Journal of Signal and Information Processing, 20 11 , 2, 18 - 25
doi:10.4236/jsip.2011.21003 Published Online February 2011 (http://www.SciRP.org/journal/jsip)
A Novel Approach to Blind I/Q Mismatch &
Carrier Offset Compensation
Sunita Panda1, S. P. Panigrahi2, D. D. Meena2, R. K. Mishra2, M. Sahu2, H. Sahu2, N. Panda2, M.
Singh2
1ECE, KIT, Berhampur, India; 2Electrical Engineering, KIST, Bhubaneswar, India.
Received December 10th, 2010; revised January 11th, 2011; accepted February 18th, 2011
ABSTRACT
This paper proposes a novel approach for I/Q mismatch compensation. Simulation results proves the superior per-
formance of the proposed technique.
Keywords: Channel Equalization, I/Q Mismatch compensation
1. Introduction
In order to increase the receiver flexibility while also
emphasizing the receiver integrability and other imple-
mentation related aspects, the design of radio receivers is
no longer dominated by the traditional superheterodyne
architecture. Instead, alternative receiver structures, like
the direct conversion [1,2] and low-IF [1,3,4] architec-
tures, are receiving more and more interest. The analog
front-end of these types of receivers is partially based on
complex or I/Q signal processing [5-7]. More specifically,
the frequency translation from radio frequencies (RF)
closer to base-band is carried out using I/Q downconver-
sion. Since, in theory, the I/Q downconversion corre-
sponds to a pure frequency translation, the fundamental
image signal problem is basically avoided during the
downconversion. In thismanner, the requirements for RF
image rejection filtering are greatly relaxed in practice
[1-4].
The assumption of infinite image signal rejection dur-
ing the downconversion is strictly valid only if ampli-
tudes and phases of the analog front-end I and Q
branches are perfectly matched [7]. In practice, this is not
the case. Some mismatch or imbalance will always exist
due to imperfections of practical analog electronics. Am-
plitude imbalances of 1% - 5% and phase imbalances of
1-5 represent feasible design Figures [1-4]. This corre-
sponds to approximately 25-40 dB image signal attenua-
tion. These levels of image attenuation are clearly insuf-
ficient in low-IF-type receivers where the image band
can carry a signal with much higher power level than the
desired signal. Some digital signal processing (DSP)-
based approaches to improve this image attenuation in IF
receivers are presented, for example, in [7-10]. In di-
rect-conversion receivers, the image signal is inherently a
self-image (the desired signal itself at negative frequen-
cies), and the analog front-end image attenuation might
be sufficient with low-order modulations.
However, with higher-order modulations, such as 16-
or 64-QAM, the distortion due to self-image cannot be
neglected and again some kind of compensation is
needed [11-13]. This is also one of the topics of this pa-
per. The idea in this paper is first to show that I/Q mis-
match causes crosstalk between the I and Q components.
Then this crosstalk or mixing of the I and Q is removed
using blind signal separation (BSS) techniques [14-16].
Compared to the other available solutions in the literature
[12,13], the proposed concept is especially attractive
since no known training signals are needed. Also the
ability to follow possible time dependencies in the mis-
match parameters (due to, e.g., temperature changes) is
another highly desirable feature.
Another challenging practical problem in radio com-
munications is carrier synchronization [6]. In practice, it
is unlikely that the frequency and relative phase of the
receiver local oscillators exactly match those of the re-
ceived carrier. In case of linear modulations, a constant
phase offset introduces a constant rotation to the received
constellation, which needs to be compensated unless dif-
ferential phase modulation is used. Even a bigger prob-
lem is caused by errors or offsets in frequency that basi-
cally results in time-varying rotation of the constellation.

A Novel Approach to Blind I/Q Mismatch & Carrier Offset Compensation
19
This is obviously unacceptable for most modulation
types and needs to be efficiently compensated for. In this
paper, the carrier offsets are shown to result in time-
varying leakage between the I and Q, and carrier tracking
is implemented with adaptive signal separation methods.
Furthermore, when combined with the previous I/Q mis-
match compensation, a single BSS stage can accomplish
both tasks jointly in a blind manner. These kinds of ap-
proaches have not been considered in the literature so far.
In addition to the above front-end-related issues, the
distortion due to transmission channel [6,15] is inevitable
in any radio receiver and needs to be addressed with care.
As will be shown, a general bandpass channel can be
viewed to cause frequency-selective crosstalk between
the transmitted I and Q data. Based on this, convolutive
mixture (or FIR-MIMO) separation techniques working
on the observed I and Q signals are applied to implement
channel equalization.
2. Direct-Conversion Receiver
2.1. Background
The fundamental tasks to be carried out in the front-end
of any communications receiver include 1) amplification
of the attenuated desired signal, 2) downconversion of
the desired signal spectrum from around the RF carrier
frequency closer to base-band, 3) attenuating the un-
wanted spectral components appearing in the antenna
signal, and 4) synchronizing the receiver local oscillators
for downconversion and sampling with the received sig-
nal. Key tasks of the baseband digital signal processing
include channel estimation, equalization, and detection.
In traditional receiver designs, these tasks are imple-
mented more or less independently of each other, aiming
at close-to-ideal operation in each signal processing stage.
This has lead to the use of the superheterodyne receiver
architecture [1,4] in order to meet the tight RF specifica-
tions in terms of image band attenuation and nonlinear
effects of the receiver front-end stages. In such receivers,
the RF signal is first downconverted to a fixed interme-
diate frequency (IF) where the receiver selectivity is im-
plemented using a fixed IF filter. The down-conversion is
usually done with a simple real mixer, and the spectral
components on the image signal band need to be attenu-
ated sufficiently by the RF stages before the mixer. Due
to the high number of discrete components and high
power consumption, the superheterodyne architecture is,
however, not the most appropriate choice for highly in-
tegrated implementations [1-4].
Furthermore, the use of fixed discrete components in
the analog front-end limits the receiver flexibility. Thus,
architectures with more simplified analog front-ends with
less RF processing are generally needed. In addition, it
has recently been demonstrated [7-9] that various
non-idealities and distortion effects due to the simplifica-
tion of the analog front-end can in general be compen-
sated by advanced DSP techniques. This is also the cen-
tral theme in this paper.
2.2. Direct-Conversion Architecture
A simple way to reduce the number of components in the
receiver and alleviate the problem of receiver complexity
is to avoid the use of intermediate frequency and quad-
rature down-convert the desired channel signal directly
from RF to baseband. Complete elimination of the IF
stage results in highly simplified structure, the so-called
direct-conversion receiver, where most of the channel
selectivity and amplification are implemented at base-
band [1,2]. Firstly, since most of the signal processing
tasks take place at low frequencies, the power consump-
tion is minimized. On the other part, very low-noise op-
eration is called for in all the remaining analog compo-
nents since the amplificatio n provided by the RF stag e is
only moderate. The direct-conversion receiver concept is
depicted in Figure 1.
Figure 1. The direct-conversion receiver architecture. The leading principle in this paper is to show that various receiver
signal processing tasks can be carried out blindly by forcing the observed I and Q signals as independent as possible using
blind signal separation.

Current Distortion Evaluation in Traction 4Q Constant Switching Frequency Converters 20
In the direct-conversion principle, since the IF is ef-
fectively zero, the image signal is actually the desired
signal itself (at negative center frequency). Ideally, with
perfect analog processing, the image band is completely
attenuated in general. However, practical analog imple-
mentations of the needed I/Q signal processing have
mismatches in the amplitude and phase responses of the I
and Q branches, leading to finite atten uation of th e image
band signal. In the direct-conversion case, the effect of
imperfect (self-)image rejection is seen as a linear trans-
formation of the original signal constellation [11-13]. As
a result, the image attenuation requirements are not ex-
tremely tight, especially if low-order modulations are
used. In effect, the 25-40 dB image attenuation of a prac-
tical analog front-end can be sufficient with low-orde
modulations.Wi th higher-order spectrally efficient modu-
lation methods, however, the distortion due to selfimage
can establish an error floor and needs to be compensated.
In practice, the use of zero IF introduces also some
other problems. The major drawback in direct-conversion
principle is the DC offset problem [1,2]. Due to zero IF,
the local oscillator frequency is o n the same frequency as
the desired chann el. Then, if the LO signal leaks into the
mixer input port, it self-mixes down to baseband causing
interfering signal components at zero frequency. These
are called DC offsets and can be orders of magnitude
larger than the desired channel signal. Besides the LO
leakage, another contributor to the offset problems is 1/f
noise of the active front-end components. For a better
receiver performance, some compensation of the DC
offsets is needed. Another analog RF-related problem is
that higher linearity is required because in a direct-con-
version receiver, second-order inter-modulation products
may fall in the signal band (in superheterodynes, the
weaker third-order intermodulation products usually set
the linearity requirements). These problems have limited
the use of direct-conversion principle in practical systems
earlier, but nowadays this approach is widely utilized in
mobile terminals. However, the introduction of higher-
order modulations in future wireless communication sys-
tems sets higher demands for the receiver performance,
and the I/Q imbalance effects, for example, to be dis-
cussed in Section 4 are likely to pose big challenges.
3. Blind Signal Separation
Currently, more and more applications call for proper
representation of multivariate data. An exten sion of prin-
cipal component analysis (PCA) called independent
component analysis (ICA) is one good example of such
techniques [16]. The ICA and its signal processing ap-
plication, blind signal separation [14,16], is applied in
this paper to communications receiver signal processing.
The purpose of this section is to introduce the basic con -
cepts and notations after which the actual applications to
I/Q mismatch and carrier offset compensation as well as
to channel equalization are discussed and analyzed in
Sections 4 and 5.
Generally speaking, blind signal separation deals with
the recovery of some interesting signals, called sources,
based on observing their linear1 mixtures, and falls under
the umbrella of multiple-input multiple-output (MIMO)
signal processing [14-16]. The term blind is used here to
emphasize the fact that no prior knowledge of the mixing
process or the temporal structure of the underlying source
signals is needed, but only the statistical properties are
utilized. The leading principle in this context is the as-
sumption of statistical independence of the original
sources [14-16]. Thus in practice the recovery consists of
forcing the separator output signals to be “as independent
as possible,” according to the selected independence-
measure. Commonly used approaches in this context are,
for example, nonlinear decorrelation and minimization of
mutual information. As is obvious, the relative order of
the recovered sources or the individual amplitude/power
levels cannot be blindly id entified.
In addition to statistical independence, another key
assumption is that only o ne of the sources (if any) can be
Gaussian [14-17]. This is easy to understand since for
Gaussian signals uncorrelatedness implies independence,
making it impossible to distinguish any (remaining) or-
thogonal/orthonormal transformation . Luckily most com-
munications signals are, indeed, non-Gaussian. In the
following, M and N denote the number of observed and
original source signals, respectively.
3.1. Instantaneous MIMO Models
Assuming the mixing process is instantaneous, the mth
observation, say


m
x
k is of the form


m
x
k





,1 1,mmNN
askas k, where,

n
x
k denotes the
nth source signal and ,1m arepresents the relative weight
of the source


n
s
k in the observation

n
x
k. Stack-
ing the observed and source signal samples at time-in-
stant k into column vectors
 
1,T N
s
ksksk


and


x
k

 
1,T N
xkx k




 respectively, the system
model can be simply written as:


x
kAsk (1)
where the M × N matrix A describes the mixing process
with


,
,mn mn
A
a. In general, the matrix A is assumed
unknown but nonsingular (full rank). This is a natural
assumption and is fundamental for the identifiability of
the model; see [14,16-18] for more details. Generally
speaking, the separator processes a sequence of the ob-
servation vectors and tries to “invert” the model in (1). In
adaptive separation, the parameters of the separator are

A Novel Approach to Blind I/Q Mismatch & Carrier Offset Compensation
21
updated iteratively which also enables tracking the pos-
sible time-variant features of the mixturemodel. The
separator output at time k is commonly written as
  
y
kWkxkWkAskTksk  (2)
Where,

Wk denotes the N × M separator matrix,
 
Tk WkA is the total equivalent system matrix (N
× N), and
 
1,T N
yky kyk

. For successful
separation, T(k) should converge to a “quasi-identity”
(permutation and scaling) matrix with only one nonzero
element on each of its row and column.
Various algorithms, with varying computational com-
plexity and performance, to determine the separator co-
efficients exist in the literature; see [14,16] for excellent
reviews.
One exciting feature of th e separation stage is the pos-
sibility for uniform performance. This being the case, the
separation performance is independent of the underlying
mixture coefficients (i.e ., the matrix A, as long as it is full
rank) and depends only on the source statistics. One such
algorithm with this desirable property is the so-called
equivariant adaptive separation via independence (EASI),
proposed originally in [19], whose performance depends
only on the certain nonlinear moments of the source sig-
nals. The EASI algorithm consists formally of two sub-
tasks; a whitening (second order decorrelation) part and a
nonlinear decorrelation part where the selection of the
used nonlinear function depends on the source statistics.
The exact alg orithm descri ption can be found i n [19]. This
algorithm is used also in this paper in Section 4. Notice,
however, that this is done only to illustrate the principal
operation of the proposed receiver concepts; thus any
other adaptive separation approach could be tested and
used as well in practice.
3.2. Convolutive MIMO Models
A more general class of signal models is obtained if the
assumption of instantaneous mixing is dropped. In other
words, the mixing process can also containmemory and
thus be frequency selective [14,16,20]. A direct exten-
sion of the model in (1) results in



l
l
x
kAskl
 (3)
Which represents a MIMO convolution of the sequences
 
,1, 1,sk sksk and

101
,,,,AAA


with each l
A
being size M × N. In other words, each
observation

n
x
k is a convolutive mixture of the
original source signals. In this case, the identifiability
condition related to the mixing process is generally for-
mulated in terms of the corresponding system (MIMO)
transfer function

l
l
l
A
zAz

; the recovery of the
original source contributions is feasible if this system
transfer f unct ion has full r an k [14,16,20 ] .
Practical recovery of the source signals consists of
multichannel filtering of the observed vectors and can
generally be written as:


l
l
y
kWkxkl



(4)
where, the N × M separator matrices

l
Wkare adapted
to minimize the predetermined dependence measure be-
tween the components of y(k). In terms of system trans-
fer functions, the relation between the source signals and
the separator output signals is of the form:


,,Tzk WzkAa (5)
where, W(z, k) is the transfer function of the separator at
time k. For successful source recovery, the total system
T(z, k) should converge to a matrix with only one non-
zero element in each row and column.
As in case of instantaneous mixtures, also here a wide
variety of different algorithms for separator adaptation
exists in the literature, and some of them can be claimed
to have the uniform separation performance [14,16,20].
One example is the natural gradient-based approach de-
scribed in [20]. This algorithm is applied also in this pa-
per in Section 5 to the channel equalization problem in
terms of the I and Q signals.
4. Blind I/Q Mismatch & Carrier Offset
Compensation
In this section, the I/Q mismatch problem due to analog
front-end nonidealities as well as the carrier synchrona-
tion task are addressed in detail for direct-conversion
receivers. The basic idea is to show that both of these
practical problems can be viewed to create dependence
between the observed I and Q signals. Then, a signal
separation algorithm is applied to remove this depend-
ence and thus to recover the original I and Q data..
4.1. Signal Models and I/Q separation-based
Compensation
I/Qmismatch:
For analysis purposes, the received RF signal, say


rt, is written as:





   
Re exp
cos sin
c
IcQc
rtztj t
ztt ztt





 (6)
where,






IQ
ztz tjzt denotes the corresponding
ideal baseband equivalent of the desired channel to be
recovered by the receiver front-end. Now, to model the
amplitude and phase mismatches of the analog front-end,
the (complex) LO signal of the I/Q down- converter is
written as:

A Novel Approach to Blind I/Q Mismatch & Carrier Offset Compensation
22

 
 
12
cos sin
exp exp
LO cc
cc
xttjg t
K
jt Kjt




 
 (7)
 
12
1 exp1 exp
,
22
g
jgj
KK

  
 (8)
where, g and φ represent the relative amplitude and phase
mismatches, respectively. The latter form of (7) indicates
that two frequency translations take place due to mis-
matches. Indeed, the down-conversion of r(t) combined
with low-pass filtering results in:



12
*
x
tKztKzt (9)
and the corresponding self-image rejection ratio is:

22
10 12
10log KK.
To examine the mismatch effect from the I and Q sig-
nal point of view, the model in (9) can be written as
 
IQ
x
txtjxt , where,
 
 
cos sin
II
QQI
xt zt
x
tgztg zt 


(10)
In other words, I/Q mismatch tends to create depend-
ence between the I and Q signals. Assuming that the
original I and Q signals are statistically independent,
which holds, for example, for square QAM type of con-
stellations, these original I and Q components can be
recovered blindly using a signal separation algorithm.
Carrier offsets:
To see the explicit effect of carrier offsets more for-
mally, the I/Q down-converter LO signal is now written
as:







 

cos sin
exp
LO cc
c
xtt jt
jt

 

 
 
(11)
where, Δω and θ model the frequency and phase offsets,
respectively, relative to th e received sign al in (6). Now, it
is common to write the down-converted signal after low-
pass filtering as:
 


expxt ztjt

  (12)
Interestingly, when written in terms of the I and Q sig-
nals, the model in (12) can be expressed as:



 
cos sin
cos sin
II Q
QQ I
x
ttztjtzt
x
ttztjtzt
 
 
 
  (13)
Thus, from the I/Q point of view, the carrier offsets
correspond to time-varying mixing of the I and Q signals,
and an adaptive signal separation algorithm can be used
to track and remove this effect.
A combined signal model incorporating both the I/Q
mismatch and the carrier offset effects is given next. Us-
ing analysis similar to those given above, it is relatively
easy to show that observable signal after down-conver-
sion and low-pass filtering appears as:






 

1
2
exp
*exp
xt Kztjt
Kz tjt






(14)
Here it is naturally assumed that the frequency offset is
smaller than the guard band between the adjacent fre-
quency channels. Now, the complex signal in (14) corre-
sponds to an I/Q signal pair of the form:





 


 

cos sin
cos
sin
IIQ
QQ
I
x
ttzttzt
xt gtzt
gt zt
 


 


(15)
Joint compensation using blind I/Q signal separa-
tion:
As given in (15), the observable I and Q signals in the
presence of I/Q mismatch and carrier offsets appear as
instantaneous and time-varying mixtures of the true I and
Q signals. Switching to discrete-time notations


I
x
k



I
x
kTs , and so forth, we introduce 2 × 1 source and
observation vectors
 
,T
IQ
s
kzkzk




 and


x
k

 
,T
IQ
xkxk



 and write the model in (15) as x(k) =
A(k)s(k), where,




cos sin
sin cos
kk
Ak gk gk
 






 



 

(16)
Notice that the frequency offset Δω refers here to the
“normalized” frequency variable Δω = 2πΔf / fS. Now an
adaptive signal separation algorithm, such as the EASI
algorithm discussed in Section 3, can be used to blindly
estimate the source signals zI (k) and zQ(k).
It is interesting to note that the identifiability of the
model in (16) is independent of the carrier offset levels
and also practically independent of the mismatch values.
This can be seen more formally by examining the deter-
minant of A(k):




det cosAk g

 (17)
Thus, the system matrix A(k) is invertible given that
0g

and /2



. The first requirement (0g

)
simply states that the downconversion stage needs to
produce two non-zero signals while the second one pre-
vents the case where the two signals after down-conver-
sion would be just scaled versions of each other. These
are more than natural requirements for any I/Q front-end
and are always fulfilled by any practical analog design.
Thus this indicates that the proposed idea is robust in the
face of dif- ferent imbalance and offset levels in terms of
identifiability.

A Novel Approach to Blind I/Q Mismatch & Carrier Offset Compensation
23
There are some further practical issues related to the
proposed compensation scheme. First of all, as discussed
in Section 2, the direct-conversion architecture suffers
from the well-known DC offset problem [1,2,4] due to
self-mixing of the LO signal leaking into the mixer RF
port. Most signal separation algorithms, in turn, assume
zero-mean data, so the DC offset needs to be compen-
sated prior to the separation stage. Another practical as-
pect is related to the amount of frequency offsets toler-
ated. On one hand, the frequency offset should be smaller
than the guard band between adjacent frequency channels.
If not, the receiver is not anymore really zero IF but
closer to low IF and the nearby channel signal (or at least
part of it located on the true image band) appears as in-
terference on top of the desired signal after downconver-
sion. On the other hand, the frequency offset determines
the dynamics of the system matrix A(k) in (16), which is
indeed the dynamics that the adaptive separation algo-
rithm needs to follow. In other words, this dynamics
should be within the tracking capability of the applied
adaptive algorithm. Commonly, this poses some limita-
tions to the used step-sizes such that a relatively large
stepsize is needed. The used step-size, in turn, is usually
directly related to the separator steady-state performance
and cannot, of course, exceed its own algorithm-specific
stability limit [14,16,19]. Thus we can conclude that even
though the frequency offset level is irrelevant from the
identifiability point of view, the tracking capability of the
practical algorithms as well as the role of the image sig-
nal limit the applicability of the proposed concept to mild
frequency offset. In other words, coarse frequency syn-
chronization should be implemented by other means
prior to the separation stage. Notice also that due to the
amplitude (sign) and ordering ambiguities mentioned in
Section 3, it is possible that the recovered constellation
still formally suffers from 1) a constant phase rotation of
(integer multiple of) 90 and/or 2) complex conjugation.
In practice, these issues can be easily resolved using a
little side information in the actual data detection stage. It
should be noted that any blind algorithm is subject to
similar ambiguities in general.
As mentioned above, these ambiguities can be reduced
in a later stage of the receiver, for example, by using a
minimal number of known symbols (pilot or training sym-
bols specified in the signaling frame structure) or by us-
ing differential coding/mapping between bits and sym-
bols. In general, for modulations other than square QAM,
the effects of possible dependence between the true I and
Q should be explored individually.
5. Simulations
Here some example results are given to illustrate the ef-
ficiency of the proposed compensation idea. In the simu-
lations, imbalance levels of 3% and 3 are used corre-
sponding to an approximate of 30 dB image attenuation
which should represent a typical practical case. Phase
offset in the system is assumed to be 20 and the (remain-
ing) frequency offset 0.0001 × 2π. Given, for example, a
10 MHz sampling frequency, this corresponds to 1 kHz
absolute frequency offset.
The data modulation is 16 QAM. The model also in-
cludes additive white Gaussian noise (AWGN) with the
signal-to-noise ratio (SNR) ranging from 0 dB to 20 dB.
The EASI algorithm discussed in Section 3 is then used
as an example algorithm in the separation stage with a
step-size of 0.01 and a third-order (cubic) nonlinearity [19].
The time-varying mixture coefficients are illustrated in
Figure 2a, followed below by an example realization of
the separator coefficients in Figure 2b with SNR of 20
dB. Clearly, the separation algorithm is able to track the
time- varying mixture coefficients successfully. The cor-
responding symbol rate output samples without and with
compensation are depicted in Figure 3. As is evident, the
signal without compensation is useless due to I/Q mis-
match and carrier offsets. The compensator output signal,
however, is a good estimate of the transmitted symbol
constellation.
The most fundamental performance measure of any
digital communication system is the bit or symbol error
rate (BER/SER). This is assessed next for the proposed
compensator as a function of additive noise SNR. The
decisions are made symbol by symbol using the mini-
mum distance detection principle. The obtained results
are depicted in Figure 4 which also shows the SER with
additive noise only for refere nce.
The corresponding SER without any compensation is
close to one, independ ently of the SNR, and is not shown
for simplicity. As is evident, the proposed compensator
can efficiently estimate the transmitted signal, bringing
the error rate close to the AWGN bound. Especially in
the raw SER levels of 10-1 to 10-2, which is the crucial
operating range of any practical system before error-
control decoding, the proposed receiver is really close
(within 1 dB) to the noise limit.
6. Conclusions
In this paper, blind I/Q signal separation-based ap-
proaches for receiver signal processing were proposed.
More specifically, the I/Q mismatches and carrier offsets
as well as the linear distortion due to general band pass
channels were shown to create crosstalk between the
transmitted I and Q signals.
Then compensation structures utilizing blind signal
separation were used to compensate for these effects.
Also some simulation results were given to illustrate the
efficiency of the proposed techniques. Combining the

A Novel Approach to Blind I/Q Mismatch & Carrier Offset Compensation
24
(a) Mixture coefficients.
(b) Separator coefficients.
Figure 2. (a) An illustration of the dynamics of the system
matrix. (b) One realization of the separator coefficients
using the EASI algorithm (step-size 0.01). The I/Q mis-
match values: g = 1.03 and φ = 3. The carrier offset levels: θ
= 20 and Δω = 2π × 0.0001. Additive noise SNR = 20 dB.
(a) Without compensation.
(b)With compensation.
Figure 3. Symbol rate output samples (16-QAM) (a) without
and (b) with compensation. The I/Q mismatch values: g =
1.03 and φ = 3. The carrier offset levels: θ = 20 and Δω = 2π
× 0.0001. Additive noise SNR = 20 dB.
Figure 4. Symbol error rate of the EASI algorithm-based
compensator for 16-QAM data. The I/Q mismatch values: g
= 1.03 and φ = 3. The ca rrier offset levels: θ = 20 and Δω =
2π × 0.0001. Also shown for reference is the symbol error
rate with additive noise only (AWGN bound).
presented I/Q mismatch and carrier offset compensation
and the channel equalizer principles into a single (or a
cascade of two) I/Q separator(s) results in a versatile re-
ceiver building block for future radio communication
systems. Future work should be directed to further veri-
fication and prototyping of the proposed approaches us-
ing measured real-world receiver frontend signals.
Generally speaking, the idea behind this paper is to
give new views for applying complex or I/Q signal proc-
essing efficiently in radio receiver design and to take full
advantage of the rich signal structure inherent to com-
plex-valued communications signals.
REFERENCES
[1] S. Mirabbasi and K. Martin, “Classical and Modern Re-
ceiver Architectures,” IEEE Communication Magazine,
Vol. 38, No. 11, 2000, pp. 132-139. doi:10.1109/35.883
502
[2] A. A. Abidi, “Direct-conversion Radio Transceivers for
Digital Communications,” IEEE Journal Solid-State Cir-
cuits, Vol. 30, No. 12, 1995, pp. 1399-1410. doi:10.1109
/4.482187
[3] J. Crols and M. S. J. Steyaert, “Low-IF Topologies for
HighPerformance Analog Front Ends of Fully Integrated
Receivers,” IEEE Transaction Circuits System II, Vol. 45,
No. 3, 1998, pp. 269-282. doi:10.1109/82.664233
[4] J. Crols and M. S. J. Steyaert, “CMOS Wireless Trans-
ceiver Design,” Kluwer Academic, Dordrecht, 1997.
[5] M. E. Frerking, “Digital Signal Processing in Communi-
cation Systems,” Chapman & Hall, New York, 1994.
[6] E. A. Lee and D. G. Messerschmitt, “Digital Communi-
cation,” Kluwer Academic, Norwell, 1988.
[7] M. Valkama, “Advanced I/Q Signal Processing for Wide-

A Novel Approach to Blind I/Q Mismatch & Carrier Offset Compensation
25
band Receivers: Models and Algorithms,” Ph.D. Disser-
tation, Tampere University of Technology, Tampere,
2001.
[8] G. Fettweis, M. L¨ohning, D. Petrovic, M. Windisch, P.
Zillmann and E. Zimmermann, “Dirty RF,” Proceeding
on Wireless World Research Forum, Oslo, June 2004. pp.
133-148.
[9] M. Valkama, J. Pirskanen and M. Renfors, “Signal Proc-
essing Challenges for Applying Software Radio Princi-
ples in Future Wireless Terminals: an Overview,” Inter-
national Journal of Communication Systems, Vol. 15,
2002, pp. 741-769. doi:10.1002/dac.561
[10] J. P. F. Glas, “Digital I/Q Imbalance Compensation in a
low-IF Receiver,” IEEE Proceeding on Global Telecom-
munications Conference, Sydney, November 1998, Vol. 3,
pp. 1461-1466.
[11] ] M. Valkama and M. Renfors, “Digital I/Q Imbalance
Compensation in Direct-Conversion Receivers,” Pro-
ceeding on Workshop on Software Radios, Karlsruhe,
March 2002, pp. 51-55.
[12] I.-H. Sohn, E.-R. Jeong and Y. H. Lee, “Data-aided Ap-
proach to I/Q Mismatch and DC-offset Compensation in
Communication Receivers,” IEEE Communication Let-
ters, Vol. 6, No. 12, 2002, pp. 547-549. doi:10.1109/
LCOMM. 2002.806451
[13] J. K. Cavers and M. W. Liao, “Adaptive Compensation
for Imbalance and Offset Losses in Direct Conversion
Transceivers,” IEEE Transaction on Vehicular Technol-
ogy, Vol. 42, No. 4, 1993, pp. 581-588. doi:10.1109
/25.260752
[14] S. Haykin, “Unsupervised Adaptive Filtering, Vol. I:
Blind Source Separation,” John Wiley & Sons, New York,
2000.
[15] S. Haykin, “Unsupervised Adaptive Filtering, Vol. II:
Blind Deconvolution,” JohnWiley & Sons, New York,
2000.
[16] A. Hyv¨arinen, J. Karhunen and E. Oja, “Independent
Component Analysis,” JohnWiley & Sons, New York,
2001. doi:10.1002/0471221317
[17] J. Eriksson and V. Koivunen, “Identifiability, Separabil-
ity, and Uniqueness of Linear ICAmodels,” IEEE Signal
Processing Letters, Vol. 11, No. 7, 2004, pp. 601-604.
doi:10.1109/LSP.2004.830118
[18] J.-F. Cardoso, “Blind Signal Separation: Statistical Prin-
ciples,” Proceeding of the IEEE, Vol. 86, No. 10, 1998,
pp. 2009-2025. doi:10.1109/5.720250
[19] J.-F. Cardoso and B. H. Laheld, “Equivariant Adaptive
Source Separation,” IEEE Transaction Signal Processing,
Vol. 44, No. 12, 1996, pp. 3017-3030. doi:10.1109/78.
553476
[20] S. Amari, S. C. Douglas, A. Cichocki and H. H. Yang,
“Multichannel Blind Deconvolution and Equalization us-
ing the Natural Gradient,” 1st IEEE Proceeding on Work-
shop on Signal Processing Advances in Wireless Com-
munications, Paris, April 1997, pp. 101-104..
[21] S. C. Douglas and S. Haykin, “On the Relationship Be-
tween Blind Deconvolution and Blind Source Separa-
tion,” 31st Proceeding on Asilomar Conference on Sig-
nals, Systems, and Computers (ASILOMAR ’97), Calif,
November 1997, pp. 1591-1595.
[22] S. Haykin, “Adaptive Filter Theory,” 3rd edition, Pren-
tice-Hall, New Jersey, 1996.