IF amplier
in order to ensure proper termination of the mixer output port.
It is important to note that the attenuation of a resistive matching network will only
be equal to the design value if the network is operated between the impedances for which
it was designed. It is important to remember this, since the resistive networks are often
found in applications where the terminating impedances are substantially dierent from the
correct values. The example cited in the previous paragraph is one such case.
6.4.4.2 Minimum-loss Resistive Matching Networks
In some cases, especially when a broadband impedance match is required, one would like to
design a resistive network that has the smallest possible attenuation (). The problem is
then to nd the solution that yields the smallest subject to the constraint that all resistive
elements have values greater than or equal to zero. This constraint ensures that the solution
corresponds to a passive network.
Consider the T-network in Figure 6.36.
R
C
=
p
R
1
R
2
sinh
(6.86)
R
B
=
R
2
tanh
p
R
1
R
2
sinh
(6.87)
R
A
=
R
1
tanh
p
R
1
R
2
sinh
(6.88)
208 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
R
A
R
1 R
2
R
B
R
C
Figure 6.36: T-network
The goal is to nd the smallest that gives values for R
A
, R
B
, R
C
which are all 0.
Note from Equation 6.86 that
sinh =
p
R
1
R
2
R
C
(6.89)
Since sinh increases monotonically as a function of , as shown in Figure 6.37, minimizing
is equivalent to minimizing sinh . Clearly then, we should use the largest value of R
C
Figure 6.37: sinh as a function of
for which R
B
0 and R
A
0 (see Equation 6.89). Using
cosh =
1 + sinh
2
(6.90)
=
1 + R
1
R
2
/R
C
2
(6.91)
we nd from Equation 6.87 and Equation 6.88
R
B
= R
2
R
2
C
R
1
R
2
+ 1 R
C
(6.92)
R
A
= R
1
R
2
C
R
1
R
2
+ 1 R
C
(6.93)
For the moment, assume R
2
> R
1
. Then, as R
C
is increased in order to decrease sinh(),
R
A
will decrease and will become equal to zero when (see Equation 6.93)
R
C
=
R
1
1 R
1
/R
2
(6.94)
6.4. THREE-ELEMENT MATCHING NETWORKS 209
Be aware that R
B
will always be greater than zero if R
2
> R
1
. So, if R
2
> R
1
, the minimum
attenuation is obtained when R
C
is given by Equation 6.94. The corresponding value for
(denoted by
min
) is obtained from Equation 6.91 and is
min
= cosh
1
R
2
/R
1
(6.95)
The value for R
B
is obtained by using Equation 6.94 in Equation 6.92 to yield
R
B
= R
2
1 R
1
/R
2
(6.96)
Notice that since R
A
= 0, the minimum-loss resistive network reduces to an L-network with
the series arm (R
B
) connected to the larger of the two terminating resistances.
The preceding discussion was based on the T-network, although the nal result was
found to be an L-network. The same result would have been obtained if the starting point
had been the Pi-network.
6.4.4.3 Summary of Resistive Minimum-loss Network
The resistive minimum-loss network reduces to an L-network with series arm connected to
the larger of the terminating resistances as in Figure 6.38.
R
series
R
big R
small
R
shunt
Figure 6.38: Resistive minimum-loss network
Loss in dB = 8.686 cosh
1
R
big
/R
small
R
shunt
=
R
small
p
1R
small
/R
big
R
series
= R
big
1 R
small
/R
big
(6.97)
210 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
6.5 References
1. Smith, Jack, Modern Communication Circuits, McGraw-Hill, 1986.
2. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw Hill, 1943.
6.6. HOMEWORK PROBLEMS 211
6.6 Homework Problems
1. You are given a black box with two output terminals (a 1-port"). You play around
with the box for awhile and make the following observations:
(a) The output voltage from the box is sinusoidal.
(b) The peak magnitude of the open circuit voltage at the output of the box is
found to be 5 V.
(c) You connect a 50 resistor across the terminals and nd that the peak magnitude
of the voltage across the resistor is 2.795 V.
(d) You short the output of the box and nd the peak magnitude of the short
circuit current is 100 mA.
Find the power available from the source. Express your result in dBm.
2. Consider a source with impedance Z
S
= R
S
+ jX
S
and suppose that the source drives
a load that is purely resistive. Denote the load resistance by R
L
.
(a) What load resistance should be used if the goal is to maximize the power delivered
to the load?
(b) Find an expression for the power delivered to the load resistance found in part 2a.
Express your result in terms of the source parameters only, i.e., P
avs
, R
S
and
X
S
.
3. Find the lossless network having the minimum number of elements to match a source
impedance of 20-j60 to a load impedance having an equivalent circuit of 100 of
resistance shunted by 50 of capacitive reactance. Draw the network and label all
parts.
4. Consider the design of a lossless L-network to match the source and load shown in Fig-
ure 6.39. All resistances and reactances are in ohms, and the current source magnitude
is the peak value.
Lossless
Matching
Network
1 mA
850 j100
j75
50
Figure 6.39: Complex source and load to be matched with a lossless L-network.
(a) Find the power available from the source. Express your result in dBm.
(b) How much power would be delivered to the load if a lossless matching network
was not used, i.e., if the load is connected directly to the source? Express your
result in dBm.
212 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
(c) There are four possible solutions for the matching network if an L-network is
used. Find all four. Sketch all solutions and indicate whether the elements are
inductors or capacitors.
(d) Verify two of your designs by plotting the path from the load to the source on a
Smith Chart.
5. Consider a source with impedance Z
S
= R
S
+ jX
S
and a load that is purely resistive.
Denote the load resistance by R
L
. Suppose that R
S
< R
L
. Under what condition(s)
are there more than 2 lossless L-networks that conjugately match the source to the
load? Derive a simple inequality that can be checked to determine whether there are
2 or 4 solutions.
6. The source and load shown in Figure 6.40 can be matched with an L-network that
consists of two capacitors. All resistances and reactances are given in ohms.
1 mA 75 j100
jX
1
jX
2
j40
20
Figure 6.40: Source and load matched with 2-capacitor L-network.
Find X
1
and X
2
that will cause all of the available source power to be delivered to
the load. Both X
1
and X
2
must be <0.
7. Design a matching network to match the source and load shown in Figure 6.41. Use a
T-network such that I
out
lags I
in
by 60
at ! = 10
7
. What is the voltage phase shift?
j100
200
j50 50
I
in
V
in
V
out
I
out
+
V2=0
Input admittance with output shorted (7.2)
Y
21
=
I
2
V
1
V2=0
Forward transfer admittance, output shorted (7.3)
Likewise, if V
1
is forced to zero by shorting the input, Y
22
and Y
12
can be measured using
Y
22
=
I
2
V
2
V1=0
Output admittance with input shorted (7.4)
Y
12
=
I
1
V
2
V1=0
Reverse transfer admittance, input shorted (7.5)
All of the Y-parameters have units of admittance. Y-parameters are determined experi-
mentally by measuring input and transfer admittances with one port driven and the other
port terminated in a short circuit. Therefore the standard, or reference, termination for the
Y-parameter set is a short circuit.
7.1. INTRODUCTION 223
7.1.2 Z-parameters
Z-parameters or impedance parameters were used in chapter 5 to characterize the proper-
ties of transformers. Z-parameters are dened such that the currents are the independent
variables and the voltages are dependent, i.e.,
V
1
= Z
11
I
1
+Z
12
I
2
V
2
= Z
21
I
1
+Z
22
I
2
(7.6)
The Z-parameters can be interpreted as follows:
Z
11
=
V
1
I
1
I2=0
Input impedance, output open circuited (7.7)
Z
21
=
V
2
I
1
I2=0
Forward transfer impedance, output open circuited (7.8)
Z
22
=
V
2
I
2
I1=0
Output impedance, input open circuited (7.9)
Z
12
=
V
1
I
2
I1=0
Reverse transfer impedance, input open circuited. (7.10)
Z-parameters have units of impedance, and measurement of Z-parameters involves deter-
mining input, output, and transfer impedances with one of the ports driven and the other
open-circuited. Thus, the reference impedance for the Z-parameters is an open circuit.
7.1.3 Hybrid (h) parameters
V
1
= h
11
I
1
+h
12
V
2
I
2
= h
21
I
1
+h
22
V
2
(7.11)
Hybrid parameters or h-parameters use both a short circuit and an open circuit as
reference terminations. The reference termination is an open circuit for the input port and
a short circuit for the output port. This is easily recognized by considering h
11
and h
22
:
h
11
=
V
1
I
1
V2=0
Input impedance with output short circuited (7.12)
h
22
=
I
2
V
2
I1=0
Output admittance with input open circuited (7.13)
Note that the parameters h
11
and h
22
have dierent units (impedance and admittance,
respectively). This is the origin of the hybrid designation.
A common characteristic of the Y-, Z-, and h-parameter sets is that the independent
variables are forced to zero when the 2-port is terminated in a standard, or reference,
impedance. For example, the independent variables in the Y-parameter representation are
the voltages at the input and output ports. To force one of the independent parameters to
zero, it is necessary to terminate one of the ports with a short circuit. The independent
variables in the Z-parameter representation are the input and output currents. Since input
or output currents are forced to zero when a port is terminated with an open circuit, the
reference impedance for the Z-parameter set is said to be an open circuit. The reference
impedance for the h-parameter set is an open circuit for the input port and a short circuit
for the output port. Thus the choice of independent variables determines the reference
impedance for the parameter set.
224 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
7.1.4 ABCD-parameters
ABCD-parameters are dened such that the port 1 variables depend on the port 2 variables:
V
1
= AV
2
BI
2
I
1
= CV
2
DI
2
(7.14)
The ABCD parameter representation diers from the Y-, Z-, and h-parameter sets in that
the independent variables are both associated with the output port. Thus, it is not possible
to dene a xed reference impedance termination for the input and output ports that will
force the independent variables to zero. Nevertheless, this parameter set is very useful
because the ABCD matrix for a cascade of 2-ports is the matrix product of the individual
ABCD matrices, as shown in the next section.
7.2 Special types of 2-ports and their matrix properties
7.2.1 Reciprocal 2-ports
A 2-port network that does not contain any sources and is constructed only from any
combination of resistors, inductors, capacitors, transformers and transmission lines will obey
the principle of reciprocity, which states that the current (voltage) response at one port due
to a voltage (current) excitation at another port is independent of which port is excited and
which port the response is measured at. Figure 7.2 illustrates this principle for the case
(a)
V
2
I
1
+
Reciprocal
2-port
(b)
V
2
I
1
+
Reciprocal
2-port
Figure 7.2: If the 2-port is reciprocal, voltage response at port 2 due to current I
1
at port
1 is the same as voltage response at port 1 due current I
1
at port 2.
of the voltage response due to a current excitation. Since the voltage response at port i
due to a current excitation at port j is Y
ij
, reciprocal 2-ports satisfy Y
12
= Y
21
. Note that
Y
11
is not necessarily equal to Y
22
in a reciprocal network. If the network is reciprocal and
symmetric, then Y
12
= Y
21
and Y
11
= Y
22
.
Figure 7.3 illustrates the reciprocity principle for the case of the current response due to
(a)
V
1
I
2
+
Reciprocal
2-port
(b)
V
1
I
2
+
Reciprocal
2-port
Figure 7.3: If the 2-port is reciprocal, current response at port 2 due to voltage V
1
at port
1 is the same as current response at port 1 due to voltage V
1
at port 2.
7.3. PARALLEL, SERIES, CASCADE CONNECTIONS OF 2-PORT NETWORKS 225
a voltage excitation. A reciprocal 2-port will have Z
12
= Z
21
. A 2-port that is reciprocal
and symmetric will have Z
12
= Z
21
and Z
11
and Z
22
.
It is left as an exercise to show that the ABCD matrix of a reciprocal network will satisfy
ADBC = 1. In other words, the determinant of the ABCD matrix is unity for a reciprocal
2-port.
The lossless and lossy matching networks discussed in Chapter 6 are all examples of
reciprocal networks. Since two of the network parameters are the same, a reciprocal network
can be fully characterized (at each frequency) by 3 network parameters.
7.2.2 Reciprocal lossless 2-ports
A lossless 2-port contains no dissipative elements, and hence the time-averaged real power
absorbed by the network is zero. A 2-port network constructed only of lossless transmission
lines, transformers, and lossless Ls and Cs will be reciprocal and lossless. The constraint
that the power absorbed by the network is zero can be written as
1
2
<(V
1
I
1
) +
1
2
<(V
2
I
2
) = 0, (7.15)
where <( ) takes the real part of its argument. Using the Z parameters, equation 7.15 can
be written as
<(Z
11
|I
1
|
2
+Z
12
I
1
I
2
+Z
21
I
1
I
2
+Z
22
|I
22
|
2
) = 0.
If the network is reciprocal, then Z
12
= Z
21
so the constraint is
R
11
|I
1
|
2
+ 2R
21
<(I
1
I
2
) +R
22
|I
2
|
2
= 0, (7.16)
where the R
ij
= <(Z
ij
). Since equation 7.16 must be satised for any set of excitations,
(I
1
, I
2
), we conclude that R
ij
= 0 for a lossless, reciprocal 2-port. In other words, the Z
ij
are
purely imaginary. Similar reasoning can be used to show that the Y
ij
are purely imaginary
as well.
7.3 Parallel, series, cascade connections of 2-port net-
works
Any 2-port network can be viewed as being constructed from simpler 2-port networks whose
ports are connected in various ways. Analysis of complex 2-port networks can be greatly
simplied if the network is decomposed into a number of simpler interconnected 2-port
networks whose 2-port parameters are easily found. In the following sections, the parallel,
series, and cascade connections are considered. Other possibilities, such as parallel-series
(input ports connected in parallel and output ports in series) and series-parallel are left as
exercises.
In all of the following discussions related to interconnection of 2-ports, it is assumed that
the interconnection does not upset the basic property that says that the current owing into
one terminal associated with any port must ow out of the other terminal, i.e. referring
back to Figure 7.1 we assume that the relationships I
1
= I
0
1
and I
2
= I
0
2
are valid for each
of the interconnected 2-ports.
226 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
7.3.1 2-ports connected in parallel
Figure 7.4 shows a 2-port created by interconnecting two 2-ports with their input ports and
output ports connected in parallel. The Y-parameters of the constituent 2-ports can be
combined to obtain the Y-parameters of their parallel combination, as shown below.
I
a
1
I
a
2
Y
a
11
Y
a
12
Y
a
21
Y
a
22
V
a
1
V
a
2
+
-
+
-
V
b
1
V
b
2
I
b
1
I
b
2
+
-
+
-
Y
b
11
Y
b
12
Y
b
21
Y
b
22
V
2
I
2
+
-
V
1
I
1
+
-
Figure 7.4: Two 2-ports interconnected such that the input and output ports are in parallel.
The parallel connection results in the following relationships:
V
1
= V
a
1
= V
b
1
, V
2
= V
a
2
= V
b
2
(7.17)
I
1
= I
a
1
+I
b
1
, I
2
= I
a
2
+I
b
2
. (7.18)
In vector notation, e.g.
V =
V
1
V
2
, I =
I
1
I
2
, Y =
Y
11
Y
12
Y
21
Y
22
,
equations 7.17 and 7.18 can be written as:
V = V
a
= V
b
(7.19)
I = I
a
+I
b
. (7.20)
The voltage and current vectors associated with the individual 2-ports must satisfy
I
a
= Y
a
V
a
(7.21)
I
b
= Y
b
V
b
. (7.22)
Equations 7.19, 7.21, and 7.22 can be used in equation 7.20 to show that
I = {Y
a
+Y
b
}V. (7.23)
Thus, the Y-parameter matrix for the parallel combination of 2-ports is the sum of the
constituent Y-parameter matrices.
7.3. PARALLEL, SERIES, CASCADE CONNECTIONS OF 2-PORT NETWORKS 227
7.3.2 2-ports connected in series
Figure 7.5 shows a 2-port created by interconnecting two 2-ports with their input ports
and output ports connected in series. The Z-parameters of the constituent 2-ports can be
combined to obtain the Z-parameters of their series combination, as shown in Figure 7.5.
I
a
1
I
a
2
Z
a
11
Z
a
12
Z
a
21
Z
a
22
V
a
1
V
a
2
+
-
+
-
V
b
1
V
b
2
I
b
1
I
b
2
+
-
+
-
Z
b
11
Z
b
12
Z
b
21
Z
b
22
V
2
I
2
+
-
V
1
I
1
+
-
Figure 7.5: Two 2-ports interconnected such that the input and output ports are in series.
The series connection results in the relationships:
I
1
= I
a
1
= I
b
1
, I
2
= I
a
2
= I
b
2
V
1
= V
a
1
+V
b
1
, V
2
= V
a
2
+V
b
2
.
In vector notation the voltage and current vectors must satisfy
I = I
a
= I
b
(7.24)
V = V
a
+V
b
. (7.25)
The voltage and current vectors for the individual 2-ports must satisfy:
V
a
= Z
a
I
a
(7.26)
V
b
= Z
b
I
b
(7.27)
Equations 7.24, 7.26, and 7.27 can be used in equation 7.26 to show:
V = {Z
a
+Z
b
}I. (7.28)
Thus, the Z-parameter matrix for the series combination of 2-ports is the sum of the
constituent Z-parameter matrices.
228 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
I
a
1
I
a
2
A
a
B
a
C
a
D
a V
a
1
V
a
2
+
-
+
-
V
b
1
V
b
2
I
b
1
I
b
2
+
-
+
-
A
b
B
b
C
b
D
b
Figure 7.6: Two 2-ports in cascade.
7.3.3 Cascaded 2-ports
Figure 7.6 shows two 2-ports connected in cascade, i.e. the output of the rst 2-port drives
the input of the second 2-port.
The cascade connection results in the following relationships:
V
a
2
= V
b
1
, I
a
2
= I
b
1
(7.29)
Dene the input vector IN =
V
1
I
1
V
2
I
2
A B
C D
S
, it will deliver its maximum power, P
avs
, to the matching
network, so P
in
= P
avs
. Because the matching network is lossless, power delivered to
the load will also be P
avs
, so P
out
= P
avs
. This means that the transducer gain of a
lossless matching network is always equal to 1.
The transducer gain, G
T
, calculated for a particular set of source and load terminations
measures the performance of the 2-port in that system relative to that of a lossless
matching network. If G
T
> 1, then the power delivered to the load is greater than
P
avs
, which is better than could ever be achieved by using a lossless matching network.
On the other hand, if G
T
< 1, more power would be delivered to the load if the 2-port
was replaced with a lossless matching network.
230 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
Lossless
Matching
Network
Z
S
Z
in
= Z
S
Z
out
= Z
L
Z
L
P
avs
P
in
= P
avs
P
avo
P
out
= P
avs
Figure 7.8: Transducer power gain
Note that G
T
is generally a more useful quantity than is G for system design calcu-
lations. To see why this is so, note that it is possible to have a situation where the
operating power gain of an amplier in a particular system is G = 10 but the trans-
ducer gain in the same system is G
T
= .5. Here the power gain G looks good since
Pout
Pin
= 10 obviously, this 2-port must be contain an active device but G
T
says
wed be better o (i.e. more power would be delivered to the load) if the 2-port was
replaced with a passive, lossless matching network which, by denition, has G
T
= 1.
To reconcile the dierence between the two gains, take a look at Equation 7.36, which
shows that the ratio G
T
/G is equal to the fraction of the P
avs
that is delivered to the
input of the 2-port, i.e. the input mismatch factor:
G
T
G
=
Pout
Pavs
Pout
Pin
=
P
in
P
avs
(7.36)
=
.5
10
= 0.05
The problem with this amplier is the input mismatch just 5 percent of the available
power from the source is actually delivered to the 2-port! The situation could be
improved by adding a lossless matching network between the source and the amplier
which would cause all of the available power from the source to be delivered to the
input of the amplier (P
in
= P
avs
). The addition of the lossless matching network
would change the source impedance seen by the amplier, but this will not change the
operating power gain because G does not depend on Z
S
. So G would still be equal to
10 and hence P
out
= 10P
in
= 10P
avs
.
The operating power gain, G, only accounts for what happens to the power after it
gets in to the 2-port. It doesnt provide any information about what fraction of the
available power actually gets in.
3. Available Power Gain
G
A
Power available from the output of 2 port
Power available from the source
=
P
avo
P
avs
(7.37)
G
A
will depend only on the source impedance and the 2-port network parameters. It
describes how much power is potentially available from the output of the 2-port.
7.4. POWER GAIN DEFINITIONS 231
It may be helpful to summarize how the dierent power gains are related. First, the trans-
ducer gain will always be less than or equal to the operating power gain since
G
T
G
=
Pout
Pavs
Pout
Pin
=
P
in
P
avs
1 (7.38)
Thus, G
T
/G is equal to the impedance mismatch factor at the input port, which is equal
to one when P
in
= P
avs
, or when the source is conjugately matched to the input of the
2-port. The ratio given in equation 7.38 can be used to determine how much the power
delivered to the 2-port could be increased if a lossless matching network is added between
the source and the 2-port.
The transducer gain will always be less than or equal to the available gain since
G
T
G
A
=
Pout
Pavs
Pavo
Pavs
=
P
out
P
avo
1 (7.39)
So G
T
/G
A
is equal to the impedance mismatch factor at the output port, which is equal
to one when the load is conjugately matched to the output of the 2-port. The ratio given
in Equation 7.39 can be used to determine how much the power delivered to the load could
be increased if a lossless matching network was added between the 2-port and the load.
The operating gain can be greater than or less than the available gain since
G
A
G
=
Pavo
Pavs
Pout
Pin
=
Pin
Pavs
Pout
Pavo
(7.40)
Depending on whether the input mismatch factor (numerator) or the output mismatch
factor (denominator) is the smaller, the ratio G
A
/G can be either > 1 or < 1. If the input
is match is better than the output match, then the ratio in Equation 7.40 will be > 1. On
the other hand, if the output match is better than the input match, then the ratio will be
< 1.
Finally, it should be noted that only the operating gain and the available gains can be
cascaded. That is, when several ampliers are connected in cascade, the overall operating
or available gain of the cascade is simply the product of the operating or available gains of
the individual ampliers. This is not true for the transducer gain.
When cascading gains, it must be kept in mind that the source impedance that terminates
the input of the second amplier is the output impedance of the rst amplier, etc. Consider
the problem of cascading a number of identical ampliers. Suppose the overall available
gain is the quantity of interest. The available gain of the rst amplier depends on the
source termination, Z
S
. The available gain of the second amplier depends on the output
impedance of the rst amplier which, in turn, depends on the source termination. The
available gain of the third amplier depends on the output impedance of the second amplier,
etc. So the available gain of a cascade of ampliers depends on the 2-port network parameters
of the individual ampliers as well as the source termination of the rst amplier. Similarly,
if the operating power gain of the cascade is desired, it will be found to depend only on
the load termination of the last amplier and the 2-port parameters of all of the individual
ampliers.
232 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
7.5 Calculation of Impedance and Gain using the Impedance
Matrix
The concepts dened in the previous section will be applied in this section where the input
and output impedances and transducer gain are derived for a 2-port represented by its
Z-parameters.
Consider a system consisting of a 2-port embedded between a source with impedance Z
S
and peak open-circuit voltage V
S
and a load with impedance Z
L
(Figure 7.9). The power
available from the source P
avs
is dened to be the power that the source would deliver to a
conjugately matched load and is
P
avs
=
|V
S
|
2
8R
S
(7.41)
The voltage and current at input and output ports are denoted by (V
1
, I
1
) and (V
2
, I
2
),
respectively, with current chosen to ow into the two port as shown in Figure 7.9.
[Z]
V
S
Z
S
Z
L V
1
V
2
I
1
I
2
+
-
+
-
Figure 7.9: A system consisting of source, 2-port, and a load.
The voltage and currents at the input and output of the 2-port are related by impedance
matrix elements, Z
ij
, i.e.
V
1
= Z
11
I
1
+Z
12
I
2
(7.42)
V
2
= Z
21
I
1
+Z
22
I
2
(7.43)
The power delivered to the load impedance is related to P
avs
by the transducer power gain
G
T
G
T
P
out
P
avs
=
1
2
Re[V
2
I
2
]
|V
S
|
2
8R
S
(7.44)
The load constraint I
2
= V
2
/Z
L
can be employed in equation 7.44 to write
G
T
= 4|
V
2
V
S
|
2
R
S
R
L
|Z
L
|
2
(7.45)
where R
S
= Re[Z
S
] and R
L
= Re[Z
L
]. Equation 7.45 can be written in terms of the
magnitude of the voltage gain, |V
2
/V
1
|, and the voltage division that occurs at the input of
the 2-port, |V
1
/V
S
|, i.e.:
G
T
= 4|
V
2
V
1
|
2
|
V
1
V
S
|
2
R
S
R
L
|Z
L
|
2
(7.46)
The voltage division can be written in terms of the input impedance of the 2-port and the
source impedance
V
1
V
S
=
Z
IN
Z
IN
+Z
S
(7.47)
7.6. APPLICATIONS OF 2-PORT ANALYSIS 233
and the input and output impedances of the 2-port can be obtained from the load constraint
and equations 7.42 and 7.43. The input impedance is:
Z
IN
=
V
1
I
1
= Z
11
Z
12
Z
21
Z
L
+Z
22
(7.48)
and the output impedance is:
Z
OUT
=
V
2
I
2
|
V
S
=0
= Z
22
Z
12
Z
21
Z
S
+Z
11
(7.49)
The voltage gain is obtained from 7.42 and 7.43 using the load constraint I
2
= V
2
/Z
L
:
A
V
=
V
2
V
1
=
Z
21
Z
L
Z
11
Z
L
+Z
11
Z
22
Z
12
Z
21
(7.50)
Now, equation 7.46 can be rewritten using 7.50 and 7.48 as:
G
T
= 4|
Z
21
Z
L
Z
11
Z
L
+Z
11
Z
22
Z
12
Z
21
|
2
|
Z
11
Z12Z21
Z
L
+Z22
Z
11
Z12Z21
Z
L
+Z22
+Z
S
|
2
R
S
R
L
|Z
L
|
2
which simplies to the nal result:
G
T
= 4
|Z
21
|
2
R
L
R
S
|(Z
11
+Z
S
)(Z
22
+Z
L
) Z
12
Z
21
|
2
(7.51)
Since the available source power P
avs
is a constant determined by the capabilities of the
source, the output power will be proportional to G
T
. The largest possible output power
with passive source and load terminations results when Z
S
and Z
L
are chosen to maximize
G
T
subject to Re[Z
S
]>0 and Re[Z
L
]>0. These optimum terminations are denoted by Z
MS
and Z
ML
and it can be shown that these terminations result in a simultaneous conjugate
match at the input and output ports, i.e. Z
MS
and Z
ML
satisfy the following equations:
Z
IN
= Z
MS
Z
OUT
= Z
ML
(7.52)
7.6 Applications of 2-port analysis
7.6.1 Losses in L-networks for impedance matching
Consider an L-network designed to match two resistive terminations, R
S
and R
L
, with
R
S
< R
L
. The design of such networks has been considered previously. Suppose that such
a network has been designed and that the series and shunt elements available for use in the
network are lossy i.e., suppose that the series and shunt elements have component Qs at
the design frequency, !
o
, denoted by Q
s
and Q
p
, respectively. The loss in the series element
can be represented by a series resistance, r
s
, and the loss in the parallel element can be
represented by a shunt resistance, r
p
. The circuit model is shown below in Figure 7.10. The
series and shunt reactances are assumed to be those necessary to match the source to the
load in the absence of any losses in the network. Thus, the load impedance seen by the
resistive 2-port will be equal to Z
L
= R
S
jX
S
.
234 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
r
p
Z = R
S
jX
S
R
S
Q
s
=
|XS|
rs
jX
S
jX
P
Z = R
S
+jX
S
r
s
Q
p
=
rp
|XP |
R
L
Figure 7.10: Circuit model for an L-network with lossy components. The loss resistances
can be lumped into a 2-port which is contained within the dashed box.
As shown in the Figure, the loss resistances can be lumped into a 2-port which is con-
tained within the dashed box. The Z-parameter matrix for the 2-port is easily shown to
be:
[Z] =
Z
11
Z
12
Z
21
Z
22
r
s
+r
p
r
p
r
p
r
p
(7.53)
Notice that the 2-port dened here is reciprocal and hence Z
12
= Z
21
. The loss caused
by the presence of the resistances r
s
and r
p
can be determined by calculating the transducer
gain of the 2-port in the system shown in Figure 7.10. The source impedance seen by the
resistive 2-port is Z
S
= R
S
+ jX
S
and the load impedance is Z
L
= R
S
jX
S
. If the
Z-parameters and the terminating impedances are inserted into equation 7.51, the result is:
G
T
=
4r
2
p
R
2
S
|(r
s
+r
p
+R
S
+jX
S
)(r
p
+R
S
jX
S
) r
2
p
|
2
(7.54)
After some algebraic manipulation, the transducer gain can be rewritten as:
G
T
=
1
|1 +
R
S
2rp
(1 +Q
2
) +
rs
2R
S
+
rs
2rp
(1 jQ)|
2
(7.55)
where the fact that |X
S
| = R
S
Q has been used to eliminate X
S
from the expression. Next,
one can make use of the following relationships to write G
T
in terms of the L-network Q
(Q) and the component Qs of the lossy series and shunt elements:
R
S
2rp
=
Q
2Qp(1+Q
2
)
rs
2R
S
=
Q
2Qs
rs
2rp
=
Q
2
2QsQp(1+Q
2
)
(7.56)
After using 7.56 in 7.55, G
T
can be written as:
G
T
=
1
|1 +
Q
2Qp
+
Q
2Qs
+
(1jQ)Q
2
2QsQp(1+Q
2
)
|
2
(7.57)
7.6. APPLICATIONS OF 2-PORT ANALYSIS 235
This result can be checked by allowing the losses to approach zero (Q
s
! 1, Q
p
! 1) in
which case G
T
!1, as expected.
In practical applications, the component Qs (Q
s
and Q
p
) will be larger than the L-
network Q (Q
s
> Q and Q
p
> Q) in which case the last term in the denominator can be
neglected compared to the second and third terms. Thus, a good approximation for the
transducer gain is:
G
T
'
1
(1 +
Q
2Qp
+
Q
2Qs
)
2
(7.58)
This result illustrates the fact that component Qs must be much larger than the L-network
Q if component losses are to be reasonably small. For example, suppose that it is necessary
to keep the matching network loss below 1 dB this means that G
T
10
(1/10)
=
0.794. Assuming that the component Qs are equal (Q
s
= Q
p
), then the constraint G
T
0.794 leads to Q
s,p
8.2Q, which means that the L-network must be implemented with
components having component Qs at least 8.2 times as large as the L-network Q. Obviously,
this will become impractical when the L-network Q is too large. Large L-network Qs are
associated with large resistance transformation ratios
1
so that it becomes more dicult to
implement an L-network with low losses as the resistance transformation ratio increases.
7.6.2 Two-winding Transformers
7.6.2.1 Equivalent Circuit Model for Two-winding Transformers
Transformers are often utilized in both narrow-band and wide-band RF applications. They
can be employed for impedance transformation, phase inversion, and dc isolation. Trans-
formers are also utilized as resonant circuit elements.
Consider the two-winding transformer shown in Figure 7.11.
I
1
I
2
V
2
V
1
M
L
2
L
1
+
-
+
-
Figure 7.11: 2-winding transformer.
Ignoring losses, the equations that describe this device are
V
1
= j!L
1
I
1
+j!MI
2
(7.59)
V
2
= j!MI
1
+j!L
2
I
2
(7.60)
where L
1
and L
2
are the self inductances of the transformer windings and M is the mutual
inductance. The dot convention is such that if current ows into the dotted terminals, the
magnetic uxes linking the two coils will reinforce each other. With this convention, M will
be a positive number. The 2-winding transformer is completely described by its open-circuit
impedance matrix,
[Z] =
j!L
1
j!M
j!M j!L
2
.
1
Recall that the Q of an L-network is Q =
q
R
big
R
small
1.
236 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
Since the circuit operation of the transformer is completely described by this impedance
matrix, any network having the same dening equations can be substituted for the trans-
former. One useful equivalent circuit is shown in Figure 7.12.
n : 1
k
2
L
1
(1 k
2
)L
1
Ideal Transformer
Figure 7.12: Equivalent circuit for a two-winding transformer
In this model the real transformer has been replaced with an ideal transformer and
equivalent inductances. The equivalent circuit parameters are k (coupling coecient), n
(turns ratio), and L
1
(self inductance of winding 1. This equivalent circuit is not unique.
For example, it is possible to derive an equivalent circuit that has the self inductance of
winding 2 as a parameter. The terminal relations for the ideal transformer are summarized
by Figure 7.13 and equations 7.61 and 7.62.
+
-
+
-
I
A
I
B
V
A
V
B
n : 1
Figure 7.13: Terminal relations for the ideal transformer
V
B
=
V
A
n
(7.61)
I
B
= nI
A
(7.62)
It is left as an exercise to demonstrate that the equivalent circuit shown in Figure 7.12 has the
same impedance matrix as the actual transformer, provided that the following relationships
hold:
k =
M
p
L
1
L
2
(7.63)
n = k
L
1
L
2
(7.64)
The k parameter is called the coecient of coupling and n is the eective turns ratio for
the transformer.
A useful approximation results when the coecient of coupling approaches 1. This will
be the situation when both windings are wound on a high permeability magnetic core. Here
the equivalent circuit reduces to the circuit shown in Figure 7.14.
7.6. APPLICATIONS OF 2-PORT ANALYSIS 237
n : 1
L
1
Ideal Transformer
Figure 7.14: Equivalent circuit when k approaches 1. In this case, the eective turns ratio
is, approximately, n =
L
1
/L
2
.
7.6.2.2 Impedance Transformation with the Two-winding Transformer
Consider the situation where a tightly-coupled transformer is to be used as an impedance
transformer as in Figure 7.15.
n : 1
L
1
Ideal Transformer
Z
L Z
in
Figure 7.15: Tightly-coupled transformer used as impedance transformer
The Z-parameter matrix for a two-winding transformer is
[Z] =
j!L
1
j!M
j!M j!L
2
j!L
1
j!L
1
k
2
n
j!L
1
k
2
n
j!L
1
k
2
n
2
.
This is the exact impedance matrix. When the transformer is tightly coupled, k ' 1, and
the impedance matrix can be approximated by
[Z] '
j!L
1
j!L
1
/n
j!L
1
/n j!L
1
/n
2
.
The impedance seen looking in to a tightly coupled transformer that is terminated in load
impedance Z
L
is now easily computed:
Z
IN
= Z
11
Z
12
Z
21
Z
22
+Z
L
= j!L
1
(j!L
1
/n)
2
j!L
1
/n
2
+Z
L
=
j!L
1
Z
L
n
2
j!L
1
+Z
L
n
2
(7.65)
The tightly-coupled transformer will behave essentially like an ideal transformer if !L
1
|Z
L
n
2
| or, since n
2
= L
1
/L
2
, if !L
2
|Z
L
|. In this case
Z
IN
n
2
Z
L
. (7.66)
238 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
1 : n
L
1
Ideal Transformer
Z
L Z
in
Figure 7.16: Transformer reversed
Similarly, if the transformer is reversed, as in Figure 7.16 then
Z
IN
=
1
n
2
s L
1
Z
L
s L
1
+ Z
L
(7.67)
And if
!L
1
Z
L
(7.68)
then
Z
IN
'
1
n
2
Z
L
(7.69)
These examples lead to a rule that is applicable when designing tightly-coupled impedance
transformers: The inductive reactance of the impedance transformer winding
should be signicantly larger than the impedance to which it is connected. Typi-
cally, the inductive reactance is chosen to be at least four times larger than the impedance at
the lowest frequency of intended operation. When this constraint is satised, the tightly cou-
pled transformer can provide an impedance transformation that is essentially independent
of frequency over a wide bandwidth.
7.6.2.3 Single-tuned Transformer
In Section 7.6.2.2 the impedance transforming properties of the tightly-coupled transformer
were examined, and a working rule was given for designing a transformer which will be-
have essentially like an ideal transformer. This is most useful when designing transformers
that must operate over a relatively wide frequency range. If narrow-band operation is de-
sired, then the inductive reactance of the transformer windings can be resonated, and ideal
transformer operation will result at the resonant frequency. For example, suppose that a
capacitance is used to resonate the primary inductance of the transformer as in Figure 7.17.
C Z
in
R
L
Figure 7.17: Actual circuit
If the capacitor resonates with L
1
at the frequency of interest, then at that frequency, the
circuit will behave like an ideal transformer, and the input impedance will be Z
in
= n
2
R
L
.
7.6. APPLICATIONS OF 2-PORT ANALYSIS 239
n : 1
L
1
Ideal Transformer
Z
in
C R
L
Figure 7.18: Single-tuned transformer using a tightly coupled transformer and a resonating
capacitance.
More generally, the primary circuit can be modeled by reecting the load resistance through
the transformer as shown in Figure7.19. The result is a parallel resonant circuit with
L
1 Z
in
C n
2
R
L
Figure 7.19: Load resistance reected through transformer
Q =
n
2
R
L
!
o
L
1
=
R
L
!
o
L
2
and
!
o
=
1
p
L
1
C
.
The input impedance at the resonant frequency is
Z
IN
= n
2
R
L
. (7.70)
The bandwidth of the single-tuned transformer circuit is the same as the bandwidth of
the equivalent parallel RLC circuit shown in Figure 7.19, i.e.,
BW =
!
o
Q
=
!
2
o
L
1
n
2
R
L
=
1
n
2
CR
L
.
The bandwidth can be controlled by adjusting the self inductance of the primary winding,
L
1
, and the turns ratio, n. For a given L
1
, the resonating capacitance, C, must be chosen to
resonate with L
1
at !
o
. This provides a useful method for designing impedance matching
networks with specied bandwidth.
7.6.3 Two Magnetically-Coupled Resonators (Doubly-Tuned Trans-
former)
The Z matrix for two coupled coils can be written in terms of the self-inductances, L
1
and
L
2
and mutual-inductance M:
[Z] =
j!L
1
j!M
j!M j!L
2
(7.71)
240 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
In this section, well employ the Z matrix to investigate the properties of a lter consisting
of two magnetically coupled resonators. The system is shown in Figure 7.20.
R r C
L
M
r
L
C
R
k=M/L
Figure 7.20: Two magnetically-coupled series resonant circuits in a system with source and
load impedances equal to R.
We assume that the two coils are identical (L
1
= L
2
= L), and that resistive losses can
be modeled by a resistance, r, in series with each of the coils. In addition, a capacitance,
C, can be added in series with each coil to form two magnetically-coupled series resonant
circuits. It is easy to verify that adding the elements r and C in series with each coil changes
the diagonal elements of the Z matrix by adding the impedances of these elements to the
self-impedance of the coils and does not aect the o-diagonal (coupling) elements of the Z
matrix. Thus, the impedance matrix of the coupled resonator system is given by:
[Z] =
r +j!L +
1
j!C
j!M
j!M r +j!L +
1
j!C
(7.72)
In a system with source and load impedance denoted by R, the transducer gain of the
coupled-resonator lter is obtained by using 7.72 and Z
S
= R, Z
L
= R in Equation 7.51:
G
T
=
4!
2
M
2
R
2
|(R +r +j!L +
1
j!C
)
2
+ !
2
M
2
|
2
(7.73)
Dene the coupling factor :
k =
M
L
, (7.74)
the resonant frequency of either resonator in isolation:
!
o
=
1
p
LC
, (7.75)
and the loaded Q of the resonator (accounting for the loss resistance and termination resis-
tance):
Q
0
=
!
o
L
R +r
. (7.76)
Equations 7.74 - 7.76 can be used in Equation 7.73 to write the transducer gain as:
G
T
=
R
2
(R +r)
2
4k
2
Q
02 !
2
!
2
o
|(1 +jQ
0
(
!
!o
!o
!
))
2
+k
2
Q
02
!
2
!
2
o
|
2
(7.77)
7.6. APPLICATIONS OF 2-PORT ANALYSIS 241
This is a complicated-looking result, so to simplify interpretation it is useful to examine the
transducer gain at the resonant frequency !
o
:
G
T
|
!=!o
=
R
2
(R +r)
2
4k
2
Q
02
(1 +k
2
Q
02
)
2
(7.78)
In the absence of resonator losses (r ! 0) the rst term is equal to one. The second term
will be equal to one also if
k =
1
Q
0
(7.79)
Thus (in the absence of resonator losses), if the coupling factor and/or loaded Q are adjusted
so that equation 7.79 is satised, we expect that the coupled resonator lter will transmit
signals unattenuated at the resonant frequency, !
o
. In this case, the resonators are said to
be critically coupled.
Figure 7.21 shows the transducer gain for a system employing resonators with loaded
Q equal to 5. Curves are plotted for 5 dierent coupling factors, two of which are smaller
than the critical coupling value of 1/5=0.2 and two of which are larger than the critical
value. Coupled resonators with k smaller than the critical coupling value are said to be
undercoupled, whereas resonators with k larger than the critical value are overcoupled.
A characteristic of the overcoupled response is a double-peaked response with separation
between the peaks that depends on the degree of overcoupling. Undercoupling, on the other
hand, yields a single-peaked response with progressively larger attenuation as the coupling
factor is decreased below the critical value. Notice that the critically coupled response
yields a bandpass lter with relatively at passband response. An obvious disadvantage of
this type of lter is the fact that the optimum coupling factor depends on the source and
load impedances. Thus, a lter that is operated with dierent terminations than it was
designed for may end up either under- or over-coupled and the shape of the lter response
may degrade signicantly under such conditions.
242 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
0
0.2
0.4
0.6
0.8
1
0.6 0.8 1 1.2 1.4 1.6 1.8 2
G
T
!/!
o
Coupled lossless resonators, loaded Q=5
k = 0.10 undercoupled
k = 0.15 undercoupled
k = 0.20 critical coupling
k = 0.30 overcoupled
k = 0.40 overcoupled
Figure 7.21: Transducer power gain for a coupled resonator lter using resonators with
loaded Q equal to 5. Critical coupling will occur when k = 1/Q
0
= 1/5 = 0.2.
7.6.4 Analysis of a Small-signal Series/Shunt Feedback Amplier
In this section we will calculate the input and output impedance and voltage gain of the
amplier shown in Figure 7.22. The amplier is redrawn in Figure 7.23 to show how the
circuit can be described as a combination of 2-ports connected in series, in cascade, and in
parallel. The hybrid- 2-port is in series with the 2-port containing the single shunt resistor
R
e
. The output of the resulting 2-port is cascaded with the input of a 2-port consisting
of an ideal transformer with 1:1 turns ratio. The ideal transformer is congured so that it
produces a 4:1 impedance transformation. The cascade is in parallel with a 2-port consisting
of the single series resistor R
f
.
Analysis proceeds as follows:
calculate the Z parameters of the hybrid- 2-port and the 2-port containing the shunt
resistor, R
e
. Add these two Z matrices to obtain the Z matrix of the series combina-
tion.
Convert the Z matrix of the series combination to an ABCD matrix. Denote this ma-
trix by [ABCD]
1
. Determine the ABCD matrix of the 4 : 1 transformer and denote
this matrix by [ABCD]
2
. The ABCD matrix of the cascade is then [ABCD]
1
[ABCD]
2
.
Note that the order of the terms in this matrix product is important and must reect
the order in which the 2-ports are cascaded.
Convert ABCD of the cascade to a Y matrix and sum this matrix with the Y matrix
for the 2-port consisting of the single series resistor, R
f
.. The result is the Y matrix
of the overall circuit.
7.6. APPLICATIONS OF 2-PORT ANALYSIS 243
Calculate Y
in
, Y
out
, and A
v
in terms of the Y parameters of the overall circuit.
+
-
+
-
+
-
+
-
Figure 7.22: Small signal model of amplier with series and shunt feedback and 4:1 output
transformer.
+
-
V
be
g
m
V
be
r
r
o
R
e
R
f
R
S
R
L
Figure 7.23: Feedback amplier drawn as a set of interconnected 2-ports.
The Z matrix for the hybrid- 2-port is:
Z
11
Z
12
Z
21
Z
22
0
g
m
r
r
o
r
o
Z
11
Z
12
Z
21
Z
22
R
e
R
e
R
e
R
e
Z
11
Z
12
Z
21
Z
22
+R
e
R
e
g
m
r
r
o
+R
e
r
o
+R
e
A B
C D
=
1
g
m
r
o
r
+R
e
(R
e
+r
) (R
e
+r
)(R
e
+r
o
) R
e
(R
e
g
m
r
o
r
)
1 r
o
+R
e
A B
C D
=
1
g
m
r
0 R
e
+r
+g
m
R
e
r
0 1
The ABCD matrix of the 4:1 transformer implemented using the ideal transformer with
1:1 turns ratio is:
A B
C D
2 0
0
1
2
A B
C D
=
1
2g
m
r
0 R
e
+r
+g
m
R
e
r
0 1
Y
11
Y
12
Y
21
Y
22
=
2g
m
r
R
e
+r
+g
m
R
e
r
1
2gmr
0
1 0
Y
11
Y
12
Y
21
Y
22
=
1
R
f
1 1
1 1
The sum of the 2 Y matrices is the Y matrix for the overall circuit:
Y
11
Y
12
Y
21
Y
22
1
Re+r+gmRer
+
1
R
f
1
R
f
2gmr
Re+r+gmRer
1
R
f
1
R
f
.
The input impedance of a 2-port can be written in terms of the Y parameters of the
2-port and the load admittance, Y
L
, as:
Y
IN
= Y
11
Y
12
Y
21
Y
22
+Y
L
.
7.7. Y, Z, H, ABCD RELATIONSHIPS 245
Insert the Y parameters of the overall 2-port and use Y
L
=
1
R
L
:
Y
IN
=
1
R
e
+r
+g
m
R
e
r
+
1
R
f
1
R
f
(
2gmr
Re+r+gmRer
1
R
f
)
1
R
f
+
1
R
L
.
The input impedance is then found from Z
IN
= Y
1
IN
. After some simplication, the input
impedance can be written as:
Z
IN
=
(R
L
+R
f
)(R
e
+r
+g
m
R
e
r
)
R
e
+r
+R
L
+R
f
+g
m
r
(R
e
+ 2R
L
)
.
Similarly, the output admittance depends on the Y parameters and the source admittance,
Y
S
:
Y
OUT
= Y
22
Y
12
Y
21
Y
11
+Y
S
.
Inserting the Y parameters of the overall 2-port, use Y
S
=
1
R
S
, and Z
OUT
= Y
1
OUT
to show:
Z
OUT
=
(R
S
+R
f
)(R
e
+r
+g
m
R
e
r
) +R
S
R
f
R
S
+R
e
+r
+g
m
r
(R
e
+ 2R
S
)
.
The voltage gain of a 2-port depends on the Y parameters and the load admittance Y
L
=
1
R
L
:
A
v
=
Y
21
Y
22
+Y
L
.
Insert the Y parameters and simplify to show:
A
v
= 2g
m
R
L
R
f
R
L
+R
f
1
1 +R
e
(g
m
+
1
r
)
+
R
L
R
L
+R
f
.
7.7 Y, Z, h, ABCD relationships
Relationships between the parameter sets described in this Chapter are given in the following
sections. The determinants of the parameter matrices are dened as follows:
D
Y
= Y
11
Y
22
Y
21
Y
12
(7.80)
D
Z
= Z
11
Z
22
Z
21
Z
12
(7.81)
D
h
= h
11
h
22
h
21
h
12
(7.82)
D
ABCD
= AD BC (7.83)
7.7.1 Converting to Y-parameters
Y
11
Y
12
Y
21
Y
22
Z
22
D
Z
Z
12
D
Z
Z
21
D
Z
Z
11
D
Z
1
h
11
h
12
h
11
h
21
h
11
D
h
h
11
D
B
D
ABCD
B
1
B
A
B
Z
11
Z
12
Z
21
Z
22
Y
22
D
Y
Y
12
D
Y
Y
21
D
Y
Y
11
D
Y
D
h
h
22
h
12
h
22
h
21
h
22
1
h
22
A
C
D
ABCD
C
1
C
D
C
h
11
h
12
h
21
h
22
D
Z
Z
22
Z
12
Z
22
Z
21
Z
22
1
Z
22
1
Y
11
Y
12
Y
11
Y
21
Y
11
D
Y
Y
11
B
D
D
ABCD
D
1
D
C
D
A B
C D
Z
11
Z
21
D
Z
Z
21
1
Z
21
Z
22
Z
21
Y
22
Y
21
1
Y
21
D
Y
Y
21
Y
11
Y
21
D
h
h
21
h
11
h
21
h
22
h
21
1
h
21
7.8 Summary
7.8.1 Z parameters
Z
IN
= Z
11
Z
12
Z
21
Z
L
+Z
22
(7.84)
Z
OUT
= Z
22
Z
12
Z
21
Z
S
+Z
11
(7.85)
A
V
=
Z
21
Z
L
Z
11
Z
L
+Z
11
Z
22
Z
12
Z
21
(7.86)
A
I
=
Z
21
Z
22
+Z
L
(7.87)
7.8.2 Y parameters
Y
IN
= Y
11
Y
12
Y
21
Y
L
+Y
22
(7.88)
Y
OUT
= Y
22
Y
12
Y
21
Y
S
+Y
11
(7.89)
A
V
=
Y
21
Y
22
+Y
L
(7.90)
A
I
=
Y
21
Y
L
Y
11
Y
L
+Y
11
Y
22
Y
12
Y
21
(7.91)
7.9. REFERENCES 247
7.9 References
1. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw-Hill Book Com-
pany, 1943.
2. Krauss, Herbert L., Charles W. Bostian, and Frederick H. Raab, Solid State Radio
Engineering, John Wiley and Sons, 1980.
3. Ludwig, Reinhold, and Pavel Bretchko, RF Circuit Design - Theory and Applications,
Prentice-Hall, Inc., New Jersey, 2000.
4. Balabanian, Norman and Theodore A. Bickart, Electrical Network Theory, John Wiley
& Sons, 1969.
248 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
7.10 Homework Problems
1. Suppose the Y-parameters of a 2-port are known.
(a) Derive an expression for the input admittance of the 2-port when it is terminated
with load admittance Y
L
.
(b) Derive an expression for the voltage gain A
v
= V
2
/V
1
. Your result will depend
on some of the Y-parameters and Y
L
.
(c) Using your results for parts 1a and 1b nd an expression for the operating power
gain of the 2-port, G. The operating power gain is dened as
G =
P
out
P
in
(7.92)
where P
out
is the power delivered to the load and P
in
is the power delivered to
the 2-port. Your result should be in terms of the Y-parameters of the 2-port and
Y
L
.
2. Suppose the ABCD-parameters of a 2-port are known.
(a) Derive an expression for the input impedance of the 2-port when it is terminated
with load impedance Z
L
.
(b) Derive an expression for the output impedance of the 2-port when it is terminated
with source impedance Z
S
.
(c) Derive an expression for the voltage gain A
v
= V
2
/V
1
. Your result will depend
on some of the ABCD-parameters and Z
L
.
(d) Find an expression for the transducer power gain, G
T
, of the 2-port when it is
driven with a source having impedance Z
S
and terminated with load impedance
Z
L
. Your result will depend on the ABCD-parameters as well as Z
S
and Z
L
.
Simplify the expression as much as possible.
3. Find the Z parameters for the 2-port in Figure 7.24.
Figure 7.24: 2-port.
4. Find the Z-parameters for the T-network in Figure 7.25. Port 1 is on the left.
5. Find the ABCD parameters for the 2-port shown in Figure 7.24.
6. Consider the unilateral hybrid-pi model shown in Figure 7.26. Find an expression for
the available power gain of this 2-port. Your result should be expressed only in terms
of Z
1
, Z
2
, g
m
and the impedance of the source Z
s
(not shown). Start from the basic
denition of available power gain.
7.10. HOMEWORK PROBLEMS 249
Z
1
Z
2
Z
3
Figure 7.25: T-network.
Z
2
Z
1 V
g
m
V
+
-
Figure 7.26: Unilateral hybrid-pi model.
7. The small signal equivalent circuit model for an amplier is shown in Figure 7.27.
The amplier is designed to operate at a center frequency of 30 MHz. Find the 3 dB
+
-
V
i
14.1pF
2H
R
L
= 3k
g
m
V
i
28.2pF 1k
1H
Figure 7.27: Small signal equivalent circuit model for an amplier.
bandwidth of the operating power gain. Express your result in MHz. Start from
the basic denition of operating power gain.
8. A lossy network has been designed to provide 10 dB of attenuation (i.e. operating
power gain G = 10 dB) when the network is terminated with a 50 load. It is
also known that the input impedance of the network is 300 when the output of the
network is terminated with a 50 load. Find the transducer gain for this 2 port in a
system with Z
S
= Z
L
= 50 .
9. The 2-port shown in Figure 7.28 is characterized by its Z-parameters, where Z
11
=
Z
12
= Z
21
= Z
22
= R. Find Z
in
. Express your result in terms of R only. Hint: The
input impedance of a 2-port terminated with load impedance Z
L
is Z
in
= Z
11
Z12Z21
Z22+Z
L
.
10. This exercise will provide some practice in building up a more complex 2-port by
combining simple 2-ports in series and in parallel. It is also intended to provide
some insight into why we utilize the broadband transformer in the amplier that is
constructed in laboratory 3. Consider the small-signal equivalent circuit shown in the
250 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
Z
11
Z
12
Z
21
Z
22
R
R
Z
in
Figure 7.28: Arbitrary 2-port in series with a 2-port consisting of a shunt resistor (a series-
feedback conguration).
Figure. The resistances R
S
and R
L
are the source and load resistances, respectively,
and are external to the amplier. This is a simple transistor amplier with series (R
e
)
and shunt (R
f
) feedback. It is similar to the amplier that you will design and build
in Lab 3, however it does not include the 1:1 transformer between the load and the
transistors collector.
+
-
+
-
(a) Find the Y-parameter matrix for the feedback amplier. Do not include R
S
and
R
L
in the amplier - they are the external source and load terminations.
(b) Find expressions for the input impedance (Z
in
), output impedance (Z
out
), and
voltage gain (A
v
) of the amplier when terminated with R
S
and R
L
.
(c) Now, follow the analysis given in section 7.6.4 to determine expressions for the
feedback resistance R
f
, and power gain G under the assumption that the param-
eters are adjusted to provide a simultaneous (approximate) impedance match at
the input and output ports when both ports are terminated with resistance R.
Make a table similar to Table 3.1 in the Lab notes. Table 3.1 lists, for each value
of R
e
, the power gain and the value of R
f
required to produce the impedance
match assuming that R
S
= R
L
= R = 50 , I
CQ
= 10 mA, and = 100. Table
3.1 was produced using equations 3.12 and 3.14 in the Lab notes. Compare the
results for the amplier analyzed here and the one analyzed in the Lab notes.
What advantage (if any) does the addition of the transformer provide?
7.10. HOMEWORK PROBLEMS 251
11. A 2-port is empirically found to have the following properties. (i) With the output port
unterminated the input impedance Z
IN
= 100 and the voltage gain A
V
=
V2
V1
= 0.5.
(ii) With the output port terminated in a short circuit, the input impedance Z
IN
=
75 and the current gain A
I
=
I2
I1
= 0.5. Find Z
IN
and A
V
when the output port
is terminated with Z
L
= 50 .
12. A resistor is added in parallel with the input of an existing 2-port, as shown in the Fig-
ure. Determine the overall ABCD matrix of the system. Your result will be expressed
A B
C D
R
in terms of the ABCD parameters of the original 2-port (denoted by A, B, C, D) and
the resistance R.
13. An impedance inverter has the property that the input impedance of the network
is equal to the inverse of the load impedance, i.e. Z
IN
= Z
1
L
. Determine what
properties an ABCD matrix must have to produce the impedance inverter function.
14. Consider the T-network shown in Figure 7.29. The ABCD parameters of the T-
Z
1
Z
2
Z
3
Figure 7.29: T-network to be used as an impedance inverter.
network are:
A B
C D
=
1
Z
3
Z
1
+Z
3
Z
1
Z
2
+Z
2
Z
3
+Z
1
Z
3
1 Z
2
+Z
3
Z
o
(8.5)
a
2
+ b
2
= V
2
/
Z
o
(8.6)
a
1
b
1
=
Z
o
I
1
(8.7)
a
2
b
2
=
Z
o
I
2
(8.8)
The independent variables are a
1
and a
2
and the dependent variables are b
1
and b
2
. Re-
member, the reference impedance for a parameter set is that impedance which will force
one of the independent variables to zero when it is used to terminate the 2-port. The choice
of a nite reference impedance, Z
o
, forces us to choose the independent variables to be
proportional to V
i
+ Z
o
I
i
, as dened in Equations 8.1 and 8.2, so that they will be forced
to zero when the 2-port is terminated in the reference impedance, Z
o
.
8.2. INTERPRETATION OF S-PARAMETERS 255
The variables dened in Equations 8.1-8.4 have been normalized with respect to the
square root of the reference impedance, Z
o
. With this normalization (and assuming that
the voltages and currents are rms values) the time-averaged power delivered to a given port
(denoted by i) can be written as P = Re[V
i
I
i
] = |a
i
|
2
|b
i
|
2
. Thus, the variables have units
of [Watts]
1/2
.
Given the denitions in Equations 8.1-8.8, the scattering parameters are dened as fol-
lows:
b
1
= S
11
a
1
+ S
12
a
2
(8.9)
b
2
= S
21
a
1
+ S
22
a
2
(8.10)
Thus
S
11
=
b
1
a
1
|
a2=0
(8.11)
S
12
=
b
1
a
2
|
a1=0
(8.12)
S
21
=
b
2
a
1
|
a2=0
(8.13)
S
22
=
b
2
a
2
|
a1=0
(8.14)
8.2 Interpretation of S-parameters
Consider S
11
:
S
11
=
b
1
a
1
|
a2=0
(8.15)
Setting a
2
= 0 is equivalent to setting
V
2
= Z
o
I
2
V
2
I
2
= Z
o
(8.16)
That is, to set a
2
= 0, we terminate port 2 with the reference impedance, Z
o
. Thus
S
11
=
b
1
a
1
|
a2=0
=
V
1
Z
o
I
1
V
1
+Z
o
I
1
|
output port terminated in Zo
(8.17)
=
V
1
/I
1
Z
o
V
1
/I
1
+Z
o
|
output port terminated in Zo
or,
S
11
=
Z
in
Z
o
Z
in
+Z
o
|
output port terminated in Zo
(8.18)
where Z
in
is the input impedance of the 2-port when the output is terminated in Z
o
.
256 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
Recall from transmission line theory that S
11
has the same form as the reection coe-
cient, , that relates the incident and reected voltage waves on a transmission line
=
V
V
+
=
reected voltage wave
incident voltage wave
(8.19)
where the total voltage on the line, V, is given by
V = V
+ V
+
(8.20)
Comparison of Equations 8.19 and 8.20 with
S
11
=
b
1
a
1
(8.21)
and
V
1
= (a
1
+ b
1
)
Z
o
(8.22)
leads to the following correspondence between the variables a
1
and b
1
and the voltage waves
on a transmission line:
a
i
=
1
p
Z
o
(Voltage wave incident on port i) (8.23)
b
i
=
1
p
Z
o
(Voltage wave reected from port i) (8.24)
This is an interesting observation, but where is the transmission line? Although there
has been no mention of transmission lines in the system so far, it is helpful to imagine
that the source and load are connected to the 2-port through sections of transmission line
with characteristic impedance Z
o
. Well assume that the sections of transmission line have
innitesimal length, so that they do not aect the electrical characteristics of the system,
as in Figure 8.2.
2-port
V
1
V
2
I
2
I
1
I
1
+
-
I
2
Z
S
Z
L
V
s
a
1
b
1
a
2
b
2
+
-
Z
o
Z
o
Figure 8.2: Source and load connected to 2-port through transmission line with impedance
Z
o
Employing this conceptual model, the a and b variables may be interpreted as repre-
senting the (normalized) voltage waves that would exist on the sections of transmission line.
When working with S parameters, the model can be used to visualize the a and b variables
as incident and reected (normalized) voltage waves, although in practice the terminations
may be connected directly to the 2-port without any intervening transmission lines. (Actu-
ally, in practice you will nd that it is virtually impossible to get signals into and out of a
8.2. INTERPRETATION OF S-PARAMETERS 257
2-port without employing transmission lines between the source and the input port and the
output port and the load.)
The model shown in Figure 8.2 makes it clear that setting Z
L
= Z
o
will make a
2
= 0,
since there will be no reection from the load termination. Notice that setting Z
s
= Z
o
would not make a
1
= 0 in the circuit shown in Figure 8.2, because the source would cause
a non-zero incident wave (a
1
) to exist on the input line.
Now consider S
21
:
S
21
=
b
2
a
1
|
a2=0
(8.25)
As discussed earlier, to set a
2
= 0, terminate port 2 with Z
o
. To aid in obtaining an
intuitive feel for the signicance of S
21
, it is helpful to consider the special circuit shown in
Figure 8.3 where the source impedance is assumed to be equal to the reference impedance.
2-port
V
1
V
2
I
2
I
1
I
1
+
-
I
2
V
s
a
1
b
1 b
2
+
-
Z
o
Z
o
Figure 8.3: A 2-port embedded in a system with Z
S
= Z
L
= Z
o
.
Note that the circuit has been drawn without showing any intervening lengths of trans-
mission line between the 2-port and the terminations. The as and bs can be thought of
as existing at the node where the the 2-port is connected to the terminations. To solve
for S
21
, rst relate b
2
to the output voltage using Equation 8.4 and the auxiliary relation
V
2
/I
2
= Z
o
:
b
2
=
1
2
p
Z
o
(V
2
Z
o
I
2
) (8.26)
= V
2
/
Z
o
Then a
1
can be related to the open circuit voltage of the source
a
1
=
1
2
p
Z
o
(V
1
+ Z
o
I
1
) (8.27)
using
I
1
= (V
s
V
1
)/Z
o
(8.28)
Thus
a
1
=
1
2
p
Z
o
(V
1
+ V
s
V
1
) (8.29)
= V
s
/2
Z
o
Finally,
S
21
=
b
2
a
1
|
a2=0
=
V
2
V
s
/2
|
Z
L
=Zo
(8.30)
258 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
So for the circuit shown in Figure 8.3, S
21
is the voltage across the load divided by 1/2 of
the open circuit source voltage. Notice that if the 2-port is removed and the load connected
directly to the source, then the voltage across the load would be V
s
/2. In other words, S
21
is
the insertion voltage gain in a system where the source and load impedances are both Z
o
.
Remember that this result is not the denition of S
21
; it is a special result that was derived
from the denition for the case when both the source and load impedances are equal to Z
o
.
This result is often useful when it is necessary to derive an expression for S
21
starting from
the circuit model of a 2-port. S
12
has a similar interpretation - it is the reverse insertion
voltage gain, i.e. the insertion gain that would be measured if the two port was inserted
into the system backwards, i.e. with the output port connected to the source and input port
connected to the load. The insertion gain interpretion is useful to remember when trying to
get an intuitive feel for published or measured values of S
21
or S
12
.
8.2.1 Example - Computing S-parameters for a given circuit model
The 2-port in Figure 8.4 consists of a series impedance. To nd S
11
, terminate the 2-port
Z
S
Figure 8.4: 2-port consisting of a series impedance.
in Z
o
and nd Z
in
as in Figure 8.5.
Z
S
Z
o
Z
in
Figure 8.5: Setup for nding S
11
.
Z
in
= Z
s
+Z
o
(8.31)
S
11
=
Z
in
Z
o
Z
in
+Z
o
=
Z
s
Z
s
+ 2Z
o
(8.32)
Because the circuit is not changed if the two ports are reversed, S
22
= S
11
.
To nd S
21
, refer to Figure 8.6.
V
2
= V
s
Z
o
2Z
o
+Z
s
(8.33)
V
2
V
s
=
Z
o
2Z
o
+Z
s
S
21
=
V
2
V
s
/2
=
2Z
o
2Z
o
+Z
s
(8.34)
8.2. INTERPRETATION OF S-PARAMETERS 259
V
s
Z
o
Z
S
Z
o
+
V
2
1
) +<(V
2
I
2
) = 0. (8.35)
In terms of the variables used to dene the scattering parameters, this constraint becomes:
|a
1
|
2
|b
1
|
2
+ |a
2
|
2
|b
2
|
2
= 0. (8.36)
Use the denitions of the S-parameters to write the b
i
in terms of the a
i
, then equation 8.36
can be written as
|a
1
|
2
+ |a
2
|
2
= (|S
11
|
2
+ |S
21
|
2
)|a
1
|
2
+ (|S
12
|
2
+ |S
22
|
2
)|a
2
|
2
+ 2<{(S
11
S
12
+S
21
S
22
)a
1
a
2
}.
260 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
Since this constraint must be true for any possible excitation, it must hold for a
1
= 0 and
for a
2
= 0. Setting a
1
= 0 we nd that
|S
12
|
2
+ |S
22
|
2
= 1. (8.37)
Setting a
2
= 0 we nd that
|S
11
|
2
+ |S
21
|
2
= 1. (8.38)
Hence, for arbitrary a
1
and a
2
we must also have
S
11
S
12
+S
21
S
22
= 0. (8.39)
These three relationships say that the dot product of a column of [S] with the conjugate of
the same column is unity, whereas the dot product of a column of [S] with the conjugate of
the other column is zero. Another way to write these relationships is
[S][S]
= [I],
where [ ]
= [S]
[S] = [I].
8.3 Applications of Scattering Parameters
Consider in Figure 8.7 a 2-port with arbitrary source and load terminations (Z
S
, Z
L
).
Z
S
Z
L
V
s
in
out
S
L
[S]
Z
o
Figure 8.7: 2-port with arbitrary source and load terminations
The reference impedance is assumed to be Z
o
. Dene
in
and
out
to be the input and
output reection coecients, i.e.,
in
=
Z
in
Z
o
Z
in
+ Z
o
(8.40)
out
=
Z
out
Z
o
Z
out
+ Z
o
(8.41)
8.3. APPLICATIONS OF SCATTERING PARAMETERS 261
The source reection coecient,
S
, is dened as
S
=
Z
S
Z
o
Z
S
+ Z
o
(8.42)
and the load reection coecient,
L
, is dened as
L
=
Z
L
Z
o
Z
L
+ Z
o
(8.43)
8.3.1 Derivation of Input and Output Reection Coecients
We will now derive expressions for
in
and
out
in terms of the source and load reection
coecients and the S-parameters of the 2-port. Consider the input reection coecient,
in
= b
1
/a
1
. We expect
in
to be a function of the four S-parameters as well as the
load reection coecient,
L
. The relationship can be derived by noting that the load
termination imposes the following relationship between a
2
and b
2
:
L
=
a
2
b
2
(8.44)
Using this relationship and Equation 8.10 we can write
a
2
= a
1
S
21
L
1 S
22
L
(8.45)
Inserting Equation 8.45 into Equation 8.9 yields
in
=
b
1
a
1
= S
11
+
S
12
S
21
L
1 S
22
L
(8.46)
Similarly, it can be shown that the output reection coecient is given by
out
=
b
2
a
2
= S
22
+
S
12
S
21
S
1 S
11
S
(8.47)
Inspection of Equation 8.46 shows that the input reection coecient reduces to S
11
if
the load reection coecient is 0 (
L
= 0). This is consistent with the discussion in the
previous section regarding the interpretation of S
11
. Similarly,
out
= S
22
if
S
= 0. In
general, however, the input and output reection coecients depend on the way the 2-port
is terminated. In the special case of a 2-port with S
12
= 0, the input and output reection
coecients are independent of the terminations. Such a 2-port is said to be unilateral, since
the device does not exhibit any reverse transmission. In other words, if a unilateral 2-port
(S
12
= 0) is excited at port 2, no response will result at port 1. Although it is usually not
possible to construct a 2-port that is perfectly unilateral, in some cases the S
12
coecient
is small enough that the 2-port can be considered to be approximately unilateral. The case
where S
21
= 0 is also unilateral but not useful, since this 2-port would not give a response
at the output when excited at the input.
8.3.2 Stability of 2-ports
Before detailed design calculations based on a particular 2-port are made, it is usually
necessary to investigate whether the 2-port is potentially unstable. Of course, stability is
only a concern when the 2-port contains active elements such as a transistor or a negative
resistance device. The questions that need to be answered are:
262 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
1. Is the 2-port unconditionally stable? That is, is there any combination of passive
source and load terminations (
L
,
S
) for which the 2-port will oscillate? If not, the
2-port is said to be unconditionally stable.
2. If the 2-port is not unconditionally stable, i.e., if it is potentially unstable, then we
would like to be able to determine which source and load terminations make it unstable.
If the 2-port is to be used as an amplier, we would avoid the unstable terminations.
If the goal is to design an oscillator, the designer would deliberately choose source and
load terminations to cause oscillation.
The stability question can be studied using the negative resistance concept. Considering
Figure 8.8, suppose that one of the impedances represents the input or output impedance
of the 2-port.
Z
1
Z
2
Figure 8.8: Z
1
or Z
2
represents input or output impedance of 2-port
Remember that the circuit will oscillate when the two impedances are connected if Re
[Z
1
+ Z
2
] 0 and Im [Z
1
+ Z
2
] = 0. Thus, to make an oscillator, one of the Z
i
s must
have a negative real part.
For a particular 2-port, if there is some load impedance that makes the real part of
Z
in
0, or if there is some source impedance that makes the real part of Z
out
0, then
the 2-port is potentially unstable because it is possible to choose a passive source or load
termination that will make the system oscillate.
An impedance with a negative resistive part corresponds to a reection coecient with
a magnitude greater than 1. For example, consider Z
in
and write Z
in
= R
in
+ j X
in
, then
in
=
Z
in
Z
o
Z
in
+ Z
o
(Z
o
= R
o
(real)) (8.48)
=
(R
in
R
o
) + jX
in
(R
in
+ R
o
) + jX
in
|
in
| =
(R
in
R
o
)
2
+ X
2
in
(R
in
+ R
o
)
2
+ X
2
in
1/2
(8.49)
Examination of Equation 8.49 leads to the following observations:
If R
in
> 0 ) |
in
| < 1 (8.50)
If R
in
0 ) |
in
| 1 (8.51)
To determine whether a 2-port is potentially unstable, we check to see if either |
in
| or
|
out
| can be larger than or equal to unity.
8.3. APPLICATIONS OF SCATTERING PARAMETERS 263
First consider |
in
|:
|
in
| = |S
11
+
S
12
S
21
L
1 S
22
L
| (8.52)
= |
S
11
L
(S
11
S
22
S
12
S
21
)
1 S
22
L
|
Dene the determinant of the S-parameter matrix, D, as follows:
D = S
11
S
22
S
12
S
21
(8.53)
Then Equation 8.52 becomes
|
in
| = |
S
11
L
D
1 S
22
L
| (8.54)
We can now set |
in
| = 1 and solve for the corresponding locus of points in the
L
plane,
i.e., we can solve for the values of
L
that make |
in
| = 1. Setting |
in
| = 1:
|S
11
L
D| = |1 S
22
L
| (8.55)
The absolute value signs can be eliminated by squaring both sides of Equation 8.55:
(S
11
L
D)(S
11
L
D
) = (1 S
22
L
)(1 S
22
L
) (8.56)
Expanding both sides of the equation and collecting terms yields
|
L
|
2
+
2 Re(
L
(S
22
DS
11
))
|D|
2
|S
22
|
2
+
|S
11
|
2
1
|D|
2
|S
22
|
2
(8.57)
In deriving Equation 8.57, use has been made of the following identity
z + z
= 2Re(z) (8.58)
where the operator Re() extracts the real part of its argument. Also note that
z z
= 2Im(z) (8.59)
where the operator Im() extracts the imaginary part of its argument. Now, it is convenient
to rewrite Equation 8.57 in terms of the real and imaginary parts of
L
.
Let
L
= U
L
+ jV
L
and substitute Equation 8.57. After some fairly extensive algebraic
manipulation, Equation 8.57 can be written in the form
(U
L
U
CL
)
2
+ (V
L
V
CL
)
2
= r
L
2
(8.60)
where
U
CL
=
Re(DS
11
S
22
)
|D|
2
|S
22
|
2
(8.61)
V
CL
=
Im(S
22
DS
11
)
|D|
2
|S
22
|
2
(8.62)
r
L
=
|S
12
S
21
|
||S
22
|
2
|D|
2
|
(8.63)
264 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
Equation 8.60 indicates that the locus of points in the
L
plane that correspond to
input reection coecients with unit magnitude is a circle with center at the complex point
C
L
= U
CL
+jV
CL
and radius r
L
. It is convenient to write the coordinates of the center of
the stability circle as a complex number, i.e. dene C
L
= U
CL
+jV
CL
. Then
C
L
=
S
22
D
S
11
|S
22
|
2
|D|
2
(8.64)
The circle, when plotted on the
L
plane, is referred to as the
L
-plane stability circle.
Consider the stability circle shown in Figure 8.9.
U
L
V
L
L
Plane
C
L
r
L
Stability
Circle
Unit
Circle
Figure 8.9:
L
-plane stability circle
Figure 8.9 contains two circles. The circle that is centered on the origin is a unit circle
(circle with radius = 1) and represents the outer boundary of the Smith Chart. Points
within that circle correspond to load reection coecients (
L
) with magnitudes less than 1
or, equivalently, load impedances with a positive real part. The other circle is the stability
circle, and values of
L
which lie on that circle will map to input reection coecients
with magnitude equal to 1 or, equivalently, input impedances that are purely reactive. The
stability circle represents the boundary between the region of the
L
plane that maps to
input reection coecients with magnitude less than one (|
in
| < 1) and the region that
maps to (|
in
| > 1). The region of the
L
plane that maps to (|
in
| > 1) is called the
unstable region of the
L
plane. The two regions (stable and unstable) correspond to
the regions inside and outside the stability circle. To decide whether the unstable region
corresponds to the region inside or outside of the stability circle, it is sucient to examine
the value of S
11
for the 2-port under consideration. Note carefully that the origin of the
L
plane (the point
L
= 0) maps to S
11
, i.e., recall
in
= S
11
+
S
12
S
21
L
1 S
22
L
(8.65)
so that when
L
= 0, we have
in
= S
11
(8.66)
8.3. APPLICATIONS OF SCATTERING PARAMETERS 265
Now consider Figure 8.9 and suppose that it is known that |S
11
| < 1. The origin in the
L
plane is outside the stability circle and this point maps to S
11
. This leads us to conclude:
(i) the region of the
L
plane that lies outside the stability circle maps to |
in
| < 1 and (ii)
the region inside the stability circle maps to |
in
| > 1. The region inside the stability circle
would therefore be referred to as the unstable region of the
L
plane.
Thus far we have treated only the so-called
L
-plane stability circle. A complete charac-
terization of the 2-ports stability requires that the
S
-plane stability circle also be studied.
The
S
-plane stability circle corresponds to the values of
S
which map to output reection
coecients (
out
) with magnitude equal to one. The
S
-plane stability circle is described
by
(U
S
U
CS
)
2
+ (V
S
V
CS
)
2
= r
S
2
(8.67)
where
C
S
= U
CS
+jV
CS
(8.68)
=
S
11
D
S
22
|S
11
|
2
|D|
2
(8.69)
r
S
=
|S
12
S
21
|
||S
11
|
2
|D|
2
|
(8.70)
A decision regarding whether the unstable region in the
S
-plane lies inside or outside
of the
S
-plane stability circle is made by examining the magnitude of S
22
and noting that
the point
S
= 0 (the origin of the
S
-plane) maps to S
22
.
It is now appropriate to dene the concept of unconditional stability. An unconditionally
stable 2-port has the property that no choice of passive source and load terminations will
make the 2-port oscillate. In other words, an unconditionally stable 2-port has |
in
| < 1 and
|
out
| < 1 for any choice of passive source and load terminations. The restriction to passive
sources and loads means that we limit our attention to sources with |
S
| 1 and loads with
|
L
| 1, i.e., those regions of the
S
- and
L
-planes that lie inside and on the unit circle
centered on the origin. A 2-port is unconditionally stable if the unstable regions of the
S
-
and
L
-planes lie completely outside of the unit circles. Once the center coordinates and
the radii of the stability circles have been computed, the circles can be plotted and, utilizing
the known magnitudes of S
11
and S
22
, we can determine whether the unstable regions lie
outside of the unit circles.
As an alternative to plotting the stability circles, it is possible to derive a relatively
simple algebraic criterion (or set of criteria) that must be satised in order for a 2-port to
be unconditionally stable. The derivation of one such criterion is given in section 8.3.4. A
summary of the various sets of criteria that have been derived is given here. In each case,
if the criterion (or set of criteria) is satised, then the 2-port is unconditionally stable:
K > 1, |S
12
S
21
| < 1 |S
11
|
2
(8.71)
K > 1, |S
12
S
21
| < 1 |S
22
|
2
(8.72)
K > 1, B
1
> 0 (8.73)
K > 1, B
2
> 0 (8.74)
K > 1, |D| < 1 (8.75)
266 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
ES
=
1 |S
11
|
2
|S
22
S
11
D| + |S
12
S
21
|
> 1 (8.76)
0
ES
=
1 |S
22
|
2
|S
11
S
22
D| + |S
12
S
21
|
> 1 (8.77)
where
K =
1 |S
11
|
2
|S
22
|
2
+ |D|
2
2 |S
12
| |S
21
|
> 1 (8.78)
D = S
11
S
22
S
12
S
21
(8.79)
B
1
= 1 + |S
11
|
2
|D|
2
|S
22
|
2
(8.80)
B
2
= 1 + |S
22
|
2
|D|
2
|S
11
|
2
(8.81)
Any one set of criteria given by 8.71-8.77 is a necessary and sucient set of criteria for a
2-port to be unconditionally stable. (It should be noted that some textbooks state that the
union of the criteria listed in 8.71 and 8.72 are necessary and sucient criteria. This results
in 3 criteria, but it can be shown that either 8.71 or 8.72 are necessary and sucient, i.e.
at most only 2 inequalities must be checked.)
If any one of the sets of stability criteria are satised, then no choice of passive source and
load terminations will make the 2-port oscillate. This statement applies only at the frequency
where the S-parameters were measured. It is possible for a 2-port to be unconditionally
stable in a certain frequency band but potentially unstable in some other frequency band.
In practice, stability should be checked at many frequencies in the bandwidth within which
the 2-port has appreciable gain. This requires that the S-parameters be measured at many
frequencies. Checking one of the sets of conditions given in equations 8.71-8.77 is easier than
plotting the stability circles, especially when one wishes to quickly determine the stability
status over a wide frequency range; however if the 2-port turns out to be potentially unstable,
then the stability circles provide detailed information on which terminations make the 2-port
unstable.
Finally, note that the parameter K dened in Equation Fac8.78 and used in criteria
8.71 through 8.75 is called the Rollet Stability Factor. While the condition K>1, by itself,
does not guarantee that a 2-port is unconditionally stable, it turns out that K plays an
important role in determining whether or not a 2-port can be simultaneously matched with
passive terminations at both the input and output ports.
8.3.3 Example - 2-port stability analysis.
A 2-port has S-parameters (Z
o
= 50 )
S
11
= 0.4\ 20
(8.82)
S
12
= 0.1\40
(8.83)
S
21
= 7.5\150
(8.84)
S
22
= 0.6\ 30
(8.85)
8.3. APPLICATIONS OF SCATTERING PARAMETERS 267
The Rollett Stability factor for this 2-port is K = 0.853 < 1, so the 2-port is potentially
unstable. The coordinates of the stability circles are
U
CS
= 0.163 V
CS
= 0.589 r
S
= 1.17
U
CL
= 0.401 V
CL
= 0.913 r
L
= 1.70
(8.86)
The stability circles for this 2-port are plotted in Figures 8.10 and 8.11. The stable region
U
S
V
S
S
Plane
C
S
r
S
Stability
Circle
Unit
Circle
Figure 8.10:
S
plane stability circle. The stable region, corresponding to values of |
S
|
that map to |
out
| < 1, is shaded.
of the
S
- and
L
-planes is shaded. For the
S
-plane stability circle, the stable region was
identied by noting that the origin of the
S
-plane (
S
= 0) maps to an output reection
coecient
out
= S
22
. Since the origin of the
S
-plane is in the interior of the stability circle
and |S
22
| < 1, we conclude that the interior of the stability circle maps to output reection
coecients |
out
| < 1; hence, the interior of the stability circle is the stable region. Similar
reasoning was used to determine that the interior of the
L
-plane stability circle was stable.
To verify that the stable and unstable regions have been correctly identied, consider the
two points labeled A and B in the
L
-plane plot. The point A is in the unstable region
and represents a load reection coecient of
L
= 0.5 + j0.7 = 0.86\54
in
= 0.70\ 83
.
The following example is identical to the previous one, except that the magnitude of S
12
has been reduced from 0.1 to 0.01.
268 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
U
L
V
L
L
Plane
C
L
r
L
Stability
Circle
Unit
Circle
A
B
Figure 8.11:
L
plane stability circle. The stable region, corresponding to values of
L
that
map to |
in
| < 1, is shaded. Point A, at
L
= 0.5 +j0.7 = 0.86\54
(8.87)
S
12
= 0.01\40
(8.88)
S
21
= 7.5\150
(8.89)
S
22
= 0.6\ 30
(8.90)
The stability factor, K, is 3.74 (> 1), and we also have
1 |S
11
|
2
> |S
12
S
21
| (8.91)
1 |S
22
|
2
> |S
12
S
21
| (8.92)
so the 2-port is unconditionally stable. (Note, only one of 8.91 and 8.92 needs to be checked.)
Stability circles for this example are shown in Figures 8.12 and 8.13. As in the rst example,
U
S
V
S
S
Plane C
S
r
S
Unstable
Region
Unit
Circle
Figure 8.12:
S
plane stability circle. The unstable region is shaded.
we have made use of the fact that the origin of the
S
- and
L
-planes map to output and
input reection coecients, respectively, with magnitudes less than 1. Hence the origin of
each plane is in the stable region. Since the origin is outside the stability circle, the region
outside the circle is the stable region and the region inside is the unstable region. Since the
unstable regions do not include any passive source or load terminations, the 2-port is said
to be unconditionally stable.
In the following section, a simple test is derived that can predict whether or not a
particular 2-port is unconditionally stable.
270 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
U
L
V
L
L
Plane
Unstable
Region
Unit
Circle
Figure 8.13:
L
plane stability circle. The unstable region is shaded.
8.3.4 Derivation of a Criterion for Unconditional Stability of 2-
ports
All of the essential information necessary to investigate the stability of a 2-port (at a par-
ticular frequency) is contained in the 4 parameters that dene the stability circles in the
S
and
L
planes. In this section, we derive a criterion that can be used to quickly check
to see if a 2-port is unconditionally stable. As already pointed out, numerous sets of nec-
essary and sucient criteria have been derived that can be used to determine whether or
not a 2-port is unconditionally stable.
1
The criterion derived in this section is particularly
interesting because it involves only a single inequality and because the numerical value of
the stability parameter has the useful property that its magnitude represents the smallest
distance between the origin of the reection coecient plane and the edge of the unstable
region.
In order for a 2-port to be unconditionally stable, it is necessary for the unstable region
in the
L
and
S
planes to lie outside of the unit circle. Consider the
L
plane rst, and
assume that |S
11
| < 1 (this is a necessary condition for unconditional stability, because if
|S
11
| 1 a negative, or zero, real part of Z
IN
is obtained by setting Z
L
= Z
o
. In this case,
the 2-port can oscillate with a passive source termination and hence is potentially unstable.)
There are two cases to consider.
1. First, suppose that the stability circle does not enclose the origin of the
L
plane
(i.e. |C
L
| > r
L
). Then the inside of the stability circle is the unstable region. If the
2-port is unconditionally stable it is necessary for the minimum distance between the
origin and the edge of the stability circle to be larger than 1, i.e. it is necessary that
|C
L
| r
L
> 1. Inserting the equations for |C
L
| and r
L
we nd that this amounts to:
|S
22
D
S
11
|
||S
22
|
2
|D|
2
|
|S
12
S
21
|
||S
22
|
2
|D|
2
|
> 1 (8.93)
1
Lombardi, G. and B. Neri, Criteria for the evaluation of unconditional stability of microwave linear
two-ports: a critical review and new proof, IEEE Trans. MTT, Vol. 47, No. 6, June, 1999, p746.
8.3. APPLICATIONS OF SCATTERING PARAMETERS 271
or
|S
22
D
S
11
| |S
12
S
21
|
||S
22
|
2
|D|
2
|
> 1 (8.94)
2. Now consider the other possibility, namely that the stability circle encloses the origin,
i.e. that |C
L
| < r
L
. In this case, the necessary condition |S
11
| < 1 implies that the
exterior of the stability circle is the unstable region. In order to have the unstable
region lie completely outside of the unit circle, we require that the stability circle
completely encloses the unit circle centered at
L
= 0, i.e.:
r
L
|C
L
| > 1
or
|S
12
S
21
| |S
22
DS
11
|
||S
22
|
2
|D|
2
|
> 1 (8.95)
Note carefully that equation 8.94 guarantees that the unstable region is outside of the unit
circle provided that |C
L
| > r
L
whereas equation 8.95 guarantees that the unstable region is
outside of the unit circle when |C
L
| < r
L
. Now consider what the statement |C
L
| > r
L
(or
|C
L
| < r
L
) implies in terms of the S parameters. Consider |C
L
| > r
L
:
|S
22
D
S
11
|
||S
22
|
2
|D|
2
|
>
|S
12
S
21
|
||S
22
|
2
|D|
2
|
or
|S
22
D
S
11
| > |S
12
S
21
|
Squaring both sides (note that squaring both sides does not lose any generality, because
both sides are always positive) yields
(S
22
D
S
11
)(S
22
DS
11
) > |S
12
S
21
|
2
and expanding the LHS yields
|S
22
|
2
S
22
DS
11
S
22
D
S
11
+ |D|
2
|S
11
|
2
> |S
12
S
21
|
2
Expand the second and third terms on the LHS using D = S
11
S
22
S
12
S
21
:
|S
22
|
2
S
11
S
22
(S
11
S
22
S
12
S
21
) S
11
S
22
(S
11
S
22
S
12
S
21
) + |D|
2
|S
11
|
2
> |S
12
S
21
|
2
and group terms on the LHS:
|S
22
|
2
2|S
11
|
2
|S
22
|
2
+S
11
S
22
S
12
S
21
+S
11
S
22
S
12
S
21
+ |D|
2
|S
11
|
2
> |S
12
S
21
|
2
(8.96)
Now, noting that
|D|
2
= |S
11
|
2
|S
22
|
2
S
11
S
22
S
12
S
21
S
11
S
22
S
12
S
21
+ |S
12
|
2
|S
21
|
2
(8.97)
Equation 8.97 can be used to eliminate the complex terms (3rd and 4th term on LHS) in
equation 8.96. The result is:
|S
22
|
2
|S
11
|
2
|S
22
|
2
+ |S
12
|
2
|S
21
|
2
|D|
2
+ |D|
2
|S
11
|
2
> |S
12
|
2
|S
21
|
2
272 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
or, after grouping terms, we determine that |C
L
| > r
L
amounts to:
(|S
22
|
2
|D|
2
)(1 |S
11
|
2
) > 0 (8.98)
Since we have already assumed that |S
11
| < 1, the second term on the LHS is always positive
so the combined requirements that the stability circle not enclose the origin ( |C
L
| > r
L
)
and that the inside of the stability circle represent the unstable region (|S
11
| < 1) means
that:
|S
22
|
2
|D|
2
> 0 (8.99)
Similarly, the requirements that the stability circle encloses the origin (|C
L
| < r
L
) and that
the inside of the stability circle represents the unstable region (|S
11
| < 1) amounts to:
|S
22
|
2
|D|
2
< 0 (8.100)
We now know that criterion 8.71 applies only when |S
22
|
2
|D|
2
> 0 and criterion 8.72
applies when |S
22
|
2
|D|
2
< 0. Note that criteria 8.71 and 8.72 dier only in the sign of the
numerator on the LHS, and that the denominator is the magnitude of |S
22
|
2
|D|
2
. Thus,
these criteria can be combined into into a single inequality which is valid for either case:
|S
22
D
S
11
| |S
12
S
21
|
|S
22
|
2
|D|
2
> 1 (8.101)
There is an apparent singularity in equation 8.101 when |S
22
|
2
|D|
2
= 0. This corresponds
to the situation where the radius of the stability circle approaches innity, i.e. when the
stability circle degenerates into a straight line. The singularity can be removed by noting
that:
|S
22
|
2
|D|
2
=
|S
22
S
11
D|
2
|S
12
S
21
|
2
1 |S
11
|
2
=
(|S
22
S
11
D| |S
12
S
21
|)(|S
22
S
11
D| + |S
12
S
21
|)
1 |S
11
|
2
(8.102)
Use equation 8.102 in equation 8.101 to write:
ES
=
1 |S
11
|
2
|S
22
D
S
11
| + |S
12
S
21
|
> 1 (8.103)
The parameter
ES
is referred to as the Edwards-Sinsky stability criterion,
2
and has a
useful geometric interpretation:
ES
is the minimum distance between the origin of the
L
plane and the unstable region. If the parameter
ES
is negative, it means that the unstable
region includes the origin of the
L
plane.
If the identical analysis is carried out for the
S
plane stability circles, we derive the
dual constraint:
0
ES
=
1 |S
22
|
2
|S
11
D
S
22
| + |S
12
S
21
|
> 1 (8.104)
The geometric interpretation of
0
ES
is the same as that of
ES
, but applied to the
S
plane.
It turns out that either criterion 8.103 or criterion 8.104 is a necessary and sucient
test for unconditional stability of a 2-port, even though each one was derived separately by
2
Edwards, M. L. and J. H. Sinsky, A new criterion for linear 2-port stability using a single geometrically
derived parameter, IEEE Trans. MTT, Vol. 40., No. 12, December 1992, p2303.
8.3. APPLICATIONS OF SCATTERING PARAMETERS 273
considering only one of the
L
plane or
S
plane stability circles. To show that this is true
we can prove that if
ES
1 then
0
ES
1 and the converse are both true statements.
Suppose
ES
1 and choose a passive load termination, call it
Lu
, that causes |
in
| to be
1 , i.e. |
in
(
Lu
)| 1. The negative resistance criterion for steady-state oscillation will be
satised at the input port if the source reection coecient is chosen to be
Su
=
in
(
Lu
)
1
because this choice causes
Su
in
(
Lu
) = 1. (As an exercise, you may wish to verify that
the condition Z
1
+ Z
2
= 0 is equivalent to
1
2
= 1.) Note that
Su
is passive, since
|
Su
| = 1/|
in
(
Lu
)| 1. Now, it is left as an exercise to show that
out
(
Su
) =
1
Lu
.
Thus
out
(
Su
)
Lu
= 1, which means that the oscillation condition is also satised at the
output port. The converse can be proven by simply changing subscripts in the preceeding
argument. It follows that if
ES
> 1 then
0
ES
> 1 and vice versa. Hence we can determine
whether or not a 2-port is unconditionally stable by checking either criterion 8.103 or 8.104.
8.3.5 Terminations for Simultaneous Conjugate Match
Whenever K > 1 it is possible to nd a combination of passive source and load termina-
tions for which both the input and output of the 2-port are conjugately matched. For a
given available source power, a simultaneous conjugate match condition leads to the highest
possible power delivered to the load. The particular source and load reection coecients
which result in a simultaneous conjugate match at both ports are denoted by
ms
and
ml
.
Equations for
ms
and
ml
can be derived by solving Equations 8.105 and 8.106, which
enforce the conjugate match relationship at the input and output of the 2-port, i.e.,
in
=
ms
= S
11
+
S
12
S
21
ml
1 S
22
ml
(8.105)
out
=
ml
= S
22
+
S
12
S
21
ms
1 S
11
ms
(8.106)
The solution to Equations 8.105 and 8.106 is given by
ms
=
B1
p
B
2
1
4|C1|
2
2C1
B
1
= 1 + |S
11
|
2
|D|
2
|S
22
|
2
C
1
= S
11
DS
22
(8.107)
ml
=
B2
p
B
2
2
4|C2|
2
2C2
B
2
= 1 + |S
22
|
2
|D|
2
|S
11
|
2
C
2
= S
22
DS
11
(8.108)
The proper choice of sign in Equations 8.107 and 8.108 is determined by the requirement
that |
ms
| < 1 and |
ml
| < 1. In other words, the correct solution in each case is the
one with a magnitude less than 1. Equations 8.107 and 8.108 will yield valid solutions
whenever K > 1. Since this is only a necessary condition for stability, it is possible to nd
a simultaneous conjugate match solution for some potentially unstable 2-ports (those with
K > 1).
274 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
Suppose
ms
and
ml
are known for a particular 2-port (with K > 1). Then the proce-
dure for designing an amplier which maximizes the power delivered to the load consists of
designing matching networks that transform the actual source and load reection coecients
into
ms
and
ml
, as in Figure 8.14.
Z
S
Z
L
V
s
S
L
[S]
Z
o
Lossless
Matching
Network
ms
Lossless
Matching
Network
ms
ml
ml
L
K > 1
Figure 8.14: An amplier that is simultaneously matched to source and load terminations.
A conjugate match exists at all ports.
If a 2-port does not satisfy the conditions for unconditional stability, then it is said
to be potentially unstable. Useful and stable ampliers can be designed with potentially
unstable 2-ports, however it is necessary to carefully choose source and load terminations
to avoid the unstable regions of the
S
- and
L
- planes. If K > 1, then one such choice of
terminations is
S
=
ms
and
L
=
ml
. If K 1, then it will not be possible to achieve
a simultaneous conjugate match, and the terminations will have to be chosen according to
a carefully measured tradeo between transducer gain and relative stability. A 2-port with
K 1 can be modied using resistive loading to raise the stability factor so that the loaded
2-port can be simultaneously conjugte-matched.
Of course, potentially unstable 2-ports can also be used to design oscillators. For exam-
ple, a basic oscillator design could be accomplished by choosing a
S
in the unstable region,
i.e., a
S
that maps to |
out
| > 1. Using this source termination, the output impedance of
the 2-port will have a negative real part. The load termination is then chosen such that the
negative resistance criterion for oscillation is satised at the output port. As shown earlier,
this will automatically cause the negative resistance criterion to be satised at the input
port.
8.3.6 Power Gains
1. Operating Power Gain (or just power gain)
G
Pout
Pin
=
|S21|
2
(1|
L
|
2
)
(1|S11|
2
)+|
L
|
2
(|S22|
2
|D|
2
)2Re(
L
C2)
(8.109)
with
D = S
11
S
22
S
12
S
21
C
2
= S
22
DS
11
Note that G depends only on the load termination,
L
.
2. Transducer Power Gain
G
T
Pout
Pavs
=
|S21|
2
(1|
S
|
2
)(1|
L
|
2
)
|(1S11
S
)(1S22
L
)S12S21
L
S
|
2
(8.110)
8.3. APPLICATIONS OF SCATTERING PARAMETERS 275
Notice that G
T
depends on both terminations,
S
and
L
.
3. Available Power Gain
G
A
Pavo
Pavs
=
|S21|
2
(1|
S
|
2
)
(1|S22|
2
)+|
S
|
2
(|S11|
2
|D|
2
)2Re(
S
C1)
(8.111)
with
D = S
11
S
22
S
12
S
21
C
1
= S
11
DS
22
Note that G
A
depends solely on the source impedance. It tells us how much power is
potentially available from the output of the 2-port.
4. Unilateral Transducer Power Gain
If the internal feedback within the device is small (S
12
0), then the device can be
considered unilateral. This approximation yields a particularly simple form for the
transducer gain:
G
TU
= G
T
|
S12=0
G
TU
=
1|
S
|
2
|1S11
S
|
2
|S
21
|
2
1|
L
|
2
|1S22
L
|
2
(8.112)
This form clearly shows the eects of input and output mismatch which are contained
in the rst and last terms of the expression.
5. Maximum Available Gain
A 2-port that is conjugately matched at both ports (as in Figure 8.14) will have
G = G
A
= G
T
= G
A,max
where:
G
A,max
=
S
21
S
12
[K
K
2
1]
(8.113)
The upper sign applies when B
1
< 0 and the lower sign applies when B
1
> 0. The
maximum available gain could also be calculated by using
ml
in the formula for G,
or
ms
in the formula for G
A
, or (
ms
,
ml
) in the formula for G
T
.
6. Maximum Stable Gain
A 2-port with K < 1 cannot be simultaneously conjugate-matched at both ports
using passive terminations. The stability factor of such a 2-port can be increased
using resistive loading at the input and/or output. If the resistive loading is chosen to
raise the stability factor of the loaded 2-port to a value slightly larger than 1, then the
loaded 2-port can be simultaneously conjugate-matched at both ports and will have
G
A,max
' |
S21
S12
|. Hence, for a 2-port with K < 1 we dene the maximum stable gain
to be:
G
MS
=
S
21
S
12
0 e
l
e
l
0
X
=
Z
X
Zo
Z
X
+Zo
Z
X
= Zo
1 +
X
1
X
Input reection coecient with arbitrary Z
L
:
in
= S
11
+
S
12
S
21
L
1 S
22
L
Output reection coecient with arbitrary Z
S
:
out = S
22
+
S
12
S
21
S
1 S
11
S
Voltage gain with arbitrary Z
L
(voltage gain does not depend on Z
S
):
A
V
=
V
2
V
1
=
S
21
(1 +
L
)
(1 +S
11
)(1 S
22
L
) +S
12
S
21
L
Current gain with arbitrary Z
L
(current gain does not depend on Z
S
):
A
I
=
I
2
I
1
=
S
21
(
L
1)
(1 S
11
)(1 S
22
L
) S
12
S
21
L
Denitions of some commonly used quantities:
D = S
11
S
22
S
12
S
21
K =
1|S
11
|
2
|S
22
|
2
+|D|
2
2|S
12
S
21
|
B
1
= 1 + |S
11
|
2
|D|
2
|S
22
|
2
C
1
= S
11
DS
22
B
2
= 1 + |S
22
|
2
|D|
2
|S
11
|
2
C
2
= S
22
DS
11
Operating Power Gain:
G
Pout
P
in
=
|S
21
|
2
`
1 |
L
|
2
(1 |S
11
|
2
) + |
L
|
2
(|S
22
|
2
|D|
2
) 2Re(
L
C
2
)
Available Power Gain:
G
A
Pavo
Pavs
=
|S
21
|
2
`
1 |
S
|
2
(1 |S
22
|
2
) + |
S
|
2
(|S
11
|
2
|D|
2
) 2Re(
S
C
1
)
Transducer Power Gain:
G
T
Pout
Pavs
=
|S
21
|
2
`
1 |
S
|
2
`
1 |
L
|
2
|(1 S
11
S
)(1 S
22
L
) S
12
S
21
S
|
2
A 2-port is unconditionally stable if:
K > 1 and 1 |S
11
|
2
> |S
12
S
21
|
278 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
or, if:
K > 1 and 1 |S
22
|
2
> |S
12
S
21
|
or, if:
K > 1 and B
1
> 0
or, if:
K > 1 and B
2
> 0
or, if:
K > 1 and |D| < 1
or, if:
ES
=
1 |S
11
|
2
|S
22
S
11
D| + |S
12
S
21
|
> 1
or, if:
0
ES
=
1 |S
22
|
2
|S
11
S
22
D| + |S
12
S
21
|
> 1
Stability circles in the
L
plane:
Center is at : C
L
=
S
22
D
S
11
|S
22
|
2
|D|
2
radius : r
L
=
|S
12
S
21
|
||S
22
|
2
|D|
2
|
Stability circles in the
S
plane:
Center is at : C
S
=
S
11
D
S
22
|S
11
|
2
|D|
2
radius : r
S
=
|S
12
S
21
|
||S
11
|
2
|D|
2
|
Source and load reection coecient for simultaneous conjugate match (K must be > 1) - in each case,
choose the sign that results in a reection coecient with magnitude less than 1:
ms =
B
1
q
B
2
1
4|C
1
|
2
2C
1
ml
=
B
2
q
B
2
2
4|C
2
|
2
2C
2
Maximum available power gain (only dened for K > 1):
G
A,max
=
S
21
S
12
[K
p
K
2
1]
G
A,max
is dened only for 2-ports that can be conjugately matched at both ports (K > 1). For uncondi-
tionally stable 2-ports, B
1
> 0, use the lower (negative) sign. For potentially unstable 2-ports, B
1
0, use
the upper (positive) sign.
8.5. REFERENCES 279
8.5 References
1. Carson, Ralph S., High Frequency Ampliers, John Wiley & Sons, New York, 1975.
2. Gonzalez, Guillermo, Microwave Transistor Ampliers: Analysis and Design, Prentice-
Hall, New Jersey, 1984.
3. Liao, Samuel Y., Microwave Circuit Analysis and Amplier Design, Prentice-Hall,
New Jersey, 1987.
4. Pozar, David M., Microwave Engineering, Addison-Wesley Publishing Company, 1990.
5. Vendelin, George D., Design of Ampliers and Oscillators by the S-parameter Method,
John Wiley & Sons, New York, 1982.
6. Vendelin, George D., Anthony M. Pavio, Ulrich L. Rohde, Microwave Circuit Design
Using Linear and Nonlinear Techniques, John Wiley & Sons, 1990.
280 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
8.6 Homework Problems
1. Find the S-parameters for the T-network in Figure 8.15. Port 1 is on the left. Denote
the reference impedance by Z
o
.
Z
1
Z
2
Z
3
Figure 8.15: T-network
2. Find expressions for the S-parameters of the 2-port in Figure 8.16. Denote the refer-
ence impedance by Z
o
.
Z
2
Z
1 V
g
m
V
+
-
Figure 8.16: Unilateral hybrid-Pi model.
3. Consider an ideal transformer as shown in Figure 8.17. This 2-port is characterized
V
1
I
1
+
-
V
2
+
-
I
2
1 : N
Figure 8.17: Ideal transformer
by the following relations:
V
2
= NV
1
(8.114)
I
2
= I
1
/N
Find expressions for the S-parameters of this 2-port. Denote the reference impedance
by Z
o
.
4. Suppose that two 2-ports are cascaded. Show that the S
21
parameter for the overall
network is given by
8.6. HOMEWORK PROBLEMS 281
S
21
=
S
(1)
21
S
(2)
21
1 S
(2)
11
S
(1)
22
(8.115)
where S
(1)
ij
and S
(2)
ij
refer to the S-parameters of the rst and second 2-ports, respec-
tively.
5. A 2-port has the following S-parameters (Z
0
= 50 ):
S
11
= 0.5\ 96
(8.116)
S
12
= 0.3\50
S
21
= 5.2\45
S
22
= 0.4\ 120
(a) What is the input impedance when the 2-port is terminated with a short circuit,
i.e., Z
L
= 0?
(b) Is this 2-port unconditionally stable? (Check one of the necessary and sucient
sets of criteria.)
(c) Find the coordinates of the center and the radius of the stability circles in the
S
and
L
planes. Sketch the stability circles and shade the regions that correspond
to those values of
S
and
L
that make |
out
| > 1 or |
in
| > 1.
6. Suppose that S
12
in Problem 5 is changed to S
12
= 0.05\50
L
that make |
out
| > 1 or |
in
| > 1. Is the 2-port unconditionally stable?
7. The stability circles in the output (
L
) plane for cases (a) and (b) are shown in Figure
8.18. Suppose that in both cases it is known that |S
11
| < 1. For each case indicate
L
Plane
Stability Circle
Unit circle (|
L
| = 1)
L
Plane
Stability Circle
Unit circle (|
L
| = 1)
Figure 8.18: Case (a), left and case (b), right.
what region of the
L
plane corresponds to the stable region. The stable region of
the
L
plane corresponds to those values of
L
that will cause the input reection
coecient to have a magnitude less than 1.
282 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
8. For a particular 2-port with |S
11
| < 1 and |S
22
| < 1, the center and radius of the
stability circles in the
S
and
L
planes are known to be
C
L
= 0.269 j1.86 (8.117)
r
L
= 3.01
C
S
= 2.23 +j4.76
r
S
= 4.19
(a) Make two sketches, one each for the
S
and
L
planes, and show the stability
circles. Shade the unstable region.
(b) Is the 2-port unconditionally stable?
9. Consider the system in Figure 8.19 where R
S
= 300 and R
L
= 100 . Suppose that
the 2-port is known to be unilateral and that S
22
= 0.8 (Z
o
= 50 ). The transducer
gain of the 2-port is G
T
= 12 dB. Find the available gain of the 2-port in this system.
R
S
V
i
V
o
R
L
+
-
+
-
V
S
2-port
Figure 8.19: System with resistive source and load.
10. Consider the system in Figure 8.19 where R
S
= R
L
= 300 . The voltage gain,
A
v
=
Vo
Vi
, in this system is found to be A
v
= 6\45
(8.118)
S
12
= 0.05\45
S
21
= 3.50\60
S
22
= 0.1\ 30
where Z
out
is the output
impedance of the 2-port with the 100 source termination. Express your result
in dBm.
13. Consider the 2-port from Problem 11. Suppose the reference impedance, Z
o
, is changed
to 500. Denote the new S-parameters for this reference impedance by S
0
11
, S
0
21
, S
0
12
and S
0
22
.
(a) Find S
0
11
.
(b) Find S
0
21
. Hint: The voltage gain A
v
= V
o
/V
i
in a system with arbitrary load
and source impedance is given by the following formula:
A
v
=
S
21
(1 +
L
)
(1 S
22
L
)(1 +
IN
)
(8.119)
where
IN
is the input reection coecient and V
o
, V
i
are the voltages measured
across the output and input terminals of the 2-port, respectively.
14. A 2-port has 50 S-parameters:
S
11
= 0.3 (8.120)
S
12
= 0.01
S
21
= 10.0
S
22
= 0.1
The Rollett Stability Factor for this 2-port is K = 4.525, and the 2-port is uncondi-
tionally stable.
(a) The 2-port is used with a source having impedance Z
s
= 300 and a load
Z
L
= 50 . The power available from the source is -3 dBm. Find the power
delivered to the load. Express your result in dBm.
(b) Now suppose a lossless matching network is used between the source and the
2-port. Find the power delivered to the load. Express your result in dBm.
284 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
(c) Suppose a lossless matching network is used between the 2-port and the load (no
matching network at the input). Find the power delivered to the load. Express
your result in dBm.
(d) Suppose both input and output matching networks are used so that both ports
are conjugately matched. Find the power delivered to the load. Express your
result in dBm.
15. Show that the maximum available gain G
A,max
for a unilateral 2-port is
G
A,max
=
|S
21
|
2
(1 |S
11
|
2
)(1 |S
22
|
2
)
(8.121)
Hint: When the input and output of the 2-port are conjugately matched, G = G
T
=
G
A,max
.
16. Consider a system with source and load reection coecients
S
and
L
as shown in
Figure 8.20. Dene the insertion gain
L
V
S
2-port
S
Z
L
Z
S
Figure 8.20: System with source and load reection coecients
S
and
L
G
I
=
P
delivered to load with network
P
delivered to load without network
(8.122)
where P
delivered to load without network
is the power delivered to the load when the source
is connected directly to the load, and P
delivered to load with network
is the power delivered
to the load when the 2-port is used between the source and the load.
(a) Find an expression for the insertion gain of an arbitrary 2-port.
(b) Find an expression for the insertion gain of a lossless matching network.
17. The T-parameters (sometimes called the chain scattering parameters) are dened
in terms of the same variables as the S-parameters. They are useful when cascading
2-ports, because the T-parameter matrix for a cascade is simply the product of the
T-matrices for the individual 2-ports. (Can you show that this is true?) The denition
of the T-parameters follows from choosing a
1
and b
1
to be the independent variables,
i.e.,
a
1
= T
11
b
2
+ T
12
a
2
(8.123)
b
1
= T
21
b
2
+ T
22
a
2
8.6. HOMEWORK PROBLEMS 285
where the as and bs are dened
a
i
=
V
i
+Z
o
I
i
2
p
Z
o
(8.124)
b
i
=
V
i
Z
o
I
i
2
p
Z
o
(8.125)
Suppose the S-parameters for a 2-port are known. Find the T-parameters in terms of
the S-parameters.
18. Consider a cascade of two identical 2-ports. The cascade is used in a system with
source and load as shown in Figure 8.21:
50
[S]
P
avs
= 10dBm
1000
[S]
Figure 8.21: Cascade of two identical 2-ports
The available source power is -10dBm and the 50 S-parameters of the 2-ports are
S
11
= 0.35 (8.126)
S
12
= 0.1
S
21
= 3.0
S
22
= 0.50
The 2-ports are unconditionally stable. Answer the following questions. For parts 18a
- 18g, express all results in dB or dBm, whichever is appropriate.
(a) The power delivered to the rst 2-port.
(b) The power delivered to the second 2-port.
(c) The power delivered to the load.
(d) The transducer gain for the cascaded 2-ports.
(e) The available gain of the cascaded 2-ports.
(f) The operating power gain of the cascaded 2-ports.
(g) Suppose three lossless matching networks are used between the source and the
rst 2-port, the rst and second 2-ports, and the second 2-port and the load.
Find the power delivered to the load.
(h) Find S
11
(Z
o
= 50) for the cascaded 2-ports. The answer is NOT 0.35!
19. Consider the system shown in Figure 8.22.
The 2-port has the following S-parameters (Z
o
= 50 ):
286 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
50
Lossless [S]
Z
o
= 50
P
avs
= 0dBm
Z
Amplier
Figure 8.22: System with adjustable, lossless matching network between the 2-port and the
load.
S
11
= 0.2, S
21
= 5.0
S
12
= 0.1, S
22
= 0.5
(a) Find the impedance Z that will maximize the power delivered to the load, Z
L
.
(b) Find the power that will be delivered to the load, assuming that the matching
network has been designed to present the Z found in part 19a to the output of
the 2-port. Express your result in dBm.
(c) Find the impedance Z that will maximize the power delivered to the input of the
2-port.
(d) Find the power that will be delivered to the load, assuming that the matching
network has been designed to present the Z found in part 19c to the output of
the 2-port. Express your result in dBm.
20. In this problem, we prove that a particular set of passive terminations that cause the
negative resistance criterion for oscillation to be satised at one port will automatically
cause the conditions for oscillation to be satised at the other port if the 2 port is
bilateral (i.e. if S
12
6= 0 and S
21
6= 0).
(a) Show that the negative resistance criterion for steady-state oscillation (Z
1
+Z
2
=
0) can be written in terms of the corresponding reection coecients as
1
2
= 1.
(b) Now, suppose that
L
plane stability circle analysis shows that at least part of
unstable region lies within the unit circle in the
L
plane. We choose a load
reection coecient within the unstable region and also inside of the unit circle.
Denote this load termination by
Lu
. Find the reection coecient of the passive
source termination,
Su
, that will cause the conditions for steady state oscillation
to be satised at the input port. Write your answer in terms of the S parameters
and
Lu
.
(c) Now, show that the conditions for oscillation are satised at the output port
when the 2-port is terminated with
Lu
and
Su
.
21. A 2-port has S parameters ( Z
o
= 50 ): S
11
= 0.2, S
12
= 0.0, S
21
= 2.0, S
22
= 0.8
(a) Suppose that the 2-port is used in a system with source impedance Z
S
= 50
and load impedance Z
L
= 50. The power available from the source P
avs
= 0
dBm. Find the power that will be delivered to the load. Express your answer in
dBm.
8.6. HOMEWORK PROBLEMS 287
(b) Calculate the input impedance of the 2-port when it is used in the system de-
scribed in part a.
(c) This 2 port is unconditionally stable. Find the power delivered to the load when
lossless matching networks are used at the input and the output of the 2 port so
that the 2 port is simultaneously matched at both ports. Express your answer in
dBm.
(d) Find the power delivered to the load if 2 of these 2 ports are cascaded and the
cascade is used in between the source and load specied in part a. Express your
answer in dBm.
22. A 2 port has the following Z parameters (all given in ):
Z
11
= 20 Z
22
= 300 Z
12
= 0.0 Z
21
= 1000
Find S
11
and S
21
(Z
o
= 50 ). Hint: the input impedance and voltage gain of a 2
port can be written in terms of Z parameters and the load impedance as follows:
Z
IN
= Z
11
Z
12
Z
21
Z
L
+Z
22
A
v
=
V
2
V
1
=
Z
21
Z
L
Z
11
Z
L
+Z
11
Z
22
Z
12
Z
21
.
23. In a particular system it is found that the operating, transducer, and available gains
of a 2-port are G = 10 dB, G
T
= 8 dB, G
A
= 14 dB. The 2-port is unilateral and is
unconditionally stable. The source impedance Z
S
= 100 and the power available
from the source is 3 dBm.
(a) Find the power delivered to the load. Express your result in Watts (NOT in
dBm! I want to see that you know the relationship between power in Watts and
dBm.)
(b) Find the power delivered to the 2-port by the source. Express your result in
dBm.
(c) Find the power that would be delivered to the load if a single lossless matching
network is used between the source and the 2-port. Express your result in dBm.
(d) Find the power that would be delivered to the load if a single lossless matching
network is used between the 2-port and the load. Express your result in dBm.
(e) Find the power that would be delivered to the load if lossless matching networks
are used at both the input and the output of the 2-port. Express your result in
dBm.
(f) Suppose it is known that the 2-port has S
22
= 0 (Z
o
= 100 ) and calculate
the power that would be delivered to the load if a cascade of 2 of these identical
2-ports is used between the source and the load. Express your result in dBm.
24. In a system with P
avs
= 0 dBm, Z
S
= 100 and unknown, but constant, Z
L
, it
is empirically determined that the operating, transducer, and available gains of a
particular 2-port are G = 12 dB, G
T
= 8 dB, G
A
= 14 dB when this single 2-
port is used to couple the source to the load. The 2-port is known to be unilateral,
unconditionally stable, and to have S
22
= 0 (Z
o
= 100 ).
288 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
(a) Calculate the power that would be delivered to the load if a cascade of 2 of these
identical 2-ports is used between the source and load described above. Express
your result in dBm.
(b) Suppose that a lossless matching network is added between the load and the
ouput of the second 2-port in the system described in part a. Determine the
power delivered to the load. Express your result in dBm.
25. Many ampliers are designed to be simultaneously conjugate-matched at both ports
when used between 50 source and load impedances. Suppose that you have obtained
such an amplier, and that the specications for the amplier state that the power
gain of the unit is 20 dB in a 50 system. Furthermore, the specications state that
the reverse isolation of the amplier is 30 dB in a 50 system. Reverse isolation is
the power attenuation of the amplier when the amplier is driven at the output port
and the load is connected to the input port. (Reverse isolation of 30 dB means that
the reverse power gain of the amplier is -30 dB.)
(a) What are S
11
and S
22
(Z
o
= 50 ) for the amplier?
(b) Suppose that the amplier is used between a 50 source and a 1000 Ohm load.
What is the transducer gain of the amplier in this system? Express your result
in dB.
(c) What is the largest input reection coecient magnitude, |
IN
|, that could ever
be seen at the input of the amplier, assuming that the amplier is terminated
with a passive load?
(d) Suppose that the amplier is used between a 200 source and a 200 load.
You do not have enough information to determine the exact transducer gain in
this system, however you do have enough information to determine upper and
lower limits for what the transducer gain could be. Specify the range of possible
transducer gains (in dB).
26. Consider a 2-port consisting of a passive, lossless ladder network that matches a 50
source to the load Z
L
= 10 j200 . Answer the following questions. Note carefully
that you can answer all parts of this question without actually designing the matching
network.
(a) Find |S
21
| (Z
o
= 50 ) for the 2-port.
(b) Find |S
11
| (Z
o
= 50 ) for the 2-port.
(c) Find S
22
(Z
o
= 50 ) for the 2-port. Find the magnitude and phase angle.
27. A unilateral amplier is simultaneously conjugate-matched at both ports and has a
transducer gain of 12 dB when used in a system with Z
S
= Z
L
= 300 .
(a) What is the transducer gain (in dB) when this amplier is used in a system with
Z
S
= Z
L
= 50 ?
(b) What is the available gain (in dB) when this amplier is used in a system with
Z
S
= 50 ?
28. Consider the unilateral hybrid-pi model shown in Figure 8.23.
8.6. HOMEWORK PROBLEMS 289
Z
2
Z
1 V
g
m
V
+
-
Figure 8.23:
(a) What constraints must be satised by Z
1
and Z
2
if the 2-port is to be uncondi-
tionally stable?
(b) Find an expression for the operating power gain of the unilateral hybrid-pi model.
Express your result in terms of g
m
, Z
1
, Z
2
, and the load impedance Z
L
. It is not
necessary to nd S-parameters (or any other 2-port parameter set) to work this
problem. Start with the denition of operating power gain.
(c) Find an expression for the maximum available gain G
A,max
for the unilateral
hybrid-pi model. Express your result in terms of g
m
, Z
1
, and Z
2
.
29. Many ampliers are designed to be simultaneously conjugate-matched at both ports
when used between 50 source and load impedances. Suppose that you have obtained
such an amplier, and that the specications for the amplier state that the power
gain of the unit is 20 dB in a 50 system. Furthermore, the specications state that
the reverse isolation of the amplier is 23 dB in a 50 system. Reverse isolation is
the power attenuation of the amplier when the amplier is driven at the output port
and the load is connected to the input port. (Reverse isolation of 23 dB means that
the reverse power gain of the amplier is -23 dB.)
(a) What are S
11
and S
22
(Z
o
= 50 ) for the amplier?
(b) Is this amplier unconditionally stable? Justify your answer.
(c) Suppose that the amplier is used between a 400 source and a 50 load. The
power available from the source is P
avs
= 10 dBm. Find the power delivered
to the load. Express your result in dBm.
(d) Suppose that the system of part b. is modied by adding a lossless impedance
matching network between the output of the 2-port and the 50 load. Find the
power delivered to the load. Express your result in dBm.
(e) Suppose that two of these ampliers are cascaded, and used between a 400
source with P
avs
= 10 dBm and a 50 load. Find the power delivered to the
load. Express your result in dBm.
290 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
Chapter 9
Filter Design
This chapter discusses the implementation of lter networks with passive, lossless compo-
nents. In particular, we will concentrate on the design of lters based on ladder networks
comprised of lossless inductors and capacitors. A discussion of general properties of lossless
lters in terms of the scattering parameters of the lter network is followed by a descrip-
tion of some useful functions which approximate the ideal rectangular lowpass response and
are often used as target lter response functions when designing practical lters. Then we
will look at how to design lowpass lters with a prescribed frequency dependence for the
transducer power gain function, G
T
(!). Finally, we will discuss two methods for trans-
forming an existing lowpass lter design into a bandpass lter. The rst method is based
on a straightforward replacement of the series inductors and shunt capacitors in a lowpass
lter with series and parallel resonators, respectively. The second approach involves replac-
ing all of the lowpass lter elements with resonators of the same type which results in a
coupled-resonator lter.
We assume that the lter network is passive, linear, and lossless, is driven with a source
having real impedance Z
o
, and is terminated with real load impedance Z
o
, as shown in
Figure 9.1. Since
S
=
L
= 0 in this system, the transducer, operating, and available
Figure 9.1: A passive, linear, lossless 2 port in a system with source and load impedances
equal to Z
o
.
power gains can then be written as follows:
G
T
=
P
out
P
avs
= |S
21
|
2
(9.1)
291
292 CHAPTER 9. FILTER DESIGN
G =
P
out
P
in
= 1 =
|S
21
|
2
1 |S
11
|
2
(9.2)
G
A
=
P
avo
P
avs
= 1 =
|S
21
|
2
1 |S
22
|
2
(9.3)
In equations 9.2 and 9.3 we have used the fact that the lter network is lossless, which
means that the operating and available power gains will be equal to one. From equations
9.2 and 9.3:
|S
21
|
2
= 1 |S
11
|
2
= 1 |S
22
|
2
. (9.4)
Figure 9.2: The transducer power gain, |S
21
|
2
, and input/output reection coecients for
a second order Butterworth lter implemented using a passive lossless network.
Equation 9.4 shows that the transducer gain frequency response function is related to the
squared magnitude of the input and output reection coecients of the lossless two-port.
Figure 9.2 shows the transducer gain and and squared magnitude of the input coecient
reection for a lowpass lter. Notice that the lters attenuation at high frequencies happens
because the input reection coecient approaches 1, i.e., the attenuation at high frequencies
occurs because the source is unable to deliver much power to the network and not because of
any signal absorption within the network itself. The relationship between transducer power
gain and the squared magnitude of the input reection coecient means that the design of
lossless lters can be accomplished by designing a network to realize the target transducer
power gain (|S
21
|
2
) function or by designing the network to realize the corresponding input
or output power reection coecient function (|S
11
|
2
or |S
22
|
2
).
Recall that S
11
is related to the input impedance of the network through
S
11
=
Z
in
Z
o
Z
in
+Z
o
|
Z
L
=Zo
. (9.5)
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 293
Notice that a lter with prescribed transfer gain function specied by |S
21
(j!)|
2
can be
realized by synthesizing the appropriate transducer voltage gain function, S
21
(j!), or equiv-
alently, by synthesizing an input impedance function, Z
in
(j!), that produces |S
11
(j!)|
2
=
1 |S
21
(j!)|
2
. Both approaches will be illustrated through an example, but rst we shall
examine some common functions that are used for the target transducer power gain function.
9.1 Butterworth, Chebyshev, Bessel-Thompson Filters
9.1.1 Butterworth
The Butterworth response function with cuto frequency !
c
has the following form:
|S
21
(j!)|
2
=
1
1 + (
!
!c
)
2n
(9.6)
The -3 dB frequency occurs at ! = !
c
and the shape of the function is dened by the
parameter, n, which is always an integer. The parameter n is called the order of the lter
because, as we shall see, it is the highest power in the denominator polynomial of the
transducer voltage transfer function, S
21
(j!). The Butterworth transducer gain functions
for !
c
= 1 and orders n = 1 through n = 6 are plotted in Figure 9.3. The tranducer gain
falls o more rapidly for larger lter order n.
The Butterworth lter response approaches the ideal rectangular lowpass lter response
function as the order, n, increases. The Butterworth response is called maximally at
because all derivatives up through order 2n 1 are equal to zero at ! = 0 and as ! ! 1.
The Butterworth approximation to the ideal rectangular response function is used when
atness of the transfer function magnitude within the passband is the highest priority for a
particular application.
9.1.2 Chebyshev
The Chebyshev response function has the following form:
|S
21
(j!)|
2
=
1
1 +
2
C
2
n
(
!
!c
)
(9.7)
where is a constant called the ripple parameter, and C
n
(!) is a Chebyshev polynomial
dened by:
C
n
(!) = 2
n1
[!
n
n
1!2
2
!
n2
+
n(n3)
2!2
4
!
n4
n(n4)(n5)
3!2
4
!
n6
+
n(n5)(n6)(n7)
4!2
8
!
n8
n(n6)(n7)(n8)(n9)
5!2
10
!
n10
...]
(9.8)
The summation is stopped when the exponents of ! are no longer positive. The Cheby-
shev polynomials for n=1 through 5 are given in Table 9.1.
Two of the Chebyshev polynomials (n=4 and n=5) are plotted in Figure 9.4.
The Chebyshev response function is plotted for !
c
= 1,
2
= 0.0233 and n=1 through
n=4 in Figure 9.5, for
2
= 0.259 and n=1 through n=4 in Figure 9.6, and for
2
= 0.585
and n=1 through n=4 in Figure 9.7.
As the gures show, the ripple parameter, , controls the amount of amplitude ripple
within the lters passband. In general, for a particular lter order, n, a larger value for
294 CHAPTER 9. FILTER DESIGN
6.0
4.0
2.0
0.0
2.0
0.1 1.0 10.0
G
T
(dB)
!/!
c
Butterworth
50.0
45.0
40.0
35.0
30.0
25.0
20.0
15.0
10.0
5.0
0.0
5.0
10.0
0.1 1.0 10.0 100.0
G
T
(dB)
!/!
c
Butterworth
Figure 9.3: Butterworth Response function for n=1, 2, 3, 4. 5, and 6. The parameters are
the same for both gures, but the lower gure shows a wider range of frequencies and gains.
The higher lter orders correspond to faster descent into the stopband.
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 295
n C
n
(x)
1 x
2 2x
2
1
3 4x
3
3x
4 8x
4
8x
2
+ 1
5 16x
5
20x
3
+ 5x
6 32x
6
48x
4
+ 18x
2
1
7 64x
7
112x
5
+ 56x
3
7x
Table 9.1: Chebyshev polynomials for n=1 through 5.
0.0
1.0
2.0
3.0
4.0
0.0 0.5 1.0 1.5
C
2
n
(x)
x
Figure 9.4: C
2
n
(x) for n = 4 and n = 5. Within the passband (x<1) the functions oscillate
between 0 and 1. The number of extrema within the passband is equal to the lter order n.
Notice that the even order polynomial (n = 4) is equal to one at x = 0. This corresponds
to nite attenuation at ! = 0 and will not be realizable using a lossless network with equal
source and load resistances.
296 CHAPTER 9. FILTER DESIGN
6.0
4.0
2.0
0.0
2.0
0.1 1.0 10.0
G
T
(dB)
!/!
c
Chebyshev 0.1 dB ripple
50.0
40.0
30.0
20.0
10.0
0.0
10.0
0.1 1.0 10.0 100.0
G
T
(dB)
!/!
c
Chebyshev 0.1 dB ripple
Figure 9.5: Chebyshev response;
2
= 0.0233 (0.1 dB ripple), n=1, 2, 3 and 4. The lower
gure shows a wider range of frequencies and gains.
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 297
6.0
4.0
2.0
0.0
2.0
0.1 1.0 10.0
G
T
(dB)
!/!
c
Chebyshev 1.0 dB ripple
50.0
40.0
30.0
20.0
10.0
0.0
10.0
0.1 1.0 10.0 100.0
G
T
(dB)
!/!
c
Chebyshev 1.0 dB ripple
Figure 9.6: Chebyshev response;
2
= 0.259 (1.0 dB ripple), n=1, 2, 3 and 4. The lower
gure shows a wider range of frequencies and gains.
298 CHAPTER 9. FILTER DESIGN
6.0
4.0
2.0
0.0
2.0
0.1 1.0 10.0
G
T
(dB)
!/!
c
Chebyshev 2.0 dB ripple
50.0
40.0
30.0
20.0
10.0
0.0
10.0
0.1 1.0 10.0 100.0
G
T
(dB)
!/!
c
Chebyshev 2.0 dB ripple
Figure 9.7: Chebyshev response;
2
= 0.585 (2.0 dB ripple), n=1, 2, 3 and 4. The lower
gure shows a wider range of frequencies and gains.
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 299
the ripple parameter results in a quicker roll-o into the stopband at the expense of larger
ripple within the passband. The Chebyshev polynomials C
n
(!) oscillate between -1 and
1 for |!| < 1, so the transducer power gain oscillates between 1 and
1
1+
2
. The ratio of
the maximum and minimum responses within the passband is therefore 1 +
2
. The ripple
amplitude expressed in dB is 10 log(1 +
2
). In practice, the ripple parameter is usually
specied by giving the ripple amplitude in dB. Thus, a Chebyshev lter with 0.5 dB ripple
corresponds to
2
= 10
0.5/10
1, or = .349, approximately.
The cuto frequency dened in equation 9.7 is not the -3 dB frequency. Instead, it is
the frequency where the response function crosses the level corresponding to the bottom of
the passband ripple on its descent into the stopband. This can be seen clearly in Figures
9.5 through 9.7.
Figure 9.8: Comparison of third order Butterworth and Chebyshev lters with 0.1 dB and
2.0 dB ripple. The 2.0 dB ripple Chebyshev lter exhibits the fastest descent into the stop
band. The 0.1 dB ripple Chebyshev lter has a slower descent into the stop band, but it is
still noticeably faster than that of the Butterworth lter.
As the ripple parameter is decreased so that the passband ripple approaches 0 dB,
the Chebyshev response approaches a maximally-at response characteristic and becomes
identical to the Butterworth response, provided that the cuto frequency is suitably scaled.
Figure 9.8 shows the n=3 response functions for Butterworth and for Chebyshev with 0.1 dB
and 2.0 dB ripple, respectively. In this plot, the frequency axis has been scaled so that the -3
dB frequency of all three response functions occurs at ! = 1.0. The stopband attenuation is
smallest for the Butterworth response and is largest for the Chebyshev response with 2.0 dB
ripple. This gure illustrates that, for a given lter order n, larger stopband attenuation can
be achieved if larger ripple can be tolerated within the passband. Notice also that the 0.1 dB
ripple Chebyshev lter exhibits a noticeable improvement in stopband attenuation compared
to the Butterworth response, so a signicant improvement in stopband attenuation can be
300 CHAPTER 9. FILTER DESIGN
gained by tolerating a relatively small amount of passband ripple.
Finally, notice that the even order Chebyshev tranducer gain functions are not equal to
1 at ! = 0. Instead, the even order transducer gain function is equal to [1 +
2
]
1
at ! = 0,
i.e. the lter has nite attenuation at ! = 0. It is not possible to realize this type of transfer
function with a lossless lowpass ladder network and equal source and load resistances. Since
the lowpass network lter network reduces to a direct connection between the source and
load terminations at ! = 0, in order to have nite attenuation at ! = 0 it is necessary to
have dierent source and load terminations, such that the mismatch loss between the source
and load terminations is equal to [1+
2
]
1
. If the source and load impedances are the same,
as assumed here, it is only possible to realize the odd order Chebyshev response functions
with a lowpass ladder network.
9.1.3 Bessel-Thompson
The Bessel-Thompson response function is desirable in some applications because it results
in a group delay function that is maximally at in the same sense that the Butterworth
response provides maximally at amplitude response. This results in nearly distortionless
transmission of pulse-type waveforms. Bessel lters are commonly employed in digital com-
munications and radar systems where it is necessary to employ a lter that will not smear
pulses out over long times.
The family of Bessel transfer functions can be indexed by an integer parameter, n, and
for each n the response function takes the form:
S
21
(s) =
B
n
(0)
B
n
(s)
(9.9)
where B
n
(x) are the Bessel polynomials, which satisfy
B
n
(x) = (2n 1)B
n1
(x) +x
2
B
n2
(x)
with
B
0
(x) = 1, B
1
(x) = x + 1.
Bessel polynomials for n=1 through n=5 are tabulated in Table 9.2.
n B
n
(x)
1 x + 1
2 x
2
+ 3x + 3
3 x
3
+ 6x
2
+ 15x + 15
4 x
4
+ 10x
3
+ 45x
2
+ 105x + 105
5 x
5
+ 15x
4
+ 105x
3
+ 420x
2
+ 945x + 945
6 x
6
+ 21x
5
+ 210x
4
+ 1260x
3
+ 4725x
2
+ 10395x + 10395
7 x
7
+ 28x
6
+ 378x
5
+ 3150x
4
+ 17325x
3
+ 62370x
2
+ 62370x
2
+ 135135x + 135135
Table 9.2: Bessel polynomials for orders n=1 through n=4.
The Bessel response for order n can be written in the form:
S
21
(s) =
1
1 +a
1
s +a
2
s
2
+a
3
s
3
+ a
n
s
n
(9.10)
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 301
The group delay of a lter is dened as the negative slope of the lters phase response.
Thus, writing the transfer function S
21
(j!) = A(!)e
j(!)
, the group delay, T
g
(!), is dened
as:
T
g
(!) =
d(!)
d!
(9.11)
For a transfer function of the form given in equation 9.10, it can be shown that the group
delay at ! = 0 is equal to the value of the coecient a
1
. If the Bessel polynomials given in
Table 9.2 are used in equation 9.9 and the resulting equation is manipulated so that it is in
the form of equation 9.10, the coecient a
1
will be equal to one, corresponding to a group
delay of 1 second. To scale a Bessel lter for a specied delay, it is necessary to scale the
coecients a
1
through a
n
by multiplying the i
0
th coecient by
i
, where is the desired
delay. Thus, a third-order Bessel lter with delay of 1 s would have coecients:
a
1
= 10
6
a
2
= (10
6
)
2
6
15
a
3
= (10
6
)
3
1
15
The Bessel response for n=1 through n=4 is plotted in Figure 9.9. The coecients of
the Bessel polynomial have been scaled dierently for each plot in order to make the -3 dB
frequencies equal to 1 Hz in all cases. The Bessel response exhibits a gradual descent into
the stopband and, for a given lter order and -3 dB frequency, results in less attenuation in
the stopband than the Bessel or Chebyshev lters.
The group delay for the lters with n=1 through n=4 is shown in Figure 9.10. Notice
that the shape of the Bessel lters group delay curve is the same as the Butterworth lters
maximally at gain response. The same (scaled) coecients that were used to produce
Figure 9.9 were used to produce Figure 9.10. Notice that when all of the lters are scaled
to have the same -3 dB frequency, as is the case here, the delay increases as the lter order
is increased.
9.2 Example: Synthesis of 4th order Butterworth lter
Suppose that it is necessary to design a passive LC ladder network that realizes the fourth
order Butterworth lowpass response function, i.e., the network transducer gain function
must be
|S
21
(!)|
2
=
1
1 + !
8
(9.12)
Notice that the cuto frequency of the lter has been set to 1 rad/s. Later we will discuss
how to scale the design to an arbitrary cuto frequency.
So far, only the magnitude of the transfer function has been specied. To design an actual
circuit that realizes the desired transfer function, we must determine the complex voltage
transfer function, S
21
(j!). Alternatively, we could determine the required complex input
reection coecient, S
11
(j!), from which we can solve for the complex input impedance
function that the network must realize.
302 CHAPTER 9. FILTER DESIGN
Figure 9.9: Bessel response for n=1 through n=4. The higher lter orders correspond to
faster descent into the stopband.The coecients of the Bessel polynomial given in Table 9.2
have been scaled for each curve so that the -3 dB frequency occurs at 1 Hz.
Figure 9.10: Group delay for the Bessel lters with n=1 through n=4.
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 303
9.2.1 Filter synthesis based on the S
21
function.
We must rst determine the form of a realizable S
21
function that will provide the desired
transducer gain response. The may be accomplished by noting that equation 9.12 can be
re-written as follows:
S
21
(j!)S
21
(j!) = S
21
(j!)S
21
(j!) =
1
1 + !
8
or, with ! =
s
j
:
S
21
(s)S
21
(s) =
1
1 + (
s
j
)
8
=
1
1 +s
8
(9.13)
The right hand side of equation 9.13 has 8 poles. To determine S
21
(s), the right hand side
of equation 9.13 must be factored into a product of two terms such that the second term
can be obtained by replacing s with s in the rst term. This amounts to deciding which
group of 4 poles must be assigned to S
21
(s). It will be necessary to ensure that the poles
assigned to S
21
(s) are in the left half of the s-plane. The poles must be in the left-half plane
because it is not possible to realize an S
21
function that has poles in the right-half plane
using a passive network. The pole locations are given by the roots of the following equation:
s
8
= 1 (9.14)
s
8
= e
j(2n+1)
(9.15)
The solutions to equation 9.14 (or, equivalently, equation 9.15) are s
k
= e
j(2n+1)
8
with
n = 0, 1, 2, 7. Thus, the poles are equally spaced around the unit circle centered on the
origin of the s-plane. The 4 poles that must be assigned to S
21
(s) are those in the left-half
plane. The pole locations can be written as follows:
s
k
= e
j(2n+1)
8
= cos(
(2n + 1)
8
) +j sin(
(2n + 1)
8
)
with the left-half plane pole locations obtained when n = 2, 3, 4, 5. S
21
(s) can now be written
as follows:
S
21
(s) =
1
(s s
2
)(s s
3
)(s s
4
)(s s
5
)
(9.16)
S
21
(s) =
1
(s e
j
5
8
)(s e
j
7
8
)(s e
j
9
8
)(s e
j
11
8
)
The denominator can be expanded to yield:
S
21
(s) =
1
s
4
+ 2.61313s
3
+ 3.41421s
2
+ 2.61313s + 1
. (9.17)
The ladder networks that can realize the transfer function with 4 poles will have 4
branches with one energy storage element per branch. The two possibilities are the lowpass
304 CHAPTER 9. FILTER DESIGN
C
2
L
1
C
4
L
3
V
s
V
o
Z
o
Z
o
Figure 9.11:
C
1
L
2
C
3
L
4
V
s
V
o
Z
o
Z
o
Figure 9.12:
networks shown in Figures 9.11 and 9.12. Note that in the limit as ! approaches 0, these
networks will have the property that S
21
!1. Thus, only the upper sign in the numerator
of equations 9.16 and 9.17 is relevant.
Consider Figure 9.11. If the termination impedances are 1 , the transducer voltage
gain for this network is easily calculated. (Assume that the voltage across the load is 1 V
and work backwords toward the source to nd V
S
. Then, S
21
= 2/V
S
.) The result is:
S
21
=
1
s
4
(
L1C2L3C4
2
) +s
3
(
L1C2L3+C2L3C4
2
) +s
2
(
L3C4+L1C4+L1C2+C2L3
2
) +s(
L3+L1+C2+C4
2
) + 1
(9.18)
Comparing the coecients in the denominator of 9.17 with those in the denominator of 9.18
yields four equations for the four unknown parameters:
L
1
C
2
L
3
C
4
2
= 1 (9.19)
L
1
C
2
L
3
+C
2
L
3
C
4
2
= 2.61313 (9.20)
L
3
C
4
+L
1
C
4
+L
1
C
2
+C
2
L
3
2
= 3.41421 (9.21)
L
1
+L
3
+C
2
+C
4
2
= 2.61313 (9.22)
Equations 9.19 through 9.22 can now be solved for the unknown parameters. The solution
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 305
is:
L
1
= 0.7654 H (9.23)
C
2
= 1.8478 F
L
3
= 1.8478 H
C
4
= 0.7654 F
These values will give a Butterworth response with cuto frequency !
c
= 1 rad/s if the
terminating impedances are 1 . This lter is referred to as a lowpass prototype lter
and can be used as the basis for a lter with arbitrary cuto frequency and termination
impedance.
9.2.2 Filter synthesis based on the input impedance function
From 9.4 we have
|S
11
(j!)|
2
= 1 |S
21
(j!)|
2
or, with s = j!:
S
11
(s)S
11
(s) = 1
1
1 + (
s
j
)
8
=
s
8
1 +s
8
.
As in the previous section we must factor the RHS to isolate a term having poles only in
the left-half plane, because an S
11
function that can be realized with passive network will
have poles only in the left-half plane. Proceeding exactly as before, we determine that
S
11
(s) =
s
4
(s s
2
)(s s
3
)(s s
4
)(s s
5
)
(9.24)
S
11
(s) =
s
4
(s e
j
5
8
)(s e
j
7
8
)(s e
j
9
8
)(s e
j
11
8
)
Expanding the denominator, as before:
S
11
(s) =
s
4
s
4
+ 2.61313s
3
+ 3.41421s
2
+ 2.61313s + 1
. (9.25)
The input impedance function, assuming 1 source and load impedance, is
z
in
(s) =
1 +S
11
(s)
1 S
11
(s)
Taking the upper (plus) sign in the numerator of 9.25
z
in
(s) =
2s
4
+ 2.61313s
3
+ 3.41421s
2
+ 2.61313s + 1
2.61313s
3
+ 3.41421s
2
+ 2.61313s + 1
.
Using long division, the input impedance can be written in continued fraction form
z
in
(s) = 0.7654s +
1
1.8478s +
1
1.8478s+
1
.7654s+1
. (9.26)
306 CHAPTER 9. FILTER DESIGN
Next, notice that the input impedance of the network shown in Figure 9.11 can be written
in continued fraction form as:
z
in
(s) = sL
1
+
1
sC
2
+
1
sL3+
1
sC
4
+1
. (9.27)
Equating the coecients in equations 9.27 and 9.26 yields the results already given in
equation 9.23.
If the lower (minus) sign is selected in the numerator of equation 9.25 the expression for
z
in
(s) will be the inverse of that given in equation 9.26. The input admittance for that case
would be equal to the right hand side of equation 9.26 and the continued fraction expansion
representation of the input admittance will correspond to the right hand side of equation
9.26. The ladder network shown in Figure 9.12 has an input admittance function of the
same form. Equating coecients as before yields the normalized component values for the
ladder network of the form shown in Figure 9.12:
C
1
= 0.7654 F (9.28)
L
2
= 1.8478 H
C
3
= 1.8478 F
L
4
= 0.7654 H
Notice that the same list of normalized component values applies to both of the networks
shown in Figures 9.11 and 9.12, but the values of the inductances in one case correspond to
the values of the capacitors in the other case, and vice versa. Therefore, when tabulating the
normalized component values for lters of a given type and order it is sucient to give one
list of component values. The elements in the list are interpreted as alternating inductance
and capacitance values, starting with the left side of the network, if the ladder network has
a series inductor on the left side of the network (as in Figure 9.11). When applied to a
ladder network with a shunt capacitor on the left side of the network (as in Figure 9.12) the
elements in the list are interpreted as alternating capacitance and inductance values.
The element values for lowpass prototype lters of any type and order can be derived
by applying the procedures illustrated in this example. The essential information that is
required to derive the component values is the locations of the left-half plane poles associated
with the Transducer power gain function. A table of component values for lowpass prototype
Butterworth lters is provided in section 9.2.3.
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 307
9
.
2
.
3
C
o
m
p
o
n
e
n
t
v
a
l
u
e
s
f
o
r
l
o
w
p
a
s
s
p
r
o
t
o
t
y
p
e
B
u
t
t
e
r
w
o
r
t
h
l
t
e
r
s
n
g
1
g
2
g
3
g
4
g
5
g
6
g
7
g
8
g
9
g
1
0
1
2
.
0
2
1
.
4
1
4
2
1
1
.
4
1
4
2
1
3
1
.
0
2
.
0
1
.
0
4
0
.
7
6
5
3
6
7
1
.
8
4
7
7
6
1
.
8
4
7
7
4
0
.
7
6
5
3
6
7
5
0
.
6
1
8
0
3
4
1
.
6
1
8
0
3
2
.
0
1
.
6
1
8
0
3
0
.
6
1
8
0
3
4
6
0
.
5
1
7
6
3
8
1
.
4
1
4
2
1
1
.
9
3
1
8
5
1
.
9
3
1
8
5
1
.
4
1
4
2
1
0
.
5
1
7
6
3
8
7
0
.
4
4
5
0
4
2
1
.
2
4
6
9
8
1
.
8
0
1
9
4
2
.
0
1
.
8
0
1
9
4
1
.
2
4
6
9
8
0
.
4
4
5
0
4
2
8
0
.
3
9
0
1
8
1
1
.
1
1
1
1
4
1
.
6
6
2
9
4
1
.
9
6
1
5
7
1
.
9
6
1
5
7
1
.
6
6
2
9
4
1
.
1
1
1
1
4
0
.
3
9
0
1
8
1
9
0
.
3
4
7
2
9
6
1
.
0
1
.
5
3
2
0
9
1
.
8
7
9
3
9
2
.
0
1
.
8
7
9
3
9
1
.
5
3
2
0
9
1
.
0
0
.
3
4
7
2
9
6
1
0
0
.
3
1
2
8
6
9
0
.
9
0
7
9
8
1
1
.
4
1
4
2
1
1
.
7
8
2
0
1
1
.
9
7
5
3
8
1
.
9
7
5
3
8
1
.
7
8
2
0
1
1
.
4
1
4
2
1
0
.
9
0
7
9
8
1
0
.
3
1
2
8
6
9
T
a
b
l
e
9
.
3
:
C
o
m
p
o
n
e
n
t
v
a
l
u
e
s
f
o
r
l
o
w
p
a
s
s
p
r
o
t
o
t
y
p
e
B
u
t
t
e
r
w
o
r
t
h
l
t
e
r
s
.
308 CHAPTER 9. FILTER DESIGN
9.3 Example - 3rd order/0.1 dB ripple Chebyshev low-
pass
The transducer power gain function for a Chebyshev lter has the form
|S
21
(j!)|
2
=
1
1 +
2
C
2
n
(!)
.
Then, with ! = s/j:
|S
21
(s)|
2
=
1
1 +
2
C
2
n
(
s
j
)
.
|S
11
(s)|
2
= 1 |S
21
(s)|
2
=
2
C
2
n
(
s
j
)
1 +
2
C
2
n
(
s
j
)
. (9.29)
Note that the -3 dB corner frequency of the Chebyshev response function does not occur
at ! = 1. Instead, ! = 1 corresponds to the frequency where the response has an attenuation
equal to the specied ripple. The response descends into the stopband at ! > 1. After the
lowpass prototype lter has been designed, the component values can be scaled to yield a
new prototype lter that has -3 dB frequency equal to ! = 1.
A third order Chebyshev lter with 0.1 dB ripple is obtained by using
2
= 10
(0.1/10)
1 =
0.023293 and C
2
3
(x) = (4x
3
3x)
2
in equation 9.29, which yields:
|S
11
(s)|
2
=
(s
3
+ 0.75s)
2
s
6
+ 1.5s
4
+ 0.5625s
2
2.68321
The left-half plane poles associated with this function are located at:
s
1
= 0.969406
s
2
= 0.484703 j1.20616
s
3
= 0.484703 +j1.20616
Thus, we have
S
11
(s) =
(s
3
+ 0.75s)
(s s
1
)(s s
2
)(s s
3
)
=
(s
3
+ 0.75s)
s
3
+ 1.93881s
2
+ 2.62949s + 1.63805
If the upper sign is chosen, then:
z
in
(s) =
1 +S
11
(s)
1 S
11
(s)
=
2s
3
+ 1.93881s
2
+ 3.37949s + 1.63805
+1.93881s
2
+ 1.87848s + 1.63805
.
Using long division:
z
in
(s) = 1.0316s +
1
1.1474s +
1
1.0316s+1
.
Thus, the element values for the lowpass prototype lter are:
g
1
= 1.0316 g
2
= 1.1474 g
3
= 1.0316.
9.3. EXAMPLE - 3RD ORDER/0.1 DB RIPPLE CHEBYSHEV LOWPASS 309
Since the network has 3 elements, it will be either a T network or a PI network. If the
T-network is employed, g
1
and g
3
represent the values of the series inductors (in Henries)
and g
2
represents the value of the shunt capacitor (in Farads). If the PI-network is employed
g
1
and g
3
are the shunt capacitor values (in Farads) and g
2
is the series inductor value (in
Henries).
It may be desirable to scale the Chebyshev prototype lters such that the -3 dB frequency
is located at ! = 1 s
1
. The third order prototype lter that has been designed so far has
attenuation equal to 0.1 dB at ! = 1 s
1
. The -3 dB frequency is located at ! = 1.3795 s
1
.
The lter can be scaled so that the -3 dB frequency is at ! = 1 by multiplying the element
values by 1.3795. The new prototype values are:
g
1
= 1.4994 g
2
= 1.6678 g
3
= 1.4994
310 CHAPTER 9. FILTER DESIGN
9
.
3
.
1
C
o
m
p
o
n
e
n
t
v
a
l
u
e
s
f
o
r
o
d
d
-
o
r
d
e
r
l
o
w
p
a
s
s
p
r
o
t
o
t
y
p
e
C
h
e
b
y
s
h
e
v
l
t
e
r
s
w
i
t
h
0
.
1
d
B
r
i
p
p
l
e
n
g
1
g
2
g
3
g
4
g
5
g
6
g
7
g
8
g
9
1
0
.
3
0
5
2
4
1
3
1
.
0
3
1
5
6
1
.
1
4
7
4
1
.
0
3
1
5
6
5
1
.
1
4
6
8
1
1
.
3
7
1
2
1
1
.
9
7
5
1
.
3
7
1
2
1
1
.
1
4
6
8
1
7
1
.
1
8
1
1
8
1
.
4
2
2
8
1
2
.
0
9
6
6
7
1
.
5
7
3
4
2
.
0
9
6
6
7
1
.
4
2
2
8
1
1
.
1
8
1
1
8
9
1
.
1
9
5
6
7
1
.
4
4
2
6
2
.
1
3
4
5
5
1
.
6
1
6
7
2
2
.
2
0
5
3
7
1
.
6
1
6
7
2
2
.
1
3
4
5
5
1
.
4
4
2
6
1
.
1
9
5
6
7
T
a
b
l
e
9
.
4
:
C
o
m
p
o
n
e
n
t
v
a
l
u
e
s
f
o
r
o
d
d
o
r
d
e
r
l
o
w
p
a
s
s
p
r
o
t
o
t
y
p
e
C
h
e
b
y
s
h
e
v
l
t
e
r
s
w
i
t
h
0
.
1
d
B
r
i
p
p
l
e
a
n
d
e
q
u
a
l
s
o
u
r
c
e
a
n
d
l
o
a
d
t
e
r
m
i
n
a
t
i
o
n
s
.
T
h
e
s
e
p
r
o
t
o
t
y
p
e
v
a
l
u
e
s
p
r
o
d
u
c
e
a
l
t
e
r
w
i
t
h
a
t
t
e
n
u
a
t
i
o
n
e
q
u
a
l
t
o
t
h
e
p
a
s
s
b
a
n
d
r
i
p
p
l
e
(
-
0
.
1
d
B
)
a
t
!
=
1
s
1
.
9.4. FREQUENCY AND IMPEDANCE SCALING 311
9.4 Frequency and Impedance scaling
So far we have synthesized lowpass prototype lters that have provide some desired response
function having a corner frequency of ! = 1 s
1
when terminated with source and load
resistance of 1 . To scale a lowpass prototype lter to a new frequency, it is necessary to
scale the component values so that they have the same reactance at the new cuto frequency
as the prototype values have at ! = !
c
= 1 rad/s. The new inductor and capacitor values
can therefore be obtained from the calculated values by dividing the calculated values by
the desired cuto frequency, in rad/s.
To scale the lter to a new impedance value, it is necessary to scale the component
impedances so they maintain the same magnitude relative to the terminating impedance.
For example, the fourth order Butterworth lowpass protoype has L
1
= 0.7654 H, so the
reactance of L
1
is equal to 0.7654 at the cuto frequency, or 0.7654 times the terminating
impedance. Thus, for a terminating impedance of 50 the inductor should have a reactance
of 0.7654(50) = 38.27 at !
c
. So inductor values have to be multiplied by the desired
terminating impedance, Z
o
. The capacitor C
2
has a reactance of 1/1.8478 = 0.5412 at the
cuto frequency. To scale for a terminating impedance of 50 the capacitor should have a
reactance of 0.5412(50) = 27.06 at the cuto frequency. Thus, the calculated capacitor
value should be divided by the new terminating impedance.
Both frequency and impedance scaling can be performed in one step as follows. Denote
the new scaled element values by primed quantities and the lowpass prototype values by
unprimed quantities. Then the new inductor and capacitor values are given in terms of the
desired termination impedance, Z
o
, and the desired cuto frequency, !
c
, by:
L
0
=
LZ
o
!
c
(9.30)
C
0
=
C
!
c
Z
o
. (9.31)
9.5 Bandpass Transformation
Bandpass lters can be realized by transforming a lowpass prototype lter. Well discuss
one simple transformation. The idea behind this transformation is to replace each branch of
the lowpass prototype with a new branch consisting of two elements. The series inductors
are replaced with series LC circuits, and the shunt capacitors are replaced with parallel
LC circuits. Recall that the series elements of the lowpass lter (inductors) look like short
circuits at the center of the lter response (at ! = 0), and the shunt elements of the lowpass
lter look like open circuits at the center of the lter response. We can transfer these
characteristics to any other center frequency, !
o
, by making sure that the series and parallel
branches of the new lter are resonant at !
o
. Likewise, at the edge of the original lters
passband, each branch had a certain impedance (relative to the terminating impedance). If
we choose the elements of the new branches such that the impedance level is the same at
the desired edges of the passband, then the shape of the bandpass lters transfer function
should look like a shifted version of the lowpass lters transfer function.
The lowpass-to-bandpass lter transformation can be dened in terms of the center
frequency and fractional bandwidth of the bandpass lter. Denoting the desired upper and
lower cuto frequencies of the bandpass lter by f
l
and f
u
and the center frequency by f
o
,
312 CHAPTER 9. FILTER DESIGN
the absolute bandwidth is:
BW = f
u
f
l
(9.32)
and the fractional bandwidth is:
bw =
BW
f
o
(9.33)
The element values for the parallel LC shunt elements in the bandpass lter are then
written in terms of the original normalized lowpass prototype lter shunt capacitor value,
C
lp
as:
C
bpshunt
=
C
lp
bw
(9.34)
L
bpshunt
=
1
C
bpshunt
(9.35)
The element values for the series LC series elements in the bandpass lter are given in
terms of the original normalized lowpass prototype series inductor value, L
lp
as:
L
bpseries
=
L
lp
bw
(9.36)
C
bpseries
=
1
L
bpseries
(9.37)
These normalized bandpass lter element values are scaled to the desired center frequency
and impedance value by using equations 9.30 and 9.31 with the cuto frequency, !
c
, replaced
with the desired center frequency, !
o
.
9.5.1 Example - Bandpass lter based on 4th order Butterworth
lowpass prototype
Suppose it is necessary to design a bandpass lter with center frequency of 70 MHz, 3 dB
bandwidth of 20 MHz. The lter will operate with source and load impedances of 50. The
fractional bandwidth is then:
bw =
20
70
= .2857 (9.38)
Using the 4th order lowpass Butterworth lter prototype derived in section 9.2, and
transforming that lter, the resulting bandpass lter will have the topology shown in Figure
9.13.
The element values for this lter can be obtained by scaling the the lowpass prototype
values for the 4th order Butterworth lter derived in section 9.2. The results are:
L
1
= 0.305 H C
1
= 17.0 pF
L
2
= 17.6 nH C
2
= 294 pF
L
3
= 0.735 H C
3
= 7.03 pF
L
4
= 42.4 nH C
4
= 126 pF
The transducer power gain of this lter is shown in Figure 9.14 on linear and logarithmic
scales. The lower plot shows that bandpass lters derived from lowpass prototypes using
9.6. COUPLED RESONATOR FILTERS. 313
Figure 9.13: Bandpass lter topology derived from the prototype lowpass lter shown in
Figure 9.11.
the transformation described in this section do not exhibit arithmetic symmetry, i.e., the
response falls o more slowly above the passband than it does below the passband. The
response would appear to be symmetric if plotted on a logarithmic frequency axis, as shown
in Figure 9.15.
In practice, the very large stopband attenuation shown in Figure 9.15 would not be
attained with a real lter. The analysis has ignored parasitic reactances associated with
real components (e.g., lead inductance and capacitance between turns in an inductor). In
addition, unmodeled coupling between elements and the input and output ports of the lter
will inevitably occur. These eects generally limit the attainable stopband attenuation to
numbers in the range 50-70 dB.
Finally, it should be noted that this method of transforming a lowpass prototype into
a bandpass lter will result in realizable component values only when the target fractional
bandwidth is larger than 10-15%. In general, as the target fractional bandwidth becomes
smaller, the ratio between the largest and smallest required component values will increase.
Notice that in this 20% bandwidth design, the ratio between the largest and smallest in-
ductor and between the largest and smallest capacitor is 42. As the target bandwidth is
decreased, the ratios increase until a point where it is no longer practical or possible to
realize the needed component values with reasonably small losses.
9.6 Coupled resonator lters.
The lowpass to bandpass transformation discussed in the previous section results in a net-
work containing multiple resonators, each of which has dierent L and C values. In addition,
it was pointed out that the required element values span a very wide range when the frac-
tional bandwidth of the lter is small. If a small fractional bandwidth (e.g. smaller than
10%) is needed it is generally better to use a lter topology that is based on identical cou-
pled resonators. For example, one such topology is shown in Figure 9.16 where 4 parallel
LC resonators are capacitively top-coupled to each other and to the source and load. This
lter topology has the same number of resonators as the one derived in the previous section,
however it provides the designer with the freedom to independently choose the properties
of the resonators. This extra freedom makes it possible to create a lter in which all of the
resonator inductors (or capacitors) are identical, greatly simplifying implementation of the
lter.
The normalized lowpass prototype element values, g
i
, provide the essential information
necessary to design a bandpass lter based on coupled resonators. For detailed information
314 CHAPTER 9. FILTER DESIGN
Figure 9.14: Transducer power gain (|S
21
|
2
) for the 70 MHz bandpass lter design based on
a fourth order Butterworth lowpass prototype. The upper and lower plots show the power
gain on linear, and logarithmic scales, respectively.
9.6. COUPLED RESONATOR FILTERS. 315
Figure 9.15: Same as Figure 9.14 except the frequency axis is plotted on a logarithmic scale
to illustrate the symmetry of the response with respect to the logarithm of frequency.
L
1
C
1
L
2
C
2
L
3
C
3
L
4
C
4
C
01
C
12
C
23
C
34
C
45
Z
o
Z
o
Figure 9.16: A coupled resonator lter based on capacitive coupling between parallel LC
resonators.
316 CHAPTER 9. FILTER DESIGN
on the design of coupled resonator lters see references [2,3,4]. Here we will simply de-
scribe the procedure for the design of a lter based on parallel LC resonators with common
resonator inductance and capacitive coupling. The design proceeds as follows -
1. Determine the required corner frequencies, !
1
and !
2
, for the prototype lter. The
bandwidth of the lter will be BW = !
2
!
1
and the center frequency will be
!
o
'
p
!
1
!
2
.
2. Choose a common resonator inductance, L
r
, to be used for all resonators. The ca-
pacitance required to resonate this inductance at the desired center frequency, !
o
, is
C
r
= (!
2
o
L
r
)
1
.
3. Calculate parameters J
i,i+1
as follows:
J
0,1
=
BWC
r
Z
o
g
1
J
i,i+1
|
i=1 to n1
= BWC
r
1
g
i
g
i+1
J
n,n+1
=
BWC
r
Z
o
g
n
4. Calculate the coupling capacitor values:
C
0,1
=
J
0,1
!
o
1 (Z
o
J
0,1
)
2
C
i,i+1
|
i=1 to n1
=
J
i,i+1
!
o
C
n,n+1
=
J
n,n+1
!
o
1 (Z
o
J
n,n+1
)
2
.
5. Calculate the shunt capacitance across each resonator inductor:
C
1
= C
r
C
0,1
1 + (!
o
Z
o
C
0,1
)
2
C
1,2
C
i
|
i=2 to n1
= C
r
C
i1,i
C
i,i+1
C
n
= C
r
C
n,n+1
1 + (!
o
Z
o
C
n,n+1
)
2
C
n1,n
9.6. COUPLED RESONATOR FILTERS. 317
9.6.1 Example - Coupled resonator lter with 2 parallel LC res-
onators based on Butterworth lowpass prototype
In this example we summarize the design of a bandpass lter based on a second order
(n = 2) Butterworth lowpass prototype. The lter is to have center frequency 50 MHz and
bandwidth 2.5 MHz (5% fractional bandwidth). It is to be used in a system with source
and load impedance Z
o
= 50 .
To obtain a passband response that is approximately symmetric around 50 MHz, we
choose the upper and lower corner frequencies as follows:
!
1
= 2(50 2.5/2) 10
6
= 3.0631 10
8
s
1
!
2
= 2(50 + 2.5/2) 10
6
= 3.2201 10
8
s
1
.
Then the target center frequency will be the geometric mean of the two corner frequencies:
!
o
=
p
!
1
!
2
= 3.1406 10
8
s
1
The target bandwidth is
BW = 2 2.5 10
6
= 1.5708 10
7
s
1
.
The lter will contain two parallel resonators as shown in Figure 9.17. We shall choose
L
1
C
1
L
2
C
2
C
01
C
12
C
23
Z
o
Z
o
Figure 9.17: Filter topology for the coupled resonator lter design example. The design
procedure that we have described assumes that the resonator inductances are equal (L
1
=
L
2
= L
r
).
a resonator inductance L
r
= L
1
= L
2
= 0.2 H. In practice, the inductance value is
typically chosen by selecting the value that results in the highest possible inductor Q
L
(and hence highest possible resonator Q) at the intended center frequency, !
o
. Since we
ignored component losses in the design, a real lters performance will approximate the ideal
predicted performance only when the component Q of each lter component is signicantly
larger than the lters Q = f
o
/BW. The freedom to independently choose the inductance
value to optimize resonator Q makes the coupled resonator lter very attractive for lters
with narrow bandwidth (high Q).
The total capacitance required to resonate the chosen resonator inductance is C
r
=
1/(!
2
o
L
r
) = 50.69 pF. The lowpass prototype component values for a second order Butter-
worth lter (n = 2) are g
1
= g
2
=
p
2.
The J
i,i+1
can now be calculated. The values are:
J
01
= 0.0033557 J
12
= 0.0005631 J
23
= 0.0033557.
318 CHAPTER 9. FILTER DESIGN
The coupling capacitances are:
C
01
= 10.84 pF C
12
= 1.793 pF C
23
= 10.84 pF.
The resonator capacitances are:
C
1
= C
2
= 38.06 pF.
The transducer power gain of the resulting lter is plotted in Figure 9.18 on both linear
and logarithmic scales. The asymmetry that is evident in the bottom plot is a result of the
capacitive top-coupling. At high frequencies the network reduces to an all-capacitor network
with roughly constant insertion loss.
9.7 References
1. Carson, Ralph S., Radio Concepts: Analog, John-Wiley and Sons, New York, 1990.
2. Zverev, Anatol I., Handbook of Filter Synthesis, John-Wiley and Sons, New York,
1967.
3. Rhea, Randall W., HF Filter Design and Computer Simulation, Noble Publishing,
Atlanta, 1994.
4. Matthaei, George L., Leo Young, E. M. T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Massachusetts, 1980.
9.7. REFERENCES 319
0.0
0.2
0.5
0.8
1.0
45.0 46.0 47.0 48.0 49.0 50.0 51.0 52.0 53.0 54.0 55.0
G
T
Frequency (MHz)
50.0
45.0
40.0
35.0
30.0
25.0
20.0
15.0
10.0
5.0
0.0
10.0 100.0
G
T
(dB)
Frequency (MHz)
Figure 9.18: Transducer power gain (|S
21
|
2
) for the 50 MHz bandpass lter based on a
second order Butterworth lowpass prototype and implemented with top-coupled parallel LC
resonators. The upper and lower plots show the power gain on linear and logarithmic scales,
respectively.
320 CHAPTER 9. FILTER DESIGN
9.8 Homework Problems
1. A gure of merit that is used to characterize lters is the shape factor, dened to
be the ratio of the transducer power gain bandwidth at the -60 dB and -6 dB points,
respectively. The shape factor will be larger than 1, and will approach 1 as the lter
response function approaches an ideal rectangular response. Find an expression for
the shape factor for Butterworth bandpass lters as a function of the lter order, n.
Give numerical values for the shape factor for n=3 and n=6.
2. Find and sketch the left-half plane pole locations for a fth-order Butterworth lter.
3. Design a second-order Butterworth lowpass lter with -3 dB frequency, 50 MHz and
terminating impedances Z
0
= 50 . Draw the lter and give component values. Use
ADS to plot the transducer power gain (|S
21
|
2
). Plot the transducer power gain in
dB against frequency (Use a logarithmic axis for frequency.) from at least 0.5 MHz to
500 MHz.
4. Transform the lter designed in 9-3 to a bandpass lter with center frequency 50 MHz
and 3 dB bandwidth 10 MHz. Draw the lter and give component values. Use ADS to
plot the transducer power gain. Plot the transducer power gain in dB against frequency
on a logarithmic frequency axis from 0.5 MHz to 500 MHz. Make another plot covering
a smaller frequency range around 50 MHz to show the detailed characteristics of the
transfer function within the passband.
5. A Butterworth lowpass lter using the smallest possible number of components (i.e.,
smallest possible lter order n) is to be designed to lter the output of a source that
produces output at a fundamental frequency f
o
= 15 MHz and at harmonics of that
frequency mf
0
(m = 2,3...). The attenuation at f
o
should be no more than 0.5 dB and
the attenuation at 2f
o
should be at least 30 dB. Specify the -3 dB cuto frequency of
the lter f
c
(in MHz) and the lter order n (an integer) that will achieve these design
goals. If your parameters are used, what would the attenuation be at the second
harmonic frequency 2f
o
?
Chapter 10
Noise in 1- and 2-ports
10.1 Noise Characterization of 1-ports
There are various physical phenomena that can produce noise in electronic circuits, in-
cluding:
Thermal noise - a result of random motion of charge carriers in a conductor. This type
of noise will be present in any dissipative electronic circuit element (e.g. resistors or
the dissipative parts of transistors and diodes), even in the absence of any externally
applied bias voltage or current. Thermal noise has a white spectrum, which means
that the power available in a frequency bandwidth df is independent of the center
frequency of the band.
Shot noise - is associated with current ow in semiconductor devices and vacuum
tubes. It will be present in these active devices when an average (DC) current ows.
In such devices, current ow can be modeled as a series of independent current pulses
occurring at random and a DC bias current will be associated with a time-varying
shot-noise current whose intensity spectrum is proportional to the DC current. The
shot noise spectrum is white at suciently low frequencies.
Flicker noise or 1/f noise any process with a noise power spectral density that
varies as 1/f for low frequencies. If present, icker noise will always dominate at
suciently low frequencies.
10.1.1 Thermal Noise in Resistors
A noise voltage, as shown in Figure 10.1, is always present at the terminals of a resistor
because the charge carriers within the device undergo Brownian motion. The random-walk
of individual charges constitutes a random current, which produces a corresponding thermal
noise voltage across the terminals of the device. This noise was identied and carefully
measured by J. B. Johnson while working at Bell Telephone Laboratories
1
. A model for
the noise measured by Johnson was published at the same time by H. Nyquist, who worked
out an expression for the rms thermal noise voltage across any conductor. The result was
1
Thermal Agitation of Electricity in Conductors, J. B. Johnson, Physical Review, Vol. 32, p. 97, July,
1928.
321
322 CHAPTER 10. NOISE IN 1- AND 2-PORTS
derived using basic thermodynamics, and a fundamental result from statistical mechanics,
and is known as Nyquists theorem.
2
R e
n
(t)
+
-
t
e
n
(t)
Figure 10.1: Noise voltage across the terminals of a resistor as it would be displayed on an
instrument having a lowpass frequency response with bandwidth, B. The rms noise voltage
is proportional to
p
B and the typical time between zero crossings is proportional to B
1
.
The noise voltage waveform is properly modeled as a zero-mean, wide-sense-stationary,
random process with Gaussian probability density function. The spectral density of the
noise voltage due to thermal uctuations is given by the Nyquist noise formula:
S
n
(f) = 4R
hf
e
hf/kT
1
(Volts
2
/Hz) (10.1)
where
f = frequency, Hz
h = 6.62 10
34
J sec, Planck
0
s constant
k = 1.38 10
23
J/K, Boltzman
0
s constant
R = resistance,
T = physical temperature of the resistor, K
(10.2)
When f kT/h so that the exponential term in the denominator of equation 10.1 is well
approximated by the rst two terms in a Taylor series expansion, Nyquists formula reduces
to:
S
n
(f) = 4kTR (Volts
2
/Hz) (10.3)
At standard temperature, T = 290 K, kT/h ' 6 10
12
Hz = 6000 GHz, which lies in
the infrared. So the low-frequency approximation will be valid for all RF and microwave
applications. The spectral densities given in equations 10.1 and 10.3 represent the mean-
square noise across the resistor within a bandwidth of 1 Hz, centered on any frequency
f. In practice, the insrument used to sense the noise voltage will have a nite bandwidth
that diers from 1 Hz. Suppose the instrument responds only to frequencies between the
upper and lower frequencies denoted by f
l
and f
h
, respectively. Then the mean-square noise
voltage measured by the instrument would be:
< e
2
n
>=
f
h
f
l
S
n
(f)df , (10.4)
or, denoting the bandwidth by B = f
h
f
l
:
< e
2
n
>= 4kTBR (Volts
2
) (10.5)
2
Thermal Agitation of Electric Charge in Conductors, H. Nyquist, Physical Review, Vol. 32, p. 110,
July 1928.
10.1. NOISE CHARACTERIZATION OF 1-PORTS 323
For example, suppose we wish to calculate the mean-square noise voltage produced
across a 100 k resistor in a bandwidth of 1 MHz at standard temperature (T = 290 K).
At standard temperature:
4kT = 1.6 (10
20
) Joules (10.6)
< e
2
n
> = 4kTBR
= 1.6 (10
20
)(10
6
)(100)(10
3
)
= 1.6 (10
9
) (Volts)
2
Thus the rms noise voltage is
e
rms
=
< e
2
n
> = 40 (10
6
) V = 40 V. (10.7)
The rms noise voltage across the resistor scales as
p
B, so the noise voltage across the resistor
in a 100 MHz bandwidth would be ten times as large, or 0.4 mV.
Equation 10.5 can be generalized to describe the open circuit noise voltage across a
complex impedance Z that is realized by lossless elements and resistances all having the
same physical temperature. Suppose the impedance is Z(f):
Z(f) = R(f) +jX(f), (10.8)
then the noise variance across that impedance within bandwidth B will be
< e
2
n
>= 4kT
B
R(f) df. (10.9)
For example, we can calculate the total mean-square noise voltage across the parallel RC
circuit shown in Figure 10.2. Here, we will make use of equation 10.9, which provides
R (noisy) C
Figure 10.2: Parallel RC circuit with a noisy resistor.
a prescription for calculating the noise voltage between two terminals associated with a
network consisting of an arbitrary conguration of dissipative and lossless elements. The
only requirement is that all dissipative elements are at the same physical temperature. The
impedance of the parallel RC combination is:
Z(!) =
R(1 j!RC)
1 + (!RC)
2
(10.10)
According to equation 10.9 the mean-square noise voltage depends only on the real part of
the impedance:
<[Z] = R(f) =
R
1 + (2fRC)
2
.
324 CHAPTER 10. NOISE IN 1- AND 2-PORTS
Using equation 10.9, the total mean-square noise voltage across the resistor is:
< e
2
n
>= 4kT
1
0
R
1 + (2fRC)
2
df =
kT
C
.
Note that the result depends only on the capacitance, C, and is independent of the resistance!
How is this reconciled with the fact that the mean-square voltage within any dierential
bandwidth, df, is proportional to the resistance? The answer is that the bandwidth of the
RC lter is inversely proportional to R; so for xed capacitance, larger resistances have
larger mean-squre voltages but proportionally smaller bandwidths, hence the total noise
power, integrated over all frequencies, is independent of the resistance.
A Thevenin equivalent circuit can be used to model a noisy resistor, as shown in Figure
10.3a. The Norton equivalent circuit shown in Figure 10.3b is an equivalent alternative
(a)
e
n
R (noiseless)
< e
2
n
>= 4kTBR
(b)
i
n
g = 1/R (noiseless)
< i
2
n
>= 4kTBg
Figure 10.3: (a) Thevenin equivalent circuit. (b) Norton equivalent circuit.
representation.
The available noise power from a noisy resistor is dened to be the power delivered to a
matched load, as in Figure 10.4. The instantaneous power delivered to the load is
(en/2)
2
R
,
e
n
R
e
n
/2 R
+
-
Figure 10.4: Circuit for calculation of noise power available from a noisy resistor.
so the average available noise power is
<e
2
n
>
4R
. Using Equation 10.5, the average available
noise power within a bandwidth B is:
< e
2
n
>
4R
=
4kTBR
4R
= kTB (W) (10.11)
This result says that the noise power available from a resistor (or network of resistors all at
the same physical temperature) is independent of the resistance. For later use, we will also
10.1. NOISE CHARACTERIZATION OF 1-PORTS 325
dene the available noise power per unit bandwidth:
N
a
= kT (W/Hz) (10.12)
It is useful to know that kT
o
= 4 10
21
Watts, or approximately 174 dBm. So the noise
power available from any resistor at standard temperature is 174 dBm per Hz.
Next, the resistor noise model is generalized to allow it to represent arbitrary one-ports,
such as an antenna.
10.1.2 Noise Representation of Arbitrary 1-ports
If a 1-port device contains sources of noise in addition to the thermal noise it is said to exhibit
excess noise. In this case it is still possible to model the device by a Thevenin or Norton
equivalent circuit, and to use the Nyquist noise equation to calculate mean-square noise
voltage, but it is necessary to dene an eective temperature (dierent than the physical
temperature) to account for the extra noise. When this model is used, the formula for the
voltage is
< e
2
n
>= 4kT
n
BR (10.13)
and the noise current is
< i
2
n
>= 4kT
n
Bg (10.14)
where R and g=1/R represent the actual resistance and conductance, but the temperature
T
n
is called the equivalent noise temperature of the device.
10.1.3 Noise Representation of a Receiving Antenna
Antennas are 1-ports (like resistors) and the noise power available from an antenna can be
represented by an equivalent noise temperature. The real part of the antenna impedance
can be written as the sum of radiation and ohmic resistances
Re[Z
ant
] = R
rad
+R
loss
(10.15)
where R
rad
accounts for power radiated and R
loss
accounts for power that is dissipated
in the antenna and its nearby environment. An antenna that transmits most of the power
delivered to it when used for transmission is called an ecient antenna. An ecient antenna
has:
R
loss
R
rad
(10.16)
The noise voltage across an antennas terminals comes from two sources:
1. Noise received from external sources.
2. Thermal noise generated in R
loss
.
The noise voltage across the antenna terminals can be represented as if it was generated by
an actual resistor, R = R
rad
+ R
loss
, at some temperature T
A
which is called the eective
antenna temperature or equivalent noise temperature of the antenna. This is essentially the
same as T
n
, which was dened in section 10.1.2; but the subscript A is used to make it clear
that the eective temperature is associated with an antenna. For an ecient antenna, T
A
primarily represents noise received from external sources and has nothing to do with the
physical temperature of the antenna. Instead, T
A
will depend on the intensity of external
326 CHAPTER 10. NOISE IN 1- AND 2-PORTS
noise signals picked up by the antenna at the particular frequency of interest. In general,
T
A
is a parameter that must be measured, or known from previous experience with the
particular antenna and location at which the antenna is to be used.
For example, suppose that an ecient antenna with radiation resistance of 200 has an
rms noise voltage of 0.1 V at its terminals when measured in a bandwidth of 10
4
Hz. The
equivalent noise temperature of the antenna can be calculated using:
e
rms
=
< e
2
n
> = 0.1 V =
4kT
A
BR.
With R = 200 and B = 10
4
:
T
A
=
(0.1 V)
2
4kBR
= 90.6 K
Figure 10.5 shows the equivalent antenna circuit (for noise analysis).
200 At 90.6 K physical temperature
Figure 10.5: Equivalent circuit for analysis of antenna noise.
Typical antenna noise temperatures are shown in Figure 10.6. Noise radiated by sources
within the earth-ionosphere cavity is the dominant source of noise below approximately
20 MHz, and is largest at night (corresponding to the curve labeled Max.) when low
ionospheric absorption allows noise to be propagated in from long distances. During the
daytime, ionospheric absorption limits the amount of noise received from distant sources
and the antenna temperature is signicantly lower than at nighttime. Above 20 Mhz, the
gure assumes that the antenna beam is directed away from the earth, so that dominant
noise sources are extraterrestrial noise sources.
If we ignore radio noise generated by human activities, the primary source of natural
noise between 20 MHz (roughly) and 1 GHz is radio noise emitted by sources in our galaxy
and, depending on the antenna pattern and the pointing direction of the antenna relative
to the sun, perhaps some noise from the sun. Antenna temperatures for antennas that
look at the sky vary with time of day because the plane that contains most of the radio
sources in our galaxy will drift through the antenna beam as the earth rotates. Antenna
temperature due to the galactic sources also varies approximately as f
2
, so that maximum
daily antenna temperatures might range from several thousand K at 100 MHz to a few tens
of K at 1 GHz. At frequencies between 1-5 GHz or so, T
A
reaches its smallest values. In fact,
the contribution from sources within our galaxy falls to negligible values in this frequency
range, and the dominant noise is contributed by a background noise that is uniform in all
directions and corresponds to an equivalent blackbody temperature (and hence, antenna
temperature) of approximately 3 K. This is the so-called microwave background radiation
which was discovered by Bell Laboratories scientists Arno Penzias and Robert Wilson in
1965. This discovery conrmed theoretical predictions based on the Big Bang model of an
expanding universe. Penzias and Wilson were awarded the Nobel Prize in physics in 1978,
in recognition of the signicance of their discovery. Their original paper had the unassuming
title A Measurement of Excess Antenna Temperature at 4080 Mc/s.
10.1. NOISE CHARACTERIZATION OF 1-PORTS 327
1
10
100
1000
10000
100000
1e + 06
1e + 07
1e + 08
1e + 09
1e + 10
1e + 11
1e + 12
1e + 13
1e + 14
1e + 15
0.1 1 10 100 1000
T
A
Frequency - MHz
Max.
Min.
Daily min.
Daily max.
Atmospheric noise
Cosmic noise
Figure 10.6: Typical antenna temperatures as a function of frequency. These curves are
adapted from Figure 11-44 in the book Electromagnetic Waves and Radiating Systems, by
Jordan and Balmain, Prentice Hall, 1968.
328 CHAPTER 10. NOISE IN 1- AND 2-PORTS
At frequencies above 5 GHz, antenna temperatures begin to rise again and can approach
several hundred K at certain frequencies, due to thermal emission from water vapor and
oxygen (O
2
) in earths atmosphere.
It can be shown that when an antenna beam looks at an object that emits thermal (black
body) radiation, the antenna temperature will be equal to the temperature of the object,
provided that the antenna beam is lled by the object. If the antenna beam views many
objects, each having dierent temperatures, then the resulting antenna temperature will be
a weighted average of the individual object temperatures, where the weighting depends on
the angular size of each object, and the angular response function of the antenna.
When an antenna beam is pointed at the earth, as in Figure 10.7a, the antenna tem-
perature will be approximately equal to the equivalent black body temperature of the earth
or, roughly, 290 K. This has implications for satellite-to-ground communications systems.
For example, when a ground station with a narrow antenna beam looks at a geostationary
satellite (as shown in Figure 10.7b ), the earth station antenna beam sees mostly the sky.
(a) (b)
Figure 10.7: (a) When a satellite antenna beam looks at the earth, the beam sees a warm
body, emitting thermal radiation with an eective temperature on the order of 290 K. (b)
When a ground station antenna looks at a satellite, the beam sees mainly cold space.
Hence, a ground station antenna operated at a frequency between roughly 1-5 GHz will have
a signicantly smaller antenna temperature than a satellite antenna looking back at earth.
Since the ground station antenna looks at the cold sky, it picks up very little external
noise and the noise generated within the ground station receiver is usually a limiting factor
in determining the receiver sensitivity. On the other hand, the satellite antenna looks at
warm earth, and is relatively noisy and commonly contributes more noise than the receiver
itself. This means that it is worthwhile to use an extremely low-noise receiver for the ground
station, but not for the satellite.
For a point-to-point terrestrial communications link with the antenna radiation pattern
directed horizontally over the surface of the earth, half of the beam will see cold sky, and
the other half will see earths surface, resulting in an antenna temperature somewhere in
between the sky temperature and 290 K. The antenna beam may also see non-thermal
sources of noise located on earths surface, which raise the eective temperature of the earth
above 290 K.
10.1.3.1 The eective antenna temperature of an inecient antenna
Consider an antenna with feed-point resistance
Re[Z
ant
] = R
rad
+R
loss
, (10.17)
10.1. NOISE CHARACTERIZATION OF 1-PORTS 329
where R
rad
, R
loss
represent radiation and loss components of the feedpoint resistance, re-
spectively. It is assumed that the presence or absence of loss does not aect the current
distribution that is responsible for the radiated elds. Hence if the antenna and all compo-
nents in the vicinity of the antenna are changed to lossless materials (i.e. lossy conductors
replaced with perfect conductors, lossy dielectrics replaced with perfect dielectrics) then
R
loss
!0, and R
rad
is unaected by this change.
Denote the antenna temperature of the (hypothetical) lossless antenna by T
A
. The noise
power per unit bandwidth (p.u.b.) available from the lossless antenna would be kT
A
. For
simplicity, assume that all lossy materials are at physical temperature T
loss
. The noise
power p.u.b. available from the loss resistance only will be kT
loss
.
An equivalent circuit for the lossy antenna is obtained by replacing noisy resistances
R
rad
and R
loss
with Thevenin noise voltage sources having rms voltages
p
4kT
A
R
rad
and
p
4kT
loss
R
loss
in series with noiseless resistances R
rad
and R
loss
, respectively. The open cir-
cuit noise voltage across the antenna terminals will have rms value
4k(T
A
R
rad
+T
loss
R
loss
),
and the noise power available p.u.b will be
N
a
=
k(R
rad
T
A
+R
loss
T
loss
)
R
rad
+R
loss
= k(T
A
+ (1 )T
loss
) (10.18)
where
=
R
rad
R
rad
+R
loss
(10.19)
is the antenna eciency. Therefore, the eective antenna temperature is
T
eff
= T
A
+ (1 )T
loss
. (10.20)
Equation 10.20 says that as the antenna eciency decreases ( ! 0) the eective antenna
temperature aproaches the temperature of the lossy components of the antenna.
Now consider the eect of antenna losses on signal to noise ratio at the antenna output.
In the lossy case, the available signal power is reduced by the factor relative to a lossless
antenna. The noise power p.u.b. available from a lossless antenna is kT
A
, whereas in the
lossy case it is kT
eff
which can be written as kT
A
[1 +((1 )/)T
loss
/T
A
]. Therefore the
presence of loss resistance degrades the signal-to-noise power ratio at the antenna output
terminals by the factor
SNR
lossy
SNR
lossless
=
1
1 +
1
T
loss
T
A
. (10.21)
This is an interesting result, because it shows that SNR degradation due to antenna losses
is determined by antenna eciency as well as the ratio T
loss
/T
A
. When T
loss
/T
A
1, then
even a small decrease in antenna eciency can signicantly degrade the SNR. Consider a
satellite ground station where T
A
may be only 30 K. With T
loss
= 290 K, and antenna
eciency = 0.9 (90%), the SNR is degraded by the factor 0.48, or more than 3 dB. On
the other hand, if T
loss
/T
A
1, then even an inecient antenna can provide essentially the
same SNR as a lossless antenna. For example, the small antennas used for AM broadcast
radios may have eciencies of 1% ( = 0.01) or less, so that the factor (1)/ ' 100 or so.
The antenna temperature, T
A
is typically very large at AM broadcast frequencies, so that
T
loss
/T
A
1. In this situation, the product [(1 )/][T
loss
/T
A
] can be 1 even when
itself is small, in which case the SNR available from a small inecient antenna is essentially
the same as that available from a much larger antenna with high eciency (provided that
the radiation patterns of the small and large antennas are similar).
330 CHAPTER 10. NOISE IN 1- AND 2-PORTS
10.2 Noise Characterization of Linear 2-ports
10.2.1 Eective Input Noise Temperature
The noise added by a 2-port (e.g., amplier, lter, mixer) can be characterized by an eective
input temperature for a given source impedance at a given frequency.
Consider in Figure 10.8 a 2-port with a hypothetical noiseless input termination, Z
s
,
where N
avo
= available output noise power (the noise power delivered to a matched load).
T
s
= 0
N
avo
Noisy
2-port
Z
S
Figure 10.8: 2-port with a hypothetical noiseless input termination, Z
S
.
Figure 10.9 shows an equivalent circuit where the noise has been removed from the
2-port and assigned to the input termination by raising the temperature of the input ter-
mination to a value denoted by T
e
.
T
s
= T
e N
avo
Noiseless
2-port
Z
S
Figure 10.9: Equivalent circuit for a 2-port with a hypothetical noiseless input termination,
Z
S
.
T
e
is the temperature which, when assigned to the input termination, produces the same
available output noise power as that of the actual 2-port. In general, T
e
is a function of Z
s
and frequency.
T
e
only characterizes the noise generated within the 2-port, not the input termination.
The operating noise temperature of a system (T
op
) is used to characterize the total amount of
output noise due to the combined contributions from the 2-port and the input termination,
i.e., in a practical system the source is also noisy and can be assigned a noise temperature,
T
s
, as in Figure 10.10. An equivalent circuit for noise analysis can be obtained by adding
T
s
> 0
N
avo
Noisy
2-port
Z
S
Figure 10.10: Realistic system with a noisy source.
the eective input temperature of the 2-port to the source temperature as in Figure 10.11.
10.2. NOISE CHARACTERIZATION OF LINEAR 2-PORTS 331
T
op
= T
s
+ T
e N
avo
Noiseless
2-port
Z
S
Figure 10.11: Eective input temperature of the 2-port is added to the source temperature.
T
s
could represent the antenna noise or the noise from the source impedance of a signal
generator. Note carefully that by adding the noise temperatures of the source and the
2-port, we have eectively added the noise powers, since noise power is proportional to
temperature. The implicit assumption is that the input noise voltage due to the source and
due to the 2-port are statistically independent, and hence, uncorrelated. In that case the
noise powers due to the two sources add.
10.2.2 Noise Factor (F) and Noise Figure (NF)
Noise Factor (F) or Noise Figure (NF) are alternatives to the eective input termperature
for characterizing the noise performance of a 2-port. The noise factor (F) can be dened in
two equivalent ways. The rst denition is based on the signal-to-noise ratio degradation
between the input and output of a noisy 2-port, assuming that the noise power is calculated
within the same bandwidth at both ports.
Consider Figure 10.12, which can be used as a basis for deriving an expression for the
Noise Factor. The Noise Factor is dened under the assumption that the noise power per
T
o
Z
S
N
avs
S
avs
N
avo
= G
A
N
avs
+ N
2port
S
avo
= G
A
S
avs
Noisy
2-port
Available Gain, G
A
Figure 10.12: System for dening F.
unit bandwidth available from the source is constant at the value kT
o
, where T
o
is the
standard temperature 290 K:
F
INPUT SNR
OUTPUT SNRTs=To=290K
(10.22)
332 CHAPTER 10. NOISE IN 1- AND 2-PORTS
where
T
o
= 290 K standard temperature
S
avs
= signal power available from the source
N
avs
= kT
o
noise power available from the source
G
A
=
Savo
Savs
available gain of the 2 port
N
2port
= kT
e
G
A
noise power (1 Hz BW) available at output due to 2 port only
(10.23)
It is important to note that F does not necessarily describe the SNR degradation in a
real system, because the denition is based on the assumption that the input termination
has eective temperature T
o
= 290 K.
The second denition of F can be derived from Equation 10.22 with reference to Figure
10.12:
F
S
avs
/N
avs
S
avo
/N
avo
=
S
avs
/N
avs
G
A
S
avs
/N
avo
=
N
avo
G
A
N
avs
=
G
A
N
avs
+N
2port
G
A
N
avs
(10.24)
The nal result can be thought of as an alternative denition of Noise Factor that does not
make reference to any particular signal. Here F is simply the ratio of the actual noise power
available at the output of the 2-port (with input termination at standard temperature) to
the noise power that would be available if the 2-port was noiseless.
The noise factor can be expressed in terms of the eective input temperature of the
2-port:
F = 1 +
N
2port
G
A
N
avs
= 1 +
N
2port
/G
A
N
avs
= 1 +
kT
e
kT
o
(10.25)
so
F = 1 +
T
e
T
o
, T
o
= 290 K
Since T
o
is a standard temperature (= 290 K), specication of eective input temperature
(T
e
) is equivalent to specifying F. For a noiseless 2-port, T
e
= 0, which means that F = 1,
i.e., there is no SNR degradation.
Noise gure (NF) is simply the noise factor expressed in dB, i.e.,
NF = 10 log(F) (10.26)
= 10 log
1 +
T
e
T
o
1
0
G
a
(f) df (10.43)
= k T
op
G
ao
1
0
G
a
(f)
G
ao
df
Equation 10.43 has the form kTBG if the equivalent noise bandwidth, B
n
is dened as
follows:
B
n
=
1
0
G
a
(f)
G
ao
df (10.44)
Then Equation 10.43 becomes
N
out
= k T
op
G
ao
B
n
(10.45)
In essence, by dening the equivalent noise bandwidth, the actual gain versus frequency
curve is replaced with an idealized, rectangular response function as seen in Figure 10.18,
where the area under the equivalent rectangular response is equal to the area under the
actual response function.
G
ao
Idealized Response
Actual Response
f
G
a
(f)
B
n
Figure 10.18: Idealized rectangular response function.
In general, the noise bandwidth of a system will be larger than the 3 dB bandwidth.
For example, it can be shown that for circuits whose frequency responses are determined by
simple series or parallel RLC circuits, the 3 dB bandwidth and B
n
are related by:
B
n
=
2
B
3dB
' 1.6B
3dB
10.3.2 Noise Floor, or Minimum Discernible Signal (MDS)
As mentioned earlier, the noise oor or MDS is dened to be the input signal level required
to give some specied output SNR. The following example illustrates how the noise oor is
calculated.
10.3. SENSITIVITY OF A RECEIVING SYSTEM 339
10.3.2.1 Example - MDS for a receiving system
Find the noise oor (MDS) for the receiving system in Figure 10.19. Assume that the
minimum required output signal-to-noise ratio that is required for detection of the signal is
0 dB (SNR
out,min
= 1). The eective input temperature of the receiver is found as follows:
S
avo
N
avo
50
T
s
= 290 K
V
i
B
n
= 2.1 kHz
NF = 8 dB
Receiver
Figure 10.19: Receiving system for deriving noise oor (MDS).
NF = 8 dB (10.46)
F = 10
NF/10
= 6.31 = 1 +
T
e
290
T
e
= (F 1) 290 = 1540 K
Calculate the operating noise temperature of the system:
T
op
= 290K + 1540 K = 1830 K (10.47)
Available output noise power:
N
avo
= k T
op
B
n
G
a
(10.48)
Available output signal power:
S
avo
= S
avs
G
a
(10.49)
Output SNR:
SNR
out
=
S
avo
N
avo
(10.50)
=
S
avs
G
a
k T
op
B
n
G
a
SNR
out
=
S
avs
k T
op
B
n
We require SNR
out,min
= 1, so the corresponding minimum available input signal power is:
S
avs,min
= k TopB
n
(10.51)
= (1.38)(10
23
)(1830)(2.1)(10
3
)
= 5.30(10
17
) Watts
340 CHAPTER 10. NOISE IN 1- AND 2-PORTS
In dBm (dB referenced to a milliwatt)
MDS S
avs,min
= 10 log
5.3 x10
17
10
3
(10.52)
MDS = 132.8 dBm.
Thus, the noise oor or MDS of the receiver is -132.8 dBm. Sometimes the open circuit
antenna voltage is given instead of the input power. Recall that S
avs
is dened to be the
available signal power from the source. The source (antenna) impedance is 50 , so
S
avs,min
=
V
2
s,min
8R
s
(10.53)
=
V
2
s,min
8 (50)
V
s,min
= 0.146 V
This is the peak value of the open circuit antenna voltage. The rms value is
V
s,min(rms)
= 0.103 V (10.54)
Note that the output SNR used in the denition of noise oor should be interpreted as
the SNR at the input to the demodulator stage. Nothing has been said about the particular
type of demodulator that is used in the system. In order to specify the SNR that is required
for detection of a signal, it is necessary to have some information about what SNR is required
by the particular demodulator that will be used.
10.3.2.2 Example - MDS for TV receiving system
In Figure 10.20 a 300 antenna is connected to a TV receiver with 300 input impedance.
The eective temperature of the antenna is 1000 K. The noise gure of the receiver is 4 dB.
The eective noise bandwidth is 5 MHz. Find the input signal level (in dBm) required to
B
n
= 5 MHz
NF = 4 dB
Output of last IF stage,
to demodulator
Receiver
T
A
= 1000 K
Figure 10.20: Television receiving system.
give a 15 dB signal-to-noise ratio at the output of the receiver. What is the corresponding
open circuit antenna voltage?
10.3. SENSITIVITY OF A RECEIVING SYSTEM 341
The desired output SNR is: SNR
out
)15dB )10
15/10
= 31.62
SNR
out
=
S
avs
k T
op
B
n
(10.55)
S
avs
= k T
op
B
n
(SNR
out
)
T
op
= T
s
+ T
e
Find T
e
from NF:
F = 10
NF/10
= 1 +
T
e
T
o
(10.56)
F = 2.51 )T
e
= 438.5 K
T
op
= 1000 + 438.5 = 1438.5 K
S
avs
= 1.38 10
23
(1438.5)(5 10
6
)(31.62)
= 3.14 10
12
W = 3.14 10
9
mW
= 85 dBm.
Thus, -85 dBm of input signal power is required for 15 dB SNR at the output of the receiver.
The next example calculates the degradation in sensitivity resulting from addition of a
lossy cable between the antenna and the receiver.
10.3.2.3 Example - TV system MDS with a lossy cable
Repeat the calculation for the case where the antenna is connected to the receiver through
a long transmission line having 10 dB loss as in Figure 10.21. Calculate the eective input
B
n
= 5 MHz
NF = 4 dB
Output of last IF stage,
to demodulator
Receiver
T
A
= 1000 K
Lossy transmission
line; 10 dB loss,
physical temp. T
o
Figure 10.21: Antenna-cable-receiver.
temperature of the lossy transmission line. First convert loss from dB to power ratio:
L = 10
10/10
= 10. Then T
e
for lossy transmission line is T
e
= T
o
(L1) = 2610 K. Eective
342 CHAPTER 10. NOISE IN 1- AND 2-PORTS
input temperature of the transmission line-receiver cascade:
T
e
= T
e1
+
T
e2
G
1
= 2610 +
438.5
1/10
(10.57)
= 6995 K (NF = 14dB)
T
op
= 1000 + 6995 = 7995 K
S
avi
= k T
op
B
n
(SNR
out
)
= 1.74 10
11
W = 1.74 10
8
mW
= 77.6 dBm
The input signal level must be 7.4 dB higher to get the same SNR as in the rst example.
The next example illustrates how one can calculate the NF required of a preamp, in
order to set the system MDS to some specied value.
10.3.2.4 Example - Calculating preamp NF required for a specied MDS.
In Figure 10.22 a preamplier has been added in front of the lossy cable, in an attempt to
make up for the loss in the cable. The preamp has available gain 10 dB, which counteracts
the 10 dB cable loss. The question is, what NF (or T
e
) must the preamp have if the overall
system is to have the same MDS as the receiver alone? The NF of this system must be 4
B
n
= 5 MHz
NF = 4 dB
Receiver
Lossy transmission
line; 10 dB loss,
physical temp. T
o
NF =?
G
A
= 10 dB
Preamp
Figure 10.22: Antenna-preamp-cable-receiver.
dB if it is to be equivalent to the receiver alone.
Preamp Gain = 10 dB ) G
A1
= 10
Coax Gain = 10 dB ) G
A2
=
1
10
, T
e2
= 2610 K
Receiver T
e3
= 438.5 K
10.4. MEASUREMENT OF NOISE TEMPERATURE 343
Want the overall T
e
= 438.5 K,
T
e
= T
e1
+
T
e2
G
1
+
T
e3
G
1
G
2
(10.58)
T
e1
= T
e
T
e2
G
1
T
e3
G
1
G
2
= 438.5 2610/10 4385/(10)(1/10)
T
e1
= 261 K < 0
T
e1
came out negative which means that it cant be done! Even a noiseless preamp would
give an overall T
e
= 699.5 K ) NF = 5.33 dB! It turns out that the gain of the preamp
would have to be at least 12 dB (for a noiseless preamp) and even higher for a real, noisy,
preamp in order to achieve a 4dB NF for the system. This example illustrates that a preamp
must more than just make up for the loss of an attenuator that follows it if the noise gure
of the system is to be preserved.
The following section describes the principles behind practical measurements of noise
temperature.
10.4 Measurement of Noise Temperature
The basic approach used for noise temperature measurements is based on a comparison of
the output noise powers obtained for two dierent eective source temperatures. The ratio
of the two resulting output noise powers can be used to determine Te and, hence, F and
NF. Refer to Figure 10.23.
Z
S
Z
S
T
H
T
C
T
e
=?
N
H
N
C
Figure 10.23: Noise temperature measurement with hot and cold loads.
T
H
= equivalent noise temperature of the hot load (10.59)
T
C
= equivalent noise temperature of the cold load
N
H
, N
C
) available output noise powers corresponding to T
H
, T
C
, respectively, where
N
H
= k(T
e
+ T
H
) B
n
G
A
(10.60)
N
C
= k(T
e
+ T
C
) B
n
G
A
344 CHAPTER 10. NOISE IN 1- AND 2-PORTS
The ratio N
H
/N
C
is called the Y factor and is the parameter that would be measured in
practice:
Y =
N
H
N
C
=
k(T
e
+ T
H
) B
n
G
A
k(T
e
+ T
C
) B
n
G
A
(10.61)
Y =
T
e
+ T
H
T
e
+ T
C
Since T
H
and T
C
are known, it is possible to solve for T
e
:
T
e
=
T
H
Y T
C
Y 1
(10.62)
Note that measurement of the eective input temperature of a 2-port is possible without
making accurate absolute power measurements. Only the ratio (relative magnitudes) of the
output powers obtained with hot and cold input terminations is needed.
In commercial noise gure meters, the cold load is usually a 50 resistor at room
temperature. A noise diode is turned on for the hot load. The diode generates excess noise
and, for the purposes of the measurement, acts like a resistor at some temperature 290K.
The excess noise ratio is specied for noise diodes is dened by
ENR = 10 log
T
H
290
290
(10.63)
A typical noise diode ENR is in the neighborhood of 5 dB.
When measuring small NFs the ENR must be known very accurately if precise mea-
surements are necessary. Also, it is important that the source impedance of the noise diode
doesnt change between the o (cold) and on (hot) states, because the eective input tem-
perature and available gain of a 2-port usually depend on the source impedance.
10.4.1 Practical Considerations
In practice the NF measurement set-up looks like Figure 10.24. What is actually measured
Backend
T
eb
, B
n
N
H
N
C
Noise
Diode
T
H
, T
C
Device
Under
Test
T
e
=?
G =?
Noise Diode
control signal
Figure 10.24: Noise Figure measurement setup.
is the eective temperature (or noise gure) of the cascade:
T
ec
= T
e
+
T
eb
G
(10.64)
10.5. A MODEL FOR THE DEPENDENCE OF T
E
ON Z
S
345
To determine T
e
, it is necessary to know T
eb
and G. These parameters can be determined
if the Y-factor for the backend alone is known, as in Figure 10.25.
Backend
T
eb
, B
n
N
H,0
N
C,0
Noise
Diode
T
H
, T
C
Noise Diode
control signal
Figure 10.25: Measuring the eective input temperature of the backend.
To nd the Y-factor of the backend, remove the device under test and measure the ratio
of N
H,0
and N
C,0
where
N
H,0
= k(T
eb
+ T
H
) B
n
G
b
(10.65)
N
C,0
= k(T
eb
+ T
C
) B
n
G
b
The Y-factor of the backend
Y
0
=
N
H,0
N
C,0
=
T
eb
+ T
H
T
eb
+ T
C
(10.66)
is used to determine the eective input temperature of the backend, T
eb
. The gain of the
device under test can be determined by noticing that with the DUT in the circuit we measure
N
H
= k (T
ec
+ T
H
) BGG
b
(10.67)
N
C
= k (T
ec
+ T
C
) BGG
b
To nd G,
N
H
N
H,0
=
T
ec
+ T
H
T
eb
+ T
H
G (10.68)
Since T
ec
, T
eb
and T
H
are known (or calculated), G can be determined. The power ratio
N
C
/N
C,0
could also be used to nd G. In practice the measurement of the backend noise
temperature (i.e., measurement of N
H,0
, N
C,0
) is done as part of the calibration procedure
and then stored. The gain that is computed using the NF meter is averaged over the eective
noise bandwidth of the backend and is therefore usually dierent from the gain measured
at one frequency using a signal generator.
10.5 A model for the dependence of T
e
on Z
S
The noise generated within a noisy 2-port can be modeled with the equivalent circuit shown
in Figure 10.26 where all noise sources have been removed from the 2-port and replaced
with equivalent noise voltage and noise current generators at the input of the 2-port. The
voltage source accounts for all of the output noise in the case where the input of the 2-port
is terminated with a short circuit and the current source accounts for all of the output noise
when the input of the 2-port is left open-circuited. In general, the current and voltage
346 CHAPTER 10. NOISE IN 1- AND 2-PORTS
Noiseless
e
n
i
n
2-port
- +
Figure 10.26: A model for noisy 2-ports.
sources must be assumed to be correlated, i.e. < e
n
i
n
>6= 0, to account for the fact that a
particular noise generation mechanism may produce output noise with both open and short
circuit input terminations.
Up until now, when we have referred to noise voltages and currents, we have implicitly
assumed that e
n
and i
n
were functions of time, i.e. that e
n
and/or i
n
represented the
actual time-domain voltage or current waveforms that would be viewed on a instrument
with some bandwidth, B. For the following discussion, it will be convenient to represent
the bandlimited noise voltage and current with a complex envelope or, in other words, as
noise phasors. For the discussion in this section, let us assume that e
n
(t) and i
n
(t) are the
voltage and current waveforms in a 1 Hz bandwidth centered on some frequency, f. We can
write
e
n
(t) = <(E
n
(t)e
j!t
)
and
i
n
(t) = <(I
n
(t)e
j!t
)
where E
n
(t) and I
n
(t) are complex baseband representations of the bandlimited noise signals
e
n
(t) and i
n
(t). The complex signals E
n
and I
n
are analogous to voltage and current phasors
that we are familiar with. The dierence is that a purely sinusoidal signal, at some frequency,
can be represented by a constant, complex, phasor whose magnitude and phase completely
describe the signal. Similarly, a noise signal that has been bandlimited to a bandwidth 1
Hz, centered on some frequency, can be represented by a slowly varying
3
complex phasor.
We can do circuit analysis using noise phasors in the same way that we do circuit analysis
using standard signal phasors. The dierence is that with noise phasors, the quantity of
interest will involve the mean-square value of the noise phasor, since its mean value is always
zero, and is not particularly useful. For example, given the noise phasor E
n
, suppose that
it is necessary to calculate the mean-square value of the corresponding time-domain noise
waveformwe may do this using the following identity:
< e
2
n
>=< (<(E
n
e
j!t
))
2
>=
< |E
n
|
2
>
2
10.5.1 The relationship between T
e
and input noise voltage and
current.
Consider a noisy 2-port terminated with a noiseless source impedance, Z
S
, as shown in
Figure 10.27.
3
The time scale for signicant variation will be on the order of B
1
, or 1 second for B = 1 Hz.
10.5. A MODEL FOR THE DEPENDENCE OF T
E
ON Z
S
347
Noiseless
E
n
I
n
2-port
- +
Z
S
Figure 10.27: Noisy 2-port terminated with noiseless input termination.
For the purpose of deriving an expression for the eective input temperature of the 2-
port, the noise generators can be associated with the source, as shown in Figure 10.28. Note
that we have now switched to representing the noise voltage and current with their complex
noise phasors. The eective input temperature can be found by calculating the noise power
available from the source and then equating the result to kT
e
.
E
n
I
n
- +
Z
S
Figure 10.28: Equivalent input noise generators lumped into the source for the purpose of
deriving an expression for the eective input temperature of the 2-port.
The available noise power from the source shown in Figure 10.28 is
kT
e
=
< |E
n
+I
n
Z
S
|
2
>
8R
S
=
1
8R
S
{< |E
n
|
2
> +2<{< E
n
I
n
> Z
S
}+ < |I
n
|
2
> |Z
S
|
2
},
(10.69)
where the operator <{} extracts the real part of its argument. Dene the complex correlation
coecient, :
=
r
+j
i
=
< E
n
I
n
>
< |E
n
|
2
>< |I
n
|
2
>
and the noise voltage and current variances,
2
e
,
2
i
:
2
e
=< |e
n
|
2
>=
< |E
n
|
2
>
2
,
2
i
=< |i
n
|
2
>=
< |I
n
|
2
>
2
Then 10.69 can be written
kT
e
=
1
4R
S
{
2
e
+ 2<{Z
S
}
e
i
+
2
i
|Z
S
|
2
}. (10.70)
Expand equation 10.70 using Z
S
= R
S
+ jX
S
and =
r
+ j
i
and complete the squares
to show that:
T
e
=
2
i
4kR
S
{(R
S
+
r
i
)
2
+ (X
S
i
)
2
+
2
e
2
i
(1 ||
2
)}. (10.71)
348 CHAPTER 10. NOISE IN 1- AND 2-PORTS
Equation 10.71 shows how the eective input temperature of a 2-port depends on the source
impedance, Z
S
. The particular source impedance that minimizes T
e
is called the optimum
source impedance for minimum noise, Z
opt
. The corresponding minimum value of T
e
is
denoted by T
e,min
. It is not hard to show that
Z
opt
= R
opt
+jX
opt
=
e
1
2
i
+j
i
i
(10.72)
T
e,min
=
e
i
2k
{
1
2
i
+
r
}. (10.73)
When written in terms of Z
opt
and T
e,min
, the eective input temperature is:
T
e
= T
e,min
+T
o
G
n
|Z
S
Z
opt
|
2
R
S
(10.74)
where the parameter G
n
is called the noise conductance and is given by
G
n
=
2
i
4kT
o
. (10.75)
For a particular 2-port, the parameters T
e,min
, Z
opt
, G
n
are constants that completely
characterize the noise performance of the 2-port. The set of 4 parameters {T
e,min
, Z
opt
=
R
opt
+jX
opt
, G
n
} are called the noise parameters of the 2-port. Equation 10.74 shows that
the eective input temperature of a 2-port increases quadratically as the source impedance
moves away from the optimum value Z
opt
. The sensitivity of T
e
to changes in the source
impedance is determined by the magnitude of the noise conductance parameter, G
n
.
Equation 10.74 can be written in terms of reection coecients instead of impedances:
T
e
= T
e,min
+ 4G
n
Z
o
T
o
|sopt|
2
(1|s|
2
)|1opt|
2
= T
e,min
+ 4
Rn
Zo
T
o
|sopt|
2
(1|s|
2
)|1+opt|
2
(10.76)
where
S
=
Z
S
Z
o
Z
S
+Z
o
= actual source reection coecient
opt
=
Z
opt
Z
o
Z
opt
+Z
o
= optimum source reection coecient
R
n
=
2
e
4kT
o
= noise resistance
In this case, the noise parameters consist of {R
n
,
opt
, T
e,min
} or {G
n
,
opt
, T
e,min
}.
Contours of constant noise temperature are circles in the source reection coecient
plane. In some cases manufacturers will provide plots of constant noise temperature (or
NF) on the device data sheet. Comparison with constant available power gain curves makes
it possible to select a source impedance that provides the best compromise between gain
and NF.
10.5. A MODEL FOR THE DEPENDENCE OF T
E
ON Z
S
349
10.5.2 Low frequency approximation and op-amp example
At lower frequencies, where op-amps are commonly employed, the noise voltage-current
correlation coecient is often ignored by assuming that it is zero, i.e. = 0, in which case:
Z
opt
= R
opt
=
e
i
(10.77)
T
e,min
=
e
i
2k
(10.78)
When the correlation coecient is zero, the eective input temperature reduces to:
T
e
=
2
i
R
S
4k
+
2
e
4kR
S
=
1
2
T
e,min
[
R
S
Ropt
+
Ropt
R
S
]
(10.79)
Notice that the voltage noise dominates when R
S
is small (when R
S
< R
opt
) and the
current noise dominates when R
S
is large (when R
S
> R
opt
). Manufacturers will provide
measurements of
i
and
e
on the device data sheet. The rms noise voltage,
e
, is estimated
from noise measurements taken with a very small source impedance and the rms noise current
is estimated from noise measurements taken with a large value of R
S
. As an example, the
following data was taken from the data sheet for a low-noise op-amp. At 1 kHz, the rms
noise voltage and current are given as:
e
= 2.3 nV/
p
Hz
i
= 160 fA/
p
Hz
The optimum source resistance for minimum noise and the associated minimum eective
input temperature predicted by equations 10.77 and 10.78 are:
R
opt
' 14.4 k
T
e,min
' 13.3 K.
The minimum noise gure is therefore
NF
min
= 10 log(1 +
13.3
290
) ' 0.2 dB.
It is interesting to calculate the noise temperature of this device when operated with a lower
source impedance, e.g. for R
S
= 600 :
T
e
=
1
2
13.3[
0.6
14.4
+
14.4
0.6
] ' 160 K.
In this case, operating with a source impedance of 600 results in a noise temperature that
is more than 10 times as large as the noise temperature with the optimum source impedance.
350 CHAPTER 10. NOISE IN 1- AND 2-PORTS
10.6 References
1. Krauss, H. L., C. W. Bostian, and F. H. Raab, Solid State Radio Engineering, John
Wiley & Sons, New York, 1980.
2. Smith, Jack, Modern Communications Circuits, McGraw Hill, 1986.
3. van der Ziel, Aldert, Noise in Measurements, John Wiley & Sons, New York, 1976.
4. van der Ziel, Aldert, Noise in Solid State Devices and Circuits, John Wiley & Sons,
New York, 1986.
5. Vendelin, George D., Anthony M. Pavio, Ulrich L. Rohde, Microwave Circuit Design
Using Linear and Nonlinear Techniques, John Wiley & Sons, 1990.
10.7. HOMEWORK PROBLEMS 351
10.7 Homework Problems
1. The input of a 2-port is terminated with a 50 resistor that is at room temperature
(290K), and it is found that the noise power delivered to a 50 resistor at the output
of the 2-port is 10
15
W. When the input resistor is then cooled down to 77 K (the
temperature of liquid nitrogen), the output noise power drops by 2.5 dB.
(a) Find the eective input temperature of the 2-port. You may assume that the
eective input temperature is constant over the range of frequencies where the
2-port has appreciable gain.
(b) How accurately must the change in output noise powers be measured (in dB), if
the eective input temperature of the 2-port is to be measured to within plus or
minus 1K?
(c) Suppose that your measurement accuracy is plus or minus 0.1 dB. What is the
uncertainty in the eective input temperature measurement?
2. A receiver has noise bandwidth B
n
= 5.0 kHz. The receiver has a built-in meter that
gives a reading proportional to the total power at the output of the last IF stage
(just before the detector). Note that the meter will indicate the total (signal + noise)
power.
Suppose that you connect a signal generator with eective source temperature T
s
=
290K to the input of the receiver. The meter reading is found to increase by 2 dB
when the available signal power from the signal generator is raised from 0 (no output)
to -130 dBm. You may assume that the noise power available from the generator is
constant, and independent of the available signal power.
What is the eective input temperature of the receiver?
3. You are given two ampliers with the following characteristics:
#1 G
1
= 4 dB (10.80)
T
e1
= 175 K
#2 G
2
= 16 dB (10.81)
T
e2
= 200 K
where G
1
and G
2
are the available gains of the ampliers and T
e1
, T
e2
are the eective
input temperatures. In what order should they be cascaded so that the cascade has
the minimum possible noise gure? For the optimum conguration, give the eective
input temperature, noise factor, and noise gure.
4. Consider a cascade of identical ampliers with available gain G and input eective
input temperature T
e
.
(a) The noise measure of an amplier is, by denition, the eective input temperature
of an innite number of the ampliers in cascade. Find an expression for the noise
measure. Give your nal result in closed form, i.e., do not leave your result in
terms of an innite series.
352 CHAPTER 10. NOISE IN 1- AND 2-PORTS
(b) Can the noise measure concept be applied to a 2-port with available gain that is
less than 1?
(c) Suppose one is confronted with a situation like that in problem 3. That is, you are
given two ampliers with eective input temperatures T
e1
and T
e2
and available
gains G
1
and G
2
, respectively. Show that in order to achieve the lowest possible
noise temperature, they should always be cascaded such that the amplier with
the lowest noise measure is in front.
5. Suppose that it is necessary to decide how to cascade N 2-ports with eective input
temperatures T
i
and available gains G
i
(> 1), so that the cascade has the minimum
possible eective input temperature. Explain how noise measure can be used to decide
how to cascade the 2-ports, and why it is more ecient to use the noise measure rather
than simply evaluating the cascaded noise temperature of all of the possible cascades.
6. The following situation is often encountered in practice: An antenna is located some
distance from the receiver. A decision must be made whether to place the preampli-
er at the antenna (before the lossy cable) or at the receiver (after the lossy cable).
Suppose the distance between antenna and receiver is 100 meters. The connection will
be made with a 100 meter length of coaxial cable having an attenuation of .05 dB per
meter. The physical temperature of the cable will be 290K. A preamplier is obtained
that has an available gain of 16 dB and a NF of 0.50 dB.
(a) Find the eective input temperature of the cable followed by the preamp.
(b) Find the eective input temperature of the preamp followed by the cable.
7. The front end of a particular receiver consists of the following stages:
Preamp:
G
a
= 10 dB (10.82)
NF = 5 dB
Mixer/LO-IF:
G
a
= 60 dB (10.83)
NF = 11 dB
(a) Suppose the preamp has an input impedance of 50 and is connected to a 50
antenna with eective temperature 100K. If the equivalent noise bandwidth of
the system is 5 kHz, nd the minimum input signal level (in dBm) required to
give a 15 dB SNR at the output of the system.
(b) Find the rms open-circuit antenna voltage (in microvolts) that will give the re-
quired SNR at the output of the system.
(c) Now assume that the preamp noise gure is only 1 dB and repeat part 7a.
8. Consider a receiving system designed to detect signals from a deep-space probe. Sup-
pose that the probe emits a signal with bandwidth of 100 Hz and that the receiving
antenna collects enough energy to make a total signal power of -150 dBm available
10.7. HOMEWORK PROBLEMS 353
from the antenna. The eective antenna temperature is 30K. A preamplier with eec-
tive input temperature of 20K and available gain of 10 dB is mounted at the antenna
terminals. The output of the preamplier is connected to the input of the receiver
through a cable with 6 dB of loss. (The physical temperature of the cable is 290K.)
Assume that all ports are conjugately matched, and nd the maximum eective input
temperature that the receiver can have, if the signal-to-noise ratio at the detector is
to be at least 3 dB. Assume that the eective noise bandwidth of the receiver is 100
Hz.
9. A spectrum analyzer is a super-heterodyne receiver with a swept LO which displays
the eective input power delivered to its input terminals. The analyzer computes the
eective input power by measuring the total (signal+noise) power delivered by the last
IF stage and dividing this power by the total power gain of all the preceding stages.
Suppose that you want to measure the eective input temperature of a 2-port using a
spectrum analyzer and you know that the available gain of the 2-port is 25 dB. You
may assume that the input and output of the 2-port are conjugately matched when
terminated in 50 , and that the input impedance of the spectrum analyzer is 50 .
With a room temperature (290K) 50 load connected directly to the input of the
spectrum analyzer, the analyzer indicates that the eective input noise power is -168
dBm in a bandwidth of 1 Hz centered on the frequency of interest. You now connect
the 50 load to the input of the 2-port and connect the output of the 2-port to
the spectrum analyzer. With this conguration you nd that the equivalent input
noise power in a 1 Hz bandwidth measured by the spectrum analyzer is -140 dBm.
Remember that the number displayed by the spectrum analyzer is the equivalent input
power at the analyzers input port and not at the input of the 2-port under test.
(a) What is the eective input temperature of the spectrum analyzer?
(b) Find the eective input temperature of the 2-port.
10. The signal distribution network for cable television systems consists of cascaded sec-
tions of lossy transmission line and ampliers. Suppose that the lossy cable and am-
pliers are both matched to 75. Each lossy transmission line has loss 0.05 dB/meter,
length 500 meters and physical temperature 290K. Each amplier has available
gain 25dB and eective input temperature T
e
= 100K. Note that the gain of the am-
plier is just enough to make up for the loss in the cable, i.e., the cascade of a lossy
t-line and an amplier results in a new 2-port with 0dB gain.
Now suppose that it is necesssary to distribute a television signal from a main distri-
bution station to a remote site. The site is 1.5km from the main station, so it will be
necessary to use 3 sections of t-line and 3 ampliers.
(a) Suppose that the t-lines and ampliers are cascaded in the order:
t line )amplier )t line )amplier )t line )amplier
Find the eective input temperature of the cascade of all of the ampliers and
cables.
(b) Suppose that the t-lines and ampliers are cascaded in the order:
amplier )t line )amplier )t line )amplier )t line
354 CHAPTER 10. NOISE IN 1- AND 2-PORTS
Find the eective input temperature of the cascade of all of the ampliers and
cables.
(c) Suppose that the source at the main station has equivalent source temperature
T
S
= 290K. Assuming that the available signal power and system bandwidth is
the same in both cases, specify the improvement (in dB) that would result in the
output SNR if cascade (10a) were changed to cascade (10b).
11. Find the equivalent noise bandwidth, B
n
, in terms of the -3 dB bandwidth, B
3dB
, for
a system with available gain function equivalent to a lowpass Butterworth response
with order n. In other words, assume that the available gain function has the following
form:
G
A
(f) =
1
1 + (
f
B
3dB
)
2n
Give results for n = 1, 2, 3, 4.
12. Derive equations 10.76 starting with equation 10.74.
Chapter 11
Mixers
Mixers may be classied according to whether they are based on active or passive devices.
Another distinction, that can apply to either passive or active mixers, is whether mixing
occurs as a result of a soft nonlinearity such as the current-voltage relationship in a diode or
transistor, or whether mixing results from a hard nonlinearity such as from a switch. Most
mixers in use today are of the switching type, whereby diodes or transistors are used to
switch the connection between the RF input and the IF output at a rate that is controlled
by the local oscillator. We will give a brief overview of several representative mixer circuits
in this chapter.
11.1 Mixers Based on Gradual Nonlinearities
11.1.1 Single-ended BJT Mixer
The single-ended BJT mixer makes use of the nonlinear relationship between the base-
emitter voltage and the collector current. If the RF and LO voltages are applied to the base
of the BJT so that the base-emitter voltage v
i
(t) has both RF and LO signal components
then the collector current will contain terms that are proportional to all powers of the total
input voltage:
i
C
= i
DC
+ k
1
v
i
(t) + k
2
v
2
i
(t) + k
3
v
3
i
(t) +. . . (11.1)
In general, this produces all possible mixing products, including:
f
RF
, f
LO
, |f
RF
f
LO
|, |2f
RF
f
LO
|, |2f
LO
f
RF
|, . . . (11.2)
The output voltage can be developed by loading the collector with a parallel resonant circuit
so that a signicant output voltage is developed by only one of the frequency components
present in the collector current waveform. The collector circuit is usually tuned to resonate
at either f
RF
+ f
LO
or |f
RF
f
LO
|. Figure 11.1 is an example of a single-ended BJT mixer.
In a practical implementation of the mixer shown in Figure 11.1, a matching network
would be used between the RF source and the RF input in order to couple the maximum
amount of RF signal to the base of the transistor. The matching network would be designed
to present a high impedance to the LO signal. Similarly, a matching network would be used
between the LO source and the LO input, and this network would be designed to present a
high impedance to the RF source. In this way, the LO source is prevented from loading the
355
356 CHAPTER 11. MIXERS
n : 1
v
LO
v
RF
V
cc
C
LO
C
RF
v
IF
Figure 11.1: BJT mixer
RF source, and vice versa. A simpler compromise is to supply the LO voltage through a
small-value coupling capacitor (C
LO
) such that the LO source does not signicantly disturb
the matching for the RF signal. For this circuit the local oscillator drive level required is on
the order of 100mV. A BJT mixer of this type has conversion gain ' g
m
R
L
where R
L
is the
load resistance seen by the collector at f
IF
. This circuit is attractive because of its simplicity,
and it has been used in inexpensive mass-produced receivers for consumer applications. It
is not a high performance mixer, however. In general, a single-ended BJT mixer will have a
higher noise gure than a mixer that employs an FET, or a properly designed passive mixer
based on a diode bridge. In addition, BJT mixers are subject to severe intermodulation
distortion. To avoid this problem, v
RF
must be kept smaller than 10 mV. In general,
BJT mixers of this type have poor large signal characteristics.
11.1.2 Single-ended FET Mixers
If a FET is operated in its constant current region, then
i
D
= I
DSS
(1
v
gs
V
p
)
2
(11.3)
The circuit is similar to the one shown in Figure 11.1 for the BJT mixer, but the FET has
two advantages over the BJT. These are:
a much lower third-order IMD, since there is no cubic term in the i
D
versus v
gs
relationship
Much higher RF input voltages are usable, i.e., up to at least 100 mV.
A disadvantage of the FET mixer is a somewhat smaller conversion gain than the BJT
circuit.
A major drawback of the single-ended BJT and FET mixers is the presence of a local
oscillator component as well as an RF component at the output of the mixer. Although the
LO and RF signals will be ltered by the tuned output circuit, some of the relatively large
LO signal will inevitably leak through the mixer to the IF stages. A large LO component
11.1. MIXERS BASED ON GRADUAL NONLINEARITIES 357
at the input to the IF stages is obviously undesirable, since it can cause overload at the
IF stages. It can also create undesirable eects such as gain compression and possible
generation of intermodulation products in the IF stages.
11.1.3 Balanced Mixers
A balanced conguration can be used to eectively remove the LO and/or RF signals from
the output of the mixer. One type of balanced mixer often used in integrated circuits is the
dierential-pair multiplier as shown in Figure 11.2.
The multiplier is based on a dierential amplier. For a single transistor biased in the
i
C1
i
C2
I
EE
v
i
+
-
Figure 11.2: A dierential-pair amplier. The transistors are assumed to be biased in the
active region. The bias network is not shown.
active mode
i
C
= I
S
e
v
BE
/V
T
. (11.4)
For the emitter-coupled pair with identical transistors (I
S1
= I
S2
, V
T1
= V
T2
)
i
C1
i
C2
= e
vi/V
T
(11.5)
where
v
i
= v
BE1
v
BE2
Since the emitters are connected to a constant current source,
i
E1
+i
E2
= I
EE
(11.6)
or
1
(i
C1
+i
C2
) = I
EE
(11.7)
From Equation 11.5
i
C1
= i
C2
e
vi/V
T
(11.8)
so using Equation 11.7
i
C1
=
I
EE
1 +e
vi/V
T
(11.9)
i
C2
=
I
EE
1 +e
vi/V
T
(11.10)
358 CHAPTER 11. MIXERS
If the output is taken between the collectors of the two transistors as in Figure 11.4, then
the output voltage will be proportional to the dierence of the emitter currents:
i
C1
i
C2
= I
EE
e
vi/2V
T
e
vi/2V
T
e
vi/2V
T
+e
vi/2V
T
(11.11)
i
C
= I
EE
tanh
v
i
2V
T
A plot of the current dierence, i
C
= i
C1
i
C2
, is shown in Figure 11.3. If the input
-V
T
V
T
V
i
Figure 11.3: Current dierence
signal levels are small, i.e., if
vi
2V
T
1, then the hyperbolic tangent can be approximated as
follows:
tanh
v
i
2V
T
'
v
i
2V
T
(11.12)
Thus
i
C
' I
EE
v
i
2V
T
(11.13)
Note that I
EE
multiplies the input signal in equation 11.13. If a second signal, i
i2
, is
added to I
EE
then the output will contain a term that is proportional to v
i
i
i2
, i.e. let
I
EE
!I
EE
+i
i2
, the the output is
i
C
'
2V
T
(I
EE
+i
i2
)v
i
(11.14)
Figure 11.4 shows a balanced mixer based on the dierential amplier. The output voltage
is v
o
= i
c
R. The tail current for the dierential pair is I
EE
+i
i2
, where i
i2
represents the
time-varying current that results from the input voltage signal v
i2
.
According to equation 11.14, the multiplier based on the dierential pair yields an output
that is proportional to (K + v
i
)v
i2
. Hence the output contains the desired product term
plus the input signal v
i2
. Suppose that a balanced multiplier of the type shown in Figure
11.4 is used with the RF signal driving the dierential-amplier input (v
i
! v
RF
) and the
LO signal driving the tail-current source (v
i2
! v
LO
). In that case the output will be
proportional to v
RF
(K + v
LO
) = Kv
RF
+ v
RF
v
LO
. Notice that the RF signal is present
in the output, along with the desired product signal. On the other hand, if v
i
! v
LO
and
v
i2
! v
RF
, then the output is proportional to v
LO
(K + v
LO
) = Kv
LO
+ v
RF
, and the LO
signal accompanies the desired product signal.
The unwanted term can be eliminated as followssuppose that we construct two iden-
tical multipliers, and drive the dierential inputs of each amplier with v
i
. Drive the tail-
current input of amplier 1 with v
i2
and the tail-current amplier of multiplier 2 with v
i2
.
11.2. MIXERS BASED ON SWITCHES 359
i
C1
i
C2
I
EE
+ i
i2
v
i1
v
i2
v
o
R R
V
CC
V
EE
Figure 11.4: Dierential-pair multiplier circuit.
The output of multiplier 1 is then proportional to Kv
i2
+v
i
v
i2
, whereas the output of multi-
plier 2 is proportional to Kv
i2
v
i
v
i2
. The dierence between the two outputs will contain
only a term proportional to v
i
v
i2
because the unwanted term will be canceled out. Multi-
pliers based on this principle are called Gilbert cell multipliers, and are named after Barrie
Gilbert who described a practical, and precise, multiplier circuit based on the technique.
1
Notice that in order to obtain an output that contained only the desired product term, and
no RF or LO term, it is necessary to use a double-balanced conguration, where both the
LO and RF inputs are applied dierentially.
11.2 Mixers Based on Switches
Most mixers used today are based on switches. The basic idea can be illustrated using a
very simple circuit as in Figure 11.5. Suppose that the switch is closed when v
LO
> 0 and
open when v
LO
< 0. In addition, let v
RF
(t) = V
RF
cos !
RF
t and v
LO
(t) = V
LO
cos !
LO
t,
then
v
o
(t) = V
RF
cos !
RF
t p(t) (11.15)
where p(t) is a switching function. The switching function for this circuit is shown in Figure
11.6.
The switching function can be expanded in a Fourier series:
p(t) =
1
2
+
1
n=1
sin n/2
n/2
cos n!
LO
t (11.16)
1
A Precise Four-Quadrant Multiplier with Subnanosecond Response, by Barrie Gilbert, IEEE Journal
of Solid-State Circuits, Vol. SC-3, No. 4, December 1968, p 265.
360 CHAPTER 11. MIXERS
R
L
v
o
(t) v
RF
(t)
Controlled by v
LO
Figure 11.5: Switching principle
p(t)
t
2/!
LO
1
Figure 11.6: Switching function.
Using the Fourier series respresentation for p(t) the output voltage can be written as
v
o
(t) = V
RF
cos !
RF
t
1
2
+
1
n=1
sin n/2
n/2
cos n!
LO
t
(11.17)
Notice that the coecient
sin n/2
n/2
is equal to 0 when n is an even number. Therefore, the
output signal has components at the following frequencies:
f
RF
, |f
RF
n f
LO
| n = 1, 3, 5, . . . (11.18)
A lter can be used to select the desired component which is usually |f
RF
f
LO
|. It is also
possible to select |f
RF
n3f
LO
| or higher order terms - this is called harmonic mixing.
There are many ways to implement a switch that is controlled by the local oscillator
signal. Diodes are often used as switches. If a relatively large LO voltage is impressed on
a diode, the diode will be forward biased on positive excursions of the LO signal, and will
be in a low-impedance state for superposed, small RF signals. On negative excursions of
the LO voltage the diode is driven into a high impedance state and behaves like an open
switch. Transistors can also be used as switches, with the LO controlling the bias current
and determining whether the device is biased into a low-impedance state, or is cuto.
The dierential multiplier circuit described in the previous section is often used to im-
plement a switching mixer. Refer back to Figure 11.4 and suppose that v
i
! v
LO
. Allow
the amplitude of v
LO
to be large compared to V
T
; then the tanh
v
LO
2V
T
term will be driven
to +1 on positive excursions of v
LO
and to 1 on the negative excursions. In this mode
of operation, the transistors are driven as switches, as shown in Figure 11.7. The output
voltage for the switch conguration shown is v
o
= (I
DC
+ i
RF
)R. When v
LO
changes
sign, the states of the switches reverse, and the output voltage is v
o
= +(I
DC
+i
RF
)R. The
output can be written as
v
o
(t) = (I
DC
+i
RF
) R p(t)
11.2. MIXERS BASED ON SWITCHES 361
i
1
i
2
Closed when
v
LO
> 0
v
o
R R
V
CC
I
DC
+ i
RF
+ -
Open when
v
LO
> 0
Figure 11.7: Equivalent circuit for dierential pair when a large LO signal is applied to the
dierential amplier.
where p(t) is the symmetrical switching function shown in Figure 11.8.
p(t)
t
2/!
LO
1
1
Figure 11.8: Symmetrical switching function
This switching function has the following Fourier series:
p(t) = 2
1
n=1
sin n/2
n/2
cos n!
LO
t. (11.19)
Writing i
RF
(t) = I
RF
cos !
RF
t:
v
o
(t) = (I
DC
+I
RF
cos !
RF
t)
2
1
n=1
sin n/2
n/2
cos n!
LO
t
(11.20)
The frequency components at the output will be
nf
LO
, |f
RF
nf
LO
| n = 1, 3, 5, . . . (11.21)
Two dierential switching pairs can be combined in a double-balanced Gilbert cell ar-
rangement to cancel the unwanted LO component. This is illustrated in Figure 11.9.
362 CHAPTER 11. MIXERS
i
1
v
o R
V
CC
I
DC
+ i
RF
+ -
i
2
R
I
DC
i
RF
Figure 11.9: Two dierential switching pairs combined in a double-balanced conguration.
Suppose that the switches are in the state shown when v
LO
> 0 and that all switches change
state when v
LO
< 0. Then, the currents are i
1
= I
DC
+ p(t)i
RF
and i
2
= I
DC
p(t)i
RF
.
The output voltage is v
o
= R(i
2
i
1
) = 2Ri
RF
(t)p(t), hence the output contains only
components at the frequencies |f
RF
nf
LO
|. Mixers of this type are often implemented in
integrated circuits.
The diode-ring double-balanced mixer shown in Figure 11.10a, b is another circuit that
produces an output signal proportional to the product of the RF input signal and a switch-
ing function, i.e. v
RF
(t)p(t). This mixer is noted for its ability to tolerate large RF signals
with relatively small nonlinear distortion, and for its relatively low noise gure. Its opera-
(a)
v
LO
v
RF
v
IF
1
2
3
4
a
b
T2 T1
(b)
v
LO
v
RF
v
IF
1
2 3
4
a b
T2
T1
c
d
Figure 11.10: Four-diode double-balanced mixer. Figures (a) and (b) are the same circuit.
Figure (b) emphasizes the fact that the mixer is based on a diode ring topology.
tion is easily understood once the operation of the 3-winding transformers (voltage baluns)
is understood. For simplicity, lets assume that the transformers are ideal 3-winding trans-
11.2. MIXERS BASED ON SWITCHES 363
formers. If the voltage across the windings is denoted by V
A
, V
B
, and V
C
, with polarity
indicated by the dots, and the current in each winding is denoted by i
A
, i
B
, and i
C
, with
positive current dened to ow into the dot, then an ideal 3-winding transformer satises
V
A
= V
B
= V
B
, i
A
+i
B
+i
C
= 0.
Physically, these relationships arise because when three windings with the same number of
turns are tightly coupled, e.g. by winding all three on a high-permability ferrite toroid,
then the magnetic ux through all 3 windings will be the same and the emf induced in each
winding will be the same. Since the ideal transformer is a lossless device, power conservation
requires that the sum of the current owing out of (or into) the dots of all windings must
be zero. In the context of the circuit described here, if current i ows into the dot of the
primary (driven) winding, the sum of the currents owing out of the dots of the secondary
windings must be i.
We can now describe the operation of the 4-diode ring mixer. The three-winding trans-
formers are assumed to be ideal, hence the voltage impressed across the primary will be
reected across both secondary windings, with polarity indicated by the dots. Refering to
Figure 11.10b, when v
LO
is positive, the voltage at the top of T2s secondary is positive
and the voltage at the bottom of T2s secondary is negative. Therefore diodes 1 and 2 are
on and in a low-impedance state as far as the small RF signal is concerned. Diodes 3 and
4 will be o, or in a high-impedance state, so the right-hand side winding of transformer
T1s secondary is eectively disconnected from the system. Refer to Figure 11.11a where the
forward biased diodes have been replaced with an incremental resistance, r
d
, and the reverse
biased diodes have been removed, as they are assumed to be eectively open circuits. The
symmetry of the circuit ensures that the voltage v
RF
across the left-hand winding of T1s
secondary will drive equal currents through diodes 1 and 2. These currents ow to ground
through the windings of T2
0
s secondary. Half of the RF current ows into the dot on the
top of T2s secondary and the other half ows into the bottom of T2s secondary. These
currents produce opposing magnetic uxes and no voltage is developed across the windings
of T2 secondary. Therefore, the RF voltage at nodes d and c is zero, and nodes c and d are
virtual grounds for the RF signal. The RF voltage applied by the left-hand winding of T1s
secondary appears across the series combination of the load resistance R
L
and the parallel
combination of diodes 1 and 2 (r
d
/2). The voltage across the load is v
IF
= v
RF
R
L
R
L
+r
d
/2
.
When the polarity of v
LO
reverses so that the voltage at the dots of T2s secondary is nega-
tive, the situation reverses, i.e. diodes 1 and 2 are o and diodes 3 and 4 are on. Nodes
c and d are still virtual grounds for the RF signal, so the voltage across the load resistor is
v
IF
= v
RF
R
L
R
L
+r
d
/2
. If the forward resistance of the diodes is small compared to the load
resistor (r
d
R
L
), then the output from the double-balanced 4-diode ring mixer can be
written as
v
IF
(t) = v
RF
(t)p(t)
where p(t) is the symmetrical switching function shown in Figure 11.8. Since the switching
function has no DC component, there is no direct leakage of the RF signal into the IF
output. Also, by virtue of the balance in the circuit, the voltage at nodes a and b due to the
LO signal is always zero so there is no leakage of the LO signal back into T1, i.e. LO-RF
isolation is high.
For diode-based switching-type mixers to work as described above, it is important that
the local oscillator signal (v
LO
) be large, so that the diodes will be biased to a low impedance
state, and so that it can be assumed that v
LO
controls the state of the diodes at all times.
364 CHAPTER 11. MIXERS
(a)
v
LO
> 0
v
RF
1
2
a
T2
T1
c
d
virtual ground
for RF signal
R
L
r
d
v
IF
= v
RF
R
L
R
L
+r
d
/2
v
RF
+
+
+
-
(b)
v
LO
< 0
v
RF
3
4
b
T2
T1
c
d
virtual ground
for RF signal
R
L
r
d
v
IF
= v
RF
R
L
R
L
+r
d
/2
v
RF +
+
+ -
Figure 11.11: 4-diode ring double-balanced mixer equivalent circuit when (a) v
LO
> 0 and
(b) v
LO
< 0.
11.3. CONVERSION LOSS IN MIXERS 365
Thus passive mixers require relatively high LO drive levels. Typically, the peak value of the
LO signal must be at least 0.7V in order to turn on the diodes completely. Assuming that
the input impedance seen by the LO source is 50 , the required input level is approximately
7 dBm. The harmonic mixing components, |mf
RF
nf
LO
|, can cause spurious responses
in a receiver in addition to the usual image response. These responses must be carefully
considered when designing a receiver. This will be addressed in section 11.4.
11.3 Conversion Loss in Mixers
Conversion loss is dened to be
L
C
= 10 log
10
P
in
P
out
where P
in
is the power delivered in to the mixer at the RF frequency, and P
out
is IF output
power delived to the load at the desired output frequency (usually |f
RF
f
LO
|). Lets
take the double-balanced diode-ring mixer as an example. Assuming that the diode forward
resistance is small compared to the load resistance, the output of the mixer is
v
IF
(t) = V
RF
cos !
RF
t
2
1
n=1
sin
n
2
n
2
cos n!
LO
t
.
If a lter selects just one output component, then the peak amplitude of the output voltage
will be
V
IF
= V
RF
sin
n
2
n
2
.
In an ideal 4-diode mixer, the RF input is always looking through the transformer primary
to one of the secondary windings and through the switches to the load. If the switches are
lossless, then, the RF input impedance will be equal to the load impedance. Hence, the RF
input voltage and the IF output voltage are developed across the same impedance, so the
ratio of P
in
and P
out
is the same as the ratio (V
RF
/V
IF
)
2
. The conversion loss, for any n,
is therefore
L
C
= 20 log
10
n
2 sin
n
2
.
The conversion loss is tabulated in Table 11.1. When the desired term is |f
RF
f
LO
|, the
conversion loss for the ideal switching mixer is 3.9 dB. In practice, losses in the transformer
and diodes usually raise this value to 5-6 dB.
n Ouput Frequency Loss (dB)
1 |f
RF
f
LO
| 3.9
3 |f
RF
3f
LO
| 13.5
5 |f
RF
5f
LO
| 17.9
Table 11.1: Conversion loss for ideal double-balanced diode-ring mixer.
366 CHAPTER 11. MIXERS
11.4 Spurious Responses in Receivers - Spur Charts
The idealized models for double-balanced switching mixers predict that the output will
contain components at frequencies given by |f
RF
nf
LO
|, with n an odd integer. When
more realistic models for the switches are considered, it is found that harmonics of the RF
signal will be generated within the mixer, and that even-order harmonics of the LO will
be present. A fairly general model for the nonlinear properties of a mixer consists of a
nonlinear 2-port with non-zero coecients k
1
, k
2
, k
3
, ... in front of an ideal multiplier which
is driven by a local oscillator signal containing the fundamental frequency component !
LO
in addition to harmonics of the local oscillator signal, 2!
LO
, 3!
LO
, .... etc.
v
RF
(t)
Nonlinear
2-port
|mf
RF
nf
LO
|
P
1
m=1
k
m
v
m
RF
(t)
P
1
n=0
a
n
cos[n!
LO
t]
Figure 11.12: A model for nonlinearities in a mixer. The multiplier shown in the gure is an
ideal multiplier. The output of the mixer will include terms resulting from mixing between
all harmonics of the RF and LO signals. Note that the n = 0 (DC) component is included
in the output of the LO in order to model the direct leakage of harmonics of V
RF
through
the mixer to the IF port.
When evaluating a particular receiver design, it is highly desirable to consider the possible
spurious responses which will occur because of the non-ideal characteristics of the mixer.
The goal of this section is to determine the frequencies of all possible input signals which
could be mixed to the IF.
The receiver will be designed such that the desired signal frequency, f
D
, is related to the
IF, f
IF
, by one of the following relationships:
f
LO
= f
D
+f
IF
(11.22)
or
f
LO
= f
IF
f
D
(11.23)
or
f
LO
= f
D
f
IF
(11.24)
Equation 11.22 corresponds to the choice of high LO for either the up- or down-
conversion receivers. Equation 11.23 corresponds to low LO for the up-conversion receiver,
and equation 11.24 corresponds to low LO for the down-conversion receiver.
A spurious signal that mixes to f
IF
satises the following equation:
|mf
S
nf
LO
| = f
IF
(11.25)
where we assume that m, n are integers; m > 0, n 0. Notice that by allowing n = 0 we
allow for possible IF responses due to direct leakage of the RF signal (or its harmonics) into
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 367
the IF passband. Taking account of the absolute value sign, equation 11.25 can be expanded
into 3 equations:
mf
S
+nf
LO
= f
IF
mf
S
nf
LO
= f
IF
mf
S
nf
LO
= f
IF
(11.26)
Now, depending on the receiver conguration of interest, one of equations 11.22, 11.23, or
11.24 may be used in equations 11.26 to eliminate f
LO
and to solve for the frequency of
possible spurious signals as a function of the receiver tuning, represented by the desired
signal frequency f
D
. Notice that in this context, f
D
may be interpreted as the receivers
dial setting since it is the number that the receiver display will indicate. In some literature
f
D
is referred to as the tuned frequency, since it is the intended frequency that the receiver
is tuned to receive.
Suppose that the case of interest is high LO. Then, using equation 11.22 in 11.26 we
nd:
mf
S
= (1 n)f
IF
nf
D
mf
S
= (1 +n)f
IF
+nf
D
mf
S
= (n 1)f
IF
+nf
D
(11.27)
Since the right-hand side of the rst and third equations in 11.27 are the same except for a
factor of -1, and we are interested in positive spurious frequencies, the rst equation is not
necessary and the relevant equations are:
mf
S
= (n 1)f
IF
+nf
D
mf
S
= (n + 1)f
IF
+nf
D
(11.28)
It is convenient to normalize all frequencies by the IF, i.e. dene:
S
f
S
f
IF
, and D
f
D
f
IF
then we obtain the equations used to generate a so-called universal spur chart for any
receiver employing high LO:
S =
n1
m
+
n
m
D
S =
n+1
m
+
n
m
D
(11.29)
For each value of m and n these equations dene two lines, each of which determines the
normalized spur frequency, S, for a particular normalized tuning frequency D. The case
m = 1, n = 1 gives the lines S = D and S = D + 2 which correspond to the desired
frequency and the regular image, respectively. These responses were discussed already in
Chapter 3. The other lines correspond to spurious responses that result from the non-ideal
nature of the mixer.
The normalized spur frequency is plotted in Figure 11.13 for all possibilities up through
fth order where the order of a spur is dened to be the value of n+m. The m, n values
for each curve are plotted on each line. The part of the plot where 0 < D < 1 corresponds
368 CHAPTER 11. MIXERS
Figure 11.13: Universal spur chart for high LO. The numbers on each line (m, n) indicate
the multiple of the RF and LO frequencies, respectively that are summed or dierenced to
produce an output at the IF. Only responses up through 5th order (m+n 5) are shown.
Normalized tuning frequencies greater than 1.0 correspond to downconversion and tuning
frequencies less than 1.0 correspond to upconversion.
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 369
to tuning frequencies that are smaller than the IF, (D = f
D
/f
IF
< 1) and, hence, to
up-conversion. The part of the plot where D > 1 corresponds to down-conversion.
Universal spur charts may also be produced for up-conversion with low LO and down-
conversion with low-LO. It turns out that the following normalized equations can be used
for both of these cases. When D < 1 the result corresponds to up-conversion/low-LO and
with D > 1 the result corresponds to down-conversion/low-LO:
S = |
1n
m
+
n
m
D|
S = |
1+n
m
n
m
D|
(11.30)
The universal spur chart for low-LO is shown in Figure 11.14.
Figure 11.14: Universal spur chart for low LO. Only responses up through 5th order are
shown. The numbers on each line (m, n) indicate the multiple of the RF and LO frequencies,
respectively that are summed or dierenced to produce an output at the IF. Normalized
tuning frequencies above 1.0 correspond to downconversion, and tuning frequencies below
1.0 correspond to upconversion.
370 CHAPTER 11. MIXERS
11.4.1 Crossovers
Notice that both of the universal spur charts exhibit crossovers, i.e. certain lines cross the
m=1, n=1 line which corresponds to S = D. A crossover at some tuned frequency D means
that a signal with frequency S = D will interfere with itself! For example notice that a
crossover occurs in Figure 11.13 when S = D = 3. In this case, the line that crosses the
m = 1, n = 1 line at S = D = 3 has m = 3, n = 2. This means that the 3 harmonic of the
desired signal at frequency S = 3 will mix with the second harmonic of the local oscillator
to give an output at the IF. A numerical example will be provided in the next section. For
now, it is sucient to point out that a crossover frequency represents a potential dead
zone in a receivers tuning range, since it may be impossible to receive strong signals at
this frequency.
11.4.2 Example - AM Broadcast band radio
Figure 11.15 shows a spur chart for a broadcast band AM radio employing high LO. For an
IF of 455 kHz, the normalized lower and upper frequencies of the AM broadcast band are
[540/455, 1700/455]=[1.187,3.736]. These limits are denoted in the gure by solid vertical
and horizontal lines.
11.4.2.1 Radio tuned to receive a signal at 910 kHz:
Suppose the radio is tuned to receive a desired signal with carrier frequency 910 kHz
(D=910/455=2.0). Consider possible interference from signals within the AM band re-
sulting from spurious responses up through 5th order (order = n+m). Examine the vertical
dotted line drawn at the normalized tuned frequency D=2 and notice that for spur frequen-
cies within the AM band there are 6 intersections with the dotted line (not counting the
intersection with the desired n=m=1 line). This means there are 6 possible frequencies that
could cause interference. If spurious response orders higher than 5 are considered, there
would be an even larger number of potential spurious responses.
Suppose it is necessary to determine the exact frequencies of each of the potential spurs.
It is not possible to obtain very accurate results using the chart, but the chart can be used
to provide guidance for determining the numerical values of the spur frequencies. Notice
that (m,n) for the 6 spurs are: (3,1), (3,2), (2,1), (3,2), (2,2), (2,2). Also notice that the
(2,1) spur is a crossover spur. To determine the exact frequencies recall that spurs satisfy:
|mf
S
nf
LO
| = f
IF
which corresponds to the three equations
mf
S
+nf
LO
= f
IF
(11.31)
mf
S
nf
LO
= f
IF
(11.32)
mf
S
+nf
LO
= f
IF
(11.33)
Since the AM radio uses high LO, equation 11.31 can never be satised and is not relevant
here. The other two equations can be re-arranged to yield:
f
S
=
f
IF
+nf
LO
m
(11.34)
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 371
Figure 11.15: A portion of the universal spur chart for high-LO relevant to an AM Broadcast-
band radio with 455 kHz IF and high LO.
372 CHAPTER 11. MIXERS
f
S
=
nf
LO
f
IF
m
(11.35)
Now, use the fact that when the radio is tuned to 910 kHz, f
LO
= 910+455 = 1365 kHz
to nd the exact frequency of each spur, i.e.:
(3, 1) ) f
S
=
455 + 1365
3
= 606.67 kHz
(3, 2) ) f
S
=
455 + 2(1365)
3
= 1061.67 kHz
(3, 2) ) f
S
=
2(1365) 455
3
= 758.33 kHz
(2, 2) ) f
S
=
455 + 2(1365)
2
= 1592.50 kHz
(2, 2) ) f
S
=
2(1365) 455
2
= 1137.5 kHz
(2, 1) ) f
S
=
455 + 1365
2
= 910 kHz
Notice that the (2,1) spur is a crossover spur, i.e. the spur frequency is equal to the desired
frequency (910 kHz in this case). The interference results from the second harmonic of the
910 kHz signal (at 1820 kHz) mixing with the LO at 1365 kHz to produce an output from
the mixer at 455 kHz. This self-interference would be expected to be most noticeable when
the desired signal at 910 kHz is a very strong signal. It has the somewhat peculiar property
that the stronger the desired signal, the more intense the interference will be, since the
second harmonic will be generated most eciently when the mixer is driven hard.
11.4.2.2 Strong signal at 1000 kHz
Next, suppose that a very strong signal is transmitting at 1000 kHz. As the receiver is
tuned across the band, numerous spurious responses may occur. In this case, the strong
signal at 1000 kHz (S=1000/455=2.1978) is represented by a horizontal dotted line. Notice
that there are 4 possible tuned frequencies where spurs from the signal at 1000 kHz would
be potentially observed. (The intersection between the horizontal dotted line and the (1,1)
line is not counted, since that represents the case where the receiver is tuned to receive 1000
kHz.) The 4 spurious responses correspond to (2,2), (3,2), (2,1), (3,2). Equations 11.32
and 11.33 can be used along with the fact that f
D
= f
LO
f
IF
to solve for the tuning dial
setting at which the spurious responses would occur:
f
D
=
m
n
f
S
f
IF
(
1
n
+ 1) (11.36)
and
f
D
=
m
n
f
S
+f
IF
(
1
n
1). (11.37)
Using f
S
= 1000 kHz and f
IF
= 455 kHz :
(2, 2) ) f
D
= 1000 + 455(
1
2
) = 772.5 kHz
(3, 2) ) f
D
=
3
2
1000 455(
3
2
) = 817.5 kHz
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 373
(3, 2) ) f
D
=
3
2
1000 + 455(
1
2
) = 1272.5 kHz
(2, 1) ) f
D
= 2(1000) 455(2) = 1090 kHz
Thus, spurious responses would potentially be observed at the following dial settings: 772.5
kHz, 817.5 kHz, 1090 kHz, and 1272.5 kHz.
374 CHAPTER 11. MIXERS
11.5 Homework Problems
1. Consider the mixer circuit shown in Figure 11.16:
R
S
R
L
v
LO
(t)
v
RF
(t)
v
o
(t)
+
1
1
h()v
i
(t )d (12.1)
where h() is the impulse response of the system. When the bandwidth of v
i
(t) is much
smaller than the bandwidth of the 2-port, then the impulse response of the 2-port can be
approximated as a delayed delta function, i.e. h(t) ' k(t
g
) where k is the small-signal
voltage gain and
g
is the group-delay of the 2-port. In this case the output waveform will
be a scaled and delayed version of the input signal, i.e.
v
o
(t) ' kv
i
(t
g
).
Likewise, if v
i
(t) = s
1
(t) +s
2
(t), then
v
o
(t) = ks
1
(t
g
) +ks
2
(t
g
) (12.2)
i.e., superposition applies.
In general, for input signals that are not small, the input-output relationship must be
considered to be nonlinear. To illustrate some of the important characteristics of nonlinear
2-ports while keeping the analysis relatively simple, we will ignore the possibility of energy
storage elements (such as inductances and capacitances) within the nonlinear 2-port. This
means we can ignore frequency-dependent phase shifts and, therefore, delays. Such a 2-port
will be frequency independent and is said to be memoryless because lack of energy storage
means that the output can depend only on the present value of the input, and not on past
375
376 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
values. In many cases, this assumption can be justied by lumping energy storage elements
into external, linear, networks. If the 2-port is time-invariant and memoryless, then the
output signal will be related to the input signal by an nonlinear transfer characteristic, i.e.
v
o
= f[v
i
], where f[ ] is a single-valued nonlinear function. Typically, the nonlinear input-
output characteristic is approximated by a Taylor series in the vicinity of an operating
point, and analysis proceeds based on the truncated Taylor series expansion. For input
signals that are not too large, only a few terms of the Taylor series are sucient. The
Taylor series approach can be viewed as a special case of the more general Volterra series
analysis which is applicable to nonlinear 2-ports with memory.
Denote the nonlinear input-output relationship for a memoryless 2-port as
v
o
(t) = f[v
i
(t)], (12.3)
where f[ ] is a single-valued, smooth (dierentiable), nonlinear function. One consequence
of the nonlinear input-output relationship for a memoryless 2-port is that superposition
cannot be applied to determine the output in response to multiple inputs. In this case, if
v
i
(t) = s
1
(t) +s
2
(t), then
v
o
(t) = f[s
1
(t) +s
2
(t)] (12.4)
6= f[s
1
(t)] +f[s
2
(t)] (12.5)
i.e., superposition does not apply. As we shall see, when the input signal contains only
one, or a few, sinusoidal components, the output signal may contain many more frequency
components, generated because of the nonlinearity of the 2-port. When generated within
receivers these distortion products can cause interference. In transmitters, nonlinearity in
the power amplier can cause signicant broadening of the signal spectrum, possibly causing
the output spectrum to bleed over into adjacent channels.
12.1 Power series model
For moderately large signals, a memoryless nonlinear input-output characteristic can be
expanded in a Taylor series about some operating, or quiescent, point. Denote the total
input and output signals by x
T
and y
T
, respectively so that y
T
= f[x
T
]. The input signal
can be decomposed into quiescent, and time-varying components, i.e.
x
T
= x
Q
+x (12.6)
where x
Q
is the DC component of the input signal and x is a zero-mean, time-varying
component. Similarly, denote the total output by y
T
where
y
T
= y
Q
+y. (12.7)
The rst term, y
Q
, is dened to be the output when x = 0. Then
y
T
= f[x
T
] (12.8)
= f[x
Q
] +x
df
dx
x=x
Q
+
1
2
x
2
d
2
f
dx
2
x=x
Q
+. . . +
1
n!
x
n
d
n
f
dx
n
x=x
Q
+ . . .
12.1. POWER SERIES MODEL 377
The part of the output that results from the time-varying part of the input is, therefore, of
the form
y = k
1
x +k
2
x
2
+k
3
x
3
+. . . (12.9)
where
k
i
=
d
i
f
dx
i
x=xq
So-called small-signal models retain only the rst term in equation 12.9, which represents a
linear relationship between x and y. This term is dominant when x is small enough so that
higher order terms are negligible. The higher order terms represent nonlinear distortion.
Also, notice that even when x has zero mean, y may not have zero mean since terms which
are even powers of the input signal will have non-zero means; for example, if the input is a
cosine function x = cos !t, the term x
2
= cos
2
!t =
1
2
(1 +cos 2!t) is not zero-mean. Hence,
the constant part of the output may contain terms contributed by the zero-mean input
signal, x. This means that the DC bias point of an amplier will shift when the amplier is
driven with a time-varying input signal.
To obtain a general understanding of nonlinear eects in 2-ports (without considering
particular circuit congurations), we will assume that the input-output relationship can be
accurately modeled by a power-series expansion
y = k
1
x +k
2
x
2
+k
3
x
3
+. . . (12.10)
where x represents the input excitation and y is the output quantity. Typically, x and y
will represent input and output voltages or currents. For signals that are not too large
it is sucient to truncate the expansion to terms of third order and lower. Thus for the
subsequent discussion well assume an input-output relationship consisting of a three-term
power series:
y ' k
1
x +k
2
x
2
+k
3
x
3
(12.11)
12.1.1 Specic Examples - BJT and FET nonlinearities
As an example, consider the input-output relationship for the bipolar junction transistor
(BJT) in Figure 12.2.
I
C
I
E
+
-
B
E
C
V
be
I
B
Figure 12.2: Bipolar junction transistor (BJT).
I
C
= I
S
e
V
be
/V
T
(12.12)
and V
T
=
kT
q
' 25 mV at room temperature (290 K). Now suppose that the base-emitter
voltage consists of a DC component (V
DC
) and a time-varying signal component (v
be
), i.e.,
V
be
= V
DC
+v
be
(12.13)
378 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
Then
I
C
= I
S
e
V
DC
/V
T
e
v
be
/V
T
(12.14)
If v
be
/V
T
< 1, we can expand e
v
be
/V
T
:
I
C
= I
S
e
V
DC
/V
T
1 +
v
be
V
T
+
1
2
v
be
V
T
2
+
1
6
v
be
V
T
3
+. . .
(12.15)
The rst term represents the DC component of collector current when v
be
= 0. Denote this
quiescent current by I
CQ
, i.e.:
I
CQ
= I
S
e
V
DC
/V
T
.
Denote the uctuating component of the collector current by i
c
, then a three-term power-
series approximation for the nonlinear relationship between i
C
and v
be
is:
i
C
=
I
CQ
V
T
v
be
+
I
CQ
2V
2
T
v
2
be
+
I
CQ
6V
3
T
v
3
be
Hence, for the BJT:
k
1
=
I
CQ
V
T
= g
m
k
2
=
I
CQ
2V
2
T
k
3
=
I
CQ
6V
3
T
(12.16)
The rst term is the familiar small-signal approximation; the time-varying part of the col-
lector current is equal to the transconductance, g
m
, times the time-varying part of the
base-emitter voltage. In this chapter we will be concerned with the implications of the
higher order terms in the input-output characteristic.
It is interesting to compare the BJT input-output characteristic with that for an ideal
eld eect transistor (FET) as in Figure 12.3.
I
D
+
-
G
S
D
V
GS
Figure 12.3: Field-eect transistor (FET).
I
D
= I
DSS
1
V
GS
V
P
2
(12.17)
= I
DSS
1
2
V
P
V
GS
+
1
V
2
P
V
2
GS
(12.18)
There are no third-order or higher terms in this idealized square-law FET characteristic. In
a more realistic model for the FET such terms would be present.
12.2. SINGLE-TONE INPUT 379
12.2 Single-tone Input
Consider a nonlinear amplier modeled with a three term power-series:
v
o
= k
1
v
i
+k
2
v
2
i
+k
3
v
3
i
Suppose the input signal consists of a single tone
v
i
(t) = a
1
cos !
1
t (12.19)
Then
v
o
(t) = k
1
a
1
cos !
1
t +k
2
a
2
1
cos
2
!
1
t +k
3
a
3
1
Acos
3
!
1
t (12.20)
Using trigonometric identities
cos
2
!t =
1
2
(1 + cos 2!t) (12.21)
cos
3
!t =
1
4
(3 cos !t + cos 3!t)
The output is written
v
o
(t) =
1
2
k
2
a
2
1
+ (k
1
a
1
+
3
4
k
3
a
3
1
) cos !
1
t+
1
2
k
2
a
2
1
cos 2!
1
t +
1
4
k
3
a
3
1
cos 3!
1
t
(12.22)
and the input and output spectra are illustrated in Figure 12.4. Since the amplitude of the
second and third harmonics is proportional to a
2
1
and a
3
1
, respectively, this analysis predicts
that for every 1 dB increase of the input signal, the second and third harmonics will increase
by 2 dB and 3 dB, respectively.
f
1
Input Spectrum Output Spectrum
2f
1
3f
1
f
1
Figure 12.4: Input and output spectra for single input signal.
12.2.1 Gain Compression
Now consider the output component at the fundamental frequency:
v
0
o
(t) = (k
1
a
1
+
3
4
k
3
a
3
1
) cos !
1
t
= k
1
(1 +
3
4
k3
k1
a
2
1
)a
1
cos !
1
t
(12.23)
380 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
The eective voltage gain of the amplier is
A
v
= k
1
(1 +
3
4
a
2
1
k
3
k
1
) (12.24)
The term in parenthesis arises because the amplier exhibits a third-order nonlinearity and
its magnitude depends on the input signal amplitude. Depending on the sign of k
3
, as
the input signal amplitude increases the eective gain of the amplier will either increase
(expansive nonlinearity) or decrease (compressive nonlinearity). If k
1
and k
3
have the same
sign, then the gain will increase with increasing input amplitude, and if k
1
and k
3
have
opposite signs, the gain decreases with increasing input amplitude. Our analysis of the BJT
in section 12.1.1 showed that k
1
and k
3
have the same sign, and hence the nonlinearity is
expansive. The analysis in section 12.1.1 is based on the assumption that the DC component
of V
be
is held constant. In practical amplier circuits, the transistor will be biased such that
the DC component of the collector current is held more-or-less constant (constant-current
bias), and the DC component of V
be
is allowed to adjust in order to maintain constant
DC collector current. Analysis based on this constraint will show that such an amplier
exhibits compressive nonlinearity. A detailed analysis of this type is carried out in Appendix
A, where the large signal transconductance is derived and shown to decrease with increasing
base-emitter voltage swing. For now, we shall simply state that in most cases k
1
and k
3
will
have opposite signs, so that the nonlinearity is compressive. For compressive nonlinearity,
the eective gain can be written as
k
1
1
3
4
a
2
1
k
3
k
1
The reduction in gain with increasing input signal amplitude caused by compressive
nonlinearity is called gain compression. The plot in Figure 12.5 shows the output power in
the fundamental component as a function of the input power for an amplier that exhibits
gain compression.
The gain compression of a 2-port is often characterized by the input power level that
causes the gain to be decreased by 1dB. This parameter is denoted by P
1dB
in Figure 12.5.
12.3 Two-tone Input
Now suppose that the input signal includes two tones:
v
i
(t) = a
1
cos !
1
t +a
2
cos !
2
t (12.25)
Then
v
o
(t) = k
1
[a
1
cos !
1
t +a
2
cos !
2
t] (12.26)
+k
2
[a
1
cos !
1
t +a
2
cos !
2
t]
2
+k
3
[a
1
cos !
1
t +a
2
cos !
2
t]
3
Using trigonometric identities the output can be re-written as
v
o
(t) = rst order terms + second order terms + third order terms
12.3. TWO-TONE INPUT 381
Linear
P
1dB
Nonlinear
P
out
(dBm)
P
in
(dBm)
1 dB
Figure 12.5: Output power in fundamental component as a function of input power.
where
rst order terms = k
1
[a
1
cos !
1
t +a
2
cos !
2
t]
second order terms = k
2
[
1
2
(a
2
1
+a
2
2
) +
1
2
a
2
1
cos 2!
1
t +
1
2
a
2
2
cos 2!
2
t
+a
1
a
2
cos(!
1
+ !
2
)t +a
1
a
2
cos(!
1
!
2
)t]
third order terms = k
3
[(
3
4
a
3
1
+
3
2
a
1
a
2
2
) cos !
1
t + (
3
4
a
3
2
+
3
2
a
2
1
a
2
) cos !
2
t
+
1
4
a
3
1
cos 3!
1
t +
1
4
a
3
2
cos 3!
2
t
+
3
4
a
1
a
2
2
(cos(2!
2
!
1
)t + cos(2!
2
+ !
1
)t)
+
3
4
a
2
1
a
2
(cos(2!
1
!
2
)t + cos(2!
1
+ !
2
)t)]
The input and output spectra are illustrated in Figure 12.6.
In narrow-band systems most of these frequency components will be removed by lters
downstream from the nonlinear 2-port, but the so-called in-band third-order terms (2f
2
f
1
,
2f
1
f
2
) cannot be ignored since they will always have frequencies close to that of f
1
and
f
2
. In wide-band systems, all of the terms can potentially be signicant at the output. In
receivers that use a low-IF, the second-order term at the dierence frequency f
2
f
1
may be
382 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
f
1
Input Spectrum
Output Spectrum
2f
2
3f
1
f
1
f
2
f
2
3f
2
2f
1
+ f
2
2f
2
+ f
1
2f
1
f
1
+ f
2
f
2
f
1
2f
1
f
2
2f
2
f
1
third
order
second
order
second
order
rst order
and third order
Figure 12.6: Input (top) and output (bottom) spectra for two input signals.
12.3. TWO-TONE INPUT 383
close to the IF, and is of special concern. For now we will consider only the in-band terms:
v
0
o
(t) =
k
1
a
1
+k
3
3
4
a
3
1
+
3
2
a
1
a
2
2
cos !
1
t (12.27)
+k
3
3
4
a
1
a
2
2
cos(2!
2
!
1
)t
+k
3
3
4
a
2
1
a
2
cos(2!
1
!
2
)t
+
k
1
a
2
+k
3
3
4
a
3
2
+
3
2
a
2
1
a
2
cos !
2
t
There are a number of phenomena that can be illustrated using this result.
12.3.1 Desensitization and Blocking
Suppose a
1
cos !
1
t is a relatively weak desired signal and a
2
cos !
2
t is a strong signal at a
nearby frequency that is not of interest. To model this situation we assume that a
2
a
1
.
Taking into account that a
2
a
1
, and assuming that the amplier exhibits compressive
nonlinearity, the amplitude of the desired cos !
1
t term is, approximately:
k
1
a
1
+k
3
3
2
a
1
a
2
2
(12.28)
= k
1
a
1
k
3
k
1
3
2
a
2
2
(12.29)
Notice that the amplitude of the desired term will be reduced as the amplitude a
2
increases.
This means that the presence of a
2
cos !
2
t causes gain compression for the desired signal,
even though the desired signal may be too weak to cause gain compression by itself. Thus,
an undesired strong signal can desensitize a receiver to the presence of weak signals by
compressing the gain. This phenomenon is called densensitization, or blocking.
12.3.2 Cross modulation
Another undesirable eect can result if both signals are modulated. For example, suppose
that the signals are amplitude modulated, i.e., a
1
)a
1
(1 +m
1
(t)) and a
2
)a
2
(1 +m
2
(t))
and that a
2
a
1
. Then the envelope of the desired signal will be given by
= k
1
a
1
(1 +m
1
(t)) +k
3
3
2
a
1
(1 +m
1
(t))a
2
2
(1 +m
2
(t))
2
(12.30)
Notice that the signal at frequency !
1
now contains amplitude modulation from signal at
frequency !
2
. The modulation from the undesired strong signal at !
2
has been transferred
to the desired signal at !
1
. This phenomenon is called cross modulation. If the signals
are angle modulated, the same mechanism will cause the angle modulation on signal 1 to
contain a term from the angle modulation on signal 2.
384 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
12.3.3 More than two tones and nonlinear terms with order higher
than 3
So far we have considered only two input signals. If there are more than two input tones,
many more intermodulation products will be generated. In general, for M input tones
({f
i
}, i = 1, 2, . . . M) an nth order non-linearity will generate all positive frequencies
given by the magnitude of the sum of n terms of the form f
i
, i 2 {1, 2, . . . M}. For
example, if there are 3 input signals at frequencies f
1
, f
2
, f
3
the third-order nonlinearity
will produce third-order products at | f
i
f
j
f
k
|, where i, j, k 2 {1, 2, 3}. When
the three input frequencies are closely spaced, the frequencies resulting when two of the
three terms have the same sign will be close to the input frequencies and, hence, are called
in-band intermodulation (IM) products. The in-band third order IM products when there
are three input tones are listed below:
2f
1
f
2
, 2f
2
f
1
2f
3
f
1
, 2f
1
f
3
2f
2
f
3
, 2f
3
f
2
f
1
+ f
2
f
3
, f
1
+ f
3
f
2
, f
2
+ f
3
f
1
f
1
, f
2
, f
3
With three input tones, a 4th order nonlinearity will generate frequencies of the form
| f
i
f
j
f
k
f
l
| where i, j, k, l 2 {1, 2, 3}, etc. Note that the even-order nonlinearities
will not produce in-band products. Only the odd-order nonlinearities are responsible for in-
band IM products. For example, with two input tones, a 5th order nonlinearity would
produce output components at |f
1
+f
1
+f
1
f
2
f
2
| = 3f
1
2f
2
, which is an in-band IM
product.
12.4 Quantitative Characterization of IM Distortion
When multiple input signals are present, nonlinearity causes complex eects. For quanti-
tative comparisons between 2-ports, an idealized situation, or special case, is often used to
provide a common basis for comparisons. The usual conditions for this test are to apply two
input signals with equal amplitudes; this is called a two-tone intermodulation distortion
(IMD) test. The input signals can be written as
v
i
(t) = a(cos !
1
t + cos !
2
t) (12.31)
It is assumed that the amplitude of the two signals is small enough so that gain compression
doesnt occur. Then the in-band products at the output of the 2-port are
v
0
o
(t) = k
1
a(cos !
1
t + cos !
2
t) (12.32)
+k
3
3
4
a
3
(cos(2!
2
!
1
)t + cos(2!
1
!
2
)t)
12.4. QUANTITATIVE CHARACTERIZATION OF IM DISTORTION 385
Note that the coecient of the fundamental terms is simply k
1
which is the result of
ignoring gain compression. Next dene the input power of each tone, P
in
. For simplicity,
we shall assume that the input impedance of the 2-port is resistive, and equal to the value
R
in
. Then
P
in
=
1
2
a
2
R
1
in
(12.33)
Assuming that the load impedance is resistive, and denoted by R
L
, the output power of
each of the desired (fundamental) components is then
P
d
=
1
2
k
2
1
a
2
R
1
L
The output power of the in-band third-order intermodulation products is
P
IM
=
1
2
k
2
3
(
3
4
)
2
a
6
R
1
L
Notice that P
d
will be proportional to P
in
, whereas P
IM
is proportional to P
3
in
. Figure 12.7
shows P
d
and P
IM
plotted against P
in
on a log-log plot (dBm versus dBm). At the lower
input power levels, P
IM
increases with a slope of 3 compared to P
d
which has a slope of
1. At higher input power, both components will eventually begin to saturate, as shown by
the deviations of the curves from linearity. It should be noted that our simple 3-term power
series model does not predict saturation of the IM products. It is necessary to include higher
order terms to predict this behavior.
P
(i)
I
(dBm)
P
in
(dBm)
P
(o)
I
P
d
P
IM
Figure 12.7: Output power (per tone) in the desired (P
d
) and the IM (P
IM
) components
verses input power per tone. The third order intercept power, P
(i)
I
, is determined by extrap-
olation from measurements at a small input power, where gain compression is negligible.
The fact that P
IM
increases faster than P
d
means that if gain compression did not
eventually occur, the two output curves would ultimately intersect, as shown in Figure 12.7.
386 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
The input power level at which the linearly extrapolated curves intersect is called the input
two-tone third-order intercept level and is denoted by P
(i)
I
, or IIP3. The corresponding
output power level is denoted by P
(o)
I
, or OIP3.
It is important to note that the intercept point is ctitious, since we ignored gain
compression. In actuality, both the P
IM
and P
d
curves would saturate at some nite value.
This is ignored when computing the third-order intercept level. Since we know the slopes
of the ideal P
IM
and P
d
curves, it is a simple matter to measure P
IM
and P
d
at a relatively
low input power level where gain compression is not important. Then the curves can be
extrapolated to the (ctitious) intercept point.
The intercept level is commonly used as a gure of merit for comparing the relative
quality of ampliers or systems. If it is known, then a system designer can determine
P
IM
/P
d
for any two-tone input level where gain compression can be ignored. Two-ports are
often operated at levels where gain compression is not important but where intermodulation
distortion is still of concern.
Equation 12.34 summarizes the relationship between the input power level and the so-
called intermodulation ratio, or IMR
IMR =
P
IM
P
d
=
P
in
P
(i)
I
2
(12.34)
Equation 12.34 can be used to determine P
IM
/P
d
given P
in
, if P
(i)
I
is known. This is
useful for system design applications. On the other hand, given measurements of P
d
, P
IM
for a known P
in
, Equation 12.34 can also be used to nd P
(i)
I
.
12.4.1 Example - Calculating IMR
An amplier has a two-tone third-order input intercept (IIP3) of 20 dBm. What is the
intermodulation ratio (in dB) for a two-tone input level of 0 dBm?
The intermodulation ratio is calculated from:
IMR =
P
IM
P
d
=
P
2
in
P
(i)2
I
(12.35)
or, in dB
IMR(dB) = 2 P
in
(dBm) 2 P
(i)
I
(dBm) (12.36)
= 2 (0) 2 (20)
= 40 dB (12.37)
This means that the IM products (in-band) are 40 dB below the desired signals at the
output, i.e., the output spectrum would look like Figure 12.8.
The next example shows how laboratory measurements of intermodulation products can
be used to calculates the intercept point of a 2-port.
12.4. QUANTITATIVE CHARACTERIZATION OF IM DISTORTION 387
In-band IM products
Desired tones
40 dB
Figure 12.8: Output spectrum showing desired and IM products.
+5 dBm
35 dBm
Figure 12.9: Output spectrum with 2 input signals, P
in
= 10 dBm.
12.4.2 Example - Calculating IIP3 (P
(i)
I
).
Figure 12.9 shows the spectrum measured at the output of a 2-port when the input consists
of 2 signals with P
in
= 10 dBm. The power gain and input intercept of the 2-port
can be found if it is known that gain compression can be neglected when interpreting the
measurements. This can be veried by increasing the power of both input tones by 1 dB, and
verifying that the desired tones at the output also increase by 1 dB, and that the in-band
3rd order intermodulation products each increase by 3 dB. If so, then
G = +5 dBm(10 dBm) (12.38)
= 15 dB
The input intercept can be found from
P
IM
P
d
= 2P
in
2P
(i)
I
(12.39)
where all quantities are expressed in dBm. Solving for the input intercept:
P
(i)
I
= P
in
+
1
2
(P
d
P
IM
) (12.40)
= 10 +
1
2
(5 (35))
= +10 dBm
388 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
When performing a measurement like the one illustrated in this example, a typical setup will
use a passive signal combiner to combine the outputs from two signal generators in order
to construct the two-tone signal. It is important to take care to ensure that the combiner
provides signicant isolation between the two signal generators, so that intermodulation
products are not generated within the signal generator output stages. It is also important
to verify that intermodulation products generated within the spectrum analyzer can be
neglected.
12.5 Dynamic Range of a Receiving System
The nonlinear eects that have been described in this chapter will determine the largest input
signal level that a receiving system can handle without serious degradation of performance.
The noise gure of the receiver, on the other hand, determines the smallest usable input
signal level. Together the nonlinear eects and the noise oor determine the dynamic range
of a receiver. Consider the receiving system in Figure 12.10.
G
A
, T
e
,
B
n
, P
(
I
i)
Receiver
to demodulator
T
s
Figure 12.10: Receiving system.
The dynamic range (DR) is dened as follows:
DR =
maximum useable input signal power level
minimum useable input signal power level
(12.41)
and is usually expressed in dB, i.e.,
DR = 10 log
P
max
P
min
(12.42)
The numerator P
max
is the largest usable input signal power level. There are various
denitions for P
max
that depend on what nonlinear eect is being considered by the designer
and what level of degradation is considered unsatisfactory. A commonly employed denition
for the dynamic range is the spurious free dynamic range where the powers P
max
and P
min
are dened as follows:
P
min
) input power for specied SNR
o,min
at system output (same as MDS)
P
max
) two-tone input power (per tone) at which SNR of in-band
3rd-order IM products is equal to SNR
o,min
at system output.
The minimum usable input signal power is equal to the MDS, which has been dened
previously:
P
min
MDS = k(T
S
+T
e
)B
n
SNR
o,min
.
12.6. INTERCEPT POINT OF A CASCADE 389
The maximum usable input signal power, P
max
, is dened based on the two-tone input
test concept. Suppose that the input consists of two equal amplitude tones. Then P
max
is the input level that would cause the in-band third-order products to be detectable at
the output of the receiver. In this context detectable means that the signal-to-noise ratio
of the third-order products would be equal to the minimum SNR required for detection,
SNR
o,min
.
Assume that P
min
has already been computed. The output signal power that results
from an input power P
min
is just G
A
P
min
. This output power represents the detection
threshold. Now we ask what two-tone input power is required to cause the in-band third
order products to have an output power equal to the detection threshold. The required
input power is, by denition, P
max
. When the input power, per tone, is P
max
, the output
power in the intermodulation products is G
A
P
min
. Using equation 12.34:
G
A
P
min
P
d
= (
P
max
P
(i)
I
)
2
.
Use the fact that P
d
= G
A
P
max
:
P
min
P
max
= (
P
max
P
(i)
I
)
2
.
Hence
P
max
= P
1/3
min
P
(i)
I
2/3
(12.43)
Thus, the dynamic range is
DR =
P
1/3
min
P
(i)
I
2/3
P
min
(12.44)
=
P
(i)
I
P
min
2/3
or, in dB
DR =
2
3
[P
(i)
I
P
min
] (12.45)
where P
(i)
I
and P
min
are expressed in dBm. Recall, P
min
is just the noise-oor of the
receiving system and that it depends on the signal-to-noise ratio that is required for the
particular type of signal and demodulator used in the system.
12.6 Intercept Point of a Cascade
Suppose that an amplier with known input intercept level is cascaded with another amplier
as in Figure 12.11.
To simplify the analysis, note that the signal level to stage 2 is higher than to stage 1, if
stage 1 has gain. So the nonlinear distortion in amplier 2 is likely to be more important.
We will ignore distortion generated in the rst stage, which is equivalent to assuming that
stage 1 is a linear preamplier (equivalent to letting P
(i)
I1
!1).
390 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
G
A1
,
P
(i)
I1
G
A2
,
P
(i)
I2
P
in2
P
in
Figure 12.11: Cascaded ampliers
Then, for the second stage alone:
P
IM
P
d
= (
P
in2
P
(i)
I2
)
2
(12.46)
For the cascade, denote P
(i)
I
as the input intercept (as yet unknown), then
P
IM
P
d
= (
P
in
P
(i)
I
)
2
(12.47)
Rewriting Equation 12.46 using P
in2
= G
A1
P
in
P
IM
P
d
=
(G
A1
P
in
)
2
P
2
I2
(12.48)
=
P
2
in
(P
I2
/G
A1
)
2
Comparing Equations 12.48 and 12.47 we conclude
P
(i)
I
=
P
(i)
I2
G
A1
(12.49)
The addition of stage 1 reduces the intercept point by a factor equal to the gain of the rst
stage. For example, if P
(i)
I2
= +6 dBm, G
A1
= 10 dB, then the cascade has input intercept
P
I
= 4 dBm. In the next section, we will show that adding gain in front of an existing
receiver will reduce the system DR.
12.6.1 The eect of adding a preamp to a receiver
Suppose that a receiver has eective input temperature T
er
, noise bandwidth B
n
, and input
intercept P
(i)
I
. The dynamic range of the receiver is then
DR
receiver
= [
P
(i)
I
k(T
s
+ T
er
)B
n
SNR
o,min
]
2/3
(12.50)
Suppose that a preamp with eective input temperature T
ep
and available gain G
A
is added
before the receiver. Assuming that distortion generated in the preamplier is negligible, the
new dynamic range is
DR
receiver with preamp
= [
P
(i)
I
/G
A
k(T
s
+T
ep
+T
er
/G
A
)B
n
SNR
o,min
]
2/3
(12.51)
12.6. INTERCEPT POINT OF A CASCADE 391
Consider the ratioDR
withpreamp
/DR
receiver
:
DR
receiver with preamp
DR
receiver
= [
T
s
+T
er
G
A
(T
s
+T
ep
) +T
er
]
2/3
. (12.52)
Now, we ask whether this ratio can exceed unity, i.e. whether adding a preamp can increase
the dynamic range of a receiving system. This requires
T
s
+T
er
> G
A
(T
s
+T
ep
) +T
er
(12.53)
Subtracting T
s
from both sides of Equation 12.53 results in the following inequality which
must be satised, if the addition of a preamp is to increase the DR of a receiver:
T
ep
< T
s
(1 G
A
)/G
A
(12.54)
This inequality can never be satised if the preamp has available gain > 1, since the RHS
of Equation 12.54 will be negative. The conclusion is that adding a preamp to an existing
receiver will always decrease the DR of the receiver (although it may decrease the noise oor
and therefore improve the sensitivity).
If the available gain of the added stage is less than 1, then the inequality can be satised.
For example, suppose that the added stage is a passive attenuator. Then T
ep
must be
replaced with T
att
(
1
G
A
1) and the inequality is satised if
T
att
< T
s
.
Therefore, adding an attenuator in front of an existing receiver will increase the dynamic
range of the system if the physical temperature of the attenuator is smaller than the source
temperature (usually the antenna temperature). Adding an attenuator in front of an existing
receiver will decrease the sensitivity (increase the MDS) of the receiver, however.
The intercept point of a receiver is usually set by the input intercept of the rst mixer.
The reason for this is as follows intermodulation distortion is most likely to be generated in
the stages ahead of the relatively narrow band IF ampliers, i.e., the preamplier and mixer
stages. Since the input signal level at the mixer is generally larger than for any preceeding
stage, the intercept point of the mixer will usually determine the largest usable input signal
level. After the mixer, the signals are ltered by the IF lters, and generation of further
intermodulation components in the IF ampliers is less likely since, presumably, the desired
signal component has been picked out from among other components. So the mixer is
the most likely source of intermodulation distortion products (in a well designed receiver).
Generation of intermodulation products in the mixer will be minimized if the gain in front
of the mixer is as small as possible.
This leads us to an important principle for receiver design: The smallest possible amount
of gain (preamp gain) should be used in front of the mixer. The minimum preamp gain will
be determined by the preamp noise temperature and/or the antenna temperature together
with the required MDS. Increasing the preamplier gain above that required to set the
noise oor to an acceptable level will only increase the input signal levels at the mixer input
and degrade the large-signal handling capability of the receiver without providing useful
improvement in sensitivity.
392 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
12.7 References
1. Carson, Ralph S., Radio Concepts: Analog, John Wiley & Sons, New York, 1990.
2. Clarke, Kenneth K. and Donald T. Hess, Communication Circuits: Analysis and De-
sign, Addison-Wesley, 1978.
3. Maas, Stephen A., Nonlinear Microwave Circuits, Artech House, 1988.
4. Smith, Jack, Modern Communications Circuits, McGraw Hill, 1986.
12.8. HOMEWORK PROBLEMS 393
12.8 Homework Problems
1. A nonlinear amplier has an input-output voltage characteristic
v
o
(t) = k
1
v
i
(t) +k
2
v
2
i
(t) (12.55)
Suppose that three signals with equal amplitudes are input to the amplier. The
signal frequencies are
f
1
= 0.6 MHz (12.56)
f
2
= 1.3 MHz
f
3
= 1.5 MHz
List the frequencies of all components that will appear at the output of the amplier.
2. Consider an AM broadcast band receiver that is designed to cover the frequency range
540 to 1700 kHz. The single-conversion superhet receiver uses an IF of 455 kHz and
High LO. The receiver uses an RF amplier that has the following nonlinear input-
output characteristic:
v
o
= k
1
v
RF
+k
2
v
2
RF
+k
3
v
3
RF
(12.57)
Suppose two signals at frequencies f
1
= 720 kHz and f
2
= 780 kHz are input to the
receiver. Assume that both signals pass through the preselector. You may assume
that the receiver stages that follow the RF amplier are linear and that the mixer can
be modeled as an ideal multiplier. List the frequencies of all possible output signals
from the RF amplier that could be detected with the receiver. Also, for each signal
that can be detected, list the receiver tuning dial setting at which the signal will be
detected.
3. A nonlinear 2-ports input-output characteristic can be modeled as
v
o
= 12v
i
v
3
i
where v
i
and v
o
represent the instantaneous values of the time-varying component of
the input and output voltage, respectively.
(a) Suppose that this 2-port is used in a system with source and load impedance of
50 . Assume that the 2-port is conjugately matched at the input and output
in this system. Find the input power level that causes 1dB of gain compression,
P
1dB
. Express your result in dBm.
(b) Suppose that two signals are input to the 2-port, i.e., that v
i
= a
1
cos !
1
t +
a
2
cos !
2
t. List the frequencies of all components that will appear at the output
of the 2-port.
(c) Now assume that the two signals in part 3b have equal amplitudes, i.e., a
1
=
a
2
= a. Find the maximum input signal power (for each signal) that will cause
the in-band third-order components to be down 30 dB with respect to the desired
signals at the output of the 2-port. You may assume that gain compression can
be ignored at this input power level. Express your result in dBm.
394 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
3 dBm
30 dBm
Figure 12.12: Spectrum analyzer display.
4. Two signals at closely spaced frequencies f
1
and f
2
are applied to the input of an
amplier. The input power for each of the signals is -20 dBm. The display in Figure
12.12 is seen when the output of the amplier is connected to a spectrum analyzer:
(a) Find the input intercept P
(i)
I
for this amplier. Assume that gain compression is
not important.
(b) What is the operating power gain of the amplier (in dB)?
5. Consider the receiving system in Figure 12.13:
Receiver
T
e
= 200 K
G
A
= 12 dB
B
n
= 50 kHz
NF = 8 dB
Preamp
T
A
= 1000 K
Figure 12.13: Receiving system.
(a) Assume that the antenna and 2-ports are matched to 50 . Find the available
signal power level from the antenna (in dBm) that is required to give a 15 dB
SNR at the output of the receiver.
(b) The two-tone third-order input intercept level for the receiver is +6.5 dBm. As-
sume that the preamplier is linear. Find the spurious-free dynamic range (DR)
of the system if a 15 dB SNR is required at the output for detection. Express
your result in dB.
6. A receiver has eective input temperature T
e
= 1500K and noise bandwidth B
n
=
3 kHz. The antenna temperature is 400K. The two-tone third-order input intercept
for the receiver is -5 dBm. A signal-to-noise ratio of 6 dB is required for detection.
(a) Compute the MDS for the receiver. Give your result in dBm.
(b) Compute the spurious-free dynamic range of the receiver. Give your result in dB.
(c) Suppose that the MDS computed in part 6a is too high for the intended ap-
plication, i.e., suppose that the required MDS is -130 dBm. To achieve this
12.8. HOMEWORK PROBLEMS 395
specication, we decide to add a preamplier at the input of the receiver. Sup-
pose that the preamplier will have an eective input temperature T
e
= 175K.
How much gain should the preamplier have? Specify the value that will result
in an MDS = -130 dBm.
(d) Use your result from part 6c and compute the spurious-free dynamic range of the
preamp-receiver combination. You may assume that the preamplier is a linear
2-port.
7. Consider a receiver with eective input temperature T
e
, two-tone third-order input
intercept level P
(i)
I
, and equivalent noise bandwidth B
n
. Suppose we put a passive,
matched attenuator in front of the receiver. The attenuator has loss L > 1. The
physical temperature of the attenuator is equal to standard temperature, i.e., T
att
=
T
o
. The addition of the attenuator will cause the cascade to have a higher input
intercept level than that of the original receiver. The new input intercept level for the
cascade will be P
(i)
I
L. Denote the source temperature by T
s
. Depending upon the
value of T
s
, the addition of the attenuator will either increase or decrease the dynamic
range of the system relative to that of the original receiver. What constraint must
be satised by T
s
in order for the addition of the attenuator to increase the dynamic
range of the system?
8. An amplier is found to have the following nonlinear input-output voltage character-
istic:
v
o
= k
1
v
i
+k
2
v
2
i
(12.58)
where k
1
= 15 and k
2
= 1. This characteristic does not include a third-order term,
k
3
= 0, so the two-tone third-order input intercept power is innite. Dene a two-tone
second-order input intercept power, P
(i)
I2
, which is the input power for each of the equal
amplitude tones that causes the output power in the second-order intermodulation
products at frequencies f
1
+ f
2
, |f
1
f
2
| to be equal to the output power in the
desired tones. Assume that the input impedance of the amplier is 50 and nd P
(i)
I2
(in dBm).
9. Two signals at closely spaced frequencies f
1
and f
2
are applied to the input of a
nonlinear amplier. Denote the input power of the tones by P
1
and P
2
, respectively.
Suppose that P
1
is decreased by 2 dB. Specify the change in the output power of each
of the frequency components: f
1
, 2f
2
, 2f
2
f
1
, 2f
1
f
2
, 3f
1
. You may assume that
P
1
and P
2
are small enough so that gain compression can be ignored.
10. Suppose that an AM broadcase receiver uses a nonlinear preamplier described by:
v
o
= k
1
v
i
+k
2
v
2
i
+k
3
v
3
i
(12.59)
and a 4-diode doubly-balanced switching-type mixer to receive a strong signal with
carrier frequency 1200 kHz. The 4-diode doubly balanced mixer produces terms with
frequencies |f
RF
nf
LO
|, n=1,3,5... at its output. The radio uses a 455 kHz IF, the
local oscillator tunes from 995-2155 kHz, and the receivers tuning dial spans 540 -
1700 kHz. Find and list all settings of the receiver tuning dial at which you could
potentially detect a signal from the station at 1200 kHz.
396 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
Chapter 13
Phase-locked Loops (PLLs)
13.1 PLL Fundamentals
Figure13.1 shows a linear model for a PLL. In this idealized linear model, the input is the
phase of the reference signal and the output is the phase of the VCO. Both the reference and
VCO signals are assumed to be cosinusoidal signals with constance amplitude (for now) and
slowly changing phase. When the loop is locked and in a steady state, the VCO (output)
r
PD
Phase Detector
V
a
F(s)
Loop Filter
V
c
V CO
Voltage Controlled
Oscillator
o
Figure 13.1: Linear model for a PLL.
frequency and reference frequency are equal, i.e.,
V
r
(t) = V
r
cos (!
r
t +
r
) (13.1)
V
o
(t) = V
o
cos (!
r
t +
o
). (13.2)
The phase detector (PD) produces an output voltage that is proportional to the phase error,
e
, which is dened to be the dierence between the phase of the reference signal and the
VCO signal:
V
a
(t) = K
d
(
r
(t)
o
(t)) (13.3)
= K
d
e
(t)
or in the s-domain:
V
a
(s) = K
d
(
r
(s)
o
(s)) (13.4)
= K
d
e
(s)
K
d
= PD gain constant (V/radian)
397
398 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
The PD output voltage is applied to the loop lter to produce the VCO control voltage, V
c
V
c
(s) = V
a
(s)F(s) (13.5)
where F(s) has a low-pass frequency response function.
The instantaneous frequency deviation of the VCO output signal is proportional to the
control voltage:
d
o
dt
= K
o
V
c
(t) (13.6)
or
s
o
(s) = K
o
V
c
(s) (13.7)
K
o
= VCO gain constant,
radians
Vsec
13.1.1 PLL Transfer functions
o
(s)
r
(s)
=
K
o
K
d
F(s)
s + K
o
K
d
F(s)
(13.8)
e
(s)
r
(s)
=
s
s + K
o
K
d
F(s)
(13.9)
Sometimes the control voltage, V
c
, is the desired output signal, e.g., when the loop is used
as a demodulator for FM:
V
c
(s)
r
(s)
=
s K
d
F(s)
s + K
o
K
d
F(s)
(13.10)
13.1.2 Loop Gain and Notation
The closed-loop transfer function H(s) as dened in Equation 13.8 is
H(s) =
o
(s)
r
(s)
(13.11)
=
K
o
K
d
F(s)
s + K
o
K
d
F(s)
Note:
e
(s)
r
(s)
=
r
(s)
o
(s)
r
(s)
(13.12)
= 1 H(s)
V
c
(s)
r
(s)
=
s
K
o
H(s)
H(s) can be written
H(s) =
K
o
K
d
F(s)/s
1 + K
o
K
d
F(s)/s
(13.13)
13.1. PLL FUNDAMENTALS 399
The quantity A(s) = K
o
K
d
F(s)/s is called the open loop gain. Note that the closed loop
transfer function, H(s), can be written in terms of the open loop gain:
H(s) =
A(s)
1 +A(s)
. (13.14)
13.1.3 Order and Type
The order of a PLL is dened by the highest power of s in the denominator of the closed
loop transfer function, i.e.,
First order:
H(s) =
K
s + a
(13.15)
Second order:
H(s) =
K
s
2
+ a s + b
(13.16)
The type of a PLL is dened by the number of poles at the origin for the open loop transfer
function, i.e.,
Type 1:
A(s) =
K
s
(13.17)
Type 2:
A(s) =
K
s
2
(13.18)
All PLLs are at least type 1, because the VCO output phase is proportional to the integral
of the control voltage.
13.1.4 Loop Filters
Commonly employed loop lters fall into four classes ranging from the simplest (no lter at
all) to active lters.
1. No lter:
F(s) = 1 (13.19)
Then
H(s) =
K
o
K
d
s + K
o
K
d
(13.20)
=
K
o
K
d
/s
1 + K
o
K
d
/s
=
!
L
/s
1 + !
L
/s
!
L
= K
o
K
d
= loop bandwidth
Notice that the open loop gain A(s) =
Ko K
d
s
has one pole at the origin, so this loop
is type 1. Since the power of s in the denominator of H(s) is 1, it is a rst-order loop.
400 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
2. The simplest lter that gives a second-order loop is a single-pole low-pass RC lter,
as shown in Figure 13.2. If the PD output is modeled as an ideal voltage source, V
a
,
V
a
+
R
C V
c
+
KoK
d
+
s
KoK
d
+ 1
(13.21)
This function can be written in the canonical form
H(s) =
1
1
!
2
n
s
2
+
2
!n
s + 1
(13.22)
where
!
n
=
K
o
K
d
(13.23)
=
!n
2KoK
d
=
1
2
p
KoK
d
(13.24)
The parameter !
n
is the loop bandwidth and is called the damping factor. To-
gether the loop bandwidth and the damping factor determine how quickly the loop
can respond to changes in the reference signals frequency or phase.
3. The lag-lead lter (Figure 13.3) is often preferred to the simple low-pass RC lter
because it leads to an improved phase margin and therefore a more stable loop transfer
function. The lag-lead loop lters transfer function is
F(s) =
1 +s
2
1 +s
1
,
where
1
= (R
1
+R
2
)C
2
= R
2
C
13.1. PLL FUNDAMENTALS 401
V
a
+
R
1
C
R
2
V
c
+
K
o
K
d
1
=
1
2
K
o
K
d
1/2
2
+
1
K
o
K
d
=
2
!
n
2
+
!
n
2 K
o
K
d
4. An active lter (often used in integrated circuit PLLs) is shown in Figure 13.4: The
transfer function of this lter is
F(s) =
A(s
2
+ 1)
s
2
+ 1 + (1 + A) s
1
,
where A is the open loop gain of the op-amp. For suciently large A the transfer
function is well approximated by
F(s) '
s
2
+ 1
s
1
. (13.26)
402 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
R
1
+
R
2
C
2
= R
2
C
1
= R
1
C
Figure 13.4: Active loop lter employing an op-amp.
Thus, for large A the loop transfer function can be written
H(s) =
2 !
n
s + !
2
n
s
2
+ 2 !
n
s + !
2
n
(13.27)
!
n
=
K
o
K
d
1
=
2
2
K
o
K
d
1/2
=
2
!
n
2
The high gain (large A) active loop lter and the passive lag-lead lter with large loop
gain (K
o
K
d
) both yield open-loop transfer functions of the form
H(s) =
2!
n
s + !
2
n
s
2
+ 2!
n
s + !
n
2
(13.28)
The high-gain active loop lter approximates a type 2 loop because the open loop gain
has two poles at the origin:
A(s) =
KoK
d
F(s)
s
' K
o
K
d
s2+1
s
2
1
(13.29)
The fact that the denominator of A(s) contains s
2
means that there are 2 integrators
in the loop. One is the VCO itself, because the phase of the VCO output signal is
the integral of the control voltage (
o
=
KoVc(s)
s
). The other integrator is the active
lter. Table 13.1 illustrates the how the parameters of the loop transfer function
depend on the elements in the circuits of the passive lag-lead lter and the active
lter. Figure 13.5 shows how the active lter is implemented in the Motorola MC4044
PLL integrated circuit.
13.1.5 Steady-state Error Analysis
Consider two types of inputs:
13.1. PLL FUNDAMENTALS 403
Passive lag-lead Active Filter
!
n
=
KoK
d
1
!
n
=
KoK
d
1
=
2!n
2
=
2!n
2
1
= (R
1
+R
2
)C
1
= R
1
C
2
= R
2
C
2
= R
2
C
Table 13.1: Comparison of passive lag-lead and active loop lters.
R
1
R
L
= 1k
5V
R
2
C
V
o
Figure 13.5: Active lter implementation in the Motorola MC4044 chip.
1. Step input in phase:
r
(t) = u(t) (13.30)
2. Step input in frequency:
f
r
(t) = f
c
+ f u(t) (13.31)
or, equivalently:
r
(t) = ! t u(t) (13.32)
The error signal
e
(t) =
r
(t)
o
(t) has transfer function
e
(s)
r
(s)
=
1
1 + K
o
K
d
F(s) /s
. (13.33)
The Laplace transform nal value theorem (for stable systems) says:
lim
t!1
e
(t) = lim
s!0
s
e
(s) (13.34)
= lim
s!0
s
2
r
(s)
s + K
o
K
d
F(s)
404 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
If
r
(t) is a step input, e.g.,
r
(t) = u(t), then
r
(s) =
1
s
(13.35)
lim
t!1
e
(t) = lim
s!0
s
s + K
o
K
d
F(s)
Now F(s) is either a constant (rst-order loop) or a low-pass lter that may include poles
at the origin, i.e.,
lim
s!0
F(s) =
K
s
n
6= 0 (13.36)
Thus
lim
t!1
e
(t) = lim
s!0
s
n+1
K
o
K
d
K
= 0 (13.37)
i.e., a stable PLL will track step changes in phase with zero steady-state error. If the
frequency changes suddenly, i.e.,
f
r
(t) = f
c
+ f u(t) (13.38)
or
r
(t) = t ! u(t) (13.39)
then
r
(s) =
1
s
2
! (13.40)
The steady-state phase error is
lim
t!1
e
(t) = lim
s!0
!
s + K
o
K
d
F(s)
(13.41)
=
!
K
o
K
d
F(0)
If F(0) = constant (dc gain of lter = constant), then the steady-state phase error is inversely
proportional to K
o
K
d
. Since a larger value for K
o
K
d
leads to a larger loop bandwidth for
all the lters considered, we can conclude that a large loop bandwidth is desirable if the
steady-state phase error is to be minimized.
The frequency error f
e
(t) =
d
dt
e
(t) will tend toward zero as t ! 1 (because
e
!
constant). If it is necessary to have zero steady-state phase error in response to a step
frequency input, then
lim
s!0
!
K
o
K
d
F(s)
= 0 (13.42)
or
lim
s!0
F(s) = 1 (13.43)
This means that F(s) must have a pole at the origin which means that the DC gain of the
lter must approach innity. In this case the system will be type 2, since
A(s) =
K
o
K
d
F(s)
s
(13.44)
13.2. STABILITY ANALYSIS 405
has two poles at the origin. We can approximate the type 2 loop with a high gain active lter
which can be used to achieve an essentially zero steady-state phase error. The addition of a
pole at the origin can, however, cause stability problems. Stability issues will be discussed
in the following section.
13.2 Stability Analysis
The stability of a loop is determined by A(s) which depends on F(s). A loops stability
can be studied by examining the complex-plane location of the poles of H(s). This requires
knowledge of the analytical form of H(s). Alternatively, stability can be studied without
direct knowledge of the pole locations or even an analytical description of the transfer
function by using Bode plots. A Bode plot (shown in Figures 13.6 and 13.7) consists of a
pair of graphs of the magnitude (in dB) and phase of the open loop gain A(s) plotted on
a logarithmic frequency scale. The Bode criterion for stability is simple and is equivalent
to the reverse of the Barkhausen criterion for oscillation. The Bode criterion states that a
loop is stable if the magnitude of A(s) falls below 1 (0 dB) before the phase-shift of A(s)
reaches 180
.
Gain (dB)
Phase (Deg)
0
-180
0
Gain crossover frequency
Phase crossover
frequency
log f
log f
Gain margin
Phase margin
Figure 13.6: Bode plot of A(s) for a stable PLL
Summary of stability-related terminology:
Phase margin is the dierence between the actual phase and 180
at the frequency
where the magnitude of the open loop transfer function is unity.
406 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
Gain (dB)
Phase (Deg)
0
-180
0
Gain crossover frequency
log f
log f
Phase crossover
frequency
Figure 13.7: Bode plot of A(s) for an unstable PLL
13.2. STABILITY ANALYSIS 407
Gain margin is the number of dB below 0 dB for the gain at the frequency where the
phase of the open loop transfer function is 180
.
The greater the phase margin, the more stable the system and the more phase lag
from parasitic eects can be tolerated. Phase compensation provided by the lag-lead
lter can often be used to stabilize a marginally stable loop. See the examples in the
following section.
13.2.1 Examples of Stability Analysis
Consider a PLL which has K
o
K
d
= 50 rad/s and which contains a simple RC low-pass
lter with corner frequency !
c
= 100 rad/s. The open-loop gain is
A(s) =
K
o
K
d
s
F(s) (13.45)
or
A(s) =
50
s(1+
s
100
)
A(j!) =
50
j!(1+
j!
100
)
(13.46)
The magnitude and phase of A(j!) are plotted in Figure 13.8. The crossover frequency,
Figure 13.8: Bode plot for lter corner frequency = 100 rad/s
where |A(j!)| = 1, is ! = 45.51 s
1
. At this frequency the phase angle of the open-loop
gain is 114.5
114.5
= 65.6
.
Suppose that the lter corner frequency was 10 rad/s instead of 100 rad/s. The magni-
tude and phase of A(j!) are shown in Figure 13.9. In this case the gain crossover frequency
is 21.3 s
1
and the phase angle at the gain crossover is 154.8
e
(t) = u(t)
o
(t) = (1 e
KoK
d
)
(13.48)
Figure 13.11 shows how the VCO output signal responds to the step change in a rst-order
loop. When the loop bandwidth K
o
K
d
is large, the time constant is small, and the loop
responds quickly to changes in the reference signals phase. Now, suppose that the reference
signals frequency undergoes a sudden step change. In that case:
r
(t) = !u(t)
o
(t) = !t
!
KoK
d
(1 e
KoK
d
)
(13.49)
13.3. TRANSIENT RESPONSE OF PLLS 409
Figure 13.10: Bode plot for lag-lead lter.
r
(t)
o
(t)
t
Figure 13.11: Time constant =
1
KoK
d
.
410 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
Figure 13.12 shows the eect the response of the VCO output phase compared to the ref-
erence phase after a frequency step. Notice that in steady-state, the VCO output phase
r
(t)
o
(t)
K
o
K
d
Figure 13.12: Reference signal phase and VCO output phase after a step change in the
reference frequency.
lags the reference signal phase by
!
KoK
d
. This means the the phase error is non-zero in
steady-state. If the loop gain, K
o
K
d
, is large then the phase error will be small.
We will now consider the transient response of second order type-2 loops. The transfer
function of such a PLL is:
H(s) =
2!
n
s + !
2
n
s
2
+ 2!
n
s + !
2
n
(13.50)
where !
n
is the loop bandwidth (or natural frequency), and is the damping factor. Both
are important in determining the frequency response and transient response of the loop.
Figure 13.13 shows the frequency response |H(j!)| for dierent damping factors.
Figure 13.13: Frequency response |H(j!)| for various damping factors
The way in which a second-order loop tracks a step change in phase or frequency can be
13.3. TRANSIENT RESPONSE OF PLLS 411
studied by computing the transient behavior of the phase error. Recall from Equation 13.12
that the phase error transfer function is given by:
e
(s)
r
(s)
= 1 H(s) (13.51)
For a step change in the reference phase:
r
(t) = u(t) (13.52)
or
r
(s) =
s
(13.53)
Thus
e
(s) =
s
s
2
+ 2!
n
s + !
2
n
(13.54)
where s has been normalized such that s = j!/!
n
.
It is not hard to show that the transient response is of the form:
e
(t) = e
!nt
{cosh[
2
1!
n
t]
2
1
sinh[
2
1!
n
t]} (13.55)
The transient response is plotted in Figure 13.14 for several dierent values of the damping
factor. Notice that (1) Increasing the damping factor results in faster settling and less
ringing, and (2) the phase error tends to zero for large times. The latter was predicted
earlier in our analysis of Equation 13.41.
Figure 13.14: Transient responses for damping factor values
It is useful to point out that these results can also be interpreted as the transient frequency
response due to a step change in frequency. To illustrate, suppose that the input is a step
change in frequency, i.e.,
!
r
= !
c
+ ! u(t) (13.56)
412 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
The transfer function relating VCO output frequency to the reference frequency is
!(s)
!r(s)
=
o(s)/s
r(s)/s
=
o(s)
r(s)
(13.57)
This means that the plot given in Figure 13.14 can be interpreted as the VCO frequency
response to a step change in the reference frequency.
Now consider what happens to the output phase when the reference frequency undergoes
a step change. Suppose that initially the loop is locked at its center frequency, !
c
. At t =
0 the reference frequency is increased by an amount !, i.e.,
!
r
= !
c
+ ! u(t) (13.58)
In terms of the reference phase, this input can be written as
d
r
(t)
dt
= ! u(t) (13.59)
Thus
r
(s) =
!
s
2
(13.60)
The phase error is therefore
e
(s) =
!
s
2
+ 2!
n
s + !
2
n
(13.61)
The transient response is
e
(t) =
!
!
n
e
!nt
1
2
1
sinh[
2
1!
n
t] (13.62)
This function is plotted in Figure 13.15. Notice that because the loop being considered is a
type 2 loop, the phase error tends to zero for large times. For type 1 loops the phase error
would tend toward a constant value.
13.3.1 Summary of Second-order Loops
There are 2 parameters to play with if the loop is second-order:
!
n
) natural frequency
) damping factor
Rules of thumb:
Large !
n
) small time constant, fast response
Large ) damped response, no ringing
Small ) ringing
Also note that:
13.4. APPLICATIONS 413
Figure 13.15: Transient response for a step change in the reference frequency.
For a second-order loop the phase margin increases with increasing .
The natural frequency, !
n
, and damping factor, , cannot be specied independently if
the loop lter is a simple low-pass lter. If an active lter or lag-lead lter is employed
then !
n
and can be specied independently.
13.4 Applications
13.4.1 Demodulation of an FM signal
Suppose an FM signal is applied to the reference input of a PLL. The loop will try to track
the deviations of the input frequency. If the loop is able to follow the deviations of the input
frequency, then the control voltage V
c
will be proportional to the instantaneous frequency
deviation of the reference signal. In this application the PLL is called a modulation tracking
loop. The transfer function relating the control voltage V
c
(s) to the signal phase
r
(s) as
given earlier in Equation 13.10 is
V
c
(s)
r
(s)
=
s K
d
F(s)
s + K
o
K
d
F(s)
(13.63)
The input signal for FM is of the form
V
r
(t) = V
r
cos
!
c
t +
t
o
m(t
0
) dt
0
(13.64)
i.e.,
r
(t) =
t
o
m(t
0
) dt
0
(13.65)
414 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
so
r
(s) =
1
s
M(s) (13.66)
Thus
V
c
(s) =
K
d
F(s)
s + K
o
K
d
F(s)
M(s) (13.67)
=
1
K
o
H(s) M(s)
The control voltage response is a scaled and ltered version of m(t).
For frequency demodulation, the detailed shape of |H(j!)| is important. Usually, a
at frequency response is desired. This means that a Butterworth-type transfer function
is desirable. A second-order loop with = 0.707 provides the maximally at Butterworth
response, as shown in Figure 13.16.
| | H(j )
= 0.707
/
n 1
Figure 13.16: Butterworth response is obtained when =
1
p
2
.
The loop bandwidth must be at least as large as the bandwidth of the modulation signal,
m(t). The best results are obtained when the loop bandwidth is substantially larger than
the modulation bandwidth. (See Phaselock Techniques, F.M. Gardner, J. Wiley, 2nd ed.,
1979, Ch. 9.)
13.4.2 PLL Response to AM
Let V
r
(t) = V
r
m(t) cos (!
r
t +
r
). To simplify the analysis, suppose that the PD is imple-
mented with an ideal multiplier so that the error voltage V
a
(t) is
V
a
(t) = K
d
m(t) sin
e
(13.68)
The average value of V
a
(t) is
< V
a
(t) >=< m(t) > K
d
sin
e
(13.69)
13.4. APPLICATIONS 415
As long as< m(t) >6= 0, the PD output (after averaging, which is carried out by the low-
pass loop lter) is proportional to sin
e
. This means that loop will lock and will track the
carrier as long as a carrier is present (< m(t) >6= 0). The averaging time should be long
compared to the time-scale over which m(t) varies.
A PLL can be used to generate a local carrier which can be used for coherent demodu-
lation of AM, as shown in Figure 13.17.
V
AM
(t)
PD
V
a
F(s)
V
c
V CO
X
LPF recovered modulation
Figure 13.17: Carrier for coherent demodulation of AM
The 90
phase shift is required because the the PD is assumed to give 0 output voltage
when the reference and VCO signals are in quadrature.
13.4.3 Carrier Recovery
The loop cant track the carrier if it isnt there. In DSB signals, the carrier is suppressed,
so there is no carrier component for the PLL to track. There are two equivalent schemes
for recovering the suppressed carrier using a PLL:
1. Squaring loop Assume V
r
(t) = m(t) cos (!
r
t +
r
) and < m(t) >= 0 (sup-
pressed carrier). A carrier component can be recovered if the signal is squared as
in Figure 13.18: The squared input signal is V
2
DSB
(t) = m
2
(t) cos
2
(!
r
t +
r
). Since
V
DSB
(t)
( )
2
PD F(s) V CO
X
2
LPF recovered modulation
Figure 13.18: Squaring loop.
< m
2
(t) > is non-zero, this signal has a component at twice the carrier frequency,
2!
r
. The loop locks to this double-frequency component. The VCO output is divided
416 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
by 2 to produce a recovered carrier component at frequency !
r
which is then used
to coherently demodulate the DSB signal. The recovered signal at the output of the
divider has a 180
phase ambiguity that results from the unknown initial state of the
divider. This means that the sign of the recovered modulation is arbitrary either
+m(t) or m(t) could be recovered. This cause of the ambiguity can be understood
by noting that the frequency divider produces a cosinusoidal signal whose argument is
1
2
of the argument of the double-frequency signal presented to its input. The double
frequency signal is written as cos(2!
r
t +2
r
+2n) with n any integer. The output of
the divider is cos(!
r
t +
r
+n), which can be written as cos(!
r
t +
r
), depending
on whether n is an even or odd integer.
2. Costas loop refer to Figure 13.19. Without the bottom loop, this would be a
m(t) cos(!
r
t +
r
)
X
X
LPF
I(t) = m(t) cos(
e
)
X
LPF
Q(t) = m(t) sin(
e
)
90
V CO
cos(!
r
t +
o
)
F(s)
Figure 13.19: Costas loop
conventional PLL and the error signal would have an average value of zero, because
< m(t) >= 0. With the second, or quadrature loop, the error signal is m
2
(t) sin(
r
o
) cos(
r
o
). Since < m
2
(t) >6= 0, the error signal is nite and operation is
like the squaring loop. The Costas loop has an important advantage over the squaring
loop, however, because it does not need any circuitry that operates at twice the carrier
frequency. When congured as shown the Costas loop automatically provides in-phase
and quadrature outputs, so the Costas loop acts as a quadrature demodulator.
13.5 Frequency Synthesis with PLLs
The simplest PLL frequency synthesizer is created by adding a frequency divider to the
loop that was considered in Figure 13.1. See Figure 13.20. When the PLL is locked both
of the phase detector input frequencies are equal to the reference frequency, f
r
, as shown.
This means that the frequency at the divider input, which is the VCO output frequency,
must be Nf
r
. The VCO output frequency is an integral multiple of the reference frequency.
The divisor, N, of the N divider is typically programmable. so the output frequency can
be changed by changing the divisor, N. The range of available output frequencies will be
13.5. FREQUENCY SYNTHESIS WITH PLLS 417
f
r
PD
V
a
F(s)
V
c
V CO
f
o
= Nf
r
N
f
r
Figure 13.20: PLL frequency synthesizer.
limited by the tuning range of the VCO. The reference frequency is typically generated by
a stable crystal oscillator, so when the loop is locked, the VCO output frequency will reect
the stability of the reference oscillator. If the reference oscillator frequency tolerance is f,
the VCO output frequency tolerance will be Nf.
When the divider is added to the loop, the transfer functions are modied as follows:
o
(s)
r
(s)
=
K
o
K
d
F(s)
s + K
o
K
d
F(s)/N
(13.70)
e
(s)
r
(s)
=
s
s + K
o
K
d
F(s)/N
(13.71)
The high gain second-order loop transfer function becomes
H(s) =
o(s)
r(s)
=
N(2!ns+s
2
)
s
2
+2!ns+!
2
n
!
n
=
KoK
d
N1
=
2!n
2
(13.72)
The loop parameters !
n
and are functions of N. As N is varied to tune the synthesizer
to dierent frequencies the resulting variations in !
n
and can result in signicant changes
in the loop dynamics. It is necessary to perform a worst case design analysis, i.e., using
N
min
and N
max
, nd !
nmin
, !
nmax
,
min
,
max
, and make sure that transient response and
rejection of spurious signals will be adequate at all frequencies of interest.
13.5.1 Noise and Spurious Signals
In practice, both the reference signal and the VCO signal will be accompanied by noise
and/or additional spurious signals. For example, to model the eect of an imperfect VCO
signal which contains phase noise, the model shown in Figure 13.21. The noise on the VCO
signals phase is represented by the noise signal
No
. The transfer function relating this
418 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
r
PD F(s) V CO +
o
N
No
Figure 13.21: Linear model including VCO phase noise.
noise to the output phase is
o
(s)
No
(s)
=
1
1 + K
o
K
d
F(s)/N
s
(13.73)
=
s
s + K
o
K
d
F(s)/N
s
This is a high-pass frequency response function slow, low frequency changes in the VCO
output phase will be ltered out by the loop and will not reach the output, but fast, high
frequency changes in the VCO output phase will be passed through to the output. This
result can be understood by considering that slow changes in the VCO output phase produce
a slowly varying error voltage which is passed by the loop lter and sent to the VCO control
input. The error signal tunes the VCO to compensate for the original change. On the other
hand a fast, high frequency, change in the VCO phase produces a high frequency error signal
which does not pass through the loop lter. High frequency phase perturbations will not be
ltered out by the loop because the loop cannot follow them. So the high-frequency phase
perturbations appear on the VCO output signal.
The 3 major noise sources in a PLL are:
Nr
the phase noise on the reference signal
Nd
noise or spurious signals at the output of the PD. This includes harmonics of the
reference signal, which are present in the output of most PDs. These harmonics will
modulate the VCO and produce unwanted discrete sidebands in the VCOs frequency
spectrum.
No
the intrinsic VCO phase noise, which has already been discussed.
Figure 13.22 shows a block diagram that includes these three noise sources.
The total output noise signal is
N
= (
Nr
+
Nd
)
K
o
K
d
F(s)
s + K
o
K
d
F(s)/N
+
No
s
s + K
o
K
d
F(s)/N
(13.74)
The loop functions as a low-pass lter for phase noise on the reference signal and for reference
frequency harmonics at the output of the phase detector. The loop functions as a high-pass
lter for VCO phase noise. A typical VCO noise spectrum is shown in Figure 13.23. To
minimize the VCO noise contribution, the loop bandwidth should be as wide as possible,
13.5. FREQUENCY SYNTHESIS WITH PLLS 419
r
+ PD + F(s) V CO +
o
N
Nr
Nd
No
Figure 13.22: Linear model for PLL with noise sources.
-40
-80
-120
-160
10
2
10
3
10
4
10
5
10
6
10
7
Narrow-band
loop response
Wide-band
loop response
Figure 13.23: Typical VCO noise spectrum
because a large loop BW means the high-pass characteristic that lters the VCO noise will
have smaller values at low frequencies. On the other hand, the loop bandwidth should be
signicantly less than the reference frequency in order to reduce spurious modulation of the
VCO due to the reference frequency and its harmonics, which are present in the PD output.
These requirements may conict if the reference frequency is too small.
The level of the sidebands induced on the VCO output due to reference frequency com-
ponents can be estimated using the following equation:
sideband level
carrier level
=
V
ref
K
o
2 !
ref
(13.75)
where V
ref
is the peak voltage value of spurious frequency at the VCO input. V
ref
is related
to the spurious voltage at the output of the phase detector by
V
ref
= V
F(s)|
s=j !
ref
(13.76)
Suppose that the loop lter is the active lter shown in Figure 13.24 and assume that
the op-amp open loop gain is very high. The loop lter transfer function will be
420 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
R
1
+
R
2
C
2
= R
2
C
1
= R
1
C
Figure 13.24: High-gain active loop lter.
F(s) '
s
2
+ 1
s
1
(13.77)
2
= R
2
C (13.78)
1
= R
1
C (13.79)
Usually, !
ref
is >
1
2
, so
F(j!
ref
) '
1
=
R
2
R
1
(13.80)
or, in terms of loop parameters,
F(j!
ref
) '
2 N !
n
K
o
K
d
(13.81)
so
|V
ref
| ' |V |
2 N !
n
K
o
K
d
(13.82)
If equation 13.82 is inserted into Equation 13.75:
sideband level
carrier level
' V
N !
n
!
ref
K
d
(13.83)
Usually !
ref
, N and K
d
are predetermined by other constraints. Only !
n
and, to a lesser
extent, , can be adjusted to diminish the reference frequency sidebands. If the VCO is
noisy, it may not be feasible to make !
n
very small because the VCO phase noise may then
become objectionable. And cannot be much smaller than 0.5 without running into phase
margin problems and excessive ringing. In some cases it may be helpful to add additional
poles to the loop lter to attenuate the reference frequency and its harmonics.
13.5.2 Phase Detectors - Digital
13.5.2.1 Exclusive-OR Phase Detector
Figure 13.25 shows an exclusive-OR phase detector. The two input signals represent the
reference and VCO signals after conversion to logic-level signals. The output of the exclusive-
OR gate is high if, and only if, one of the two inputs is high. In the gure the phases of the
input signals dier by one quarter of a cycle, so
e
= /2. The output signal consists of a
13.5. FREQUENCY SYNTHESIS WITH PLLS 421
Figure 13.25: Exclusive-OR phase detector
pulse stream with 50% duty cycle and twice the frequency the input signals. If the phase
dierence decreases (increases) from the nominal value of /2, the width of the output pulses
will decrease (increase). When the phase dierence is equal to the input signals are always
dierent, and the output is constantly high, and when the phase dierence is zero, the input
signals are always the same and the output is always zero. The lowpass loop lter averages
the phase detector output waveform over many cycles, so the output of the loop lter will
be approximately equal to the mean of the ex-OR output waveform. Hence, only the mean
of the output waveform is important for determining the loop dynamics. The average PD
output value versus phase shift
e
is shown in Figure 13.26.
e
V
m
/2
/2
V
m
2
Figure 13.26: Average output versus phase shift
The exclusive-OR has an operating range of /2. The center of the operating range is
at
r
o
=
e
= /2, i.e., the loop locks with the reference and VCO signals in quadrature.
If the phase error ever enters the region between and 2 where the slope of the PD
characteristic is negative, the loop will lose lock because the error voltage drives the VCO
in the wrong direction, tending to increase the error signal rather than decrease it.
Ideally, the output contains no energy at the input frequency, but it has signicant com-
ponents at twice the reference frequency at also at multiples thereof. The second harmonic
content is maximum at
e
= /2 and has a peak-to-peak amplitude 1.27 times the peak-to-
peak value of the phase detector characteristic. If these reference frequency harmonics are
not suciently attenuated by the loop lter they can produce signicant frequency mod-
ulation of the VCO and potentially objectionable discrete sidebands on the VCO output
signal.
13.5.2.2 Hold-in Range of PLL, !
H
The hold-in range is also called the lock range, tracking range, or synchronization range.
It is the maximum reference frequency range over which the PD stays within its operating
range. Suppose the PD operating range is
m
radians. The output frequency range is then
K
d
K
o
F(0)
m
(13.84)
422 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
or, dening K
L
= F(0) as the DC gain of loop lter,
!
H
= K
d
K
o
K
L
m
(13.85)
The loop will track reference frequencies over a range of !
H
.
13.5.2.3 Set-reset (SR) ip-op
A set-reset ip-op can be used as a phase detector (Figure 13.27). The input signals f
A
and
f
A
f
B
S
R
Q
Figure 13.27: S-R ip-op
f
B
consist of narrow pulses, as in Figure 13.28. Signal B resets the ip-op, and signal A sets
T
A t
t B
t
OUT
Figure 13.28: Signals f
A
and f
B
it. The average output voltage as a function of phase error is shown in Figure 13.29. The
SR ip-op has twice the operating range of the exclusive-OR (). The loop locks with
o
= . The output contains a component at the input frequency which has a maximum
value at
r
o
= . The amplitude of this component is 1.27 times the peak-to-peak value
of the output characteristic.
Trade-os between the exclusive-OR(EX-OR) and the S-R ip-op are:
EX-OR small operating range but no output at fundamental frequency, f
r
S-R Larger operating range but output at fundamental frequency
The S-R ip-op has some inherent problems if the input pulses to the S-R ip-op
have nite width, there will be at spots in the output characteristic. If the input signal
goes away, the ip-op output will remain high (or low). The loop interprets this condition
13.5. FREQUENCY SYNTHESIS WITH PLLS 423
V
/2
2 4
e
V
Figure 13.29: Average output as function of phase error
as a large phase error and changes the VCO frequency. Eventually the VCO will bang up
against its limiting frequency and sit there.
13.5.2.4 Quad-D Phase-Frequency Detector
The quad-D phase-frequency detector (PFD) is a more sophisticated digital phase detector
consisting ip-ops and additional logic. Widely available in integrated circuit form, it
is called a phase-frequency detector because it provides an indication of the sign of the
frequency error when the loop is out-of-lock. The PFD has two output terminals labeled
U and D (for up and down). U and D can be high simultaneously, but only one can
be active (low) at any time. Dene the duty ratio d
u
or d
D
as the fractional time either
terminal is active (low). The U output is only active when the phase error is greater than
0, and the U output is active when the phase error is negative. The phase characteristic
when locked looks like Figure 13.30. So d
u
d
D
looks like Figure 13.31.
A PFD has unique properties. The active phase error range is 360
which is double
that of the S-R ip-op and 4 times as large as the XOR. If the loop is unlocked, only
the U or the D output pulls low (active). The active output indicates the direction of the
frequency error. Both outputs are quiescent at the equilibrium tracking point (
e
= 0),
i.e., if the loop is locked at the center of its tracking range there are no reference frequency
harmonics. In the near vicinity of equilibrium one or the other of the outputs pulls down
with a small duty cycle. Thus the spurious signal is a narrow pulse at the input frequency.
This is desirable, since it is much easier to lter narrow pulses than it is to lter square
waves. There may be some crossover distortion in the PFD characteristic around
e
= 0. If
an input signal transition is missing, the PFD interprets this as loss of lock. Since the PFD
has memory, the eects propagate for more than one cycle.
The PFD has two outputs (U and D) which must be subtracted and then averaged in
order to generate the VCO control voltage. Since each output can be either high or low, the
dierence between the two outputs has three states, positive, zero, or negative. Typically,
these states are used to control a charge pump a constant current source that can source
or sink a current I
o
or be in an o state. The maximum source/sink current is typically on
the order of a few mA.
424 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
d
U
e 2
d
D
e 2
Figure 13.30: Locked phase characteristic
d
U
d
D
e 2 2
Figure 13.31: Quad-D ip-op phase detector characteristic.
13.5. FREQUENCY SYNTHESIS WITH PLLS 425
13.5.3 Examples
A PLL synthesizer has the following parameters:
f
out
= 100 MHz (13.86)
f
ref
= 100 kHz
K
o
= 1 MHz/V )2 x 10
6
rad
Vsec
The PD is an S-R ip-op with output voltage between 0 and 5 V. Consider what the loop
lter attenuation must be at 100 kHz in order to keep the reference frequency sidebands
more than 40 dB down with respect to the carrier (-40 dBc) at the VCO output:
40 dB )
sideband level
carrier level
(13.87)
= 10
40/20
= 0.01
We need 0.01 V
|F(j!
ref
)|
Ko
2!
ref
0.01, where V
=
1
2
(1.27) (5) (13.88)
= 3.175 V
Then we need
|F(j!
ref
)|
2 !
ref
K
o
1
3.175
0.01 (13.89)
=
2.10
5
10
6
1
3.175
0.01
= 6.30 x 10
4
= 64 dB
Suppose the loop lter is a simple RC low-pass lter with F(s) of the form
F(s) =
1
1 +
s
!c
(13.90)
and the loop lter is followed by a DC amplier, as in Figure 13.32.
So
!
c
=
K
o
K
d
K
L
p
2N
(13.91)
426 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
K
L
o
Figure 13.32: Loop lter followed by DC amplier.
To get -40 dB sidebands, we need
|F(j!
ref
)| = 6.3 x 10
4
(13.92)
or, assuming !
ref
!
c
K
L
!
c
!
ref
= 6.3 x 10
4
(13.93)
Then, using
K
d
= 5V/cycle (13.94)
=
5
2
V
rad
K
d
= 2 x 10
6
rad/V-sec
N = 1000
Combining Equations 13.91 and 13.93
!
c
=
K
o
K
d
!
ref
6.3 x10
4
p
2N
= 1.183 10
3
(13.95)
or
f
c
= 188 Hz
If the phase margin is to be 45
, we want
|A(j!)| = 1 when ! ' !
c
(13.98)
Thus
1 =
K
o
K
d
K
L
/N
!
c
p
2
(13.99)
From Equation 13.93:
K
L
=
!
ref
!
c
6.3 10
4
= 0.34 (13.100)
The natural frequency of the loop (loop bandwidth) is
!
n
=
K
o
K
d
K
L
!
c
N
= 1418 (13.101)
or
f
n
= 226 Hz
The damping factor for this loop would be
=
1
2
!
c
!
(13.102)
= .42
The settling time for the loop will be
7
!
n
= 5 ms (13.103)
13.5.4 Pre-scalers
A number of factors may combine to make the simple synthesizer discussed in section 13.5.3
unsuitable. For example: (i) the required channel spacing (f
r
) may be too small to allow
for adequate suppression of spurious components that appear at the PD output; (ii) a small
f
r
may result in a lock-up time which is too long; (iii) programmable dividers may not be
available for operation at very high frequencies. Problem (iii) can be addressed by rst
dividing by a xed-modulus divider (pre-scaler) as shown in Figure 13.33. It is possible
f
r
PD F(s) V CO
f
o
= MNf
r
M N
f
r
Figure 13.33: PLL frequency synthesizer with xed pre-scaler (M) and programmable di-
vider (N).
428 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
to obtain xed-modulus dividers (pre-scalers) that operate well into the microwave region.
The output of the M pre-scaler will be at a lower frequency, where the fully programmable
N can operate. When the loop is locked
f
r
=
f
o
MN
(13.104)
or
f
o
= N (Mf
r
) (13.105)
A pre-scaler allows a loop to operate at higher frequencies but the output frequency can
only be changed in increments of Mf
r
. To get better resolution we must decrease f
r
. This
will tend to make problems (i) and (ii) worse.
13.5.5 Dual Modulus Dividers
A method for obtaining good frequency resolution while operating at high output frequencies
uses a variable modulus pre-scaler. Typically, the variable modulus divider is a high-speed
divider with two choices for the modulus. A synthesizer using the dual modulus scheme is
shown in Figure 13.34 where the high-speed divider can divide by either P or P +Q. The
divide ratio is selected by a signal from the low-speed divider A. Here is a summary of how
f
r
PD F(s) V CO
f
o
= (AQ+PN)f
r
P
P +Q
A N
Figure 13.34: PLL synthesizer with dual modulus pre-scaler. The frequency at the inputs
of the N and A counters is much smaller than the VCO output frequency, so these
counters do not have to be very fast. The dual modulus pre-scaler does have to be fast,
since it accepts the VCO output signal and must divide it by either P or P +Q. The dual
modulus pre-scaler accepts a binary input from counter A, which instructs it to divide by
P or P +Q. Note that N will always be greater than A.
the dual modulus pre-scaler operates:
1. Load N and A counters with N and A, respectively. The dual modulus pre-scaler
is set to divide by P +Q.
(a) After P +Q pulses, both N and A counters are decremented.
(b) Continue until the A counter reaches zero. This will occur after (P +Q)A cycles
of the VCO output signal, f
o
.
(c) The A counter then instructs the dual modulus pre-scaler to divide by P. The
N counter contains N A. (We must have N > A for this scheme to work).
13.5. FREQUENCY SYNTHESIS WITH PLLS 429
(d) Continue until the N counter reaches zero. This takes P(N A) cycles of f
o
.
(e) The N counter outputs 1 pulse. The cycle is complete.
(f) Begin again at step (1).
The total number of f
o
pulses in one complete divide cycle is
(P +Q)A+P(N A) = AQ+PN (13.106)
i.e., for every AQ + PN pulses in, we get 1 pulse out to the PD. Thus, the three counters
together eectively divide by AQ+PN. When the loop is locked
f
o
= [AQ+PN]f
r
(13.107)
Recall that N must be greater than A for the method to work.
A frequently used divide ratio is P = 10, P +Q = 11. Then
f
o
= [10N + A] f
r
(13.108)
Suppose that A can be set to any value from 0 to 9. Then N must be at least 10 to keep
N > A. This means that the lowest output frequency is 100f
r
, but any integer multiple
larger than 100f
r
can be achieved. Thus the 10/11 dual modulus pre-scaler can be used
to obtain integer multiples of f
r
even when the output frequency is too large for a fully
programmable divider to handle.
Consider an example. Suppose that we need a frequency synthesizer to cover the fre-
quency range 100 MHz to 1099 MHz in 1 MHz increments. Since the step size is 1 MHz,
it is necessary to choose f
r
1 MHz. Let f
r
= 1 MHz. Suppose that 1099 MHz is too
fast for a fully programmable divider, however a fast (P = 10, P + Q = 11) dual modulus
pre-scaler is available. Then we use the following parameters:
f
r
= 1 MHz (13.109)
N = 100
A ) 0 to 99
f
o
= [1000 + A] f
r
Notice that the largest frequency at the input to the N and A counters will be 109.9
MHz.
Now, suppose we need to cover 1000.00 to 1000.99 MHz in 10 kHz increments. Then we
need f
r
= 10 kHz. The minimum divide ratio required would be
1000
0.01
= 10
5
(13.110)
=
f
o
(min)
f
r
Since A must run from 0 to 99, N
min
is 100. Parameters that would work for this situation
are:
N = 10000 (13.111)
P = 10
P +Q = 11
430 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
The maximum frequency at the input to the fully programmable dividers would be 0.100099
MHz in this case.
Consider an example using parameters available from a commercially available synthe-
sizer chip. The chip has a 5-bit A counter, so that A can range from 0 to 31. The N counter
is a 13-bit counter and N can be set to any value in the range 3 to 8191. The dual-modulus
pre-scaler uses P = 8, P +Q = 9. The output frequency will be
f
o
= (8N +A)f
r
,
where 0 A 31, 3 N 8191, and N > A. Suppose that it is necessary to design a
synthesizer that covers 230240 MHz in 100 kHz steps and we wish to select possible values
for N and A in order to tune the synthesizer to a particular output frequency. For example,
to tune the synthesizer to the middle of the range (235 MHz) the eective divide ratio must
be 235/0.1 = 2350 = (8N + A). The N counter can be set to the integer part of 2350/8,
which is N = 293. Then, 8N = 8(293) = 2344, so the A counter must be set to A = 6. To
tune to 239.6 MHz, (N, A) = (299, 4) would work, as would (298, 12) or (297, 24).
13.6. REFERENCES 431
13.6 References
1. Best, Roland E., Phase Locked Loops, McGraw Hill, 1984.
2. Egan, William F., Frequency Synthesis by Phaselock, J. Wiley & Sons, New York,
1981.
3. Gardner, Floyd M., Phaselock Techniques, (2nd ed.), J. Wiley & Sons, New York,
1979.
4. Manassewitsch, Vadim J., Frequency Synthesizers, Theory and Design, J. Wiley &
Sons, New York, 1987.
5. Smith, Jack, Modern Communications Circuits, McGraw Hill, 1986.
432 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)
13.7 Homework Problems
1. Consider the rst-order PLL shown in Figure 13.35. The gain constants for the VCO
and phase detector are K
o
= 10
7
radians/Volt-second, and K
d
= 10 Volt/radian,
respectively. The center frequency (free-running frequency) of the VCO is f
c
= 35.0
MHz. Assume that the loop is locked.
Figure 13.35: First-order PLL frequency synthesizer.
(a) If N = 10 and the reference frequency f
r
= 1.0 MHz, what is the output frequency,
f
o
?
(b) What is the steady-state output voltage from the phase detector?
(c) What is the steady-state phase error
e
=
r
o
? Give your result in degrees.
Appendix A
Circuit Models for BJT and FET
A.1 Hybrid-pi equivalent circuit for bipolar junction tran-
sistor (BJT)
I
C
I
E
+
-
B
E
C
V
be
I
B
Figure A.1: BJT
The following approximate relationship is useful when the base-emitter junction is for-
ward biased:
I
C
= I
S
exp[
V
be
V
T
], (A.1)
where
V
T
=
kT
q
' 25mV at room temperature. (A.2)
The base current, I
b
, is related to the collector current by
I
b
=
I
C
. (A.3)
These relationships can be used to solve for the small-signal input resistance, r
(see Fig-
433
434 APPENDIX A. CIRCUIT MODELS FOR BJT AND FET
ure A.2)
r
= [
@I
b
@V
be
]
1
V
be
=V
beq
(A.4)
=
V
T
I
CQ
r
'
.025
I
CQ
and the small-signal transconductance
g
m
= [
@I
C
@V
be
]
V
be
=V
beq
(A.5)
=
I
CQ
V
T
=
r
g
m
' 40I
CQ
where I
CQ
is the quiescent collector current.
A useful linear model for the behavior of small high frequency signals superimposed on
the DC bias point is the small-signal hybrid-pi model shown in Figure A.2. Typically r
o
r
x
C
+
V
g
m
V
r
o C
o
B C
E
Figure A.2: Hybrid-pi small-signal model for BJT
(tens to hundreds of k), r
x
(a few tens of ), and r
>r
o
. Since r
o
and r
are relatively
large resistances, and each is shunted by a capacitance, r
o
and r
is often given as C
ob
, which is the output capacitance
in the common-base conguration. Typical values for C
+C
, (A.6)
A.2. HYBRID-PI EQUIVALENT CIRCUIT FOR FIELDEFFECT TRANSISTOR (FET)435
and f
T
is the frequency where the short-circuit current gain has a magnitude of unity. The
value of f
T
depends on g
m
and, therefore, on how the transistor is biased.
The -3dB frequency for the short-circuit current gain is denoted by f
and is given by
f
=
1
2
1
r
(C
+C
)
(A.7)
Note that the -3dB frequency and the unity gain frequency are related as follows:
f
T
f
= (A.8)
For f < f
the short-circuit
current gain is approximately f
T
/f.
A.2 Hybrid-pi equivalent circuit for eld eect transistor
(FET)
Figure A.3 looks very similar to the BJT model (Figure A.2). The gate-source resistance,
r
gs
, is generally large compared to the impedance of C
gs
. When this condition holds (i.e.
at suciently high frequencies) r
gs
can be omitted from the model.
C
gs
+
v
gs
r
gs
C
dg
g
m
v
gs
r
ds C
ds
G D
S
Figure A.3: Hybrid-pi equivalent circuit for the FET
For a junction FET (JFET), the transconductance is proportional to the square root
of the drain current, I
D
. The proportionality constant depends on the saturation drain
current, I
DSS
, and the pincho voltage, V
P
, i.e.
g
m
=
2
|V
P
|
I
DSS
I
D
. (A.9)
The parameters I
DSS
and V
P
are usually available from device data sheets.
A.3 Large-signal transconductance of a BJT with sinu-
soidal V
be
When a BJT is driven with a sinusoidal signal such that the base-emitter voltage swing ap-
proaches or exceeds a few tens of mV, the collector current waveform becomes non-sinusoidal.
436 APPENDIX A. CIRCUIT MODELS FOR BJT AND FET
It is useful to study the eect of sinusoidal base-emitter voltage swing on the collector cur-
rent waveform. The following approximate relationship between base-emitter voltage V
be
and collector current I
C
can be used as a starting point
I
C
= I
S
e
V
be
q/kT
(A.10)
where kT/q is approximately 25mV for T=290K, and I
S
is a constant. Decompose the
base-emitter voltage and collector current into quiescent and time-varying components, i.e.,
I
C
= I
DC
+ i
C
(A.11)
V
be
= V
DC
+ v
be
(A.12)
where lower-case letters refer to the time-varying component of the quantity. Later, we will
make the assumption that the transistor bias network acts to keep the DC component of
the collector current at a nearly constant value. Suppose that the time-varying component
of the base-emitter voltage is sinusoidal, i.e.,
V
be
= V
DC
+ v
1
cos!t (A.13)
and let x = v
1
q/kT = v
1
/25mV (at room temperature). Then
I
C
= I
S
exp[V
DC
q/kT] exp[ xcos!t] (A.14)
The term exp[ x cos !t] is a non-sinusoidal periodic function of time and can be expanded
in a Fourier series. The series is
exp[ x cos !t] = I
o
(x) + 2
1
n=1
I
n
(x) cos(n!t) (A.15)
where the coecients I
n
(x) are values of the modied Bessel function of the rst kind. Using
this relationship, the collector current waveform can be written as
I
C
= I
DCo
[I
o
(x) + 2
n=1
I
n
(x) cos(n!t)] (A.16)
Here I
DCo
is the DC component of collector current when the time-varying component of
the input signal is equal to zero (v
1
= 0). The DC component of the collector current when
the time-varying component of the input signals is not zero is given by
I
DC
= I
DCo
I
o
(x). (A.17)
This function is plotted in Figure A.4, which shows that I
o
(x) grows very rapidly when the
base-emitter voltage swing exceeds a few tens of mV. This analysis has assumed that the DC
component of the base-emitter voltage is held constant. Practical amplier and oscillator
circuits will employ either constant current source bias for I
DC
or negative feedback (by
including a resistor in series with the emitter and ground) to force I
DC
to be nearly constant.
In such cases the DC component of V
be
(denoted by V
DC
) will decrease with increasing v
1
such that I
DC
is more or less independent of v
1
. With constant-current bias we can write:
I
C
= I
DC
[1 + 2
1
n=1
I
n
(x)
I
o
(x)
cos(n!t)] (A.18)
A.3. LARGE-SIGNAL TRANSCONDUCTANCE OF ABJT WITHSINUSOIDAL V
BE
437
2 4 6 8 10
100
200
300
400
500
600
700
x
I
0
(x)
Figure A.4: The function I
o
(x) governs how the DC component of the collector current
grows as the amplitude of sinusoidal base-emitter voltage increases (x = v
1
q/kT), assuming
that the DC component of the base-emitter voltage is held constant.
where I
DC
is treated as a constant. Examination of A.18 shows that the collector current
waveform has components at DC, the fundamental frequency, and harmonics of the funda-
mental frequency. The relative amplitude of the harmonics and fundamental gives an indica-
tion of how sinusoidal the collector current waveform will be. Assuming constant-current
bias, the amplitude of the fundamental component is proportional to 2I
1
(x)/I
0
(x). This
quantity is plotted in Figure A.5. The amplitude of the fundamental approaches a constant
when the base-emitter voltage swing exceeds 25 mV (i.e. for x > 1). The relative strengths
of the second and third harmonics to the fundamental (I
2
(x)/I
1
(x) and I
3
(x)/I
1
(x), respec-
tively) are also shown in Figure A.5. Notice that the harmonic amplitudes grow rapidly
5 10 15 20
0.25
0.5
0.75
1
1.25
1.5
1.75
2
x
2I
1
(x)
I
0
(x)
I
2
(x)
I
1
(x)
I
3
(x)
I
1
(x)
Figure A.5: 2I
1
(x)/I
o
(x) is the relative amplitude of the fundamental component of the
collector current. The other curves show the ratio of second and third harmonic amplitudes
to the fundamental amplitude.
when x is greater than 1 (v
1
> 25 mV). At large values of x the fundamental amplitude
approaches twice the DC bias current, and the harmonic amplitudes approach that of the
fundamental. Thus it becomes apparent that the base-emitter voltage swing must be kept
as small as possible if the collector current waveform is to be sinusoidal.
This approximate analysis of the nonlinear characteristics of the BJT can also help us to
gain some intuitive feeling for the saturation mechanism that limits the growth of oscillations
in self-limiting oscillators, and for the reduction in apparent gain that results when a BJT
amplier is driven by a large amplitude input signal. For these purposes we consider only
438 APPENDIX A. CIRCUIT MODELS FOR BJT AND FET
the collector current component at the fundamental frequency, under the assumption that
a resonant network eectively removes the harmonics. Then we can compute a large-signal
transconductance for the transistor. With a sinusoidal input signal the transconductance
is simply the ratio of the amplitudes of the fundamental component of i
C
and v
1
. The
small-signal transconductance can be obtained in the limit as x approaches 0, i.e.,
g
m
= lim
x!0
I
DC
2 I
1
(x)
v
1
I
0
(x)
=
I
DC
kT/q
(A.19)
At room temperature kT/q = 25 mV, so
g
m
=
I
DC
25 mV
40I
DC
(A.20)
The large signal transconductance is
G
m
(x) = I
DC
2 I
1
(x)
v
1
I
0
(x)
= I
DC
q
kT
2 I
1
(x)
x I
0
(x)
= g
m
2
x
I
1
(x)
I
0
(x)
(A.21)
The ratio of the large signal to small-signal transconductance is shown in Figure A.6. This
2 4 6 8 10
0.2
0.4
0.6
0.8
1
G
m
(x)
g
m
x
Figure A.6: Ratio of large signal to small-signal transconductance
result shows that the large signal transconductance of the transistor decreases from the
small-signal value as the base-emitter voltage swing (x) increases. Thus in an oscillator
application, as the oscillation amplitude grows, the eective transconductance decreases.
The oscillations will continue to grow until the transconductance has been reduced to a value
that causes the (large-signal) loop gain to be equal to 1. The large-signal transconductance
is also important in determining the behavior of stable (non-oscillating) transistor circuits
under large signal conditions. For example, the decrease of transconductance under large
signal excitation is responsible for gain compression in transistor ampliers when large
signals are applied. These eects will discussed in some detail in Chapter 12.
Appendix B
Three-winding Transformer
The three-winding transformer nds many applications in RF circuits. Examples include
balanced-to-unbalanced conversions (baluns), power splitters and combiners and voltage
adders and subtractors. This chapter will explore some of these applications. Through-
out this discussion it will be assumed that the transformer is ideal in other words, we
assume that the device is lossless, that the windings are perfectly coupled, and that the
self-inductance of each winding is innite. In practice, perfect coupling can be approxi-
mated by constructing all three windings on a common high-permeability core. Innite
self-inductance is a good approximation if the inductance of each winding is large compared
to the impedance connected across the winding
The schematic representation for an ideal three-winding transformer is shown in Figure
B.1. Two parallel lines running alongside the three windings remind us that a practical
implementation of the transformer requires all three coils to be wound on a common core
that has large relative magnetic permeability so that all three coils are linked by the same
magnetic ux. By virtue of the perfect coupling between windings, the voltages across
windings b and c can be written in terms of the voltage across winding a as follows:
V
b
=
N
b
N
a
V
a
, V
c
=
N
c
N
a
V
a
(B.1)
Since the transformer is assumed to be lossless, the total power delivered to the device must
be zero, i.e.
<(V
a
I
a
+V
b
I
b
+V
c
I
c
) = 0,
where <() takes the real part of its argument. Using the voltage relationships, the voltages
can be written in terms of the voltage across any one of the windings. Using V
a
the lossless
condition becomes
<(V
a
(I
a
+
N
b
N
a
I
b
+
N
c
N
a
I
c
)) = 0.
The lossless condition must hold for any applied voltage, V
a
, so the currents must satisfy
N
a
I
a
+N
b
I
b
+N
c
I
c
= 0. (B.2)
Equations B.1 and B.2 are called the ideal transformer equations, and they fully characterize
the ideal 3-winding transformer.
439
440 APPENDIX B. THREE-WINDING TRANSFORMER
I
a
+
V
a
I
b
+
V
b
I
c
+
V
c
N
a
N
b
N
c
Figure B.1: Schematic representation of a three-winding transformer. The polarity of the
windings is indicated by a dot. The dot convention is such that when current ows into the
dot in each winding, the resulting magnetic uxes within the core add constructively. The
turns ratio is N
a
: N
b
: N
c
for windings a, b, and c, respectively.
kk
V
4
+
Z
4
+
V
b
V
1
+
Z
1
I
a
+
V
a
V
2
+
Z
2
I
b
V
c
+
Z
3
I
c
+
V
3
N : 1
1
2
3
4
Figure B.2: The three-winding transformer congured as a 4-port device.
B.1. CONJUGATE PORTS 441
For simplicitys sake, well consider a system with N
a
: N
b
: N
c
= N : N : 1, and with
windings a and b connected so that they share a common terminal as shown in Figure B.2.
This conguration results in a 4-port device.
Taking into account the current directions dened in Figure B.2, the ideal transformer
relations can be written
V
a
= V
b
= N V
c
(B.3)
N(I
a
+ I
b
) = I
c
(B.4)
The loop equations are
V
1
+ Z
1
I
a
+ V
a
+ (I
a
I
b
)Z
4
+ V
4
= 0 (B.5)
V
4
+ (I
b
I
a
)Z
4
+ V
b
+ Z
2
I
b
+ V
2
= 0 (B.6)
V
c
+ I
c
Z
3
+ V
3
= 0. (B.7)
Use B.3 andB.7 in B.5 and B.6 to eliminate V
a
, V
b
, V
c
:
I
a
(Z
1
+ Z
4
) + I
b
(Z
4
) + I
c
(N Z
3
) = V
1
V
4
NV
3
(B.8)
I
a
(Z
4
) + I
b
(Z
2
+ Z
4
) + I
c
(N Z
3
) = V
4
V
2
NV
3
(B.9)
Equations B.4, B.8 and B.9 can be written in matrix form:
Z
1
+Z
4
Z
4
NZ
3
Z
4
Z
2
+Z
4
N Z
3
N N 1
I
a
I
b
I
c
V
1
V
4
N V
3
V
4
V
2
N V
3
0
(B.10)
B.1 Conjugate Ports
We now consider some useful special cases. It is possible to select the termination impedances
so that there is isolation between ports. For example, suppose it is desired to isolate port 4
from port 3 so that there will be no output at port 4 when a signal is applied at port 3. To
proceed, we turn o V
1
, V
2
, V
4
and nd the condition for no output at port 4.
The current at port 4 will be I
a
I
b
. Using Cramers Rule to solve for I
a
:
I
a
=
D
a
D
=
N V
3
Z
4
N Z
3
N V
3
Z
4
+Z
2
N Z
3
0 N 1
D
where D is the system determinant and D
a
is the determinant of the matrix formed by
replacing the rst column with the source column vector (the right-hand side of equation
442 APPENDIX B. THREE-WINDING TRANSFORMER
B.10). Evaluating D
a
:
D
a
= N V
3
[(Z
4
+ Z
2
) N
2
Z
3
] + Z
4
[N V
3
] + N Z
3
[N
2
V
3
] (B.11)
= V
3
N(2 Z
4
+ Z
2
)
Similarly, I
b
is obtained from:
I
b
=
D
b
D
=
Z
1
+Z
4
N V
3
N Z
3
Z
4
N V
3
N Z
3
N 0 1
D
D
b
= (Z
1
+ Z
4
)[N V
3
] + N V
3
[Z
4
N
2
Z
3
] + N Z
3
[N
2
V
3
] (B.12)
= V
3
N (2 Z
4
+ Z
1
)
The current owing in the impedance Z
4
is
I
a
I
b
= (D
a
D
b
) /D (B.13)
= V
3
N (Z
2
Z
1
) /D.
The current in Z
4
will be zero if Z
1
= Z
2
. Thus, port 4 is isolated from port 3 if Z
1
= Z
2
.
It is left as an exercise to show that the isolation works both ways, i.e. that port 3 will be
isolated from port 4 when Z
1
= Z
2
. When ports 3 and 4 are isolated, we say that ports 3
and 4 are conjugate.
Suppose we wish to isolate port 1 from port 2. To nd the necessary constraint on the
terminating impedances, set V
1
= V
3
= V
4
= 0 and solve for I
a
due to V
2
. Setting I
a
= 0
will yield the desired condition:
I
a
=
D
a
D
=
0 Z
4
N Z
3
V
2
Z
4
+Z
2
N Z
3
0 N 1
D
= [Z
4
V
2
+ N Z
3
(N V
2
)] /D (B.14)
= V
2
[Z
4
N
2
Z
3
] /D
Thus, port 1 will be isolated from port 2 if
Z
4
= N
2
Z
3
(B.15)
B.2. HYBRID TRANSFORMER 443
Again, it is left as an exercise to show that condition B.15 also causes port 2 to be isolated
from port 1, i.e. I
b
= 0 with V
1
6= 0 and V
2
= V
3
= V
4
= 0. That is, the isolation works
both ways. When ports 1 and 2 are isolated, we say that ports 1 and 2 are conjugate.
If ports 3 and 4 are conjugate and ports 1 and 2 are conjugate then the system is called
the biconjugate transformer.
B.2 Hybrid Transformer
Recall that the conditions for a biconjugate system are
Z
1
= Z
2
(B.16)
Z
3
=
1
N
2
Z
4
. (B.17)
The system of equations for the biconjugate transformer is
Z
1
+Z
4
Z
4
Z
4
/N
Z
4
Z
4
+Z
1
Z
4
/N
N N 1
I
a
I
b
I
c
V
1
V
4
N V
3
V
4
V
2
N V
3
0
(B.18)
In addition to isolation between ports, it is sometimes desired to have all 4 ports matched
for maximum power transfer. This means that a conjugate match exists between each port
and its terminating impedance. Consider the input impedance seen by the source that is
connected to port 1. For a conjugate match at port 1, we require that Z
in1
= Z
1
. The
impedance Z
in1
is equal to the voltage at terminal 1, which is V
1
I
a
Z
1
, divided by the
current owing into this terminal, which is I
a
. Thus, Z
in1
=
V1
Ia
Z
1
. When solving for I
a
we set V
2
, V
3
, V
4
= 0 in the equations B.28, because only port 1 is driven when calculating
Z
in1
. Solving for I
a
:
I
a
=
V
1
Z
4
Z
4
/N
0 Z
4
+Z
1
Z
4
/N
0 N 1
D
I
a
= (2Z
4
+Z
1
)V
1
/D (B.19)
where
D = (2Z
4
+Z
1
)
2
(B.20)
Thus
V
1
I
a
= 2Z
4
+Z
1
. (B.21)
The input impedance is then
Z
in1
= 2Z
4
. (B.22)
For a conjugate match at port 1 we need
Z
in1
= Z
1
(B.23)
444 APPENDIX B. THREE-WINDING TRANSFORMER
so
Z
4
=
1
2
Z
1
(B.24)
Notice that the combination of the biconjugate conditions (equations B.17) and equation
B.24 determine the terminations at ports 2, 3, and 4 once the termination at port 1 is
specied. There are no additional degrees of freedom left once biconjugacy and a conjugate
match at one port are enforced. This implies that the input impedances at the other ports
(2, 3, and 4) are determined once the conjugate match at port 1 is enforced. It is not hard to
show that a biconjugate system with one port matched for maximum power transfer will au-
tomatically be conjugate matched at all 4 ports. How wonderful! A biconjugate transformer
that is conjugately matched at all 4 ports is called a hybrid transformer. The terminating
impedances for a hybrid transformer system must satisfy the following conditions:
Z
1
= Z (B.25)
Z
2
= Z (B.26)
Z
3
= Z
/2N
2
Z
4
= Z
/2
If port 1 is terminated with a resistance, R. Then port 2 must also be terminated with R,
port 3 must be terminated with R/(2N
2
) and port 4 must be terminated with R/2. Notice
that if the turns ratio is chosen to be N =
1
p
2
then Z
1
= Z
2
= Z
3
= R and Z
4
= R/2. We
will assume that the port terminations are resistive and that N =
1
p
2
in the discussions to
follow.
B.3 Applications of the Hybrid Transformer
B.3.1 Power Splitters
B.3.1.1 180-degree splitter
Suppose a signal is applied to port 3 as in Figure B.3.
R/2
+
V
b
R
I
a
+
V
a
R
I
b
V
c
+
R
I
c
+
V
3
1 :
p
2
1
2
3
4
Figure B.3: A hybrid transformer driven at port 3.
B.3. APPLICATIONS OF THE HYBRID TRANSFORMER 445
The system of equations for this hybrid transformer is
3
2
R
1
2
R
1
p
2
R
1
2
R
3
2
R
1
p
2
R
1
p
2
1
p
2
1
I
a
I
b
I
c
1
p
2
V
3
1
p
2
V
3
0
(B.27)
Since port 4 is conjugate to port 3, we already know that there is no response at port 4.
Hence, the current leaving port 4 must be zero, which implies that I
b
= I
a
. Lets solve for
I
a
:
I
a
=
1
p
2
V
3
1
2
R
1
p
2
R
1
p
2
V
3
3
2
R
1
p
2
R
0
1
p
2
1
D
.
The solution is
I
a
= I
b
= V
3
p
2
4R
.
Denoting the voltages at terminals 1 and 2 by V
1
and V
2
, respectively, we have V
1
= I
a
R
and V
2
= I
b
R, or
V
1
=
p
2
4
V
3
, V
2
=
p
2
4
V
3
.
Hence, the voltages at ports 1 and 2 have equal amplitudes, and are 180 degrees out of
phase. The power delivered to terminations 1 and 2 is
P
1
= P
2
=
1
2
|V
1
|
2
R
=
1
16
|V
3
|
2
R
.
Note that P
1
and P
2
are each exactly one half of the power available from the source
connected to port 3. This conguration of the hybrid transformer is called the 180-degree
power splitter.
B.3.1.2 In-phase splitter
Suppose a signal is applied to port 4 as in Figure B.4. The system of equations for this
hybrid transformer is
3
2
R
1
2
R
1
p
2
R
1
2
R
3
2
R
1
p
2
R
1
p
2
1
p
2
1
I
a
I
b
I
c
V
4
V
4
0
(B.28)
Since port 4 is conjugate to port 3, we already know I
c
= 0. Lets solve for I
a
:
I
a
=
V
4
1
2
R
1
p
2
R
V
4
3
2
R
1
p
2
R
0
1
p
2
1
D
.
446 APPENDIX B. THREE-WINDING TRANSFORMER
V
4
+
R/2
+
V
b
R
I
a
+
V
a
R
I
b
V
c
+
R I
c
1 :
p
2
1
2
3
4
Figure B.4: A hybrid transformer driven at port 4.
The solution is
I
a
= V
4
1
2R
.
Similarly,
I
b
= +V
4
1
2R
.
The voltages at terminals 1 and 2 are V
1
= I
a
R and V
2
= I
b
R, or
V
1
=
1
2
V
4
, V
2
=
1
2
V
4
.
Hence, the voltages at ports 1 and 2 have equal amplitudes, and are in phase. The power
delivered to terminations 1 and 2 is
P
1
= P
2
=
1
2
|V
1
|
2
R
=
1
8
|V
4
|
2
R
.
Note that P
1
and P
2
are each exactly one half of the power available from the source
connected to port 4. This conguration of the hybrid transformer is called the in-phase
power splitter.
B.3.2 Sum or Dierence Combiners using a Hybrid Transformer
The hybrid transformer has the property that the phase shift between three of the four
ports is zero and the phase shift to the remaining port will be 180
3
2
R
1
2
R V
1
1
2
R
3
2
R V
2
1
p
2
1
p
2
0
D
.
The solution is
I
c
=
p
2
4R
(V
1
V
2
)
The voltage at port 3 is V
3
= I
c
R, so
V
3
=
p
2
4
(V
1
V
2
).
The outport at port 3 is proportional to the dierence between the applied signals.
The output at port 4 will be proportional to I
a
I
b
. Using the fact that ports 1 and 2
are conjugate, and noting that generators 1 and 2 each see a matched impedance, R, it is
obvious that I
a
=
V1
2R
and I
b
=
V2
2R
. Hence
V
4
= (I
a
I
b
)
R
2
=
1
4
(V
1
+V
2
).
The output at port 4 is proportional to the sum of the applied signals.
B.4 References
1. Smith, Jack, Modern Communications Circuits, Second Edition, WCB/McGraw-Hill,
1998.
2. Sartori, Eugene P., Hybrid Transformers, IEEE Transactions on Parts, Materials
and Packaging, Vol. PMP-4, No. 3, September 1968, pp 59-66.
448 APPENDIX B. THREE-WINDING TRANSFORMER
Appendix C
Appendix: Useful Constants and
Trigonometric Identities
speed of light in free-space c = 2.998 10
8
m s
1
permittivity of free-space
0
= 8.854 10
12
F m
1
permeability of free-space
0
= 4 10
7
H m
1
Boltzmann constant k = 1.381 10
23
J K
1
elementary charge q = 1.602 10
19
C
Planck constant h = 6.626 10
34
J s
Table C.1: Some useful constants
449
450APPENDIXC. APPENDIX: USEFUL CONSTANTS ANDTRIGONOMETRIC IDENTITIES
sin a =
e
ja
e
ja
2j
cos a =
e
ja
+e
ja
2
tan a =
sin a
cos a
sec a =
1
cos a
csc a =
1
sin a
sin(a b) = sin a cos b cos a sin b
cos(a b) = cos a cos b sin a sin b
tan(a b) =
tan atan b
1tan a tan b
sin a sin b =
1
2
cos(a b)
1
2
cos(a +b)
sin a cos b =
1
2
sin(a +b) +
1
2
sin(a b)
cos a cos b =
1
2
cos(a +b) +
1
2
cos(a b)
cos
2
a =
1
2
(1 + cos(2a))
sin
2
a =
1
2
(1 cos(2a))
cos
3
a =
3
4
cos a +
1
4
cos(3a)
sin
3
a =
3
4
sin a
1
4
sin(3a)
sinh a =
e
a
e
a
2
cosh a =
e
a
+e
a
2
tanh a =
sinh a
cosh a
=
e
a
e
a
e
a
+e
a
Table C.2: Some trigonometric identities.