240 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO.

1, JANUARY 2007

Joint Compensation of Transmitter and

Receiver Impairments in OFDM Systems
Alireza Tarighat, Member, IEEE, and Ali H. Sayed, Fellow, IEEE

Abstract— The implementation of OFDM-based systems suf- amplitudes. The performance of a receiver can be severely
fers from impairments such as in-phase and quadrature-phase limited by such IQ imbalances at the transmitter and receiver.
(IQ) imbalances in the front-end analog processing. Such im- The effect of receiver IQ imbalances on OFDM systems
balances are caused by the analog processing of the radio
frequency (RF) signal and can be present at both the transmitter and the resulting performance degradation have been investi-
and receiver. The resulting IQ distortion limits the achievable gated in [3], [4]. Several compensation algorithms have been
operating SNR at the receiver and the achievable data rates. proposed in [5]–[9]. Developing receiver algorithms in the
In this paper, the effect of both the transmitter and receiver digital domain that compensate for IQ imbalances in integrated
IQ imbalances in an OFDM system is studied and algorithms wireless systems can lead to more efficient systems in terms
are developed to compensate for such distortions in the digital
domain. The algorithms include post-FFT least-squares and of area-power-cost [5]. In the the recent works [5], [9] by the
adaptive equalization, as well as a pre-distortion scheme at the authors, compensation algorithms for OFDM receivers with
transmitter and a pre-FFT correction at the receiver. IQ imbalances have been developed for both cases of SISO
Index Terms— Compensation algorithms for analog impair- and MIMO communications. All these previous studies have
ments, orthogonal frequency division multiplexing (OFDM), focused on the problem of IQ imbalances at the receiver only.
transmitter and receiver in-phase and quadrature-phase (IQ) In this paper, we first develop a framework that models the
imbalances. effect of IQ imbalances at both the transmitter and receiver
of an OFDM system. Two compensation approaches are then
I. I NTRODUCTION presented. In one approach, both the transmitter and receiver
distortions are compensated jointly at the receiver after the
Limiting issue in the implementation of wireless systems
A is the impairment associated with analog processing
due to component imperfections. Most of the impairments
FFT operation. In the other approach, the transmitter imbal-
ance is compensated at the transmitter by using a proposed
pre-distorter, while the receiver imbalance is compensated at
cannot be efficiently nor entirely eliminated in the analog the receiver. The latter approach addresses the transmitter error
domain due to power-area-cost trade-offs. Therefore, efficient vector magnitude (EVM) in standardized OFDM systems.
compensation schemes in the digital baseband domain are The paper is organized as follows. The next section de-
desirable for wireless transceivers. scribes the model used for transmitter and receiver IQ imbal-
A major source of impairments in high-frequency wireless ances. Section III formulates the effect of IQ imbalances on
system implementations is the imbalance between the In- an OFDM system. Joint compensation schemes at the receiver
phase (I) and Quadrature-phase (Q) branches; or equivalently, are presented in Section IV. A pre-distortion scheme at the
the real and imaginary parts of the complex signal [1]. transmitter is discussed in Section V. Simulation results are
This imbalance can be introduced at both the transmitter shown in Section VI and conclusions are given in Section VII.
(during frequency up-conversion) and the receiver (during fre-
quency down-conversion). Both the up-conversion and down-
II. F ORMULATION OF IQ I MBALANCES
conversion are implemented in the analog domain by what
is known as complex up-conversion and complex down- Let y(t) represent the received baseband complex signal
conversion (for more information see [1], [2]). To perform before being distorted by the IQ imbalance at the receiver.
the complex frequency conversion, both the sine and cosine The distorted signal in the time domain can be modeled as
oscillating waveforms are required. The IQ imbalance is [3], [4]:
basically any mismatch between the I and Q branches from the yd (t) = μr y(t) + νr y ∗ (t) (1)
ideal case, i.e., from the exact 90o phase difference and equal where the distortion parameters, μr and νr , are related to the
Manuscript received February 28, 2005; revised July 23, 2005; accepted amplitude and phase imbalances between the I and Q branches
February 10, 2006. The associate editor coordinating the review of this paper in the RF/Analog demodulation process at the receiver. A
and approving it for publication was H. Li. Part of this work appeared in simplified model for the distortion parameters can be written
the proceedings of the IEEE Workshop on Signal Processing Advances in
Wireless Communications (SPAWC), 2005. This work was supported in part as [4]:
by NSF grants CCF-0208573 and ECS-0401188. μr = cos(θr /2) + jαr sin(θr /2)
(2)
A. Tarighat was with the Electrical Engineering Department, University of νr = αr cos(θr /2) − j sin(θr /2)
California, Los Angeles, CA 90095 USA (e-mail: [email protected]). He
is now with WiLinx, Los Angeles, CA 90025 USA. where θr and αr are respectively the phase and amplitude
A. H. Sayed is with the Electrical Engineering Department, University of
California, Los Angeles, CA 90095 USA (e-mail: [email protected]). imbalance between the I and Q branches at the receiver. The
Digital Object Identifier 10.1109/TWC.2007.05135 phase imbalance is any phase deviation from the ideal 90o
1536-1276/07$20.00 
c 2007 IEEE
TARIGHAT AND SAYED: JOINT COMPENSATION OF TRANSMITTER AND RECEIVER IMPAIRMENTS IN OFDM SYSTEMS 241

between the I and Q branches. The amplitude imbalance is where Hc is an N × N circulant matrix whose first row is
defined as:  
aI − aQ h0 h1 · · · hL 0 . . . 0 (10)
αr =
aI + aQ and v̄ is additive white noise at the receiver. It is known that
where aI and aQ are the gain amplitudes on the I and Hc can be diagonalized by the N -point DFT matrix as
Q branches. When stated in dB, the amplitude imbalance
Hc = F∗ ΛF (11)
is computed as 10 log(1 + αr ). For instance, an amplitude
imbalance of 0dB corresponds to the ideal case of αr = 0. where
The values of θr and αr are not known at the receiver since Λ = diag{λ} (12)
they are caused by manufacturing inaccuracies in the analog
and the vector λ is related to h via
components.  
√ ∗ h
A similar expression can be used to model IQ imbalances λ= N F (13)
at the transmitter. Let s(t) represent the transmitted baseband 0(N −(L+1))×1
complex signal before being distorted by IQ imbalances. Then Substituting (6) and (7) into (9) leads to
the distorted baseband signal in the time domain will be given
by–see Fig. 1: ȳ = Hc [μt s̄ + νt conj(s̄)] + v̄ (14)

sd (t) = μt s(t) + νt s∗ (t) (3) or

ȳ = Hc [μt F∗ s + νt conj(F∗ s)] + v̄ (15)
where the distortion parameters μt and νt are defined as in
(2). The design of OFDM receivers in the presence of both The received block of data ȳ after being distorted by receiver
transmitter and receiver IQ imbalances is discussed next. IQ imbalances will be transformed into (using (1)):
z̄ = μr ȳ + νr conj(ȳ) (16)
III. OFDM S IGNALS W ITH IQ I MBALANCES
Now remember that the N -point DFT of the complex conju-
We extend the approach of [5], [9] to the case in which gate of a sequence is related to the DFT of the original se-
IQ imbalances are present at both the transmitter and the quence through a mirrored relation. For notational simplicity,
receiver. Thus recall that in OFDM systems, a block of data we denote the operation which gives the DFT of the complex
is transmitted as an OFDM symbol. Assuming a block size conjugate of a vector by the superscript #, i.e., for a vector
equal to N (where N is a power of 2), the transmitted block X of size N we write
of data is denoted by ⎡ ⎤ ⎡ ⎤
X(1) X ∗ (1)
Δ
s = col{s(1), s(2), . . . , s(N )} (5) ⎢ X(2) ⎥ ⎢ X ∗ (N ) ⎥
⎢ ⎥ ⎢ ⎥
⎢ .. ⎥ ⎢ .. ⎥
⎢ . ⎥ ⎢ ⎥
Each block is passed through the IDFT operation: ⎢ ⎥ ⎢ ∗ . ⎥
⎢ X(N/2) ⎥ ⎢ X (N/2 + 2) ⎥
X =⎢ ⎢ ⎥ # ⎢ ⎥
s̄ = F∗ s (6) ⎥ =⇒ X = ⎢ X ∗ (N/2 + 1) ⎥
⎢ X(N/2 + 1) ⎥ ⎢ ⎥
⎢ X(N/2 + 2) ⎥ ⎢ X ∗ (N/2) ⎥
where F is the unitary discrete Fourier transform (DFT) matrix ⎢ ⎥ ⎢ ⎥
⎢ .. ⎥ ⎢ .. ⎥
of size N defined by ⎣ . ⎦ ⎣ . ⎦
Δ √ X(N ) X ∗ (2)
[F]ik = √1N exp −j2πikN , j = −1 (17)
i, k = {0, 1, . . . , N − 1} so that
A cyclic prefix of length P is added to each transformed block X = Fx =⇒ X # = Fconj (x) (18)
of data and then transmitted through the channel–see Fig. 1. It can be verified similarly that
Due to IQ imbalances at the transmitter, as modeled by (3),
the distorted transmitted vector is given by: x = F∗ X =⇒ x# = F∗ conj (X) (19)

s̄d = μt s̄ + νt conj(s̄) (7) Now substituting (11) into (15) gives

ȳ = F∗ ΛF [μt F∗ s + νt conj(F∗ s)] + v̄
where the notation conj(.) denotes a column vector (or matrix)
whose entries are the complex conjugates of its argument. An = μt F∗ Λs + νt F∗ ΛFconj(F∗ s) + v̄
(20)
FIR model with L + 1 taps is assumed for the channel, i.e., = μt F∗ Λs + νt F∗ Λs# + v̄
h = col{h0 , h1 , . . . , hL } (8) = F∗ diag{λ} μt s + νt s# + v̄

with L ≤ P in order to preserve the orthogonality between where we used the fact that Fconj(F∗ s) = (FF∗ s)# = s# .
tones. At the receiver, the received samples corresponding Moreover, using (14), we can write
to the transmitted block s̄ are collected into a vector, after conj(ȳ) = conj(Hc ) [μ∗t conj(s̄) + νt∗ (s̄)] + conj(v̄) (21)
discarding the received cyclic prefix samples. The received
block of data before being distorted by receiver IQ imbalances where conj(Hc ) is again a circulant matrix defined in terms
is given by [5]: of conj(h) as in (10) so that
 
ȳ = Hc s̄d + v̄ (9) conj(Hc ) = F∗ diag λ# F (22)
242 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 1, JANUARY 2007

 
μr μt λ(k) + νr νt∗ λ∗ (N − k + 2) μr νt λ(k) + νr μ∗t λ∗ (N − k + 2)
Γk = (4)
νr∗ μt λ(k) + μ∗r νt∗ λ∗ (N − k + 2) νr∗ νt λ(k) + μ∗r μ∗t λ∗ (N − k + 2)

" #
z(2 )

Discard tones 1 & N/2+1

|
z2 =
s (1 ) z* (N )
s (2) ȳ z
.. s IFFT s̄ Analog Transmitter s̄ d Analog Receiver
h

FFT

{z
IQ Distortion IQ Distortion
s ( N – 1) s̄ d = μ t s̄ + νt conj (s̄) ȳd = μr ȳ + νr conj ( ȳ)
" #
s (N ) z( N /2)
| {z } zN /2 =
s z* (N / 2 +2)

}
Fig. 1. An OFDM system with both transmit and receive IQ imbalances and the notation used in the derivations. Note that the addition and removal of the
cyclic prefix are not shown in this figure for ease of notations.

where   and  
∗ conj(h) # v(k)
F =λ (23) vk =
0(N −(L+1))×1 v∗ (N − k + 2)
Substituting the above into (21) results in for k = {2, . . . , N/2} and the 2 × 2 matrix Γk is given by
  (4).
conj(ȳ) = F∗ diag λ# F [μ∗t conj(s̄) + νt∗ s̄] + conj(v̄)
  The objective is to recover sk from zk in (27) for k =
= F∗ diag λ# μ∗t s# + νt∗ s + conj(v̄) {2, . . . , N/2} or, equivalently, s from z. Several algorithms,
(24)
adaptive and otherwise, for estimating channel/disortion para-
where we substituted Fs̄ = s and Fconj(s̄) = s# using (6)
meters and recovering the sk for the special case with ideal
and (18).
transmitter (μt = 1 and νt = 0) were proposed in [5]. The
After applying the DFT operation to the received block of
main difference here in relation to [5] is in the form of the
data z̄ given by (16) (as is done in an OFDM receiver) and
matrix Γk , which contains contributions from IQ distortions
using (20) and (24), we obtain
from both the transmitter and receiver. In the sequel, we
Δ extend some of the schemes of [5] to the more general case
z = Fz̄
of imbalances at both the transmitter and receiver.
=μr diag{λ} μt s + νt s# + (25)
 
νr diag λ# μ∗t s# + νt∗ s + μr v + νr v#
IV. J OINT TX/RX C OMPENSATION AT THE R ECEIVER
where the vector v is given by v = Fv̄ and the vector v# is A. Least-Squares Compensation
defined according to the transformation (17). After rearranging
terms, The least-squares estimate of sk , k = {2, . . . , N/2}, de-
noted by ŝk , is given by [10]:
 
z = μr μt diag{λ} + νr νt∗ diag λ# s+
  ŝk = (δI + Γ∗k Γk )−1 Γ∗k zk (31)
μr νt diag{λ} + νr μ∗t diag λ# s# + (26)
μr v + νr v# where the parameter δ > 0 is added for regularization. In
order to implement the solution (31), the channel information
This result gives the exact input-output relation in an OFDM (λ) and the distortion parameters (μt ,νt ,μr ,νr ) are required.
system with both transmitter and receiver IQ imbalances as a Training symbols are required to enable the receiver to esti-
function of the channel taps {λ} and the distortion parameters mate those values. Thus note that we may use equation (27)
μr , μt , νr , and νt . Note that (26) collapses to the input-output for channel estimation by rewriting it as:
relation derived in [5] for μt = 1 and νt = 0, as a special case  
s(k) 0 s∗ (N − k + 2) 0
where ideal IQ branches were assumed at the transmitter. zk = ×
0 s(k) 0 s∗ (N − k + 2)
As seen from (26), the vector z is no longer related only ⎡ ⎤
to the transmitted block s through a diagonal matrix, as is the μr μt λ(k) + νr νt∗ λ∗ (N − k + 2)
⎢ νr∗ μt λ(k) + μ∗r νt∗ λ∗ (N − k + 2) ⎥
⎢ ⎥
⎣ μr νt λ(k) + νr μ∗t λ∗ (N − k + 2) ⎦ + vk
case in an OFDM system with ideal I and Q branches. There
is also a contribution from s# . Note however that the system
of equations defined by (26) can be reduced to independent νr∗ νt λ(k) + μ∗r μ∗t λ∗ (N − k + 2)
(32)
2 × 2 systems of equations as follows [5]:
Assuming nT r OFDM symbols are transmitted for training,
zk = Γk sk + vk (27) then nT r realizations of the above equation can be collected to
perform the least-squares estimation of the elements forming
where Γk . The estimated Γk can then be substituted into (31) for data
   
z(k) s(k) estimation. The same training data used for channel estimation
zk = , sk =
z∗ (N − k + 2) s∗ (N − k + 2) in standard OFDM systems can be used in this scheme as the
(28) training symbols for joint channel and distortion estimation.
TARIGHAT AND SAYED: JOINT COMPENSATION OF TRANSMITTER AND RECEIVER IMPAIRMENTS IN OFDM SYSTEMS 243

∗
2 2 2 λ(k) 2 2 λ (k) ∗ ∗ ∗
σv2 (|μr | + |νr | )(|μt | + | λ(N −k+2) | |νt | ) + 4Re( λ∗ (N −k+2) μr μt νr νt )
Rs̃k (1, 1) = (29)
|λ(k)|2 (|μr |2 − |νr |2 )2 (|μt |2 − |νt |2 )2
⎛ ∗
⎞
λ (k)
(|μr |2 + |νr |2 )(|μt |2 + | λ(Nλ(k) 2 2 ∗ ∗ ∗
−k+2) | |νt | ) + 4Re( λ∗ (N −k+2) μr μt νr νt )
Loss in SNR = −10 log ⎝ ⎠ (30)
(|μr |2 − |νr |2 )2 (|μt |2 − |νt |2 )2

A performance analysis on the achievable SNR using this where wk and wN −k+2 are 1×2 equalization vectors updated
compensation scheme is now given and compared to that according to an adaptive algorithm (for instance LMS or some
of a receiver with ideal IQ branches. Using (27) and the other adaptive form) for k = {2, . . . , N/2} [10]. To better
corresponding least-squares estimate given by (31), the error illustrate the update equations, we introduce the time (or
(i) (i)
in the estimation of sk is given by (assuming δ ≈ 0): iteration) index i. As a result, let wk and wN −k+2 represent
(i)
Δ
s̃k = ŝk − sk = Γ−1 the equalization vectors at time instant i. Furthermore, let zk
k vk
(33)
represent the vector zk defined in (28) at time instant i. Now,
and, consequently, the covariance matrix of the error vector is the equalization coefficients for k = {2, . . . , N/2} are updated
Δ
Rs̃k = E (s̃k s̃∗k ) = σv2 (Γ∗k Γk )−1 (34) according to the LMS rules:
 ∗
(i+1) (i) (i) (i)
where Rvk was substituted by σv2 I2×2 . Now let us consider wk = wk + μLMS zk ek (38)
the (1, 1) element of Rs̃k , which denotes the error variance  ∗
in estimating s(k). It can be verified that the (1, 1) element (i+1) (i)
wN −k+2 = wN −k+2 + μLMS zk
(i) (i)
eN −k+2 (39)
of Rs̃k is given by (29).
Note that the expression (29) collapses to (i) (i) (i) (i)
where ek = dk − wk zk is the error signal generated at
2 (i)
σv iteration i for the tone index k using a training symbol dk ,
(35) (i)
|λ(k)|2 where the training symbol dk can be different for different
(i)
for a receiver with ideal transmitter and receiver IQ branches tone indices k. A similar relation holds for eN −k+2 . Moreover,
(μr = μr = 1 and νr = νt = 0), as is expected. The difference μLMS is the LMS step-size parameter.
between the error variance given by (29) for a receiver with An important property of the schemes proposed in this
least-squares equalization and the error variance given by (35) section is that they compensate for both transmit and re-
for a receiver with ideal IQ branches is defined as the loss in ceive imbalances jointly at the receiver. In other words,
SNR (in dB) and is given by (30) where μr , μt , νr , and νt are the transmitter is not necessarily required to achieve good
related to the physical imbalances αr , αt , θr , and θt via (2). IQ matching. This is an advantage for proprietary systems
A similar expression can be derived for the error variance in where the transmitter and the receiver are designed by the
estimating the other element of sk , namely, s(N − k + 2). same manufacturer, since it can significantly relax the design
Note that the above expression also depends on the ratio specification on the transmitter. However, this is not desired
between the channel values λ(k) and λ(N − k + 2). To better for standardized systems where the transmitters and receivers
illustrate the dependence of the SNR loss on the physical may be designed and manufactured by different manufacturers.
distortion parameters (as an upper bound on the achievable In such systems, the transmitted signal’s distortion has to be
performance), we evaluate the expression (30) for several below a certain level specified by the standard, namely the
values of distortion parameters assuming λ(k) ≈ λ(N −k+2), error vector magnitude (EVM), so that receivers by other man-
i.e., ufacturers can correctly decode it. In this case, the transmitter
Loss in SNR has to meet a certain level of IQ matching. To address this
  issue, we suggest below a scheme at the transmitter, referred
(|μr |2 + |νr |2 )(|μt |2 + |νt |2 ) + 4Re(μ∗r μ∗t νr νt∗ )
= −10 log to as pre-distortion, such that the final transmitted signal is
(|μr |2 − |νr |2 )2 (|μt |2 − |νt |2 )2 sufficiently close to an ideal transmitter.
(36)
The above expression is evaluated for different values of
transmitter and receiver distortion parameters–see Table I at V. P OST-IFFT AND P RE -FFT C OMPENSATION
the top of the next page. For instance, the SNR degradation in Recalling (1) as the model for the distorted signal, let us
the presence of both transmitter and receiver imbalances (for define the following operation [5]:
reasonably large values of θr = 2.0 , αr = 1.0dB, θt = 2.0 ,
o o
Δ
and αr = 1.0dB) is lower bounded by ∼ 0.5dB. yc (t) = yd (t) − ηr yd∗ (t) (40)
where
Δ νr
B. Adaptive Equalization ηr = ∗ (41)
μr
The adaptive estimation of s(k) and s∗ (N − k + 2) in (28)
can be attained as follows: Then it can be verified that
 
ŝ(k) = wk zk |νr |2
(37) yc (t) = μr − ∗ y(t) (42)
ŝ∗ (N − k + 2) = wN −k+2 zk μr
244 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 1, JANUARY 2007

TABLE I
L OSS IN SNR ( IN DB) A CCORDING TO (36) E VALUATED FOR D IFFERENT VALUES OF T RANSMITTER AND R ECEIVER D ISTORTION PARAMETERS

θt = 0o θt = 0.4o θt = 0.8o θt = 1.2o θt = 1.6o θt = 2.0o

αt = 0dB αt = 0.2dB αt = 0.4dB αt = 0.6dB αt = 0.8dB αt = 1.0dB
θr = 0o , αr = 0dB 0.00 0.01 0.03 0.06 0.11 0.18
θr = 0.4o , αr = 0.2dB 0.01 0.02 0.05 0.10 0.16 0.23
θr = 0.8o , αr = 0.4dB 0.03 0.05 0.10 0.15 0.22 0.30
θr = 1.2o , αr = 0.6dB 0.06 0.10 0.15 0.21 0.29 0.38
θr = 1.6o , αr = 0.8dB 0.11 0.16 0.22 0.29 0.37 0.47
θr = 2.0o , αr = 1.0dB 0.18 0.23 0.30 0.38 0.47 0.57

s (1)
s (2) s̄ ȳ
.. Digital Analog Transmitter s̄ d
h
IFFT

Pre-distortion IQ Distortion
s ( N − 1) s̄ c = s̄ − η t conj(s̄ ) s̄ d = μ t s̄ c + νt conj (s̄ c )
s (N )
| {z }
s

z (1 )
ȳ ȳd Digital ȳc z z (2)
Analog Receiver ..

FFT
IQ Distortion Compensation
ȳd = μ r ȳ + νr co nj(ȳ ) ȳc = ȳd − η r conj( ȳd ) z (N − 1)
z( N )
| {z }
z

Fig. 2. An OFDM system with post-IFFT transmit and pre-FFT receive compensation for IQ imbalances. The gray blocks depict the compensation operations.
Note that the addition and removal of the cyclic prefix are not shown in the figure for ease of notations.

This relation suggests that the receiver IQ distortion can be It can be verified that these elements can be reorganized as a
removed by using the above scaling given that the value of function of ηt and ηr :
ηr is provided. Note that only the ratio between νr and μ∗r is
ρ11 (k) = β(k) + ηr ηt∗ β ∗ (N − k + 2)
needed to calculate (40), and not the individual values.
A similar approach can be used at the transmitter but in ρ12 (k) = ηt β(k) + ηr β ∗ (N − k + 2)
(47)
a different order. In this case, the transmitted signal is pre- ρ21 (k) = ηr∗ β(k) + ηt∗ β ∗ (N − k + 2)
distorted in the digital domain before transmission in such ρ22 (k) = ηr∗ ηt β(k) + β ∗ (N − k + 2)
a way that the final transmitted signal is sufficiently free of
distortion. Recalling (3), let us assume that the transmitted where
Δ
signal s(t) is first pre-distorted according to β(k) = μr μt λ(k)
Δ
(48)
Δ β(N − k + 2) = μr μt λ(N − k + 2)
sc (t) = s(t) − ηt s∗ (t) (43)
Using the input-output relation (27) and by transmitting some
where training sequence, the channel matrix Γk can be estimated. A
Δ νt
ηt = (44) special pilot pattern is proposed in [5] that can be used here to
μt simplify the process of estimating the elements of Γk . Thus,
Then using sc (t) in (3) gives assume that sufficiently accurate estimates of the elements of
  Γk are achievable. Denote them by ρ̂11 (k), ρ̂12 (k), ρ̂21 (k),
|νt |2 and ρ̂22 (k). Once these estimates are available, they can be
sd (t) = μt − ∗ s(t) (45)
μt substituted in the system of equations (47) to solve for the
four unknowns ηt , ηr , β(k), and β(N − k + 2). There are
In this way, the transmitted signal is a multiple of the desired
two points here to consider: 1) The system of equations (47)
signal and is free of IQ distortion. A block diagram of an
can be repeated for different values of k = {2, . . . , N/2}.
OFDM system using the above operations is depicted in Fig. 2.
Since ηt and ηr are independent of k, the estimates of ηt
The issue now becomes how to estimate the compensation
and ηr for different values of k can be averaged at the end for
parameters ηt and ηr . Let us revisit the channel matrix given
better estimates. 2) It can be verified that the four equations in
by (4). Let ρ11 (k), ρ12 (k), ρ21 (k), and ρ22 (k) denote the four
(47) collapse to only two independent equations when β(k) =
elements in Γk , i.e,
  β(N − k + 2) or, equivalently, λ(k) = λ(N − k + 2), with
ρ11 (k) ρ12 (k) three unknowns left ηt , ηr , and β(k). In other words, with a
Γk = (46)
ρ21 (k) ρ22 (k) flat (single-path) channel, not enough information is gathered
TARIGHAT AND SAYED: JOINT COMPENSATION OF TRANSMITTER AND RECEIVER IMPAIRMENTS IN OFDM SYSTEMS 245

o o o o
QAM=16, Phase Im(RX)=3 , Gain Im(RX)=1dB, Phase Im(TX)=3 , Gain Im(TX)=1dB, n training=20, μLMS =0.1 QAM=64, Phase Im(RX)=2 , Gain Im(RX)=0.8dB, Phase Im(TX)=2 , Gain Im(TX)=0.8dB, n training=20, μLMS =0.1
0 0
10 10
Ideal IQ Branches Ideal IQ Branches
IQ Imbalance/No Compensation IQ Imbalance/No Compensation
IQ Imbalance/Joint Least − Squares IQ Imbalance/Joint Least −Squares
IQ Imbalance/Joint Adaptive −1 IQ Imbalance/Joint Adaptive
−1
10
10

−2
10
−2
10

−3

Uncoded BER
10
Uncoded BER

−3
10
−4
10

−4
10
−5
10

−5 −6
10 10

−7
−6 10
10 0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30
SNR (dB)
SNR (dB)

(a) AWGN flat channel (non-fading). (a) AWGN flat channel (non-fading).
o o QAM=64, Phase Im(RX)=2o, Gain Im(RX)=0.8dB, Phase Im(TX)=2o, Gain Im(TX)=0.8dB, n =40, μ =0.1
QAM=16, Phase Im(RX)=3 , Gain Im(RX)=1dB, Phase Im(TX)=3 , Gain Im(TX)=1dB, n =20, μ =0.1 0
training LMS
training LMS
0 10
10
Ideal IQ Branches Ideal IQ Branches
IQ Imbalance/No Compensation IQ Imbalance/No Compensation
IQ Imbalance/Joint Least − Squares IQ Imbalance/Joint Least − Squares
IQ Imbalance/Joint Adaptive IQ Imbalance/Joint Adaptive

−1
−1
10
10
Uncoded BER
Uncoded BER

−2
10
−2
10

−3
10
−3 0 5 10 15 20 25 30 35 40
10
0 5 10 15 20 25 30 SNR (dB)
SNR (dB)
(b) 4-tap Rayleigh fading channel (fading).
(b) 4-tap Rayleigh fading channel (fading).
Fig. 4. BER vs. SNR simulated for the following configuration: 64QAM
Fig. 3. BER vs. SNR simulated for the following configuration: 16QAM constellation, training length of 40 OFDM symbols in least-squares and LMS
constellation, training length of 20 OFDM symbols in least-squares and LMS solutions, LMS step-size of μLM S = 0.1, transmitter phase imbalance
solutions, LMS step-size of μLM S = 0.1, transmitter phase imbalance of θt = 2o , transmitter amplitude imbalance of αt =0.8dB, receiver phase
of θt = 3o , transmitter amplitude imbalance of αt =1dB, receiver phase imbalance of θr = 2o , and receiver amplitude imbalance of αt =0.8dB. ‘IQ
imbalance of θr = 3o , and receiver amplitude imbalance of αt =1dB. ‘IQ Imbalance/Joint Least-Squares’ refers to the compensation scheme given by
Imbalance/Joint Least-Squares’ refers to the compensation scheme given by (31) and ‘IQ Imbalance/Joint Adaptive’ refers to the scheme given by (37).
(31) and ‘IQ Imbalance/Joint Adaptive’ refers to the scheme given by (37). In both schemes, the transmitter and receiver imbalances are compensated
In both schemes, the transmitter and receiver imbalances are compensated jointly at the receiver.
jointly at the receiver.

and ηt can be estimated from

to estimate the unknown values. Therefore, it is necessary to  ∗
have a multi-path channel in order to use (47) to estimate ηt ρ12 (k) ρ21 (k)
, (50)
and ηr . ρ11 (k) ρ22 (k)
A simpler scheme can be used to estimate ηt and ηr if a If a transmitter with ideal IQ branches is available, then
transmitter or receiver with sufficiently ideal IQ branches is substituting ηt = 0 in (47) results in
available during the estimation process. In this case, an ideal
transmitter can be used to estimate ηr or vice versa (an ideal ρ11 (k) = β(k)
receiver can be used to estimate ηt ). If a receiver with ideal IQ ρ12 (k) = ηr β ∗ (N − k + 2)
branches is available, then substituting ηr = 0 in (47) results (51)
ρ21 (k) = ηr∗ β(k)
in
ρ22 (k) = β ∗ (N − k + 2)
ρ11 (k) = β(k)
ρ12 (k) = ηt β(k) and ηr can be estimated from
(49)  ∗
ρ21 (k) = ηt∗ β ∗ (N − k + 2) ρ12 (k) ρ21 (k)
, (52)
ρ22 (k) = β ∗ (N − k + 2) ρ22 (k) ρ11 (k)
246 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 1, JANUARY 2007

o o
QAM=16, Phase Im(RX)=3 , Gain Im(RX)=1dB, Phase Im(TX)=3 , Gain Im(TX)=1dB, n training=20, μLMS =0.1
10
0 viewed as a self-calibration process by the device before it
Ideal IQ Branches
IQ Imbalance/No Compensation starts communicating.
IQ Imbalance/Pre − Distorter

VI. S IMULATIONS
10
−1 A typical OFDM system (similar to IEEE802.11a) is simu-
lated to evaluate the performance of the proposed schemes in
Uncoded BER

comparison to an ideal OFDM system with no transmit-receive

IQ imbalance and a receiver with no compensation scheme.
The parameters used in the simulation are: OFDM symbol
length of N = 64, cyclic prefix of P = 16. All the figures
−2
10

present simulation results for uncoded BER. To better depict

the performance of the proposed schemes, each simulation
configuration is repeated for two different channel profiles
10
−3 for the sake of comparison. : 1) additive white Gaussian
0 5 10 15 20 25 30
noise (AWGN) channel with a single tap unity gain and 2)
SNR (dB)
a multipath channel with (L + 1) = 4 taps where the taps
Fig. 5. BER vs. SNR simulated for the following configuration: 4-tap are chosen independently with complex Gaussian distribution.
complex Gaussian fading channel, 16QAM constellation, training length of Every channel realization is independent of the previous one
20 OFDM symbols, transmitter phase imbalance of θt = 3o , transmitter
amplitude imbalance of αt =1dB, receiver phase imbalance of θr = 3o , and and the BER results depicted are from averaging the BER
receiver amplitude imbalance of αt =1dB. ‘IQ Imbalance/Pre-Distorter’ refers curves over hundreds of independent channels. The simulation
to the compensation scheme in defined by (40) and (43). results are presented in the following two sub-sections for the
algorithms presented in Section IV and Section V:
QAM=64, Phase Im(RX)=2o, Gain Im(RX)=0.8dB, Phase Im(TX)=2o, Gain Im(TX)=0.8dB, n =40, μ =0.1
training LMS
0
10
Ideal IQ Branches
IQ Imbalance/No Compensation A. Joint TX/RX compensation at the Receiver
IQ Imbalance/Pre− Distorter

The BER versus SNR are simulated and shown in Figs. 3

and 4 for 16QAM and 64QAM constellations. In all fig-
10
−1 ures, ‘Ideal IQ Branches’ legend refers to a receiver with
perfect IQ branches and perfect channel knowledge and ‘IQ
Uncoded BER

Imbalance/No Compensation’ refers to a system with IQ

imbalance but no compensation scheme. ‘IQ Imbalance/Joint
Least-Squares’ refers to the compensation scheme in sub-
section IV-A given by (31) where the matrix Γk is estimated
−2
10

using (32). ‘IQ Imbalance/Joint Adaptive’ refers to the scheme

presented in sub-section IV-B given by (37). Each system
configuration is repeated for an AWGN channel and a 4-tap
10
−3 complex Gaussian multipath channel for comparison purposes.
0 5 10 15 20 25 30 35 40
SNR (dB)
The values used for phase and amplitude imbalances for both
the transmitter and the receiver are typical values achievable
Fig. 6. BER vs. SNR simulated for the following configuration: 4-tap in practical integrated circuit implementations.
complex Gaussian fading channel, 64QAM constellation, training length of
40 OFDM symbols, transmitter phase imbalance of θt = 2o , transmitter
amplitude imbalance of αt =0.8dB, receiver phase imbalance of θr = 2o , B. Pot-IFFT Transmitter and Pre-FFT Receiver Compensa-
and receiver amplitude imbalance of αt =0.8dB. ‘IQ Imbalance/Pre-Distorter’
refers to the compensation scheme in defined by (40) and (43).
tions
The BER versus SNR are simulated and shown in Figs. 5
and 6 for 16QAM and 64QAM constellations. Only a 4-
Once one of the correction coefficients ηr or ηt has been multipath Rayleigh fading channel is simulated in this part. ‘IQ
estimated, the other one can be obtained by using the system Imbalance/Pre-Distorter’ refers to the compensation scheme
of equations (47). in Section V defined by (40) and (43). An ideal receiver is
Note that the schemes proposed in this section can be either assumed to estimate ηt using (50) and an ideal transmitter is
implemented using the receiver at the receiving device or using assumed to estimate ηr using (52 as explained in Section V.
the receiver available in the transmitting device. In the first
case, the estimated parameters need to be sent back to the VII. C ONCLUSIONS
transmitter. The feedback overhead will be negligible since The paper studied the problem of transmitter and receiver
these parameters change at a very slow rate. In the latter case, IQ imbalances in OFDM systems. An input-output relation is
the estimation is performed using the receiver available in the derived as a function of both transmit and receiver distortion
transmitting device and no feedback is required. In this case, parameters. The input-output relation is then used to develop
the parameter estimation process is performed during the time compensation algorithms for the IQ imbalances in the digital
that the device is not communicating. This process can be domain. Different compensation schemes are presented. Two
TARIGHAT AND SAYED: JOINT COMPENSATION OF TRANSMITTER AND RECEIVER IMPAIRMENTS IN OFDM SYSTEMS 247

of the proposed algorithms are implemented at the receiver Alireza Tarighat (S’00-M’05) received the B.Sc.
side and compensate for both transmitter and receiver dis- degree in electrical engineering from Sharif Uni-
versity of Technology, Tehran, Iran, in 1998. He
tortions in one step. In another approach, the transmitter received the M.Sc. and Ph.D. degrees in electrical
distortion is compensated through a pre-distortion operation engineering from the University of California, Los
and the receiver distortion is compensated for at the receiver. Angeles (UCLA) in 2001 and 2005, respectively.
During the summer of 2000, he was with Broad-
com, El Segundo, CA, where he worked on design-
R EFERENCES ing IEEE 802.11 transceivers. From 2001 to 2002 he
was with Innovics Wireless, Los Angeles, working
[1] B. Razavi, RF Microelectronics. Upper Saddle River, NJ: Prentice Hall, on systems and ASIC development of advanced
1998. antenna diversity and rake processing for 3G WCDMA mobile terminals.
[2] A. A. Abidi, “Direct-conversion radio transceivers for digital commu- Since 2005, he has been with WiLinx, Los Angeles, CA, working on systems
nications,” IEEE J. Solid-State Circuits, vol. 30, pp. 1399–1410, Dec. and silicon development of UWB wireless networks. His research interests
1995. are in communications theory and signal processing, including MIMO OFDM
[3] A. Baier, “Quadrature mixer imbalances in digital TDMA mobile radio systems, multi-user MIMO wireless networks, algorithms for impairments
receivers,” in Proc. International Zurich Seminar on Digital Commu- compensation, and experimental and practical communication systems.
nications, Electronic Circuits and Systems for Communications, Mar. Mr. Tarighat was the recipient of the Gold Medal of the National Physics
1990, pp. 147–162. Olympiad, Iran, 1993, and the Honorable Mention Diploma of the 25th
[4] C. L. Liu, “Impacts of I/Q imbalance on QPSK-OFDM-QAM detection,” International Physics Olympiad, Beijing, China, 1994. He received the 2006
IEEE Trans. Consumer Electron., vol. 44, pp. 984–989, Aug. 1998. Outstanding PhD Award in electrical engineering from UCLA.
[5] A. Tarighat, R. Bagheri, and A. H. Sayed, “Compensation schemes
and performance analysis of IQ imbalances in OFDM receivers,” IEEE Ali H. Sayed (F’01) received the Ph.D. degree in
Trans. Signal Processing, vol. 53, no. 8, pp. 3257–3268, Aug. 2005. electrical engineering in 1992 from Stanford Univer-
[6] A. Schuchert, R. Hasholzner, and P. Antoine, “A novel IQ imbalance sity, Stanford, CA. He is Professor and Chairman of
compensation scheme for the reception of OFDM signals,” IEEE Trans. electrical engineering at the University of Califor-
Consumer Electron., vol. 47, pp. 313–318, Aug. 2001. nia, Los Angeles. He is also the Principal Investi-
[7] S. Fouladifard and H. Shafiee, “Frequency offset estimation in OFDM gator of the UCLA Adaptive Systems Laboratory
systems in presence of IQ imbalance,” in Proc. International Conf. (www.ee.ucla.edu/asl). He has over 250 journal and
Commun., May 2003, pp. 2071–2075. conference publications, is the author of the textbook
[8] J. Tubbax, B. Come, L. V. der Perre, S. Donnay, M. Engels, M. Moonen, Fundamentals of Adaptive Filtering (New York:
and H. D. Man, “Joint compensation of IQ imbalance and frequency Wiley, 2003), is coauthor of the research mono-
offset in OFDM systems,” in Proc. Radio and Wireless Conference, graph Indefinite Quadratic Estimation and Control
Aug. 2003, pp. 39–42. (Philadelphia, PA: SIAM, 1999) and of the graduatelevel textbook Linear
[9] A. Tarighat and A. H. Sayed, “MIMO OFDM receivers for systems Estimation (Englewood Cliffs, NJ: PrenticeHall, 2000). He is also coeditor of
with IQ imbalances,” IEEE Trans. Signal Processing, vol. 53, no. 9, pp. the volume Fast Reliable Algorithms for Matrices with Structure (Philadel-
3583–3596, Sep. 2005. phia, PA: SIAM, 1999). He has contributed several articles to engineering and
[10] A. H. Sayed, Fundamentals of Adaptive Filtering. New York: Wiley, mathematical encyclopedias and handbooks and has served on the program
2003. committees of several international meetings. His research interests span
several areas, including adaptive and statistical signal processing, filtering
and estimation theories, signal processing for communications, interplays
between signal processing and control methodologies, system theory, and fast
algorithms for largescale problems.
Dr. Sayed is recipient of the 1996 IEEE Donald G. Fink Award, a 2002
Best Paper Award from the IEEE Signal Processing Society, the 2003 Kuwait
Prize in Basic Science, the 2005 Frederick E. Terman Award, the 2005 Young
Author Best Paper Award from the IEEE Signal Processing Society, and two
Best Student Paper awards at international meetings. He is also a member of
the technical committees on Signal Processing Theory and Methods (SPTM)
and on Signal Processing for Communications (SPCOM), both of the IEEE
Signal Processing Society. He is a member of the editorial board of the IEEE
Signal Processing Magazine. He has served as Editor-in-Chief of the IEEE
Transactions on Signal Processing (2003-2005) and is now serving as Editor-
in-Chief of the EURASIP Journal on Applied Signal Processing. He is serving
as General Chairman of ICASSP 2008.