EECS 242:

Receiver Architectures
Outline
! Complex baseband equivalent of a bandpass signal
! Double-conversion single-quadrature (Superheterodyne)
! Direct-conversion (Single-conversion single-quad,
homodyne, zero-IF)
! Weaver; Double-conversion double-quad
! Low-IF
! References
EECS 242
Complex Baseband
! Any passband waveform can be written in the following form:
EECS 242
s
p
(t) = a(t) cos [
c
t + (t)]
s
p
(t) = a(t) cos
c
t cos (t) a(t) sin
c
t sin(t)
s
p
(t) =

2s
c
(t) cos
c
t

2s
s
(t) sin
c
t
s
c
(t) a(t) cos (t) = I(t) s
s
(t) a(t) sin(t) = Q(t)
a(t) = |s(t)| =

s
2
c
(t) + s
2
s
(t)
(t) = tan
1
s
s
(t)
s
c
(t)
s(t) = s
c
(t) + js
s
(t) = I(t) + jQ(t)
s
p
(t) = Re

2s(t)e
j
c
t

! We define the complex baseband signal and show that all operations at
passband have a simple equivalent at complex baseband:
||s||
=
||s
p
||
2
Orthogonality I/Q
! An important relationship is the orthogonality between the modulated I
and Q signals. This can be proved as follows (Parsevals Relation):
EECS 242
x
c
(t) =

2s
c
(t) cos
c
t x
s
(t) =

2s
s
(t) sin
c
t
< x
c
, x
s
>=< X
c
, X
s
>= 0 < X
c
, X
s
>=

X
c
(f)X

s
(f)df
cos =
1
2
(e
j
+ e
j
)
x
c
(t) =
1

2
(s
c
(t)e
j
c
t
+ s
c
(t)e
j
c
t
) X
c
(f) =
1

2
(S
c
(f f
c
) +S
c
(f +f
c
))
X
s
(f) =
1

2j
(S
s
(f f
c
) S
s
(f + f
c
))
< X
c
, X
s
>=
1
2j

((S
c
(f f
c
) + S
c
(f + f
c
)) (S

s
(f f
c
) S

s
(f + f
c
))) df
Orthogonality (Freq. Dom.)
! In the above integral, if the carrier frequency is larger than
the signal bandwidth, then the frequency shifted signals do
not overlap
EECS 242
S
c
(f f
c
)S

s
(f + f
c
) 0
S
c
(f + f
c
)S

s
(f f
c
) 0
< X
c
, X
s
>=
1
2j

S
c
(f f
c
)S

s
(f f
c
)df

S
c
(f + f
c
)S

s
(f + f
c
)df

< X
c
, X
s
>=
1
2j

S
c
(f)S

s
(f)df

S
c
(f)S

s
(f)df

= 0
! Due to this orthogonality, we can double the bandwidth of
our signal my modulating the I and Q independently.
Also, we have
< u
p
, v
p
>=< u
c
, v
c
> + < u
s
, v
s
>= Re (< u, v >)
Complex Baseband Spectrum
! Since the passband signal is real, it has a conjugate symmetric
spectrum about the origin. Lets define the positive portion as follows:
EECS 242
S
+
p
(f) = S
p
(f)u(f)
S(f) =

2S
+
p
(f + f
c
)
S
p
(f) =
S(f f
c
) + S

(f f
c
)

2
v(t) =

2s(t)e
j
c
t
V (f) =

2S(f f
c
)
S
p
(t) = Re(v(t)) =
v(t) + v(t)

2
S
p
(f) =
V (f) +V

(f)
2
=
S(f f
c
) +S

(f f
c
)

2
! Then the spectrum of the passband and baseband complex signal are
related by:
! Since:
The Image Problem
! After low-pass filtering the mixer output, the IF is given by
LNA
LO
IF IF
IF
m
r
(t) cos(
LO
+
IF
)tcos(
LO
)t =
1
2
m
r
(t) (cos(2
LO
+
IF
)t + cos(
IF
)t)
m
i
(t) cos(
LO

IF
)tcos(
LO
)t =
1
2
m
i
(t) (cos(2
LO

IF
)t + cos(
IF
)t)
IF
output
=
1
2
(m
i
(t) + m
r
(t)) cos(
IF
)t
EECS 242
Image Problem (Freq Dom)
! Complex modulation shifts in only one direction real
modulation shifts up and down
EECS 242
RF
+
()
RF

() RF
+
(
0
)
RF
+
( +
0
)
IM
+
()
IM

()
IM
+
(
0
)
IM

( +
0
)
LO
LO
IF IF
e
jLOt
e
jLOt
RF

(
0
)
IM

(
0
)
LO
LO
LO
LO
IM

()
e
jLOt
e
jLOt
RF

()
RF

( +
0
)
RF
+
()
IM
+
() IM
+
( +
0
)
Complex Modulation (Positive Frequency)
Complex Modulation (Positive Frequency)
Real Modulation
Superheterodyne Architecture
! The choice of the IF frequency dictated by:
! If the IF is set too low, then we require a very high-Q image reject
filter, which introduces more loss and therefore higher noise figure
in the receiver (not to mention cost).
! If the IF is set too high, then subsequent stages consume more
power (VGA and filters)
! Typical IF frequency is 100-200 MHz.
LNA
Of chip
Passive
BPF
Image
Filter
IF
Filter
EECS 242
LO Planning in Superhet
! Two separate VCOs and synthesizers are used. The IF LO is fixed, while the
RF LO is variable to down-convert the desired channel to the passband of
the IF filter (SAW).
! This typically results in a 3-4 chip solution with many off-chip components.
! LO
1
should never be made close to an integer multiple of LO
2
for any
channel. The N
th
harmonic of the the fixed LO2 could leak into the RF
mixer and cause unwanted mixing.
LNA
Of chip
Passive
BPF
Image
Filter
IF
Filter
PLL
1
LO
1
PLL
2
LO
2
I
Q
IF
IF
LO
2
LO
1
n LO
2
IF
nLO
2
leaks
into RF mixer
EECS 242
The ! IF Problem
! Assume that there is a blocker half-way between the LO
and the desired channel. Due to second-order non-linearity
in the RF front-end:
EECS 242

m
blocker
(t) cos(
LO
+
1
2

IF
)t

2
= (m
blocker
(t))
2
+ (m
blocker
(t))
2
) cos(2
LO
+
IF
)t
! If the LO has a second-order component, then this signal
will fold right on top of the desired signal at IF:
IF
LO
IF
IF

(m
blocker
(t))
2
) cos(2
LO
+
IF
)t

cos(2
LO
)t = (m
blocker
(t))
2
cos(
IF
)t +
Note: Bandwidth expansion of blocker due to squaring operation.
Half-IF Continued
! If the IF stage has strong second-order non-linearity, then
the half-IF problem occurs through this mechanism:
EECS 242
IF
DC
IF
IF
DC
IF
2

m
blocker
(t) cos(
1
2

IF
)t

2
= (m
blocker
(t))
2
+ (m
blocker
(t))
2
cos(
IF
)t
! This highlights the importance of frequency planning. One
should select the IF by making sure that there is no strong
half-IF blocker. If one exists, then the second-order non-
linearity must be carefully managed.
Dual-Conversion Single-Quad
! Disadvantages:
! Requires bulky off-chip SAW filters
! As before, two synthesizers are required
! Typically a three chip solution (RF, IF, and Synth)
! Advantages:
! Robust. The clear choice for extremely high sensitivity radios
! High dynamic range SAW filter reduces/relaxes burden on active
circuits. This makes it much easier to design the active circuitry.
! By the same token, the power consumption is lower (< 25mA)
EECS 242
LNA
Of chip
Passive
BPF
Image
Filter
IF
Filter
PLL
1
LO
1
PLL
2
LO
2
I
Q
Complex Mixer
! A complex mixer is derived by simple substitution.
! Note that a complex exponential only introduces a
frequency shift in one direction (no image rejection
problems).
EECS 242
+
+
+

+
+
Hilbert Architecture
! Image suppression by proper phase shifting.
EECS 242
+

IF

A
B
C
LNA
RF = m
r
(t) cos(
LO
+
IF
)t +m
i
(t) cos(
LO

IF
)t
A = RF cos(
LO
t) =
1
2
m
r
(t) (cos(2
LO
+
IF
)t + cos(
IF
)t) +
1
2
m
i
(t) (cos(2
LO

IF
)t + cos(
IF
)t)
B = RF sin(
LO
t) =
1
2
m
r
(t) (sin(2
LO
+
IF
)t sin(
IF
)t) +
1
2
m
i
(t) (sin(2
LO

IF
)t + sin(
IF
)t)
C =
1
2
m
r
(t) (cos(2
LO
+
IF
)t + cos(
IF
)t) +
1
2
m
i
(t) (cos(2
LO

IF
)t cos(
IF
)t)
IF
+
= A+C = m
r
(t) cos(
IF
t)
IF

= AC = m
i
(t) cos(
IF
t)
Sine/Cosine Together
! Since the sine treats positive/negative frequencies
differently (above/below LO), we can exploit this behavior
! A 90 phase shift is needed to eliminate the image
! 90 phase shift equivalent to multiply by j sign(f)
EECS 242
IF IF LO LO
IF IF LO LO
Cosine Modulation
Sine Modulation
IF IF LO LO
Delayed Sine Modulation
/j
/j
/j
/j
Hilbert Implementation
! Advantages:
! Remove the external image-reject SAW filter
! Better integration
! Requires extremely good matching of components (paths
gain/phase). Without trimming/calibration, only ~40dB
image rejection is possible. Many applications require
60dB or more.
! Power hungry (more mixers and higher cap loading).
EECS 242
+

IF
A
B
C

D
LNA
Note: A real
implementation
uses 45/135 degree
phase shifters for
better matching/
tracking.
Gain/Phase Imbalance
IF = m
r
(t)

cos(
IF
t) cos(

2
) sin(
IF
t) sin(

2
)

+ m
i
(t)

cos(
IF
t) cos(

2
) sin(
IF
t) sin(

2
)

EECS 242
A = RF (1 +) cos(
LO
t +

2
) =
1
2
m
r
(t)(1 +)

cos(2
LO
t +
IF
t +

2
) + cos(
IF
t

2
)

+
1
2
m
i
(t)(1 +)

cos(2
LO
t
IF
t +

2
) + cos(
IF
t +

2
)

B = RF (1 ) cos(
LO
t

2
) =
1
2
m
r
(t)(1 )

sin(2
LO
t +
IF
t

2
) sin(
IF
t

2
)

+
1
2
m
i
(t)(1 )

sin(2
LO
t
IF
t

2
) + sin(
IF
t

2
)

C =
1
2
m
r
(t)(1 )

cos(2
LO
t +
IF
t

2
) + cos(
IF
t

2
)

+
1
2
m
i
(t)(1 )

cos(2
LO
t
IF
t

2
) cos(
IF
t

2
)

IF = A+C =
m
r
(t)
2

(1 +) cos(
IF
t

2
) + (1 ) cos(
IF
t +

2
)

+
m
i
(t)
2

(1 +) cos(
IF
t +

2
) (1 ) cos(
IF
t

2
)

Image-Reject Ratio
! Level of image rejection depends on amplitude/phase
mismatch
! Typical op-chip values of 30-40 dB achieved (< 5, < 0.6
dB)
EECS 242
IR(dB) = 10 log

cos

2
sin

2
cos

2
+ sin

2

2
0 1 2 3 4
0
2
4
6
8
10
RF/IF Phase Shift, Fixed LO
! This requires a 90 degree phase shift across the band. Its
much easier to shift the phase of a single frequency (LO).
! Polyphase filters can be used to do this, but a broadband
implementation requires many stages (high loss)
EECS 242
+

IF

A
B
LNA
A
RF
RF

Weaver Architecture
! Eliminates the need for a phase shift in the signal path.
Easier to implement phase shift in the LO path.
! Can use a pair of quadrature VCOs. Requires 4X mixers!
! Sensitive to second image.
EECS 242
RF = m
r
(t) cos(
LO
1
+
IF
1
)t +m
i
(t) cos(
LO
1

IF
1
)t
IF
1
= LO
1
RF
IF = LO
2
IF
1
= LO
2
LO
1
+RF = RF (LO
1
LO
2
)
A
LPF
= cos
LO
1
t RF =
m
r
2
cos(
IF
1
)t +
m
i
2
cos(
IF
1
)t
B
LPF
= sin
LO
1
t RF =
m
r
2
sin(
IF
1
)t +
m
i
2
sin(
IF
1
)t
C
LPF
= Acos
LO2
t =
m
r
4
cos(
IF
)t +
m
i
4
cos(
IF
)t
D
LPF
= B sin
LO2
t =
m
r
4
cos(
IF
)t +
m
i
4
cos(
IF
)t
IF = C D =
m
r
2
cos
IF
t
+

IF
A
B
LNA
RF
C
D
LO
1
LO
2
+
Direct Conversion (Zero-IF)
! The most obvious choice of LO is the RF frequency, right?
IF = LO RF = DC?
! Why not?
! Even though the signal is its own image, if a complex
modulation is used, the complex envelope is asymmetric
and thus there is a mangling of the signal
EECS 242
LNA
LO
RF DC
DC
Let
RF
=
LO
=
0
m
r
(t) cos(
RF
)t cos(
LO
)t =
1
2
m
r
(t) (1 + cos(2
0
)t)
Direct Conversion (cont)
! Use orthogonal mixing to prevent signal folding and retain
both I and Q for complex demodulation (e.g. QPSK or
QAM)
! Since the image and the signal are the same, the image-
reject requirements are relaxed (its an SNR hit, so
typically 20-25 dB is adequate)
EECS 242

LNA
I
Q
Problems with Zero-IF
! Self-mixing of the LO signal is a big concern.
! LO self-mixing degrades the SNR. The signal that reflects
from the antenna and is gained up appears at the input of
the mixer and mixes down to DC.
! If the reflected signal varies in time, say due to a changing
VSWR on the antenna, then the DC offset is time-varying
EECS 242

LNA
I
Q
LO = p(t) cos(
LO
t +(t))
LO LO = p(t)
2
+ p(t)
2
cos(2
LO
t + 2(t))
DC
Dynamic DC
Ofset
DC Offset
! DC offsets that appear at the baseband experience a large
gain. This signal can easily saturate out the receive chain.
! A large AC coupling capacitor or a programmable DC-
offset cancellation loop is required. The HPF corner
should be low (kHz), which requires a large capacitor.
! Any transients require a large settling time as a result.
EECS 242

LNA
I
Q
60 dB Gain
1mV Ofset 1V Ofset
Sensitivity to 2
nd
Order Disto
! Assume two jammers have a frequency separation of !f:
EECS 242
s
1
= m
1
(t) cos(
1
t)
s
2
= m
2
(t) cos(
1
t +t)
(s
1
+ s
2
)
2
= (m
1
(t) cos
1
t)
2
+ (m
2
(t) cos
2
t)
2
+ 2m
1
(t)m
2
(t) cos(
1
t) cos(
1
+)t
LPF{(s
1
+ s
2
)
2
} = m
1
(t)
2
+ m
2
(t)
2
+ m
1
(t)m
2
(t) cos()t
! The two produce distortion at DC. The modulation of the
jammers gets doubled in bandwidth.
! If the jammers are close together, then their inter-
modulation can also fall into the band of the receiver.
! Even if it is out of band, it may be large enough to saturate
the receiver.
Sensitivity to 1/f Noise
! Since the IF is at DC, any low frequency noise, such as 1/f
noise, is particularly harmful.
! CMOS has much higher 1/f noise, which requires careful
device sizing to ensure good operation.
! Many cellular systems are narrowband and the entire
baseband may fall into the 1/f regime!
EECS 242
Noise
Density
1/f Slope, 20dB/dec
1
k
H
z
1
0

k
H
z
1
0
0

k
H
z
2
0
0

k
H
z
n
i
2
100 n
i
2
Example: GSM has a 200 kHz bandwidth. Suppose
that the flicker corner frequency is 100 kHz. The in-
band noise degradation is thus:
v
2
ave
=
1
200kHz

100kHz
1kHz
a
f
df +

200kHz
100kHz
bdf
a = 1kHz 100 v
2
i
b = v
2
i
v
2
ave
=
1
200kHz

a ln
100k
1k
+ b(200k 100k)

=
1
200kHz
(11.5a + 100kb) = 6.25v
2
i
Low IF Architecture
! Instead of going to DC, go a low IF, low enough so that
the IF circuitry and filters can be implemented on-chip, yet
high enough to avoid problems around DC (flicker noise,
offsets, etc). Typical IF is twice the signal bandwidth.
! The image is rejected through a complex filter.
EECS 242

LNA
I
Q
C
o
m
p
l
e
x

I
m
a
g
e
R
e
j
e
c
t

F
i
l
t
e
r
Double-Conversion Double-Quad
! The dual-conversion double-quad architecture has the
advantage of de-sensitizing the receiver gain and phase
imbalance of the I and Q paths.
EECS 242

+
+
+

+
+
I
Q
A
B
C
D
E
F
LNA
Analysis of Double/Double
! Assuming ideal quadrature and no gain errors:
EECS 242
RF = m
r
(t) cos(
LO1
+
LO2
+
IF
)t +m
i
(t) cos(
LO1
+
LO2

IF
)t+
A = LPF{RF cos(
LO1
t)} =
1
2

m
r
(t) cos(
LO2
+
IF
)t+
m
i
(t) cos(
LO2

IF
)t
B = LPF{RF sin(
LO1
t)} =
1
2

m
r
(t) sin(
LO2
+
IF
)t
m
i
(t) sin(
LO2

IF
)t
C = LPF{Acos(
LO2
t)} =
1
2

m
r
(t) cos(
IF
)t+
m
i
(t) cos(
IF
)t
D = LPF{Asin(
LO2
t)} =
1
2

m
r
(t) sin(
IF
)t+
m
i
(t) sin(
IF
)t
E = LPF{B cos(
LO2
t)} =
1
2

m
r
(t) sin(
IF
)t+
m
i
(t) sin(
IF
)t
F = LPF{B sin(
LO2
t)} =
1
2

m
r
(t) cos(
IF
)t+
m
i
(t) cos(
IF
)t
I = C F = (m
r
(t) +m
i
(t)) cos(
IF
)t
Q = D +E = (m
r
(t) m
i
(t)) sin(
IF
)t
Gain Error Analysis
! The gain mismatch is reduced since due to the product of
two small numbers (amplitude errors).
EECS 242
RF = m
r
(t) cos(
LO1
+
LO2
+
IF
)t +m
i
(t) cos(
LO1
+
LO2

IF
)t+
A = LPF{RF

1 +
a
1
2

cos(
LO1
t)} =
1
2

1 +
a
1
2

m
r
(t) cos(
LO2
+
IF
)t+
m
i
(t) cos(
LO2

IF
)t
B = LPF{RF

1
a
1
2

sin(
LO1
t)} =
1
2

1
a
1
2

m
r
(t) sin(
LO2
+
IF
)t
m
i
(t) sin(
LO2

IF
)t
C = LPF{A

1 +
a
2
2

cos(
LO2
t)} =
1
2

1 +
a
1
2

1 +
a
2
2

m
r
(t) cos(
IF
)t+
m
i
(t) cos(
IF
)t
D = LPF{A

1
a
2
2

sin(
LO2
t)} =
1
2

1 +
a
1
2

1
a
2
2

m
r
(t) sin(
IF
)t+
m
i
(t) sin(
IF
)t
E = LPF{B

1 +
a
2
2

cos(
LO2
t)} =
1
2

1
a
1
2

1 +
a
2
2

m
r
(t) sin(
IF
)t+
m
i
(t) sin(
IF
)t
F = LPF{B

1
a
2
2

sin(
LO2
t)} =
1
2

1
a
1
2

1
a
2
2

m
r
(t) cos(
IF
)t+
m
i
(t) cos(
IF
)t
I = C F = (1 +a
1
a
2
)(m
r
(t) + m
i
(t)) cos(
IF
)t
Q = D +E = (1 a
1
a
2
)(m
r
(t) m
i
(t)) sin(
IF
)t
Phase Error Analysis
! The phase error impacts the I/Q channels in the same way,
and as long as the phase errors are small, it has a minimal
impact on the gain of the I/Q channels.
EECS 242
RF = m
r
(t) cos(
LO1
+
LO2
+
IF
)t +m
i
(t) cos(
LO1
+
LO2

IF
)t+
A = LPF{RF cos(
LO1
t +
1
)} =
1
2

m
r
(t) cos(
LO2
+
IF
+
1
)t+
m
i
(t) cos(
LO2

IF
+
1
)t
B = LPF{RF sin(
LO1
t
1
)} =
1
2

m
r
(t) sin(
LO2
+
IF

1
)t
m
i
(t) sin(
LO2

IF
+
1
)t
C = LPF{Acos(
LO2
t +
2
)} =
1
2

m
r
(t) cos(
IF
+
1
+
2
)t+
m
i
(t) cos(
IF
+
1
+
2
)t
D = LPF{Asin(
LO2
t
2
)} =
1
2

m
r
(t) sin(
IF

1

2
)t+
m
i
(t) sin(
IF

1

2
)t
E = LPF{B cos(
LO2
t +
2
)} =
1
2

m
r
(t) sin(
IF
+
1
+
2
)t+
m
i
(t) sin(
IF
+
1
+
2
)t
F = LPF{B sin(
LO2
t
2
)} =
1
2

m
r
(t) cos(
IF

1

2
)t+
m
i
(t) cos(
IF

1

2
)t
Q = D +E = (m
r
(t) m
i
(t)) cos(
1
+
2
) sin(
IF
)t
Double-Quad Low-IF
! Essentially a complex mixer topology. Mix RF I/Q with
LO I/Q to form baseband I/Q
! Improved image rejection due to desensitization to
quadrature gain and phase error.
EECS 242

+
+
+

+
+
I
Q
A
B
C
D
LNA
References
! Fundamental of Digital Communication, U. Madhow, Cambridge 2008
! O. Shanaa, EECS 290C Course Notes, 2005.
! A.A. Abidi, Radio frequency integrated circuits for portable
communications, Proc. of CICC, pp. 151-158, May 1994.
! A.A. Abidi, Direct conversion radio transceivers for digital communications,
Proc. of ISSCC, pp. 186-187, Feb. 1995.
! J. Crols and M. Steyaert, A single-chip 900MHz receiver front-end with high
performance low-IF topology, IEEE JSSC, vol. 30, no. 12, pp. 1483-1492,
Dec. 1995.
! B. Razavi, RF Microelectronics, Prentice Halls, 1998.
! J. Crols, M. Steyaert, CMOS Wireless Transceiver Design, Kluwer Academic
Publishers, 1997.
! Practical RF System Design, Willian F. Egdan, Wiley-IEEE Press, 2003.
EECS 242