0% found this document useful (0 votes)

65 views

Complex Modulation Basics

Fundamentals of Complex Modulation

Uploaded by

Kashif Virk

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

65 views

Complex Modulation Basics

Fundamentals of Complex Modulation

Uploaded by

Kashif Virk

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 69

Advanced I/Q Signal Processing for

Communication Systems
Mikko Valkama and Markku Renfors

Tampere University of Technology

Institute of Communications Engineering
P.O.Box 553, FIN-33101 Tampere, FINLAND
Tel: +358-40-8490-756, Fax: +358-3-3115-3808
E-mail: valkama@cs.tut.fi and mr@cs.tut.fi

SDR-03 Technical Conference, Nov. 2003, Orlando, FL

CONTENTS

1. Complex Signals and Systems

– Basic concepts and definitions
– Analytic signals and Hilbert transforms
– Frequency translations and mixing
– Complex signals and sampling

2. Sampling and Multirate DSP with Bandpass and I/Q signals

– Sampling of bandpass signals
– Multirate processing of bandpass and I/Q signals
– Efficient polyphase structures

3. I/Q Mismatch Problems in Analog I/Q Signal Processing

– I/Q signal processing in receivers
– Signal models for I/Q imbalance
– Imbalance effects in receiver front-ends

Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved
SDR-03 Technical Conference, Nov. 2003, Orlando, FL
4. Adaptive DSP for I/Q Imbalance Compensation
– Blind signal estimation based imbalance compensation
– Simulation examples

5. Second-Order Sampling and its Enhancements

– Basic second-order sampling scheme
– Imbalance problem and image rejection
– Enhancing the image rejection using DSP

6. I/Q Signal Processing in Frequency Synthesizers

– Digitally synthesized complex tone and analog I/Q mixing
– I/Q imbalance problem revisited

Summary
References

SDR-03 Technical Conference, Nov. 2003, Orlando, FL

Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved
1. COMPLEX SIGNALS AND SYSTEMS
Background and Motivation
– All physical signals and waveforms are real-valued
→ so why bother to consider complex-valued signals and systems ?
– The original complex signal concepts can be traced back to the introduction of
lowpass equivalent notation, i.e., analysis of bandpass signals and systems
using their lowpass/baseband equivalents
→ in general, a real-valued bandpass signal/system has a complex-valued
lowpass equivalent
• for example, linear I/Q modulation and demodulation principles are based on
these ideas
• also all advanced frequency translation techniques and thus the related
receiver architectures (low-IF, direct-conversion, etc.) utilize complex signals
• sampling and efficient multirate processing of bandpass signals is another
good example

1 (131)

Basic Concepts and Definitions

– By definition, the time domain waveform or sequence x(t) of a complex signal is
complex-valued, i.e., x(t) = xI(t) + jxQ(t).
– In practice, this is nothing more than a pair of two real-valued signals xI(t) and
xQ(t) carrying the real and imaginary parts.
– Similarly, a complex system is defined as a system with complex-valued
impulse response.
– In the frequency domain, real-valued signals have always symmetric amplitude
spectrum
→ complex signals don’t (need to) have any symmetry properties in general
→ the spectral support (region of non-zero amplitude spectrum) can basically be
anything
– One basic operation related to complex quantities is complex-conjugation
* *
→ if the spectrum of x(t) is denoted by X(f), then the spectrum of x (t) is X (−f)
→ this simple-looking result is surprisingly useful when interpreting some
properties of complex signals in the continuation
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

2 (131)
→ an immediate consequence is that if you consider the real part of x(t), i.e., y(t)
= Re[x(t)] = (x(t) + x*(t))/2, its spectrum is Y(f) = (X(f) + X *(−f))/2
*
• if X(f ) and X (−f ) are not overlapping, y(t) = Re[x(t)] contains all the
information about x(t)
• this result will find good use, e.g., in understanding frequency translations
– Another key operation related to linear systems in general is convolution.
– In the general complex case, this can be written as
x (t ) ∗ h(t ) = ( xI (t ) + jxQ (t )) ∗ (hI (t ) + jhQ (t ))
= xI (t ) ∗ hI (t ) − xQ (t ) ∗ hQ (t ) + j ( xI (t ) ∗ hQ (t ) + xQ (t ) ∗ hI (t ))
– In other words, 4 real convolutions are needed in general.
– Obvious simplifications occur if either the filter input or the filter itself is real
valued
→ in these cases, only two real convolutions need to be calculated

3 (131)

Analytic Signals and Hilbert Transforms

– Hilbert transformer is generally defined as an allpass linear filter which shifts the
phase of its input signal by 90 degrees.
– The (anticausal) impulse and frequency responses can be formulated as
continuous-time discrete-time
1  0, n even
hHT (t ) = hHT (n) = 
πt  2 /(π n), n odd
− j, f ≥ 0 − j, 0 ≤ ω < π
H HT ( f ) =  H HT (e jω ) = 
+ j , f < 0 + j , −π ≤ ω < 0
– In practice this behaviour can be well approximated over a finite bandwidth.
– One fascinating property related to Hilbert filters/transformers is that they can be
used to construct signals with only positive frequency content.
– This kind of signals are generally termed analytic and they are always complex.
– The simplest example is to take a cosine wave Acos(ω1t) whose Hilbert transform
is Asin(ω1t) (just a 90 degree phase shift!)
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

4 (131)
→ these together when interpreted as I and Q components of a complex signal
result in Acos(ω1t) + jAsin(ω1t) = Aexp(jω1t) whose spectrum has an impulse
at ω1 but nothing on the other side of the spectrum
– The “elimination” of the negative frequencies can more generally be formulated
as follows.
– Starting from an arbitrary signal x(t) we form a complex signal x(t) + jxHT(t) where
xHT(t) denotes the Hilbert transform of x(t).
– Then the spectrum of the complex signal is X(f)[1 + jHHT(f)] where
continuous-time
1 + j × (− j ), f ≥0 2, f ≥0
1 + jH HT ( f ) =  =
 1 + j × j, f <0 0, f <0
which shows the elimination of the original negative frequency content. Similar
concepts carry on to discrete-time world and we can write
discrete-time
1 + j × (− j ), 0 ≤ ω < π  2, 0 ≤ ω < π
1 + jH HT (e jω ) =  =
 1 + j × j, −π ≤ ω < 0 0, −π ≤ ω < 0
5 (131)

– This idea of using Hilbert transform to generate analytic signals is further

illustrated graphically in the following figure.
– In practice the Hilbert filtering causes a delay and a corresponding delay needs
to be included also in the upper (I) branch.

input spectrum

I
f
input output
output spectrum
HT Q

6 (131)
– Design Example: Hilbert transformer of order 50, design bandwidth 0.1π … 0.9π
(π denotes half the sampling frequency), Remez design
→ this results in about 87 dB attenuation for the negative frequencies (wrt.
corresponding positive band)

Hilbert Transformer (HT)

0.5

−0.5

−1
−30 −20 −10 0 10 20 30
n

7 (131)

Hilbert Transformer (HT)

1.5
Amplitude

0.5

0
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Frequency ω / π
2
Phase wrt. π/2

−1

−2
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Frequency ω / π

8 (131)
Complex (1+j×HHT) Response
1.5

Amplitude
1

0.5

0
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Frequency ω / π

0
Amplitude [dB]

−50

−100

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

Frequency ω / π

9 (131)

Frequency Translations and Mixing

– One key operation in communications signal processing is the shifting of a signal
spectrum from one center-frequency to another
→ conversions between baseband and bandpass representations are special
cases of this
– The basis of all the frequency translations lies in multiplying a signal with a
complex exponential, generally referred to as complex or I/Q mixing.
– This will indeed cause a pure frequency shift, i.e.,
y (t ) = x(t )e jωLOt ⇔ Y ( f ) = X ( f − f LO )
which forms the basis for all the linear modulations.
– This is illustrated in frequency domain below.
input spectrum output spectrum

f f
fCReserved
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights

10 (131)
– In general, four real mixers are needed to implement a complex mixer as
x(t )e jωLOt = ( xI (t ) + jxQ (t ))(cos(ωLO t ) + j sin(ωLO t ))
= xI (t )cos(ωLOt ) − xQ (t )sin(ωLOt ) + j ( xQ (t )cos(ωLO t ) + xI (t )sin(ωLO t ))
→ in the special case of a real input, only two mixers needed
– Real mixing is obviously a special case of the previous complex one and results
in two frequency translations:
y (t ) = x(t )cos(ωLOt )
1 1 1
= x(t ) (e jωLOt + e− jωLOt ) ⇔ Y ( f ) = X ( f − f LO ) + X ( f + f LO )
2 2 2
– Here, the original spectral component appear twice in the mixer output, the two
replicas being separated by 2fLO in frequency.
– In receivers, this results in the so called image signal problem since the signals
from both fC + fLO and fC − fLO will appear at fC after the real mixing stage
→ if real mixing is used, the image signal needs to be attenuated before the
actual mixer stage
→ we’ll talk about this in more detail in the receiver architecture section
11 (131)

input spectrum

f
fC − fLO fC fC + fLO

real mixing

output spectrum

f
fC − 2fLO fC − fLO fC fC + fLO fC + 2fLO

12 (131)
– Linear I/Q modulation methods are basically just a special case of complex
mixing.
– Given a complex message signal x(t) = xI(t) + jxQ(t), it is first modulated as
x(t)exp(jωCt), after which only the real part is actually transmitted:
1 1
y (t ) = Re[ x(t )e jωC t ] = xI (t )cos(ωC t ) − xQ (t )sin(ωC t ) = x(t )e jωC t + x* (t )e− jωCt
2 2
→ interpretation #1: xI(t) and xQ(t) are modulated onto two orthogonal (cosine
and sine) carriers; nice from the implementation point of view
→ interpretation #2: x(t) and x*(t) are modulated onto two complex exponentials
exp(jωCt) and exp(−jωCt); key in building general understanding and
recovering x(t) back from y(t)
– Notice that both terms/spectral components (at +fC and −fC) contain all the
original information (i.e., x(t)).
– This process, also termed lowpass-to-bandpass transformation, is pictured in the
figure below.

13 (131)

input spectrum output spectrum

f f
−fC fC

– I/Q demodulation: In the receiver, the goal is to recover the original message
x(t) from the modulated signal y(t).
– Based on the previous discussion, it’s easy to understand that either of the signal
components at +fC or −fC can be used for that purpose, while the other one
should be rejected.
– Since
1 1 1 1
y (t )e− jωC t = ( x(t )e jωC t + x* (t )e− jωC t )e− jωC t = x(t ) + x* (t )e− j 2ωC t
2 2 2 2
the message can be fully recovered by simply lowpass filtering the complex
receiver mixer output.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

14 (131)
– Formal block-diagrams for the modulator and demodulator in terms of complex
signals are presented below.

e jωC t
MODULATOR:

input Re[·] output

e − jωC t
DEMODULATOR:

input LPF output

15 (131)

Complex Signals and Sampling

– In periodic sampling (sample rate fS) the resulting discrete-time signal has a
periodic spectrum where the original continuous-time spectrum is replicated
around the integer multiples of the sampling frequency.
– Interestingly, any of these spectral replicas (or “images”) can be considered as
the useful part and thus be used for further processing.
– Consequently, sampling (and multirate operations in general) can also be used,
in addition to mixing techniques, in performing frequency translations.
– Let B denote the double-sided bandwidth of a complex-valued baseband signal
(i.e., the spectrum is nonzero only for − Bneg ≤ f ≤ B pos , B = Bneg + Bpos)
→ to avoid aliasing, the sampling frequency fS should simply be high enough
such that the spectral images don't overlap, i.e.,
f S − Bneg ≥ B pos ⇔ f S ≥ Bneg + B pos ⇔ fS ≥ B
→ this is the traditional Nyquist sampling theorem
→ naturally, since the signal to be sampled is complex-valued, there exist two
real-valued sample streams (I and Q) both at rate fS
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

16 (131)
→ if the signal to be sampled consists of multiple frequency channels, sampling
rates below f S = B are possible if only some of the channels are of interest
• the sampling frequency should simply be selected in such a manner that
aliasing is avoided on top of those interesting frequency bands
– Example spectra which both have the same lower limit f S = B for the sampling
frequency are depicted below.

original sampled

B B

f f
−2fS −fS fS 2fS

B B

f f
−2fS −fS fS 2fS

17 (131)

– Conclusion: It doesn’t matter whether the signal is real or complex or whether it

is located “symmetrically” with respect to origin
→ the minimum sampling rate is always f S = B

– So in general the traditional statement “signal should be sampled at least at rate

two times its highest frequency component” can be concluded inaccurate
→ what really matters is the double-sided bandwidth
→ a good example is the sampling of a real-valued lowpass signal, say x(t), with
spectral support –W … W
• when sampled directly, the minimum sampling rate is fS = 2W
• as an alternative, you can form an analytic signal x(t) + jxHT(t), where xHT(t)
denotes the Hilbert transform of x(t), for which the minimum sampling rate is
only fS = W (even though the highest frequency component present in both
signals is W)

18 (131)
2. SAMPLING AND MULTIRATE DSP WITH
BANDPASS AND I/Q SIGNALS
SAMPLING OF BANDPASS SIGNALS
– Starting point is the traditional Nyquist sampling theorem: Any signal occupying
the band –Bneg … Bpos [Hz] is completely characterized by its discrete-time
samples given that the sampling rate is at least Bneg + Bpos (two-sided bandwidth).
– People commonly interpret this that if the highest frequency component in a
signal is fMAX, you need to take at least 2fMAX samples per second
→ strictly speaking, this is inaccurate (as concluded before)
→ i.e., sampling at or above rate 2fMAX is clearly always sufficient but e.g. in
case of bandpass signals we can also use (usually much) lower sample rate
→ more specifically, sampling at rate below 2fMAX will indeed result in aliasing
but as long as all the information about the original signal is present in the
samples, we are doing good
• keep in mind also that the Nyquist (“accessible”) band for any sample rate fS is
–fS /2 … fS /2, so with below 2fMAX sampling rates it is really one of the images
that appear on this band !!!
19 (131)

→ these kind of techniques are generally referred to as subsampling

– The one and only principle to remember in sampling is that the resulting signal
has a periodic spectrum and any part of that spectrum can be selected/used for
further processing.
– More specifically, in communications receivers, aliasing due to sub-sampling
can be taken advantage of to bring the signal closer to baseband.
– We consider two cases; starting from a real-valued bandpass signal, the
resulting sample stream is either 1) real-valued or 2) complex-valued.

Real Subsampling
– Basic setup: real-valued bandpass signal, bandwidth B, center-frequency fC,
upper band-edge fU = fC + B/2 and lower band-edge fL = fC – B/2.

original
B

f
–fU –fC –fL fL fC fU
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

20 (131)
– Now sampling at some rate fS results in a signal where the previous spectrum is
replicated at integer multiples of the sampling rate (the basic effect of sampling).
– With fS < 2fU, aliasing will take place but as long as the aliasing components don’t
fall on top of each other, everything is OK !!
– So an example spectrum of the sampled signal could look like in the figure
below, when there is no harmful aliasing and yet fS < 2fU.

sampled
B

••• ••• ••• •••

f
–fU –fC –fL fL fC fU
–fL +nf S –fU +(n+1)f S

– Based on the above figure, it is easy to formulate the regions of allowable

sampling rates. These are in general of the form
2 fC + B 2f −B 2f −B
≤ fS ≤ C where 0 ≤ n ≤ floor ( C )
n +1 n 2B
21 (131)

– Comments:
→ as can be seen, the possible values of the sampling rate depend on both the
bandwidth B and the center-frequency fC
→ for n = 0 we get f S ≥ 2 fC + B = 2( f C + B / 2) = 2 fU which is the traditional
Nyquist sampling theorem (the upper limit becomes infinity)
→ for n > 0 we are really sampling at lower frequency than given by the
traditional Nyquist theorem
→ for n > 0 aliasing does occur but with given values of fS, not on top of the
desired signal band (no harmfull aliasing)
→ the lowest possible sampling rate is in general given by
2 fC + B 2 fC + B 2 fC + B 2 fC + B
fS ≥ = = =
nmax + 1 floor ( 2 fC − B ) + 1 floor ( 2 fC − B + 1) floor ( 2 fC + B )
2B 2B 2B
2 fC + B
• the “ultimate” sampling rate fS = 2B is utilizable iff is an integer (then
2B
2 fC + B 2 fC + B
and only then floor ( )= )
2B 2B
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

22 (131)
– Numerical example: fC = 20 kHz and B = 10 kHz, so
40 − 10
→ 0 ≤ n ≤ floor ( ) = floor (1.5) = 1 and the possible values for fS are
20
→ n = 0: 50 kHz ≤ f S ≤ ∞
→ n = 1: 25 kHz ≤ f S ≤ 30 kHz
• try e.g. with fS = 27 kHz and you see that no harmfull aliasing occurs

Complex subsampling
– Instead of sampling directly the real-valued signal, the idea is to sample the
corresponding analytic signal !!!
– So the sampling structure looks like (HT denotes Hilbert transformer)

fS I

input

HT Q
fS

23 (131)

– Now since the analytic signal is free from negative frequency components,
sampling frequency of fS = B (or any rate above) is always (independently of the
center-frequency fC !) sufficient to avoid harmfull aliasing !!!
– Some example spectral figures with the same input signal as in the previous
subsection:

fS = B: sampled
B

••• ••• ••• •••

f
fL fC fU

fS > B: sampled
B

••• ••• ••• •••

f
fL fC fU
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

24 (131)
– Notice: If the center-frequency fC is an integer multiple of the sample rate fS (i.e.,
fS = fC / k), the center-frequency of the k-th spectral replica will coincide with zero
frequency and a direct bandpass-to-lowpass transformation is obtained !!!
→ this is easy to understand based on spectral interpretations but can also be
seen as follows:
• the real bandpass input, say r(t), can be written in terms of its baseband
equivalent z(t) as
1 1
r (t ) = Re[ z (t )e jωC t ] = z (t )e jωC t + z * (t )e − jωC t
2 2
• then the corresponding analytic signal is of the form
1 1 1 1
r (t ) + jrHT (t ) = z (t )e jωC t + z * (t )e − jωC t + j ( − j z (t )e jωC t + j z * (t )e − jωC t )
2 2 2 2
jωC t
= z (t )e
• thus sampling at fS = fC / k (with k integer) results in

r ( nTS ) + jrHT (nTS ) = z (nTS )e jωC nTS = z ( nTS )e j 2π fC nTS = z ( nTS )e j 2π nk = z (nTS )
which are indeed just the samples of the baseband equivalent

25 (131)

MULTIRATE PROCESSING OF BANDPASS AND I/Q SIGNALS

– The two fundamental multirate operations (decimation/down-sampling and
interpolation/up-sampling) enable the sample rate to be altered digitally.
– In addition to this, they also offer an interesting alternative to mixing in
performing frequency translations.
Decimation or Down-Sampling with Complex Signals
– The basic block-diagram to reduce the sample rate by an integer factor L is
presented in the figure below, with fS denoting the original sample rate.
– In the down-sampling (↓ L), every L-th sample of the input sequence is picked up
to form the output sequence and the new sample rate becomes fS /L.
– As a consequence, all the frequency bands located at the integer multiples of
fS /L (within –fS /2 … fS /2 of course) are aliased down to baseband.
– These are also illustrated in the following figure.
– Thus, if the (generally complex) input signal is of lowpass type (passband
–Bneg … Bpos), the decimation filter H(z) should attenuate all the frequency bands
of width B=Bneg +Bpos located correspondingly at the previous critical frequencies.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

26 (131)
– Naturally, the bandwidth B=Bneg +Bpos needs to be smaller than fS /L
→ but there’s no restriction such that, e.g., Bpos should be smaller than (fS /L)/2
→ as long as fS /L ≥ B, the down-sampled signal is free from harmful aliasing
– On the other hand, if the desired signal is originally a bandpass signal (with a
generally complex-valued baseband equivalent), the inherent aliasing can be
exploited to change the signal center-frequency.
– Now, the decimation filter H(z) is a bandpass filter selecting the desired
frequency band, and aliasing can be used to bring the signal closer to baseband
→ as a special case of this, if the signal is centered at any multiple of the output
sample frequency fS /L, an analytic bandpass filter and decimation will result
in a direct bandpass-to-lowpass transformation
→ this basically represents a digital equivalent of the complex (I/Q) subsampling
scheme of the previous section
– Two example cases follow in the figures below.

27 (131)

H(z) ↓L
fS fS fS /L

before down-sampling: B

••• •••

f
−fS /2 −2fS /L −fS /L fS /L 2fS /L fS /2

after down-sampling:

••• •••

f
−2fS /L −fS /L B fS /L 2fS /L
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

28 (131)
H(z) ↓L
fS fS fS /L

before down-sampling: B

••• •••

f
−fS /2 −2fS /L −fS /L fS /L 2fS /L fS /2

after down-sampling:

••• •••

f
−2fS /L −fS /L B fS /L 2fS /L

29 (131)

Interpolation or Up-Sampling with Complex Signals

– The basic block-diagram to increase the sample rate by an integer factor L is
presented in the figure below, with fS again denoting the original sample rate.
– The output sequence of the up-sampler (↑ L) is formed by adding L – 1 zeros
between each original input sample.
– As a consequence, all the spectral images (within –LfS /2 … LfS /2) of the input
spectrum appear now at the output signal at the multiples of the input sample
rate fS.
– These are illustrated in the following figure.
– Traditionally, the interpolation filter H(z) attenuates these extra images retaining
only the original spectral component.
– This is, however, not the only possibility. More precisely, any of the spectral
images within –LfS /2 … LfS /2 can be considered to be the desired one.
– As an example, the original lowpass signal can be transformed into a bandpass
signal by simply using a proper bandpass filter as the interpolation filter H(z).
This filter now retains the spectral image at the desired center-frequency and
attenuates
Proceeding of the others.
the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

30 (131)
↑L H(z)
fS LfS LfS

••• •••

f
−LfS /2 −2fS −fS fS 2fS LfS /2

••• •••

f
−LfS /2 −2fS −fS fS 2fS LfS /2
31 (131)

EFFICIENT POLYPHASE STRUCTURES

– Polyphase filtering represents one interesting approach to implement decimation
or interpolation in a flexible yet computationally efficient way.
– In the traditional approaches, the decimation/interpolation filters operate at the
higher sampling rate, i.e., either before down-sampling or after up-sampling.
– Given that the down-/up-sampling ratio is L, the idea in the polyphase structures
is to split the related filtering into L parallel stages operating at the lower
sampling rate
→ in applications requiring very high operation speeds, this can be a crucial
benefit
– Furthermore, the polyphase structures are quite flexible if used, e.g., in
channelization applications.
– These aspects will be explained in more detail in the following.

32 (131)
Polyphase Decomposition of FIR filters
– The so called polyphase decomposition of a finite length filter H(z) is usually
formulated as
L −1
H ( z ) = ∑ z −i H i ( z L )
i =0

– To put it in words, the output of any FIR filter in general can be constructed as a
sum of the outputs of the filters H0(zL), H1(zL), …, HL–1(zL) whose input signals are
delayed by z0, z –1, …, z –(L – 1).
– Given that the impulse response of H(z) is h(n), the impulse responses of H0(z),
H1(z), …, HL–1(z) are simply
h0 (n) = {h(0), h( L), ...}
h1 (n) = {h(1), h( L + 1), ...}
!
hL−1 (n) = {h( L − 1), h(2 L − 1), ...}

33 (131)

– Example: Two-phase decomposition of H(z)=h(0)+h(1)z–1 +…+h(14)z–14 is

H ( z ) = h(0) + h(2) z −2 + h(4) z −4 + h(6) z −6 + h(8) z −8 + h(10) z −10 + h(12) z −12 + h(14) z −14
+ z −1 (h(1) + h(3) z −2 + h(5) z −4 + h(7) z −6 + h(9) z −8 + h(11) z −10 + h(13) z −12 )
= H 0 ( z 2 ) + z −1 H1 ( z 2 )
where

h0(n)={h(0), h(2), h(4), h(6), h(8), h(10), h(12), h(14)}

h1(n)={h(1), h(3), h(5), h(7), h(9), h(11), h(13)} =

H0(z2) +

z –1 H1(z2)

34 (131)
Basic Polyphase Decimators and Interpolators
– In decimation applications, the down-sampler which traditionally operates on the
filter output can now be transferred to the front of the branch filters given that the
L time delays in the filters H0(zL), …, HL–1(zL) are replaced by ordinary unit delays
→ this operation is intuitively clear and theoretically justified by the famous
noble identity of multirate signal processing
– So in the i-th branch, the input signal is delayed by z –i, down-sampled by L, and
filtered using Hi(z).
– Finally, the outputs of the L branches are summed to form the final decimated
output signal.
– Notice that all the branch filters operate at the lower sampling rate !
– This is illustrated in the following using a simple example.

35 (131)

– Example (cont’d): Down-sampling by L=2.

H0(z2) ↓2

z –1 H1(z2)

↓2 H0(z)

z –1 ↓2 H1(z)

– Since in general the delay z –i is different in every branch and the down-samplers
operate synchronously, the low-rate data sequences entering the branch filters
are all disjoint.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

36 (131)
– Thus, the front-end delays and down-samplers can actually be discarded by
simply feeding every L-th (with a proper time-shift) sample to the polyphase
branches using a commutative switch.
– This final structure is depicted in the figure below.

H0(z)

H1(z)
+
fS
... fS /L

H(z) L
fS fS fS /L HL
1(z)

(a) (b)
Figure. (a) The basic decimation (down-sampling) scheme. (b) The corresponding polyphase
implementation.

37 (131)

– Similar reasoning can be used also in the interpolation case.

– The traditional approach based on zero-padding and filtering maps into a
polyphase structure where the input signal to be interpolated is fed directly into
the branch filters H0(z), H1(z), …, HL–1(z).
– The outputs of these filters are then up-sampled and delayed, the delay being z –i
in the i-th branch, and finally summed to form the interpolator output.
– Due to zero-packing and different branch delays, only one branch filter output
actually contributes to the final output (with other outputs being zero) at any
given time instant.
– Thus, as in the decimation case, the final structure can be implemented by
simply multiplexing the branch filter outputs using a commutator.
– This approach is pictured below.

38 (131)
H0(z)

H1(z)

fS
... LfS

L H(z)
fS LfS LfS HL
1(z)

(a) (b)
Figure. (a) The basic interpolation (up-sampling) scheme. (b) The corresponding polyphase
implementation.

39 (131)

– One interesting interpretation of the previous polyphase structures is related to

aliasing and the amplitude and phase characteristics of the polyphase filters.
– As an example, consider the decimation case. Since the input signal is directly
down-sampled in each polyphase branch, all the bands located at the integer
multiples of the final output sampling rate alias to baseband.
– However, due to the relative delays of the different branches as well as different
phase characteristics of the polyphase filters, only the information within the
passband of the prototype filter H(z) will appear in the final decimator output.
– To be more specific, all the polyphase branch filters are actually allpass filters
whose amplitude response is ideally constant.
– Furthermore, the phase delays of the different filters differ by 1/L with H0(z)
having the largest delay.
– This is easy to see even intuitively when considering how the polyphase filters
are obtained from the “prototype” H(z).

40 (131)
Bandpass Polyphase Structures
– The previous discussion is basically valid for both lowpass and bandpass
decimator/interpolator structures.
– The only difference is, of course, related to the characteristics and design of the
filter H(z).
– A straight-forward way to handle the lowpass and bandpass cases is to consider
them separately.
– However, the full flexibility of the polyphase structures can only be capitalized by
treating them together.
– To illustrate the basic idea, assume the filter H(z) is designed for a specific
lowpass decimation scenario (passband –Bneg … Bpos in general).
– Now, suppose we wish to change the structure to process a bandpass signal
located around a center-frequency fC which is an integer multiple of the output
rate fS /L.
– In terms of the normalized frequency variable ω = 2πf/fS, this means that ωC is an
integer multiple of 2π/L.

41 (131)

– In general, an analytic filter G(z) to extract the interesting band can be obtained
simply by frequency translating the corresponding lowpass prototype filter H(z)
as g(n) = h(n)exp{jωCn}.
– The impulse responses of the corresponding polyphase implementation of g(n)
are then obtained as presented before, i.e., g0(n)={g(0), g(L), …}, g1(n)={g(1),
g(L+1), …}, etc.
– However, after some manipulations, the polyphase impulse responses for the
analytic filter G(z) can simply be written as
g 0 (n) = h0 (n)
g1 (n) = h1 (n) × exp{ jk (2π ) / L}
g 2 (n) = h2 (n) × exp{ jk (2π )2 / L}
!
g L −1 (n) = hL −1 (n) × exp{ jk (2π )( L − 1) / L}
where the integer k is the “channel index”, i.e., ωC = k(2π/L).
=> the same polyphase representation as in the lowpass case except for the
constant complex multipliers exp(jk(2π)i/L) !!!
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

42 (131)
– Example: Bandpass down-sampling by L=4 with ωC = π/2 (i.e., fC = fS / 4),
g(n) = h(n)exp(j(π/2)n)
π π π
j 0 j 4 j 8
g 0 ( n) = {h(0)e 2 , h ( 4) e 2 , h(8)e 2 , ...} = {h(0), h(4), h(8), ...}
π π π π
j 1 j 5 j 9 j
g1 (n) = {h(1)e 2 , h(5)e 2 , h(9)e 2 , ...} = {h(1), h(5), h(9), ...}e 2

π π π
j 2 j 6 j 10
g 2 ( n) = {h(2)e 2 , h ( 6) e 2 , h(10)e 2 , ...} = {h(2), h(6), h(10), ...}e jπ
π π π 3π
j 3 j 7 j 11 j
g 3 ( n) = {h(3)e 2 , h (7 ) e 2 , h(11)e 2 , ...} = {h(3), h(7), h(11), ...}e 2

– So as stated before, the modulating exponentials in the polyphase filters reduce

to constant multipliers wi,k = exp{jk(2π)i/L}, i = 1, 2, …, L–1.
– This effect, in turn, can be implemented by simply multiplying the outputs of the
lowpass polyphase filters by the constants wi,k !!!
– As a consequence, the same polyphase front-end (the same filters !!!) can be
used to extract any channel located at the multiple of the output rate.
– This general polyphase decimator structure is illustrated in the figure below.
43 (131)

H0(z)
w1, k

H1(z)
+

+
fS
... wL
1, k
fS /L

HL
1(z)
+

Figure. General polyphase decimator.

– Similar type reasoning can again be used also in the interpolation (up-sampling)
case.

44 (131)
Matlab Demo
– The previous ideas of bandpass polyphase filtering and decimation are illustrated
using Matlab.
– As an example case, we have three channels
→ #1: ωC1 = 0.25π, 8PSK modulated signal, 16 samples per symbol
→ #2: ωC2 = 0.50π, 4PSK modulated signal, 16 samples per symbol
→ #3: ωC3 = 0.75π, 16QAM modulated signal, 16 samples per symbol
– In all the cases, raised-cosine pulse-shape with 35% roll-off is used.
– Since the channels are centered at integer multiples of π/4 (i.e., fS /8), we use a
polyphase structure with L=8
→ after downsampling and polyphase filtering, a further decimation by 2 is
included to get symbol rate samples (in order to plot the output constellations)
– The simulation setup and the results are illustrated in the following figures.

45 (131)

H0(z)
2π k=1:
exp( j k ⋅1)
8
f H1(z)
exp( j
2π
k ⋅ 2) ↓2
8 k=2:
H2(z)

•••
2π k=3:
exp( j k ⋅ 7)
8
H7(z)

Figure. Demonstration set-up, L = 8.

46 (131)
Real bandpass signal and the analytic channel filter, k = 1

0.8
Amplitude

0.6

0.4

0.2

0
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
Frequency ω / π

47 (131)

0.2 0.2
h (n)

h (n)

0 0
0

−0.2 −0.2
0 5 10 0 5 10
0.2 0.1
h (n)

h (n)

0 0
2

−0.2 −0.1
0 5 10 0 5 10
0.1 0.1
h (n)

h (n)

0 0
4

−0.1 −0.1
0 5 10 0 5 10
0.2 0.2
h (n)

h (n)

0 0
6

−0.2 −0.2
0 5 10 0 5 10
n n

48 (131)
−10 −10

|H0| [dB]

|H1| [dB]
−20 −20

−30 −30
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
−10 ω/π −10 ω/π
|H2| [dB]

|H3| [dB]
−20 −20

−30 −30
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
−10 ω/π −10 ω/π
|H4| [dB]

|H5| [dB]
−20 −20

−30 −30
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
−10 ω/π −10 ω/π
|H6| [dB]

|H7| [dB]
−20 −20

−30 −30
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ω/π ω/π

49 (131)

Phase delays of the polyphase filters

6.8
−arg(Hi) / ω

6.6

6.4

6.2

5.8
0 0.2 0.4 0.6 0.8 1
ω/π

50 (131)
Symbol rate output samples, channel #1
1.5

0.5
IM

−0.5

−1

−1.5
−1.5 −1 −0.5 0 0.5 1 1.5
RE

51 (131)

Real bandpass signal and the analytic channel filter, k = 2

0.8
Amplitude

0.6

0.4

0.2

0
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
Frequency ω / π

52 (131)
Symbol rate output samples, channel #2
0.8

0.6

0.4

0.2
IM

−0.2

−0.4

−0.6

−0.8
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
RE

53 (131)

Real bandpass signal and the analytic channel filter, k = 3

0.8
Amplitude

0.6

0.4

0.2

0
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
Frequency ω / π

54 (131)
Symbol rate output samples, channel #3
4

1
IM

−1

−2

−3

−4
−4 −3 −2 −1 0 1 2 3 4
RE

55 (131)

– Notice that even though analytic channel filters are shown in the three figures,
the actual implementation is really based on the polyphase concepts.
– In other words, there is no single filter with the given response but the polyphase
decimator effectively implements that kind of function.
– Remember also that the branch filters for each channel (k = 1, 2, 3) are identical,
only the “scaling” coefficients wi,k after the filters depend on the channel index.
– Notice also that the structure works similarly for complex input signals as
well (in the previous example, the input is real-valued).

56 (131)
3. I/Q MISMATCH PROBLEMS IN ANALOG
I/Q SIGNAL PROCESSING
I/Q Signal Processing in Receivers
– In communication receivers, one of the key front-end functionalities is to down-
convert the desired channel signal from RF closer to baseband, in the
presence of other channels/signals.
– In this context, the fundamental problem of image signal attenuation is a
major concern.
– Traditionally, in superheterodyne (and its variants) receivers, the image band is
attenuated using RF filtering before the down-conversion stage
→ the basic image signal problem is illustrated in the following figure where the
target is to translate the desired channel signal to an intermediate frequency
(IF) fIF
• the desired channel is illustrated in grey and it’s image in dark, separated in
frequency by 2fIF

57 (131)

2fIF 2fIF

f
−fLO fLO

f f
−fIF fIF −fIF fIF

– The lower part illustrates the signal after down-conversion and lowpass filtering
without (left) and with (right) RF image rejection filtering.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

58 (131)
– In practice a tradeoff is needed in the selection of the IF:
→ the higher the IF, the easier it is to implement the RF image rejection filter
(since the separation of the desired and image bands is 2fIF)
→ on the other hand, the lower the IF, the easier it is to implement the channel
selectivity filtering
– In order to reduce the needed RF image rejection filtering, complex (I/Q)
down-conversion can be used instead of real mixing
→ theoretically a pure frequency translation and image problems are avoided
during the frequency shift
→ this idea can be used in the receiver front-end simply for down-conversion
purposes, independently of the desired channel modulation
– A generalized receiver based on this idea appears in figure below.
– The 90°° phase shift can basically be introduced either between the local
oscillator (LO) signals (point A) or between the input signal branches (point B)
→ point A: cos(.) and –sin(.) LO signals
→ point B: a wideband Hilbert transformer (and two in-phase LOs)

59 (131)

AGC
I
LPF A/D

RF LNA

LO DSP

A AGC
B Q
LPF A/D

– In theory: the resulting I and Q channels should have equal amplitudes and a
phase difference of 90°
→ infinite attenuation for the image signal band
– In practice: real-world analog components, such as the mixers, LPFs, etc., can
never be perfectly matched
→ some imbalance will always exist
→ the image attenuation is finite (formal proof will be given later)
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

60 (131)
– 1-2° and 1-2% imbalance values are realistic, resulting in 20- 40 dB image
attenuation.
– To be more precise, let’s start to look at the following situation in more detail:
→ wideband front-end, no image rejection filtering (idealized case)
→ multichannel (see the following figure) bandpass signal (bandwidth B)
centered at fLO
→ target: produce a wideband baseband equivalent of the received signal
(wideband downconversion)
→ how: using I/Q signal processing
– Obtain a formal characterization of the imbalance effects due to the
mismatches of practical analog electronics
→ first some basic results are given in the case when the amplitudes and/or
phases of the two components (I and Q) of a complex signal become
mismatched
→ these results are then applied to the wideband receiver case

61 (131)

R(f)
B

f
−fLO fLO

Z(f) Z'(f)

?
f f
−fIF fIF −fIF fIF
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

62 (131)
Narrowband Imbalance Model
– Consider an ideal single frequency signal z (t ) = e jω 0t = cos(ω 0t ) + j sin(ω 0t ) .
– Here, we assume that the mismatched I and Q components have
→ relative amplitudes g1 and g 2
→ relative phases φ1 and φ 2
– In other words, we write the imbalanced signal z ′(t ) as
z ′(t ) = g1 cos(ω 0t + φ1 ) + jg 2 sin(ω 0t + φ 2 )

e jx + e − jx e jx − e − jx
– Recap: Euler’s formulas cos( x) = and sin( x) = .
2 2j
– Then, using the Euler’s formulas, z ′(t ) can be written as

 g1e jφ1 + g 2 e jφ2  jω 0t  g1e − jφ1 − g 2 e − jφ2  − jω 0t

z ′(t ) =  e +  e
 2   2 

63 (131)

– So, in addition to the original component e jω 0t , z ′(t ) consists also of the mirror
frequency component e − jω 0t .
– Considering the relative strengths of the two frequency components, z ′(t )
can also be written as
 g j (φ −φ )   g 2 − j (φ2 −φ1 ) 
1 + 2 e 2 1  1 − e 
g g
z ′(t ) =  1 g e 1e 0 + 
1
j φ j ω t 1  g e − jφ1 e − jω 0t
 2   2  1
   
   
 g j (φ −φ )   g 2 − j (φ2 −φ1 ) 
1 + 2 e 2 1  1 − e 
 g1  j (ω t +φ )  g1  g e − j (ω 0t +φ1 )
= g1e 0 1
+
 2   2  1
   
   
– So we notice that the relative strengths of the two frequency components depend
only on the relative amplitude and phase imbalances
→ g 2 / g1 and φ 2 − φ1
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

64 (131)
– Therefore, to simplify the notations, we assume
→ g1 = 1 , g 2 = g and
→ φ1 = 0 , φ 2 = φ
– Then, the model can be written in a more simple form as
 1 + ge jφ   1 − ge − jφ  − jω 0 t
′ 
z (t ) =  e jω 0t +  e
 2   2 
Wideband Imbalance Model
– Theorem: Consider a complex-valued signal z (t ) = z I (t ) + jzQ (t ) and its complex
conjugate z * (t ) = z I (t ) − jzQ (t ) whose Fourier transforms are given by
→ z (t ) ⇔ Z ( f ) = Z I ( f ) + jZ Q ( f )
→ z * (t ) ⇔ Z * (− f ) = Z I* (− f ) − jZ Q* (− f ) = Z I ( f ) − jZ Q ( f )

Then, the Fourier transform of Z ′( f ) = H I ( f ) Z I ( f ) + jH Q ( f ) Z Q ( f ) can be

written as

65 (131)

 H I ( f ) + HQ ( f )   H ( f ) − HQ ( f )  *
Z ′( f ) =   Z ( f ) +  I  Z (− f )
 2   2 
– Proof: Direct substitution (next page) will yield

 H I ( f ) + HQ ( f )   H I ( f ) − HQ ( f )  *
Z ′( f ) =   Z ( f ) +   Z (− f )
 2   2 
 H I ( f ) + HQ ( f )   H ( f ) − HQ ( f ) 
=  (Z I ( f ) + jZ Q ( f ) ) +  I (Z I ( f ) − jZ Q ( f ) )
 2   2 
 H I ( f ) Z I ( f ) + jH I ( f ) Z Q ( f ) + H Q ( f ) Z I ( f ) + jH Q ( f ) Z Q ( f ) 
=  
 2 
 H I ( f ) Z I ( f ) − jH I ( f ) Z Q ( f ) − H Q ( f ) Z I ( f ) + jH Q ( f ) Z Q ( f ) 
+  
 2 
= H I ( f ) Z I ( f ) + jH Q ( f ) Z Q ( f ) q.e.d.

66 (131)
– As can be observed, the mismatches cause a signal component relative to
z * (t ) to appear in addition to the original signal component z (t ) !
– Again, only the difference between H I ( f ) and H Q ( f ) contribute to the relative
strength of the mirror component Z * (− f ) (i.e., z * (t ) ).
– Clearly, a generalization of the narrowband case.

Imbalance Effects in Receiver Front-Ends

– In general, all the analog components such as the
→ quadrature mixing stage
→ branch filters
→ A/D converters
affecting the I and Q branch signals contribute to the effective amplitude and
phase mismatches.
– Motivated by this, the model of the following figure is used hereafter.

67 (131)

xLO,I (t) = cos(2π fLO t)

xI (t) zI' (t)
HNOM ( f ) HI ( f )

r(t)

xQ (t) zQ' (t)

HNOM ( f ) HQ ( f )

xLO,Q (t) = –gsin(2π fLO t + φ)

– The effect of quadrature demodulator: the local oscillator signal xLO (t) of an
imbalanced quadrature demodulator is here modelled as
xLO (t ) = cos(2π f LO t ) − jg sin(2π f LOt + φ )
= K1e− j 2π f LOt + K 2 e j 2π f LOt
where g and φ represent the demodulator amplitude and phase imbalances,
respectively.
– The mismatch coefficients K1 and K2 are given by
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

68 (131)
ideally
K1 = [1 + ge − jφ ]/ 2 = 1,
ideally
K 2 = [1 − ge jφ ]/ 2 = 0 .

– For more details, see the narrowband case in the beginning of the material (use
g ← –g).
– The effect of branch components: the branch component mismatches can be
easily modelled as imbalanced lowpass filters (LPF) as given by
H LPF , I ( f ) = H NOM ( f ) H I ( f )
H LPF ,Q ( f ) = H NOM ( f ) H Q ( f )

– HNOM (f) is the nominal LPF response rejecting the high-frequency components.
– HI (f) and HQ (f) represent the actual mismatch effects due to branch filters,
AGCs, A/Ds, etc.
→ with perfect matching (ideal case), HI (f) = HQ (f)

69 (131)

– Now, to explicitly characterize the imbalance effects on the individual

channel signals, we write the multichannel received signal r(t) as
r (t ) = 2 Re[ z (t )e j 2π f LOt ] = z (t )e j 2π f LOt + z * (t )e − j 2π f LOt
– As a model, the received signal r(t) is down-converted to baseband by mixing it
with xLO (t).
– Assuming that HNOM (f) = 1 for |f| ≤ B/2 and HNOM (f) = 0 for |f| > B/2, the
downconverted signal x(t) can be easily written as
x(t ) = K1 z (t ) + K 2 z * (t )
– To analyze the effect of branch mismatches, the signal x(t) can first be written
as x(t) = xI (t) + jxQ (t), where
xI (t ) = z I (t )
xQ (t ) = g cos(φ ) zQ (t ) − g sin(φ ) z I (t )

– Then, in terms of Fourier transforms, the signal z'(t) = zI'(t) + jzQ'(t) after branch
mismatches is given by
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

70 (131)
Z ′( f ) = Z I′ ( f ) + jZ Q′ ( f )
= H I ( f ) X I ( f ) + jH Q ( f ) X Q ( f )
= H I ( f ) Z I ( f ) + jH Q ( f )[ g cos(φ ) Z Q ( f ) − g sin(φ ) Z I ( f )]
= [ H I ( f ) − jH Q ( f ) g sin(φ )]Z I ( f ) + j[ H Q ( f ) g cos(φ )]Z Q ( f ).
= AI ( f ) Z I ( f ) + jAQ ( f ) Z Q ( f )
– After some manipulation, the above result can be written in a more convenient
form as (see the general theorem in the wideband imbalance model section)
Z ′( f ) = G1 ( f ) Z ( f ) + G2 ( f ) Z * (− f )
where
G1 ( f ) = [ AI ( f ) + AQ ( f )]/ 2
= [ H I ( f ) + H Q ( f ) ge − jφ ]/ 2

G2 ( f ) = [ AI ( f ) − AQ ( f )]/ 2
= [ H I ( f ) − H Q ( f ) ge jφ ]/ 2

71 (131)

– In the above model, the term relative to Z*(–f) is caused by the imbalances and
represents the image aliasing effect.
→ with perfect matching, I/Q processing allows us to consider negative and
positive frequencies separately
→ this separabililty of negative and positive frequencies (the ideal case) is lost
due to mismatches
– Now, the image attenuation of the analog front-end can be defined as
L( f ) = | G1 ( f ) |2 / | G2 ( f ) |2
– With practical analog electronics (as stated earlier), this attenuation is usually in
the order of 20…40 dB.
– The important question is whether this 20…40 dB attenuation is sufficient
→ depends on the architecture
→ direct-conversion (zero-IF) receiver: sufficient, especially with low-order
modulations
→ low-IF receiver: insufficient, even though the system specs help to some
extent
→ general
Proceeding wideband
of the SDR receiver:
03 Technical Conference clearly
and Product insufficient
Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

72 (131)
R(f)

f
−fLO fLO
Z(f) Z*(–f)

f f
−fIF fIF −fIF fIF
Z'(f) = G1(f)Z(f) + G2(f)Z*(–f)

f
−fIF fIF

73 (131)

– Possible solutions to enhance the image attenuation:

→ analog RF image reject filtering (real or complex)
→ difficult to implement, complicates the analog front-end
→ limits the integrability
→ limits the flexibility
→ advanced DSP at baseband (referred to as imbalance compensation)
→ estimate the amplitude and phase mismatches and correct them
→ training signals needed usually
→ difficulties with frequency-selective mismatches
→ novel statistical signal processing based methods to remove the
image interference
→ signal estimation vs. mismatch estimation
→ basic tools: adaptive interference cancellation or blind signal separation
→ blind, i.e., no training signals needed
→ can easily cope with frequency- and time-dependent imbalances
→ to be introduced next
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

74 (131)
4. ADVANCED DSP FOR I/Q IMBALANCE COMPENSATION
Baseband Signal Model for Digital Imbalance Compensation
– Here, in our formulation, the task of imbalance compensation is to enhance
the finite image attenuation L(f) of the analog processing.
– More precisely, the target is to obtain an image-free observation of a
specific channel (referred to as the desired channel) signal located at non-zero
intermediate frequency (IF) after the initial wideband downconversion
→ this signal estimation based approach is different from traditional imbalance
compensation techniques
• traditional approach is to try to estimate the imbalance parameters and use
them in some kind of a correction network
• here we directly estimate the final “quantity of interest”; the desired signal
– Some notations
→ desired channel (grey) baseband equivalent signal s (t )
→ image channel (dark) baseband equivalent signal i (t )
→ PX = E(|x(t)|2) in general

75 (131)

– As in the ideal (perfect matching) case, the imbalanced multichannel signal Z'(f)
contains a
→ desired signal component around +fIF
→ an image component around –fIF
– Due to imbalances (see the previous figure), Z'(f) has
→ also a destructive image signal component around +fIF
→ also a desired signal component around –fIF
– Motivated by this, we generate two baseband observations, d(t) and v(t)
→ d(t) is the baseband observation of the combined signal around +fIF
→ v(t) is the mirrored (complex-conjugated) baseband observation of the
combined signal around –fIF
→ this observation generation is illustrated in the following figure
→ the desired signal s(t) is then estimated as
sˆ(t ) = L{d (t ), d (t − 1), d (t − 2),..., v(t ), v(t − 1), v(t − 2),...}
→ what kind of processing L{.} is used, is not defined yet
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

76 (131)
Z'( f ) D( f ) V( f )

f f f
−fIF fIF

d(n)
HLP (z)

COMPENSATION
I(n)

ALGORITHM
e − jω IF n
sˆ(n)
j
Q(n) e + jω IF n
v(n)
HLP (z) (.)*

77 (131)

– After some manipulation, the frequency domain expression for these

observations can be written using matrix formulation as
X( f ) = A( f )S( f )
where X(f) = [D(f) V(f)]T, S(f) = [S(f) I *(–f)]T, and s(t) and i(t) denote the
baseband equivalents of the desired and image signals, respectively.
– The matrix A(f) is given by
 G ( f + f IF ) G2 ( f + f IF ) 
A( f ) =  *1 *  P( f , BC )
G
 2 ( − f − f IF ) G1 ( − f − f )
IF 

where BC denotes the individual channel bandwidth and P(f,BC ) = diag[Π (f,BC ),
Π (f,BC )] with Π (f,BC ) = 1 for | f| ≤ BC /2 and zero otherwise.
– If HI ( f ) = HQ ( f ), the general model reduces to an instantaneous mixture model

d (t )  K1 PS K 2 PI   s1 (t ) 
 v(t )  =  K * P 
K1* PI   s 2 (t )
   2 S
where s1(t) = s(t)/sqrt(PS ), s2(t) = i*(t)/sqrt(PI ), and PX = E(|x(t)|2) in general.
– This is a valid
Proceeding model
of the SDR if the
03 Technical I/Q demodulator
Conference is Copyright
and Product Exposition. the main source
© 2003 ofAllimbalance.
SDR Forum. Rights Reserved

78 (131)
– Notice that compensation is actually needed only if the image signal is more
powerful than the desired signal, i.e.,
→ sˆ(t ) = d (t ) is a good estimator if PS >> PI
– Even though the baseband model was here derived in continuous-time domain,
the observations can in practice be generated digitally (after A/D)
→ excess imbalance effects are avoided
– Consequently, discrete-time notations d(n), v(n), s(n), and i(n) are used
hereafter.
– In general, the observations x1(n) = d(n) and x2(n) = v(n) appear as convolutive
mixtures of the effective source sequences s1(n) = s(n) and s2(n) = i*(n)
→ the instantaneous mixture model is a special case of this

79 (131)

Adaptive Interference Cancellation (IC) Based Compensation

– If the desired channel signal is originally more powerful than the image signal
(PS >> PI), the image attenuation of analog processing is sufficient
→ the desired channel observation d(n) can be used directly as an estimate of
s(n)
– On the other hand, in the difficult case of a strong image signal (PI >> PS),
→ the attenuation of analog processing is insufficient
→ v(n) is highly correlated with the interfering signal component (see the
previous figure) but only weakly correlated with the desired signal component
of d(n)
– Motivated by this, adaptive interference canceller can be used to estimate s(n)
as

sˆIC (n) = d (n) − ∑kN=IC0 wk (n)v(n − k )

– The filter coefficients wk (n), k = {0, …, NIC }, can be adapted with any practical
algorithm, such as the well-known least-mean-square (LMS) or recursive least-
squares (RLS) algorithms.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

80 (131)
Multichannel Blind Deconvolution (MBD) Based Compensation
– Assumption: Signals in different frequency channels are statistically independent.
– Observation: With practical imbalance values, the general mixture model is
invertible (i.e., A(f) is non-singular).
→ blind signal separation (or multichannel blind deconvolution in general)
algorithms can be used to estimate the source vector s(n) = [s(n) i*(n)]T as

sˆ (n) = ∑ kN=MBD
0 Wk ( n ) x( n − k )

where x(n) = [d(n) v(n)]T.

→ the actual desired channel signal s(n) is then estimated as
sˆMBD (n) = eTi sˆ (n) , i = 1 or 2,
where e1 = [1 0]T or e2 = [0 1]T is used, depending on the possible source
permutation (i.e., MBD produces also an estimate of the image signal).
– In general, there exists a wide variety of different approaches to measure the
independence of the separated output signals, and thus, to adapt the demixing
matrices Wk (n), k = {0, …, NMBD}.

81 (131)

Comparisons
– In general, the IC based compensator is only utilizable if the image signal is
more powerful than the desired channel signal.
– Consequently, some kind of power estimation of the different channel signals is
needed to decide when to switch the IC structure on and off.
– This problem, usually referred to as signal leakage, can in theory be avoided
using the MBD based compensator, though, no compensation is actually needed
if the image signal is weak.
– On the other hand, the performance of the IC based solution is likely to be more
insensitive to the effects of additive noise and symbol timing errors, and,
especially, to different interferer types.
– Naturally, there is also the issue of computational complexity.
– Both techniques
→ are blind, i.e., no training signals needed !!
→ are able to handle frequency-selective mismatches
→ are also able to cope with time-variant mismatch effects due to inherent
adaptive
Proceeding nature
of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

82 (131)
Simulation Example
– Front-End Parameters:
→ The received signal consists of the desired and image channels of bandwidth
0.2π located originally around 0.7π and 0.3π, respectively.
→ The desired and image signals are QPSK- and 8PSK-modulated,
respectively, with raised-cosine pulse-shapes (roll-off 0.35).
→ The relative power difference is –40 dB.
→ In translating the desired channel signal to an IF of 0.2π, imbalance values of
g = 1.02 and φ = –2° are used for the quadrature demodulator.
→ After that, the branch mismatches are modelled as HI (z) = 0.01 + z –1 + 0.01z –2
and HQ (z) = 0.01 + z –1 + 0.2z –2.
→ Finally, the symbol rate baseband observations d(n) and v(n) are generated
by proper frequency translations of ±0.2π, lowpass filtering, and decimation.

83 (131)

– IC Simulation:
→ The standard RLS algorithm with a forgetting factor of 0.999 is used to adapt
the IC filter of length 5 (NIC = 4).
→ The total number of samples is 2,000 to guarantee steady-state operation.

– MBD Simulation:
→ The natural gradient based algorithm of Amari et al. is used with a step-size
of 0.001 to adapt the demixing filters of length 5 (NMBD = 4).
→ The total number of samples is 12,000 to quarantee steady-state operation.
→ To stabilize the adaptation, the source separator input signals x1(n) = d(n) and
x2(n) = v(n) are normalized as (ad-hoc)
x1(n) ← x1(n)/sqrt(P1),
x2(n) ← x2(n)/sqrt(P2),

where the power estimates P1(n) and P2(n) are obtained recursively as
Pi (n) = 0.995Pi (n–1) + 0.005|xi (n)|2, i = 1 and 2.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

84 (131)
– Comments:
→ Front-end imbalance properties used in the simulations are illustrated in
Figure 1 and Figure 2.
→ Single realizations of the absolute value of the “center taps” w2(n) and W2,ij(n)
= [W2(n)]ij are presented in Figure 3 and Figure 4 to illustrate the
convergence properties.
→ Clearly, the IC algorithm converges much faster than the MBD algorithm.
→ However, as verified by Figure 5, there is no difference in the steady-state
operation between the two methods (only the desired signal estimate is
shown for the MBD method, it also produces an estimate of the original image
signal).

85 (131)

1.2

1.1
|H | / |H |
I

1
Q

0.9

0.8
−1 −0.5 −0.2 0 0.2 0.5 1

20
arg(HQ) − arg(HI)

−10

−20
−1 −0.5 −0.2 0 0.2 0.5 1
Frequency ω / π

Figure 1. Relative amplitude and phase mismatches of the branch components. Positive (grey) and
negative (dark) IF bands are illustrated in different colours.

86 (131)
1.2

1.1

|G1|
1

0.9

0.8
−1 −0.5 −0.2 0 0.2 0.5 1

0.12

0.1
|G2|

0.08

0.06
−1 −0.5 −0.2 0 0.2 0.5 1
Frequency ω / π

Figure 2. General imbalance coefficients G1 and G2. Positive (grey) and negative (dark) IF bands are
illustrated in different colours.

87 (131)

10
Absolute value

W2,11

W2,21
0
0 2000 4000 6000 8000 10000 12000

10
Absolute value

W2,12

5
W2,22
0
0 2000 4000 6000 8000 10000 12000
Iteration number n
Figure 3. Center tap values W2,ij(n) = [W2(n)]ij of the demixing filters using the natural gradient algorithm
(step-size 0.001).

88 (131)
0.2

Absolute value
0.15
w2
0.1

0.05

0
0 20 40 60 80 100
Iteration number n

Figure 4. Center tap value w2(n) of the RLS (forgetting factor 0.999) based interference canceller.

ŝIC ŝMBD
1 1

0.5 0.5

0 0

−0.5 −0.5

−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1

Figure 5. Compensated output samples after convergence of both the IC (left) and MBD (right) based
compensators. Also shown are the ideal QPSK symbol locations (white asterisks).

89 (131)

Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved
5. SECOND-ORDER SAMPLING AND ENHANCEMENTS
– Starting point: Complex (I/Q) sampling of bandpass signals.

fS I
input

HT Q
fS

– Idea: Approximate the needed 90 degree phase shift (Hilbert transformer) using
a simple time delay of one quarter of the carrier cycle.

fS I’
input

DELAY Q’
fS

90 (131)

– The time delay of ∆T = 1/(4fC) corresponds to a frequency-dependent phase

shift of
π f
−2π f ∆T = −
2 fC
→ thus the ideal Hilbert filtering is achieved only at the carrier-frequency ±fC
→ as a result, the supression of the negative frequency content (image
components) is generally imperfect

input spectrum input spectrum

f f
output spectrum output spectrum

91 (131)
Filtering Effect of the Delay Processing
– We write the input bandpass signal r (t ) in terms of its baseband equivalent
signal z (t ) = z I (t ) + jzQ (t ) as (scaling by 2 is irrelevant and could be ignored)

r (t ) = 2 Re[ z (t )e j 2π fC t ] = z (t )e j 2π fC t + z * (t )e − j 2π fC t
= 2 z I (t ) cos(2π f C t ) − 2 zQ (t ) sin( 2π f C t )

– Then the complex signal, say x(t), entering the sampler(s) is generally of the
form x(t ) = r (t ) + jr (t − ∆T ) where ∆T = 1 /( 4 f C ) .
– The spectrum of this signal can be written as
X ( f ) = [1 + jH D ( f )]R( f )
where
H D ( f ) = e− j 2π f ∆T .
– This effect is illustrated in frequency domain in the following figure.

92 (131)

WITH HILBERT TRANSFORMER

1+jHHT (f)

f
−fC fC

WITH DELAY
1+jHD (f)

f
−fC fC

93 (131)
– Now suppose you use a subharmonic sampling frequency fS = fC / r, where the
integer r ≥ 1 denotes the subsampling ratio.
→ the signal is aliased directly to baseband
→ in case of delay processing, the component from negative frequencies
falls directly on top of the desired component
• kind of self-interference
• actually depends on the structure of the signal
o in case of a single frequency channel (as shown below), this interference is
not really problematic, especially with low-order modulations
o but how about in the wideband/multichannel case ?

WITH HILBERT TRANSFORMER WITH DELAY

••• ••• ••• •••

f f
−fS fS −fS fS

94 (131)

– When applied to multichannel signals, the image band signal ban be much
more powerful than the desired channel
→ obvious problem
→ quantitative measures given shortly

1+jHHT (f)

f f
−fC fC −fIF fIF

1+jHD (f)

95 (131)
Image Attenuation Analysis
– After some manipulations, the sampled complex signal (with fS = fC / r, r integer)
can be written as
x(nTS ) = r (nTS ) + jr (nTS − ∆T )
= ...
= 2 z I (nTS ) + j 2 zQ (nTS − ∆T )
= I ′(n) + jQ′(n)
– Thus it is interesting to note that the front-end delay maps directly into a
corresponding delay of the baseband observation.
– For analysis purposes, we can formally think that the discrete-time signal is
obtained by sampling the corresponding continuous time baseband signal
z ′(t ) = 2 z I (t ) + j 2 zQ (t − ∆T )
– This can be written in a more informative form as
z′(t ) = z (t ) + z (t − ∆T ) + z * (t ) − z * (t − ∆T )

96 (131)

→ the part including z(t) and z(t – ∆T) corresponds to the signal component
originating from positive frequencies
→ the part including z*(t) and z*(t – ∆T) corresponds to the signal component
originating from negative frequencies
– To get the exact image attenuation, we write the Fourier Transform of z'(t) as
Z ′( f ) = (1 + e − j 2πf∆T ) Z ( f ) + (1 − e − j 2πf∆T ) Z * ( − f )
– Then, the image attenuation L 2 (f) provided by the second-order sampling is
given by
2
1 + e − j 2 π f ∆T 1 + cos(2π f ∆T )
2

L2 ( f ) = = ... =
1 − e − j 2 π f ∆T
2
sin(2π f ∆T )

– To illustrate, if the center frequency fC =100 MHz and the bandwidth B=25 MHz,
the image attenuation L 2(f) at the band edge is only around 20 dB
→ in case of multichannel downconversion, the power difference of the
individual channel signals can be even 50…100 dB
→ then, it is clear that the image attenuation L 2(f) and thus, the basic second-
order sampling scheme, are not sufficient as such for multichannel receivers!!
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

97 (131)
Second-Order Sampling and Enhanced Image Rejection
– In narrowband single channel receivers, the image signal is inherently a "self-
image" and the attenuation of the basic second-order sampling scheme as such
can be adequate.
– In multichannel receivers, as discussed before, the image band signal can be up
to 50-100 dB stronger than the desired channel signal, and the image attenuation
of the basic second-order scheme alone is clearly insufficient.
– Two alternative methods to enhance this image attenuation are presented
next.
– For simplicity, the following compensation methods are analyzed in continuous-
time domain.

98 (131)

Interference Cancellation
– The basic idea: To enhance the obtainable image attenuation, and thus, to
reproduce an accurate baseband observation of the multichannel signal z (t ) , the
interference canceller –type of compensation structure is proposed.

I'(n)=r(nTS)
fS
r(t) z′(nTS ) + zˆ(nTS )

Q'(n)=r(nTS – ∆T) –
DELAY
fS

(.)* C(e jω)

99 (131)
– As illustrated, the idea is to use a fixed compensation filter C ( f ) together with
the complex conjugate of z ′(t ) as a reference signal to estimate and subtract the
image signal interference.
– Based on the previous signal model for Z ′( f ) and on the proposed
compensation strategy, the frequency response of zˆ (t ) can be written as

Zˆ ( f ) = Z ′( f ) − C ( f ) Z ′ * (− f )
= G1, IC ( f ) Z ( f ) + G2, IC ( f ) Z * (− f )
where
G1, IC ( f ) = 1 + e − j 2π f ∆T − C ( f )(1 − e − j 2π f ∆T )

G2, IC ( f ) = 1 − e − j 2π f ∆T − C ( f )(1 + e − j 2π f ∆T )
– As a result, the image signal components of zˆ (t ) are attenuated with respect to
the desired signal components by
2
G1, IC ( f )
LIC ( f ) = 2
G2, IC ( f )

100 (131)

– Then, to completely cancel the image signal components, the compensator

frequency response C ( f ) should be selected in such a way that G2, IC ( f ) = 0 .

– This solution, referred to as the "zero forcing" –solution C ZF ( f ) , is given by

1 − e − j 2πf∆T
C ZF ( f ) = .
1 + e − j 2πf∆T
– After some manipulation, this can be written as
j sin( 2πf∆T )
C ZF ( f ) = .
1 + cos(2πf∆T )
– The above frequency response can be well approximated using a real-valued
impulse response with odd symmetry.
– Therefore, no "cross-filtering" between the I and Q components of the input
signal z ′(t ) is needed.
– Using this optimum compensator, the gain G1,ZF ( f ) of the desired signal
components can be easily shown to be of the form

101 (131)
− j 2πf∆T − j 2πf∆T 4e − j 2πf∆T
G1, ZF ( f ) = 1 + e − C ZF ( f )( 1 − e )= .
1 + e − j 2πf∆T
– With reasonable bandwidth-to-center frequency ratios ( B / f C ), this describes
practically a constant gain, linear phase frequency response.
– As a consequence, the effect of image components Z * (− f ) can indeed be
successfully compensated without causing any notable distortion to the desired
components Z ( f ) .
– Thus, an accurate reproduction of z (t ) is obtained, i.e., zˆ (t ) ≈ z (t ) (up to a delay
and multiplication by a constant).
– In the actual digital implementation, C ZF ( f ) gives the desired response which
should be approximated using a finite order discrete-time FIR/IIR filter.
– This can be done, e.g., using a type III FIR filter (odd symmetry).
– A design example is given below:
→ fC = 100 MHz, B = 25 MHz, fS = 50 MHz
→ type III FIR filter, length 9, LS and minimax approximations of the given
optimum response

102 (131)

40
Attenuation [dB]

100

120
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
Frequency ω / π

– dash-dotted: basic second-order sampling (L2)

– solid: enhanced scheme with LS –optimized IC compensation filter (LIC)
– dashed: enhanced scheme with minimax –optimized IC compensation filter (L )
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights ReservedIC

103 (131)
Fractional-Delay Filtering
– Motivated by the original signal model, the image attenuation of the basic
second-order sampling scheme can also be enhanced by properly delaying the I
branch signal r(nTS) relative to the Q branch signal r(nTS – ∆T).
– Using a sampling frequency fS = fC /r, the needed delay ∆T = TS /(4r) is only a
fraction of the sampling interval TS.
– Therefore, a digital fractional delay filter D(z) is a natural choice for the delay
implementation.
– The basic block-diagram of the modified second-order sampling based
quadrature demodulator utilizing this idea is presented below.

I'(n)=r(nTS)
D(e jω)
fS
r(t) zˆ(nTS )

Q'(n)=r(nTS – ∆T)
DELAY
fS

104 (131)

– To analyze the image attenuation of the complete structure, we proceed as

follows (in continuous-time domain for simplicity).
– First, let D(f) denote the frequency response of the fractional delay
compensation filter.
– Then, the Fourier transform of the compensated signal zˆ (t ) is given by

Zˆ ( f ) = 2 Z I ( f ) D( f ) + j 2 Z Q ( f )e − j 2πf∆T
– After some manipulations, this can be written in a more convenient form as
Zˆ ( f ) = G1, FD ( f ) Z ( f ) + G2, FD ( f ) Z * (− f )
where
G1, FD ( f ) = D( f ) + e − j 2π f ∆T

G2, FD ( f ) = D ( f ) − e − j 2π f ∆T
– As a result, the enhanced image attenuation LFD (f) of fractional delay processing
is given by

105 (131)
2 2
G1, FD ( f ) D ( f ) + e − j 2 π f ∆T
LFD ( f ) = 2
= 2
G2, FD ( f ) D ( f ) − e − j 2 π f ∆T

– Naturally, the ideal frequency response of the needed fractional delay filter is,
indeed, e − j 2πf∆T for which LFD → ∞.
– Based on the previous analysis, this would result in a perfect (up to a delay and
multiplication by a constant) reconstruction of the desired baseband equivalent
signal as zˆ (t ) = 2 z (t − ∆T ) .
– Considering the actual digital implementation, let µ = ∆T/TS = 1/(4r) denote the
normalized fractional delay.
– Then, the frequency response of the ideal digital FD filter is given by
DOPT (e jω , µ ) = e − jωµ for ω ≤ πB / f S ,

where ω = 2πf/fS .
– Again, the task is to approximate this ideal response using a finite order FIR/IIR
filter.

106 (131)

– Previous design example:

→ fC = 100 MHz, B = 25 MHz, fS = 50 MHz ( → µ = 1/8)
→ FIR filter, length 8, LS and minimax approximations of the given optimum
response
→ results are illustrated in the following figure

– General conclusion: Based on the presented design examples, image

attenuations in the order of 80 … 100 dB can be easily achieved (at least
theoretically) using either of the presented techniques (IC/FD).

107 (131)
0

40
Attenuation [dB]

100

120
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
Frequency ω / π

– dash-dotted: basic second-order sampling (L2)

– solid: enhanced scheme with LS –optimized FD compensation filter (LFD)
– dashed: enhanced scheme with minimax –optimized FD compensation filter (LFD)

108 (131)

Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved
6. I/Q SIGNAL PROCESSING IN FREQUENCY SYNTHESIZERS
– The design of frequency synthesizers is challenging for wireless applications:
→ spectral purity, high frequency range, fast tuning, power consumption, etc.
– Here the idea of combining digital and analog synthesis techniques for
achieving these goals is discussed and analyzed
→ the proposed architecture uses I/Q modulation to translate a digitally
synthesized tuneable low frequency tone to the final frequency range
– Practical problems: unavoidable mismatches between the amplitudes and
phases of the I and Q branches result in imperfect sideband rejection degrading
the spectral purity of the synthesized signal.
– A compensation structure based on digital pre-distortion of the low frequency
tone is presented to enhance the signal quality
→ practical algorithms for updating the compensator parameters are proposed
based on minimizing the envelope variation of the synthesizer output signal
→ simulation results are also presented to illustrate the efficiency of the
proposed synthesizer concept

109 (131)

Synthesizer Architecture
Fixed
LO
Pre-Compensation ILO QLO
INCO D/A x1(t) xI (t)
a11 LPF

a21 + Output
NCO / DDS x(t)
a12 –

a22
QNCO D/A x2(t) xQ (t)
LPF

{a11 , a12 , a21 , a22}

Digital A(t) Envelope

A/D
Processing Detector

Figure. Synthesizer architecture where digital pre-compensation is used to compensate for the
non-idealities of the analog part.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

110 (131)
– Signal Analysis:
→ NCO signals: INCO(t)=cos(ωNCOt) and QNCO(t)=sin(ωNCOt)
→ I/Q mixer: ILO(t)=cos(ωLOt) and QLO(t)=gsin(ωLOt + φ) where g and φ denote
the gain and phase imbalances of the mixing stage
→ furthermore, let θ1 and θ2 denote the relative phase shifts due to the D/A
converters and branch filters.
– Now, the synthesizer output signal x(t) can be easily shown to be of the form
x(t)=Re[xLP(t)exp(jωLOt)] where the lowpass equivalent xLP(t) is given by
xLP (t ) = αe jω NCOt + βe − jω NCOt
– Thus, the synthesizer output consists of two spectral components:
→ the desired tone at ωLO + ωNCO and
→ the image tone at ωLO – ωNCO
whose relative strengts |α |2 and |β |2 depend on g, φ, and θ = θ2 – θ1, as well as on
the compensation parameters aij.

111 (131)

– With practical analog electronics, the image tone attenuation defined here as
L = | α |2 | β |2
is only 20-40 dB if no compensation is used (i.e., a11 =a22 =1 and a12 =a21 =0)
→ in wireless systems, these levels of synthesizer spurious tones can result in
severe interchannel interference (ICI)
→ in general, image tone attenuations in the order of 50-80 dB are needed in
wireless applications
– The image tone at the synthesizer output can, however, be canceled with
proper pre-compensation parameters aij.
– In fact, setting β = 0 and α = exp(jθ1) will force the output to be a pure sinusoidal
(with relative phase θ1).
– These optimum parameters aij can be easily shown to be of the form
sin(θ ) cos(θ )
a11 = 1, a12 = tan(φ ), a 21 = − , a 22 =
g cos(φ ) g cos(φ )

112 (131)
– If the only motivation is really just to cancel the image tone (i.e., to set β = 0), a
more simple solution is also available.
– Using the notation ψ = φ – θ, the solution can be formulated as
1
a11 = 1, a12 = tan(ψ ), a 21 = 0, a 22 =
g cos(ψ )
for which also α ≈ exp(jθ1) with practical imbalance values.
– Thus, only two actual compensation parameters a12 and a22 need to be
implemented.
– Notice that only the phase error difference ψ = φ – θ and the gain mismatch g
are actually contributing to the image tone, and therefore, on the ideal
compensation parameters.

113 (131)

Practical Compensator Update

– In practice, the imbalances g, φ, and θ needed to calculate the previous
compensators are unknown and need to be measured or estimated somehow.
– To emphasize implementation simplicity, the approach taken here is to carry
out the estimation using only the envelope of the synthesizer output:

A(t ) = | xLP (t ) | = | αe jω NCOt + βe − jω NCOt |= | α |2 + | β |2 +2 Re[αβ *e j 2ω NCOt ]

→ the envelope is flat only if the image tone is completely attenuated (β = 0)

– Thus, instead of estimating g and ψ, a more simple approach is to directly

adapt a12 and a22 to minimize the envelope variation.
– One possibility is to consider the envelope peak-to-peak (PP) value
d A = max{A(t )} − min{ A(t )}
– Given that |α|>|β|, which is always the case with practical imbalance values, the
above peak-to-peak value can also be expressed as

114 (131)
– Though not strictly parabolic, the dA -surface having a unique minimum lends
itself well to iterative minimization.
– The true gradient of dA (derivative with respect to a12 and a22) depends, however,
on g and ψ which are unknown.
– A practical approach is then to adapt only one parameter (either a12 or a22)
at a time.
– The direction (sign) of the needed one dimensional gradient at each iteration can
be determined based on observing the behaviour of dA between two previous
adaptations of the corresponding parameter.
– Assuming that a12 and a22 are updated at odd and even adaptation instants,
respectively, this leads to the following update rule
d A ( n)
aˆ12 (n) = aˆ12 (n − 2) + K 12 (n)λ1
λ 2 + d A ( n)
aˆ 22 (n) = aˆ 22 (n − 1)

for n odd with K12 ∈{+1,–1} and

115 (131)

aˆ12 (n) = aˆ12 (n − 1)

d A ( n)
aˆ 22 (n) = aˆ 22 (n − 2) + K 22 (n)λ1
λ 2 + d A ( n)
for n even with K22 ∈{+1,–1}.
– In the above adaptation scheme, K12(n)=K12(n–2) if dA(n–1)≤dA(n–2) and
K12(n)=–K12(n–2) if dA(n–1)>dA(n–2) with similar reasoning for K22.
– Step-size parameters λ1 >0 and λ2 ≥0 are used here to control the adaptation
speed and steady-state accuracy. Notice that the magnitude of the update term
is always between zero and λ1, depending on the value of dA.

116 (131)
Example Simulations
– To demonstrate the proposed synthesizer concept with digital pre-compensation,
some computer simulations are carried out.
– In the simulations, imbalance values of g=1.05, φ =6°, and θ =1° (ψ = φ – θ =5°)
are used, corresponding to an image tone attenuation of approximately 26 dB.
– Pre-compensation parameters are initialized as a12 =0 and a22 =1 and are then
iteratively updated one-by-one to minimize the peak envelope variation using the
approach described above.
– In general, updating is carried out once per envelope cycle and step-size values
of λ1 =0.01 and λ2 =0.02 are used.
– With these example values, the dA -surface is illustrated in Figure 1 as a function
of a12 and a22.
– Also shown in the a12 ,a22 -plane is the behaviour of the pre-distortion parameters
during one simulation realization.

117 (131)

0.2

0.15
dA

0.1

0.05

0
1.1
0.2
1 0.15
a22 0.1
0.9 0.05
0 a12
0.8 −0.05
Figure 1: Peak envelope variation dA as a function of the compensation parameters a12 and a22 for
g = 1.05 and ψ = 5°. Also shown in the a12 ,a22 -plane is one realization of the compensation
parameters during the adaptation.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

118 (131)
– The corresponding output envelope is depicted in Figure 2 verifying successful
synthesizer operation.
– With these imbalance and step-size values, the steady-state operation is
reached in 50 iterations or so and the steady-state image attenuation is in the
order of 100-150 dB.

Output Envelope
1.1
Amplitude

1.05

0.95
0 20 40 60 80 100
Time in Envelope Cycles

Figure 2: Envelope of the synthesizer output signal during the adaptation (g = 1.05 and ψ = 5°).

119 (131)

– To illustrate the ability to perform the compensation also in time-varying

environments, an abrupt change in the imbalances is tested.
– The gain mismatch coefficient g is changed from 1.05 to 0.98 and the I/Q mixer
phase imbalance φ from 6° to 4°.
– The resulting output envelope is presented in Figure 3 evidencing fast and
accurate synthesizer operation also in case of time-varying imbalances.

Output Envelope
1.1

1.05
Amplitude

0.95

0.9
0 20 40 60 80 100
Time in Envelope Cycles
Figure 3: Envelope of the synthesizer output signal during the adaptation when the initial
imbalances g = 1.05 and ψ = 5° are changed to g = 0.98 and ψ = 3°, respectively.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

120 (131)
Some Practical Matters
– The output envelope is periodic with fundamental frequency 2ωNCO
→ if dA is determined and compensation parameters updated once per envelope
cycle (or once per several cycles), the frequency range of the digital synthesis
part should be selected in such a manner that the envelope variation rate
does not limit the whole synthesizer's settling time
→ on the other hand, higher frequencies in the digital part always imply higher
sampling rates and higher power consumption
→ proper compromise between these two issues is needed
– Data and coefficient wordlengths:
→ according to preliminary results, image tone attenuations in the order of 80 dB
are achievable with 16 bits, more detailed analysis still needed
→ one important issue in this context is also the available accuracy of the
envelope variation measurements
– Local oscillator leakage: creates an additional spectral component at ωLO, can be
compensated by adding proper constants to the low frequency (NCO) signals.

121 (131)

Conclusions
– In this section, frequency synthesizer design for wireless applications was
considered.
– To achieve fast switching capabilities with high operating frequencies and
reasonable power consumption, the approach was to use I/Q modulation of
digitally tunable low frequency tone.
– The practical imbalance problem with analog I/Q signal processing was then
considered and analyzed.
– Based on this analysis, a digital pre-compensation structure was presented
together with some simple yet efficient approaches to determine the
compensator coefficients.
– The proper operation of the whole synthesizer was demonstrated using
computer simulations both in time-invariant and time-varying situations.
– Future work should be directed to more detailed evaluation of the finite
wordlength effects (hardware prototyping, etc.).

122 (131)
GENERAL SUMMARY
– Efficient processing of complex (I/Q) signals is the key in developing and building
sophisticated communication systems.
– Frequency translations and advanced sampling techniques as well as efficient
multirate DSP are good examples where complex signal concepts are especially
useful
→ Hilbert transform and the notion of analytic signals
→ polyphase filtering
→ second-order sampling and its enhancements
– One important practical aspect is the mismatch problem associated with analog
I/Q signals
→ image attenuation analysis
→ digital compensation techniques
– Frequency synthesizers represent another interesting application where I/Q
signal processing can be utilized.

123 (131)

SOME REFERENCES
[1] A. A. Abidi, “Direct conversion radio transceivers for digital communications,” IEEE J. Solid-State Circuits, vol. 30, pp.
1399-1410, Dec. 1995.
[2] A. A. Abidi, “CMOS wireless transceiver: The new wave,” IEEE Commun. Mag., vol. 37, pp. 119-124, Aug. 1999.
[3] A. A. Abidi, “Low-power radio-frequency ICs for portable communications,” in L. E. Larson, Ed., RF and Microwave
Circuit Design for Wireless Communications. Norwood, MA: Artech House, 1996, pp. 43-98.
[4] S. Amari, S. C. Douglas, A. Cichocki, and H. H. Yang, “Multi-channel blind deconvolution and equalization using the
natural gradient,” in Proc. IEEE Workshop Signal Processing Adv. Wireless Commun., Paris, France, Apr. 1997, pp. 101-104.
[5] R. Ansari and B. Liu, “Multirate signal processing”, in S. K. Mitra and J. F. Kaiser, Eds., Handbook for Digital Signal
Processing. New York: John Wiley & Sons, 1993.
[6] D. Babic, J. Vesma, and M. Renfors, “Decimation by irrational factor using CIC-filter and linear interpolation”, in Proc.
IEEE Int. Conf. Acoust., Speech, Signal Processing, Salt Lake City, UT, May 2001.
[7] E. Buracchini, “The software radio concept,” IEEE Commun. Mag., vol. 38, pp. 138-143, Sept. 2000.
[8] C. Caballero Gaudes, M. Valkama, M. Renfors, and J. Ajanki, “Fast frequency synthesizer concept based on digital tuning
and I/Q signal processing,” in Proc. IEEE Int. Conf. Digital Signal Processing, Santorini, Greece, July 2002, pp. 1317-1320.
[9] J.-F. Cardoso, “Blind source separation: Statistical principles,” Proc. IEEE, vol. 86, pp. 2009-2025, Oct. 1998.
[10] J.-F. Cardoso and B. H. Laheld, “Equivariant adaptive source separation,” IEEE Trans. Signal Processing, vol. 44, pp. 3017-
3030, Dec. 1996.
[11] J. K. Cavers and M. W. Liao, “Adaptive compensation for imbalance and offset losses in direct conversion transceivers,”
IEEE Trans. Veh. Technol., vol. 42, pp. 581-588, Nov. 1993.
[12] E. Cetin, I. Kale, and R. C. S. Morling, “Adaptive compensation of analog front-end I/Q mismatches in digital receivers,” in
Proc. IEEE Int.
Proceeding Symp.
of the SDRCircuits Syst.,Conference
03 Technical Sydney, Australia, May
and Product 2001, pp.Copyright
Exposition. 370-373.© 2003 SDR Forum. All Rights Reserved
124 (131)
[13] J. Y. C. Cheah, “Introduction to wireless communications applications and circuit design,” in L. E. Larson, Ed., RF and
Microwave Circuit Design for Wireless Communications. Norwood, MA: Artech House, 1996, pp. 17-41.
[14] C. Chien, Digital Radio Systems on a Chip – A Systems Approach. Norwell, MA: Kluwer Academic Publishers, 2001.
[15] S. Choi, A. Cichocki, and S. Amari, “Flexible independent component analysis,” in Proc. IEEE Signal Processing Soc.
Workshop Neural Networks Signal Processing, Cambridge, UK, Aug. 1998, pp. 83-92.
[16] A. J. Coulson, “A generalization of nonuniform bandpass sampling,” IEEE Trans. Signal Processing, vol. 43, pp. 694-704,
Mar. 1995.
[17] A. J. Coulson, R. G. Vaughan, and M. A. Poletti, “Frequency-shifting using bandpass sampling,” IEEE Trans. Signal
Processing, vol. 42, pp. 1556-1559, June 1994.
[18] J. Crols and M. S. J. Steyaert, “An analog integrated polyphase filter for a high performance low-IF receiver,” in IEEE Symp.
VLSI Circuits Dig. Tech. Papers, Kyoto, Japan, June 1995, pp. 87-88.
[19] J. Crols and M. S. J. Steyaert, “A 1.5 GHz highly linear CMOS downconversion mixer,” IEEE J. Solid-State Circuits, vol.
30, pp. 736-742, July 1995.
[20] J. Crols and M. S. J. Steyaert, “A single-chip 900 MHz CMOS receiver front-end with a high performance low-IF topology,”
IEEE J. Solid-State Circuits, vol. 30, pp. 1483-1492, Dec. 1995.
[21] J. Crols and M. S. J. Steyaert, “Low-IF topologies for high-performance analog front ends of fully integrated receivers,”
IEEE Trans. Circuits Syst. II, vol. 45, pp. 269-282, Mar. 1998.
[22] J. Crols and M. S. J. Steyaert, CMOS Wireless Transceiver Design. Dordrecht, the Netherlands: Kluwer Academic
Publishers, 1997.
[23] C. Dick and f. harris, “Configurable logic for digital communication: Some signal processing perspectives,” IEEE Commun.
Mag., vol. 37, Aug. 1999.
[24] V. Eerola, T. Ritoniemi, and H. Tenhunen, “Second-order sampling and oversampled A/D- and D/A-converters in digital data
transmission,” in Proc. IEEE Int. Symp. Circuits Syst., Singapore, June 1991, pp. 1521-1524.

125 (131)

[25] C. W. Farrow, “A continuously variable digital delay element,” in Proc. IEEE Int. Symp. Circuits Syst., Espoo, Finland, June
1988, pp. 2641-2645.
[26] M. Faulkner, T. Mattsson, and W. Yates, “Automatic adjustment of quadrature modulators,” Electron. Lett., vol. 27, pp. 214-
216, Jan. 1991.
[27] W. Feixue, Y. Shaowei, and G. Guirong, “Mixer-free digital quadrature demodulation based on second-order sampling,”
Electron. Lett., vol. 34, pp. 854-855, Apr. 1998.
[28] M. E. Frerking, Digital Signal Processing in Communication Systems. New York: Chapman & Hall, 1994.
[29] A. Gatherer, T. Stetzler, M. McMahan, and E. Auslander, “DSP-based architectures for mobile communications: Past,
present and future,” IEEE Commun. Mag., vol. 38, pp. 84-90, Jan. 2000.
[30] J. P. F. Glas, “Digital I/Q imbalance compensation in a low-IF receiver,” in Proc. IEEE Global Telecommun. Conf., Sydney,
Australia, Nov. 1998, pp. 1461-1466.
[31] R. A. Green, “An optimized multi-tone calibration signal for quadrature receiver communication systems,” in Proc. IEEE
Workshop Statist. Signal Array Processing, Pocono Manor, PA, USA, Aug. 2000, pp. 664-667.
[32] R. A. Green, R. Anderson-Sprecher, and J. W. Pierre, “Quadrature receiver mismatch calibration,” IEEE Trans. Signal
Processing, vol. 47, pp. 3130-3133, Nov. 1999.
[33] J. E. Gunn, K. S. Barron, and W. Ruczczyk, “A low-power DSP core-based software radio architecture,” IEEE J. Select.
Areas Commun., vol. 17, pp. 574-590, Apr. 1999.
[34] S. L. Hahn, Hilbert Transforms in Signal Processing. Norwood, MA: Artech House, 1996.
[35] L. Hanzo, “Bandwidth-efficient wireless multimedia communications,” Proc. IEEE, vol. 86, pp. 1342-1382, July 1998.
[36] f. harris, C. Dick, and M. Rice, “Digital receivers and transmitters using polyphase filter banks for wireless communications,”
IEEE Trans. Microwave Theory and Techniques, vol. 51, pp. 1395-1412, Apr. 2003.
[37] f. harris and M. Rice, “Multirate digital filters for symbol timing synchronization in software defined radios,” IEEE J. Select.
Areas Commun.,
Proceeding vol.
of the SDR19,03pp. 2346-2357,
Technical Dec. 2001.
Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

126 (131)
[38] f. harris, “Digital signal processing in radio receivers and transmitters,” Int. J. Wireless Information Networks, vol. 5, pp.
133-145, 1997.
[39] f. harris, “Digital filter equalization of analog gain and phase mismatch in I-Q receivers,” in Proc. IEEE Int. Conf. Univ.
Pers. Commun., Cambridge, MA, USA, Sept. 1996, pp. 793-796.
[40] S. Haykin, Adaptive Filter Theory, 3rd ed. Upper Saddle River, NJ: Prentice-Hall, 1996.
[41] S. Haykin, Ed., Unsupervised Adaptive Filtering, vol I: Blind Source Separation. New York: John Wiley & Sons, 2000.
[42] T. Hentschel and G. Fettweis, “Software radio receivers,” in F. Swarts, P. van Rooyan, I. Oppermann, and M. P. Lötter, Eds.,
CDMA Techniques for Third Generation Mobile Systems. Norwell, MA: Kluwer Academic Publishers, 1999, pp. 257-283.
[43] T. Hentschel, M. Henker, and G. Fettweis, “The digital front-end of software radio terminals,” IEEE Pers. Commun., vol. 6,
pp. 40-46, Aug. 1999.
[44] D. Hilborn, S. P. Stapleton, and J. K. Cavers, “An adaptive direct conversion transmitter,” in Proc. IEEE Veh. Technol. Conf.,
Denver, CO, USA, June 1992, pp. 764-767.
[45] A. Hyvärinen, J. Karhunen, and E. Oja, Independent Component Analysis. New York: John Wiley & Sons, 2001.
[46] S. A. Jantzi, K. W. Martin, and A. S. Sedra, “Quadrature bandpass delta-sigma modulation for digital radio,” IEEE J. Solid-
State Circuits, vol. 32, pp. 1935-1950, Dec. 1997.
[47] Y.-C. Jeng, “Perfect reconstruction of digital spectrum from nonuniformly sampled signals,” IEEE Trans. Instrum. Meas.,
vol. 46, pp. 649-652, June 1997.
[48] H. R. Karimi and B. Friedrichs, “Wideband digital receivers for multi-standard software radios,” in IEE Coll. Adaptable
Multi-Standard Mobile Radio Terminals, London, UK, Mar. 1998, pp. 5/1-5/7.
[49] T. E. Kolding, “Multi-standard mixed-signal transceivers for wireless communications – A research overview,” Department
of Communication Technology, Aalborg University, Aalborg, Denmark, Tech. Rep. R-96-1005, Dec. 1996.
[50] T. I. Laakso, V. Välimäki, M. Karjalainen, and U. K. Laine, “Splitting the unit delay,” IEEE Signal Processing Mag., vol. 13,
pp. 30-60, Jan. 1996.

127 (131)

[51] E. A. Lee and D. G. Messerschmitt, Digital Communication. Boston, MA: Kluwer Academic Publishers, 1988.
[52] J. P. Y. Lee, “Wideband I/Q demodulators: Measurement technique and matching characteristics,” IEE Proc. Radar, Sonar,
Navigation, vol. 143, pp. 300-306, Oct. 1996.
[53] Y.-P. Lin and P. P. Vaidyanathan, “Periodically nonuniform sampling of bandpass signals,” IEEE Trans. Circuits Syst. II,
vol. 45, pp. 340-351, Mar. 1998.
[54] A. Lohtia, P. A. Gould, and C. G. Englefield, “An adaptive digital technique for compensating for analog quadrature
modulator / demodulator impairments,” in Proc. IEEE Pacific Rim Conf. Commun., Comput., Signal Processing, Victoria,
BC, Canada, May 1993, pp. 447-450.
[55] W. W. Lu, “Compact multidimensional broadband wireless: The convergence of wireless mobile and access,” IEEE
Commun. Mag., vol. 38, pp. 119-123, Nov. 2000.
[56] H. Meyr, M. Moeneclay, and S. A. Fechtel, Digital Communication Receivers. New York: John Wiley & Sons, 1998.
[57] S. Mirabbasi and K. Martin, “Classical and modern receiver architectures,” IEEE Commun. Mag., vol. 38, pp. 132-139, Nov.
2000.
[58] R. L. Mitchell, “Creating complex signal samples from a band-limited real signal,” IEEE Trans. Aerosp. Electron. Syst., vol.
25, pp. 425-427, May 1989.
[59] J. Mitola, Software Radio Architecture: Object-Oriented Approaches to Wireless Systems Engineering. New York: John
Wiley & Sons, 2000.
[60] H. Nguyen, “Improving QPSK demodulator performance for quadrature receiver with information from amplitude and phase
imbalance correction,” in Proc. IEEE Wireless Commun. Networking Conf., Chicago, IL, USA, Sept. 2000, pp. 1440-1444.
[61] T. Okanobu, H. Tomiyama, and H. Arimoto, “Advanced low voltage single chip radio IC,” IEEE Trans. Consumer Electron.,
vol. 38, pp. 465-475, Aug. 1992.
[62] J. M. Paez-Borrallo and F. J. Casajus Quiros, “Self adjusting digital image rejection receiver for mobile communications,” in
Proc. IEEE Veh. Technol. Conf., Phoenix, AZ, USA, May 1997, pp. 686-690.
Proceeding of the SDR 03 Technical Conference and Product Exposition. Copyright © 2003 SDR Forum. All Rights Reserved

128 (131)
[63] J. W. Pierre and D. R. Fuhrmann, “Considerations in the autocalibration of quadrature receivers,” in Proc. IEEE Int. Conf.
Acoust., Speech, Signal Processing, Detroit, MI, USA, May 1995, pp. 1900-1903.
[64] R. Porat and f. harris, “Resolving and correcting gain and phase mismatch in transmitters and receivers for wideband ofdm
systems,” in Proc. Asilomar Conf. Signals, Syst., Computers, Pacific Grove, CA, Nov. 2002, pp. 1005-1009.
[65] K.-P. Pun, J. E. Franca, and C. Azeredo-Leme, “Wideband digital correction of I and Q mismatch in quadrature radio
receivers,” in Proc. IEEE Int. Symp. Circuits Syst., Geneva, Switzerland, May 2000, pp. 661-664.
[66] C. M. Rader, “A simple method for sampling in-phase and quadrature components,” IEEE Trans. Aerosp. Electron. Syst., vol.
20, pp. 821-824, Nov. 1984.
[67] T. S. Rappaport, Wireless Communications: Principles and Practice. Upper Saddle River, NJ: Prentice-Hall, 1996.
[68] B. Razavi, “Design considerations for direct-conversion receivers,” IEEE Trans. Circuits Syst. II, vol. 44, pp. 428-435, June
1997.
[69] B. Razavi, “RF IC design challenges,” in Proc. Design Autom. Conf., San Francisco, CA, USA, June 1998, pp. 408-413.
[70] U. L. Rohde, J. Whitaker, and T. T. N. Bucher, Communications Receivers: Principles and Design, 2nd ed. New York:
McGraw-Hill, 1997.
[71] S. J. Roome, “Analysis of quadrature detectors using complex envelope notation,” IEE Proc. Radar and Signal Processing,
vol. 136, pp. 95-100, Apr. 1989.
[72] J. C. Rudell et al., “A 1.9-GHz wide-band IF double conversion CMOS receiver for cordless telephone applications,” IEEE J.
Solid-State Circuits, vol. 32, pp. 2071-2088, Dec. 1997.
[73] J. Sevenhans, B. Verstraeten, and S. Taraborrelli, “Trends in silicon radio large scale integration: Zero IF receiver! Zero I&Q
transmitter! Zero discrete passives!,” IEEE Commun. Mag., vol. 38, pp. 142-147, Jan. 2000.
[74] D. K. Shaeffer and T. H. Lee, The Design and Implementation of Low-Power CMOS Radio Receivers. Boston, MA: Kluwer
Academic Publishers, 1999.

129 (131)

[75] E. Song, S.-I. Chae, and W. Kim, “A 2 GHz CMOS double conversion downconverter with robust image rejection
performance against the process and temperature variations,” in IEEE Symp. VLSI Circuits Dig. Tech. Papers, Honolulu, HI,
USA, June 2000, pp. 38-41.
[76] M. S. J. Steyaert and R. Roovers, “A 1-GHz single-chip quadrature modulator,” IEEE J. Solid-State Circuits, vol. 27, pp.
1194-1197, Aug. 1992.
[77] M. S. J. Steyaert et al., “A 2-V CMOS cellular transceiver front-end,” IEEE J. Solid-State Circuits, vol. 35, pp. 1895-1907,
Dec. 2000.
[78] J. Tsui, Digital Techniques for Wideband Receivers. Norwood, MA: Artech House, 1995.
[79] H. Tsurumi and Y. Suzuki, “Broadband RF stage architecture for software-defined radio in handheld terminal applications,”
IEEE Commun. Mag., vol. 37, pp. 90-95, Feb. 1999.
[80] H. Tsurumi et al., “Broadband and flexible receiver architecture for software defined radio terminal using direct conversion
and low-IF principle,” IEICE Trans. Commun., vol. E83-B, pp. 1246-1253, June 2000.
[81] M. Valkama, Advanced I/Q Signal Processing for Wideband Receivers: Models and Algortihms. Ph.D. Dissertation,
Tampere, Finland: Tampere University of Technology, Dec. 2001.
[82] M. Valkama, M. Renfors, and V. Koivunen, “Advanced methods for I/Q imbalance compensation in communication
receivers,” IEEE Trans. Signal Processing, vol. 49, pp. 2335-2344, Oct. 2001.
[83] M. Valkama, J. Pirskanen, and M. Renfors, “Signal processing challenges for applying software radio principles in future
wireless terminals: An overview,” Int. J. Commun. Syst., vol. 15, pp. 741-769, Oct. 2002.
[84] M. Valkama and M. Renfors, “A novel image rejection architecture for quadrature radio receivers,” IEEE Trans. Circuits
Syst. II, to appear.
[85] M. Valkama and M. Renfors, “Advanced DSP for I/Q imbalance compensation in a low-IF receiver,” in Proc. IEEE Int.
Conf. Communications, New Orleans, LA, June 2000, pp. 768-772.

130 (131)
[86] M. Valkama, M. Renfors, and V. Koivunen, “Blind source separation based I/Q imbalance compensation,” in Proc. IEEE
Symposium 2000 on Adaptive Systems for Signal Processing, Communications and Control, Lake Louise, AL, Canada, Oct.
2000, pp. 310-314.
[87] M. Valkama, M. Renfors, and V. Koivunen, “Compensation of frequency-selective I/Q imbalances in wideband receivers:
Models and algorithms,” in Proc. Third IEEE Signal Processing Workshop Signal Processing Adv. Wireless Commun.,
Taoyuan, Taiwan, R.O.C., March 2001, pp. 42-45.
[88] M. Valkama and M. Renfors, “Second-order sampling of wideband signals,” in Proc. IEEE Int. Symp. Circuits Syst., Sydney,
Australia, May 2001, pp. 801-804.
[89] R. G. Vaughan, N. L. Scott, and D. R. White, “The theory of bandpass sampling,” IEEE Trans. Signal Processing, vol. 39,
pp. 1973-1984, Sept. 1991.
[90] J. A. Weppman, “Analog-to-digital converters and their applications in radio receivers,” IEEE Commun. Mag., vol. 33, pp.
39-45, May 1995.
[91] S. Wu and B. Razavi, “A 900-MHz/1.8-GHz CMOS receiver for dual-band applications,” IEEE J. Solid-State Circuits, vol.
33, pp. 2178-2185, Dec. 1998.
[92] D. T. S. Yim and C. C. Ling, “A 200-MHz CMOS I/Q downconverter,” IEEE Trans. Circuits Syst. II, vol. 46, pp. 808-810,
June 1999.
[93] L. Yu and W. M. Snelgrove, “A novel adaptive mismatch cancellation system for quadrature IF radio receivers,” IEEE Trans.
Circuits Syst. II, vol. 46, pp. 789-801, June 1999.
[94] K. C. Zangi and R. D. Koilpillai, “Software radio issues in cellular base stations,” IEEE J. Select. Areas Commun., vol. 17,
pp. 561-573, Apr. 1999.
[95] M. Zeng, A. Annamalai, and V. K. Bhargava, “Recent advances in cellular wireless Communications,” IEEE Commun. Mag.,
vol. 37, pp. 128-138, Sept. 1999.

131 (131)

Spacecraft Lithium-Ion Battery Power Systems (Thomas P. Barrera)
50% (2)
Spacecraft Lithium-Ion Battery Power Systems (Thomas P. Barrera)
339 pages
Elektor 1984 05
No ratings yet
Elektor 1984 05
58 pages
A Pipelined Multi-Core MIPS Machine Hardware Implementation and Correctness Proof (Mikhail Kovalev, Silvia Melitta Muller, Wolfgang J. Paul)
No ratings yet
A Pipelined Multi-Core MIPS Machine Hardware Implementation and Correctness Proof (Mikhail Kovalev, Silvia Melitta Muller, Wolfgang J. Paul)
359 pages
M1623 Radar
No ratings yet
M1623 Radar
52 pages
Active RF and Microwave Devices
100% (1)
Active RF and Microwave Devices
4 pages
Yaesu Vx-3r Service Manual
100% (1)
Yaesu Vx-3r Service Manual
44 pages
2 Complex Signals and Radios Short 2pp
No ratings yet
2 Complex Signals and Radios Short 2pp
32 pages
Sampling and Multirate Techniques For Complex and Bandpass Signals
No ratings yet
Sampling and Multirate Techniques For Complex and Bandpass Signals
58 pages
1.1 The Communication Process: Source Information User
No ratings yet
1.1 The Communication Process: Source Information User
14 pages
Hilbert Transform, SSB, VSB Systems From Simon Haykin 3rd Edition
No ratings yet
Hilbert Transform, SSB, VSB Systems From Simon Haykin 3rd Edition
25 pages
ee321_lec1
No ratings yet
ee321_lec1
39 pages
Complex Signals
No ratings yet
Complex Signals
17 pages
DSP LAB-Experiment 4: Ashwin Prasad-B100164EC (Batch A2) 18th September 2013
No ratings yet
DSP LAB-Experiment 4: Ashwin Prasad-B100164EC (Batch A2) 18th September 2013
9 pages
Representation of Signals & Systems: 1. Hilbert Transform
No ratings yet
Representation of Signals & Systems: 1. Hilbert Transform
13 pages
SSB RCV Signals
No ratings yet
SSB RCV Signals
9 pages
Notes 3
No ratings yet
Notes 3
3 pages
Formulas Esism
No ratings yet
Formulas Esism
6 pages
World's Largest Science, Technology & Medicine Open Access Book Publisher
No ratings yet
World's Largest Science, Technology & Medicine Open Access Book Publisher
34 pages
E416 L4
No ratings yet
E416 L4
28 pages
Complex Baseband Representation: Saravanan Vijayakumaran Sarva@ee - Iitb.ac - in
100% (1)
Complex Baseband Representation: Saravanan Vijayakumaran Sarva@ee - Iitb.ac - in
20 pages
Band Pass Systems, Phasors and Complex Representation of Systems
No ratings yet
Band Pass Systems, Phasors and Complex Representation of Systems
23 pages
DSP Lecture 2
No ratings yet
DSP Lecture 2
77 pages
W236: T D H T: A B T: HE Iscrete Ilbert Ransform Rief Utorial
No ratings yet
W236: T D H T: A B T: HE Iscrete Ilbert Ransform Rief Utorial
16 pages
Signal Representation
No ratings yet
Signal Representation
13 pages
Lecture 4 Introduction To Analog and Digital Communications, S. Haykin, M. Moher, 2nd
No ratings yet
Lecture 4 Introduction To Analog and Digital Communications, S. Haykin, M. Moher, 2nd
10 pages
Ramesh Babu PDF
No ratings yet
Ramesh Babu PDF
303 pages
Ramesh Babu
No ratings yet
Ramesh Babu
303 pages
Complexbaseband Representation of Bandpass Signals
No ratings yet
Complexbaseband Representation of Bandpass Signals
18 pages
2 Band Pass Signals
No ratings yet
2 Band Pass Signals
23 pages
TLT-5400/5406 DIGITAL TRANSMISSION, 3rd Matlab-Exercise: Mapping Transmit Filter R
No ratings yet
TLT-5400/5406 DIGITAL TRANSMISSION, 3rd Matlab-Exercise: Mapping Transmit Filter R
3 pages
Signal Processing and Correlation Techniques: Preamble
No ratings yet
Signal Processing and Correlation Techniques: Preamble
15 pages
Complexbaseband Representation of Bandpass Signals
No ratings yet
Complexbaseband Representation of Bandpass Signals
18 pages
Complexbaseband Representation of Bandpass Signals
No ratings yet
Complexbaseband Representation of Bandpass Signals
18 pages
Control The World by Adding and Multiplying: Andres Kwasinski
No ratings yet
Control The World by Adding and Multiplying: Andres Kwasinski
52 pages
M8 - Discrete Time Fourier Transform
No ratings yet
M8 - Discrete Time Fourier Transform
30 pages
EE501 Adaptive Filter Design: Instructor: Dr. Farhan Khalid
No ratings yet
EE501 Adaptive Filter Design: Instructor: Dr. Farhan Khalid
26 pages
Signal Processing: Understanding Digital
No ratings yet
Signal Processing: Understanding Digital
7 pages
Lec31&32
No ratings yet
Lec31&32
28 pages
Chap4 Lec1
No ratings yet
Chap4 Lec1
24 pages
Advanced Signals and Systems - Hilbert Transform: Gerhard Schmidt
No ratings yet
Advanced Signals and Systems - Hilbert Transform: Gerhard Schmidt
24 pages
Topic1 LTI PDF
No ratings yet
Topic1 LTI PDF
221 pages
Lec 17
No ratings yet
Lec 17
21 pages
Frequency Response and Continuous-Time Fourier Transform
No ratings yet
Frequency Response and Continuous-Time Fourier Transform
25 pages
HilbertTransform & Bandpass
No ratings yet
HilbertTransform & Bandpass
13 pages
Networks and Systems Profvgkmurti Department of Electrical Engineering Indian Institute of Technology, Madras Lecture - 03 Introductory Concepts
No ratings yet
Networks and Systems Profvgkmurti Department of Electrical Engineering Indian Institute of Technology, Madras Lecture - 03 Introductory Concepts
22 pages
1 Discrete-Time Fourier Transform (DTFT) : Indian Institute of Technology Bombay
No ratings yet
1 Discrete-Time Fourier Transform (DTFT) : Indian Institute of Technology Bombay
14 pages
Lecture 2
No ratings yet
Lecture 2
33 pages
ee2000_ch4_fourier_transform
No ratings yet
ee2000_ch4_fourier_transform
48 pages
Digital Signal Processing - Lecture 1_ Introduction
No ratings yet
Digital Signal Processing - Lecture 1_ Introduction
69 pages
report_SP (2)
No ratings yet
report_SP (2)
18 pages
Basicsofdigitalfilters
No ratings yet
Basicsofdigitalfilters
65 pages
ECE 414 Tutorial 1: Review: Random Process Fourier Transform Ouetaso
No ratings yet
ECE 414 Tutorial 1: Review: Random Process Fourier Transform Ouetaso
16 pages
Lecture 1-5 EEEN323 March2022
No ratings yet
Lecture 1-5 EEEN323 March2022
139 pages
Lecture3-SystemsChannelsFSFT
No ratings yet
Lecture3-SystemsChannelsFSFT
17 pages
(eBook PDF) Digital Signal Processing First, eBook, Global Edition - Get the ebook in PDF format for a complete experience
100% (1)
(eBook PDF) Digital Signal Processing First, eBook, Global Edition - Get the ebook in PDF format for a complete experience
57 pages
Lecture 1 Introduction
No ratings yet
Lecture 1 Introduction
39 pages
Doc11-B14-dspcourse-lec1
No ratings yet
Doc11-B14-dspcourse-lec1
33 pages
Formula Table Digital Signal Processing
No ratings yet
Formula Table Digital Signal Processing
43 pages
Ece113 Lec04 Mixers
No ratings yet
Ece113 Lec04 Mixers
45 pages
Course 01 - Introduction
No ratings yet
Course 01 - Introduction
56 pages
Analog Dialogue, Volume 47, Number 2
From Everand
Analog Dialogue, Volume 47, Number 2
Analog Dialogue
No ratings yet
Student Solutions Manual to Accompany Economic Dynamics in Discrete Time, secondedition
From Everand
Student Solutions Manual to Accompany Economic Dynamics in Discrete Time, secondedition
Yue Jiang
4.5/5 (2)
Loop-shaping Robust Control
From Everand
Loop-shaping Robust Control
Philippe Feyel
No ratings yet
Laplace Transforms Essentials
From Everand
Laplace Transforms Essentials
Morteza Shafii-Mousavi
3.5/5 (3)
Lectures on Integral Equations
From Everand
Lectures on Integral Equations
Harold Widom
3.5/5 (1)
Multiple Integrals, A Collection of Solved Problems
From Everand
Multiple Integrals, A Collection of Solved Problems
Steven Tan
No ratings yet
A Glossary of the Tribes and Castes of the Punjab and North-West Frontier Province
No ratings yet
A Glossary of the Tribes and Castes of the Punjab and North-West Frontier Province
624 pages
Canals, Colonies and Class
No ratings yet
Canals, Colonies and Class
342 pages
Islam, Immigration & Identity
No ratings yet
Islam, Immigration & Identity
275 pages
Absorption of Beta and Gamma Radiation Lab
No ratings yet
Absorption of Beta and Gamma Radiation Lab
36 pages
Grading System of Danish General Upper Secondary Education
No ratings yet
Grading System of Danish General Upper Secondary Education
8 pages
Partition and Locality Violence, Migration, and Development in Gujranwala and Sialkot, 1947-1961 (Dr. Ilyas Chattha)
No ratings yet
Partition and Locality Violence, Migration, and Development in Gujranwala and Sialkot, 1947-1961 (Dr. Ilyas Chattha)
321 pages
Punjab - A History From Aurangzeb To Mountbatten (Rajmohan Gandhi)
No ratings yet
Punjab - A History From Aurangzeb To Mountbatten (Rajmohan Gandhi)
728 pages
A PRACH Preamble Generation and Detectio
No ratings yet
A PRACH Preamble Generation and Detectio
8 pages
CC1000 Data Sheet 2 2
No ratings yet
CC1000 Data Sheet 2 2
50 pages
Receiver
No ratings yet
Receiver
9 pages
HSMS 2825 TR1G Avago Datasheet 10022522
No ratings yet
HSMS 2825 TR1G Avago Datasheet 10022522
5 pages
How To Create Your Signal Jammer
No ratings yet
How To Create Your Signal Jammer
6 pages
Service Manual: Uhf Transceiver
No ratings yet
Service Manual: Uhf Transceiver
40 pages
Image Rejecting Front-End For GSM Applications UAA2072M: Philips Semiconductors Product Specification
No ratings yet
Image Rejecting Front-End For GSM Applications UAA2072M: Philips Semiconductors Product Specification
16 pages
GBPPR 'Zine - Issue #63
No ratings yet
GBPPR 'Zine - Issue #63
78 pages
Intelligent Block Upconverter
No ratings yet
Intelligent Block Upconverter
196 pages
Lecture 27: Mixers. Gilbert Cell: Mixers Shift The Frequency Spectrum of An Input Signal
No ratings yet
Lecture 27: Mixers. Gilbert Cell: Mixers Shift The Frequency Spectrum of An Input Signal
0 pages
IC 756pro Serv
No ratings yet
IC 756pro Serv
93 pages
Icf 310
No ratings yet
Icf 310
42 pages
Get Electronics 2nd Edition B. V. Narayana Rao Free All Chapters
100% (14)
Get Electronics 2nd Edition B. V. Narayana Rao Free All Chapters
57 pages
EICO 324 Signal Generator User Manual
83% (6)
EICO 324 Signal Generator User Manual
15 pages
Mixer Design
No ratings yet
Mixer Design
42 pages
Illustrated International Ham Radio Dictionary 2019
No ratings yet
Illustrated International Ham Radio Dictionary 2019
260 pages
Microelectronics
No ratings yet
Microelectronics
4 pages
Mud 2 K
No ratings yet
Mud 2 K
10 pages
Detector de Metales
No ratings yet
Detector de Metales
7 pages
Sebastian Hoyos: ELEN 689:special Topics Advanced Mixed-Signal Interfaces
No ratings yet
Sebastian Hoyos: ELEN 689:special Topics Advanced Mixed-Signal Interfaces
59 pages
08-An Integrated 100-GHz FMCW Imaging Radar For Low-Cost Drywall Inspection
No ratings yet
08-An Integrated 100-GHz FMCW Imaging Radar For Low-Cost Drywall Inspection
15 pages
Sub Harmonic Mixer Design With Ansoft Designer
No ratings yet
Sub Harmonic Mixer Design With Ansoft Designer
35 pages
Instruction Manual - PMC Mixing Machine
No ratings yet
Instruction Manual - PMC Mixing Machine
9 pages
Satellite Transponder
No ratings yet
Satellite Transponder
8 pages
Radio Frequency and Wireless Communications
No ratings yet
Radio Frequency and Wireless Communications
12 pages
RF Prime Microwave: Surface Mount Mixer Data Sheet Lrfms-2-10 5 To 1000 MHZ
No ratings yet
RF Prime Microwave: Surface Mount Mixer Data Sheet Lrfms-2-10 5 To 1000 MHZ
1 page

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.

Complex Modulation Basics

Uploaded by

Complex Modulation Basics

Uploaded by

Advanced I/Q Signal Processing for

Tampere University of Technology

SDR-03 Technical Conference, Nov. 2003, Orlando, FL

1. Complex Signals and Systems

2. Sampling and Multirate DSP with Bandpass and I/Q signals

3. I/Q Mismatch Problems in Analog I/Q Signal Processing

5. Second-Order Sampling and its Enhancements

6. I/Q Signal Processing in Frequency Synthesizers

SDR-03 Technical Conference, Nov. 2003, Orlando, FL

Basic Concepts and Definitions

Analytic Signals and Hilbert Transforms

– This idea of using Hilbert transform to generate analytic signals is further

Hilbert Transformer (HT)

Hilbert Transformer (HT)

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

Frequency Translations and Mixing

input spectrum output spectrum

input Re[·] output

input LPF output

Complex Signals and Sampling

– Conclusion: It doesn’t matter whether the signal is real or complex or whether it

– So in general the traditional statement “signal should be sampled at least at rate

→ these kind of techniques are generally referred to as subsampling

••• ••• ••• •••

– Based on the above figure, it is easy to formulate the regions of allowable

••• ••• ••• •••

••• ••• ••• •••

MULTIRATE PROCESSING OF BANDPASS AND I/Q SIGNALS

Interpolation or Up-Sampling with Complex Signals

EFFICIENT POLYPHASE STRUCTURES

– Example: Two-phase decomposition of H(z)=h(0)+h(1)z–1 +…+h(14)z–14 is

h0(n)={h(0), h(2), h(4), h(6), h(8), h(10), h(12), h(14)}

h1(n)={h(1), h(3), h(5), h(7), h(9), h(11), h(13)} =

– Example (cont’d): Down-sampling by L=2.

– Similar reasoning can be used also in the interpolation case.

– One interesting interpretation of the previous polyphase structures is related to

– So as stated before, the modulating exponentials in the polyphase filters reduce

Figure. General polyphase decimator.

Figure. Demonstration set-up, L = 8.

Phase delays of the polyphase filters

Real bandpass signal and the analytic channel filter, k = 2

Real bandpass signal and the analytic channel filter, k = 3

 g1e jφ1 + g 2 e jφ2  jω 0t  g1e − jφ1 − g 2 e − jφ2  − jω 0t

Then, the Fourier transform of Z ′( f ) = H I ( f ) Z I ( f ) + jH Q ( f ) Z Q ( f ) can be

Imbalance Effects in Receiver Front-Ends

xLO,I (t) = cos(2π fLO t)

xQ (t) zQ' (t)

xLO,Q (t) = –gsin(2π fLO t + φ)

– Now, to explicitly characterize the imbalance effects on the individual

– Possible solutions to enhance the image attenuation:

– After some manipulation, the frequency domain expression for these

Adaptive Interference Cancellation (IC) Based Compensation

sˆIC (n) = d (n) − ∑kN=IC0 wk (n)v(n − k )

where x(n) = [d(n) v(n)]T.

– The time delay of ∆T = 1/(4fC) corresponds to a frequency-dependent phase

input spectrum input spectrum

WITH HILBERT TRANSFORMER

WITH HILBERT TRANSFORMER WITH DELAY

••• ••• ••• •••

(.)* C(e jω)

– Then, to completely cancel the image signal components, the compensator

– This solution, referred to as the "zero forcing" –solution C ZF ( f ) , is given by

– dash-dotted: basic second-order sampling (L2)

– To analyze the image attenuation of the complete structure, we proceed as

– Previous design example:

– General conclusion: Based on the presented design examples, image

– dash-dotted: basic second-order sampling (L2)

{a11 , a12 , a21 , a22}

Digital A(t) Envelope

Practical Compensator Update

A(t ) = | xLP (t ) | = | αe jω NCOt + βe − jω NCOt |= | α |2 + | β |2 +2 Re[αβ *e j 2ω NCOt ]

– Thus, instead of estimating g and ψ, a more simple approach is to directly

for n odd with K12 ∈{+1,–1} and

aˆ12 (n) = aˆ12 (n − 1)

– To illustrate the ability to perform the compensation also in time-varying

You might also like