You are on page 1of 5

1

Mobile WiMAX: Impact of channel estimation error on


the Performance of Limited Feedback Linear Precoding
Mai Tran, Andrew Nix, and Angela Doufexi
Centre for Communications Research, Merchant Venturers Building,
University of Bristol, Bristol BS8 1UB, UK

Abstract— The mobile WiMAX standard (802.16e) uses MIMO Mobile WiMAX system with numerous antennas,
multiple-input multiple-output (MIMO) limited feedback linear subcarriers, and a rapidly changing channel. The need to
precoding to exploit the channel state information at the reduce the amount of CSI feedback information motivates
transmitter. Although the performance of limited feedback
the use of a codebook based linear precoding technique [5,
linear precoding in relation to traditional open-loop MIMO-
OFDM has been extensively studied in the literature, these 6]. Here, the mobile station (MS) calculates the optimal
studies commonly assume perfect channel estimation at the precoding matrix for each subcarrier and feeds back the
receiver. In a practical OFDM-based system, the estimated matrix, rather than the CSI, to the base station (BS).
channel matrix often differs from the actual channel matrix due Specifically, the optimal precoding matrix is constrained to
to errors incurred in the channel estimation process. This one of N distinct matrices, which are referred to as codebook
results in degraded performance relative to the case with
entries, designed offline and known to both the MS and BS.
perfect channel estimation. To date, few researchers have
studied the impact of channel estimation error on the The MS identifies the optimal precoding matrix based on the
performance of an OFDM limited feedback linear precoding current CSI. Since the codebook is known at the BS, the MS
system. This paper investigates the channel estimation error only needs to feedback a binary index of the optimal
using 1) an MMSE channel estimator that takes into account precoding matrix, rather than the entire precoding matrix
the subcarrier correlation when estimating the channel, 2) a itself. For each combination of the number of transmit ( NT )
Low Rank (LR) channel estimator that relaxes the requirement
for a perfect channel covariance matrix in the MMSE receiver, and receive ( N R ) antennas, the 802.16e standard defines
and 3) a ZF estimator where this correlation information is two codebooks: one with 8 entries and the other with 64
ignored. Simulation results show that with the MMSE
entries [7]. These correspond to 3-bit and 6-bit codebook
estimator the system suffers very little array gain loss with a
performance degradation of 0.2dB SNR. Compared to the indices for each precoding matrix respectively.
MMSE estimator, the LR estimator incurs a small performance The performance improvement of codebook based linear
loss of around 0.5dB. Finally, when the ZF estimator is precoding MIMO-OFDM systems has been previously
implemented, a significant performance degradation is reported in the literature [8]. However, results are often
observed with approximately 4-5dB loss in array gain loss. based on the assumption that the channel is perfectly
Index Terms—802.16e, WiMAX, MIMO, linear precoding, estimated at the receiver. In practice, the OFDM channel
limited feedback. estimator at the receiver always leads to imperfect channel
I. INTRODUCTION estimation. If the channel is not perfectly known at the
receiver, the performance of the linear precoding technique
The first WiMAX systems were based on the IEEE will degrade due to two reasons: 1) the receiver will select
802.16-2004 standard [1]. This targeted fixed broadband the optimal precoding matrix based on the errored channel
wireless applications via the installation of Customer
estimation Ĥ , and not the true channel H, and 2) the use of
Premises Equipment (CPE). In December 2005 the IEEE
completed the 802.16e-2005 [2] amendment, which added errored channel information at the receiver (e.g., Ĥ is used
new features to support mobile applications. instead of H in the MMSE receiver of the SM system). This
Mobile WiMAX now supports both open-loop and paper investigates the impact of channel estimation error on
closed-loop multiple-input multiple-output (MIMO) the performance of a linear precoding mobile WiMAX
techniques. Open-loop techniques, such as space time block system using the distributed PUSC subcarrier permutation
coding (STBC) and spatial multiplexing (SM), can be used scheme.
to increase diversity gain or system throughput without the The paper is organized as follows: Section II describes
need for channel state information (CSI) at the transmitter. important parameters used in the mobile WiMAX simulator.
However, recent work [3, 4] has reported further increases in An overview of precoded spatial multiplexing and dominant
system performance (both diversity and array gain) and eigenbeamforming for mobile WiMAX systems is described
throughput by applying linear precoding closed-loop in Section III. Training-based channel estimation techniques
techniques at the transmitter that exploit knowledge of the supported in the mobile WiMAX standard are described in
CSI. Section IV. Section V investigates the impact of channel
The key idea behind linear precoding is to customize the estimation error on the performance of a precoded mobile
transmit signal by pre-multiplication with a precoding WiMAX system. Finally, conclusions are presented in
matrix. It is well-known that singular value decomposition Section VI.
(SVD) linear precoding provides the highest achievable II. LINK LEVEL MOBILE WIMAX SIMULATOR
performance [4]. However, the SVD approach requires A detailed downlink Mobile WiMAX link-level simulator
perfect CSI at the transmitter, which cannot be achieved in a
2

[9] using the PUSC subcarrier permutation and spread over NT transmit antennas by multiplying by an
convolutional coding with soft Viterbi decoding has been NT × M precoding matrix Fk ), and n k is an N R ×1 noise
implemented by the authors based on the 802.16e-2005
vector whose entries are complex, independent and
standard [2]. The simulator models a cell with an omni-
identically distributed (i.i.d) additive white Gaussian noise
directional basestation (BS) and three mobile stations (MS)
randomly situated in the cell. In the downlink, each MS is (AWGN) samples with zero mean and variance σ 2 .
randomly allocated 5 out of a total of 15 subchannels. The x1

BS transmits data simultaneously to 3 MS, with each sharing


a common OFDMA symbol. Table I summarises the
OFDMA parameters used in the Mobile WiMAX simulator. xNT
A detailed description of the simulator can be found in [9].
TABLE I: OFDMA PARAMETERS
y1
Parameter Value
Carrier frequency (GHz) 2.3
FFT size 512
yNR
Channel bandwidth (MHz) 5
Sampling frequency Fs (MHz) 5.6
Sampling period 1/ Fs (µs) 0.18 Fig. 1: Linear precoding spatial multiplexing system block diagram
Subcarrier frequency spacing ∆ f =Fs/NFFT (kHz) 10.94
In this paper the received symbol vector y k is decoded
Useful symbol period Tb = 1/ ∆ f (µs) 91.4
Guard Time Tg = Tb/8 (µs) 11.4
using an MMSE linear decoder G k , given by
-1
OFDMA symbol duration Ts =Tg+Tb (µs) 102.9   M σ n2   * *
Number of used subcarriers (Nused) 421 G k = Fk* H*k H k Fk +   I M  Fk H k . (2)
Number of pilot subcarriers 60
  Es  
Number of data subcarriers 360 The optimal precoding matrix Fopt is determined for each
Number of data subcarriers in each subchannel 24 subcarrier using the minimum mean square error (MSE)
Number of subchannels 15 criterion [5] as
−1
Number of users (Nusers) 3 Es  Es 
MSE ( Fk ) =  I M + Fk*H*k H*k Fk*  (3)
Number of subchannels allocated to each user (NALSU) 5 M  M σ n2 
Based on the ETSI 3GPP2 spatial channel model (SCM) where
[10, 11], urban micro and urban macro tapped delay line
(TDL) channels were generated for use in this analysis. The
Fopt = arg min trace MSE Fki
Fki ∈Q
( ( )) (4)
TDL comprises 6 taps with non-uniform delays. Each tap and Q is the codebook (which is known to both the BS and
experiences Rayleigh fading based on an MS velocity and
the traditional Jake’s Power Doppler Spectrum [12]. The MS). Q is constructed using the methods described in
antenna element separation is 10 λ at the BS and 0.5 λ at section 8.4.5.4.10.15 of [2].
the MS, where λ represents the carrier wavelength. B. Dominant Eigenbeamforming (DE)
III. LINEAR PRECODING The second linear precoding system considered in this
paper is dominant eigenbeamforming (DE), as illustrated in
This section summarizes two different linear precoding
Fig. 2 for the k-th subcarrier. Here the BS transmits a single
systems, namely linear precoding spatial multiplexing (SM
spatial stream across the NT transmit antennas.
PRE) and dominant eigenbeamforming (DE), both of which
are implemented in the mobile WiMAX simulator. x1

A. Linear precoding spatial multiplexing (SM PRE)


For purposes of simplicity, a generic linear precoding
spatial multiplexing system for a single subcarrier is xNT

illustrated in Fig. 1.
In the case of an OFDM mobile WiMAX system, the k-th
subcarrier is allocated a precoding matrix Fk , and the y1

N R × 1 receive symbol vector y k is given by


Es
yk = H k Fk s k + n k (1) yN R
M
where k is the subcarrier index, Es is the total transmit
power for the k-th subcarrier, M is the number of spatial Fig. 2: Dominant eigenbeamforming system block diagram
streams ( M < NT ), H k is the N R × NT normalised channel In a precoded mobile WiMAX system, the k-th subcarrier
matrix, s k is an M × 1 transmit data symbol vector (which is is assigned a NT × 1 precoding vector f k . The receive
3

symbol vector y k for a DE system can be expressed as where Rhh=E{hh*} denotes the auto-covariance matrix of
the channel vector h and I denotes the N × N identity
y k = Es H k fk sk + n k . (5)
matrix.
The main drawback of the MMSE estimator is that it
In this paper y k is decoded using a traditional maximum
requires a perfect channel covariance matrix at the receiver.
ratio combiner g [6]. In practice the receiver does not often have this information
g k = H k fk / H k f k 2 (6) in advance, and hence this too needs to be estimated. The
The optimal precoding vector fopt is determined from (7) work in [17] proposes a low-rank MMSE estimator (LR)
that uses the channel covariance matrix estimated from a
using the criterion defined in [6].
2
uniform power-delay profile (pdf). The estimated correlation
fopt = arg max H k f ki (7) between the m-th and the n-th subcarrier in this case is given
2
fki ∈Q
by
IV. CHANNEL ESTIMATION IN MOBILE WIMAX 1, if m = n
 m− n
In order to perform channel estimation, the mobile  − j 2π L
rm ,n = 1 − e N , if m ≠ n (11)
WiMAX standard supports a training-based technique [2,  m−n
13] where known symbols are transmitted to aid the  j 2π L
receiver’s channel estimation algorithm. There are two ways  N
to transmit training symbols: 1) transmitting preamble-based where L is the number of samples in the guard interval, and
symbols in which known preambles are sent at the beginning N is the number of subcarriers within an OFDM symbol. It
of each frame, and 2) transmitting pilot-based symbols can be seen that this estimator only requires knowledge of
where several known pilots are inserted into each OFDM the guard interval length and the number of subcarriers in
symbol within a frame in order to track the changing channel the system. Results in [17] also show that the wrong channel
between OFDM symbols. statistics (due to the use of a uniform pdf) only incurs a
Our mobile WiMAX simulator assumes a block fading small performance loss relative to the case with perfect
channel where the channel remains constant over a WiMAX knowledge of the covariance matrix.
transmission frame, but changes between frames. Therefore In a MIMO-OFDM system the received signal at each
only the preamble is needed in our simulator to estimate the receive antenna is the superposition of transmit signals from
channel. With a frame duration of 5ms, as defined in the NT transmit antennas. Therefore, in order to differentiate
standard, this block fading assumption is valid for mobile the preamble signals transmitted from each antenna, an
applications with velocities up to 80 km/h (i.e., a coherence independent pattern for transmitting preamble signals [18] is
time of 6 ms). A preamble-based OFDM channel estimation implemented as illustrated in Fig. 3. The independent pattern
system with N subcarriers is often modelled as transmits preamble signals from each antenna at a time when
y = Xh + n (8) the other antennas keep silent. By doing this all the
where X is an N × N diagonal matrix whose diagonal preambles are received at the receiver without interfering
elements are the pilot symbols in the frequency domain, h is with one another.
an N × 1 complex channel vector whose entries are the
frequency response of N subcarriers, and n is an N × 1 noise
vector of independent and identically distributed complex,
zero-mean Gaussian noise variables with variance σ 2 .
Without loss of generality, we assume that the channel is
normalised such that E hk{ } = 1 and E { X } = 1 . The
2
k ,k
2

channel estimate ĥ can be obtained using the zero forcing


(ZF) or the minimum mean-square error (MMSE) channel
estimator [14-16]. For example, by using the ZF channel
Fig. 3: Independent pattern for transmitting preambles [18] for a MIMO
estimator, the channel estimate vector ĥ is given by system with 2 transmit antennas
hˆ ZF = X −1y . (9)
1.4
The ZF channel estimator is implemented with very low
1.2
complexity but fails to consider the potentially significant
correlation between subcarriers, and therefore suffers from a 1
Channel amplitude

high mean-square error [14]. In order to improve the quality 0.8

of the channel estimate, an MMSE based channel estimator 0.6

[14, 16] that minimizes the mean-square error by leveraging 0.4 Perfect channel
the subcarrier correlation, can be used. The MMSE channel Estimation at 0dB SNR
0.2 Estimation at 10dB SNR
estimate hˆ MMSE in the frequency domain is given by [14] Estimation at 20dB SNR
0

( )
0 50 100 150 200 250 300 350 400
−1
hˆ MMSE = R hh R hh + σ 2 I hˆ ZF (10) Subcarrier index

Fig. 4: MMSE channel estimation at various SNRs


4

Fig. 4 and Fig. 5 compare the actual channel with the estimator demonstrated in [17]. Finally, the ZF channel
channels estimated using the MMSE and ZF estimators, estimation scheme suffers significant performance
correspondingly, for various SNR values. It can be seen that degradation with a 4dB loss in array gain.
the MMSE estimator achieves a very accurate channel Fig. 8 illustrates the impact of MMSE, LR and ZF channel
estimate. It can maintain a reasonable channel estimate estimations on the PER performance of a 2 × 2 QPSK and
accuracy even at 0dB SNR. However when the ZF estimator 16QAM 3/4 rate Alamouti system [19]. Comparing Fig. 7
is used the channel estimate performance becomes very poor with Fig. 8 it can be seen that different channel estimation
at low SNRs. At an SNR of 0dB the ZF channel estimate is algorithms have a similar impact on the performance of
unusable. closed-loop and open-loop MIMO diversity systems. This
3
similar performance degradation initially seems counter-
Perfect channel intuitive because the channel estimation in a closed-loop
Estimation at 0dB SNR
2.5
Estimation at 10dB SNR MIMO system causes both an incorrect precoding matrix
2
Estimation at 20dB SNR
fˆopt at the transmitter and an errored channel Ĥ at the
Channel amplitude

receiver. This is expected to result in a higher performance


1.5
degradation than the open-loop MIMO system, where only
1 an errored channel Ĥ is experienced at the receiver. In fact,
a closed-loop MIMO system in this case is equivalent to an
0.5
open-loop MIMO system operating over a channel Hfˆ opt
0
0 50 100 150 200 250 300 350 400 ˆ ˆ at the receiver. This
with an errored channel estimate Hf
Subcarrier index opt
Fig. 5: ZF channel estimation at various SNRs results in a similar performance loss to the open-loop
Fig. 6 compares the mean-squared error (MSE) between system.
the MMSE, LR and ZF estimators. As expected, the MSE of 0
10 QPSK 3/4 DE PRE 2x2 Perf H
the MMSE and LR estimators are much smaller than that of 16QAM 3/4 DE PRE 2x2 Perf H
QPSK 3/4 DE PRE 2x2 MMSE H
the ZF estimator. 16QAM 3/4 DE PRE 2x2 MMSE H
0 -1 QPSK 3/4 DE PRE 2x2 LR uni
10 10 16QAM 3/4 DE PRE 2x2 LR uni
LR uniform estimator QPSK 3/4 DE PRE 2x2 ZF H
PER

MMSE estimator 16QAM 3/4 DE PRE 2x2 ZF H

ZF estimator
-2 -2
10 10
MSE

-4 -3
10 10
-4 -2 0 2 4 6 8 10 12 14 16 18
SNR (dB)
Fig. 7: PER performance of 2x2 DE QPSK and 16QAM 3/4 rate systems
-6
with MMSE, LR and ZF channel estimation
10 0
0 5 10 15 20 25 30 35 40 10 QPSK 3/4 ST22 Perf H
SNR (dB) 16QAM 3/4 ST22 Perf H
QPSK 3/4 ST22 MMSE H
Fig. 6: MSE of MMSE, LR and ZF estimators 16QAM 3/4 ST22 MMSE H
QPSK 3/4 ST22 LR uni
-1
10 16QAM 3/4 ST22 LR uni
V. SYSTEM PERFORMANCE ANALYSIS QPSK 3/4 ST22 ZF H
16QAM 3/4 ST22 ZF H
PER

This section studies the packet error rate (PER)


performance of the precoded MIMO mobile WiMAX 10
-2

system with MMSE, LR and ZF channel estimation. Linear


precoding with channel estimation errors can be simulated
by assuming that the receiver selects the optimal precoder 10
-3

( )
-2 0 2 4 6 8 10 12 14 16 18 20
matrix F = f Hˆ using knowledge of the channel estimate SNR (dB)
Fig. 8: PER performance of 2x2 Alamouti QPSK and 16QAM 3/4 rate
matrix Ĥ , and not the true channel H. systems with MMSE, LR and ZF channel estimation
0
The ideal case with perfect channel knowledge at the 10 QPSK 3/4 DE PRE 4x2 Perf H
16QAM 3/4 DE PRE 4x2 Perf H
receiver (denoted as Perf H) is demonstrated as a benchmark QPSK 3/4 DE PRE 4x2 MMSE H
16QAM 3/4 DE PRE 4x2 MMSE H
for PER comparison. QPSK 3/4 DE PRE 4x2 LR uni
-1 16QAM 3/4 DE PRE 4x2 LR uni
10
A. Dominant Eigenbeamforming QPSK 3/4 DE PRE 4x2 ZF H
16QAM 3/4 DE PRE 4x2 ZF H
PER

Fig. 7 shows the PER performance of the 2 × 2 QPSK


and 16QAM 3/4 rate DE systems using MMSE, LR and ZF 10
-2

channel estimators. It can be seen that the performance for


the MMSE channel estimate is quite close to that of the ideal
channel system (degraded by approximately 0.2dB). The LR 10
-3

-6 -4 -2 0 2 4 6 8 10 12 14 16
estimator, although using inaccurate channel statistics, only SNR (dB)
suffers an approximate 0.5dB loss relative to the MMSE Fig. 9: PER performance of 4x2 DE QPSK and 16QAM 3/4 rate systems
approach. This result agrees with the performance of the LR with MMSE, LR and ZF channel estimation
5

Fig. 9 illustrates the PER performance of a 4 × 2 QPSK compared to the ideal system with perfect channel
and a 16QAM 3/4 rate DE system using MMSE, LR and ZF knowledge. The LR channel estimator, based on a uniform
channel estimators. It demonstrates a performance channel correlation, only incurred a small loss of
degradation of 0.2dB, 1dB, and 5dB for MMSE, LR and ZF performance (i.e., 0.5dB) relative to the MMSE, even
systems respectively. though it used incorrect channel statistics. Finally, a ZF
B. Linear precoding spatial multiplexing estimator with no channel statistic information degraded
system performance by 4-5dB.
Fig. 10 studies the impact of MMSE, LR and ZF
estimators on the PER performance of 4 × 2 SM PRE QPSK VII. ACKNOWLEDGEMENTS
and 16QAM 3/4 rate systems. It presents the same The authors would like to thank the Centre for
performance degradation as observed for the DE system. Communications Research for providing a range of high-
Compared to the ideal system with perfect channel performance computing facilities.
knowledge, the use of an MMSE estimator only incurs a
small array gain loss of 0.2dB. The LR estimator suffers VIII. REFERENCES
very little loss relative to the MMSE estimator, and the ZF [1] "IEEE Std 802.16-2004 (Revision of IEEE Std 802.16-2001)," pp.
0_1-857, 2004.
estimator degrades by approximately 4dB in array gain. [2] "IEEE Std 802.16e-2005 and IEEE Std 802.16-2004/Cor 1-2005,"
0
10 QPSK 3/4 SM PRE 4x2 Perf H
pp. 0_1-822, 2006.
16QAM 3/4 SM PRE 4x2 Perf H [3] A. Scaglione, P. Stoica, S. Barbarossa, G. B. Giannakis, and H.
QPSK 3/4 SM PRE 4x2 MMSE H Sampath, "Optimal Designs for Space-Time Linear Precoders and
16QAM 3/4 SM PRE 4x2 MMSE H
Decoders," IEEE Transactions on Signal Processing, vol. 50, pp.
QPSK 3/4 SM PRE 4x2 LR uni
-1
10 16QAM 3/4 SM PRE 4x2 LR uni
1051-1064, 2002.
QPSK 3/4 SM PRE 4x2 ZF H [4] E. Sengul, E. Akay, and E. Ayanoglu, "Diversity Analysis of Single
and Multiple Beamforming," IEEE Transactions on
PER

16QAM 3/4 SM PRE 4x2 ZF H

Communications, vol. 54, pp. 990-993, 2006.


-2
10 [5] D. J. Love and R. W. Heath, "Limited Feedback Unitary Precoding
for Spatial Multiplexing Systems," IEEE Transactions on
Information Theory, vol. 51, pp. 2967-2975, 2005.
[6] D. J. Love, R. W. Heath, Jr., and T. Strohmer, "Grassmannian
-3
10 beamforming for multiple-input multiple-output wireless systems,"
0 5 10 15 20 25 IEEE Transactions on Information Theory, vol. 49, pp. 2735-2747,
SNR (dB)
2003.
Fig. 10: PER performance of 4x2 SM PRE QPSK and 16QAM 3/4 rate [7] Q. Li, X. E. Lin, and J. Zhang, "MIMO Precoding in 802.16e
systems with MMSE, LR, and ZF channel estimation WiMAX," Journal of Communications and Networks, vol. 9, 2007.
[8] E. Akay, E. Sengul, and E. Ayanoglu, "Performance Analysis of
Fig. 11 demonstrates the PER performance of a 2 × 2 Beamforming for MIMO OFDM with BICM," IEEE International
Conference on Communications, vol. 1, pp. 613-617, 2005.
open-loop SM system using different channel estimators. As
[9] D. H. Mai Tran, Andrew Nix, Angela Doufexi, Mark Beach, "Mobile
expected, a similar performance loss to the SM PRE system WiMAX: Downlink Performance Analysis with Adaptive MIMO
(Fig. 10) is observed. Switching," presented at IEEE Mobile WiMAX Symposium 2009,
10
0 California, USA, 2009.
QPSK 3/4 SM 2x2 Perf H
16QAM 3/4 SM 2x2 Perf H
[10] 3GPP, "Spatial channel model for multiple input multiple output
QPSK 3/4 SM 2x2 MMSE H (MIMO) simulations (TR 25.996 Rel. 6)," 2003.
16QAM 3/4 SM 2x2 MMSE H [11] G. Calcev, D. Chizhik, B. Goransson, S. Howard, H. Huang, A.
QPSK 3/4 SM 2x2 LR uni
10
-1
16QAM 3/4 SM 2x2 LR uni
Kogiantis, A. F. Molisch, A. L. Moustakas, D. Reed, and H. Xu, "A
QPSK 3/4 SM 2x2 ZF H Wideband Spatial Channel Model for System-Wide Simulations,"
PER

16QAM 3/4 SM 2x2 ZF H Vehicular Technology, IEEE Transactions on, vol. 56, pp. 389-403,
2007.
-2
10 [12] M. J. Gans, "A power-spectral theory of propagation in the mobile-
radio environment," Vehicular Technology, IEEE Transactions on,
vol. 21, pp. 27-38, 1972.
-3
[13] WiMAX Forum, "Mobile WiMAX - Part 1: A Technical Overview
10 and Performance Evaluation," August, 2006.
4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
SNR (dB) [14] J.-J. Van de Beek, O. Edfors, M. Sandell, S. K. Wilson, and P. O.
Borjesson, "On channel estimation in OFDM systems," presented at
Fig. 11: PER performance of 2x2 SM QPSK and 16QAM 3/4 rate systems Vehicular Technology Conference, 1995 IEEE 45th, 1995.
with MMSE, LR and ZF channel estimation [15] Y. Shen and E. Martinez, "Channel Estimation in OFDM Systems,"
Freescale Semiconductor AN3059, Rev. 0, 1/2006, 2006.
VI. CONCLUSIONS [16] D. Lowe and X. Huang, "Adaptive Low-Complexity MMSE Channel
Estimation for OFDM," presented at Communications and
This paper has studied the impact of channel estimation Information Technologies, 2006. ISCIT '06. International Symposium
error on the performance of a linear precoding Mobile on, 2006.
WiMAX system. Three different channel estimators were [17] O. Edfors, M. Sandell, J.-J. Van de Beek, S. K. Wilson, and P. O.
Borjesson, "OFDM channel estimation by singular value
implemented: 1) an MMSE estimator with a perfect channel decomposition," Communications, IEEE Transactions on, vol. 46,
covariance matrix, 2) an LR estimator with an estimated pp. 931-939, 1998.
channel covariance matrix, and 3) a ZF estimator where the [18] J. G. Andrews, A. Ghosh, and R. Muhamed, Fundamentals of
correlation factor is ignored. Spatial Multiplexing and WiMAX: Understanding broadband wireless networking, 1 ed:
Dominant Eigenbeamforming systems with 2 × 2 and 4 × 2 Prentice Hall, 2007.
[19] S. M. Alamouti, "A Simple Transmit Diversity Technique for
antenna configurations were studied. Results have shown Wireless Communications," IEEE Journal on Selected Areas in
that a linear precoding Mobile WiMAX system using Communications Sac, vol. 16, pp. 1451-1458, 1998.
MMSE channel estimation maintains a very good
performance with only a 0.2dB array gain degradation

You might also like