Joint TX and RX IQ Imbalance Compensation of OFDM Transceiver in Mesh Network

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Joint Tx and Rx IQ Imbalance Compensation of

OFDM Transceiver in Mesh Network


Chia-Horng Liu
Telecommunication Laboratories
Chunghwa Telecom Co., Ltd., Taiwan

Abstract—The direct conversion is an attractive architecture, mismatched issues, joint TX and RX IQ imbalance problem
as compared to the traditional heterodyne approach, for both in multipath fading channel is modeled and studied in [9].
transmitter and receiver design. In spite of its great advantages The rest of this paper is organized as follows. In section
in hardware implementation, direct conversion results in several
negative effects, such as IQ imbalance, flicker noise, DC offset, II, we setup the system model and formulate the problem. In
etc. In this paper, we study the IQ imbalance problems occur Section III, we study joint TX and RX IQ imbalance prob-
at both transmitter and receiver in an OFDM-based mesh lems and propose end-to-end estimation schemes. Computer
network. The contributions of this paper are three-fold. First, simulations are conducted to evaluate the performance of the
we propose a self-calibration scheme at transmitter. Therefore, proposed schemes and the results are shown in Section IV.
the IQ imbalance at transmitter is compensated before communi-
cation. Second, we design a standard-independent IQ imbalance Finally, some concluding remarks are given in Section V.
estimation scheme at receiver. Finally, three different equalization II. P ROBLEM F ORMULATION
schemes for IQ-mismatched receiver are studied. It is shown from
our simulation results that the proposed schemes are feasible for A. Joint TX and RX IQ Imbalance System Model
solving Tx-Rx IQ imbalance problems in OFDM systems. Consider an OFDM system with IQ mismatched quadrature
modulator and demodulator. At transmitter, the local oscilla-
I. I NTRODUCTION
tor(LO) signal is
Direct conversion is a major approach to achieve compact
xLO (t) = (1 + εT )cos(wct + φT ) + j(1 − εT )sin(wct − φT )
and low cost transceiver design for current and next generation
wireless communication systems. In spite of the advantages in = AT e jwc t + BT e− jwc t (1)
hardware implementation, direct conversion results in several where
negative effects, such as In-phase/Quadrature (IQ) imbalance, 
AT = cos(φT ) + jεT sin(φT )
DC offset, even-order modulation, flicker noise, phase noise, (2)
BT = εT cos(φT ) − jsin(φT )
and self mixing [1]. The IQ imbalance arises at the quadrature
transceiver, in which the signals are de-multiplexed into in- ,wc = 2π fc , fc is the carrier frequency, εT and φT are amplitude
phase and quadrature parts. When in-phase and quadrature- and phase imbalance between the transmitter I and Q branches,
phase local reference signals are mismatched, the amplitude respectively. Two transmitter IQ imbalance coefficients, AT
and phase of output signals are imbalance. This will result and BT , are used to model the IQ imbalance system. Here,
in severe distortion in signal constellation and performance we use notation (.)∗ to denote complex conjugate. The radio
degradation. The problem of IQ imbalance have been modeled frequency(RF) signal at transmitter is [12]
and addressed in the previous works [2], [3].
xT (t) = 2Re{s(t)xLO (t)}
A performance degradation analysis in IQ mismatched
multi-carrier direct receivers is given in [4], in which theoret- = 2Re{(s(t)AT + s∗ (t)B∗T )e jwc t } (3)
ical symbol error rate of M-ary QAM modulated orthogonal where notation Re{x} denotes the real part of x and s(t) is the
frequency division multiplexing (OFDM) signal in multipath source signal.
fading channel is derived. Besides analytical study, several At receiver, RF signal interfered by multipath fading chan-
schemes have been proposed to solve the IQ imbalance prob- nel is represented as
lem. In [5], [6], the IQ imbalance parameters are estimated by
solving 2 linear complex equations with the aid of a known r(t) = xT (t) ∗ h(t) + wn (t) (4)
training sequence or pilot symbols. The received signal is where notation ∗ denotes linear convolution, h(t) is the channel
then corrected by algebra calculation. The adaptive signal impulse response(CIR), and wn (t) is additive white Gaussian
processing, such as least-squares and least mean squares noise(AWGN). The received signal is then down-converted to
(LMS) algorithms, is another approach for resolving the IQ baseband directly. Similar to (1) and (2), the receiver local
imbalance compensation [7]. For these approaches, perfect oscillator(LO) signal is
channel state information is assumed to be available at the re-
ceiver. An unified compensation scheme for different OFDM- yLO (t) = (1 + εR )cos(wct + φR ) − j(1 − εR )sin(wct − φR )
based systems is derived in [8]. In addition to TX or RX IQ = AR e− jwc t + BR e jwc t (5)

978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 1


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
si(t)

s(t)

sq(t)

I(t)

(Iˆ, Hˆ )
Q(t)
yi(t)

yq(t)

Fig. 1. The block diagram of the proposed self-calibration system for IQ imbalance.

y ( n) yi (n) I T

z * ( n) y q (n)

HˆT

Fig. 2. The block diagram of the proposed IQ imbalance estimation scheme within the self-calibration system.

where and worldwide inter-operability for microwave ac-


 cess(WiMAX) systems, IQ imbalance at transmitter is
AR = cos(φR ) − jεR sin(φR )
(6) usually negligible. This is because base stations are
BR = εR cos(φR ) + jsin(φR )
implemented with superior modules, while cost-effective
, εR and φR are amplitude and phase imbalance between components are with terminals. In this case, AT  1 and
the receiver I and Q branches, respectively. Two receiver IQ BT  0.
imbalance coefficients, AR and BR , are used to model the IQ 2) Similarly, IQ imbalance at receiver is negligible in the
imbalance system. From (4)-(6), the direct down-converted uplink (UL) of a centralized network. In this case, AR 
signal is represented as 1 and BR  0.
y(t) = LP[2(AR e− jwc t + BR e jwc t )(xT (t) ∗ h(t) + wn (t))] 3) In a mesh or ad-hoc network, such as some wireless
local area networks (WLAN) and wireless sensor net-
= (AR h(t)AT + BR h∗ (t)BT ) ∗ s(t)
work(WSN), both transmitters and receivers are usually
+(BR h∗ (t)A∗T + AR h(t)B∗T ) ∗ s∗ (t) implemented with cost-effective components. Therefore,
+AR w(t) + BR w∗ (t) (7) IQ mismatch at both transmitters and receivers must be
considered.
where notation LP[.] denotes the low pass filtering, w(t) is
baseband AWGN with zero mean and variance σ2w . After
III. J OINT E STIMATION AND C OMPENSATION S CHEMES
analog-to-digital conversion, the signal is transformed from
FOR TX AND RX IQ I MBALANCE
the time domain to the frequency domain at the FFT block
and (7) becomes A. Self-Calibration at Transmitter

Y (k) = (AR H(k)AT + BR H−1 (k)BT )S(k) Although commercial transceivers are produced to meet
+(BR H−1 (k)AT + AR H(k)B∗T )S−1
∗ ∗ ∗
(k) the requirements of specifications, the accuracy will change
∗ as time goes by. Furthermore, different environment factors,
+ARW (k) + BRW−1 (k). (8)
such as temperature and humidity, will give rise to different
Here, H(k) is channel frequency response(CFR) of the kth accuracies. To solve these problems, we propose the self-
sub-carrier, S(k) is the source data in the frequency domain, calibration scheme. Consider the simplified quadrature direct
S−1 (k) is the mirror of S(k) with respect to the center sub- conversion transceiver in Fig. 1, in which the upper arm is
carrier, H−1 (k) is the mirror CFR, and W−1 (k) is the mirror the transmission path and the lower arm is the reception path.
noise. For self-calibration stage, the switches are put at ”S1” and a
training sequence is generated by the Baseband Module. At the
B. Remarks upper arm, the training sequence is multiplied by the reference
Three different kinds of IQ imbalance systems are discussed signal generated by LO block and then up-converted to RF
below. directly. Instead of emitting the RF signal, the system loops
1) In the downlink (DL) of a centralized network, such the training signal back to the lower arm. The training signal
as wideband code division multiple access(WCDMA) is then down-converted directly to baseband using the same

978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 2


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
reference signal. A sophisticated analog circuit is required for where Y (k) is the output of FFT block. By substituting of (6)
RF signal feedback. and (8) into above equation, we could derive the estimators of
The proposed IQ imbalance estimator scheme for self- εR and φR , as shown below.
calibration is shown in Fig.2. Suppose the complex-valued 
√ 1 − 1 − 4ℜ{Δ}
training signal z(n) = 1/ 2pn (1 + j) is transmitted, where ε̂R = −1 (17)
{p0 , p1 , ..., pN−1 } is PN sequence and N is the length of PN 2ℜ{Δ}
sequence. In Fig.2, the input signal,y(n), is multiplied by z∗ (n) 1 (1 + ε̂R )
φ̂R = arcsin{2ℑ{Δ} }. (18)
and then the IQ signals are split for the following process. 2 (1 − ε̂R )
Since same reference signal is used at mixer, εT = εR and In practical, ensemble average in (16) is implemented by time
φT = φR . By substituting of (2) and (6) into (7), we have average over N samples. It is noted that (17) is valid if and only
i (n)
yi (n) = (1 + ε2T ) + (1 − ε2T )sin(2φT ) + w (9) if ℜ{Δ} ≤ 1/4. This condition is held when εR ≤ 1. Therefore,
(17) is valid for general cases. Furthermore, since εR ≤ 1,
 
yq (n) = −2εT + wq (n) (10)
the other estimator of amplitude imbalance from second order
and equation is invalid. It is noted that the estimators in (17) and
1 (18) are standard-independent and applicable for both initial
i (n)= √ pn {{(1 + εT )cos(φT ) + (1 − εT )sin(φT )}wi (n)
w acquisition and tracking. By substituting of (17) and (18) into
2
(6), we obtain the estimations of AR and BR .
+{(1 − εT )cos(φT ) + (1 + εT )sin(φT )}wq (n)} (11)
Next, we continue to estimate the CFR. The output of FFT
1
q (n)= √ pn {{(1 − εT )sin(φT ) − (1 + εT )cos(φT )}wi (n)
w block is modeled as
2
+{(1 − εT )cos(φT ) − (1 + εT )sin(φT )}wq (n)}. (12) Y0 (k) = U{P0 (k)H(k) + W0 (k)}, 0 ≤ k ≤ (N/2 − 1) (19)
∗ (k)]T are the received training symbols
where Y0 = [Y (k) Y−1
Since wi (n) and wq (n) are linear combinations of wi (n) and

, H(k) = [H(k) H−1 (k)]T , W0 = [W (k) W−1 ∗ (k)]T ,
wq (n), they are also Gaussian distributed with zero mean and
variance  
AR BR
σ2w U= (20)
1 B∗R A∗R
σ2wi {(1 + ε2T ) + (1 − ε2T )sin(2φT )}
=
2 2
and
σ2w 1  
σ2wq = {(1 + ε2T ) − (1 − ε2T )sin(2φT )}. (13) P(k) 0
2 2 P0 (k) = . (21)
0 ∗ (k)
P−1
The maximum likelihood estimator(MLE) of εT is [10] [11]
−1 P(k) is the known training symbol at the kth sub-carrier. It
ε̂T =
yq (n)].
E[ (14) is noted that Y−1 (k) = Y (N − k) for 1 ≤ k ≤ (N/2 − 1) and
2
Y−1 (0) = Y (N/2). Therefore, MLE of H(k) is
According to the invariance property of the MLE [11], the
MLE of φT is  = P−1 −1
H(k) 0 (k)U Y0 (k) (22)
1 −1 E[yi (n)] − ε̂2T −1 where AR and BR are estimated from (17) and (18). At the
φ̂T = sin ( ). (15)
2 1 − ε̂2T acquisition stage, preambles are utilized for channel estimation
It is observed from (14) and (15) that the MLEs for amplitude and (22) is applicable to general OFDM systems. At the
and phase imbalances are asymptotically unbiased. The mean tracking stage, only part of sub-carriers in one OFDM symbol
squared error(MSE) of ε̂T is σ2wq /(4N). The MSE of φ̂T is is assigned for pilots. In this case, pilots must be assigned
symmetrically with respect to center sub-carrier. One of this
MSEφT = E[|φT − φ̂T |2 ] ≈ σ2wi /{4N(1 − ε2T )2 }. Finally, ε̂T and
kind of system is IEEE 802.11a/g.
φ̂T are used to compensate the IQ imbalance at the Local
Oscillator block. It is noted that the initial self-calibration is C. Equalization at Receiver
a standard-independent scheme. In this subsection, we study three different equalization
B. Receiver IQ Imbalance schemes. The first one is direct zero forcing (DZF) scheme
[9], as shown below:
After self calibration process, IQ imbalance at transmitter
becomes negligible and we continue to cope with receiver IQ  = Y (k)
S(k) (23)
imbalance. We propose the receiver IQ imbalance estimation AR H(k)
scheme, which is modified from [8] and applicable for both where AR and H(k) are estimated from last subsection. For
initial acquisition and tracking. Consider the following equa- DZF scheme, image signal is regarded as interference and not
tion compensated.
E[Y (k)Y−1 (k)] The second scheme is joint detection equalizer(JDE) [5].
Δ= ∗ (k)|2 ] (16)
E[|Y (k) +Y−1 This approach is similar to the channel estimation scheme

978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 3


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
shown in last subsection. The output of FFT block is modeled Fig.3 that the simulated MSE is close to the theoretical value
as derived in Section II A. Furthermore, longer training sequence
results in smaller MSE. When the length of training sequence
Y1 (k) = V(k)S0 (k) + W0 (k), 0 ≤ k ≤ (N/2 − 1) (24)
is 4096 and SNR is 10dB, IRR of the compensated transmitter
where Y1 = [Y (k) Y−1 ∗ (k)]T are the received data symbols, is raised from 13.5dB to 52dB. In general, IQ imbalance at

S0 (k) = [S(k) S−1 (k)]T are the original data symbols, V(k) = transmitter becomes negligible after self-calibration.
UH1 (k), and Figure 4 depicts MSE of the channel estimation scheme
  illustrated in Section II B. Figure 4(a) shows the results in
H(k) 0
H1 (k) = ∗ (k) . (25) ITU pedestrian channel model A and Fig.4(b) in vehicular
0 H−1
channel model A. The MLE without considering image signal,
The JDE of S0 (k) is derived from the above linear model and 
i.e. H(k) = (AR P(k))−1Y (K), is also plotted for comparison.
is given by We denote this estimator as MLE0 . When SNR is 40 dB,
 the MSE of the proposed scheme is about 10−4 for these
S0 (k) = H−1 −1
1 (k)U Y1 (k) (26)
two channel models. However, estimator without considering
where H1 (k) is estimated from (22). image signal performs poor for small IRR. It is also observed
The third equalization scheme is the proposed successive that the proposed channel estimation scheme works well for
interference cancelation(SIC). The SIC scheme is illustrated both large and small IQ imbalance.
below: Figure 5 shows SER of IQ imbalance OFDM receiver
1) Compare the magnitude of H(k) and H−1 (k) and detect in ITU pedestrian channel model A. The system without
the sub-carrier with larger channel gain is detected first, IQ imbalance and with ICI is also plotted for comparison.
as shown below: The IRR is 36.8dB in Fig.5(a) and 13.6dB in Fig.5(b). The
 (k) results of QPSK and 64QAM are plotted on the same figure
Ŝ(k) = Q( ARYH(k) ), i f |H(k)| ≥ |H−1 (k)| for comparisons. The OFDM receiver without IQ mismatch,
Y (k) (27)
Ŝ−1 (k) = Q( AR H (k) ), i f |H(k)| < |H−1 (k)| which is labeled as Ideal, is also shown. It is observed from
−1
Fig.5 that performance of SIC is close to JDE and better
where notation Q(.) denotes the quantization function
than DZF. Figure 6 shows SER of IQ imbalance OFDM
for signal detection.
2) Then the detected signal is subtracted from the received receiver in ITU vehicular channel model A. The IRR is 36.8dB
signal, as shown below: for Fig.6(a) and 13.6dB for Fig.6(b). The simulation results
 are similar to that of Fig. 5. It is observed from Fig.5 and
Y−1 (k) = Y−1 (k) − BR H ∗ (k)Ŝ∗ (k), i f |H(k)| ≥ |H−1 (k)|
(28) Fig.6 that JDE and SIC schemes work well for both QPSK
Y (k) = Y (k) − BR H−1
∗ (k)Ŝ∗ (k), i f |H(k)| < |H (k)|
−1 −1
and 64-QAM in the frequency selective fading channel. In
3) Finally, detect the remaining signal, as shown below: compensating Rx IQ imbalance, we assume Tx IQ imbalance

⎨ Ŝ (k) = Q( Y−1 (k) ), i f |H(k)| ≥ |H (k)| is negligible. However, there might be no self-calibration at
−1 AR H−1 (k) −1
(29) Tx or the transmitter is calibrated incorrectly. We consider this
⎩ Ŝ(k) = Q( Y(k) ), i f |H(k)| < |H−1 (k)| case in Fig. 7. The Tx IRR is 36.8dB in Fig.7(a) and 13.6dB in
AR H(k)
Fig.7(b). The IRR at Rx is 36.8dB. It is shown that the effect of
IV. P ERFORMANCE A NALYSIS AND S IMULATION R ESULTS
mismatched Tx in Fig.7(a) is little. On the other hand, severe
A. Computer Simulation Results performance degradation occurs in Fig.7(b). Therefore, Tx
Regarding the IQ imbalance effects, image rejection ra- IQ imbalance must be compensated before transmission. The
tio(IRR) is usually utilized to assess the impacts on system proposed self-calibration scheme is designed for this purpose.
performance. IRR is the power ratio of the desired and image
signal. The definition of IRR is [2] B. Complexity Comparisons
AR AT + BR BT 2
IRR = | | (30) In complexity evaluation, division is considered as a kind of
BR A∗T + AR B∗T
multiplication. Furthermore, the trigonometric and square root
where channel effects are not taken into account. AT , BT , AR operation are carried out by the look-up table. The complexity
and BR are the IQ imbalance parameters shown in (2) and (6) of three different equalization schemes is shown in Table I, in
for transmitter and receiver, respectively. The ITU pedestrian which Nu is the number of useful sub-carriers. It is shown that
channel model A and vehicular channel model A [13] are the complexity of SIC is less than JDE and close to DZF. This
simulated for performance evaluation. Figure 3 shows MSEs is because matrix inversion is required for JDE scheme.
of the amplitude and phase imbalance estimation using the
proposed self-calibration scheme at transmitter. The amplitude TABLE I
and phase imbalance simulated here are 1dB and 50 , which COMPLEXITY OF THREE EQUALIZATION SCHEMES
are the maximum imbalance value of off-the-shelf components Complexity DZF JDE SIC
[1]. The definition of amplitude imbalance is shown in [2]. For Multiplications 2Nu 5Nu +6 3Nu
Additions 0 Nu +1 Nu /2
this scenario, the IRR is evaluated to be 13.6dB. It is shown in

978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 4


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
−1 −1 0 0
10 10 10 10
Simulated,N=64 Simulated,N=64
Simulated,N=256 Simulated,N=256
Simulated,N=1024 Simulated,N=1024
−2 Simulated,N=4096 −2 Simulated,N=4096
10 Theoretical,N=64 10 Theoretical,N=64
Theoretical,N=256 Theoretical,N=256
Theoretical,N=1024 Theoretical,N=1024
Mean Squared Error −1 −1

Mean Squared Error


Theoretical,N=4096 Theoretical,N=4096 10 10
−3 −3
10 10

SER

SER
−4 −4
10 10
−2 −2
10 10

−5 −5
10 10 DZF, 64QAM DZF, 64QAM
JDE, 64QAM JDE, 64QAM
SIC, 64QAM SIC, 64QAM
Mismatched Tx, DZF, 64QAM Mismatched Tx, DZF, 64QAM
Mismatched Tx, JDE, 64QAM Mismatched Tx, JDE, 64QAM
−6 −6 Mismatched Tx, SIC, 64QAM Mismatched Tx, SIC, 64QAM
10 10 −3 −3
0 5 10 0 5 10 10 10
0 10 20 30 40 0 10 20 30 40
SNR(dB) SNR(dB) SNR(dB) SNR(dB)
(a) (b)

Fig. 7. The effects of mismatched Tx on symbol error rate in ITU vehicular


Fig. 3. Mean squared error of the proposed self-calibration scheme at
channel model A: (a)IRR at Tx is 36.8dB and (b)IRR at Tx is 13.6dB.
transmitter for different training sequence lengths: (a) amplitude estimation
and (b) phase estimation.
Based on this model, the end-to-end IQ imbalance estimation
10
1
IRR=36.8dB, proposed
10
1
IRR=36.8dB, proposed
and compensation schemes are studied. Computer simulations
IRR=36.8dB, MLE0 IRR=36.8dB, MLE0
IRR=13.6dB, proposed
IRR=13.6dB, MLE0
IRR=13.6dB, proposed
IRR=13.6dB, MLE0
are conducted to evaluate the performance of the proposed
10
0
10
0
schemes. By the proposed self-calibration scheme and 4096
training samples, IRR at transmitter is raised from 13.6dB
Mean Squared Error

Mean Squared Error

−1 −1
10 10
to 52dB. Similar enhancement is obtained from receiver IQ
−2 −2
imbalance estimation scheme. Based on the estimated im-
10 10
balance parameters, DZF, JDE and SIC detection schemes
−3 −3
are designed for channel equalization. It is shown that the
10 10
performance of SIC is close to JDE and better than DZF.
10
−4
10
−4
Furthermore, the implementation complexity of SIC is lower
0 10 20
SNR(dB)
30 40 0 10 20
SNR(dB)
30 40
than JDE. Therefore, SIC is suggested for equalization in the
Fig. 4. Mean squared error of the channel estimation for IQ mismatched IQ mismatched OFDM receiver.
receiver: (a)ITU pedestrian channel model A and (b)ITU vehicular channel
model A. R EFERENCES
[1] B. Razavi, RF Microelectronics. Upper Saddle River, NJ: Prentice Hall,
10
0
10
0
1998.
64QAM [2] C-L Liu, “Impacts of I/Q imbalance on QPSK-OFDM-QAM detection,”
10
−1
64QAM
10
−1 IEEE Trans. on Consum. Electron., vol. 44, no. 3, Aug. 1998, pp. 984-
989.
−2 QPSK −2 QPSK
[3] B. Razavi, “Design considerations for direct-converstion receivers,” IEEE
10 10
Trans. on Circuits and System II, vol. 44, no. 6, pp. 428-435, Jun. 1997.
SER

SER

[4] M. Windisch and G. Fettweis, “Performance degradation due to I/Q


−3 −3
10
Ideal, QPSK
10
Ideal, QPSK
imbalance in multi-carrier direct converstion receivers: a theoretical
DZF, QPSK
JDE, QPSK
DZF, QPSK
JDE, QPSK analysis,” in Proc. IEEE Int. Conf. Commun. (ICC), Istanbul, Turkey, pp.
SIC, QPSK
10
−4 SIC, QPSK
Ideal, 64QAM
10
−4
Ideal, 64QAM 257-262, June 2006.
DZF, 64QAM DZF, 64QAM
JDE, 64QAM JDE, 64QAM [5] A. Tarighat, R. Bagheri, and A. H. Sayed, “Compensation schemes and
SIC, 64QAM SIC, 64QAM
10
−5

0 10 20 30 40
10
−5

0 10 20 30 40
performance analysis of IQ imbalances in OFDM receivers,” IEEE Trans.
SNR(dB) SNR(dB)
Signal Proc., vol. 53, no. 8, pp. 3257-3268, Aug. 2005.
Fig. 5. Symbol error rate of the IQ mismatched receiver in ITU pedestrian [6] J. Tubbax, B. Côme, L. V. Perre, S. Donnay, M. Eagles, H. D. Man, and
channel model A: (a)IRR=36.8dB and (b)IRR=13.6dB. M. Moonen, “Compensation of IQ imbalance and phase noise in OFDM
systems,” IEEE Trans. Wireless Commun., vol. 4, no. 3, pp. 872-877,
0
10
0
10 May 2005.
[7] A. Schuchert and R. Hasholzner, “A novel IQ imbalance compensation
−1
10
64QAM −1
10
64QAM scheme for the reception of OFDM signals,” IEEE Trans Consum.
Electron., vol. 47, no. 3, pp. 313-318, Aug. 2001.
−2
QPSK
−2
QPSK
[8] M. Windisch and G. Fettweis, “Standard-independent I/Q imbalance
10 10
compensation in OFDM direct-conversion receivers,” in Proc. 9th
SER

SER

International OFDM Workshop, Dresden, Germany, Sep. 2004, pp. 57-61.


−3 −3
10
Ideal, QPSK
10
Ideal, QPSK
[9] Tim C. W. Schenk, Peter F. M. Smulders, and Erik R. Fledderus, “Es-
DZF, QPSK
JDE, QPSK
DZF, QPSK
JDE, QPSK
timation and compensation of frequency selective TX/RX IQ imbalance
−4
10
SIC, QPSK
Ideal, 64QAM
−4
10 SIC, QPSK
Ideal, 64QAM
in MIMO OFDM systems,” in Proc. IEEE Int. Conf. Commun. (ICC),
DZF, 64QAM
JDE, 64QAM
DZF, 64QAM
JDE, 64QAM
Helsinki, Finland, June 2006, pp. 251-256.
−5
10
0
SIC, 64QAM

10 20 30 40
−5
10
0
SIC, 64QAM

10 20 30 40
[10] M. Morelli and U. Mengali, “A comparison of pilot-aided channel
SNR(dB) SNR(dB)
estimation methods for OFDM systems,” IEEE Trans. Signal Proc., vol.
Fig. 6. Symbol error rate of the IQ mismatched receiver in ITU vehicular 49, no. 12, pp. 3065-3073, Dec. 2001.
channel model A: (a)IRR=36.8dB and (b)IRR=13.6dB. [11] S. M. Kay, Fundamentals of Statistical Signal Processing: Estimation
Theory. Upper Saddle River, NJ: Prentice Hall, 1993.
V. C ONCLUSIONS [12] J. G. Proakis, Digital Communications. 4th Ed., McGraw-Hill, Inc.,
2001.
In this paper, we study Tx-Rx IQ imbalance problems in [13] ITU “Guidelines for evaluation of radio transmission technologies for
OFDM systems. The Tx-Rx IQ imbalance model is setup. IMT-2000,” Rec. ITU-R M.1225, 1997.

978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 5


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

You might also like