1 Frequency Response of Electronic Circuits

Frequency Response of
Electronic Circuits
Historically, electronic circuits were characterized by exciting them with an oscillator’s
output, and then with an oscilloscope determing how the circuit affected the magnitude and
phase of it’s input signal. This technique is typically only useful for linear circuits. For
non-linear circuits containing transistors, it is useful when the signals are small enough so
the transistors can be adequately characterized by their operating points and small linear
changes about their operating points; that is: the non-linear transistors can be accurately
described using small-signal analysis. The use of this technique for describing electronic
circuits has led to the development of frequency-domain techniques for characterizing lin-
ear, and weakly non-linear, systems, a technique which is currently used ubiquitously in sys-
tem analysis.

1.1 Frequency Response of Linear Systems

Consider a linear system whoose transfer function, H ( s ) (in the Laplace domain) is
known, being excited by an input signal having Laplace transform X in(s) . The Laplace
transform of the output signal, X out(s) , is given by

X out(s) = H ( s )X in(s) (1.1)

In the time domain, assuming x ( t ) is the inversere Laplace Transform of X in ( s ) , and h ( t )
is the inversere Laplace Transform of H ( s ) (often called its impulse response), we have

x out(t) = x in(t) • h ( t ) (1.2)

That is, the output signal is the convolution of the input signal with the impulse response of
the system’s transfer function. The convolution operation is defined as
∞
x in(t) • h ( t ) = ∫ xin(τ)h ( t – τ ) dτ (1.3)
–∞

The merit of frequency-domain analysis is that for particular inputs it is often easier to use
(1.1) and the inverse Laplace Transform to calculate the expected output of a circuit rather
than (1.3). Examples of this are when the inputs are pure sinusoids, or step inputs.

© K.W. Martin, 2001 March 15, 2001

 2 Frequency Response of Electronic Circuits

Sinusoidal Inputs
The output of an oscillator can often be characterized as a sinusoidal signal given by

( e jωin t + e – j ωin t )
x in(t) = A in cos ( ω in t ) = A in -------------------------------------- (1.4)
2
where e x is defined as

e x = cos ( ωt ) + j sin ( ωt ) (1.5)

and j is defined as

j = –1 (1.6)
is a purely imaginary number of unit magnitude.
In the frequency domain, X in(s) consists of two impulses, one at

s = jω in (1.7)
and one at

s = – jω in (1.8)
with each impulse having a magnitude A in ⁄ 2 .
The output of a linear system having such an input, also consists of two impulses in the
frequency domain, but now each impulse will be multimplied by H ( s ) where s = jω in for
the positive frequency impulse, and s = – jω in for the negative frequency impulse. After
taking the inverse Laplace Transform of the output signal, it is seen that
A in jω t
- ( e in H(iω in) + e – j ωin t H(– i ω in) )
x out(t) = -------
2
A in
= -------- H(iω in) ( e jωin t e jφ + e – j ωin t e – j φ )
2 (1.9)
A in
= -------- H(iω in) ( e j ( ωin t + φ ) + e – j ( ωin t + φ ) )
2
= A in H(iω in) cos ( ω in t + φ )

where, the Transfer function H(jω in) has been writtin in terms of its magnitude response and
phase response; that is

H(jω in) = H(jω in) e jφ (1.10)

and φ = ∠H(jω in) .
The development just presented has been given in terms of the Laplace Transform for
the particular case where s = jω in . It is often presented in terms of Fourier Transforms,
which are equivalent to Laplace Transforms with s = jω in , but in the notation, the j sym-

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 3

bol denoting that the frequency is complex is usually not shown. For examaple, (1.10) would
normally be written as

H(ω in) = H(ω in) e jφ (1.11)

where φ = ∠H(ω in) .
The interpretation of (1.11) is as follows: when a linear electronic circuit is excited with
the output of a sinusoidal oscillator, the output of the circuit will also be a sinusoid at the
same frequency, but it’s magnitude and phase will be different. In particular: the magnitude
of the output sinusoid will be equal to the magnitude of the input sinusod multiplied by
H(ω in) where H(ω in) is the Fourier Transform of the circuit commonly called the Transfer
Function. In addition, the phase difference between the output sinusoid and the input sinu-
soid will be equal to φ where again φ = ∠H(ω in) .

Example 1.1 Consider a linear circuit having a transfer function given by

A0
H(s) = ---------------
- (1.12)
s
1 + ------
ω0
where A 0 = 10 and ω 0 = 2π × 100 rad./s. Find the magnitude and phase
of the output sinusoid assuming the input sinusoid has a peak voltage of
1V for the frequencies 10Hz, 100Hz, and 1kHz.
Solution The transfer function given by (1.12) is often called a first-order
low-pass transfer function and is the most-commonly encountered transfer
function in electronic circuits. It is an often used model for an operational
amplifier. We can re-write (1.12) with s = jω using Fourier Transforms as

A0 A0
H(ω) = -----------------
- = ---------------------------- e jφ (1.13)
ω ω 2
1 + j ------
ω0 1 +  ------
 ω 0

where
ω
φ = – tan–1  ------ (1.14)
 ω 0

Note that

A0
H(ω) = ---------------------------
- (1.15)
 ω 2
1 + ------
 ω 0

Using (1.14) and (1.15), for the case f = 100Hz , where

© K.W. Martin, 1998 March 15, 2001

 4 Frequency Response of Electronic Circuits

ω = 2πf = 628.3rad , we have

10 10 10
H(ω) = --------------------------------- = -------------------------- = ------------------------------ = 9.95 = 19.96dB (1.16)
1 +  ----------- 1 +  ----
2πf 2 f 2 10  2
 --------
 100-
 2πf 0  f 0 1 +

where converting a gain to dB implies taking the log (base 10) and multi-
plying by 20. The phase shift is also given by

φ = – tan–1  --------- = – 0.00997rad = – 5.7°

10
(1.17)
 100

The fact that the phase difference between the input and output is nega-
tive, implies the output sinusoid comes after the input sinusoid by – 5.7° , or
in other words ‘lags’ the input sinusoid. For f in = 100Hz , we have

10
H(ω) = ------- = 0.7071 × 10 = 20dB – 3.01dB (1.18)
2
and

φ = – tan–1 ( 1 ) = – 0.785rad = – 45° (1.19)

Note that at f in = f 0 or equivalently, for ω in = ω 0 , the magnitude gain is
down by -3dB, and the phase is delayed by – 45 ° . Finally, for f in = 1kHz ,
we have

10 10 10
H(ω) = -------------------------- = --------------------------------- = ------------- = 20dB – 20.04dB = – 0.04dB (1.20)
10.05
1 +  ----
f 2  1000 
2
 f 0 1 + -----------
 100  -

and

φ = – tan–1  ------------ = – 1.47rad = – 84°

1000
(1.21)
 10 

Thus, the gain is approximately 20dB less than its low-frequency value
and the phase shift is approaching – 90° .

Step Response of First-Order Circuits

Another common means of characterizing linear circuits is to excite them with step
inputs. This would normally be done in practice by exciting them with a square wave having
a low-enough frequency (or equivalently long-enough period), so the circuit settles between
edges, and therefore each edge effectively excites the circuit similar to a step input.

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 5

Consider a step input x in(t) = A in u(t) where u ( t ) , the step-input function, is defined as

0, t ≤ 0
u(t) =  (1.22)
1, t > 0

The step function u ( t ) has a Laplace Transform given by

1
U ( s ) = --- (1.23)
s
Therefore, the Laplace Transform of the output of a linear system having transfer function
T ( s ) and a step input is given by
H(s)
X out(s) = A in ------------ (1.24)
s
This is normally easy to calculate especially if H ( s ) is expressed as a sum of first-order
terms using residues.
Consider the special case were a first-order low-pass filter has a transfer function given
by

A0
H(s) = ---------------
- (1.25)
s
1 + ------
ω0
Using (1.24), we have
A in A 0
X out(s) = -------- ---------------- (1.26)
s s
1 + ------
ω0

Equation (1.26) can be re-written in terms of its residues as

1 1
X out(s) = A in A out --- – --------------- (1.27)
s s + ω0
It is now simple to take the Inverse Laplace Transform to get

x out(t) = u ( t )A in A 0 [ 1 – e –t / τ ] (1.28)
where τ = 1 ⁄ ω 0 .
In a similar manner it is straight-forward to show that the step response of a general
first-order circuit having a transfer function given by

© K.W. Martin, 1998 March 15, 2001

 6 Frequency Response of Electronic Circuits

 1 + ----- s
-
 ω z
H(s) = A 0  ---------------- (1.29)
 1 + ----- s
-
 ω 0

has an output having Laplace Transform

 1 + ----- s
-
A in A 0  ω z
X out(s) = --------------  ---------------- (1.30)
s 
1 + ------
s
 ω 0

and its step response is given by

ω –t / τ
x out(t) = u ( t )A in A 0 1 –  1 – -----0- e (1.31)
 ω z
where τ = 1 ⁄ ω 0
Note that the step response for t very slightly greater than 0 is denoted by x out(0 +) and
using (1.31) is easily found to be given by
ω0
x out(0 +) = A in A 0 ------ (1.32)
ωz

This is easily verified using the Laplace Transform property

ω0
x out(0 +) = lim sX out(s) = lim H(s) A in = A in A 0 ------ (1.33)
s→∞ s→∞ ωz

together with equation (1.30). In a similar manner, it is easily found that after a long time,
when the first-order circuit has settled, its response denoted x out(∞) is given by
x out(∞) = A in A 0 (1.34)
Once again, this can be verified using the Laplace Transform property
x out(∞) = lim sX out(s) = lim H(s) A in (1.35)
s→0 s→0

together with (1.30). Using (1.31),(1.32) and (1.35), one can derive the general equation giv-
ing the step response of any first-order circuit as

x(t) = x ( ∞ ) – [ x ( ∞ ) – x ( 0 + ) ]e –t / τ (1.36)

Example 1.2 Consider the first-order lowpass circuit shown in Fig. 1.1. Assume
R = 1kΩ and C = 1nF . Find the -3dB frequency of the circuit and the
time constant of the circuit. Assuming the input signal is a 0.5V step input

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 7

R
Vin Vout
C

Fig. 1.1 A RC first-order lowpass circuit.

at time 0, what is the output voltage at 1.5µs .

Solution The transfer function of this circuit is easily found using the admit-
tance divider formula; that is

V out(s) G 1(s)
H ( s ) = -----------------
- = ---------------------------------- (1.37)
V in(s) G 1(s) + G 2(s)
where G 1(s) is the interconnecting admittance between the input and out-
put and the denominator is the sum of all admittances connected to the
output node (in this case G 1(s) + G 2(s) ). For this case, we have
G 1(s) = 1 ⁄ R and G 2(s) = sC and therefore

1⁄R 1
H ( s ) = ------------------------- = --------------------- (1.38)
sC + 1 ⁄ R 1 + sRC
The – 3dB in radians is given by

1 1 6
ω 0 = --------- = --------------------------------- = 1 ×10 rad = 1Mrad (1.39)
RC 1 ×10 1 ×10
3 –9

In Hz, the -3dB frequency is given by

1 1
f -3dB = ------ --------- = 159kHz (1.40)
2π RC
The circuit time constant is given by

1
τ = ------ = RC = 1µs (1.41)
ω0
Next, using (1.28), we have the output voltage at 1.5µs given by

x out(t) = 0.5 [ 1 – e –1.5 ×10 ⁄ 1 ×10

–6 –6
] = 0.5 [ 1 – e –1.5 ] = 0.39V (1.42)

Example 1.3 Consider the first-order circuit circuit shown in Fig. 1.2. Find equa-
tions for the transfer function and for the step-response. What constraint is
necessary and sufficient for the step-response to be an ideal step. For the
special case of C 1 = 5pF , C 2 = 10pF , R 1 = 2kΩ , and R 2 = 10kΩ , plot

© K.W. Martin, 1998 March 15, 2001

 8 Frequency Response of Electronic Circuits

R1

C1
Vin Vout

R2 C2

Fig. 1.2 A RC first-order circuit.

the step response assuming the input signal is a 2V step. How long does it
take for the circuit to settle within 70% of its final value?
Solution The transfer function can once again be found using the admit-
tance-divider formula. We have
V out(s) sC 1 + G 1
H ( s ) = ------------------ = -----------------------------------------------------
sC 1 + G 1 + sC 2 + G 2
V in(s)
C1
1 + s ------- (1.43)
G1 G1
=  -------------------- -------------------------------------
 G 1 + G 2 ( C1 + C2 )
1 + s ------------------------
G1 + G2

where G 1 = 1 ⁄ R 1 and G 2 = 1 ⁄ R 2 . Equation (1.43) can be re-written in

the same form as (1.29) with the substitutions

G1 R2
A 0 = -------------------
- = -------------------- (1.44)
G1 + G2 R1 + R2

C
τ z = ω z–1 = ------1- = R 1 C 1 (1.45)
G1
and

C1 + C2
- = ( R 1 || R 2 ) ( C 1 + C 2 )
τ 0 = ω 0–1 = ------------------- (1.46)
G1 + G2
In order for the step response (given by (1.31)) to be an ideal step, we
need ω z = ω 0 . Using (1.45) and (1.46), we see the necessary conditions
are

C1 C1 + C2
------- = -------------------- (1.47)
G1 G1 + G2

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 9

which implies

C2 G2 R1
------ = ------- = ------ (1.48)
C1 G1 R2
For the case, of C 1 = 5pF , C 2 = 10pF , R 1 = 2kΩ , and R 2 = 10kΩ , we
8 7
have A 0 = 0.833 , ω z = 1 ×10 rad , and ω 0 = 4 ×10 rad . Using (1.33) and
(1.43), we see

C1
x out(0 +) = A in lim H ( s ) = A in -------------------- = 2  ------------------------------ = 0.67V
5pF
(1.49)
s→∞ C 1 + C 2
 5pF + 10pF
and

R2
x out(∞) = A in lim H ( s ) = A in -------------------- = 2  -------------------------------- = 1.67V
10kΩ
(1.50)
s→0 R1 + R2  2kΩ + 10kΩ
The circuit time constant is 1 ⁄ ω 0 = 25ns . Using these values, it is possi-
ble to sketch the step response as shown in Fig. 1.3. Note that in one time

1.67V

25ns,1.51V
0.67V

Fig. 1.3 The step-response of the circuit of Fig. 1.2.

constant, the output voltage has settled to within 63% of its total change
(i.e. x out(∞) – ( x out(∞) – x out(0 +) )e –1 = 2 + – 1.33 e –1 = 1.51V ).For
the circuit to settle within 70% of its final value, we need

1 – e –t / τ = 0.7 (1.51)
which implies

t = τ ln  ---------------- = 1.2τ = 30ns

1
(1.52)
 1 – 0.7

© K.W. Martin, 1998 March 15, 2001

 10 Frequency Response of Electronic Circuits

Example 1.4 Consider an amplifier having a small signal transfer function that
approximately is given by

A0
A ( s ) = ---------------------- (1.53)
s
1 + -------------
ω -3dB
What is the approximate unity-gain frequency and the phase shift at the
5 3
unity-gain frequency for A 0 = 1 ×10 and ω -3dB = 1 ×10 ?
Solution At frequencies s = jω » jω -3dB , which is the case for most signal fre-
quencies, the s term in the denomimator dominates and we have
A 0 ω -3dB
A ( s ) ≅ -------------------- (1.54)
s
and the open-loop transfer function of the amplifier is very similar to that of
an ideal integrator. The magnitude of the Fourier Transform is found by
subsituting s = jω and taking the magitude. To find the unity gain fre-
quency, this magnitude should be set to unity and then ω (often called
ω ta ) should be solved for. We have

A 0 ω -3dB
-------------------- = 1 (1.55)
ω ta
which implies

ω ta = A 0 ω -3dB (1.56)
The phase shift of the transfer function is given by
ω –1
∠A ( ω ) = – tan  ------------- (1.57)
 ω -3dB

Noting that since A 0 » 1 , we also have ω ⁄ ω -3dB » 1 , we have

∠A ( ω ta ) ≅ – 90° . (1.58)
8
Also, for our example, using (1.56) and (1.58), we have ω ta ≅ 1 ×10 rad .

High-Order Transfer Functions

High-order transfer functions (of order greater than first) are normally written in one of
two forms: the first form is a ratio of polynomials in s, the Laplace Transform variable; the
second form is usually written as a ratio of products of first or second-order terms; this is
often called the root form. The first form may be expressed as

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 11

a0 + a1 s + … + am s m
H ( s ) = ------------------------------------------------------ (1.59)
b0 + b1 s + … + bn s n
Normally, m ≤ n , and all of a i and b i are real. Further, assuming the transfer function is
stable, all of the b i will be positive. The alternative form of writing a transfer function might
look like
( s + z 1 ) ( s + z m )… ( s + z m )
H ( s ) = K ----------------------------------------------------------------------- (1.60)
( s + ω 1 ) ( s + ω m )… ( s + ω n )

This is often referred to as root form. However, to be exact note that the coefficients z 1 ,
z 2 …z m are not the actual roots of the numerator, rather they are equal to the negative roots.
For example, s = – z 1 , is a root, not s = z 1 . For transfer functions of Real systems, the
coefficients in either the numerator or the denominator will either be real or occur in com-
plex-conjugate pairs. Further the real-parts of all the denominator coefficients will be posi-
tive for stable transfer functions.

Second-Order Low-Pass Transfer Functions

Second-order low-pass transfer functions are often encountered when characterizing
electronic circuits. For example, operational amplifiers, having feedback, and perhaps being
inadequately compensated, are often modelled as second-order low-pass functions. Three
alternative formulations for second-order low-pass transfer functions (assuming a real
numerator) are

K
H ( s ) = ---------------------------------------------
( 1 + sτ 1 ) ( 1 + sτ 2 )
K
= -------------------------------------------------
 1 + -------- s 
- 1 + ---------
s (1.61)
 ω p1  ω p2
Kω p1 ω p2
= ----------------------------------------------
-
( s + ω p1 ) ( s + ω p2 )
The coefficients τ 1, τ 2 or ω p1, ω p2 are either real and positive, or occur in com-
plex-conjugate pairs. The transfer function may also be written in the form

Kω p1 ω p2
H ( s ) = ---------------------------------------------------------------------
- (1.62)
ω p1 ω p2 + s ( ω p1 + ω p2 ) + s 2
Equation (1.62) may be written in the alternative (and popular) form

© K.W. Martin, 1998 March 15, 2001

 12 Frequency Response of Electronic Circuits

Kω 02
H ( s ) = ----------------------------------
- (1.63)
ω0
ω 0 + s ------ + s
2 2
Q
Note that K is the d.c. gain of the transfer function. This latter form is often the preferred
form of experienced filter designers.
We will now consider two special cases: first when ω p1, ω p2 are both real and distinct,
and secondly when they occur as a complex-conjugate pair.
i) Consider Real Poles where ω p1 < ω p2
We can set the denominator of (1.63), D ( s ) , equal to the denominator of (1.61), and we
have

ω
D ( s ) = ω 02 + s -----0- + s 2
Q

=  1 + ---------  1 + ---------
s s (1.64)
 ω p1   ω p2
= ω p1 ω p2 + s ( ω p1 + ω p2 ) + s 2
Setting the coefficients equal, we get

ω 02 = ω p1 ω p2 (1.65)
and

ω0
------ = ω p1 + ω p2 (1.66)
Q
Solving (1.65) and (1.66), we get
ω0
ω p1, ω p2 = -------- ( 1 ± 1 – 4Q 2 ) (1.67)
2Q
We can express the Fourier Transform of (1.61) as
K
H ( s ) = --------------------------------------------------------------e jφ (1.68)
ω ω 2
1 +  --------- 1 +  ---------
2
 ω p1  ω p2

where
ω ω
φ = – tan–1  --------- – tan  --------- (1.69)
 ω p1  ω p2

This form makes it easy to calculate H ( ω ) and ∠H ( ω ) .

ii) Widely-Spaced Real Poles where ω p1 « ω p2

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 13

Note that for the special case of widely-spaced poles where ω p1 « ω p2 , we have Q « 1 ,
and we can make the approximation

1 – 4Q 2 ≅ 1 – 2Q 2 (1.70)
Using (1.70) and (1.67) gives

ω p1 ≅ ω 0 Q (1.71)

ω0
ω p2 ≅ ------ (1.72)
Q
Using (1.68) and (1.69), it is possible to simply approximate the magnitude and phase
response of low-pass transfer functions in a number of different frequency regions. For
ω « ω p1 , we have H ( ω ) = K . At ω = ω p1 , we have H ( ω p2 ) = K ⁄ 2 , and
∠H ( ω ) = – 45° . Next, for ω p1 « ω « ω p2 , we have
Kω p1
H ( ω ) ≅ ------------- (1.73)
ω
If we express the magnitude response in terms of decibels (dB), we

H ( ω ) dB ≅ 20 log ( Kω p1 ) – 20 log ( ω ) (1.74)

Thus, the magnitude response is decreasing -20dB for every decade increase in ω . In addi-
tion, in this region, we have ∠H ( ω ) ≅ – 90° . Next, at ω = ω p2 , we have
Kω p1
H ( ω p2 ) ≅ ---------------- (1.75)
2ω p2
and ∠H ( ω ) = – 135° . The final region is when ω p2 « ω . In this region, we have
Kω p1 ω p2
H ( ω ) ≅ ----------------------
- (1.76)
ω2
Expressing the magnitude gain in dB, we have

H ( ω ) dB ≅ 20 log ( Kω p1 ω p2 ) – 40 log ( ω ) (1.77)

and we see the gain is decreasing -40 dB for every decade increase in ω . In this region, we
also have ∠H ( ω ) ≅ – 180° .

Bode Plots
The previous observations are the basis for a methodology useful for constructing
approximate plots of the magnitude gain (in dB) and phase versus log frequency for
low-pass transfer functions having real poles. These plots are called Bode Plots. The method
is based on the following rules.

© K.W. Martin, 1998 March 15, 2001

 14 Frequency Response of Electronic Circuits

a) As the frequency becomes larger than a pole frequency, the slope of the magnitude
gain decreases by -20dB.
b) Each pole contributes an additional – 45° phase shift at the frequency of the pole. At
frequencies substantially higher than a pole frequency, it contributes an additional
– 90° .

Example 1.5 Consider a second-order low-pass transfer function having

4
K = 1 ×10 , ω p1 = 10 , and ω p2 = 100 . Plot its approximate magnitude
gain in dB and phase versus log ω .
Solution Based on the just presented rules, the following plots result.

H ( ω ) dB
-20dB/dec
80dB

40dB
-40dB/dec

log ( ω )
1 3
∠H ( ω )
1 3
log ( ω )
– 90°

– 180°

Fig. 1.4 Magnitude gain (in dB) and phase versus log ω .

This technique is easily generalized for use with higher-order transfer

functions.

Example 1.6 Construct the Bode Plots for the closed-loop configuration shown in
Fig. 1.5
At mid-band frequencies, C 1 can be assumed to be a short circuit (since
its very large), and C 2 can be approximated by an open circuit since it’s
very small. The reason for including C 1 is that at very-very-low frequen-
cies, it operates as an open circuit. Thus, at d.c., the gain is only unity. The
low gain at d.c. (of unity) helps minimize bothersome ‘clicks’ due to d.c.
offsets inherent in the amplifier that would be amplified by the large gain if
C 1 had not been included. C 2 limits the high frequency response. It also

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 15

V in
V out
R2

2kΩ Z2
C2

100pF
R1
100Ω
Z1
C1

100µF

Fig. 1.5 The feedback configuration often used for stereo power amplifiers.

helps stabilize the op-amp, but this is beyond the scope of the current sub-
ject.
At moderately-low frequencies where C 1 is having an effect, C 2 can cer-
tainly be ignored as it is so small. The circuit at these frequencies has
almost the response as the simplified circuit shown in Fig. 1.6. The transfer

V in
V out
R2
Z2
2kΩ
R1
100Ω
Z1
C1

100µF

Fig. 1.6 A simplified circuit that has the same frequency response for low and
mid-band frequencies.

function low and mid-band frequencies is found by analyzing this simplified

circuit. The closed-loop transfer function is given by

© K.W. Martin, 1998 March 15, 2001

16 Frequency Response of Electronic Circuits

V out(s) Z 1(s) + Z 2(s)

H ( s ) = -----------------
- = -------------------------------- (1.78)
V in(s) Z 1(s)
where

1
Z 1(s) = R 1 + ---------- (1.79)
sC 1
and
Z 2(s) = R 2 (1.80)
We have
1
R 1 + R 2 + ----------
sC 1
H ( s ) = ------------------------------------
1
R 1 + ---------- (1.81)
sC 1
1 + s ( R 1 + R 2 )C 1
= -------------------------------------------
-
1 + sR 1 C 1

We see the simplifed circuit is first order with a zero at

1
ω z1 = -------------------------------- = 4.76rad (1.82)
( R 1 + R 2 )C 1
and a pole at

1
ω p1 = ------------- = 100rad (1.83)
R1 C1
At mid-band frequencies, the gain can be found by (1.81) and assuming
sC 1 is very large. We easily see
R1 + R2
H ( s ) ≅ -------------------- (1.84)
R1

At mid and high-frequencies, when C 1 can be certainly ingnored, but C 2 ,

can not necessarily be ignored, then the response of the circuit is very
close to the response of the simplified circuit shown in Fig. 1.7. We now
have

Z 1(s) = R 1 (1.85)
and

1
Z 2(s) = ------------------------------ (1.86)
1 ⁄ R 2 + sC 2

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 17

V in
V out
R2

2kΩ Z2
C2

100pF
R1
100Ω
Z1

Fig. 1.7 A simplified circuit that has the same frequency response for mid-band
and high frequencies.

Using (1.78), we have (after a little algebra)

Y 1(s) + Y 2(s)
H ( s ) = ---------------------------------
Y 2(s)
1 1
------ + ------ + sC 2
R1 R2
= ------------------------------------- (1.87)
1
------ + sC 2
R2
R 1 + R 2 1 + s ( R 1 || R 2 )C 2
=  --------------------  -------------------------------------------
 R1   1 + sR 2 C 2 

We see the simplifed circuit is again first order with a zero at

1 8
ω z1 = ------------------------------- = 1.05 ×10 rad (1.88)
( R 1 || R 2 )C 2
and a pole at

1 6
ω p1 = ------------- = 5 ×10 rad (1.89)
R2 C2
Once again the mid-band gain is found (in this case assuming C 2 is small)
to be
R1 + R2
H ( s ) ≅ -------------------- (1.90)
R1

An alternative approach would be to consider the actual circuit of Fig. 1.5

© K.W. Martin, 1998 March 15, 2001

18 Frequency Response of Electronic Circuits

that is valid at all frequencies and substitute

1
Z 1(s) = R 1 + ---------- (1.91)
sC 1
and

1
Z 2(s) = ------------------------------ (1.92)
1 ⁄ R 2 + sC 2
directly into (1.78) without simplifying. After some algebra, we get

1 + s { ( R 1 + R 2 )C 1 + R 2 C 2 } + s 2 R 1 R 2 C 1 C 2
H ( s ) = ---------------------------------------------------------------------------------------------------------------- (1.93)
( 1 + sR 1 C 1 ) ( 1 + sR 2 C 2 )
If we now make the assumption that the numerator, N ( s ) , has two widely
separated zeros ω z1 « ω z2 , then we may express the numerator as

N ( s ) =  1 + ---------  1 + ---------
s s
ω z1 ω z2
s2
= 1 + s  --------- + --------- + ------------------
1 1
 ω z1 ω z2 ω z1 ω z2 (1.94)

s s2
≅ 1 + --------- + ------------------
ω z1 ω z1 ω z2

Setting (1.94) equal to (1.93), it is easy to solve for ω z1 and ω z2 . We get

1 1
ω z1 ≅ ---------------------------------------------------- ≅ -------------------------------- (1.95)
( R 1 + R 2 )C 1 + R 2 C 2 ( R 1 + R 2 )C 1

and

1 R1 + R2 1
ω z2 ≅ ------------------------------------- ≅ --------------------- = ------------------------------- (1.96)
ω z1 R 1 R 2 C 1 C 2 R 1 R 2 C 2 ( R 1 || R 2 )C 2
We also have two widely-spaced poles given by

1
ω p1 = ------------- (1.97)
R1 C1
and

1
ω p2 = ------------- (1.98)
R2 C2

Using the circuit values, we have ω z1 = 4.76rad ≅ 5rad ,

8 8 6
ω z2 = 1.05 ×10 rad ≅ 1 ×10 rad , ω p1 = 100rad , and ω p2 = 5 ×10 rad .

© K.W. Martin, 2001 March 15, 2001

 1.1 Frequency Response of Linear Systems 19

Note that both the zeros and the poles are widely separated. At mid-band
frequencies, the gain (as given by (1.84)) is 21 (i.e. 26.4dB). These are the
same results we got when considering the simplified circuits valid at low
and high frequencies only. These values lead us to the following Bode
Plots.

H ( ω ) dB

20dB

10dB

0dB log ( ω )
2 4 6 8
∠H ( ω )
90°
0°
2 4 6 8 log ( ω )
– 90°

Fig. 1.8 Approximate magnitude and phase responses of the circuit of Fig. 1.5.

i) Step Response for Real Poles

The step response for a second-order low-pass transfer function can be found by expand-
ing (1.61) into partial fractions (or residues). It is not too difficult to show that

ω p2 ω p1 ω p1 ω p2
H ( s ) = K ------------------------ ------------------ – ------------------------ ------------------ (1.99)
ω p2 – ω p1 s + ω p1 ω p2 – ω p1 s + ω p2
Equivalently, in terms of time constants τ 1 = 1 ⁄ ω p1 and τ 2 = 1 ⁄ ω p2

τ1 1 τ2 1
H ( s ) = K ---------------- ----------------- – ---------------- ----------------- (1.100)
τ 1 – τ 2 1 + sτ 1 τ 1 – τ 2 1 + sτ 2
It is now straightforward to express the step response of the second-order system in terms of
the individual step responses of two first-order systems. We have

Kω p2 Kω p1
x out(t) = A in ------------------------ [ 1 – e –tωp1 ] – A in ------------------------ [ 1 – e –tωp2 ] (1.101)
ω p2 – ω p1 ω p2 – ω p1
and we see the step response is composed of two first-order modes.
ii) Consider Complex-Conjugate Poles where ω p1 = ω p2∗
Using (1.67), we see this case occurs when Q > 1 ⁄ 2 and we have
ω0
ω p1, ω p2 = -------- [ 1 ± j 4Q 2 – 1 ] (1.102)
2Q

© K.W. Martin, 1998 March 15, 2001

 20 Frequency Response of Electronic Circuits

Note that
ω0
ω p1 = ω p2 = -------- [ 1 + 4Q 2 – 1 ] 1 / 2 = ω 0 (1.103)
2Q
The step response can now be found by substituting

ω p1 = ω r + jω q (1.104)
and

ω p2 = ω r + – j ω q (1.105)
into (1.101) where

ω
ω r = -------0- (1.106)
2Q
and
ω0
ω q = -------- 4Q 2 – 1 (1.107)
2Q
We get
A in K
- { ( ω r – jω q ) [ 1 – e –ωr t e – jωq t ] – ( ω r + jω q )e –ωr t e jωq t }
x out(t) = --------------
– 2jω q
A in K
= ------------ { jω q [ 2 – e –ωr t ( e jωq t + e – jωq t ) ] – ω r [ e –ωr t ( e jωq t – e – jωq t ) ] } (1.108)
2jω q
ωr
= A in K 1 – e –ωr t cos ( ω q t ) – ------e –ωr t sin ( ω qt )
ωq

Thus, we see the step response has two sinusoidal term having envelopes with magnitudes
exponentially decreasing with time constants equal to the inverse of the real parts of the
poles. The oscillatory frequency of the sinusoids are determed by the imaginary parts of the
poles.
It is also possible to determine the peak of the step response by setting

d
x (t) = 0 (1.109)
d t out
The solution having the smallest magnitude is the peak of the step response. The details of
this analysis are left to the reader. The end result is
π
–  -----------------------
 4Q 2 – 1
x out(t) = A in Ke (1.110)
max

Equation (1.110) is valid only for Q ≥ 0.5 . For Q ≤ 0.5 , there is no overshoot. Q = 0.5 is

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 21

called a maximally-damped response. These different cases have been plotted approximately
in Fig. 1.9.

(c)

(b)

(a)

Fig. 1.9 The step responses of a second-order low-pass circuit for (a) Q ≤ 0.5 , (b)
Q = 0.5 , and (c) Q ≥ 0.5 .

1.2 Frequency Response of Transistor Circuits

When analyzing electronic circuits containing transistors to determine their frequency
response, it is implicitly assumed that signals are small enough that small-signal approxima-
tions about an operating point accurately reflect how the circuit operates. Given this assump-
tion, the procedure is:
i) Do an operating point analysis for all currents and voltages. Make sure all relevant
transistors are in the active region.
ii) Replace all transistors with their small-signal circuits that include parasitic capaci-
tances for modelling high-frequency limitations.
iii) Analyze the resulting linearized small-signal circuit for its frequency response.

High-Frequency MOS Small-Signal Model

The small-signal high-frequency model for MOS transistors takes the low-frequency
model and adds parasitic capacitances to model charge-storage effects of MOS transistors, A
cross-sectional view of a MOS transistor showing the physical relationship of these parasitic
capacitances is given in the chapter on Modelling and is repeated in Fig. 1.10. The high-fre-
quency model for MOS transistors in the active region is shown in Fig. 1.11. The largest
capacitance in the model is the gate-source capacitance C gs . Its value can be calculated
from the approximate formula

2
C gs = --- WLC ox + WL ov C ox (1.111)
3

© K.W. Martin, 1998 March 15, 2001

 22 Frequency Response of Electronic Circuits

VSB = 0 VGS > Vtn

Polysilicon VDG > -Vtn

Cgd SiO2
Al

Cgs n+
n+
Lov
p+ Field Cs-sw C′ sb Lov C′ db Cd-sw
Implant -
p Substrate

Fig. 1.10 A cross-section of an n-channel MOS transistor showing the small signal
capacitances.

C gd

vg vd

C gs
v gs gm vgs gsvs rds C db

vs
rs
C sb

vs ′

Fig. 1.11 The high-frequency model for MOS transistors in the active region.

where C ox is the gate capacitance per unit area and is given by

K ox ε 0
C ox = -------------
- (1.112)
t ox
W is the transistor width, L is the transistor length, L ov is the effective overlap of the gate
and junction conductive regions, K ox is the relative dielectric constant of silicon-dioxide
– 12
(i.e. 3.9), ε 0 is the dielectric constant of free-space (i.e. ε 0 = 8.854 × 10 F ⁄ m ) and t ox

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 23

is the thickness of the gate SiO 2 insulation layer. The capacitance C gd is primarily due to
the effective1 overlap capacitance between the gate layer and the conductive drain region. Its
value is given by
C gd = WL ov C ox (1.113)

The other capacitances in the small-signal model are due to the depletion capacitances of
reverse-biased junctions. We have

C sb = ( A s + WL )C js + P s C j-sw (1.114)

where A s is the area of the source junction, and P s is the effective perimeter of the source
junction. C js is the bottom-plate depletion capacitance of the source junction per unit area
and is given by
C j0
C js = ----------------------------------
- (1.115)
1 + V SB ⁄ Φ 0
where Φ 0 is approximately 0.9V. Note the approximate inverse square-root dependance on
the source-bulk (or substrate) voltage V SB . Note also that the effective area used for the
source junction in (1.114) has the channel area added (i.e. WL )) as the source junction and
the channel are electrically connected with low impedance. The equation for the drain-bulk
capacitance is similar.
C db = A d C jd + P d C j-sw (1.116)

where

C j0
C jd = ----------------------------------
- (1.117)
1 + V DB ⁄ Φ 0
Note that the effective drain junction area no longer includes the area of the channel (i.e.
WL ).
The resistor, r s , is often ignored in most texts, but can be important in modern technolo-
gies, where more-lightly-doped junctions may be present, and where velocity-saturation can
also occur (which is modelled by including r s ). These effects can be especially important
for higher-frequency circuits. In this text, as in practically own preceding texts, r s will be
assumed negligible to simplify analyses; for most-accurate modelling this decision may
need to be re-visited in the future.

1. Part of this capacitance is due to fringe electric fields, and therefore L ov is usually taken larger than its
actual physical overlap to more accurately give an effective value for overlap capacitances.

© K.W. Martin, 1998 March 15, 2001

 24 Frequency Response of Electronic Circuits

High-Frequency BJT Small-Signal Model

The high-frequency model for BJT transistors in the active region is shown in Fig. 1.12.

rb Ccb ic
Base Collector
vb vc
ib
Cbe rπ vbe gmvbe ro Ccs

ie
ve
Emitter

Fig. 1.12 The high-frequency model for BJT transistors in the active region.

The largest parasitic capacitance is now the base-emitter capacitance. It is composed of two
components: one due to the base-emitter depletion capacitance and denoted C j , and the sec-
ond and normally larger component, C d , due to the minority charge in the base region
responisble for the collector diffusion current. Thus,

C be = C j + C d (1.118)
where the diffusion capacitance is given by
IC
C d = τ b ------ = g m τ b (1.119)
VT
The parameter τ b is approximately the base-transit time-constant. Its value is normally
determined based on empirical measurements and is one of the important Spice parameters
to be specified. The base-emitter depletion capacitance is often approximated as twice the
depletion capacitance per unit area at 0V bias conditions multiplied by the effective emitter
area. The resistor r b Fig. 1.12 is due to the intrinsic base resistance. This resistor is often
ignored in low-frequency analysis (in most previous texts), but is often important in
high-frequency circuits. It might be mentioned that it is often the major source of noise in
bipolar circuits. The capacitances C cb and C cs and the collector-base and collector-sub-
strate depletion capacitances. The latter is normally larger, but the former can often be the
major frequency limitation in, for example, common-emitter stages, as we shall see.

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 25

High-Frequency Analysis of Common-Source Amplifier

The first high-frequency analysis that will be presented is for an MOS common-source
stage. The high-frequency analysis of the bipolar common-emitter amplifier is very similar.
An example common-source configuration is shown in Fig. 1.13. It is assumed that the bias

Q3 Q2

I bias
V out

CL
R in
V in Q1

Fig. 1.13 A common-source amplifier.

voltage of the input signal is such that both Q 1 and Q 2 are in the active region. Based on
this assumption, a small-signal equivalent circuit for high-frequency analysis of the com-
mon-source amplifier of Fig. 1.13 is shown in Fig. 1.14. Here, C gs1 is the gate-to-source

Rin v1 Cgd1
vout
+ C
vin v gs1 gs1 g v R2 C2
m1 gs1
-

Fig. 1.14 A small-signal model for high-frequency analysis of the com-

mon-source amplifier.

capacitance of Q 1 and C gd1 is the gate-to-drain capacitance of Q 1 . Note that it has been
assumed that the output capacitance of the input signal source can be ignored. The capaci-
tance C 2 is made up of the parallel connection of the drain-to-bulk capacitances of Q 1 and
Q 2 , the gate-drain capacitance of Q 2 , and the load capacitance C L . Usually, C L dominates.

© K.W. Martin, 1998 March 15, 2001

 26 Frequency Response of Electronic Circuits

Miller Theorem
The most commonly presented method for analyzing common-source and com-
mon-emitter amplifiers to determine the – 3dB frequency is based on the Miller Theorem.
This theorem is shown pictorially in Fig. 1.15. This theorem states that the two circuits

I1 I2
Y

V1 V2
–A

(a)
V1 V2
–A
I1 I2

Y1 Y2

(b)
Fig. 1.15 The Miller-Theorem states circuit (b) is equivalent to circuit (a) assum-
ing Y 1 and Y 2 are chosen appropriately.

shown are equivalent assuming Y 1 and Y 2 are chosen appropriately. To be specific if

Y 1(s) = Y(s) ( 1 + A ) (1.120)

and

Y 2(s) = Y(s)  1 + ----

1
(1.121)
 A
then the I-V relations are identical for the two circuits. To be more specific

I 1 = Y(s) ( V 1 – V 2 ) = Y(s) ( 1 + A )V 1 (1.122)

and

I 2 = Y(s) ( V 2 – V 1 )V 2 = Y ( s )  1 + ---- V 2
1
(1.123)
 A

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 27

for both circuits.

The power of the Miller Theorem is it greatly simplifies the analysis for the dominant
pole of a circuit having a capacitor between the input and output of an inverting high-gain
amplifier, as we shall see. It should be cautioned, however, the Miller Theorem can not be
used when estimating the second pole of most amplifiers; it implicitly assumes the fre-
quency response between nodes V 1 and V 2 is frequency independant. This assumption is
not valid when capacitive loading at node V 2 , which is responsible for the second pole,
becomes appreciable and causes A(ω) to decrease.
The Miller Theorem can be used to simplify the analysis of the small-signal circuit of
Fig. 1.14. The simplified circuit is shown in Fig. 1.16. This simplification is based on noting

R in v1 vout

vin + C gs1 g m1 v gs1

v gs1 R2
C2
-

C gd1  1 + -----------------
1
C gd1 ( 1 + g m1 R 2 )
 g m1 R 2

Fig. 1.16 A simplified small-signal model for the common-source amplifier based
on using the Miller Theorem

that the low-frequency gain, A , is given by

A = g m1 R 2 (1.124)
The admittance Y is equal to Y(s) = sC gd1 . Therefore, we have

Y 1(s) = sC gd1 ( 1 + A ) = sC gd1 ( 1 + g m1 R 2 ) (1.125)

and

Y 2(s) = sC gd1  1 + ---- = sC gd1  1 + -----------------

1 1
(1.126)
 A   g m1 R 2

The time constant at node v 1 certainly dominates because of the large capacitance there.
This gives

τ 1 = R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] (1.127)

The estimate for the -3dB frequency of the circuit is then given by

© K.W. Martin, 1998 March 15, 2001

 28 Frequency Response of Electronic Circuits

1 1
ω –3dB = ----- = --------------------------------------------------------------------------- (1.128)
τ1 R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ]
As a final note regarding the Miller Theorem, this technique of realizing a large grounded
capacitor by placing a smaller capacitor between the input and output of a high-gain invert-
ing amplifier has been often used to effectively realize large grounded on-chip capacitors.

Exact Nodal Analysis

To exactly analyze the small-signal circuit at very-high frequencies, nodal analysis can
be used. At node v 1 , summing all of the currents leaving the node and setting the sum equal
to zero, we have
v 1 ( G in + sC gs1 + sC gd1 ) – v in G in – v out sC gd1 = 0 (1.129)
where G in = 1 ⁄ R in 2. Also, at the output node, we have
v out ( G 2 + sC gd1 + sC 2 ) – v 1 sC gd1 + g m1 v 1 = 0 (1.130)
where we have used v 1 = v gs1 .
Before proceeding, it is worth mentioning here that to minimize making circuit equation
errors, a consistent methodology should be maintained when writing nodal equations. The
methodology employed here is as follows. The first term is always the node at which the
currents are being summed. Multiplying this node voltage is the sum of all admittances con-
nected to the node. The next negative terms are the adjacent node voltages with each being
multiplied by the connecting admittance. The last terms are any current sources with a posi-
tive sign being used if the current sources flow out of the node.
From (1.130), we have

[ G 2 + s ( C gd1 + C 2 ) ]
v 1 = v out ---------------------------------------------------- (1.131)
– g m1 + sC gd1
Substituting (1.131) into (1.129) gives

[ G 2 + s ( C gd1 + C 2 ) ] [ G in + s ( C gs1 + C gd1 ) ]

v out --------------------------------------------------------------------------------------------------------------- – v out sC gd1 = v in G in (1.132)
– g m1 + sC gd1
which implies

 G in G 2 + s [ G 2 ( C gs1 + C gd1 ) + G in ( C gd1 + C 2 ) + g m1 C gd1 ] 

v out  
 + s 2 [ ( C gs1 + C gd1 ) ( C gd1 + C 2 ) – C gd12 ]
 (1.133)
= – v in G in ( g m1 – sC gd1 )

2. Whenever a variable is designated G i , it is implicitly assumed that it is an admittance and that G i = 1 ⁄ R i , where R i
is the resistance of the same component.

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 29

v out { G in G 2 + s [ G 2 ( C gs1 + C gd1 ) + G in ( C gd1 + C 2 ) + g m1 C gd1 ] + }

(1.134) mme
s 2 [ ( C gs1 + C gd1 ) ( C gd1 + C 2 ) – C gd1
2 ]

= – v in G in ( g m1 – sC gd1 )
Continuing, we have
C gd1
– g m1 R 2  1 – s ------------
v out  g m1 
---------- = --------------------------------------------------
- (1.135)
v in 1 + sa + s 2 b
where
a = R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] + R 2 ( C gd1 + C 2 ) (1.136)
and
b = R in R 2 ( C gd1 C gs1 + C gs1 C 2 + C gd1 C 2 ) (1.137)
At frequencies when the gain has started to decrease, but is still much greater than unity,
the first-order term in the numerator ( – s ( C gd1 ⁄ g m1 ) ) and the second-order term in the
denominator ( s 2 b ) can be ignored. For this case, we have

v out – g m1 R 2
A ( s ) = ---------- ≅ ------------------------------------------------------------------------------------------------------------------------------------------- (1.138)
v in 1 + s { R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] + R 2 ( C gd1 + C 2 ) }
The low frequency gain is as expected, – g m1 R 2 . Also, setting s = jω –3dB and solving for
A
A ( jω –3dB ) = ------0- (1.139)
2
gives
1
ω –3dB ≅ ----------------------------------------------------------------------------------------------------------------------- (1.140)
R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] + R 2 ( C gd1 + C 2 )
This approximation is close to that obtained previously using the Miller Approximation, but
now includes a term due to the time constant at the output node.
At higher frequencies, when the gain is not much greater than unity, then the second pole
and the zero should be considered. The frequency for the second pole can be found by
assuming the poles are real and widely separated and that the denominator can therefore be
expressed as

s2
D ( s ) =  1 + ---------  1 + --------- ≅ 1 + --------- + -------------------
s s s
(1.141)
 ω p1  ω p2 ω p1 ω p1 ω p2
The coefficients of (1.141) can then be placed equal to the coefficients of the denominator of
(1.135). Setting (1.141) equal to the denominator of (1.135), we have
1 1
ω p1 ≅ --- = ----------------------------------------------------------------------------------------------------------------------- (1.142)
a R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] + R 2 ( C gd1 + C 2 )

© K.W. Martin, 1998 March 15, 2001

 30 Frequency Response of Electronic Circuits

In addition, we have

1
ω p2 ≅ ------------- (1.143)
ω p1 b
Using (1.137), (1.142), and (1.143), we see
g m1 C gd1
ω p2 ≅ ---------------------------------------------------------------------------- (1.144)
C gs1 C gd1 + C gs1 C 2 + C gd1 C 2
For the special case where the load capacitance and therefore C 2 is large, we have

C gd1 g m1
ω p2 ≅ ------------------------------ --------- (1.145)
C gs1 + C gd1 C 2
Note also, that at high frequencies, the zero defined by

g m1
ω z = – -----------
- (1.146)
C gd1
can be important. Because the sign is negative, this zero causes phase lag rather than phase
lead. This is important when compensating CMOS op amps having common-source stages.
Finally, note that both ω p2 and ω z are proportional to the transistor’s transconductance
g m1 ; having a large transconductance is important in minimizing the detrimental effects of
the second pole and the zero.

Zero-Value Time-Constant Analysis

As an aside, it is of interest to note that the previous result for the -3dB frequency is the
same as one would obtain if the zero-value time-constant analysis technique was used [Gray,
1993]. In this technique, one calculates a time-constant for each capacitor by assuming all
other capacitors are zero. The procedure is as follows.
i For each capacitor in turn, with all other capacitors taken to be zero, find a corre-
sponding time-constant. To do this, replace the capacitor in question with a volt-
age source, and then calculate the resistance ‘seen by that capacitor’ by taking the
ratio of the voltage source to the current flowing from it. Note that in this analysis
step, there are no capacitors in the circuit. The time-constant seen by the capacitor
is then simply the capacitor multiplied by the resistance seen by it.
ii The -3dB frequency for the complete circuit is then given by one over the sum of
the individual capacitor time-constants. That is

1
ω –3dB = ------------------ (1.147)
∑ Ri Ci
where R i is the capacitance seen by the i’th capacitor.3

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 31

For complicated circuits, this technique is very powerful in estimating a circuits band-
width with minimal complication and also in determing which nodes are most important and
need to have their associated time-constants minimized.
For the common-source amplifier, the resistance seen by C gs1 is the input source imped-
ance R in , the resistance seen by C gd1 can be shown to be given by R in ( 1 + g m1 R 2 ) + R 2 ,
and the resistance seen by C 2 is R 2 . Taking the -3dB frequency as the inverse of the sum of
the time constants results in (1.140).
Often, unless R in « R 2 , the first term in the denominator of (1.147) dominates and we
have
1
ω –3dB ≅ -------------------------------------------------------------- (1.148)
R in [ C gs1 + C gd1 ( 1 + A ) ]
where A = g m1 R 2 is the magnitude of the low-frequency gain. Again, this results agrees
with that found earlier using the Miller Theorem. The term C gd1 ( 1 + A ) is often called the
Miller capacitance. Because the size of C gd1 is effectively multiplied by one plus the gain
of the amplifier, the importance of having C gd1 small is obvious.

Example 1.7 Assume all transistors have W ⁄ L = 100µm ⁄ 1.6µm in Fig. 1.13,
and that µ n C ox = 90µA ⁄ V 2 , µ p C ox = 30µA ⁄ V 2 , I bias = 100µA ,
r ds-n(Ω) = 8000L(µm) ⁄ I D(mA) , and r ds-p(Ω) = 12000L(µm) ⁄ I D(mA) .
Also, assume R in = 180kΩ , C L = 0.3pF , C gs1 = 0.2pF ,
C gd1 = 0.015pF , C db1 = 20fF , C gd2 = 22fF and C db2 = 36fF . Esti-
mate the -3dB frequency of the common-source amplifier of Fig. 1.13.
Solution We have
R 2 = r ds1 || r ds2 = 77kΩ (1.149)
and
C 2 = C L + C db1 + C db2 + C gd2 = 0.38pF (1.150)
The time-constant due to R in , namely R in [ C gs1 + C gd1 ( 1 + A ) ] , is now
equal to 0.26µs. The time-constant due to R 2 , namely R 2 ( C gd1 + C 2 ) is
equal to 0.03µs. The -3dB frequency (in hertz) is equal to
1 1
f –3dB ≅ ------ -----------------------------------------------------------------------------------------------------------------------
2π R in [ C gs1 + C gd1 ( 1 + g m1 R 2 ) ] + R 2 ( C gd1 + C 2 )

= 550kHz (1.151)

3. This procedure exactly calculates the coefficient of the ‘s’ term in the denominator. The approximation is
in assuming the inverse of this coefficient is the -3dB frequency and that higher-order terms can be ignored.

© K.W. Martin, 1998 March 15, 2001

 32 Frequency Response of Electronic Circuits

Common-Gate/Common-Base Amplifier
The frequency response of the common-gate and common-base stage is usually superior
to that of the common-source (or common-emitter) stage due to the low-impedance, r in , at
the source node assuming G L is not considerably smaller than g ds1 . In addition, there is no
Miller Capacitance coupling from the input node to the high-gain output node. An example
of a common-gate bipolar amplifier is shown in Fig. 1.17. Its small-signal circuit is shown in

V CC

RL

V out

CL
Q

Rs
V in
Cs = ∞
I bias

V EE
Fig. 1.17 A bipolar common-gate amplifier.

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 33

Fig. 1.18. We will use the technique of zero-value time-constant analysis to estimate the

C cb
vout

rπ C be vbe
ro RL
ve g m v be
Rs C 3 = C cs + C L

vin

Fig. 1.18 The high-frequency model for a BJT common-gate amplifier.

bandwidth of this amplifier.

First we will estimate the time-constant associated with each capacitor. Consider C be .
This capacitor is connected between the emitter and ground. The small-signal resistance
seen by it is simply the resistance seen looking into the emitter. Consider the simplified
small-signal model at the emitter node as shown in Fig. 1.19. Note the direction of the con-

ie
ve

Rπ gm ve

Fig. 1.19 A simplfied small-signal model at the emitter-node.

trolled-source has been changed, and the control voltage is correspondingly v e rather than
v be . We have

ie = ve ( gπ + gm ) (1.152)

© K.W. Martin, 1998 March 15, 2001

 34 Frequency Response of Electronic Circuits

where g π = 1 ⁄ r π . Therefore,

v β β α
r e = ----e- = r π || ------- = ------- || ------- =  ------------ ------- = ------- ≅ -------
1 1 1 1
(1.153)
ie gm gm gm  β + 1 gm  gm gm

which is a relatively small impedance. Thus, the time-constant associated with C be ,

denoted τ 1 , is simply given by

α C be
τ 1 = ( r e || R s )C be ≅ r e C be = ------- C be ≅ --------- (1.154)
gm gm
This assumes r e « R s , but in the event this assumption is not justified, the time constant
would be smaller and therefore less important. Consider next the resistance seen-by C cb
with all other capacitances set to zero. Once again this capacitor is grounded. The resistance
seen by it is

r 2 = R L || r c (1.155)
where r c is the impedance looking into the collector, without any capacitors but with emit-
ter-degeneration, where the degeneration resistor is R S . From our previous analysis of the
output impedance of the common-gate amplifier at low frequencies, we know this resistance
is given by

r c = r o [ 1 + g m ( r π || R S ) ] (1.156)
Normally, this impedance will be much greater than R L , and with a good approximation we
have r 2 ≅ R L , and

τ 2 ≅ R L C cb (1.157)

Using an identical analysis, we see the time constant τ 3 associated with C 3 is

τ 3 ≅ R L C 3 = R L ( C cs + C L ) (1.158)
Noramally, C L will dominate here. The total time constant τ tot is now given by

C be
τ tot ≅ --------- + R L ( C cb + C cs + C L ) (1.159)
gm
The – 3dB frequency will be approximately given by 1 ⁄ τ tot with good accuracy. Note the
absence of any time constants having capacitive terms effectively multiplied by the low-fre-
quency gain of the amplifier unlike the common-emitter stage. This results in significantly
superior high-frequency operation.
In the next section we will see how by combining a common-base configuration with a
common-emitter configuration (or a common-gate stage with a common-source stage), even
further improvements are achieved. This combination not only results in no Miller capaci-

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 35

tance, but also has a significantly-higher input impedance than the common-base (or com-
mon-gate) stage.

Cascode Gain-Stage
In modern IC design, a commonly used configuration for a single-stage amplifier is a
cascode configuration. This configuration consists of a common-source-connected transistor
feeding into a common-gate-connected transistor. Two examples of cascode amplifiers are
shown in Fig. 1.20. The configuration in Fig. 1.20(a) has both an n-channel common-source
transistor, Q 1 , and an n-channel common-gate cascode transistor Q 2 . This configuration is

Ibias Ibias1

Vout
Q2
Vbias Q2 CL Vin Q1 Vbias

Vout
Vin Q1
Ibias2 CL

(a) (b)

Fig. 1.20 (a) A telescopic cascode amplifier and (b) a folded-cascode amplifier.

sometimes called a telescopic cascode amplifier. The configuration shown in Fig. 1.20(b)
has an n-channel input (or drive) transistor but a p-channel transistor is used for the cascode
or common-gate transistor. This configuration is usually called a folded-cascode stage. It
allows the dc level of the output signal to be the same as the dc level of the input signal.
Unfortunately, it is usually slower than the telescopic cascode amplifier because although
parasitic capacitances at the source of the cascode transistor are similar in both cases, the
impedance levels of the folded-cascode stage are about 2.5 to 3 times larger due to the
smaller transconductance of p-channel transistors as compared to n-channel transistors.
There are two major reasons for the popularity of cascode stages. The first is that they
can have quite large gain for a single stage due to the large impedances at the output. For this
to be the case, the current sources connected to the output nodes would normally be realized
using high-quality cascode current-mirrors. Normally this high gain is obtained without any
degradation in speed or sometimes with an improvement in speed. The second major reason
for their use is that they limit the voltage across the input drive transistor. This minimizes

© K.W. Martin, 1998 March 15, 2001

 36 Frequency Response of Electronic Circuits

any short-channel effects which can become more important with modern technologies hav-
ing very short channel-length transistors.
The analysis of the cascode gain stage will be done for the telescopic stage of Fig.
1.20(a). The same analysis, with only minor modifications, also applies for the
folded-cascode stage of Fig. 1.20(b).
From the section on cascode current-mirrors, we know that the impedance looking into
the drain of cascode transistor Q 2 is approximately given by
r d2 ≅ g m2 r ds1 r ds2 (1.160)
The total impedance at the output node will be r d2 in parallel with R L where R L is the out-
put impedance of the bias current source, I bias . Assuming I bias is a high-quality source
with an output impedance on the order of
R L ≈ g m-p r ds-p
2 (1.161)
then the total impedance at the output node will be
g m r ds 2
R out ≈ --------------- (1.162)
2
We have dropped the indices here under the assumption that the transistors are somewhat
matched and to simplify matters since we are only deriving a somewhat approximate
answer.
To find the approximate low-frequency gain, we can make use of some of the analysis
done previously for the common-gate stage. We found earlier that the low-frequency imped-
ance looking into the source of the common-gate or cascode transistor, Q 2 , was given by
g m2 + g s2 + g ds2 g m2
Y in2 = ------------------------------------------ ≅ -------------------- (1.163)
g ds2 g ds2
1 + ----------- 1 + -----------
GL GL
Note that the indices are now changed to reflect the fact that Q 2 is the common-gate transis-
tor rather than Q 1 . Substituting (1.161) into (1.163) and again assuming all elements are
somewhat matched so that the indices can be dropped, we have
gm
Y in2 ≈ ----------------------------- ≈ g ds (1.164)
g ds
1 + -------------------
2 ⁄g
-
g ds m
The gain from the input to the source of Q 2 is therefore given by
v s2 g m1 gm
-------- = – ---------------------------- ≈ – ----------- (1.165)
v in g ds1 + Y in2 2g ds
The overall gain can then be found using (1.165) and the equation developed earlier for the
gain from the source to the drain of a common-gate stage. We have
v s2 v out gm g m2 g m g m2 1 gm 2
A V = -------- ---------- ≈ – ----------- ------------------------ ≈ – ----------- ----------- ≈ – ---  -------- (1.166)
v in v s2 2g ds G L + g ds2 2g ds g ds2 2  g ds

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 37

The reader should be cautioned that (1.166) is only approximate primarily due to the dif-
ficulty of accurately determining the output admittance g ds for the different transistors. For
example, one problem in estimating g ds is that it is voltage dependant. Therefore, prudent
designers should never do a design where successful operation is dependant on knowing the
gain accurately rather than just knowing it will be greater than some minimum value.

Example 1.8 Assuming g m is on the order of 0.5mA/V and r ds is on the order of

100kΩ , what is the gain of the cascode amplifier.
Solution Using (1.166), we have A V = – 1250 . This is a fairly representative
number.

High-Frequency Analysis of Cascode Gain Stage

The exact high-frequency analysis of a cascode gain stage is usually left to simulation on
a computer, however, an approximate analysis is not too complicated. At high frequencies,
the time-constant due to the output node almost always dominates since the impedance is so
large at that node. The total capacitance at the output node, C out , is the parallel combination
of C gd2 + C db2 , the load capacitance C L , and the output capacitance of the bias current
source, C bias (normally, C L is the major contributor). Assuming, the time-constant at the
output stage dominates, the -3dB frequency is approximately equal to the inverse of the
time-constant. In other word, we have

2g ds 2
1
ω –3dB ≅ ------------------ ≈ -------------- (1.167)
R out C L g m C L
A more accurate estimate (though still not exact) may be found using the zero-value
time-constant analysis method. An advantage of this estimation technique is that it gives
some insight into the relative importance of each capacitor in determining the overall -3dB
frequency. The small-signal model being analyzed is shown in Fig. 1.21 where
C s2 = C db1 + C sb2 + C gs2

C d2 = C gd2 + C db2 + C L + C bias

(1.168)
In the zero-value time-constant analysis, all independent sources are set to zero (i.e. here,
V in is set to 0 volts) and each capacitor is considered in turn with all other capacitors set to
zero. As before, the corresponding time-constant is found and labeled τ Ci . Then the -3dB
frequency, ω –3dB , is estimated to be one over the sum of all the time-constants.
In this circuit, the first time-constant found is that corresponding to C gs1 and is labelled
τ Cgs1 . The resistance seen by C gs1 is R in and therefore
τ Cgs1 = C gs1 R in (1.169)

© K.W. Martin, 1998 March 15, 2001

 38 Frequency Response of Electronic Circuits

vout

gm2vs2
Rin vg1 Cgd1 Cd2
rds2 GL
vin vs2

Cgs1 gm1vg1
rds1 Cs2

Fig. 1.21 The small-signal model of the cascode gain stage.

The calculation of the time-constant corresponding to C gd1 is more involved, although

it is similar to the analysis just presented for the time-constant of the common-base stage
associated with C cb . C gd1 is replaced by a voltage source V x . Next, the resistance seen by
C gd1 is found by calculating the ratio of V x to i x (the current leaving V x ). The final
time-constant is then given by this resistance multiplied by C gd1 . The small-signal model
for this analysis is shown in Fig. 1.22(a) where the resistance R d1 is the parallel combina-

vx
ix ix
vg1

Rin gm1vg1 Rd1 vx Rd1

gm1vy
vy
Rin
(a) (b)

Fig. 1.22 Two equivalent small-signal models for calculating the resistance seen
by C gd1 .

tion of r ds1 and the impedance seen looking into the source of Q 2 (the cascode transistor)
at low frequencies. The circuit of Fig. 1.22(a) can be redrawn as the equivalent circuit shown
in Fig. 1.22(b). In this transformation, the node called ground was changed (which node is
called ground is arbitrary in an analysis) and the direction of the voltage-controlled cur-
rent-source was changed. The perceptive reader might notice that the circuit of Fig. 1.22(b)
is essentially the same as the used previously for finding the output impedance of a
source-degenerated current-source. We have

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 39

v y = i x R in (1.170)
Also
i x = ( v x – v y )G d1 – g m1 v y (1.171)
Substituting (1.170) into (1.171) and solving for v x ⁄ i x gives
v
r Cgd1 = ----x- = R d1 [ 1 + R in ( G d1 + g m1 ) ] (1.172)
ix
Now the admittance looking into the source of the cascode transistor, Q 2 , was found previ-
ously when analyzing the cascode amplifier at low frequencies. Namely,
Y s2 ≈ g ds (1.173)
The impedance R d1 is the parallel combination of this admittance and r ds1 . Therefore, we
have
r ds
R d1 ≈ ------- (1.174)
2
Substituting this result into (1.172), we have
r ds
r Cgd1 ≈ ------- [ 1 + R in ( 2g ds + g m1 ) ]
2
(1.175)
r ds
≈ ------- ( 1 + g m1 R in )
2
and therefore
r ds
τ Cgd1 ≈ C gd1 ------- ( 1 + g m1 R in ) (1.176)
2
If R in is large, say on the order of a transistor output impedance r ds , then this time-constant
is approximately given by
g m r ds 2
τ Cgd1 ≈ C gd1 --------------- (1.177)
2
This time-constant can be almost as large as the corresponding time-constant for a com-
mon-source amplifier — a fact not well known.
The resistance seen by capacitor C s2 is r ds1 in parallel with the impedance seen looking
into the source of Q 2 which, from (1.173), is approximately r ds . Thus, we have

r ds
τ Cs2 ≈ C s2 ------- (1.178)
2
The resistance seen by C d2 is the output impedance of the cascode amplifier which is

approximately given by ( g m r ds
2 ) ⁄ 2 from previous analysis at low frequencies. Thus, the

time-constant due to C d2 is given by

g m r ds 2
τ Cd2 ≈ C d2 --------------- (1.179)
2

© K.W. Martin, 1998 March 15, 2001

 40 Frequency Response of Electronic Circuits

Note that this time-constant has the same form as (1.177), but normally C d2 will be much
larger than C gd1 (because C L is often large) making τ Cd2 dominate.
The sum of the time-constants is then given by

τ total ≈ τ Cgs1 + τ Cgd1 + τ Cs2 + τ Cd2

g m r ds 2
r ds g m r ds 2 (1.180)
≈ C gs1 R in + C gd1 --------------- + C s2 ------- + C d2 ---------------
2 2 2
The -3 dB frequency is estimated to be one over this time-constant (i.e. ω –3dB = 1 ⁄ τ total ).

Example 1.9 Assume that for both the input transistor and the cascode transistor,
g m = 1mA ⁄ V , r ds = 100kΩ , R in = 180kΩ , C L = 5pF , C gs = 0.2pF ,
C gd = 15pfF , C sb = 40fF , C db = 20fF and C bias = 20fF . Estimate the
-3dB frequency of the cascode amplifier of Fig. 1.20(a).
Solution The time-constants associated with each capacitor are readily evalu-
ated using (1.180). First note that

C s2 = C db1 + C sb2 + C gs2 = 0.26pF

(1.181)
C d2 = C gd2 + C db2 + C L + C bias = 5.055pF
We have

τ Cgs1 = C gs1 R in = 36ns

g m r ds2
τ Cgd1 = C gd1 --------------- = 75ns
2
r ds (1.182)
τ Cs2 = C s2 ------- = 13ns
2
g m r ds 2
τ Cd2 = C d2 --------------- = 25.3us
2

As expected, the time-constant at the output node dominates and the sec-
ond most important time-constant is that due to C gd1 although its effect on
the -3dB frequency is negligible. Therefore, the -3dB frequency is very
accurately given by ω –3dB ≅ 1 ⁄ τ Cd2 = 2π × 6.3kHz .

Before leaving this section, some comments should be made about the high-frequency
operation of cascode-gain stages. As we saw above, typically one pole dominates and thus
we can reasonably model the amplifier gain as

Av
A(s) = -----------------------------------
- (1.183)
1 + ( s ⁄ ω –3dB )
Thus, at frequencies substantially larger than ω –3dB , which is usually the frequency band of

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 41

operation, the gain is approximately given by

Av g m1
A ( s ) ≅ ---------------------- ≅ – ---------
- (1.184)
s ⁄ ω –3dB sC L
using (1.166) and (1.167). It should also be noted that the approximations of (1.167) and
(1.184) are quite good unless either the source impedance or source capacitance is very
large. In addition, at frequencies much larger than the -3dB frequency, the admittance at the
source of Q 2 can be found by making use of (1.163) where G L is replaced by G L + sC L .
Such a substitution results in

g m2
Y in2 = ---------------------------------
g ds2
1 + ----------------------- -
G L + sC L
G L + sC L (1.185)
= g m2  -----------------------------------------
 g ds2 + G L + sC L
G L + sC L
≅ g m2  ---------------------------
 g ds2 + sC L

At frequencies where ω » 1 ⁄ ( r ds C L ) , the terms in s dominate and Y in2 ≅ g m2 . The

approximate time-constant due to the node at the source of Q 2 is then simply given by the
total capacitance at that node divided by g m2 . The total capacitance at the source of Q 2 is
C gs2 in parallel with C db1 in parallel with C gd1 . Because this capacitance is not exces-
sively large and the impedance at the node, 1 ⁄ g m2 , is small, the time-constant at that node
can usually be neglected. However, in amplifiers with a small source impedance, this node
would still be the primary factor determining the second pole of a cascode amplifier. It is
very easy to derive an upper-bound on the time-constant at the source of Q 2 . Almost always,
C db1 in parallel with C gd1 will be less than C gs2 . Therefore, the total capacitance at the
source of Q 2 is equal to KC gs2 where K is between 1 and 2 (usually closer to 1). Using

g m2 = µ p C ox  ----- V eff2
W
(1.186)
 L 2
for the folded-cascode amplifier (for the telescopic-cascode amplifier substitute µ n for µ p )
and
2
C S2 = KC gs2 = K --- ( WL ) C ox (1.187)
3 2
gives the approximate frequency of the second pole (ignoring the time-constants at other
non-dominant nodes) is given by
1 g m2 3µ p V eff2 3µ p V eff2
ω p2 ≅ -------- = --------- - > --------------------
- = -------------------- - (1.188)
τ S2 C S2 2KL 22 4L 22

© K.W. Martin, 1998 March 15, 2001

 42 Frequency Response of Electronic Circuits

This equation is an upper-limit on the unity-gain frequency of any amplifier that uses a
cascode gain stage. Note that (1.188) is relatively independent of that actual design once
V eff2 is chosen which is usually determined by maximum signal-handling requirements.
Also note the very strong dependance on the channel length.

Example 1.10 Estimate the lower-bound on the frequency of the second-pole of a

folded-cascode amplifier for a 0.8µm technology where a typical value of
0.25V is chosen for V eff2 .
Solution Normally, a minimum length of a cascode transistor in an analog cir-
cuit might be 25% to 50% the minimum length of transistors used in digital
circuits. Therefore, assuming L 2 = 1.5 ⋅ 0.8µm = 1.2µm , and using
2
µ p = 0.02M ⁄ V ⋅ s , and V eff2 = 0.25V , gives
9
ω p2 > 1.7 ×10 rad = 2π ⋅ 276MHz . For a telescopic-cascode amplifier,
the upper-bound would be 690MHz . In most practical op-amp designs,
the unity-gain frequency of a typical design might be limited to around
one-half the frequency of the lower-bound on the frequency of the sec-
ond-pole, or 138MHz and 345MHz , respectively.

Finally, the time-constant at the gate of the input transistor can be important when the
source resistance is large, though not as important as in a common-source stage since a cas-
code gain-stage does not suffer as much from a “Miller Effect”. In other words, at high fre-
quencies, the effective impedance at the source of Q 2 has decreased to 1 ⁄ g m2 and there is
not much gain from the gate of Q 1 to the source of Q 2 . Recall that in a common-source
amplifier, the effective size of C gd1 was magnified by the gain of the common-source
amplifier.

Source-Follower Amplifier
Before proceeding, it should be stated here that the high-frequency analysis of
source-follower amplifiers is somewhat involved. It will be shown that these types of ampli-
fiers can have complex poles and thus a designer should be careful that the circuit does not
exhibit too much overshoot and ringing. Also shown is a compensation circuit that will
result in only real axis poles and therefore no overshoot and ringing. However, due to the
complexity of the material, it is suggested that only advanced readers cover this sub-section
in detail.
The frequency response of the source follower can be found by modelling the source as
a Norton equivalent circuit and adding a load capacitance as shown in Fig. 1.23. The
small-signal model used for this circuit, which includes the parasitic capacitances, is shown
in Fig. 1.24. Capacitor C s includes both the load capacitor, C L , and the parasitic capacitor
C sb1 . Similar to what was done at low-frequencies, r ds1 , r ds2 , and the voltage-controlled

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 43

Vout
Iin Rin Cin
Ibias CL

Fig. 1.23 The configuration used to analyze the frequency response of the source
follower.

Cgd1 vd1

iin Cin vgs1 rds1

Rin Cgs1 g m1v gs1 gs1vs1
vout
vs1
rds2 Cs

Fig. 1.24 An equivalent small-signal model for the source-follower.

© K.W. Martin, 1998 March 15, 2001

 44 Frequency Response of Electronic Circuits

current source modelling the body-effect current-source can be modelled by a single resis-
tor. This allows us to analyze the simplified small-signal model shown in Fig. 1.25 where

Yg
vg1

iin vgs1
Rin C' in Cgs1 g m1v gs1 vout

C' in = C in + C gd1 Rs1 Cs

Fig. 1.25 A simplified equivalent small-signal model for the source-follower.

again R s1 = r ds1 || r ds2 || 1 ⁄ g s1 and the input capacitance is given by C' in = C in + C gd1 .
Nodal analysis is possible, but it is very complicated for this example. The analysis will
proceed in three steps. First, the gain from v g1 to v out will be found. Next, the admittance,
Y g , looking into the gate of Q 1 , but not taking into account C gd1 , will be found. Then, the
gain from i in to v g1 will be found. Finally, the overall gain from v in to v out will be found
and the results interpreted.
At node, v out , setting the sum of the currents leaving the node to zero, we have
v out ( sC s + sC gs1 + G s1 ) – v g1 sC gs1 – g m1 ( v g1 – v out ) = 0 (1.189)
Solving for v out ⁄ v g1 , we have
v out sC gs1 + g m1
---------- = ----------------------------------------------------------------- (1.190)
v g1 s ( C gs1 + C s ) + g m1 + G s1
The next step is to calculate the admittance, Y g , looking into the gate of Q 1 , but not taking
into account the current going into C gd1 . The input current is given by
i g1 = ( v g1 – v out )sC gs1 (1.191)
Using (1.190) to eliminate v out in (1.191) and solving for Y g = i g1 ⁄ v g1 , we have
i g1 sC gs1 ( sC s + G s1 )
Y g = -------
- = ----------------------------------------------------------------- (1.192)
v g1 s ( C gs1 + C s ) + g m1 + G s1
Also, an equation can be written relating the input current, i in , to the gate voltage, v g1 , as
i in = v g1 ( sC' in + G in + Y g ) (1.193)
Substituting (1.192) into (1.193) and rearranging gives

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 45

v g1 s ( C gs1 + C s ) + g m1 + G s1
-------- = ----------------------------------------------------------------- (1.194)
i in a + sb + s 2 c
where
a = G in ( g m1 + G s1 )
b = G in ( C gs1 + C s ) + C' in ( g m1 + G s1 ) + C gs1 G s1 (1.195)
c = C gs1 C s + C' in ( C gs1 + C s )

Using (1.190) and (1.194), we then have

v out sC gs1 + g m1
A ( s ) = ---------
- = ------------------------------
- (1.196)
i in a + sb + s 2 c
Thus, we see that the transfer-function is second-order. Specifically, it has two poles (roots
of the denominator) which may be either real or complex-conjugate. If they are com-
plex-conjugate, then the step response of the circuit will exhibit over-shoot and possibly
ringing. This potential problem is a disadvantage when using source-followers.
To determine if the transfer function will exhibit ringing, (1.196) can be written in the
form

N(s)
A ( s ) = A ( 0 ) ---------------------------------2- (1.197)
s s
1 + ----------- + ------2
ω0 Q ω0

where ω 0 and Q can be found by equating the coefficients of (1.197) to the coefficients of
(1.196). Here, parameter ω o is called the resonant frequency and parameter Q is called the
Q-factor4 [Sedra, 1991]. It is well known that if Q < 1 ⁄ 2 ≈ 0.707 , then the magnitude of
the transfer-function will have its maximum at dc and there will be no peaking (assuming the
zero is at a very-high frequency and therefore has negligible effect). Furthermore, for
Q = 1 ⁄ 2 , the -3dB frequency is equal to ω o . When the time-domain response is investi-
gated, restrictions on the Q-factor can also be found to guarantee no peaking for a step input.
Specifically, for there to be no peaking in the step-response, it is necessary that both poles be
real which is equivalent to the requirement that Q ≤ 0.5 . In the case where Q > 0.5 , the per-
centage overshoot of the output voltage can be shown to be given by

π
– -----------------------
2
4Q – 1
% overshoot = 100e . (1.198)
For the source-follower, equating the coefficients of (1.197) to the coefficients of (1.196) and
solving for ω 0 and Q results in

4. The Q-factor is 1/2 times the inverse of the damping-factor. The damping-factor is an alternative method of indicating
the pole locations in second-order transfer-functions.

© K.W. Martin, 1998 March 15, 2001

 46 Frequency Response of Electronic Circuits

G in ( g m1 + G s1 )
ω0 = ----------------------------------------------------------------- (1.199)
C gs1 C s + C' in ( C gs1 + C s )

G in ( g m1 + G s1 ) [ C gs1 C s + C' in ( C gs1 + C s ) ]

Q = ------------------------------------------------------------------------------------------------------------------ (1.200)
G in C s + C' in ( g m1 + G s1 ) + C gs1 G s1
If Q is greater than 0.5, the poles will be complex-conjugate and the circuit will exhibit
overshoot. Although this Q equation is rather complex, it is interesting to note that if C s
and/or C' in becomes large (i.e. a large load and/or input capacitor), then Q becomes small
and there will be no overshoot (though the circuit will be slow). As well, when C' in and
G s1 becomes small ( G s1 becomes small when the transistor’s source is connected to the
substrate which eliminates the body-effect), then the circuit will have a large Q (i.e. large
ringing) when G in becomes small and C s ≈ C gs1 . In summary, source follower (and
emitter-follower) circuits can exhibit large amounts of overshoot and ringing under cer-
tain conditions. Fortunately, the parasitic capacitances and output impedances in practical
microcircuits typically result in only moderate overshoot for worst-case conditions.
Finally, note also that the numerator zero of the transfer-function is on the negative
real-axis at a frequency given by

– g m1
ω z = ------------
- (1.201)
C gs1
and is typically at a much higher frequency than ω o .

Example 1.11 Using the parameters in Example 1.7, and also assuming
R in = 180kΩ , C L = 10pF , C gs1 = 0.2pF , C gd1 = 15fF , C sb1 = 40fF ,
and C in = 30fF , find ω 0 , Q and the frequency of the zero for the
source-follower of Fig. 1.23.
Solution It is straightforward to show g m1 = 1.06mA ⁄ V , r ds1 = 128kΩ ,
r ds2 = 192kΩ , and g s1 = 0.16mA ⁄ V . Thus, we have
C' in = C in + C gd1 = 45fF (1.202)
G s1 = g s1 + g ds1 + g ds2 = 0.173mA ⁄ V (1.203)
C s = C L + C sb1 = 10.04pF (1.204)
and so we can find ω o as
G in ( g m1 + G s1 )
ω0 = -----------------------------------------------------------------
C gs1 C s + C' in ( C gs1 + C s )
7
= 5.24 ×10 rad ⁄ s = 2π × 8.34 MHz (1.205)

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 47

G in ( g m1 + G s1 ) [ C gs1 C s + C' in ( C gs1 + C s ) ]

Q = ------------------------------------------------------------------------------------------------------------------
G in C s + C' in ( g m1 + G s1 )
= 1.17 (1.206)
This results in an overshoot for a step input given by
π
– -----------------------
2
4Q – 1
% overshoot = 100e = 22.7% (1.207)
which is quite large though the numbers used are reasonable values and
could easily occur in practice. The zero frequency is found using (1.201) to
be 844MHz and thus it can almost certainly be ignored.

A bipolar transistor emitter-follower is very similar to a CMOS source-follower and

therefore its high-frequency analysis will not be included here. However, it should be men-
tioned that the problem of complex-conjugate poles can be more severe for bipolar emit-
ter-followers especially when resistive loads are used.
For both CMOS source-followers and bipolar emitter-followers, when complex conju-
gate poles occur, they can be eliminated by adding a compensation network. To see this, note
that (1.192) can be re-written as

1
Y g = sC 2 + --------------------------- (1.208)
1
– R 1 – ----------
sC 1
where
C gs1 ( C s g m1 – C gs1 G s1 ) g m1 C gs1 C s
C 1 = -------------------------------------------------------------- ≅ -------------------------------------------------------------
( g m1 + G s1 ) ( C gs1 + C s ) ( g m1 + G s1 ) ( C gs1 + C s )
( C gs1 + C s ) 2 ( C gs1 + C s ) 2
R 1 = -------------------------------------------------------------- ≅ -------------------------------- (1.209)
C gs1 ( C s g m1 – C gs1 G s1 ) C gs1 C s g m1
C gs1 C s
C 2 = ------------------------
-
C gs1 + C s
and the approximation is due to the fact that typically C s > C gs1 and g m1 > G s1 . This is the
same admittance as the circuit shown in Fig. 1.26. Thus, the input admittance is same as a
capacitor in parallel with a series combination of a negative capacitor and a negative resistor.
If a third network consisting of a capacitor of size C 1 and a resistor of size R 1 , in series,
was connected to the gate of the source-follower as shown in Fig. 1.27, then the negative ele-
ments would be cancelled. The resulting input admittance would then simply be C 2 as given
in (1.209). In this case (1.194) becomes

© K.W. Martin, 1998 March 15, 2001

 48 Frequency Response of Electronic Circuits

-C1
Yg C2
-R1

Fig. 1.26 A circuit having the same admittance as the input impedance looking
into the gate of a source-follower (ignoring C gd )

C1 Vout
Iin
Rin C in
R1 CL
Ibias

Fig. 1.27 Adding a compensation network ( C 1 and R 1 ) to compensate for the

negative components of the admittance looking into the gate of the source-fol-
lower.

v g1 1
-------- = ----------------------------------------------------------------- (1.210)
i in C gs1 C s 
G in + s  C' in + ------------------------ -
 C gs1 + C s

and (1.196) becomes

C gs1
 1 + s ----------- -
v out  g m1   g m1 
A ( s ) = ---------- = R in ------------------------- ------------------------------------------ (1.211)
 g m1 + G s1 
1 + -----  1 + -----
i in s s
 p1   p 2
where
G in G in
p 1 = ----------------------------------------- ≅ --------------------------- (1.212)
C gs1 C L C' in + C gs1
C' in + ------------------------ -
C gs1 + C L

© K.W. Martin, 2001 March 15, 2001

 1.2 Frequency Response of Transistor Circuits 49

g m1 + G s1 g m1 + G s1
p 2 = ------------------------- ≅ ------------------------- (1.213)
C gs + C L CL
The approximation is accurate when C s » C gs1 . Irrespective of the approximation, the poles
are now guaranteed real and no over-shoot will occur.
Therefore, when designing source-followers (or emitter followers), the recommended
procedure is to check to see if the poles are complex using either (1.200) or a SPICE tran-
sient analysis to look for overshoot. When the poles are complex, then either increase C in
and/or C s or, alternatively, add the compensating network as shown Fig. 1.27.

Example 1.12 Using the same parameters as in Example 1.11, find the compensa-
tion network and the resulting first and second poles of the source follower
of Fig. 1.23.
Solution Using (1.209), we have
g m1 C gs1 C s
C 1 ≅ ------------------------------------------------------------- = 0.170fF (1.214)
( g m1 + G s1 ) ( C gs1 + C s )
and
( C gs1 + C s ) 2
R 1 ≅ -------------------------------- ≅ 49.3kΩ (1.215)
C gs1 C s g m1
The capacitor is a reasonable value to be realized on chip. The resistor
could be realized by a MOS transistor biased in the triode (i.e. linear)
region. Assuming the compensation network is used, the poles of the
transfer function would then become
G in
p 1 ≅ ------------------------------ = 2π × 3.61MHz (1.216)
C gs1 + C gd1
and
g m1 + G s1
p 2 = ------------------------- = 2π × 19.3MHz (1.217)
C gs + C L
The speed penalty paid for using the compensation network is quite high
as the pole frequency without compensation was around 8MHz while
here the dominant pole is at 3.6MHz .

Finally, it should be mentioned here that if the source-follower buffer is intended to be

used in an op-amp (and thus feedback will be placed around the buffer), and if the resonant
frequency of the source-follower is substantially greater than the unity-gain frequency of the
amplifier, then the overshoot can be tolerated and no compensation network is necessary.

© K.W. Martin, 1998 March 15, 2001

 50 Frequency Response of Electronic Circuits

1.3 Frequency Response of DIfferential Amplifiers

Before considering the frequency response of differential amplifiers it is expedient to
introduce the high-frequency T-models for the BJT and MOS transistors which simplify
some analyses.

High-Frequency T-Model
Similar to low-frequency analysis, there exists a T-model for high-frequency modelling
that sometimes results in simpler analyses and greater insight, especially when the gain is
not large. Consider the small-signal model for a BJT model shown repeated in Fig. 1.28 and

rb Ccb
vb
vb ′ vc
i bx
Cbe rπ gmvbe ro Ccs
vbe

i ex

ve

Fig. 1.28 A small-signal model a BJT.

its corresponding T-model shown in Fig. 1.29. In both cases, the small-signal circuits have
been drawn to make their equivalences more understandable. For the circuit of Fig. 1.28, we
have

i bx = v be ( sC be + g π ) (1.218)
where g π = 1 ⁄ r π . For the circuit of Fig. 1.29, we also have

i bx = v be ( sC be + g e ) – g m v be (1.219)
= v be ( sC be + g e – g m )
= v be ( sC be + g π )
since g e = 1 ⁄ r e = g m + g π . In addition, for the circuit Fig. 1.28, we have

© K.W. Martin, 2001 March 15, 2001

 1.3 Frequency Response of DIfferential Amplifiers 51

vc
Ccb gmvbe
Ccs
rb i bx
vb
vb ′
ro
Cbe re
vbe

i ex

ve
Fig. 1.29 A high-frequency T-model model for a BJT.

i ex = v be ( sC be + g π ) + g m v be
= v be ( sC be + g π + g m ) (1.220)
= v be ( sC be + g e )
But this equation also holds for the circuit of Fig. 1.29. In addition, the current-voltage rela-
tions for all other components are identical. Thus, the current-voltage relations of the two
circuits are identical and therefore the circuits are equivalent. The T-model significantly sim-
plifies the analysis of differential-pair-based amplifiers especially when the transistor output
resistors can be ignored. A variant of the high-frequency T-model is based on the fact that
i e ≅ i ex . Using this approximation, we have
ie
v be ≅ ------------------------- (1.221)
g e + sC be
This allows for an alternative equation for the voltage-dependant current source. Namely,

gm gm ⁄ ge 1
g m v be ≅ -------------------------i e ≅ ----------------------i e ≅ ----------------------i e (1.222)
g e + sC be C be C be
1 + s --------- 1 + s ---------
gm gm
Finally, note that in many analyses, the base-resistance, r b , is often ignored (or taken into
account by taking R s , the input source resistance, larger than its actual size), but in applica-
tions where the source resistance is small, it can be important.

© K.W. Martin, 1998 March 15, 2001

 52 Frequency Response of Electronic Circuits

The high-frequency T-model for MOS transistors is similar and is shown in Fig. 1.30.

vd
g m v gs
C gd C db
gs vs
r ds
vg
1
r s = -------
C gs gm v gs

vs

rs C sb

vs ′

Fig. 1.30 A high-frequency T-model model for an MOS transistor.

Again, a popular variant is to replace the depedant source g m v gs with

1
g m v gs ≅ ---------------------- i s (1.223)
C gs
1 + s ---------
gm
In most analyses, the source resistance r s is usually ignored, but it can be important in some
high-frequency analyses.

Symetric Differential Amplifier

A symetric bipolar differential amplifier with resistive loads is shown in Fig. 1.31. A
small-signal model of this amplifier using the T-model is shown in Fig. 1.32. Here we have
ignored both the base resistances and the transistor output resistances in order to simplify
the circuit. Neither of these simplifications is really necessary as the former could be
accounted for by taking R s1 and R s2 larger, and the latter could be accounted for by taking
R c1 and R c2 smaller.
Assuming R c1 = R c2 , R s1 = R s2 , and that the transistors are matched, then the cir-
- +
cuit is perfectly symetric. If one also assumes that v in = – v in , then based on the syme-
try, the node voltage v e will never change similar to if it was connected to a small-signal
ground. Indeed, if the node v e was connected to a small-signal ground, then circuit will

© K.W. Martin, 2001 March 15, 2001

 1.3 Frequency Response of DIfferential Amplifiers 53

V cc

R c1 = R c2
- +
V out V out
+ -
V in Q1 Q2 V in
R s1 R s2

I bias

Fig. 1.31 A symetric BJT differential amplifier.

R c1 R c2

- +
v out v out
g m1 v be1

C cb1 C cb2
R s1 g m1 v be2 R s2
+ -
v in v in
r e1 C be1 C r e2
be2
v be1 v be2

ve
Fig. 1.32 The small-signal model of the symetric differential amplifier of Fig.
1.31.

operate identically. This allows one to simplify the analysis to that of the half-circuit; with
the half-circuit being identical to that of a common-emitter amplifier which we have studied
previously. Given this equivalence, one can state immediately, the amplifier bandwidth,
ω -3dB , is given by

1
ω -3dB = ------------------------------------------------------------------------------------------------ (1.224)
( R s1 || r π1 ) [ C be1 + C cb1 ( 1 + g m1 R c1 ) ]

© K.W. Martin, 1998 March 15, 2001

 54 Frequency Response of Electronic Circuits

Single-Ended Differential Amplifer

Consider next the frequency response of the single-ended differential amplifier shown in
Fig. 1.33. Before we analyze this amplifier for its frequency response, note that this configu-

V cc

Rc

V out
Rs
+
V in Q1 Q2

I bias

Fig. 1.33 A single-ended output BJT differential amplifier.

ration can be considered a two-stage amplifier consister of an emitter follower cascaded by a

common-base stage. The small-signal model for this amplifier is shown in Fig. 1.34. Note

Rc

+
v out
g m1 v be1

C cb1 C cb2
Rs g m1 v be2
+
v in
r e1 C be1 C r e2
be2
v be1 v be2

ve
Fig. 1.34 The small-signal model of the single-ended differential amplifier of Fig.
1.33.

both the similarities and the differences of the circuits shown in Fig. 1.32 and Fig. 1.34. The

© K.W. Martin, 2001 March 15, 2001

 1.3 Frequency Response of DIfferential Amplifiers 55

differences, though they appear minor result in a significantly large bandwidth (and approx-
imately half the gain).
Before starting the analysis, note that between the gate of Q 1 and ground are two identi-
cal networks, namely r e1 in parallel with C be1 , and r e2 in parallel with C be2 . This obser-
vation justifys simplifying the circuit into the equivalent small-signal model shown in Fig.

Rs v b1
v in v b1 v out
C be C cb1 g m --------
2r e ---------- 2 C cb2
2 v b1 Rc
g m --------
2

Fig. 1.35 A simplified and equivalent small-signal model of the amplifier of Fig.
1.33.

1.35. This simplification assumes the transistor parameters are matched which allows one to
drop the subscripts. The reader is encouraged to spend a little time convincing themselves
the circuits are equivalent; it is suggest this would be time well spent. Next notice that the
left-most dependant current source is proportional to the voltage across it; this make it
equivalent to a negative resistor (given the direction of the current source) of size – 2 ⁄ g m .
We can simplify the circuit even more by noting that the parallel equivalent of 2r e with
– 2 ⁄ g m , denoted r eq , has value

1
r eq = ---------------------------------
ge ⁄ 2 – gm ⁄ 2
2 (1.225)
= ------
gπ
= 2r π
The final simplifified circuit is now shown in Fig. 1.36 where C 1 = C be ⁄ 2 + C cb1 . The
transfer function is now easily found almost by inspection to be

2r π   g m 
 -------------------- – - Rc
 2r π + R s-  ------ 2 
H(s) = --------------------------------------------------------------------------------------- (1.226)
( 1 + sC 1 ( R s || 2r π ) ) ( 1 + sC cb2 R c )
The first pole frequency due to the time constant at the base of Q 1 is given by

© K.W. Martin, 1998 March 15, 2001

 56 Frequency Response of Electronic Circuits

Rs v b1
v in v b1 v out
g m --------
2 C cb2
2r π Rc
C1

Fig. 1.36 A simplified and equivalent small-signal model of the amplifier of Fig.
1.33.

1
ω p1 = -------------------------------
C 1 ( R s || 2r π )
-
(1.227)
1
= -----------------------------------------------------------------
( C be ⁄ 2 + C cb1 ) ( R s || 2r π )
Note that there is no Miller-multiplication term (i.e. 1 + A 0 ) multiplying the capacitance
C be . For this reason, the compound stage is must faster than a common-emitter stage.

MOS Differential Amplifier

An MOS differential amplifier is shown in Fig. 1.37. For this case, rather than analyze

Q3 Q4

v out

v in Q1 Q2

I bias

Fig. 1.37 A differential-input single-ended-output MOS amplifier.

the circuit properly, a rather simple-minded guess will be made at predicting the nodal time
constants and then the bandwidth will then be predicted as the inverse of the sum of the time
constants.

© K.W. Martin, 2001 March 15, 2001