Anda di halaman 1dari 31

2ND QUARTER 2007, VOLUME 9, NO.

www.comsoc.org/pubs/surveys

CHANNEL ESTIMATION FOR


WIRELESS OFDM SYSTEMS
MEHMET KEMAL OZDEMIR, LOGUS BROADBAND WIRELESS SOLUTIONS, INC. AND
HUSEYIN ARSLAN, UNIVERSITY OF SOUTH FLORIDA
ABSTRACT
Orthogonal frequency division multiplexing (OFDM) is a special case of
multi-carrier transmission and it can accommodate high data rate requirement of multimedia based wireless systems. Since channel estimation is an
integral part of OFDM systems, it is critical to understand the basis of
channel estimation techniques for OFDM systems so that the most appropriate method can be applied. In this article, an extensive overview of channel estimation techniques employed in OFDM systems are presented. In
addition, the advantages, drawbacks, and relationship of these estimation
techniques with each other are analyzed and discussed. As the combination
of multiple input multiple output (MIMO)-OFDM systems promises higher
data rates, estimation techniques are further investigated for these systems.
Although the existing proposed techniques differ in terms of computational
complexity and their mean squared error (MSE) performance, it has been
observed that many channel estimation techniques are indeed a subset of
LMMSE channel estimation technique. Hence, based on a given systems
resources and specifications, a suitable method among the presented techniques can be applied.

riven by multimedia based applications, anticipated


future wireless systems will require high data rate
capable technologies. Novel techniques such as
OFDM and MIMO stand as promising choices for future high
data rate systems [1, 2].
OFDM divides the available spectrum into a number of
overlapping but orthogonal narrowband subchannels, and
hence converts a frequency selective channel into a nonfrequency selective channel [3]. Moreover, ISI is avoided by
the use of CP, which is achieved by extending an OFDM
symbol with some portion of its head or tail [4]. With these
vital advantages, OFDM has been adopted by many wireless standards such as DAB, DVB, WLAN, and WMAN [5,
6].
MIMO, on the other hand, employs multiple antennas at
the transmitter and receiver sides to open up additional subchannels in spatial domain. Since parallel channels are established over the same time and frequency, high data rates
without the need of extra bandwidth are achieved [7, 8]. Due
to this bandwidth efficiency, MIMO is included in the standards of future BWA [9]. Overall, these benefits have made
the combination of MIMO-OFDM an attractive technique for
future high data rate systems [1012].

18

1553-877X

As in many other coherent digital wireless receivers, channel estimation is also an integral part of the receiver designs
in coherent MIMO-OFDM systems [13]. In wireless systems,
transmitted information reaches to receivers after passing
through a radio channel. For conventional coherent receivers,
the effect of the channel on the transmitted signal must be
estimated to recover the transmitted information [14]. As long
as the receiver accurately estimates how the channel modifies
the transmitted signal, it can recover the transmitted information. Channel estimation can be avoided by using differential
modulation techniques, however, such systems result in low
data rate and there is a penalty for 34 dB SNR [15 19]. In
some cases, channel estimation at user side can be avoided if
the base station performs the channel estimation and sends a
pre-distorted signal [20]. However, for fast varying channels,
the pre-distorted signal might not bear the current channel
distortion, causing system degradation. Hence, systems with a
channel estimation block are needed for the future high data
rate systems.
Channel estimation is a challenging problem in wireless
systems. Unlike other guided media, the radio channel is highly dynamic. The transmitted signal travels to the receiver by
undergoing many detrimental effects that corrupt the signal

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

CFR

DFT/IDFT

Coefficients

Coefficients

CIR

Tap index

Subcarrier index

n Figure 1. Time and frequency domain channels representation for OFDM based systems.
and often place limitations on the performance of the system.
Transmitted signals are typically reflected and scattered, arriving at receivers along multiple paths. Also, due to the mobility
of transmitters, receivers, or scattering objects, the channel
response can change rapidly over time. Most important of all,
the radio channel is highly random and the statistical characteristics of the channel are environment dependent. Multipath
propagation, mobility, and local scattering cause the signal to
be spread in frequency, time, and angle. These spreads, which
are related to the selectivity of the channel, have significant
implications on the received signal. Channel estimation performance is directly related to these statistics. Different techniques are proposed to exploit these statistics for better
channel estimates. There has been some studies that cover
these estimation techniques, however these are limited to the
comparison of few of the channel estimation techniques
[2124]. This paper focuses on an extensive overview of the
channel estimation techniques commonly applied to OFDM
based multi-carrier wireless systems.

OFDM CHANNEL ESTIMATION


Channel estimation has a long and rich history in single carrier communication systems. In these systems, the CIR is typically modeled as an unknown time-varying FIR filter, whose
coefficients need to be estimated [14]. Many of the channel
estimation approaches of single carrier systems can be applied
to multi-carrier systems. However, the unique properties of
multi-carrier transmission bring about additional perspectives
that allow the development of new approaches for channel
estimation of multi-carrier systems.
In OFDM based systems, the data is modulated onto the
orthogonal frequency carriers. For coherent detection of the
transmitted data, these sub-channel frequency responses must
be estimated and removed from the frequency samples. Like
in single carrier systems, the time domain channel can be
modelled as a FIR filter, where the delays and coefficients can
be estimated from time domain received samples, which are
then transformed to frequency domain for obtaining the CFR.
Alternatively, radio channel can also be estimated in frequency domain using the known (or detected) data on frequency
domain sub-channels. Instead of estimating FIR coefficients,
one tap CFR can be estimated (Fig. 1).
Channel estimation techniques for OFDM based systems
can be grouped into two main categories: blind and non-blind.
The blind channel estimation methods exploit the statistical
behavior of the received signals and require a large amount of
data [25]. Hence, they suffer severe performance degradation
in fast fading channels [26]. On the other hand, in the nonblind channel estimation methods, information of previous
channel estimates or some portion of the transmitted signal
are available to the receiver to be used for the channel esti-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

mation. In this article, only the non-blind channel estimation


techniques will be investigated.
The non-blind channel estimation can be studied under
two main groups: data aided and DDCE. In data aided channel estimation, a complete OFDM symbol or a portion of a
symbol, which is known by the receiver, is transmitted so that
the receiver can easily estimate the radio channel by demodulating the received samples. Often, frequency domain pilots
are employed similar to those in new generation WLAN standards (802.11a and HYPERLAN2) [27]. The estimation accuracy can be improved by increasing the pilot density. However,
this introduces overhead and reduces the spectral efficiency.
In the limiting case, when pilot tones are assigned to all subcarriers of a particular OFDM symbol, an OFDM training
symbol can be obtained (block type pilot arrangement). This
type of pilot arrangement is usually considered for slow channel variation and for burst type data transmission schemes,
where the channel is assumed to be constant over the burst.
The training symbols are then inserted at the beginning of the
bursts to estimate the CFR (e.g. WLAN and WiMAX systems) [28, 29]. When channel varies between consecutive
OFDM symbols, either the training symbols should be inserted regularly within OFDM data symbols with respect to the
time variation of the channel (Doppler spread), or the channel should be tracked in a decision directed mode to enhance
the receiver performance.
In the DDCE methods, to decode the current OFDM symbol the channel estimates for a previous OFDM symbol are
used. The channel corresponding to the current symbol is
then estimated by using the newly estimated symbol information. Since an outdated channel is used in the decoding process, these estimates are less reliable as the channel can vary
drastically from symbol to symbol [31, 32]. Hence, additional
information is usually incorporated in DDCE such as periodically sent training symbols. Channel coding, interleaving, and
iterative type approaches are also commonly applied to boost
the performance of DDCE~techniques.
There are numerous approaches to estimate the channels
for OFDM subcarriers. The direct estimation of the channel
for subcarriers treats each subcarrier as if the channels are
independent. However, in practice, the CFR is often oversampled via the subcarriers, and hence the estimated frequency
domain channel coefficients are correlated. On the other
hand, the noise in these subcarriers can be independent. By
utilizing the correlation of CFR in subcarriers, the noise can
be reduced significantly. Therefore, the channel estimation
accuracy can be improved [28]. Several approaches have been
proposed to exploit this correlation. These approaches and
their relationship with each other will be discussed in the subsequent sections to provide a unified understanding. Similarly,
the subcarrier correlation in time and spatial domain can be
exploited since the noise can be considered to be independent
in time and spatial domain as well.

19

Cyclic
prefix
X1

S
/
P

IFFT
Kpoint

Remove
Cyclic prefix
P
/
S

Ant #1

Y1
IFFT
Kpoint

P
/
S
K

Wireless
channel

Data
bits

Ant #1

S
/
P

Deinterleaving,
demodulation,
decoding

Coding,
modulation,
interleaving
Remove
Cyclic prefix

Cyclic
prefix

XNtx

S
/
P

IFFT
Kpoint

P
/
S

Ant #Ntx

Ant #Nrx

S
/
P

IFFT
Kpoint

Output
bits

K
P
/
S

CSI
YNrx

n Figure 2. MIMO-OFDM transceiver model.


Although it is a common approach to assume the channel
to be constant over an OFDM symbol duration [9, 27], for
fast fading channels the same assumption leads to ICI [33],
which degrades the channel estimation performance. Hence,
the methods employed in data-aided and decision directed
channel estimation need to be modified so that the variation
of the channel over the OFDM symbol is taken into account
for better estimates. External interfering sources also affect
the performance of channel estimation. The effect of interfering sources can be mitigated by exploiting their statistical
properties. Although most systems treat ICI and external
interference as part of noise, better channel estimation performance can be obtained by more accurate modeling [34].
There are basically three basic blocks affecting the performance of the non-blind channel estimation techniques. These
are the pilot patterns, the estimation method, and the signal
detection part. Each method covered in this article either
tackles one of the above basic block or several at a time. The
specific choice depends on the wireless system specifications
and the channel condition. The aspects of each method are
presented such that a suitable method can easily be selected
for a given wireless system and channel conditions. It can be
observed that each method can be approximated to the other
methods by using the same set of variables. For example, in
this paper it is shown that each estimation method is indeed a
subset of LMMSE technique.
In the literature, initial channel estimation methods have
been mostly developed for SISO-OFDM systems, that is, single antenna systems. With the emergence of MIMO-OFDM,
these methods need some modifications as the received signal
in MIMO-OFDM is the superposition of all the transmitted
signals of a given user. In many cases, the methods of SISOOFDM are easily adopted for MIMO-OFDM but novel methods exploiting space-time codes or other MIMO specific
elements are also introduced.
In the rest of the article, starting from a generic system
model, the channel estimation techniques will be presented
starting from the less complicated techniques. More emphasis
will be given on data aided channel estimation as it provides
some unique approaches for OFDM systems. Discussions on
ICI, external interferers, and MIMO systems as well as related
issues will also be given. Finally, some concluding remarks and
potential research areas will be given at the end of the article.

NOTATION
Matrices and the vectors are denoted with boldface letters,
where the upper/lower letters will be used for frequency/time

20

domain variables; (.) H denotes conjugate-transpose; E{.}


denotes expected value; diag(x) stands for diagonal matrix
with the column vector x on its diagonal; 0 ab denotes a
matrix of a b zero entries; IN denotes N N identity
with
matrix; and j= 1.

SYSTEM MODEL
A generic block diagram of a basic baseband-equivalent
MIMO-OFDM system is given in Fig. 2. A MIMO-OFDM
system with Ntx transmit and Nrx receive antennas is assumed.
The information bits can be coded and interleaved. The coded
bits are then mapped into data symbols depending on the
modulation type. Another stage of interleaving and coding
can be performed for the modulated symbols. Although the
symbols are in time domain, the data up to this point is considered to be in the frequency domain. The data is then demultiplexed for different transmitter antennas. The serial data
symbols are then converted to parallel blocks, and an IFFT is
applied to these parallel blocks to obtain the time domain
OFDM symbols. For the transmit antenna, tx, time domain
samples of an OFDM symbol can be obtained from frequency
domain symbols as
(1) [ n, m ] =
xtx

IFFT { Xtx [ n, k ]}

(1)

K 1

= Xtx [ n, k ]e j 2 mk / K 0 k, m K 1

(2)

k =0

where X tx [n, k] is the data at the kth subcarrier of the nth


OFDM symbol, K is the number of subcarriers, and m is the
time domain sampling index. After the addition of CP, which
is larger than the expected maximum excess delay of the channel, and D/A conversion, the signals from different transmit
antennas are sent through the radio channel.
The channel between each transmitter/receiver link is modelled as a multi-tap channel with the same statistics [3]. The
typical channel at time t is expressed as,
L 1

h(t , ) = l (t ) ( l ),

(3)

l =0

where L is the number of taps, l is the lth complex path gain,


and l is the corresponding path delay. The path gains are
WSS complex Gaussian processes. The individual paths can be

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

variables in matrix notation, for rxth receive antenna and nth


OFDM symbol, we get

correlated, and the channel can be sparse.


At time t, the CFR of the CIR is given by,

H (t , f ) =

N tx

h(t , )e j 2 f d .

(4)

Yrx =

H [ n, k ] H (nT f , k f ) = h[ n, l ]F ,
kl
K

(5)

l =0

where h[n, l] = h(nT f , kt s ), F K = e j2/K , T f is the symbol


length including CP, f is the subcarrier spacing, and ts = 1/Df
is the sample interval. In matrix notations, for the nth OFDM
symbol, Eq. 5 can be rewritten as
H = Fh

(6)

where H is the column vector containing the channel at each


subcarrier, F is the unitary FFT matrix, and h is the column
vector containing the CIR taps.
At the receiver, the signal from different transmit antennas are received along with noise and interference. After perfect synchronization, down sampling, and the removal of the
CP, the simplified received baseband model of the samples
for a given receive antenna, rx, can be formulated as
N tx L 1
m
yrx [ n, m ] = xtx [ n, m l ]hrxtx [ n, l ]
tx =1 l = 0

(7)

+ irx [ n, m ] + wrx [ n, m ],
where rx =1, , Nrx, the time domain effective CIR,
l],
over an OFDM symbol is given as time-variant linear filter
depending on the time selectivity of the channel. Please note
that n represents OFDM symbol number, while m denotes the
m
sampling index in time domain so that h rxtx[n, l] is the CIR at
the sampling time index m for the symbol n. When the CIR is
m
constant over an OFDM symbol duration, then h rxtx[n, l] will
be the same for all m values, and hence the superscript m can
be dropped. Moreover, i rx [n, m] is the term representing
external interference, w rx [n, m] is the AWGN sample with
2
zero mean and variance of w. After taking FFT of the time
domain samples of Eq. 7, the received samples in frequency
domain can be expressed as,

yrx [n, m]e

2 km
K

(8)

m=0

1
m
= xtx [ n, m l ]hrxtx [ n, l ]
K m =0 tx =1 l =0
K 1 N tx L 1

+ irx [ n, m ] + wrx [ n, m ]] e
N tx

1
tx =1 K

2 km
K

[ n, k ] e j 2 ( m l ) k / K
tx

m = 0 l = 0 k = 0

h [ n, l ] e

m
rxtx

K 1 L 1 K 1

2 km
K

Xtx + Irx + Wrx .

(12)

tx =1

Here, Yrx is column vector storing the received signal at each


subcarrier, F is the unitary FFT matrix with entries

ej2mk/KK with m and k being the row and column index and
= F rxtx F H , which can be considered as the equivalent
channel between each received and all the transmitted subcarriers. Moreover Xtx denotes the column vector for transmitted
symbols from txth transmit antenna, Irx is the column vector
for interferers, Wrx is the column vector for noise, and rxtx is
the matrix containing the channel taps at each m index. The
entries of are given by

rxtx

h 0 [ n, 0 ]
0
rxtx
1
1
hrxtx [ n,1]
hrxtx [ n, 0 ]

=
L 1
L 1
hrxtx [ n, L 1] hrxtx [ n, L 2 ]

0
0

0
hrxtx [ n, 2 ]

m
h rxtx[n,

K 1

N tx

L 1

1
K

(11)

tx =1

With proper CP and timing, the CFR can be written as [3],

Yrx [ n, k ] =

Frxtx F H Xtx + Irx + Wrx ,

(9)

(10)

+ I rx [ n, k ] + Wrx [ n, k ]

where Irx[n, k] and Wrx[n, k] are the corresponding frequency


domain components calculated from irx[n, m]s and wrx[n, m]s,
respectively. After arranging the terms, and representing the

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

1
hrxtx [ n, 2 ]

(13)

0
hrxtx [ n,1]

1
hrxtx [ n, 3]

K
K
hrxtx1[ n, L 1] hrxtx1[ n, 0 ]

When the channel is assumed to be constant over one


OFDM symbol and the CP is larger than the CIR length, then
h m [n, l] is the same for all ms, making rxtx a circulant
rxtx
matrix [35]. The multiplication of FrxtxF H then results in a
diagonal matrix, and hence no cross-terms between subcarriers exist, that is, no ICI occurs. In this case, h is equivalent to
the first column of . However, when the channel varies over
an OFDM symbol, then ICI occurs, and for the equalization
the channel at each time sample of OFDM symbol is needed,
that is, at each m samples. For the frequency domain estimation, this requirement translates into the knowledge of the
channel coefficients at each carrier frequency as well as their
cross-terms. The number of unknowns in time domain estimation are KL, whereas the number of unknowns in frequency
domain (the entries of ) are K2. In either case, the number
of unknowns will be higher than the number of equations, and
hence a system of under-determined equations will result in.
Simplifications are needed so that the unknowns in the system
of equations are reduced. Different approaches will be
described in detail in the subsequent sections.
Once the received signals for each transmit antennas are
detected with the help of channel estimation, the reverse
operation at the receiver is performed, that is, they are
demodulated, de-interleaved, and decoded. As it will be seen
later, the information at different stages of decoding process
can be exploited to enhance the performance of channel estimation methods.

21

Frequency

Frequency

Time
Training symbols

Time
Data symbols

Data subcarriers

Pilot subcarriers

(a)

(b)

n Figure 3. Typical training symbols and pilot subcarriers arrangement.


OFDM CHANNEL ESTIMATION TECHNIQUES
There are several basic techniques to estimate the radio channel in OFDM systems. The estimation techniques can be performed using time or frequency domain samples. These
estimators differ in terms of their complexity, performance,
practicality in applications to a given standard, and the a priori information they use. The a priori information can be subcarriers correlation in frequency [36], time [3], and spatial
domains [37]. Moreover, the transmitted signals being constant modulus [38], CIR length [39], and using a known alphabet for the modulation can also be a priori information [40,
41]. The more the a priori information is exploited, in general
the better the estimates are [42].
For frequency domain channel estimates, MSE is usually
considered as the performance measure of channel estimates,
and it is defined by
^

MSE = E{|H[n, k] H [n, k]|2},

(14)

where H [n, k] is the estimate of equivalent channel at kth subcarrier of nth OFDM symbol. Although MSE is used extensively, sometimes, other measures like BER performance are
also used [43, 44]. BER performance is mainly used when the
performance of OFDM system with the channel estimation
error is to be evaluated [45, 46].
Before introducing the estimation techniques, it is worthwhile to look at the data aided channel estimation in general
and the pilot allocation mechanisms.

DATA AIDED CHANNEL ESTIMATION


In this subsection, we will review commonly used methods in
the data aided channel estimation. Initially, we will consider
the methods developed for SISO-OFDM. ICI is assumed not
to exist and the CIR is assumed to be constant for at least one
OFDM symbol. Hence, is a diagonal matrix, where each
diagonal element represents the channel between the corresponding received and the transmitted subcarriers. In this
case, for the nth OFDM symbol, the channel given in Eq. 5 at
each subcarrier can be related to as

22

H[n, k] = [k, k].

(15)

Furthermore, the external interference is folded into the noise


with noise statistics being unchanged. With the above assumption, the expression in (12) can be expressed as
Y = diag(X) H + W,

(16)

or
Y[n, k] = H[n, k] X[n, k] + W[n, k].

(17)

Here H and W are the column vectors representing the channel and the noise at each subcarrier for the nth OFDM symbol, respectively.
In data aided channel estimation, known information to
the receiver is inserted in OFDM symbols so that the current
channel can be estimated. Two techniques are commonly
used: sending known information over one or more OFDM
symbols with no data being sent, or sending known information together with the data. The previous arrangement is usually called channel estimation with training symbols while the
latter is called pilots aided channel estimation (Fig. 3).
Channel estimation employing training symbols periodically sends training symbols so that the channel estimates are
updated [29]. In some cases training symbols can be sent
once, and the channel estimation can then be followed by
decision directed type channel estimation. The details of the
decision directed will be given later in the article.
In the pilots aided channel estimation, the pilots are multiplexed with the data. For time domain estimation, the CIR is
estimated first. The estimate of the CIR are then passed
through a FFT operation to get the channel at each subcarrier
for the equalization in frequency domain. For frequency
domain estimation, the channel at each pilot is estimated, and
then these estimates are interpolated via different methods.
Pilots Allocation for Data Aided Channel Estimation
For the pilot aided channel estimation, the pilot spacing needs
to be determined carefully. The spacing of pilot tones in frequency domain depends on the coherence frequency (channel
frequency variation) of the radio channel, which is related to
the delay spread. According to the Nyquist sampling theorem,

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

0.9

Correlation coefficient (abs)

0.8

0.7

0.6

0.5

0.4
0.3
0

10

20

30
40
Subcarrier index

50

60

70

n Figure 4. Periodic behavior of subcarriers cross-correlation for


K = 64.

the number of subcarrier spacing between the pilots in frequency domain, Dp, must be small enough so that the variations of the channel in frequency can be all captured, that is,
1
Dp
(18)
max df
where max is the maximum excess delay of channel. When the
above is not satisfied, then the channel available at the pilot
tones does not sample the actual channel accurately. In this
case, an irreducible error floor in the estimation technique
exists since this causes aliasing of the CIR taps in the time
domain [47].
When the channel is varying across OFDM symbols, in
order to be able to track the variation of channel in time
domain, the pilot tones need to be inserted at some ratio that
is a function of coherence time (time variation of channel),
which is related to Doppler spread. The maximum spacing of
pilot tones across time is given by
1
Dt
(19)
2 fdmaxTf
where f d max is the maximum Doppler spread and T f is the
OFDM symbol duration. For comb-type pilot arrangements,
the pilot tones are often inserted for every OFDM symbols.
When the spacing between the pilot tones does not satisfy the
Nyquist criteria, then the pilots can still be exploited in a combined pilot-plus DDCE [48].
The pilots can be sent continuously for each OFDM symbol. Since the channel might be varying both in time and frequency domains, for the reconstruction of the channel, this
2-D function needs to be sampled at least a Nyquist rate.
Hence, the rate of insertion of pilots in frequency domain and
from one OFDM symbol to another cannot be set arbitrarily.
The spacing of pilots should be according to Eq. 18 and Eq.
19. In general, within an OFDM symbol the number of pilots
in frequency domain should be greater than the CIR length
(maximum excess delay), which is related to the channel delay
spread. Over the time, the Doppler spread is the main criteria
for the pilot placement.
Many studies are performed in order to get the optimum
pilot locations in time-frequency grid given a minimum number of pilots that sample the channel in 2-D at least Nyquist
rate. This optimality is in general based on the MSE of the LS
estimates [6, 49]. It should be noted that an optimum pilot
allocation is a trade-off between wasted energy in unnecessary
pilot symbols, the fading process not being sampled sufficient-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

ly, the channel estimation accuracy, and the spectral efficiency of the system [50]. Hence, an optimum pilot allocation for
a given channel might not be optimal for another channel as
the fading process will be different.
In addition to minimizing MSE of the channel estimates,
pilots also need to simplify the channel estimation algorithms
so that the system resources are not wasted. For example, it
is noted that the use of constant modulus pilots simplify the
channel estimation algorithms as matrix operations become
less complex [38, 51].
Some other important elements for pilot arrangements
are the allocation of power to the pilots with respect to the
data symbols, the modulation for the pilot tones etc. In many
cases, the power for pilot tones and data symbols are equally
distributed. The channel estimation accuracy can be improved
by transmitting more power at the pilot tones compared to
the data symbols [52]. For a given total power, this reduces
the SNR over the data transmission. As for the pilot power
at different subcarriers, studies show that based on the MSE
of the LS estimates pilots should be equipowered [6, 53].
Moreover, due to the lack of the pilot subcarriers at the
edge of OFDM symbols, the estimation via the extrapolation
for the edge subcarriers results in a higher error [54, 55]. Simulations also reveal that the channel estimation error at the
edge subcarriers are higher than those at the mid-bands due
to this extrapolation [5658]. One quick solution would be to
increase the number of pilot subcarriers at the edge subcarriers [58], however this would decrease the spectral efficiency of
the system [57]. Due to the periodic behavior of the Fourier
Transform, the subcarriers at the beginning and the end of
the OFDM symbol are correlated, and this can be used to
improve the channel estimates at the edge subcarriers (Fig. 4).
Simulations exploiting this property are reported to enhance
the estimation accuracy of the edge subcarriers [57].
Another issue related to pilot arrangement is the pattern
of the pilots, that is, how to insert the pilots to efficiently
track the channel variation both in time and frequency
domains. The selection of a pilot pattern may affect the channel estimation performance, and hence the BER performance
of the system.
Equation 18 states that the pilot spacing in frequency
domain needs to satisfy the Nyquist criteria. More insight into
Eq. 18 reveals that the number of required pilots in frequency
domain can be taken as the CIR length. At a first glance, this
does not pose any restriction on the pilot spacing that a sufficient number of pilots can be inserted in adjacent subcarriers.
However, when the MSE of the time domain LS estimation,
which is covered in the next subsection, is analyzed, it is
observed that the minimum MSE is obtained when the pilots
are equispaced with maximum distance [6, 31, 39]. This is due
to the reason that when the pilots are inserted in adjacent
subcarriers, then the FFT matrix used in the time domain LS
estimation approaches to an ill conditioned matrix, making
the system performance vulnerable to the noise effect [39].
Hence, from the MSE of LS estimation, the pilots in frequency domain need to be equipowered, equispaced, and their
number should not be less than the CIR length. Since the use
of pilots is a trade-off between extra overhead and the accuracy of the estimation, adaptive allocation of pilots based on the
channel length estimation can offer a better trade-off [52, 56,
59]. As will be seen later in the article, with MIMO and ICI
additional requirements will be observed on the pilot subcarriers spacing and properties.
When it comes to the pilot allocation for subsequent
OFDM symbols, either the set of subcarriers chosen in a previous OFDM symbol or a different set of pilots can be used
(Fig. 3). The use of the same subcarriers as the pilots is a

23

widely used pilot arrangement. In such a pilot arrangement,


first the channel between subcarriers is estimated via interpolation in frequency domain. This is followed by interpolation
over OFDM symbols in time domain. In some cases, interpolation can be first performed in time domain, followed by the
frequency domain interpolation. The details of different interpolation techniques will be given later in this section.
The analysis of MSE of time domain LS estimation over
several OFDM symbol indicates that for a lower MSE, the
pilots should be cyclically shifted for the next OFDM symbol
[6, 60]. This pilot allocation is similar to those used in DTV
applications, and is similar to the pilot scheme given in Fig.
13. In this pilot allocation scheme, the interpolation is first
performed in frequency domain, followed by the interpolation
in time domain. Similar to the pilot scheme used in DTV, a
hexagonal type pilot scheme is also proposed [6163]. In both
schemes, different subcarriers are utilized for each OFDM
symbol, and hence the possibility of sticking into terribly fading subcarriers is eliminated, that is, diversity is exploited.
In addition to the above pilot schemes, different types of
pilot schemes are tested through simulations [56]. The pilots
having more density than the others, those utilizing different
subcarriers over time and at the edge subcarriers are expected
to perform better for channels varying both in time and frequency domains.
The previous pilot allocation schemes were solely based on
the MSE analysis of the channel estimation. In some cases,
other system parameters can also be considered for the pilots
to be used. For example, due to the IFFT block at the transmitter side, PAPR of OFDM systems can be very high. It is
observed that different training symbols (not scattered pilots)
results in different PAPR [64]. Moreover, different scattered
pilot allocation schemes can result in different PAPR when
multiplexed with data. Since the data is random, the optimum
allocation for minimum PAPR will be different for each transmission. However, pre-defined pilot allocation schemes can be
tested for the best PAPR [65]. With such a scheme however,
the information about the pilot scheme needs to be conveyed
to the receiver side, and this reduces the spectral efficiency of
the system.
It is clear from the discussion about the pilot allocation
that a better system performance can be obtained when the
system is adaptive [52, 59, 60, 66]. In this case, the information about the channel statistics becomes very critical. The
pilot allocation in the frequency domain requires the delay
spread estimation, whereas the one in over OFDM symbols
(over time evolution) requires Doppler spread estimation. If
these estimates are available, then a pilot scheme using just
the right amount of pilots can yield an acceptable performance. If this information is not available, then the pilot
scheme can be designed based on the worst channel condition, that is, the maximum expected delay and Doppler
spreads. In addition to unknown channel statistics, randomly
generated pilots can be utilized for the reduction of interference from adjacent cells. However, it is shown via simulations
that such pilots cause severe degradation in the channel estimation MSE [67].
So far the pilots in the frequency domain are discussed. In
some cases, the estimation can be performed using the data in
time domain, that is, data before the FFT block at the
receivers. Training symbols for this case can be set to all 1s in
frequency domain that result in an impulse in the time
domain. When this impulse is passed through the channel,
then CIR can be obtained. By careful arrangement of 1s in
frequency domain, the multiple replicas of the CIR can be
obtained, and these can be improved through noise averaging.
In a similar way, PN sequences superimposed with the data

24

can be utilized for the channel estimation. In such a case, correlators at the receiver can be used for the expected samples
of the OFDM symbols [6870]. However, it is shown that
superimposing training with data is not optimal for channel
estimation [71].
Having reviewed the pilot schemes employed in OFDM
systems, it is time to look at the channel estimation techniques. Starting from the methods using the least a priori
information, in this article we will review channel estimation
methods such as LS estimation, ML, transform domain techniques, and LMMSE. Simple interpolation techniques will be
covered along with LS estimation technique.

LS ESTIMATION
Before going into the details of the estimation techniques, it is
necessary to give the LS estimation technique as it is needed
by many estimation techniques as an initial estimation. Starting from system model of SISO-OFDM given in Eq. 17 as[72]
Y[n, k] = X[n, k]H[n, k] + W[n, k],

(20)

the LS estimation of H[n, k] is


Y [n,k ]
W[ n,k ]

H LS [ n,k ] =
= H [n,k ]+
.
X[ n,k ]
X[ n, k]

(21)

In matrix notations,
^

H LS = diag(X)1Y + diag(X)1W.

(22)
^

Note that this simple LS estimate for H LS does not exploit


the correlation of channel across frequency carriers and across
OFDM symbols.
The MSE of LS estimation of Eq. 22 is given by [73]
MSE LS =

K
E H SNR

(23)

where EH = E{H[n, k]}.


LS method, in general, is utilized to get initial channel estimates at the pilot subcarriers [72], which are then further
improved via different methods.
It is also common to introduce CIR to Eq. 16 to exploit
CIR length for a better performance [21, 74]. In this case, Eq.
16 can be modified as [74]
Y = diag(X)Fh + W
where H = Fh. The LS estimation of Eq. 24 is then
^

H = QLSFH diag(X)HY

(25)

where
QLS = (FH diag(X)Hdiag(X)F)1.

(26)

The above LS estimation will be referred as time domain LS.


When no assumptions on the number of the CIR taps or
length are made, then the time domain LS reduces to that of
frequency domain, and it does not offer any advantages. However, with the assumption that there are only L number of
channel taps, which then reduces the dimension of the matrices F and hence Q, an improved performance due to the
noise reduction can be obtained [75, 76]. The resultant LS
estimation has higher computational complexity than the frequency domain LS but the performance increase is the plus
side of the approach. The increase in the performance can be
considered as the exploitation of subcarrier correlation. A
comparison study showed that when the frequency domain LS
also exploits the correlation of the subcarriers, then its performance can be that of time domain LS (21). Further compari-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

son studies showed that based on the SNR information, either


method can be used [74]. For example if the SNR is low then
the time domain LS can be less accurate as additional filtering
in time domain is based on less accurate CIR length. In this
case, the probability of not accounting for all the taps and discarding some of them are high. However, for other SNR
regions, the time domain LS gives better results as it utilizes a
more accurate CIR length. The use of time domain LS
becomes inevitable when OFDM is combined with MIMO
systems [77]. This will be explored more when channel estimation techniques for MIMO systems are presented.
Similar to the time domain LS, the ML estimate of the
CIR taps for the same system model given in Eq. 24 can be
derived. With the assumption of L channel taps and Np number of pilot subcarriers, the ML estimate of the channel coefficients is shown to be [58, 78],
^

H
H
HML = (Fp Fp)1Fp diag(X)HY

(27)

where F p is N p L truncated unitary Fourier matrix. In the


above formulation, for the sake of simplicity, it is assumed
that pilots symbols are from PSK constellation and hence
diag(X)Hdiag(X) = IK, and they do not appear in the parenthesis for the inverse operation. It can be observed that when
the number of pilots is greater than the channel length and
the noise is AWGN, the time domain LS estimate in Eq. 25 is
equivalent to the ML estimate given in Eq. 27 [58, 79]. Furthermore, it should be noted that the ML estimate given in
(27) makes the assumption about the CIR length, which
improves the performance of the estimation accuracy [80].
Unlike LMMSE channel estimation, both LS and ML are
based on the assumption that the CIR is a deterministic quantity with unknown parameters. This implies that LS and ML
techniques do not utilize the long term channel statistics and
hence are expected to perform worse than the LMMSE channel estimation method [58]. However, the computational complexity is the main trade-off factor between the two groups of
the channel estimation techniques.
Before introducing the other channel estimation techniques, it is worthwhile to review the methods used for the
training sequences as well as the pilot subcarriers. The corresponding implications on the channel estimation techniques
will also be covered briefly.

this is one of the algorithms employed for IEEE 802.11a/b/g


and fixed WiMAX systems. However, these approaches introduce an error floor for non-constant channels, that is, outdoor
channels. The highest performance degradation occurs at the
symbols farthest from the training symbols. For video transmission systems, the critical information can be sent over the
symbols closer to the training symbols, while non-critical
information can be sent over those farther from the training
symbols [29, 30]. It is observed that such an arrangement can
improve the performance without increasing the number of
training blocks. However, for systems requiring equal priority
packets like data networks, such an approach cannot be taken.
In this case satisfactory results can be obtained by increasing
the rate at which the training symbols are sent at the expense
of system efficiency.
For the fast varying channels, interpolation methods can be
utilized in time domain. Interpolating the channel linearly
between the training symbols is one simple solution [59, 72,
82]. The disadvantage with such an approach is the latency
introduced in the system [83]. Indeed, if the system can tolerate more latency, then the channel estimation for non-training
OFDM symbols can be improved by higher order polynomials
[66, 84, 85].

CHANNEL ESTIMATION TECHNIQUES IN PILOT MODE


In the pilot mode, only few subcarriers are used for the initial
estimation process. Depending on the stage where the estimation is performed, estimation techniques will be considered
under time and frequency domains techniques.
In frequency domain estimation techniques, as a first step,
CFR for the known pilot subcarriers is estimated via (22).
These LS estimates are then interpolated/extrapolated to get
the channel at the non-pilot subcarriers. The process of the
interpolation/extrapolation can be denoted as
^

H = QHLS

(28)

where Q is the interpolation/extrapolation matrix. The goal of


the estimation technique is to obtain Q with lower computational complexity but at the same time is to achieve higher
accuracy for a given system. In this subsection, the calculation
of matrix Q for simple interpolation techniques will be discussed.

CHANNEL ESTIMATION TECHNIQUES IN TRAINING MODE


As mentioned before, in the training mode, all the subcarriers
of an OFDM symbol are dedicated to the known pilots. In
some systems like WLAN or WiMAX, two of the symbols are
reserved for the training. If the training symbols are employed
over two OFDM symbols, for very slowly varying channels, the
channels at two OFDM symbols for the same subcarriers can
be assumed to be the same. In this case, the estimates can be
averaged for further noise reduction [72]. If the noise variances of the OFDM symbols are different, then Kalman filtering can be used such that noise variances are exploited as
weighting parameters [81].
Once the channel is estimated over the training OFDM
symbols, it can be exploited for the estimation of the channels
of the OFDM symbols sent in between the training symbols.
Depending on the variation of the channel along time, different techniques can be utilized.
A very common method is to assume the channel being
unchanged between OFDM training symbols [23, 2830, 69].
In this method, the channel that is estimated at training symbols is used for the subsequent symbols until a new training
sequence is received. The channel is then updated by using
the new training sequence, and the process continues. In fact,

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Piecewise Linear Interpolation Two of the simplest ways


of interpolation are the use of piecewise constant [86] and linear interpolation [22, 84, 87, 88]. In the piecewise constant
interpolation, the CFR between pilot subcarriers is assumed to
be constant, while in piecewise linear interpolation the channel
for non-pilot subcarriers is estimated from a straight line
between two adjacent pilot subcarriers. Mathematically, for
piecewise constant interpolation, Q is a matrix consisting of
columns made up from shifted versions of the column vector
c = [1,1,,1, 0,, 0 ]T ,
Dp

where D p is the spacing of the pilots. For the the piecewise


linear interpolation, Q consists of coefficients that are a function of the slope of the line connecting two pilot subcarriers
and the distance of the pilots to the subcarrier for which the
channel is to be estimated.
In the first method, acceptable results can be obtained if
the CFR is less frequency selective or the CIR maximum
excess delay is very small. Such a constraint makes the CFR at
the subcarriers very correlated that CFR at a group of subcarriers can be assumed to be the same.

25

In piecewise linear interpolation some variation is allowed


between the pilot subcarriers. Such an approach can result in
a lower MSE since noise averaging is performed. Moreover,
when the channel becomes more frequency selective, the
piecewise linear interpolation results in a better performance
compared to the piecewise constant [8689]. For a better
insight into the performance of the piecewise linear interpolation, its MSE is derived and is expressed in terms of the channel statistics and the pilot spacing as [87]
1
1
2
MSE = (5 + )(1 + ) R f [ 0 ] + (2 + 2 ) w
3
3
D p 1

l =0

1
(1 1){ R f [l ]} + (1 2 ){ R f [ D p ]}
3

(29)

where 1/ = Dp, Rf[l] is the frequency domain correlation of


2
CFR, denotes the real part of a complex number, and w is
the noise variance. When the piecewise linear interpolation is
to be performed between OFDM symbols over time, then the
parameters above need to be replaced with their time domain
equivalence. As can be seen from the expression, lower MSE
results in:
When many pilots are used
When the noise is low
When the channel is very correlated
Higher Order Polynomial Fitting Piecewise linear interpolation requires more pilot subcarriers for an acceptable performance in highly frequency selective channels [52, 86, 89].
However, by using a priori information about the frequency or
time selectivity of the channel, the use of higher order polynomial can result in better performance. Higher order polynomials indeed can approximate the wireless channels accurately,
since the channel itself is smooth in both time and frequency
domains [66]. The degree of this smoothness depends on the
selectivity of the channel. For highly time and frequency selective channels, the higher the polynomial order, the better the
estimation at the expense of higher computational complexity
[23]. However, when the channel is changing very slowly both
in frequency and time, then the use of very high order polynomials can degrade the performance, as the modelling uses
noise as a means to represent the channel [66]. This behavior
also suggests dynamic polynomial fitting based on the channel
statistics [23]. Simulations show that adaptive polynomial fitting performs better than the static polynomial fitting when
the channels become more selective [23]. In a move towards
reducing the computational complexity of such an adaptation,
instead of estimating the true channel statistics, variation of
channel between two adjacent subcarriers can be monitored,
and an idea of how fast the channel is changing can be
obtained [90]. Further computational complexity can be
achieved if the coefficients of Q are made power of 2 to eliminate the multiplication/division via bit shifting. It is observed
that such an approach can yield accurate channel estimates
[90].
In the higher order polynomial approaches, the entries of
the Q are calculated by using more information about the
channel. Higher order polynomial fitting uses more than two
pilot subcarriers for the CFR estimation. While some of the
polynomial fitting methods utilize no channel statistics [52,
91], others assume to have some information about the statistics [66, 85]. The most common higher order interpolation
methods are spline interpolation [22, 60, 89], Gaussian interpolation [22], and polynomial fitting [66, 85, 91, 93, 94]. In the
spline interpolation, basis function of some orders or Beizer
curve are defined over a group of subcarriers [60, 84]. These

26

basis functions are determined such that they are unity at the
pilot locations at which they are defined for, and vanishes at
the other pilot locations. The channel at non-pilot subcarriers
can then be found as
Np

H [ n, k ] =

B p [n, k ]H [n, p],


p =1

(30)

where Np is the number of pilots over a range, Bp[n, k] is the


basis function at subcarrier k, and H[n, p] is the CFR at the
pilot location p. The rows of the interpolation matrix, Q, are
then formed using B p[n, k]s. For more frequency selective
channels the order of the basis functions, B p[n, k]s, can be
increased for a better performance. This corresponds to having more columns in Q, and implies the use of more pilot subcarriers for the estimation of a single subcarrier.
Gaussian interpolation is another interpolation technique,
where the coefficients of Q are obtained from a Gaussian
function [95]. The Gaussian function resembles the sinc function, the ultimate function for ideal low pass filtering. The
Gaussian function can be considered as an approximation to
the sinc function. The width of the Gaussian function or
equivalently the coefficients used in the interpolation are
dependent on the frequency selectivity of the channel. Hence,
as with many approaches, the knowledge of the channel statistics can improve the performance of the Gaussian interpolation.
Similar to the Gaussian interpolation, radial basis functions
utilizing Gaussian function are also used for the interpolation
purpose [96]. The coefficients of the radial basis functions are
determined through some non-linear training mechanism similar to those used in neural networks. Overall, the goal is to
find the coefficients of the interpolation using the Gaussian
function as a basis, and the training process indeed reflects
the information about the channel statistics to the coefficients
to be used in the interpolation. Hence, the approach of the
radial basis function interpolation can be considered as an
adaptive low-pass filtering. The improved performance due to
this adaptation comes at the cost of training process using
pilot subcarriers.
2-D regression models for the pilot subcarriers scattered in
frequency and time domains are also studied [85, 94]. In these
models, a 2-D polynomial whose coefficients are obtained
using the channel correlation and the initial LS estimates at
the pilot subcarriers is developed. Although higher order polynomials can be used, second order approximation is found to
yield close to ideal BER performance for certain channels
[85].
All of the above interpolators can be seen as a simple lowpass filter. This is due to the fact that CIR has a finite length
that is in general much smaller than the number of subcarriers. The above interpolation methods are not ideal low-pass
filters, and hence they introduce an error floor due to either
the suppression of some of the channel taps or the inclusion
of noise whose effect becomes effective at high SNR regions.
A low-pass filtering can eliminate the noise in non-tap locations, which in turn means the elimination of most of the
noise in the estimated subcarriers. For example, it is shown
that the use of raised cosine filter as a low-pass filter provides
accurate channel estimates for WLAN systems [28]. The
sharper the low-pass filtering the better the estimates are.
Since the Fourier Transform of a rectangular function (or a
window) is the sinc function, the sinc interpolator with the
known CIR length provides ideal low-pass filtering. However,
sinc interpolator is not realizable in practical implementations.
Moreover, it is computationally heavy as it requires more
CFR samples.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Tap coefficients

-1
0

10

20

30

40
Tap index

50

60

70

n Figure 5. Typical CIR samples with noise.


The low pass interpolation utilizes the extra information
about the CIR length. Further improvement can be achieved
with the information of other channel statistics [37, 97, 98].
However, if the channels are less frequency and time selective, then there is no need for very complicated estimation
techniques, and the use of simple interpolators will do the job.
Since computation of the information of channel statistics will
need extra processing, systems unable to get the statistics can
assume a worst case scenario for the typical application. Such
systems can use an interpolator based on the assumed statistics throughout the application.

TRANSFORM DOMAIN TECHNIQUES


It was mentioned that in general the CIR length is much
smaller than the number of pilot subcarriers, that is, L <
N p . When an orthogonal transformation is applied to the
CFR at the pilot subcarriers, the transform domain contains
L number of significant values, that is, values relatively having more energy or magnitude than the noise. Since the
noise is assumed to be AWGN in frequency domain, it is
AWGN in transform domain as well. If the significant values of the transform domain signal are retained, and the
non-significant ones are treated as zero, then the noise term
will be eliminated significantly especially when L < N p
<
[99]. For this operation, some sort of threshold is needed to
differentiate between the significant values of the signal and
noise terms. The CFR can then be obtained by applying the
inverse of the orthogonal transformation, since such an
operation will also achieve interpolation for non-pilot subcarriers [99, 100].
Once the CFR is obtained via a transform domain technique, the channel at subsequent OFDM symbols (over time)
can be obtained via different methods. The filtering process of
transform domain is usually followed by linear interpolation in
time domain [101]. Wiener filtering is also found to be effective in noise reduction in time domain [102]. With Wiener filtering being optimum in the sense of minimizing MSE, it has
been applied in both domains as well. This will be covered in
the subsequent sections.
The transform domain techniques exploit the information
about the number of significant values in the transform
domain and their location. Moreover, more number of pilot
subcarriers are used for the interpolation process. Hence, they
perform better than the simple interpolation techniques in
general [103]. Different transform domain techniques are

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

studied for the channel estimation of OFDM based systems.


Fourier [31, 84, 101, 104, 105], Hadamard [106, 107], Discrete
Cosine [108, 109], Karhunen-Loeve Transform KLT [110],
and 2-D Fourier Transformation [111] are few to name.
In Fourier Transform, two techniques are investigated. In
the very common approach, first the IFFT of the CFR is
taken. The resultant transform domain is the time domain,
where typically the channel taps are concentrated into a subregion. Figure 5 shows a typical CIR with significant taps and
noise. By zeroing the terms out of this subregion that corresponds to noise, only the significant taps are retained. The
noise reduced signal is then transformed back into the frequency domain via FFT operation.
Mathematically, the process of the Fourier Transform
technique for the channel estimation of OFDM based systems
can be represented as [76],
H

H FT = K / N p FD FT FP H LS

(31)

where
K / Np
is for the normalization, Fp is the Fourier matrix with the rows
corresponding to the subcarrier index of the pilot tones, and
DFT is given by
DL
D FT =
0 ( K L L )

0( L K L )
.
0( K L K L )

(32)

The expression for DFT given above is applied in many transform domain based approaches using Fourier Transform,
which in general neglects the correlation between CIR taps.
By incorporating the CIR tap correlation and appropriately
choosing the coefficients of DL, a full matrix that can result in
lower channel estimation MSE can be obtained [76].
When the channel statistics are not available, then DL = IL
can be used. In this case, Eq. 31 can be expressed as [76],
H

H FT = K / N p FL FLp H LS

(33)

where FL and FLp contain only the first L columns of F and


Fp, respectively.
By examining Eq. 31, it can be anticipated that the pilots
need not be chosen such that they sample CFR uniformly.
However, the condition number of Fourier matrices increases
for closely spaced pilots. Hence, when the pilots are not closely spaced, Fourier matrices with higher condition numbers can
result in higher MSE in the presence of noise [31, 39]. Theoretically, in the presence of AWGN, the MSE of the Fourier
Transform approach is shown to be minimum when the pilots
are equispaced with maximum distance [39].
In terms of implementation complexity, Fourier Transform
approach is computationally very efficient, thanks to FFT
algorithms. The computational complexity of FFT algorithms
are further reduced via the radix 4 operation with restrictions
on the pilot spacing to be a power of 2 [52, 112].
Similar to the above Fourier Transform approach, instead
of taking IFFT as an initial transformation, first FFT is performed over CFR samples [59, 29, 102, 113]. The equivalent
transform domain taps are not concentrated into a subregion
as they are in the CIR taps. The equivalent taps of this alternative Fourier Transform approach are shown in Fig. 6 for a 5
tap channel.
As seen in Fig. 6, only the center region of the transform
domain needs to be zeroed, meaning the identification of one
more region. Mathematically, the above procedure can be

27

Non-sample spaced
Sample spaced
5

4
Tap coefficients

Equivalent tap coefficients

0
0

10

20

30
40
Tap index

50

60

70

10

15
20
Tap index

25

30

35

n Figure 7. Equivalent CIR taps from CFR when the real taps

n Figure 6. Equivalent taps of fourier transform of CFR.

are not sample spaced.

written as

H FT = K / N p F H D FT 2 Fp H LS ,

(34)

where the entries of the DFT2 are the cyclically shifted entries
of DFT along its diagonal elements.
Studies are performed to compare the performance of
both of the Fourier Transform approaches. When the number
of significant taps are estimated correctly, both methods perform the same. However, since in the second Fourier Transform approach three regions need to be identified, in the
presence of noise it is more prone to errors [59, 114].
In both of the Fourier Transform approaches, the noise
within the significant taps is not eliminated. If the noise in the
transform domain is completely independent, then there is
not much to be done for removing the noise within the taps.
However, if the noise in the transform domain is correlated,
then by using the information about the noise in the non-tap
locations, the noise in tap locations can be reduced. Studies
show that this additional processing provides further improvement in the MSE of the channel estimation [115].
The Fourier Transform approach assumes the knowledge
of the tap location. It is assumed that the taps are equally
spaced with sampling time of the OFDM symbol. When the
IFFT of the CFR is taken, the equivalent CIR is given by,
L 1

h( , t ) =

l (t )
l =0

sin( BW ( l ))
,
( l )

(35)

where BW is the bandwidth of the OFDM symbol. If the CIR


taps are sample spaced, then CFR is band-limited or CIR is
time-limited, and hence the performance of the Fourier Transform approach is very close to the ideal low-pass filtering.
However, if the CIR taps are not sample spaced, that is, l,
then as can be seen from Eq. 35, the energy of the non-sample
spaced tap is leaked to the other taps. With this, CFR is not
bandlimited anymore, and there is aliasing. When the noiseonly taps are eliminated, this leaked energy is also removed,
and hence some degradation occur as the total energy of the
CIR taps is not preserved. Figure 7 shows the equivalent CIR
taps from CFR when the real taps are non sample spaced.
The aliasing due to non-sample spacing can be considered
as the high frequency terms in the domain from which the
transformation is performed. For this reason, windowing in
frequency domain is applied in order to mitigate for the aliasing effect [116, 117]. In contrast to the conventional Fourier

28

Transform approaches that use rectangular windowing, other


window types like Hamming and Hanning can be used in the
frequency domain. The effect of this windowing is removed
when the CFR is transformed back from the transform
domain. It is observed that this additional processing improves
the performance of channel estimation when the taps are not
sample spaced. However, since the aliasing effect is not completely removed, still at high SNR regions an error floor
occurs.
Another transform domain technique is studied via DCT
[108, 109]. The main reason behind the use of this approach is
to get better channel estimates when CIR taps are not sample
spaced. In this approach, the effect of high frequency terms is
mitigated by exploiting the property of DCT algorithm. DCT
equivalently takes the symmetry of the CFR samples and
introduce a 2K sequence with a smoother transition between
the elements of 2K sequence. Here, the CFR samples are first
passed through a DCT operation. Similar to the Fourier
Transform approach, the energy corresponding to equivalent
CIR taps is concentrated in subregions, whose length are
much smaller than the dimension of the orthogonal transformation. The regions corresponding to non-tap locations are
zeroed out for the noise elimination. The resultant transform
domain signal is then passed through an IDCT operation to
get CFR. Simulation results show that better results than the
conventional Fourier Transform domain approach can be
obtained [109]. However, as in the case of windowing operation for CFR samples, since the aliasing is not completely
removed, an error floor still exists at high SNR region. The
mathematical description of the DCT approach is similar to
the second Fourier Transform approach with Fourier matrices
being replaced by the DCT matrices. Although the same
approach of taking the symmetry of samples can also be
applied to the Fourier Transform approach, faster DCT algorithms than the FFT algorithms can offer a better computational complexity [109]. However, since OFDM based systems
already has FFT algorithms on IC, additional IC will be
required for the DCT implementation, increasing the cost of
the modules.
Very similar to the DCT approach, Hadamard and KLT
transforms are also studied [106, 107, 110]. In these methods,
the same steps as in Fourier Transform approach are taken so
that the signal and the noise subspaces are separated. It
should be noted that KLT has not been applied to OFDM
based systems but is tested for single carrier MIMO systems.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Analysis showed that SNR improvement can be achieved, suggesting its application to MIMO-OFDM systems.
Practical systems such as WLAN and WiMAX introduce
guard bands in OFDM symbols via the elimination of the use
of the subcarriers at the edge of the OFDM symbols. The
transform domain techniques suffer from these unused subcarriers or suppressed subcarriers as this corresponds to rectangular windowing in frequency domain that results in sinc
convolution in transform domain or equivalent time domain.
Hence, the taps are leaked to one another due to sinc interpolation, and the taps orthogonality is lost. Since the transform
domain techniques assume a certain channel length, L, K L
taps are zeroed out during time windowing. When this windowing in time domain is applied for reducing the noise plus
interference, it will cause Gibbs phenomenon when the signal
in the transform domain is transferred back to the frequency
domain. In other words, the channel frequency response will
have ripples around the edge carriers. The reason for this is
the truncation of the sinc function in time domain. Although
no studies have been reported to overcome this issue, simple
extrapolation via the use of correlation properties of the subcarriers (Fig. 4) can be employed before any transform
domain approach.
Although different transform domain techniques are studied, it should be noted that since the noise is assumed to be
AWGN in the original domain, it will have its AWGN characteristic in the transform domain. The equivalent channel taps
will be concentrated only in a small region. Therefore, as long
as the equivalent taps are correctly identified in all the transform domain methods, the same performance will be achieved.
Since the Fourier Transform approach utilizes fast algorithms
in the OFDM transceiver structure, the use of Fourier Transform offers a better trade-off among the transform domain
channel estimation techniques.
When the CFR samples are also available over several
OFDM symbols at the pilot subcarriers, 2-D Fourier Transform technique can be used [111, 118]. Here, the CFR samples at the pilot subcarriers are passed through a 2-D Fourier
Transform. The transform domain signal is expected to have a
diamond shape concentrated in a 2-D subregion. By zeroing
out the signal values outside this subregion, noise reduction
can be achieved. The resultant signal can then be transformed
into the original domain via an inverse 2-D Fourier Transform. The performance of 2-D transform domain approaches
depends on the appropriate transform domain filtering, which
is related to the channel statistics.
The transform domain techniques have been successfully
applied to the cases where impairments to the system are
modelled as Gaussian noise. For example, an OFDM system
with significant PAPR can be improved via clipping, where
the large peaks are replaced with a pre-defined envelope, A.
The system model can then be modified as [119]
Y = (cr diag(X) + )H + W

A/x,

(36)

x is the average energy


where the clipping ratio, cr =
of the input signal, and is the distortion caused by the clipping. With cr known, the effect of distortion can be reduced
via transform domain techniques since the distortion
approaches complex Gaussian distribution with zero mean.
Simulations performed for this scenario showed that with the
use of Fourier Transform technique, accurate channel estimation can be obtained [119].
The information about the CIR length is important in
achieving higher performance in transform domain approaches. A CIR length taken to be smaller than the actual CIR
length will eliminate the significant taps, while a channel
taken to be longer will result in less noise suppression. How-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

ever, the first case is more critical than the second, as and
hence in practice the CIR length is usually taken to be the
length of the CP [36, 120, 121].
For more accurate results, algorithms are also developed
to estimate the number of significant channel taps. This is
needed especially in high data rate communication where the
channel length can be long but the number of channel taps
can be smaller, that is, sparse channel. The corresponding
channel taps and their location can be searched based on a
cost function assuming channel tap locations and the corresponding coefficients [122125]. In many cases, since the
problem is similar to finding the tones of a signal, ESPRIT
and MDL algorithms are employed to get the number and the
location of the channel taps [11, 124126]. However, these
approaches can yield degraded performance when the number
of taps increases. Hence, other approaches based on iterative
ML are proposed both to reduce the computational complexity and to get accurate estimation [123]. Moreover, taps
searching based on the energy of the taps is also studied [102,
113, 127, 128]. Iterative algorithms like Newton-Lapsons
method can also be utilized to get the channel taps [129, 130].
The use of PN codes superimposed with the data to reveal
the channel taps is also widely applied [70]. The output of
these correlators is related to CIR, which can be averaged further when the correlators result in multiple CIR copies. However, the performance of this approach suffer from the
influence of the transmitted data that suggests an increase in
the power of the PN sequence or time-consuming iterative
methods [31]. Hence, transmitting data and the pilots over
different subcarriers, and then using the correlators before the
FFT block at the receiver is proposed [131]. In this method, a
time domain signal obtained via the IFFT of the pilot signal is
utilized to be correlated with the received signal. Ideally, the
output of the correlator are the delayed impulses whose
amplitude and delay are related to CIR.
Having realized that the performance of the transform
domain techniques are heavily dependent on the CIR tap
locations, an inaccurate assumption or calculation of CIR tap
locations can degrade any of the transform domain techniques
drastically. Hence, a transform domain method which inherently uses the information of the channel taps is expected to
provide better results. For this purpose, unitary transform
based on the eigenmatrices of the auto-correlation of CFR of
different channel PDPs is shown to give better results than
the transform domain techniques presented above [107].
When the exact channel PDP is not available, then a channel
PDP can be assumed and the transformation can be done
accordingly. However, if the exact PDP is known then eigendecomposition of auto-correlation matrix of CFR can provide
the optimum transform. In the following section, this optimum transform, a special form of LMMSE, is presented in
detail.

LMMSE CHANNEL ESTIMATION


LMMSE is widely used in the OFDM channel estimation
since it is optimum in minimizing the MSE of the channel
estimates in the presence of AWGN. LMMSE uses additional
information like the operating SNR and the other channel
statistics. LMMSE is a smoother/interpolater/extrapolater, and
hence is very attractive for the channel estimation of OFDM
based systems with pilot subcarriers. However, the computational complexity of LMMSE is very high due to extra information incorporated in the estimation technique [22, 36, 58].
For a given linear system model in the form of
y = Ax + w,

(37)

29

LMMSE of the variable x is given by,


^
x

1
= Ryx Ryy y

(38)

where Ryx is the cross-covariance between variables y and x.


When the expression in Eq. 38 is applied to the OFDM channel estimation given in Eq. 16 with equal pilot spacing, Dp,
^

HLMMSE = RHHp
^
2
(RHpHp + w (diag(X) diag(X)H)1)1 HLS

(39)

can be obtained. Here, Hp is the CFR at the pilot subcarriers,


RHHp represents the cross-correlation between all the subcarriers and the pilot subcarriers, and R H p H p represents the
auto-correlation between the pilot subcarriers. As can be
seen in Eq. 39, LMMSE uses additional information in its
estimation process such as the correlation between subcarriers and SNR.
The LMMSE estimation of H in Eq. 39 is computationally
very heavy. For example, the dependency on the transmitted
symbols due to the matrix inversion required at each estimate
needs many operations. Moreover, large sized, full matrix
multiplication required for a single estimate increases the
computational complexity of LMMSE as well. The non-trivial
matrix inversion required in the LMMSE estimation is another factor increasing the computational complexity of LMMSE.
Therefore, although LMMSE is optimal, without reducing its
computational complexity, it is hard to realize its application
in practical systems.
The complexity of LMMSE can be significantly reduced if
the LMMSE expression is made independent of the transmitted symbols. Although the expression inside the inversion
operation also contains the term R H pH p, which is the autocovariance of the CFR at the pilot tones, R H p H p does not
change for a large number of OFDM symbols since it is a
function of channel PDP. Therefore, for a given large number
of OFDM symbols, the term RHpHp can be assumed to be constant, leaving (diag(X)diag(X)H)1 as the constantly changing
parameter from symbol to symbol. By assuming the transmitted symbols use the same signal constellation, the expression
(diag(X)diag(X)H)1 in Eq. 39 can be replaced by the expected
value of (diag(X)diag(X)H)1 [36]. That is,
E {(diag(X )diag(X ) H )1} = I N p
| 2}

SNR

(40)

|2},

E{1/|Xk
with Xks being the constelwhere = E{|Xk
lation points. Then, Eq. 39 becomes,

H LMMSE = R HH p R H p H p +
I N p H LS .

SNR

(41)

It is recommended in some studies that such an approximation should not be assumed for the whole OFDM subcarriers,
as the noise level can be different for various portions of the
symbol [132]. In this case, a windowing approach can be
applied to suppress the noise so that over the whole symbol
the noise level is almost constant. The advantages of this
approach comes at the expense of SNR estimation for each
subcarrier and the additional filtering. Since it is observed via
simulations that the approximation given in Eq. 40 has negligible performance degradation for the OFDM channel estimation, the SNR estimation for each subcarrier or subcarrier
block is usually omitted [36, 120].
Although the expression in Eq. 41 is simpler, it still needs
to be updated with the changing operating SNR. Moreover,
the expression in Eq. 41 needs to be recalculated whenever

30

the channel PDP changes. Either due to the change in SNR


or PDP, the channel estimation via Eq. 41 is computationally
complex since it requires multiplications in the order of O(K3)
for the channel estimate of a single subcarrier.
The complexity of LMMSE is even higher for 2-D channel
estimation of OFDM systems since the number of the total
subcarriers is higher [24, 133, 134]. Hence, the complexity of
LMMSE can be computationally prohibitive for practical 2-D
channel estimation. For this reason, computationally efficient
methods are proposed so that the benefits of LMMSE are
realized both in 1-D and 2-D channel estimation.
Subspace methods are investigated for the computational
complexity and noise subspace reduction for the LMMSE
channel estimation. With subspace methods, the number of
multiplications required for the channel estimate of a single
subcarrier is reduced by exploiting SVD [24, 36, 120, 135,
136]. Subspace methods applied to the LMMSE channel estimation reveal the degree of independency of the subcarriers
auto and cross-correlation matrices. Since the subcarrier correlation is a function of the channel delay spread, it ultimately
reveals long-term significant CIR taps or channel PDP.
Without subspace methods, the complexity of the channel
estimation using LMMSE can be reduced significantly by
assuming a pre-defined channel length [75]. However, for
sparse channels this would mean unnecessary computation
when the significant number of channel taps is smaller than
the channel length [122, 126]. With the CIR length being
much smaller than the number of the subcarriers, SVD of the
auto and cross correlation matrices of CFR result in only as
many significant singular values as the significant number of
CIR taps. As the noise is assumed AWGN in frequency
domain, the SVD decomposition results in equivalent singular
values for the noise terms. Hence, it can be anticipated that
the noise in frequency domain is equally distributed in the
subspace domain with equal energy in all dimension of the
subspace. If the subspace due to the noise is eliminated, then
noise reduction is achieved [126]. Moreover, due to the formulation of LMMSE, less number of multiplications will be
required after the SVD operation. This will be seen more
clearly with the following derivations.
Starting from an all-pilot case that all the subcarriers are
pilot tones, the LMMSE channel estimation can be re-written
as,

H LMMSE = R H p H p R H p H p +
I K H LS .

SNR

(42)

If SVD is to be performed over Hermitian RHpHp,


RHpHp = UUH

(43)

can be written. Here U is a unitary matrix and is a diagonal


matrix bearing the singular values 0, 1, , K1 in descending orders. Then in Eq. 42 can be re-written as,
^

HLMMSE = UUHHLS,

(44)

where is a diagonal matrix with entries

i
i =
,
i +

SNR

i = 0,1,, K 1.

(45)

With the above formulation, the corresponding MSE is


expressed as [36],

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

MSE LMMSE

1
=
K
+

1
K

K 1

i (1 i ) + SNR i2

i =0

^r
Hlmmse

K 1

where the entries of the diagonal matrix

i
i =r

i = 0,1,, r 1
,

(47)

i = r,, K 1

where r represents the number of the significant singular values. The above is nothing but the result obtained from the
low-rank approximation of the R H p H p . With the low-rank
approximation, the number of required multiplications
reduces from the order O(K3) to O(rK2).
Although the low-rank approximation via a SVD of the
auto-covariance matrices reduces the number of multiplication for the channel estimation, obtaining the SVD of the
auto-covariance matrices by itself is computationally very
complex and is in the order of O(K3) [137]. Therefore, it will
be no use to exploit the low-rank approximation if the SVD is
to be performed for every estimation process. Although, the
auto-covariance matrix, R HpHp, is a function of the channel
PDP that can be assumed to be constant for a good number
of OFDM symbols [138], when the channel PDP changes the
re-computation of SVD of RHpHp can be non-practical especially when the number of the subcarriers is large.
For this reason different approaches are proposed in order
to eliminate the need for SVD operation. The so called robust
channel estimation methods are developed for this purpose
[36, 139, 140]. In these methods, a channel PDP is assumed
for the system under the consideration, and the auto-covariance matrix and its SVD are then pre-calculated for the
assumed channel PDP. The most common assumed channel
PDPs are uniform and exponential, with uniform PDP being
used more extensively [13, 126]. Simulation results show that
robust LMMSE channel estimation results in acceptable performance degradation for certain systems when compared to
the LMMSE with perfect channel knowledge [13, 111]. The
degree of degradation increases as the true channel deviates
significantly from the assumed channel. By pre-calculating the
SVD of the auto-covariance matrices for more possible channel PDPs, this degradation can be mitigated. In this case, by
looking at the delay spread of the channel, the closest channel
PDP for which the SVD is pre-calculated can be used. With
additional computational complexity needed for the delay
spread estimation, this approach is found to improve the MSE
performance of the robust LMMSE channel estimator by a
factor of 2 dB [139].
The low-rank approximation for LMMSE channel estimation has also been investigated for the pilot symbol aided
channel estimation [24, 36, 136]. It is shown that for the pilot
symbols similar simple expressions to those of all pilot case
are also possible [120]. For the pilot case,

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

(48)
r

are given by,

(46)

In the decomposition of Eq. 44, the number of multiplication is still in the order of O(N3). The number of multiplication can be reduced if only significant singular values of or
are considered.
It should be noted that the number of significant singular
values is related to the number of long-term significant taps.
The relationship between these two will be given shortly.
Since the number of significant taps is much lower than the
number of the subcarriers, there will be only a few significant
singular values in or . Therefore, the entries of matrix
can be approximated as
i

i = i +
SNR

= Prr(Qr)HHLS,

K
i
Dp

i = K +
D p i SNR

i = 0,1,, r 1

.
(49)

i = r,, N p 1

Here, the superscript r represents only the first r columns of


matrices, P = FV and Q = FpV. Here, V is the unitary matrix
from the SVD of Rhh, the auto-correlation matrix of CIR. The
singular values of Rhh are denoted by 0 1 , , K1. The
MSE of the above estimator is shown to be [120],

1 r 1
MSE (r ) = K i 1 i

K i =0
Dp

1
K

K 1

2
+ i2

SNR

(50)

K i .
i =r

The low-rank approximation for the pilot case requires


r(Np + K) multiplications for the channel estimation of all the
subcarriers. Although this is computationally very efficient
compared to the multiplications in the order of O(K3), still the
need for the SVD of the auto and cross-covariance matrices
of CFR makes the real-time LMMSE estimation almost
impossible. Therefore, the methods enabling real-time
LMMSE estimation are needed.
A close examination of low-rank LMMSE channel estimation shows that for the real-time LMMSE unitary matrices Pr
and Qr are needed as well as the SNR and r significant singular values of RHpHp or Rhh. Since Pr and Qr matrices are related to each other, if one of them is obtained the other can be
calculated easily [120].
It is noted that subspace tracking can enable real time lowrank LMMSE channel estimation of OFDM systems [120].
Subspace tracking has been introduced in adaptive filtering
when the signal under consideration has a subspace with a
dimension less than the number of the data snapshots. This is
a very common case for oversampled systems. In such scenarios, the adaptive filtering requires SVD of large-sized matrices. Subspace tracking avoids SVD of large-sized matrices by
tracking the significant singular values and the corresponding
singular vectors [141, 142]. In OFDM based systems, since the
number of significant taps determines the dimension of the
signal (CFR samples) subspace, and since in most cases this is
much smaller than the number of subcarriers and pilot tones,
oversampling will be observed. Hence, subspace tracking can
be applied to track the few significant singular values and the
corresponding vectors of the matrices RHH or RHpHp, that is,
tracking Pr, r, or Qr with a computational complexity in the
order of O(Kr). However, subspace tracking can only be started after some initial channel estimates that need to be
obtained via some other methods like transform domain techniques.
Subspace tracking has been investigated for OFDM based
systems in [121] and [138] via different approaches. In the first
study, for example, the channel estimates at the pilot tones
are transformed into the time domain by using the singular
vectors of the auto-covariance of the CFR. This approach is
applied in all the pilot case, and produces no mismatches
when the CIR taps are uncorrelated and monotonically
decreasing. In this case, the P matrix is simply the unitary

31

Fourier matrix and is used for the transformation of channel


estimates in frequency domain. The transformation using the
matrix consisting singular vectors of the auto-covariance
matrix of CFR eliminates the noise subspace and hence
improves the performance. The second study, on the other
hand, projects the CIR estimates onto the delay subspace so
that only the significant taps are considered. The method
bears the fact that there will be only small number of time
instances at which the value of the CIR taps are significant.
These time instances are then tracked using the channel estimates at the pilot tones. The tracking of the delay subspace
can be considered as a pre-filtering before
interpolation/extrapolation that improves the performance of
the LS estimate at the pilot subcarriers. Different subspace
tracking methods are analyzed for the tracking of the delay
subspace, and it is observed that delay subspace tracking in
the order of O(Kr) can give accurate results [143]. Since the
subspace of CFR or CIR are related, both studies essentially
perform the same task: noise subspace elimination. Since subspace tracking is more towards oversampled systems, the
number of pilot subcarriers need to be higher than the number of significant taps, otherwise performance degradation will
be observed especially in high SNR region.
Similar to subspace tracking, LMMSE coefficients are
tracked via NLMS and RLS algorithms [144]. In these methods, the CIR taps are updated based on the cost functions
defined for NLMS and RLS. Although NLMS is less complex
and less accurate compared to RLS, care must be taken in
RLS algorithm for oversampled systems, as the performance
can be faulty due to implicit matrix inversion needed during
update operation [145].
2-D LMMSE Since the computational complexity of 2-D
LMMSE is high, several methods are proposed to eliminate
this heavy computation. Among the methods is the use of two
cascading LMMSE filters, thanks to the separability of the
channel correlation in frequency and in time [24, 133, 146,
147]. It is demonstrated that the use of two cascaded LMMSE
performs as accurate as the 2-D LMMSE [24, 133, 148]. This
way the computational complexity is reduced from O((2N)3)
to O(N 3 ), that is, an 87 percent decrease in computational
complexity. While one of the filters uses the frequency domain
correlation between subcarriers, the second filter uses the correlation of the subcarriers over the time. The filtering can
either be done first in frequency or time domain, followed by
the filtering in the other domain. The correlation of the subcarriers over the time depends on the mobility of the stations,
and hence the Doppler the shift. Therefore, 2-D LMMSE
requires channel statistics in time domain as well.
In some cases, the need of channel statistics in time
domain can be eliminated by using simple linear interpolation
[149]. Although computationally more efficient, for very time
selective channels more pilots will be needed over the time for
an acceptable performance.
Similar to 1-D LMMSE, low-rank 2-D LMMSE channel
estimation is also studied [24]. It is observed that the low-rank
approximation yields better results with lesser computational
complexity when compared to non-low rank 2-D LMMSE.
Since the computational complexity is highly dependent on
the number of subcarriers, via the use of lesser subcarriers,
the computational complexity can be reduced significantly.
Such a thinking has brought the use of 2-D LMMSE over
many subregions [24, 147] or piecewise LMMSE concept [87].
In fact, piecewise LMMSE is first considered for 1-D channel
estimation in frequency domain where the noise level is considered to be different at each subregion [150]. Subregioning
implies that the subcarrier correlation is only limited to the

32

those within a given neighborhood, ignoring the correlation


between the subcarriers in different subregions. While for
more frequency and time selective channels no significant gain
is possible with the use of correlation between all subcarriers,
for relatively less frequency and time selective channels, better
performance can be achieved with the use correlation between
subcarriers in different subregions. The 2-D LMMSE filter
order is therefore very critical for obtaining more accurate
results. Doppler and delay spread information can be used for
this purpose. Once these two parameters are incorporated, it
is observed that more accurate channel estimation than the
conventional LMMSE is possible [133, 147, 150]. With
Doppler and delay spread parameters being available, pilot
spacing can also be determined so that the pilots sufficiently
sample the channel response both in time and frequency
domains. When the Doppler and delay spread estimation cannot be done in real-time, as a rule of thumb, the required
pilot numbers can be taken twice of that required by the sampling theorem [134].
Robust channel estimation is also investigated for the 2-D
LMMSE channel estimation [13, 134]. Similar to the channel
statistics assumed in the frequency domain estimation, the
auto-correlation of the subcarrier over time evolution is
assumed to follow the zero order first kind Bessel function.
This function is then based on a maximum assumed Doppler
shift, resulting in the well-known Jakes Doppler spectrum [13,
33]. Simulation results also show that robust 2-D LMMSE
channel estimation results in an acceptable performance
degradation in some mismatch scenarios [13, 111]. The reason
that the robust channel estimation gives good results even in a
mismatch scenario is due to the use of channel statistics that
can be considered as a coarse approximation of the true
statistics. As it will be seen later, the role of the channel
statistics in LMMSE will be visible with accurate estimation of
SNR. When SNR is not available, then the use of channel
statistics alone in robust LMMSE will not bring a desired
improvement given the computational complexity. This topic
will be more elaborated in the next section.

DECISION DIRECTED CHANNEL ESTIMATION


DDCE is one of the earliest methods studied for OFDM,
mainly because of its popularity in legacy systems. In the earlier studies, DDCE was applied mostly in training based systems, where one or more OFDM symbols were used as the
training symbols. The main idea behind DDCE is to use the
channel estimation of a previous OFDM symbol for the data
detection of the current estimation, and thereafter using the
newly detected data for the estimation of the current channel
[31, 151, 152]. Data detection can be based on hard or soft
decision [153]. While for the hard decision a specific constellation point is forced, for the soft decision bitwise detection is
utilized [154]. Since soft decision utilizes more information
about the incoming signal and soft-decoding, for the sufficient
number of iterations, near ideal performance can be observed
[155, 156]. Once the data at the subcarriers is detected, any
methods described in the previous subsections can be used to
estimate the current channel.
Although DDCE is simple, it inherently introduces two
basic problems: the use of outdated channel estimates, and
the assumption of correct data detection. The use of outdated
channel estimates does not pose a serious issue when the
channel is varying very slowly. In this case, the channel can be
assumed quasi-static over a number of OFDM symbols [32,
72]. It was observed that such an approach results in acceptable performance as long as the channel variation is slow over
the time. However, when the channel starts varying faster,

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

100

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

BRSK
QAM-4
QAM-18
QAM-84
10-1

BER

then the outdated channel estimates for the previous OFDM


symbol are no longer valid for the use of the data detection in
the current OFDM symbol [55, 73]. In this case, the data
detection would be incorrect, so are the newly estimated
channel coefficients. Hence, the error in the channel estimation and data detection build up to make the system performance unacceptable [157]. This error propagation becomes
more critical when the the number of incorrect decisions
increases at low SNR regions [158, 159].
As a quick solution to overcome the problem related to the
outdated channel estimates, the training symbols can be sent
more often. The time instances at which the training symbols
should be sent can be based on different criteria. Training
symbols can be sent periodically where the period is predetermined for a given system [159]. Moreover, the change in
channel estimates can be monitored to determine whether a
channel estimation is indeed needed [160]. As the channel
varies fast over the time, the need to send the training symbols more frequently has a high penalty in terms of the overall
system efficiency [49]. In this case, training symbols can be
replaced by the pilot subcarriers [161].
When pilot subcarriers are used in DDCE, the conventional channel estimation methods can be used as long as the
number of pilot subcarriers is sufficient. The pilots can be
sent over each OFDM symbol, or with the knowledge of channel statistics, they can be sent with certain OFDM symbols.
For example, a reliable subset of data subcarriers can be
employed as the pilot subcarriers [162, 163]. The channel estimated at the reliable subcarriers can be used to estimate the
channel at the other subcarriers via interpolation. Although
sounds to be a good solution, the lack of reliable information
about the subcarriers and the high probability of the reliable
subcarriers being non-uniformly distributed over the OFDM
symbol reduce the performance of this approach. By increasing the power level of the reliable subcarriers, the performance degradation can be mitigated to some extent [163].
Another approach can be the use of prediction algorithms
on the channel estimation. The channel estimated in previous
OFDM blocks can be used to predict the channel in the next
block [73, 164, 165]. Prediction algorithms can be applied
either on the channel taps or the channels at the subcarriers
[81]. The previous has the advantage that the number of variables to predict is much smaller but needs IFFT to get the
channel at the subcarriers [166]. The latter requires no transformation but it requires the prediction of more number of
variables, that is, subcarriers. Besides, since the prediction is
performed individually for each subcarrier, the correlation
properties of the subcarriers are not utilized that results in a
worse performance when compared to the prediction for the
time domain channel coefficients [72, 73]. In frequency
domain prediction, the number of subcarriers for which the
prediction to be performed can be reduced so that the channel at the other subcarriers can be estimated by simple interpolation techniques. Depending on the complexity of the
prediction algorithm, such an approach can result in a lower
computational complexity. Linear relationship based prediction approaches can give good results in slowly varying channels, but their performance becomes unacceptable once the
channel varies fast over time. Hence, prediction algorithms
taking into account the channel statistics can perform better
[6, 72]. Indeed, it was demonstrated that the use of ARMA
modelling and Kalman filtering can result in substantial performance improvement in DDCE methods with prediction
[159, 165, 166].
Whether it is through prediction or the use of detected signals, the channel estimation at each subcarrier can be passed
through some filtering to obtain better estimates. Among the

10-2

10-3

10-4
-10

-5

10
SNR [dB]

15

20

25

30

n Figure 8. BER performance of an OFDM system in 5 tap


Rayleigh fading channel with different modulations.

methods, the transform domain methods, LMMSE, or the low


pass filtering are widely applied [132, 167169]. Moreover,
averaging the channel estimates at the subcarriers over a
number of subcarriers is also shown to yield accurate results
for slowly varying channels [170].
In the preceding paragraphs, it was stated that the performance of DDCE can be improved with the information of
channel statistics. In some cases this information may not be
available, or it may be desirable to improve the system performance further. In this case, efficient approaches to the data
detection portion can be introduced for a more reliable channel estimation.
Coding theory is probably one of the most widely fields
applied to the data detection portion of the OFDM systems
[146]. Figure 8 shows the performance of an OFDM system in
Rayleigh fading in a multi-path channel without any coding. It
can be seen that the performance is the same as the single
carrier systems with flat fading. In the case of multipath, single carrier systems can use complex equalizers to improve the
performance of the system significantly. The BER performance of OFDM systems shown in Fig. 8 is therefore unacceptable for practical systems employing high order
modulation [171]. Hence, OFDM systems need to employ
coding for an acceptable performance. In most studies,
OFDM with coding is called as COFDM [154, 155, 171]. With
coding available in OFDM systems, DDCE can exploit this
information to improve the data detection.
The typical coding mechanisms are RS, convolutional, trellis, turbo, and LPDC coding [35, 146, 155, 171175]. The output of the decoder can be further processed for the FEC to
increase the performance of the detection process [32, 168,
176]. Simulation results of many studies showed that the BER
performance of the channel estimation with coding is drastically improved compared to those without coding [146, 155,
171].
Below is a quick summary of the some of the coding techniques applied to OFDM systems. Since the coding techniques performance in OFDM based systems is out of scope
of the current article, the papers cited in this subject and references therein can be referred for more detailed information.
Some Coding Techniques Applied to OFDM Systems
RS (outer coding) and CC (inner coding) are usually applied
back to back in OFDM based systems [9, 177]. RS codes are
linear block codes and are suitable for burst type errors. Their
decoding process is relatively less complex. The CC are binary

33

Coherent
detection
Channel
estimator

Deinterleaver

Decoder
Output

Channel
estimator

Replica
generator

n Figure 9. Joint iterative DDCE.


error correcting codes where input bits are mapped to another
set of bits, by not only using the present bits to be encoded
but also by the previous information bits. Viterbi decoder is
usually employed in the decoding process. The CC are studied
extensively for OFDM based systems, and are shown to yield
improved BER performance with even improved PAPR performance [161, 178, 179].
Trellis codes are introduced by Ungerboeck [180] and are
a special type of CC. Trellis codes provide a better performance/complexity tradeoff than lattices in the bandwidth-limited regime, although the difference is not as dramatic. The
key ideas in the invention of trellis codes were the use of minimum squared Euclidean distance as the design criterion and
the coding on subsets of signal sets using CC principles [181].
Trellis codes are usually combined with STC to improve the
performance of MIMO-OFDM systems [35, 182, 183].
Turbo codes are introduced by Berrou et al., [184], and are
a new class of iterated short CC. They are built from the parallel concatenation of two recursive systematic CC using a
feedback decoder. Turbo codes can achieve a BER performance close to the Shannon limit [185] in an AWGN channel
[178]. Since there are many subcarriers in an OFDM based
system, and that the probability of all subcarriers to fade is
low, Turbo codes applied to OFDM systems can lead to a
high performing wireless system [186].
The LPDC codes are first proposed by Gallager in 1960s
[187], but it has been ignored due to its high computational
complexity when long codewords are to be designed [175].
Due to its parallelized structure, LPDC codes are easy implement via basic elements. LPDC codes have excellent performance in AWGN channels, and under certain conditions they
can perform better than the turbo codes [175]. Hence, LPDC
codes are being re-invented for OFDM based systems, as the
new systems require relatively low complex LPDC codes. For
example, IEEE 802.16e standard employs tile structure in
uplink direction, where relatively less number of subcarriers
are used for the pilots and data [177]. Moreover, the use of
LPDC coding with STC is shown to yield 5-6 dB coding gain
for MIMO-OFDM systems with reasonable computational
complexity [174].
The channel estimation performance has an impact on the
decoding process of the OFDM receivers [188]. Studies in this
area model the channel estimation error as a Gaussian noise,
where the noise power can come from different sources such
as ICI as well as the channel estimation method itself
[188190]. The analyses show that the channel estimation
error becomes effective in system performance when the
channel estimation error is greater that 30 dB [191]. This
additional noise reduces the effective operating SINR of the
decoder block in the receiver, thereby causing performance
degradation. In fact, the analysis via simulations show that
under the channel estimation error, the coding gain of the
decoders reduces significantly [188]. Similarly, the BER analyses also indicate performance degradation due to channel estimation error (less than 3dB for uncoded case) [45, 192].
Hence, as pointed out in the introduction, for a better performing OFDM system, channel estimation block in the
receiver should be very accurate.
Iterative channel estimation algorithms can be exploited to
minimize the channel estimation errors [156]. In these

34

approaches, the channel estimation can be found via any of


the methods described in the preceding sections, and the estimates can be improved by remodulating the detected signals
[154]. It is clear that when the number of iterations is one,
then the approach is the same as the conventional approaches. However, for more iterations better performance is
achieved at the expense of more~computation.

EM ALGORITHM
Among the DDCE channel estimation of COFDM, the EM is
one of the most attractive methods. Mostly because EM algorithm also utilizes the error probabilities that are already
being utilized by the decoders. For example, the maximum a
posteriori (MAP) decoder used in turbo decoding can provide
the probabilities of the transmitted symbols, which is exactly
what the EM algorithm is looking for the channel estimation
[193195]. Hence, computational complexity of EM algorithm
is reduced significantly, making EM a good match for the
COFDM channel~estimation.
The EM algorithm consists of two steps: an expectation
and a maximization step. The motive of the expectation step
is to estimate the corresponding component of the transmit
signal in the received signal, whereas the motive of the maximization step is to estimate the channel given the transmitted
signals, which can either be the pilots or the detected symbols.
With the pilots being available more accurate results can be
obtained [196].
In the above description, the detected signal can come
from a decoder. The decoder itself however, requires channel
estimate (Fig. 9). This chicken-and-the-egg problem can be
solved iteratively with some initial values either assigned to
the channel or the detected signals. While it is common to
employ EM algorithm in the channel estimation part, for the
data detection part different decoders can be employed. For
example, in parallel to the mostly applied turbo coding and
the corresponding decoders, it was shown that EM algorithm
can also be integrated with a QRD-M algorithm using CC
[197]. Iteratively performing the channel estimation and data
detection with sufficient number of iterations are shown to
give very close BER performances to the ideal case [193].
The inherent iterative approach of EM does not necessarily need the channel statistics. With a sufficient number of iterations, EM algorithm converges to the ML algorithm [40],
which was shown in the previous sections to be equivalent to
LS. However, due to EMs iterative nature, the computational
complexity is relatively less [40]. For the EM algorithms to
converge rapidly, the initial assumed/estimated values are critical.
The above iterative scheme described for the EM algorithms can also be generalized as joint and iterative channel
estimation (Fig. 9). Joint and iterative channel estimation
techniques are introduced when the conventional DDCE
channel estimation techniques are unable to estimate the
channel in fast fading, when it is desired to reduce the pilot
overhead, or when the non-linear distortions like power
amplifier non-linearities make the conventional approaches
ineffective [153, 154]. For example, by using relatively less
number of pilots (less than the Nyquist rate), it is shown that
the joint iterative approaches can detect the symbols, and estimate the channel simultaneously with more computational
complexity [48]. For non-linear distortions, iterative methods
can extract the distortion, and the channel estimates can be
improved accordingly [153].
Based on Fig. 9, different combination of the coherent
detection, channel estimator, and the decoder algorithms are
proposed. For example, Kalman filtering is employed to esti-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

102

UNIFICATION OF LMMSE WITH TIME


DOMAIN LS APPROACH
(51)

Then LMMSE can be written as [75],

(52)

With Rhh being invertible,

HLS = F[FHdiag(X)Hdiag(X)F]1 FHdiag(X)HY.


2
w

(53)
(54)

is very small, and


Note that when SNR is very high,
LMMSE reduces to time domain LS approach. Hence, if SNR
information is not available, there is no need to employ
LMMSE with SNR set to a high value since the same performance can be achieved with less computational complexity
offered by time domain LS.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

10-1

10-2

10-3
-20

-15

-10

-5

0
SNR [dB]

10

15

20

n Figure 10. MSE of some of the channel estimation methods in


Rayleigh fading for a 5 tap channel with QAM-4 modulation
and pilot spacing, Dp = 4.

As it was demonstrated before that ML is the same as LS


when the noise is AWGN, LMMSE also reduces to ML when
SNR is very high.

UNIFICATION WITH TRANSFORM DOMAIN TECHNIQUES


In this subsection, the unification of LMMSE with the Fourier
Transform technique will be presented. The unification of
LMMSE with the other transform domain techniques is also
possible if the LMMSE estimation is performed in the corresponding transform domain technique. Since in the preceding
sections frequency domain LMMSE is presented, the unification with the Fourier Transform technique will be presented.
As most of the transform domain techniques are exploited
for the pilot subcarrier aided OFDM channel estimation, the
LMMSE formulation for the pilot subcarriers with equal spacing will be used, and the case of non-equal pilot spacing is
very similar. Starting with the matrix equation,
(55)

the auto-covariance matrix of CFR when all the subcarriers


are used as the pilots can be expressed as
E {HHH} = E{Fh(Fh)H}
RHH = FRhhFH.

Rhh = FnVK(FnV)H,
where

Fn

= F/K. Let

RHH =

HLMMSE = F
2
[(FHdiag(X)Hdiag(X)FRhh)1w + IK]1
Hdiag(X)Hdiag(X)F)1FHdiag(X)HY
(F
can be obtained. The LS estimation was given by [75],

100

(56)
(57)

The SVD of Hermitian Rhh is in the form of, Rhh = VVH.


By normalizing F column-wise (making each column unity
norm),

HLMMSE = FRhh
2
[(FHdiag(X)Hdiag(X)F)1w + Rhh]1
(FHdiag(X)Hdiag(X)F)1
FHdiag(X)HY

101

Hp = Fph,

The formulation of time domain LS was presented previously.


By rewriting the OFDM system model in terms of the CIR,
Y = diag(X)Fh + W

LS
Linear
Transform domain
LMMSE

MSE

mate the time domain channels, while QRD-M is employed


for the data detection part [198]. Due to its inherent iterative
approach, RLS is also combined with DDCE techniques to
estimate the channel, and is shown to provide accurate
results [55, 176, 199, 200]. Since the number of tracking
parameters are less in time domain channel and that RLS
can be erroneous in oversampled systems [142, 145], the RLS
is not recommended over the frequency domain channel
parameters.
The transform domain techniques, LMMSE, and ML
algorithms are also studied for the joint iterative channel
estimation where the channel estimates and data detection
are improved over each iteration [26, 99, 154, 192]. The use
of space-time and space-frequency block decoder is more
common in MIMO-OFDM systems in these iterative
approaches [35, 176, 201203]. For example, Kalman tracking coupled to the Viterbi decoder in the decoding of a
space-time trellis coded MIMO-OFDM system is observed
to give accurate results, where the ambiguity of the start of
the decoding process is eliminated by using a single pilot
subcarrier [35]. With the space-time codes being orthogonal, it is shown that LMS can be successfully applied to
MIMO-OFDM systems [204].
Unification of Channel Estimation Approaches The channel estimation techniques proposed so far basically offer a
trade-off between the complexity and the performance. Figure
10 shows the performance of methods in an exponentially
decaying 5 tap channel. While LMMSE offers the best performance especially at low SNR regions, its complexity is the
highest among the given techniques. This can be attributed to
the fact that LMMSE uses more information during the estimation process. If LMMSE uses as much information as any
other method, then the performance should be very similar if
not the same. Hence, LMMSE can be considered as the generalized channel estimation method. In this section, circumstances for which LMMSE reduces to the other methods will
presented.

FnV

PKPH.

(58)
= P, then
(59)

The above equation is nothing but the SVD of R HH. When


the CIR taps are uncorrelated and are monotonically decreasing, then the unitary matrix P is simply the unitary FFT
matrix. Similarly,
R HH p = P

K
Q H
Dp

(60)

RH pH p = Q

K
Q H ,
Dp

(61)

n
where Q = F p V and pilot spacing is chosen to be D p . With

35

CIR having L taps, is in the form of


L
|
0 L ( K L )

(62)
=
|

,
0

( K L ) L | 0( K L )( K L )
Since it is assumed that Np > L, by truncating to the size
K N p to form N p , and Q to the size of N p N p to form
QNp. Then,
R HH p = P

K
N pQ H p ,
N
Dp

K
= QN p
N pQ H p .
N
Dp

RH pH p

(63)

(64)

Since the first Np columns of Q form a unitary matrix, the


overall equations denote the SVD of R HH p and R H p H p . By
replacing the SVDs of the R HHp and R HpHp into Eq. 64, we
get
^

H
HLMMSE = PQNpHLS,

(65)

where the entries of the diagonal matrix are given by,


K
Dp
i =
, i = 0,1,, N p 1.

K
i +
Dp
SNR

(66)

In case of low-rank approximation, only r significant singular values of Rhh will be considered. Then,
^r
HLMMSE

= PrGr(Qr)HHLS,

(67)

where the entries of the diagonal matrix r are given by Eq.


49.
For high SNR, approaches to a diagonal matrix with

diagonals being Dp. Moreover, when the CIR taps are uncorrelated and there are only L number of significant taps, then
R hh is a diagonal matrix. In this case, V matrix becomes an
identity matrix, making P and Q matrices simply F and F p ,
respectively. Moreover, the SVD of Rhh results in L number
of significant singular values, making r = L. For equal spaced
comb-type pilots
H
H FT = D p FDFp H LS .

(68)

With the conditions described above, low-rank LMMSE


becomes a transform domain technique using Fourier Trans
form. Here, Dp comes from the normalization due to downsampled Fp.
As can be seen from different methods, the use of more
information increases the performance of the channel estimates at the expense of computational complexity. It is noted
in the above sections that when the SNR information is not
available and is set to a high value, then the performance of
LMMSE reduces to the those of not utilizing SNR, with
LMMSE still having high computational complexity. Hence,
the use of other methods in case of no SNR information
offers a better trade-off in terms of the performance and computational complexity.

OFDM CHANNEL ESTIMATION WITH


INTERFERENCE
So far the effect of ICI, ISI, and external interferers were
ignored, and the estimation techniques were performed

36

accordingly. In this section, the effect of interferers will be


treated separately. First the effect of ICI will be considered,
followed by the inclusion of external interferers in the channel
estimation process. A short discussion of ISI is presented
when ICI due to frequency synchronization error is covered.

OFDM CHANNEL ESTIMATION WITH ICI


Again starting from the system of SISO-OFDM, the received
signal in the presence of ICI can be expressed as,
Y = FFHX + W,

(69)

where the external interferers are folded into the AWGN


term. Here, since CIR is not constant over the OFDM symbol, is not a circulant matrix anymore. Hence, the product
of FFH is not a diagonal matrix [205]. In this case, a received
signal at a subcarrier k is affected by the transmitted signals of
all the subcarriers, increasing the number of unknowns by
K*(K 1). This also implies that when the number of subcarriers increases, the ICI increases as well [148]. The ICI power
mainly depends on the product of maximum Doppler frequency and OFDM symbol duration [33]. Hence, while the long
symbol duration of OFDM symbols avoids ISI significantly,
under very fast changing channels, this advantageous parameter turns into a disadvantageous parameter due to ICI
enhancement. The ICI power at the center subcarriers is
expected to be higher than the edge subcarriers since they are
affected more by the ICI of the other subcarriers.
ICI also occurs when there is a frequency offset due transmitter/receiver oscillator mismatch, phase noise, and/or the
non-linear power amplifier effect. The oscillator mismatch or
the phase noise cause the received signal to be sampled at
incorrect positions, and thereby taking the effect of all the
subcarriers [79, 206], that is, orthogonality loss. When left
without compensation, this effect reduces the performance of
channel estimation methods, especially those based on fixed
channel statistics [25].
Either due to the frequency offset or the fast-varying
nature of the CIR taps, ICI needs to be compensated so that
reliable channel estimation is obtained. When higher order
modulation techniques are employed, the effect of ICI is
more severe as the detection of the modulated signal needs to
differentiate many closely spaced constellation points. In this
article, these two effect will be presented independently, and
the details are given in the subsequent sections.
ICI Due to Frequency Offset ICI due to frequency offset
mostly occurs due to the loss of synchronization of the subcarriers or the phase noise of the oscillators. In WLAN and
WiMAX standards, in the preamble, two short duration
OFDM symbols are provided for the synchronization purposes [9, 78, 170]. These short symbols can also be used for the
frequency offset estimation.
Under the synchronization errors (both time and frequency), the correlation properties of the OFDM subcarrier
change in time and frequency domains, the performance of
LMMSE channel estimation algorithms degrade significantly
as these estimation algorithms utilize the correlation properties of the subcarriers. It is shown that synchronization error
can cause up to 5 dB MSE degradation of LMMSE channel
estimation [207]. Hence, the synchronization errors need to be
compensated for OFDM based systems.
The compensation of ICI due the frequency offset is relatively less challenging compared to the compensation of the
ICI due to fast channel variation since the value of the frequency offset parameter is constant over all the subcarriers.
The received signal of a SISO-OFDM in the presence of fre-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Y = Spdiag(X)H + W,

(70)

where Sp is the interference matrix representing ICI due to


the normalized frequency offset p. Here, the entries of the
interference matrix are given by
S p ( m, n ) =

sin ( m n + p ) j ( m n+ p )
e
.

K sin ( m n + p )
K

(71)

Although frequency offset estimation and its compensation


have been studied in numerous articles, we will only consider
those with the channel estimation. In these studies, the channel estimation, frequency offset estimation, and its compensation are performed jointly.
Bearing the fact that the auto-correlation of CFR decreases as the frequency offset increases due to the random behavior of transmitted signals, an iterative binary searching
algorithm based on the diagonal element of the Sep is performed by assuming maximum and minimum values for the
frequency offset [208]. At each iteration step the CFR is estimated based on the assumed frequency offset and so on. Simulation results show that the frequency offset can correctly be
estimated, improving the CFR estimates at the subcarriers.
Moreover, by realizing that the channel estimation error is
minimized when the correct length of the CIR is incorporated
into the frequency offset expression, an iterative method aiming to find the first minimum of the MSE of the channel estimation based on Fourier Transform is developed [25]. With
the use of Blackman window for filtering of the CIR taps, it is
observed that frequency offset can be estimated and compensated with the proposed iterative method.
Frequency offset compensation can be performed before
the FFT block in the receiver side [206]. By comparing the
CFRs with the compensated and uncompensated received signals, the frequency offset of the current symbol can be detected and then can be linearly interpolated to get the frequency
offset of the all the subcarriers. The estimated offset value can
then be used to predict the next frequency offset parameter.
With a more computational complexity algorithm, studies
exploited Kay filters for the frequency offset estimation by
oversampling the pilot subcarriers [79]. Improved performance can be obtained via a prediction algorithm assuming
linear variation over time. Since the frequency offset is
assumed to be the same for all of the subcarrier, averaging
can be introduced to reduce the noise significantly [79].
While frequency synchronization causes ICI, timing synchronization destroys OFDM symbol orthogonality and causes
ISI. Hence, timing synchronization also needs to be considered when performing channel estimation. Timing synchronization error causes both carrier and time dependent phase
rotations [209]. Therefore, the single pilot tracking used for
common phase rotation is not sufficient to compensate for the
timing synchronization error. The compensation for this case
needs at least two OFDM subcarriers to be tracked both in
frequency and time domains so that the slope of variation of
the phase rotation is determined [209]. For efficient system
utilization, time and frequency synchronization and channel
estimation can performed jointly [210, 211].
ICI Due to Fast Fading Channel When the CIR taps vary
over the duration of OFDM symbols, for an accurate channel
estimation, the CIR taps values corresponding at each sampling instance need to be obtained so that the corresponding
CFR is estimated. As mentioned earlier, this implies an
underdetermined set of equations as the number of unknowns

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

CIR
taps

Relative singular values strength

quency offset can be expressed as [208]

ICI
I
Noise
K

Transform domain

n Figure 11. Classification of singular values representing different system parameters.

is more than the number of equations.


In order to reduce the number of equations, the CIR taps
corresponding to each time sample of the OFDM symbol can
be correlated via some basis functions. The knowledge of CIR
taps at couple of sampling points can then be sufficient to
estimate the CIR taps at the other time instances. In this case,
a set of reduced CIR parameters, r , can be related to the
complete CIR, , parameters as [212],
= Q r

(72)

where Q is the interpolation matrix. Different approaches


are studied for the CIR taps evolution over the time. The
most frequently used method is to assume CIR taps varying
linearly [82, 212, 213]. Moreover, interpolation via low-pass
filtering can be utilized for a better estimate in time selective
channels [205]. If the CIR taps follow the Jakes channel
model [214], taps variation then follows first-kind zero-order
Bessel function [212]. In this case, the parameters of the
Bessel function can be found by locating its first zero crossing
via the examination of the subcarrier correlation evolution
over time. At the expense of more computation, the CIR taps
can be modelled as an AR process [215], whose coefficient
can obtained from the channel statistics.
Some studies tried to model the ICI as AWGN and applied
the methods which give good performance under AWGN
[189]. In one of such studies, 1-D and/or 2-D LMMSE is
employed in the channel estimation of OFDM in the presence
of ICI [149, 216]. It is observed that since ICI increases the
noise level, the number of pilot subcarriers required for the
same MSE performance of no ICI case increases by a significant amount. Hence, one way of compensation of ICI is to
increase the number of pilots in the frequency domain.
In fact, when the singular values of the auto-covariance of
CFR under the presence of ICI and noise is analyzed, it can
be observed that the singular values can be grouped under
three categories. The first group will have L number of similar
singular values with L being the number of significant CIR
taps. The second group will have I (I < K) number of similar
singular values, and the rest of the singular values correspond
to the AWGN. In these grouping, the values of singular values
are largest for the first group and smallest for the last group
assuming that the CIR taps are not buried in the noise. Figure
11 shows this scenario. When a low-rank LMMSE is to be
performed on a system with ICI, then most of the ICI will be
cancelled except those overlapping with the singular values

37

0.8

Channel coefficient

0.6

0.4

0.2

-0.2

-2

Subcarriers

n Figure 12. Typical four orthogonal OFDM subcarriers. Note


that sampling at the incorrect points leads significant interference.

that correspond to the CIR taps. Still, the use of low-rank


LMMSE would give a low MSE since it eliminates most of the
subspace corresponding to the ICI. Hence, LMMSE is used
widely in the channel estimation of OFDM with ICI [33, 148,
149, 217].
For the OFDM channel estimation using transform domain
techniques, the information about the channel length is therefore very important in order to reduce the effect of ICI. With
a known CIR length, it is observed that the use of transform
domain techniques reduce the ICI significantly [113, 130].
However, efficient methods obtaining the CIR length need to
be developed. In addition to the methods described in transform domain techniques about the CIR tap identification,
similar methods are proposed when ICI exists. For example,
the CIR length under the presence of ICI is found iteratively
starting from a longer CIR length than expected [25]. Similarly, the channel length is obtained by correlating the first two
short OFDM symbols in the preamble of the WLAN systems
with the local short symbols [80, 167]. In this correlation process, similar to the methods using PN sequences, the channel
taps are revealed, so is their length.
Pilot Spacing in the Presence of ICI In the previous subsection, it was mentioned that in the presence of ICI more
pilot subcarriers are needed for an acceptable performance. If
the number of pilots is to be increased, then it is more appropriate to place the additional pilots next to the existing ones
since the ICI is more severe in the adjacent subcarriers (Fig.
12). Bearing this observation in mind, a small subset of subcarriers are considered to be responsible for the ICI in a subcarrier within the group [218]. Simulation results show that
such pilot arrangement improves the channel estimation performance significantly. Similarly, in early studies two out of
phase adjacent subcarriers were employed as the pilots to mitigate for the effect of ICI [91].
Instead of finding the optimum pilot locations via simulations, for a frequency selective channel, theoretical approaches are carried out for the pilot placement under the presence
of ICI [205, 213]. The approaches showed that in the presence
of ICI the pilots should be all grouped for the optimum elimination of ICI. However, for a frequency selective channel this
would not sample the CFR appropriately, and hence performance degradation would occur for the frequency selective
channels. In order not to have degradation for the frequency

38

selective channels, the clustered pilot scheme is offered to be


the optimum solution. In this scheme, the group of pilots
would be equispaced over the OFDM symbol. This theoretical
finding is nothing but the solution found via simulations in
[91, 218].
The need for the clustering can be explained as follows.
When the CIR taps vary over the OFDM symbol, they need
to be sampled frequent enough in time domain so that the
corresponding CFR can be obtained. For example, if uniform
time domain pilots are employed, then their Fourier Transform would give concentrated pilots in the transformed
domain. In fact, when all the time domain samples of OFDM
are assigned to be pilots, then their Fourier Transform would
give an impulse in the frequency domain. Hence, in order to
compensate both time and frequency domains channel variation, the pilots needs to be grouped and then uniformly distributed in the frequency domain [82].
The analysis performed for the channel estimation of ICI
demonstrates that the performance improvement can be
achieved with the information of channel statistics. This is
either needed for the optimum pilot allocation and the lowrank LMMSE or the transform domain techniques intended
for the low-pass filtering.

OFDM CHANNEL ESTIMATION WITH


EXTERNAL INTERFERENCES
The channel estimation techniques presented in the previous
sections treated the interference from other systems or
sources to be part of the AWGN. As long as the interference
is like AWGN, the methods described in the preceding sections can be utilized safely as they are mostly developed for
the AWGN. However, OFDM systems can suffer from the
impulse noise, which completely destroys the information carried over the subcarriers [219, 220]. In such circumstances,
instead of trying to estimate the channel at the subcarriers via
the sent data, the estimates at the impulse-free pilot subcarriers can be utilized. Based on the channel selectivity, a number
of good estimates at the neighborhood of the destroyed subcarriers can be used both in time and frequency domains, and
using the past and future estimates. The pilot subcarriers
affected by the impulse noise can be detected by looking at
their energy level, as their energy will be much higher in the
presence of impulse noise.
Similarly, the performance of OFDM channel estimation is
investigated in the presence of narrowband interference [34]. By
modeling the narrow-band interference in frequency domain as
a complex Gaussian variable, an overall noise term including the
narrow-band interference with a modified variance can be
obtained. With the use of a generalized ML estimation, that is,
M-estimation method, results better than those not accounting
for the narrowband interference can be obtained when narrowband interference exists in the system [34].
OFDM channel estimation is also performed for the synchronous and asynchronous interference where the noise term
in OFDM system model is defined as [221, 222],
Ni

W [ n, k ] = W [ n, k ] + I q [ n, k ]

(73)

q =1

where N i is the number of interferers, and I q[n, k] is the q th


interference, which can be synchronous or asynchronous interference. It is assumed that for the synchronous case the interferers CPs are aligned with the users CP, while for
asynchronous case the CPs are not aligned with the users CP.
A ML estimation algorithm can be applied but the second
order statistics of the interferers are needed. Efficient non-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Pilots

Cyclically shifted pilots

Space

Ant #1

Ant #2

Ant #3

Ant #4
Frequency
Pilot subcarriers

Data subcarriers

Subcarriers

n Figure 13. Typical pilots for MIMO-OFDM.

n Figure 14. Cyclically shifted pilots for MIMO-OFDM systems..

iterative algorithms are developed for this purpose, and are


tested through the simulations successfully [221, 222].

methods are successfully applied in MIMO-OFDM systems


using pilots as in Fig. 13 [52, 104].
For SISO-OFDM systems, there was an upper bound on
the pilot spacing that the pilot spacing should not be too large
to cause an undersampled CFR function. For MIMO-OFDM
systems using the pilot schemes given in Fig. 13, a lower
bound is dictated so that the interference from the other
antennas is eliminated. The pilot spacing, Dp, is then

MIMO-OFDM
CHANNEL ESTIMATION TECHNIQUES
MIMO-OFDM channel estimation is a challenging task as the
received signal is the superposition of the signals from multiple transmit antennas, (Eq. 12). For the methods to be presented in this section, the ICI and other types of interference
are folded into the AWGN term for the sake of simplicity.
The MIMO-OFDM system model then becomes,
N tx

Yrx =

diag(X )tx Hrxtx + Wrx ,

(74)

tx =1

With the introduction of MIMO, the pilot arrangement has


to be modified so that the existing multiple channels can be
estimated. In the subsequent subsections, first pilot allocation
for MIMO-OFDM, and then the corresponding techniques
will be presented.

PILOT ALLOCATION IN MIMO-OFDM SYSTEMS


When MIMO-OFDM started to draw attention in wireless
communication area, pilot allocation schemes that convert the
channel estimation of MIMO-OFDM into the channel estimation of SISO-OFDM are proposed widely. In these pilot
schemes, at a given pilot scheme, only one of the transmitter
antennas sends its pilot signal at a given subcarrier while the
others remain silent [72, 158]. Such a pilot scheme is shown in
Fig. 13. WiMAX systems also use a similar pilot scheme that
is suitable for two antenna case [9].
With the pilot scheme given in Fig. 13, it can be seen from
Eq. 74 that the MIMO-OFDM received signal at the pilot
subcarriers for a given receive antenna is reduced to
Yrx[n, k] = Hrxtx[n, k]Xtx[n, k] + Wrx[n, k]

(75)

where k Ptx with Ptx holding the pilot subcarrier indices for
the transmit antenna tx. With the pilot subcarrier of each
transmit antenna being disjoint, the received signal for disjoint
pilot subcarrier indices results in as many SISO-OFDM equations as the number of pilot subcarriers. From that point on,
the methods described in the previous sections can be applied
for the channel estimation. For example, Transform domain

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Ntx Dp K/L.

(76)

If WLAN standards are to be employed in a MIMO system, then the pilot allocation in two of the long OFDM symbol in the preamble for the channel estimation can be
designed for a better performance [12]. Since in a typical
WLAN environment, the channel varies very slowly, it can be
assumed that the channel is constant over training OFDM
symbols [103]. In this case, the pilots arranged for the first
OFDM symbol can be cyclically shifted so that the CFR is
sampled uniformly at more points [54, 103, 223, 224]. Such an
arrangement can also mitigate for the edge subcarrier errors
since each antenna can transmit at least one pilot subcarrier
close to the edge subcarriers. Figure 14 shows this scenario
for Ntx = 4. In general, the pilots can be cyclically shifted by
Ntx/NO,where NO is the number of OFDM symbols over which
the channel is assumed to be constant. With the assumption
of the channel being constant over the training phase, noise
reduction can also be achieved via averaging [12]. During the
averaging, better performance can be achieved if the channel
samples are weighted according to their MSE performance or
their noise [54. 91].
Although the comb-type pilots given in Fig. 13 for MIMOOFDM symbols simplify the channel estimation process, they
introduce some drawbacks. Clearly, they reduce the spectral
efficiency since many of the subcarriers are assigned to pilots,
with most of them being the silent pilots. Moreover, the use
of silent pilots increases the PAPR [60], a critical parameter
for the power amplifier block in the transmitters. Hence, in
contrast to the pilot scheme in Fig. 13, the transmission of the
pilots for the same set of subcarriers are proposed (Fig. 15).
When CFR estimation is to be performed over such a pilot
arrangement, Ntx * Np unknowns are at present with only Np
equations being available. Hence, instead of direct CFR estimation, CIR estimation of each MIMO channel is proposed.
A receiver antenna then needs to estimate N tx CIR, each
assumed to have the same L. It should be noted that this is a
valid assumption for MIMO downlink as the transmit and
receive antennas are co-located and hence are expected to

39

K 1

qtx1tx2 [ n, l ] =

*
Xtx1 [n, k ]Xtx2 [n, k ]FK kl .

(83)

ncy

k =0

Freq
ue

Finally, p can be expressed as,

Space

p1

p
p =

p Ntx

(84)

where
Time
Pilot subcarriers

ptx = (ptx[n, 0], ptx[n, 1], , ptx[n, L])T


and

Data subcarriers

K 1

n Figure 15. Overlapping pilots for MIMO-OFDM.

ptx [ n, l ] =

N tx L 1

k =0

tx =1 l = 0

kl

Yrx [n, k ] hrxtx [n, l ]FK Xtx [n, k ]

(77)

*
FK kl Xv [ n, k ] = 0

where Xtx and Xv are the pilot subcarriers for the txth and vth
transmit antennas with v = 1,, N tx . By rearranging the
terms, at the time instant n in matrix notation
^
Qh = p

(78)

or
^

h = Q1p.

(79)
^

Here, the entries of the h, Q, and p are expressed as


h

rx 1

h=

rx Ntx

(80)

where ^rxtx is given as in Eq. 3.


h
Q11

Q=
Q
Ntx 1

Q1Ntx

Q Ntx Ntx

(81)

Qtx1tx2
and

40

qtx1tx2 [ n, L + 1]

qtx1tx2 [ n, 0 ]

(86)

As can be seen, the dimension of the Q matrix is (N tx L


NtxL), meaning more computational complexity with increasing number of transmit antennas and the CIR taps. This also
implies that the knowledge of the CIR length is critical in
attaining a low complex and more accurate channel estimation. It should be also noted that the entries of Q represent
some form of cross-correlation between the pilot subcarriers,
while those of p represent some form of cross-correlation
between the received signal and the pilot subcarriers. Different approaches are proposed in order to eliminate the computational complexity associated with the calculation of the Q1.
One of the approaches is the use of constant modulus signals
for the pilot subcarriers so that the diagonal submatrices of Q
are identity matrices [38, 51]. The matrix inversion is then
eliminated by using a single step iterative approach that utilizes the previous channel estimates. Although this approach
is less complex, performance degradation incur in such an
approach in fast-fading channels. When the cross-correlation
between the pilot subcarriers is zero, that is, orthogonal
sequences, then further simplification arises in the channel
estimate as the need for the previous channel estimate is eliminated [38, 51].
Based on the MSE of LS estimates, the pilots for different
transmit antennas need to be phase orthogonal in addition to
being equispaced and equipowered for a given minimum number of pilot tones and power [6]. For this purpose, Hadamard
[67, 103, 227, 228], Golay [64], and exponential type [49, 60,
63] orthogonal sequences are used in many studies. Hadamard
type codes have good auto and cross correlation characteristics and therefore are popular in communication systems
requiring orthogonality. The Golay codes are found to yield
lower PAPR [64]. The exponential type pilots do not only
introduce orthogonal codes, but also simplify the channel estimation process [6, 38, 51]. The exponential type pilots are
found to be optimum pilots and are given by
ttx [ n, k ] = t1[ n, k ]e

where
qtx tx [ n, 0 ]
1 2

=
q
tx1tx2 [ n, L 1]

*
Yrx [n, k ]Xtx [n, k ]FK kl .

k =0

have the same channel PDP [3, 225]. With this, the number of
unknowns that a receiver antenna has to estimate is Ntx * L.
In CIR estimation, the pilots in the frequency domain now
occupy the same subcarriers, and hence there is interference.
Figure 15 shows the overlapped pilots.
By assuming that the CFR for a given transmit antenna
can be expressed as in Eq. 5, minimization of MSE in Eq. 74
with respect to CIR coefficients for a given receive antenna rx
can be expressed as [77, 226],
K 1

(85)

(82)

2 kL
K

(87)

where ttx represent the pilots for the txth antenna.


A close look at the exponential type pilots reveals some
interesting behavior of the pilots when CIR is to be estimated.
It is known that a phase shift in frequency domain corresponds to a time shift in the time domain. Hence, the phase
of the pilots can be modified in such a way that when their
IFFT is taken, their equivalent CIR representation is delayed
in time domain. A careful design of the pilots can put the CIR
parameters in distinct positions in time domain so that the

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

CIR corresponding to each transmit antenna can be separated


easily. This property is initially proposed in [6] and [51], and
was later investigated by Auer in different studies [229231].
It is concluded that such pilot schemes indeed provide accurate channel estimates when the channel is sample spaced. It
can be seen in these approaches that for the separation of all
the CIR taps belonging to different transmit antennas, each
CIR tap needs to correspond to a distinct time position, which
suggests thatN tx L K. The above idea can be extended to
SISO-OFDM systems such that exponential type pilots result
in multiple replicas of the CIR channel in the time domain
signal. These replicas can be averaged in time domain to get
better estimates [128, 232].
Shifting the phase of the pilots works very well in the sample spaced channels, however, significant performance degradation can occur when the channel is not sampled spaced. In
this case, the paths interfere with each other, and the methods
that can separate different taps will be needed. Windowing
operation and IPIC-DLL methods studied for single antenna
systems can be applied to compensate for the aliasing occurring due to non-sample spaced taps [117]. Moreover, Wiener
filtering can be applied in time domain estimates for the separation of the CIR taps [233].
The CIR channels estimated via Eq. 79 can be further
improved if it is passed through an optimum filter. An optimum filter coefficient however requires the information about
channel PDP. Since in MIMO systems, the existence of multiple channels introduces multiple replicas of the same PDP, a
quick and more accurate estimation of PDP can be obtained
for the use in optimum filtering [225].
The space-time and space-frequency codes are also utilized
in the channel estimation of the MIMO-OFDM systems.
Before going into the details of these pilots scheme, it is
worthwhile to visit the Alamouti type coding that pioneers the
space coding [234]. Starting from two antenna case, the Alamouti schemes transmits two different signals at the same time
instances. In the next time instance, a modified version of
these signals are transmitted from the other antenna. This way
transmitter diversity is achieved both in time and space. These
transmitted symbols are called Alamouti codes that are more
generally termed as STBC. For two transmit antennas, these
codes are given by,
s
S = 1
s2

*
s2

*
s1

(88)

where * represent complex conjugate. For the transmitted


symbols to be estimated, the channel need to stay constant by
as many OFDM symbols as the number of transmitter antennas. Then, the channel at the subcarriers can be obtained via
a sufficient set of linear equations, and can be further
improved via enhanced techniques such as Wiener filtering
[135, 136]. When the channel is not constant by as many as
the number of transmitter antennas, then this scheme suffers
significant performance degradation. The Alamouti scheme is
mostly investigated for two antenna schemes [10, 11, 135]
although it can be generalized for more antennas.
The use of Alamouti codes is mostly applied to the OFDM
subcarriers in frequency domain, resulting in SFBC [128, 235].
SFBCs eliminate the need for the channel to stay constant by
as many OFDM symbols as the number of transmit antennas
but requires the channel in frequency domain to be constant
by as many subcarriers as the number of transmit antennas. It
can be observed that when the codes are applied to the subcarriers over several OFDM symbols, then the diversity due to
the Doppler spread is utilized. In the case of SFBC, the diver-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Baseband
modulator

IFFT

CP

DK/2

CP

n Figure 16. Transmitter diversity with shifted pilots after IFFT


by the amount K/2.

sity due to the delay spread is exploited [35]. In the application of space-frequency Alamouti coding, a group of subcarriers by as many as the number of transmit antennas are
assigned to a group of Alamouti codes. The key assumption is
that the CFR is constant over the group of the subcarriers.
Such a scheme results in Ntx equations with Ntx unknowns per
CFR for each subcarrier block.
Depending on the system environment either STBC or
SFBC coding scheme can be used. When the length of CIR is
very small, then the use of SFBC can result in a good performance since the assumption of constant CFR over a number
of subcarriers holds. However, for more frequency selective
channels since the assumption of the constant channel no
longer holds, performance degradation will result in. In this
case, if the channel is less time selective, then the STBC can
be applied in time domain.
Similar to the SFBC, by assuming that the channel is constant by as many subcarrier as the number of transmit antennas, the pilot sequence after the IFFT of a transmit antenna is
shifted by K/Ntx, and CP is added thereafter as shown in Fig.
16 [236, 237]. With such a scheme, only one block of IFFT
can be used instead of Ntx IFFT blocks.
The shift by K/N tx results in the symbols with different
phase shifts in the frequency domain, which are used to separate the channel for each transmit antenna. Considering the
two transmit and one receive antenna system, the received signal can be written as,
Y[n, k] = [H11[n, k] + H12ejk] X[n, k] + W[n, k]

(89)

where X[n, k] is the only pilot symbol used for both antennas. With the assumption that the CFR is constant by as
many subcarriers as the number of transmitter antennas, Eq.
89 can be written for two consecutive subcarriers, with two
unknowns H11[n, k] and H12[n, k], which can be solved with
two equations. As can be seen such an approach is nothing
but some special version of the SFBC. This approach is also
simulated for many transmit antennas, and it is observed
that as long as the channel is not too frequency selective,
then the performance of the estimation is acceptable [238,
239]. Similar to Alamouti coding, rate-one non-orthogonal
space-time codes based on Hadamard codes are found to
give accurate channel estimation with the latter offering less
complexity [240].

MIMO-OFDM WITH SPATIAL CORRELATION


The use of multiple antennas in OFDM systems brings another dimension: spatial dimension. As with the frequency and
time domains correlation, spatial domain correlation can also
be exploited in the channel estimation of MIMO-OFDM systems. With uncorrelated CIR taps, the spatial correlation
between the subcarriers having the same indices is just the
spatial correlation between the antenna elements [37].
LMMSE filtering can be applied to the subcarriers across the
space. It is observed that the use of spatial correlation can
provide additional gain when the correlation is beyond 0.8 as

41

No spatial filtering
=0.99
=0.95
=0.90
=0.85

MSE

10-1

10-2

10

12
SNR [dB]

14

16

18

20

n Figure 17. MSE vs. antenna spatial correlation coefficients.


shown in Fig. 17. The use of spatial correlation is also investigated via Kalman filtering approach for the channel tracking
in time domain [164, 166]. The studies showed that in the
presence of spatial correlation channel tracking can still be
performed with the state equations incorporating the effect of
spatial correlation. In addition to these studies, spatial correlation is also found to improve the channel estimate of MIMO
systems via a pre-filtering in time domain [98], where a time
domain LMMSE channel estimation is exploited.

CONCLUSION
In this article, we present the most common methods applied
in the channel estimation of SISO and MIMO-OFDM systems. The SIMO and MISO systems are not covered separately as the methods for SISO and MIMO can be easily modified
to be applicable to SIMO and MISO systems. Throughout the
analysis it is seen that there are three basic blocks affecting
the performance of the channel estimation. These are the
pilot patterns, the estimation method, and the signal detection
part when combined with the channel estimation. As in many
systems, each block can promise an improved performance at
the cost of additional resources. Hence, the best combination
of these three parameters depends on the typical application
[60, 241]. Although the estimation techniques presented in
this article are shown to be a subset of LMMSE channel estimation technique, instead of promoting one of the channel
estimation techniques, the methods are presented for the scenarios they perform the best. Thus, a fully adaptive system can
be developed by using each block when necessary.

FUTURE DIRECTIONS
With OFDM now standing as a solid technology for future
wireless systems, OFDM channel estimation techniques can
be improved by incorporating the features of new technologies. It is well-known that one of the promising technologies is
MIMO. However, channel estimation methods studied for
MIMO-OFDM systems mostly overlook the effect of ICI due
to high speed mobile and external interferers. The models
that approximate ICI and external interferers as AWGN
might simplify the estimation process but better results can be
obtained by developing more accurate modelings.
Moreover, the standards such as WLAN and WiMAX do
not use certain subcarriers known as guard subcarriers. The
use of transform domain techniques do not provide better

42

performance with guard bands since transform domain techniques introduce CIR path leaks due to the suppression of
unused subcarriers. Methods can be developed to eliminate
the leakage problem by extrapolating the channel for the
unused subcarriers, followed by a transform domain technique. Such an approach can reduce the path leaking significantly. An elegant combination of an extrapolation method
and a transform domain technique can be developed so that a
practical estimation method can be realized for WLAN or
WiMAX systems.
As adaptation is key to many systems, channel estimation
techniques can be made adaptive by using the information
from other physical layer blocks. For example, the information available at blocks such as timing offset estimation, frequency offset estimation, and the output of the decoder can
all be used to determine the most appropriate channel estimation technique.
Lastly, mobile version of WiMAX uses OFDMA in its
uplink direction. The subcarriers in a given OFDMA symbol
are distributed among different users based on a given tile
structure and subchannels [177]. The pilot subcarriers for different tiles are no longer adjacent and the subcarrier spacing
between tiles can vary. Although linear interpolation can easily
be used for the channel estimation, utilization of long term
channel statistics can improve the channel estimation performance. With the tile assignment changing continuously during
the uplink transmission of OFDMA, the application of the
existing OFDM channel estimation methods is not straightforward. Research can be performed on how to practically incorporate long term channel statistics on the uplink channel
estimation of OFDMA systems for a better performing system.

REFERENCES
[1] G. J. Foschini And M. J. Gans, On Limits of Wireless Communications in a Fading Environment When Using Multiple Antennas, Wireless Pers. Commun., vol. 6, no. 3, Mar. 1998, pp.
31135.
[2] L. J. Cimini, Analysis and Simulation of a Digital Mobile Channel Using Orthogonal Frequency Division Multiplexing, IEEE
Trans. Commun., vol. 33, no. 7, July 1985, pp. 66575.
[3] Y. Li, J. H. Winters, and N. R. Sollenberger, Mimo-Ofdm For
Wireless Communications, Signal Detection With Enhanced
Channel Estimation, IEEE Trans. Commun., vol. 50, no. 9,
Sept. 2002, pp. 147177.
[4] M. Engels, Wireless OFDM Systems: How To Make Them
Work?, Kluwer Academic Publishers, 2002.
[5] I. Koffman and V. Roman, Broadband Wireless Access Solutions Based on OFDM Access in IEEE 802.16, IEEE Commun.
Mag., vol. 40, no. 4, Apr. 2002, pp. 96103.
[6] I. Barhumi, G. Leus, and M. Moonen, Optimal Training Design
For MimoOfdm Systems in Mobile Wireless Channels, IEEE
Trans. Signal Processing, vol. 51, no. 6, June 2003, pp.
161524.
[7] P. W. Wolniansky et al., V-Blast: An Architecture for Realizing
Very High Data Rates Over the RichScattering Wireless Channel, Proc. ISSSE98, vol. 1, Pisa, Italy, Sept. 1998.
[8] G. D. Golden et al., Detection Algorithms and Initial Laboratory Results Using V-Blast Space-Time Communication Architecture, IEE Electron. Lett., vol. 35, no. 1, Jan. 1999.
[9] W. Tong and B. Johnson, Views on IEEE 802.16 Evolution:
IEEE 802.16 Broadband Wireless Access Working Group, Apr.
2004, Available: Http://IEEE802.Org/16.
[10] P. J. Sartori, K. L. Baum, and F. W. Vook, Impact of Spatial
Correlation on the Spectral Efficiency of Wireless OFDM Systems Using Multiple Antenna Techniques, Proc. IEEE Vehic.
Tech. Conf., vol. 3, Birmingham, AL, May 2002, pp. 115054.
[11] R. Chen and K. B. Letaief, Channel Estimation for SpaceTime Coded OFDM Systems in Non-Sample-Spaced Multipath
Channels, Proc. IEEE Wireless Commun. and Networking
Conf., vol. 1, Orlando, FL, Mar. 2002, pp. 6166.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

[12] A. V. Zelst and T. C. W. Schenk, Implementation of a Mimo


OFDM Based Wireless Lan System, IEEE Trans. Signal Processing, vol. 52, no. 2, Feb. 2004, pp. 48394.
[13] Y. Li, L. J. Cimini, and N. R. Sollenberger, Robust Channel
Estimation for OFDM Systems with Rapid Dispersive Fading
Channels, IEEE Trans. Commun., vol. 46, no. 7, July 1998, pp.
90215.
[14] H. Arslan and G. E. Bottomley, Channel Estimation in Narrowband Wireless Communication Systems, Wireless Commun. and Mobile Comp., vol. 1, no. 2, Apr. 2001, pp. 20119.
[15] T. Himsoon, S. Weifeng, and K. J. R. Liu, Single-Block Differential Transmit Scheme for Broadband Wireless MIMOOFDM
Systems, IEEE Trans. Signal Processing, vol. 54, no. 9, pp.
330514, Sept. 2006.
[16] S. Weifeng and K. J. R. Liu, Differential SpaceFrequency
Modulation for MIMO-OFDM Systems Via a smooth Logical
Channel, Proc. IEEE Globecom Conf., vol. 2, San Francisco,
CA, Dec. 2004, pp. 9137.
[17] A. C. M. X. G. B. Giannakis, Block-Differential Modulation
over Doubly Selective Wireless Fading Channels, IEEE Trans.
Commun., vol. 53, Dec. 2005, pp. 215766.
[18] P. Tarasak and V. K. Bhargava, 1/4/4-Shifted Differential
Qpsk Space-Time Block Codes, Electronis Letters, vol. 39, no.
18, Sept. 2003, pp. 132729.
[19] W. Hashim et al., Performance Comparison of Differential
Space-Time Signalling Schemes for OFDM Systems, Proc. IEEE
Malaysia Intl. Conf. Commun. and Intl. Conf. Networks, vol.
2, Kuala Lumpur, Malaysia, Nov. 2005, pp. 16.
[20] A. I. El-Arabawy and S. C. Gupta, Reduced Mobile Complexity Scheme for Fast Fading Channel Estimation in OFDM-FDD
Mobile Communication Systems, Proc. IEEE Intl. Conf. Universal Personal Commun., vol. 1, San Diego, CA, Oct. 1997, pp.
27478.
[21] I. Harjula, A. Mammela, and Z. Li, Comparison of Channel
Frequency and Impulse Response Estimation for Space Time
Coded OFDM Systems, Proc. IEEE Vehic. Tech. Conf., vol. 4,
Vancouver, Canada, Sept. 2002, pp. 208185.
[22] S. G. Kang, Y. M. Ha, and E. K. Joo, A Comparative Investigation on Channel Estimation Algorithms for OFDM in Mobile
Communications, IEEE Trans. Broadcast., vol. 49, no. 2, June
2003, pp. 14249.
[23] S. M. Lee, D. H. Lee, and H. J. Choi, Performance Comparison of Space-Time Codes and Channel Estimation in OFDM
Systems with Transmit Diversity for Wireless Lans, Proc. AsiaPacific Conf. Commun., vol. 1, Penang, Malaysia, Sept. 2003,
pp. 2124.
[24] M. Sandell and O. Edfors, A Comparative Study of PilotBased Channel Estimators for Wireless OFDM, Lulea University
of Technology, Lulea, Research Report Tulea, Sept. 1996.
[25] S. Wu and Y. Bar-Ness, OFDM Channel Estimation in the
Presence of Frequency Offset and Phase Noise, Proc. IEEE
Intl. Conf. Commun., vol. 5, Anchorage, AK, May 2003, pp.
336670.
[26] B. Han et al., An Iterative Joint Channel Estimation and Symbol Detection Algorithm Applied in OFDM System with High
Data To Pilot Power Ratio, Proc. IEEE Intl. Conf. Commun.,
vol. 3, Anchorage, AK, May 2003, pp. 207680.
[27] I. W. Group, IEEE 802.11 Wireless Local Area Networks,
May 2001,
[Online]. Available: Http://Grouper.IEEE.Org/Groups/802/11.
[28] H. Tang, K. Y. Lau, and R. W. Brodersen, Interpolation-Based
Maximum Likelihood Channel Estimation Using OFDM Pilot
Symbols, Proc. IEEE Globecom Conf., vol. 2, Taipei, Taiwan,
Nov. 2002, pp. 186064.
[29] Y. Sun, X. Wang, and K. J. R. Liu, A Joint Channel Estimation and Unequal Error Protection Scheme for Video Transmission in OFDM Systems, Proc. IEEE Intl. Conf. Image
Processing, vol. 1, Rochester, NY, Sept. 2002, pp. 54952.
[30] Y. Sun, X. Wang, and K. J. R. Liu, A Joint Channel Estimation and Unequal Error Protection Scheme for Image Transmission in Wireless OFDM Systems, Proc. IEEE Wksp. Multimedia
Signal Processing, vol. 1, St. Thomas, US Virgin Islands, Dec.
2002, pp. 38083.
[31] M. Bossert, A. Donder, and A. Trushkin, Channel Estimation
and Equalization in Orthogonal Frequency Division Multiplex-

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

ing Systems, Proc. ITG-Fachberitcht, Mobile Kommunikation,


vol. 1, Neu-Ulm, Germany, Sept. 1995, pp. 48592.
[32] V. Mignone, A. Morello, and M. Visintin, Cd3OFDM: A New
Channel Estimation Method to Improve the Spectrum Efficiency in Digital Terrestrial Television Systems, Proc. Intl. Broadcasting Convention, vol. 1, Amsterdam, the Netherlands, Sept.
1995, pp. 12228.
[33] C. Sgraja and J. Lindner, Estimation of Rapid Time-Variant
Channels for OFDM Using Wiener Filtering, Proc. IEEE Intl.
Conf. Commun., vol. 4, Anchorage, AK, May 2003, pp. 23905.
[34] W. Bai et al., Robust Channel Estimation in MIMOOFDM
Systems, Electronis Letters, vol. 39, no. 2, Jan. 2003, pp.
24244.
[35] R. J. Piechocki, A. R. Nix, and J. P. Mcgeehan, Joint SemiBlind Detection and Channel Estimation in SpaceFrequency
Trellis Coded MIMOOFDM, Proc. IEEE Intl. Conf. Commun.,
vol. 5, Anchorage, AK, May 2003, pp. 338286.
[36] O. Edfors et al., OFDM Channel Estimation By Singular Value
Decomposition, IEEE Trans. Commun., vol. 46, no. 7, July
1998, pp. 93139.
[37] M. K. Ozdemir, H. Arslan, and E. Arvas, MIMOOFDM Channel Estimation for Correlated Fading Channels, Proc. IEEE
Wireless and Microwave Technol. Conf., vol. 1, Clearwater, FL,
Apr. 2004, pp. 15.
[38] Y. Li, Simplified Channel Estimation for OFDM Systems with
Multiple Transmit Antennas, IEEE Trans. Wireless Commun.,
vol. 1, no. 1, Jan. 2002, pp. 6775.
[39] R. Negi and J. Cioffi, Pilot Tone Selection for Channel Estimation in a Mobile OFDM System, IEEE Trans. Consumer Electron., vol. 44, no. 3, Aug. 1998, pp. 112228.
[40] J. Cai, X. Shen, and J. W. Mark, EM Channel Estimation
Algorithm for OFDM Wireless Communication Systems, Proc.
IEEE Intl. Symp. Personal, Indoor and Mobile Radio Commun.,
vol. 1, Beijing, China, Sept. 2003, pp. 80408.
[41] J. H. Lee et al., Pilot Symbol Initiated Iterative Channel Estimation and Decoding for Qam Modulated OFDM Signals,
Proc. IEEE Vehic. Tech. Conf., vol. 2, Jeju, Korea, Apr. 2003,
pp. 132226.
[42] B. M. Sadler, R. J. Kozick, and T. Moore, Bounds on MIMO
Channel Estimation and Equalization with Side Information,
Proc. IEEE Intl. Conf. Acoust., Speech, and Signal Processing,
vol. 4, Salt Lake City, UT, May 2001, pp. 214548.
[43] I. Bradaric and A. P. Petropulu, Performance of
TrainingBased OFDM Systems in the Presence of Time Varying
FrequencySelective Channels, Proc. IEEE Intl. Conf. Acoust.,
Speech, and Signal Processing, vol. 4, Montreal, Canada, May
2004, pp. 73740.
[44] M. Li, J. Tan, and W. Zhang, A Channel Estimation Method
Based on FrequencyDomain Pilots and TimeDomain Processing for OFDM Systems, IEEE Trans. Consumer Electron., vol.
50, no. 4, Nov. 2004, pp. 104957.
[45] I. Gaspard, Impact of the Channel Estimation Onto the
BerPerformance of PsamOFDM Systems in Mobile Radio
Channels, Proc. IEEE Vehic. Tech. Conf., vol. 1, Rhodes,
Greece, May 2001, pp. 67377.
[46] R. Chen, C. Hou, and J. Guo, Symbol Timing and Channel
Estimation of IEEE802.11a Based on OFDM, IEEE Canadian
Conf. Electr. and Computer Eng., vol. 3, Montreal, Canada,
May 2003, pp. 154750.
[47] H. Arslan and T. Yucek, Estimation of Frequency Selectivity
for OFDM Based New Generation Wireless Communication Systems, Proc. World Wireless Congress, vol. 1, San Francisco,
Ca, May 2003.
[48] S. T. Brink, F. Sanzi, and J. Speidel, TwoDimensional Iterative App Channel Estimation and Decoding for OFDM Systems, Proc. IEEE Globecom Conf., vol. 2, San Francisco, Ca,
Dec. 2000, pp. 74145.
[49] Q. Sun et al., Estimation of Continuous Flat Fading MIMO
Channels, IEEE Trans. Wireless Commun., vol. 1, no. 4, Oct.
2002, pp. 54953.
[50] J. K. Cavers, An Analysis of Pilot Symbol Assisted Modulation for Rayleigh Fading Channels, IEEE Trans. Vehic. Tech.,
vol. 40, Nov. 1991, pp. 68693.
[51] Y. Li, Optimum Training Sequences for OFDM Systems with
Multiple Transmit Antennas, Proc. IEEE Globecom Conf., vol.

43

3, San Francisco, CA, Nov. 2000, pp. 147882.


[52] A. Dowler, A. Doufexi, and A. Nix, Performance Evaluation
of Channel Estimation Techniques for a Mobile Fourth Generation Wide Area OFDM System, Proc. IEEE Vehic. Tech. Conf.,
vol. 4, Vancouver, Canada, Sept. 2002, pp. 203640.
[53] I. Barhumi, G. Leus, and M. Moonen, Optimal Training
Sequences for Channel Estimation in MIMO OFDM Systems in
Mobile Wireless Channels, Proc. Intl. Zurich Seminar on
Broadband Commun., vol. 44, Zurich, Switzerland, Feb. 2002,
pp. 16.
[54] M. Shin, H. Lee, and C. Lee, Enhanced ChannelEstimation
Technique for MIMOOFDM Systems, IEEE Trans. Vehic. Tech.,
vol. 53, no. 1, Jan. 2004, pp. 26165.
[55] S. Boumard and A. Mammela, Channel Estimation Versus
Equalization in An OFDM Wlan System, Proc. IEEE Vehic.
Tech. Conf., vol. 1, Rhodes, Greece, May 2001, pp. 65357.
[56] F. Tufvesson and T. Maseng, Pilot Assisted Channel Estimation for OFDM in Mobile Cellular Systems, Proc. IEEE Vehic.
Tech. Conf., vol. 3, Phoenix, AZ, May 1997, pp. 163943.
[57] Z. Jianhua and Z. Ping, An Improved 2Dimensional
PilotSymbols Assisted Channel Estimation in OFDM Systems,
Proc. IEEE Vehic. Tech. Conf., vol. 3, Jeju, Korea, Apr. 2003,
pp. 159599.
[58] M. Morelli and U. Mengali, A Comparison of PilotAided
Channel Estimation Methods for OFDM Systems, IEEE Trans.
Signal Processing, vol. 49, no. 12, Dec. 2001, pp. 306573.
[59] C. Hou, S. Song, and D. Cao, Channel Estimation for Adaptive OFDM System and Effects of Estimation Error on System
Performance, Proc. IEEE Int. Wksp. Mobile and Wireless Commun. Network, vol. 1, Stockholm, Sweden, Sept. 2002, pp.
2004.
[60] A. Dowler and A. Nix, Performance Evaluation of Channel
Estimation Techniques in a Multiple Antenna OFDM System,
Proc. IEEE Vehic. Tech. Conf., vol. 2, Orlando, FL, Oct. 2003,
pp. 121418.
[61] M. J. F. G. Garcia, J. M. P. Borrallo, and S. Zazo, Efficient
Pilot Patterns for Channel Estimation in OFDM Systems over Hf
Channels, Proc. IEEE Vehic. Tech. Conf., vol. 4, Amsterdam,
the Netherlands, Sept. 1999, pp. 219397.
[62] M. J. F. G. Garcia, S. Zazo, and J. M. P. Borrallo, Pilot Patterns for Channel Estimation in OFDM, IEE Electron. Lett., vol.
36, no. 12, June 2000, pp. 104950.
[63] W. G. Jeon, K. H. Paik, and Y. S. Cho, Two-Dimensional
MMSE Channel Estimation for OFDM Systems with Transmitter
Diversity, Proc. IEEE Vehic. Tech. Conf., vol. 3, Atlantic City,
NJ, Oct. 2001, pp. 168285.
[64] C. Suh, C. S. Hwang, and H. Choi, Preamble Design for
Channel Estimation in MIMO-OFDM Systems, Proc. IEEE
Globecom Conf., vol. 1, San Francisco, CA, Dec. 2003, pp.
31721.
[65] M. J. F. G. Garcia, O. Edfors, and J. M. P. Borrallo, Joint
Channel Estimation and Peak-To-Average Power Reduction in
Coherent OFDM: a Novel Approach, Proc. IEEE Vehic. Tech.
Conf., vol. 2, Rhodes, Greece, May 2001, pp. 81519.
[66] X. Wang and K. J. R. Liu, OFDM Channel Estimation Based
on Time-Frequency Polynomial Model of Fading Multipath
Channel, Proc. IEEE Vehic. Tech. Conf., vol. 1, Atlantic City,
NJ, Oct. 2001, pp. 46064.
[67] L. Maniatis, T. Weber, A. Sklavos, and Y. Liu, Pilots for Joint
Channel Estimation in Multi-User OFDM Mobile Radio Systems, Proc. IEEE Intl. Symp. Spread Spectrum Tech. and App.,
vol. 1, Prague, Czech Republic, Sept. 2002, pp. 4448.
[68] M. S. Zimmerman and A. L. Kirsch, The AN/GSC-10 (Kathryin)
Variable Rate Data Modem for HF Radio, IEEE Trans. Commun., vol. 15, no. 2, Apr. 1967, pp. 197204.
[69] C. S. Yeh and Y. Lin, Channel Estimation Using Pilot Tones in
OFDM Systems, IEEE Trans. Broadcast., vol. 45, no. 4, Dec.
1999, pp. 40009.
[70] D. Wang, G. Zhu, and Z. Hu, A Combined Channel Estimation in Time-Frequency Domain for OFDM System in Mobile
Channel, Proc. IEEE Intl. Symp. Personal, Indoor and Mobile
Radio Commun., vol. 1, Beijing, China, Sept. 2003, pp. 94044.
[71] C. Budianu and L. Tong, Channel Estimation for Space-Time
Orthogonal Block Codes, IEEE Trans. Signal Processing, vol.
50, no. 10, Oct. 2002, pp. 251528.

44

[72] W. G. Jeon, K. H. Paik, and Y. S. Cho, An Efficient Channel


Estimation Technique for OFDM Systems with Transmitter
Diversity, Proc. IEEE Intl. Symp. Personal, Indoor and Mobile
Radio Commun., vol. 2, London, UK, Sept. 2000, pp. 124650.
[73] M. Bossert, A. Donder, and V. Zyablov, Improved Channel
Estimation with Decision Feedback for OFDM Systems, IEE
Electron. Lett., vol. 34, no. 11, May 1998, pp. 106465.
[74] Z. Cheng and D. Dahlhaus, Time Versus Frequency Domain
Channel Estimation for OFDM Systems with Antenna Arrays,
Proc. Intl. Conf. Signal Proc., vol. 2, Beijing, China, Aug. 2002,
pp. 134043.
[75] J. J. Van De Beek et al., On Channel Estimation in OFDM
Systems, Proc. IEEE Vehic. Tech. Conf., vol. 2, Rosemont, IL,
July 1995, pp. 81519.
[76] O. Edfors, et al., Analysis of DFT-Based Channel Estimators
for OFDM, Personal Wireless Commun., Kluwer Academic
Publisher, vol. 12, no. 1, Jan. 2000, pp. 5570.
[77] Y. Li, N. Seshadri, and S. Ariyavisitakul, Channel Estimation
for OFDM Systems with Transmitter Diversity in Mobile Wireless Channels, IEEE JSAC, vol. 17, no. 3, Mar. 1999, pp.
46171.
[78] E. G. Larsson et al., Joint Symbol Timing and Channel Estimation for OFDM Based Wlans, IEEE Commun. Lett., vol. 5,
no. 8, Aug. 2001, pp. 32527.
[79] F. D. Nunes and J. M. N. Leitao, Nonlinear Channel Estimation in Fading OFDM Systems, Proc. IEEE Globecom Conf.,
vol. 4, Rio De Janeireo, Brazil, Dec. 1999, pp. 215761.
[80] E. G. Larsson et al., An Algorithm for Joint Symbol Timing
and Channel Estimation for OFDM Systems, Proc. IEEE Statistical Signal Processing Wksp., vol. 1, Singapore, Aug. 2001, pp.
39396.
[81] Z. Yuanjin, A Novel Channel Estimation and Tracking
Method for Wireless OFDM Systems Based on Pilots and
Kalman Filtering, IEEE Trans. Consumer Electron., vol. 49, no.
2, May 2003, pp. 27583.
[82] H. Yu, M. S. Kim, and S. K. Lee, Channel Estimation and
Equalization for High Speed Mobile OFDM Systems, Proc.
Asilomar Conf. Signals, Systems and Computers, vol. 1, Monterey, CA, Nov. 2003, pp. 69397.
[83] F. Said and H. Aghvami, Linear Two Dimensional Pilot Assisted Channel Estimation for OFDM Systems, Proc. IEE Telecommun. Conf., vol. 451, Edinburgh, Uk, Apr. 1998, pp. 3236.
[84] S. Coleri et al., Channel Estimation Techniques Based on
Pilot Arrangement in OFDM Systems, IEEE Trans. Broadcast.,
vol. 48, no. 3, Sept. 2002, pp. 22329.
[85] M. X. Chang and Y. T. Su, ModelBased Channel Estimation
for OFDM Signals in Rayleigh Fading, IEEE Trans. Commun.,
vol. 50, no. 4, Apr. 2002, pp. 54044.
[86] J. Rinne and M. Renfors, Pilot Spacing in Orthogonal Frequency Division Multiplexing Systems on Practical Channels,
IEEE Trans. Consumer Electron., vol. 42, no. 3, Nov. 1996, pp.
95962.
[87] C. R. N. Athaudage and A. D. S. Jayalath, LowComplexity
Channel Estimation for Wireless OFDM Systems, Proc. IEEE
Intl. Symp. Personal, Indoor and Mobile Radio Commun., vol.
1, Beijing, China, Sept. 2003, pp. 52125.
[88] F. Said and H. Aghvami, Two Dimensional Pilot Assisted
Channel Estimation for Turbo Coded OFDM Systems, Proc. IEE
Colloquium on Turbo Codes in Digital Broadcast., vol. 19, London, Uk, Nov. 1999, pp. 16.
[89] S. Coleri, M. Ergen, A. Puri, and A. Bahai, A Study of Channel Estimation in OFDM Systems, Proc. IEEE Vehic. Tech.
Conf., vol. 2, Vancouver, Canada, Sept. 2002, pp. 89498.
[90] T. Onizawa et al., A Simple Adaptive Channel Estimation
Scheme for OFDM Systems, Proc. IEEE Vehic. Tech. Conf., vol.
1, Amsterdam, the Netherlands, Sept. 1999, pp. 27983.
[91] A. Huang and Y. Zhao, Estimating Channel Response From
Pilot Subcarrier Pairs for OFDM Systems, Proc. Midwest Symp.
on Circuits and Systems, vol. 2, Sacramento, CA, Aug. 1997,
pp. 77477.
[92] J. K. Moon and S. I. Choi, Performance of Channel Estimation Methods for OFDM Systems in a Multipath Fading Channels, IEEE Trans. Consumer Electron., vol. 46, Feb. 2000, pp.
16170.
[93] X. Ma, H. Kobayashi, and S. C. Schwartz, An Enhanced

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

Channel Estimation Algorithm for OFDM: Combined EM Algorithm and Polynomial Fitting, Proc. IEEE Intl. Conf. Acoust.,
Speech, and Signal Processing, vol. 4, Hong Kong, China, Apr.
2003, pp. 68083.
[94] M. X. Chang and Y. T. Su, 2D Regression Channel Estimation
for Equalizing OFDM Signals, Proc. IEEE Vehic. Tech. Conf.,
vol. 1, Tokyo, Japan, May 2000, pp. 24044.
[95] T. M. Lehmann, C. Gnner, and K. Spitzer, Survey: Interpolation Methods in Medical Image Processing, IEEE Trans. Med.
Imag., vol. 18, no. 11, Nov. 1999, pp. 104975.
[96] X. Zhou and X. Wang, Channel Estimation for OFDM Systems Using Adaptive Radial Basis Function Networks, IEEE
Trans. Vehic. Tech., vol. 52, no. 1, Jan. 2003, pp. 4859.
[97] L. Tong, B. M. Sadler, and M. Dong, Pilot-Assisted Wireless
Transmissions General Model, Design Criteria, and Signal
Processing, IEEE Signal Processing Mag., vol. 21, no. 6, Nov.
2004, pp. 1225.
[98] M. Kiessling et al., Statistical Prefiltering for MMSE and ML
Receivers with Correlated MIMO Channels, Proc. IEEE Wireless
Commun. and Networking Conf., vol. 2, New Orleans, LA,
Mar. 2003, pp. 91924.
[99] J. S. Chow, J. M. Cioffi, and J. A. C. Bingham, Equalizer
Training Algorithms for Multicarrier Modulation Systems,
Proc. IEEE Intl. Conf. Commun., vol. 2, Geneva, Switzerland,
May 1993, pp. 76165.
[100] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, DiscreteTime Signal Processing, 2nd Ed. Prentice Hall, 1999.
[101] M. J. F. G. Garcia, J. M. Paez-Borrallo, and S. Zazo,
DFTBased Channel Estimation in 2D-Pilot-Symbol-Aided
OFDM Wireless Systems, Proc. IEEE Vehic. Tech. Conf., vol. 2,
Rhodes, Greece, May 2001, pp. 81014.
[102] C. W. Wong, C. L. Law, and Y. L. Guan, Channel Estimator
for OFDM Systems with 2-Dimensional Filtering in the Transform Domain, Proc. IEEE Vehic. Tech. Conf., vol. 1, Rhodes,
Greece, May 2001, pp. 71721.
[103] S. Sun et al., Training Sequence Assisted Channel Estimation for MIMO OFDM, Proc. IEEE Wireless Commun. and Networking Conf., vol. 1, New Orleans, LA, Mar. 2003, pp. 3843.
[104] Y. Zhao and W. L. and W. Wu, An Efficient Channel Estimation Method for OFDM Systems with Multiple Transmit
Antennas, Proc. Intl. Conf. Info-Tech and Info-Net, vol. 2,
Beijing, China, Nov. 2001, pp. 33539.
[105] A. A. Tahat and D. R. Ucci, An Extrapolated Matched-Filter
Approach to Multi-User Channel Estimation for OFDM in
SDMA, Proc. IEEE Antennas and Propagat. Soc. Conf., vol. 2,
San Antonio, TX, June 2002, pp. 63639.
[106] Y. Li and N. R. Sollenberger, Clustered OFDM with Channel
Estimation for High Rate Wireless Data, Proc. IEEE Intl. Wksp.
Mobile Multimedia Commun., vol. 1, San Diego, CA, Nov.
1999, pp. 4350.
[107] Y. Li and N. R. Sollenberger, Clustered OFDM with Channel
Estimation for High Rate Wireless Data, IEEE Trans. Commun.,
vol. 49, no. 12, Dec. 2001, pp. 207176.
[108] Y. H. Yeh and S. G. Chen, Efficient Channel Estimation
Based on Discrete Cosine Transform, Proc. IEEE Intl. Conf.
Acoust., Speech, and Signal Processing, vol. 4, Hong Kong,
China, Apr. 2000, pp. 67679.
[109] Y. H. Yeh and S. G. Chen, DctBased Channel Estimation
for OFDM Systems, Proc. IEEE Intl. Conf. Commun., vol. 4,
Paris, France, June 2004, pp. 24426.
[110] M. Stege, P. Zillmann, and G. Fettweis, MIMO Channel
Estimation with Dimension Reduction, Proc. Intl. Symp.Wireless Personal Multimedia Commun., vol. 2, Honolulu, HI, Oct.
2002, pp. 41721.
[111] Y. Li, Pilot-Symbol-Aided Channel Estimation for OFDM in
Wireless Systems, Proc. IEEE Vehic. Tech. Conf., vol. 2, Houston, TX, May 1999, pp. 113135.
[112] S. He and M. Torkelson, Computing Partial DFT for Comb
Spectrum Evaluation, IEEE Signal Processing Lett., vol. 3, no.
6, June 1996, pp. 17375.
[113] Y. Zhao and A. Huang, A Novel Channel Estimation
Method for OFDM Mobile Communication Systems Based on
Pilot Signals and Transform-Domain Processing, Proc. IEEE
Vehic. Tech. Conf., vol. 3, Phoenix, AZ, May 1997, pp.
208993.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

[114] T. F. Wong and B. Park, Training Sequence Optimization in


MIMO Systems with Colored Interference, IEEE Trans. Commun., vol. 52, no. 11, Nov. 2004, pp. 193947.
[115] H. Z. Jafarian, M. J. Omidi, and S. Pasupathy, Improved
Channel Estimation Using Noise Reduction for OFDM Systems,
Proc. IEEE Vehic. Tech. Conf., vol. 2, Jeju, Korea, Apr. 2003,
pp. 130812.
[116] B. Yang et al., Windowed Dft Based Pilot-Symbol-Aided
Channel Estimation for OFDM Systems in Multipath Fading
Channels, Proc. IEEE Vehic. Tech. Conf., vol. 2, Tokyo, Japan,
May 2000, pp. 148084.
[117] B. Yang, Z. Cao, and K. B. Letaief, Analysis of Low-Complexity Windowed Dft-Based MMSE Channel Estimator for
OFDM Systems, IEEE Trans. Commun., vol. 49, no. 11, Nov.
2001, pp. 197787.
[118] Y. Li, Pilot-Symbol-Aided Channel Estimation for OFDM in
Wireless Systems, IEEE Trans. Vehic. Tech., vol. 49, no. 4, July
2000, pp. 120715.
[119] B. Wu and S. Cheng, Clipping Effects on Channel Estimation and Signal Detection in OFDM, Proc. IEEE Intl. Symp.
Personal, Indoor and Mobile Radio Commun., vol. 1, Beijing,
China, Sept. 2003, pp. 53134.
[120] M. K. zdemir, H. Arslan, and E. Arvas, Towards RealTime
Adaptive LowRank Lmmse Channel Estimation of
MIMOOFDM Systems, IEEE Trans. Wireless Commun., vol. 5,
no. 10, Oct. 2006, pp. 267578.
[121] J. X. Du and Y. Li, MIMO-OFDM Channel Estimation Based
on Subspace Tracking, Proc. IEEE Vehic. Tech. Conf., vol. 2,
Jeju, Korea, Apr. 2003, pp. 108488.
[122] B. Han et al., A Novel Channel Estimation Method for
OFDM Systems in Multipath Fading, Proc. IEEE Globecom
Conf., vol. 1, Taipei, Taiwan, Nov. 2002, pp. 696700.
[123] T. A. Thomas and F. W. Vook, Broadband MIMOOFDM
Channel Estimation Via Near Maximum Likelihood Time of
Arrival Estimation, Proc. IEEE Intl. Conf. Acoust., Speech, and
Signal Processing, vol. 3, Orlando, Fl, May 2002, pp. 256972.
[124] Z. J. Wang, Z. Han, and K. J. R. Liu, MIMOOFDM Channel
Estimation Via Probabilistic Data Association Based Toas,
Proc. IEEE Globecom Conf., vol. 2, San Francisco, Ca, Dec.
2003, pp. 62630.
[125] B. Yang et al., Robust and Improved Channel Estimation
for OFDM Systems in Frequency Selective Fading Channels,
Proc. IEEE Globecom Conf., vol. 5, Rio De Janeireo, Brazil, Dec.
1999, pp. 2499503.
[126] B. Yang et al., Channel Estimation for OFDM Transmission
in Multipath Fading Channels Based on Parametric Channel
Modeling, IEEE Trans. Commun., vol. 49, no. 3, Mar. 2001,
pp. 46779.
[127] H. Minn and V. K. Bhargava, An Investigation Into TimeDomain Approach for OFDM Channel Estimation, IEEE Trans.
Broadcast., vol. 46, no. 4, Dec. 2000, pp. 24048.
[128] H. Minn, D. I. Kim, and V. K. Bhargava, A Reduced Complexity Channel Estimation for OFDM Systems with Transmit
Diversity in Mobile Wireless Channels, IEEE Trans. Commun.,
vol. 50, no. 5, May 2002, pp. 799807.
[129] T. Sekit et al., The Channel Estimation Technique Based on
the Propagation Path Model of An OFDM System, Proc. IEEE
Intl. Symp. Personal, Indoor and Mobile Radio Commun., vol.
2, San Diego, Ca, Oct. 2001, pp. E88E92.
[130] M. Nakamura et al., New Estimation and Equalization
Approach for OFDM Under Doppler-Spread Channel, Proc.
IEEE Intl. Symp. Personal, Indoor and Mobile Radio Commun.,
vol. 2, Lisboa, Portugal, Sept. 2002, pp. 55560.
[131] T. Yamamura and H. Harada, High Mobility OFDM Transmission System By a New Channel Estimation and ISI Cancellation Scheme Using Characteristics of Pilot Symbol Inserted
OFDM Signal, Proc. IEEE Vehic. Tech. Conf., vol. 1, Amsterdam, the Netherlands, Sept. 1999, pp. 31923.
[132] S. Y. Park and C. G. Kang, DecisionDirected Channel Estimation for MQamModulated OFDM Systems, Proc. IEEE
Vehic. Tech. Conf., vol. 3, Birmingham, AL, May 2002, pp.
174246.
[133] P. Hoeher, S. Kaiser, and P. Robertson, 2-D Pilot-Symbol
Aided Channel Estimation By Wiener Filtering, Proc. IEEE Intl.
Conf. Acoust., Speech, and Signal Processing, vol. 3, Apr.

45

1997, pp. 184548.


[134] R. Nilsson et al., An Analysis of TwoDimensional
PilotSymbol Assisted Modulation for OFDM, Proc. IEEE Intl.
Symp. Personal, Indoor and Mobile Radio Commun., vol. 1,
Bombay, India, Dec. 1997, pp. 7174.
[135] Y. Gong and K. B. Letaief, Low Rank Channel Estimation
for Space-Time Coded Wideband OFDM Systems, Proc. IEEE
Vehic. Tech. Conf., vol. 2, Atlantic City, NJ, Oct. 2001, pp.
77276.
[136] Y. Gong and K. B. Letaief, Low Complexity Channel Estimation for Space-Time Coded Wideband OFDM Systems, IEEE
Trans. Wireless Commun., vol. 2, no. 5, Sept. 2003, pp.
87682.
[137] S. W. Ellingson, Detection of Tones and Pulses Using a
Large, Uncalibrated Array, Proc. IEEE Antennas and Propagat.
Soc. Conf., vol. 4, Columbus, OH, June 2003, pp. 1969.
[138] O. Simeone, Y. BarNess, and U. Spagnolini, Pilot-Based
Channel Estimation for OFDM Systems by Tracking the DelaySubspace, IEEE Trans. Wireless Commun., vol. 3, no. 1, Jan.
2004, pp. 31525.
[139] I. Tolochko and M. Faulkner, Real Time Lmmse Channel
Estimation for Wireless OFDM Systems with Transmitter Diversity, Proc. IEEE Vehic. Tech. Conf., vol. 3, Vancouver, Canada,
Sept. 2002, pp. 155559.
[140] M. H. Hsieh and C. H. Wei, Channel Estimation for OFDM
Systems Based on Comb-Type Pilot Arrangement in Frequency
Selective Fading Channels, IEEE Trans. Consumer Electron.,
vol. 44, no. 1, Feb. 1998, pp. 21725.
[141] P. Comon and G. H. Golub, Tracking a Few Extreme Singular Values and Vectors in Signal Processing, Proc. IEEE, vol.
78, no. 8, Aug. 1990, pp. 132743.
[142] P. Strobach, Low-Rank Adaptive Filters, IEEE Trans. Signal
Processing, vol. 44, no. 12, Dec. 1996, pp. 293247.
[143] L. Yue et al., An Improved Channel Estimation Scheme for
OFDM Systems By Tracking the Subspace, Proc. IEEE Intl.
Symp. Personal, Indoor and Mobile Radio Commun., vol. 2,
Beijing, China, Sept. 2003, pp. 110913.
[144] D. Schafhuber, G. Matz, and F. Hlawatsch, Adaptive
Wiener Filters for TimeVarying Channel Estimation in Wireless
OFDM Systems, Proc. IEEE Intl. Conf. Acoust., Speech, and
Signal Processing, vol. 4, Hong Kong, China, Apr. 2003, pp.
68891.
[145] P. Strobach, Bi-Iteration Svd Subspace Tracking Algorithms, IEEE Trans. Signal Processing, vol. 45, no. 5, May
1997, pp. 122240.
[146] P. Hoeher, Tcm on FrequencySelective LandMobile Fading Channels, Proc. Tirrenia Intern. Wksp. Digital Commun.,
vol. 1, Tirrenia, Italy, Sept. 1991, pp. 31728.
[147] V. P. Kaasila, Performance Analysis of An OFDM System
Using DataAided Channel Estimation, Proc. IEEE Vehic. Tech.
Conf., vol. 3, Houston, TX, May 1999, pp. 230307.
[148] F. Shu et al., TimeFrequency Channel Estimation for Digital Amplitude Modulation Broadcasting Systems Based on
OFDM, Proc. IEE, vol. 150, no. 4, Aug. 2003, pp. 25964.
[149] A. A. Hutter, R. Hasholzner, and J. S. Hammerschmidt,
Channel Estimation for Mobile OFDM Systems, Proc. IEEE
Vehic. Tech. Conf., vol. 1, Amsterdam, the Netherlands, Sept.
1999, pp. 30509.
[150] J. H. Ryu and Y. H. Lee, Design of Implementation-Efficient
Channel Estimation Filter for Wireless OFDM Transmission,
Proc. IEEE Vehic. Tech. Conf., vol. 3, Jeju, Korea, Apr. 2003,
pp. 159094.
[151] A. F. Kurpiers, Improved Channel Estimation and Demodulation for OFDM on Hf Ionospheric Channels, Proc. IEE Intl.
Conf. Hf Radio Systems and Techniques, vol. 1, Guildford, Uk,
July 2000, pp. 6569.
[152] H. H. Hmimy, Channel Estimation Based on Coded Pilot
for OFDM, Proc. IEEE Vehic. Tech. Conf., vol. 3, Phoenix, AZ,
May 1997, pp. 137579.
[153] S. C. Schwartz, J. Zhang, and D. Gu, Iterative Channel Estimation for OFDM with Clipping, Proc. Intl. Symp.Wireless
Personal Multimedia Commun., vol. 3, Honolulu, HI, Oct. 2002,
pp. 130408.
[154] M. Flament, B. Mielczarek, and A. Svensson, Joint Channel
Estimation and Turbo Decoding for OFDM-Based Systems,

46

Proc. Intl. Symp.Wireless Personal Multimedia Commun., vol.


3, Honolulu, HI, Oct. 2002, pp. 1299303.
[155] L. Jarbot, Combined Decoding and Channel Estimation of
OFDM Systems in Mobile Radio Networks, Proc. IEEE Vehic.
Tech. Conf., vol. 3, Phoenix, AZ, May 1997, pp. 160104.
[156] S. Y. Park et al., Iterative Receiver with Joint Detection and
Channel Estimation for OFDM System with Multiple Receiver
Antennas in Mobile Radio Channels, Proc. IEEE Globecom
Conf., vol. 5, San Antonio, TX, Nov. 2001, pp. 308589.
[157] J. Cho et al., A Novel Channel Estimation Method for
OFDM in HighSpeed Mobile System, Proc. IEEE Intl.
Symp.Ind. Electron., vol. 1, Pusan, Korea, June 2001, pp.
57174.
[158] J. Siew et al., A Channel Estimation Method for
MIMOOFDM Systems, Proc. London Commun. Symp., London, England, Sept. 2002, pp. 14.
[159] S. B. Bulumulla, S. A. Kassam, and S. S. Venkatesh, A Systematic Approach in Detecting OFDM Signals in a Fading
Channel, IEEE Trans. Commun., vol. 48, no. 5, May 2000, pp.
72528.
[160] Y. Wang et al., Threshold Channel Estimation for OFDM in
Wireless Systems, Proc. IEEE Vehic. Tech. Conf., vol. 3, Jeju,
Korea, Apr. 2003, pp. 158689.
[161] J. Kim, G. L. Stuber, and Y. Li, Iterative Joint Channel Estimation and Detection Combined with PilotTone Symbols in
Convolutionally Coded OFDM Systems, Proc. IEEE Intl. Symp.
Personal, Indoor and Mobile Radio Commun., vol. 1, Beijing,
China, Sept. 2003, pp. 53539.
[162] W. Lee and J. Zhu, Channel Estimation for High-Speed
Packet-Based OFDM Communication Systems, Proc. Intl.
Symp.Wireless Personal Multimedia Commun., vol. 3, Honolulu, HI, Oct. 2002, pp. 129398.
[163] J. Zhu and W. Lee, Channel Estimation with Power-Controlled Pilot Symbols and Decision-Directed Reference Symbols, Proc. IEEE Vehic. Tech. Conf., vol. 2, Orlando, FL, Oct.
2003, pp. 126872.
[164] T. Roman, M. Enescu, and V. Koivunen, Recursive Estimation of Time-Varying Channel and Frequency Offset in MIMO
OFDM Systems, Proc. IEEE Intl. Symp. Personal, Indoor and
Mobile Radio Commun., vol. 2, Beijing, China, Sept. 2003, pp.
193438.
[165] M. Sternad and D. Aronsson, Channel Estimation and Prediction for Adaptive OFDM Downlinks, Proc. IEEE Vehic. Tech.
Conf., vol. 2, Orlando, FL, Oct. 2003, pp. 128387.
[166] T. Roman, M. Enescu, and V. Koivunen, Time-Domain
Method for Tracking Dispersive Channels in MIMO OFDM Systems, Proc. IEEE Intl. Conf. Acoust., Speech, and Signal Processing, vol. 4, Hong Kong, China, Apr. 2003, pp. 39396.
[167] T. Fukuhara et al., A Novel Channel Estimation Method for
OFDM Transmission Technique Under Fast Time-Variant Fading
Channel, Proc. IEEE Vehic. Tech. Conf., vol. 4, Jeju, Korea,
Apr. 2003, pp. 234347.
[168] A. Chini, Y. W. M. El-Tanany, and S. Mahmoud, Filtered
Decision Feedback Channel Estimation for OFDM-Based Dtv
Terrestrial Broadcasting System, IEEE Trans. Broadcast., vol.
44, no. 1, Mar. 1998, pp. 211.
[169] X. Ma, H. Kobayashi, and S. C. Schwartz, EM-Based Channel Estimation for OFDM, Proc. IEEE Pacific Rim Conf. Commun., Computers, and Signal Processing, vol. 2, Victoria,
Canada, Aug. 2001, pp. 44952.
[170] H. K. Song et al., Frequency-Offset Synchronization and
Channel Estimation for OFDM-Based Transmission, IEEE Trans.
Broadcast., vol. 4, no. 3, pp. 957, Mar. 2000.
[171] H. Sari, G. Karam, and I. Jeanclaud, FrequencyDomain
Equalization of Mobile Radio and Terrestrial Broadcast Channels, Proc. IEEE Globecom Conf., vol. 1, San Francisco, CA,
Dec. 1994, pp. 15.
[172] Y. Xie and C. N. Georghiades, An EMBased Channel Estimation Algorithm for OFDM with Transmitter Diversity, Proc.
IEEE Globecom Conf., vol. 2, San Antonio, TX, Nov. 2001, pp.
87175.
[173] J. W. Jwa and H. S. Lee, Interleaved TC 8dpsk/OFDM with
Decision-Directed Channel Estimation on Frequency-Selective
Rayleigh Fading Channels, IEEE Commun. Lett., vol. 6, no. 10,
pp. 41315, Oct. 2002.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

[174] J. Ha et al., LDPC Coded OFDM with Alamouti/SVD Diversity Technique, Proc. Intl. Symp.Wireless Personal Multimedia
Commun., vol. 3, Aalborg, Denmark, Sept. 2001, pp. 134550.
[175] Y. Kim, K. Kim, and J. Ahn, Iterative Estimation and Decoding for an LDPCCoded OFDMA System in Uplink Environments, Proc. IEEE Intl. Conf. Commun., vol. 4, Paris, France,
June 2004, pp. 247882.
[176] K. Y. Kuan, S. Suthaharan, and A. Nallanathan, An Adaptive Channel Estimation Scheme for OFDM System with Transmitter Diversity, Proc. IEEE Globecom Conf., vol. 2, San
Francisco, CA, Dec. 2003, pp. 87377.
[177] I. S. for Local and Metropolitan Area Networks Part 16, Air
Interface for Fixed and Mobile Broadband Wireless Access Systems Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed
Bands and Corrigendum 1, IEEE Std 802.16e2005 and IEEE
Std 802.162004/Cor 12005 (Amendment and Corrigendum
to IEEE Std 802.162004) Std., 2004, available:
Http://IEEE802.Org/16
[178] L. Lin et al., Comparison of Convolutional and Turbo Codes
for OFDM with Antenna Diversity in High-Bit-Rate Wireless
Applications, IEEE Commun. Lett., vol. 4, no. 9, Sept. 2000,
pp. 27779.
[179] L. Sichao and Y. Dongfeng, Reducing Papr of OFDM with
Convolutional Code and 8-Ask Mapping, Wireless Commun.
and Mobile Comput., vol. 1, Montreal, Canada, Sept. 2005,
pp. 25356.
[180] G. Ungerboeck, Channel Coding with Multilevel/Phase Signals, IEEE Trans. Inform. Theory, vol. 28, no. 1, Jan. 1982, pp.
5567.
[181] MIT, Introduction to Lattice and Trellis Codes, 2005, MIT
Opencourseware, http://Ocw.Mit.Edu.
[182] S. Liu and J. W. Chong, Improved Design Criterion for
Space-Frequency Trellis Codes in MIMO-OFDM Systems, Proc.
IEEE Vehic. Tech. Conf., vol. 2, Los Angeles, CA, Sept. 2004,
pp. 139195.
[183] R. Roberts et al., Performance of Space-Diversity Trellis
Codes in Hiperman and 802.16a Broadband Wireless Access
Systems, Proc. IEEE Vehic. Tech. Conf., vol. 2, Milan, Italy,
May 2004, pp. 86670.
[184] C. Berrou, A. Glavieux, and P. Thitimajshima, Near Shannon
Limit ErrorCorrecting Coding: Turbo Codes, Proc. IEEE Intl.
Conf. Commun., vol. 1, Geneva, Switzerland, May 1995, pp.
106470.
[185] C. E. Shannon, A Mathematical Theory of Communications, Bell System Tech. J., vol. 27, Oct. 1948, pp. 379423
and pp. 62356.
[186] A. G. Burr and G. P. White, Performance of Turbo-Coded
OFDM, Turbo Codes in Digital Broadcasting Could It Double
Capacity? IEE Electron. Lett., no. 165, Nov. 1999, pp. 8/18/8.
[187] R. G. Gallager, Low-Density Parity-Check Codes, Ph.D. Dissertation, Massachusetts Institute of Technology, Cambridge,
MA, 1961.
[188] K. I. Ahmed, C. Tepedelenlioglu, and A. Spanias, Performance of Precoded OFDM with Channel Estimation Error,
IEEE Trans. Signal Processing, vol. 54, no. 3, Mar. 2006, pp.
116571.
[189] A. A. Hutter, E. De Carvelho, and J. M. Cioffi, On the
Impact of Channel Estimation for Multiple Antenna Diversity
Reception in Mobile OFDM Systems, Proc. Asilomar Conf. Signals, Systems and Computers, vol. 1, Pacific Grove, CA, Nov.
2000, pp. 182024.
[190] S. Takaoka, H. Gacanin, and F. Adachi, Impact of Imperfect
Channel Estimation on OFDM/Tdm Performance, Proc. IEEE
Vehic. Tech. Conf., vol. 1, Stockholm, Sweden, June 2005, pp.
44246.
[191] A. Leke and J. M. Cioffi, Impact of Imperfect Channel
Knowledge on the Performance of Multicarrier Systems, Proc.
IEEE Globecom Conf., vol. 2, Sydney, Australia, Nov. 1998, pp.
95155.
[192] P. Chen and H. Kobayashi, Maximum Likelihood Channel
Estimation and Signal Detection for OFDM Systems, Proc. IEEE
Intl. Conf. Commun., vol. 3, New York, NY, May 2002, pp.
164045.
[193] X. Zhuang and F. W. Vook, Iterative Channel Estimation

IEEE Communications Surveys & Tutorials 2nd Quarter 2007

and Decoding for a Turbo-Coded OFDM System via the EM


Algorithm, Proc. IEEE Intl. Conf. Acoust., Speech, and Signal
Processing, vol. 3, Orlando, FL, May 2002, pp. 233740.
[194] X. Ma, H. Kobayashi, and S. C. Schwartz, An EM-Based
Estimation of OFDM Signals, Proc. IEEE Wireless Commun.
and Networking Conf., vol. 1, Orlando, FL, Mar. 2002, pp.
22832.
[195] E. Jaffrot and M. Siala, Turbo Channel Estimation for
OFDM Systems on Highly Time and Frequency Selective Channels, Proc. IEEE Intl. Conf. Acoust., Speech, and Signal Processing, vol. 5, Istanbul, Turkey, June 2000, pp. 297780.
[196] Y. Xie and C. N. Georghiades, Two EM-Type Channel Estimation Algorithms for OFDM with Transmitter Diversity, IEEE
Trans. Commun., vol. 51, no. 1, Jan. 2003, pp. 10615.
[197] J. Yue et al., Channel Estimation and Data Detection for
MIMOOFDM Systems, Proc. IEEE Globecom Conf., vol. 2, San
Francisco, CA, Dec. 2003, pp. 58185.
[198] K. J. Kim and J. Yue, Joint Channel Estimation and Data
Detection Algorithms for MIMO-OFDM Systems, Proc. Asilomar Conf. Signals, Systems and Computers, vol. 2, Monterey,
CA, Nov. 2002, pp. 185761.
[199] M. Ito et al., An OFDM Receiver with DecisionDirected
Channel Estimation for the Scattered Pilot Scheme in Fast Fading Environments, Proc. IEEE Vehic. Tech. Conf., vol. 1, Jeju,
Korea, Apr. 2003, pp. 36872.
[200] M. Munster and L. Hanzo, RLS-Adaptive Parallel Interference Cancellation Assisted Decision-Directed Channel Estimation for OFDM, Proc. IEEE Wireless Commun. and Networking
Conf., vol. 1, New Orleans, LA, Mar. 2003, pp. 504.
[201] X. Ma, H. Kobayashi, and S. C. Schwartz, An EM-Based
Channel Estimation Algorithm for Space-Time and Space-Frequency Block Coded OFDM, Proc. IEEE Intl. Conf. Acoust.,
Speech, and Signal Processing, vol. 4, Hong Kong, China, Apr.
2003, pp. 38992.
[202] J. Kim et al., A Decision Directed Receiver for Alamouti
Coded OFDM Systems, Proc. IEEE Vehic. Tech. Conf., vol. 1,
Orlando, FL, Dec. 2003, pp. 66265.
[203] A. Grant, Joint Decoding and Channel Estimation for Linear
MIMO Channels, Proc. IEEE Globecom Conf., vol. 3, Chicago,
IL, Sept. 2000, pp. 100912.
[204] I. Harjula and A. Mammela, Channel Estimation Algorithm
for Space-Time Block Coded OFDM Systems, Proc. IEEE Globecom Conf., vol. 1, San Francisco, CA, Dec. 2003, pp. 14347.
[205] D. Hu and L. Yang, Time-Varying Channel Estimation Based
on Pilot Tones in OFDM Systems, Proc. Intl. Conf. Neural Networks and Signal Processing, vol. 1, Nanjing, China, Dec.
2003, pp. 70003.
[206] J. Rinne and M. Renfors, An Equalization Method for
Orthogonal Frequency Division Multiplexing Systems in Channels with Multipath Propagation, Frequency Offset and Phase
Noise, Proc. IEEE Globecom Conf., vol. 2, London, UK, Nov.
1996, pp. 144246.
[207] H. A. Mahmoud and H. Arslan, Improved Channel Estimation for Wireless OFDM Systems Under Synchronization
Errors, 2006, Submitted to IEEE Commun. Letters.
[208] T. Yucek and H. Arslan, ICI Cancellation Based Channel Estimation for OFDM Systems, Proc. IEEE Radio and Wireless
Conf., vol. 1, Boston, MA, Aug. 2003, pp. 11114.
[209] H. Arslan, Channel Frequency Response Estimation Under
the Effect of RF Impairments in OFDM Based Wireless Systems, Proc. IEEE Vehic. Tech. Conf., Montreal, Canada, Sept.
2006.
[210] J. Sun, H. M. A. Issa, and P. Qiu, Frequency and Timing
Synchronization and Channel Estimation in Preamble Based
OFDM System, Proc. IEEE Intl. Conf. Commun., vol. 2, New
York, NY, June 2002, pp. 106368.
[211] M. M. Freda, J. F. Weng, and T. LeNgoc, Joint Channel
Estimation and Synchronization for OFDM Systems, Proc. IEEE
Vehic. Tech. Conf., vol. 3, Los Angeles, CA, Sept. 2004, pp.
167377.
[212] W. G. Song and J. T. Lim, PilotSymbol Aided Channel Estimation for OFDM with Fast Fading Channels, IEEE Trans.
Broadcast., vol. 49, no. 4, Dec. 2003, pp. 398402.
[213] A. Stamoulis, S. N. Diggavi, and N. Al-Dhahir, Estimation of
Fast Fading Channels in OFDM, Proc. IEEE Wireless Commun.

47

and Networking Conf., vol. 1, Orlando, FL, Mar. 2002, pp.


46570.
[214] W. C. Jake, Microwave Mobile Communications, Piscataway,
NJ: IEEE Press, 1994.
[215] J. Cai, X. Shen, and J. W. Mark, Robust Channel Estimation
for OFDM Wireless Communication Systems An H
Approach, Proc. IEEE Vehic. Tech. Conf., vol. 4, Vancouver,
Canada, Sept. 2002, pp. 210610.
[216] S. Y. Park and C. G. Kang, Performance of PilotAssisted
Channel Estimation for OFDM System Under Time-Varying
MultiPath Rayleigh Fading with Frequency Offset Compensation, Proc. IEEE Vehic. Tech. Conf., vol. 2, Tokyo, Japan, May
2000, pp. 124549.
[217] P. K. Yoon, P. H. Kar-Ming, and N. C. Sum, Channel Estimation for Mobile OFDM System with Different Detectors
Under Time-Varying Rayleigh Fading Channel, Proc. IEEE Intl.
Conf. Commun., vol. 1, New York, NY, Nov. 2002, pp. 29498.
[218] W. T. Ng and V. K. Dubey, On Coded Pilot Based Channel
Estimation for OFDM in Very Fast Multipath Fading Channel,
Proc. IEEE Pacific Rim Conf. Commun., Computers, and Signal
Processing, vol. 2, Honolulu, HI, Dec. 2003, pp. 85963.
[219] J. Rinne, A. Hazmi, and M. Renfors, Impulse Burst Position
Detection and Channel Estimation Schemes for OFDM Systems, IEEE Trans. Consumer Electron., vol. 49, no. 3, Aug.
2003, pp. 53945.
[220] J. Rinne and A. Hazmi, Impulse Burst Position Detection
and Channel Estimation Schemes for OFDM Systems, Proc.
IEEE Intl. Conf. Consumer Electron., vol. 1, Los Angeles, CA,
June 2003, pp. 29293.
[221] A. Jeremic, T. A. Thomas, and A. Nehorai, OFDM Channel
Estimation in the Presence of Interference, Proc. IEEE Sensor
Array and Multichannel Signal Processing Wksp., vol. 1, Newport, RI, Aug. 2002, pp. 1548.
[222] A. Jeremic, T. A. Thomas, and A. Nehorai, OFDM Channel
Estimation in the Presence of Asynchronous Interference,
Proc. IEEE Intl. Conf. Acoust., Speech, and Signal Processing,
vol. 4, Hong Kong, China, Apr. 2003, pp. 68487.
[223] M. S. Kang and W. J. Song, A Robust Channel Equalizer for
OFDM Tv Receivers, IEEE Trans. Consumer Electron., vol. 44,
no. 3, Aug. 1998, pp. 112933.
[224] M. Shin et al., A New Training Symbol Structure to
Enhance the Performance of Channel Estimation for
MIMOOFDM Systems, Proc. IEEE Intl. Conf. Acoust., Speech,
and Signal Processing, vol. 4, Hong Kong, China, Apr. 2003,
pp. 397400.
[225] Y. Li and H. Wang, Channel Estimation for MIMOOFDM
Wireless Communications, Proc. IEEE Intl. Symp. Personal,
Indoor and Mobile Radio Commun., vol. 3, Beijing, China,
Sept. 2003, pp. 289195.
[226] J. X. Du and Y. Li, Channel Estimation for D-Blast OFDM
Systems, Proc. IEEE Globecom Conf., vol. 1, Taipei, Taiwan,
Nov. 2002, pp. 34347.
[227] T. L. Tung, K. Yao, and R. E. Hudson, Channel Estimation
and Adaptive Power Allocation for Performance and Capacity
Improvement of MultipleAntenna OFDM Systems, Proc. IEEE
Wksp. Signal Processing Advances, vol. 1, Taoyuan, Taiwan,
Mar. 2001, pp. 8285.
[228] A. Doufexi et al., Use of Space Time Block Codes and Spatial Multiplexing Using Tdls Channel Estimation to Enhance the
Throughput of OFDM Based Wlans, Proc. IEEE Vehic. Tech.
Conf., vol. 1, Jeju, Korea, Apr. 2003, pp. 70408.
[229] G. Auer, Channel Estimation for OFDM Systems with Multiple Transmit Antennas by Filtering in Time and Frequency,
Proc. IEEE Vehic. Tech. Conf., vol. 2, Orlando, FL, Oct. 2003,
pp. 120408.
[230] G. Auer, Channel Estimation in Two Dimensions for OFDM
Systems with Multiple Transmit Antennas, Proc. IEEE Globecom Conf., vol. 1, San Francisco, CA, Dec. 2003, pp. 32226.
[231] G. Auer, Analysis of Pilot-Symbol Aided Channel Estimation
for OFDM Systems with Multiple Transmit Antennas, Proc.
IEEE Intl. Conf. Commun., vol. 6, Paris, France, June 2004, pp.
322125.
[232] C. S. Yeh, Y. Lin, and Y. Wu, OFDM System Channel Estimation Using Time-Domain Training Sequence for Mobile
Reception of Digital Terrestrial Broadcasting, IEEE Trans.

48

Broadcast., vol. 46, no. 3, pp. 21520, Sept. 2000.


[233] G. Auer, A. Dammann, and S. Sand, Channel Estimation
for OFDM Systems with Multiple Transmit Antennas By Exploiting the Properties of the Discrete Fourier Transform, Proc.
IEEE Intl. Symp. Personal, Indoor and Mobile Radio Commun.,
vol. 2, Beijing, China, Sept. 2003, pp. 195458.
[234] S. M. Alamouti, A Simple Transmitter Diversity for Wireless
Communications, IEEE JSAC, vol. 16, no. 8, Oct. 1998, pp.
145158.
[235] S. Liang and W. Wu, Channel Estimation Based on Pilot
Subcarrier in Space-Time Block Coded OFDM System, Proc.
IEEE Intl. Conf. Commun. Technol., vol. 2, Beijing, China, Apr.
2003, pp. 911.
[236] S. B. Slimane, Channel Estimation for Hiperlan/2 with
Transmitter Diversity, Proc. IEEE Intl. Conf. Commun., vol. 1,
New York, NY, May 2002, pp. 2530.
[237] S. B. Slimane, Channel Estimation for OFDM Based Systems
with Transmitter Diversity, Proc. IEEE Intl. Conf. Systems,
Man, and Cybernetics, vol. 1, Hammamet, Tunisia, Oct. 2002,
pp. 397402.
[238] S. Kang and J. S. Lehnert, Channel Estimation for OFDM
Systems with Transmitter Diversity for a Quasi-Static Fading
Channel, Proc. IEEE Mil. Commun. Conf., vol. 1, Boston, MA,
Oct. 2003, pp. 30913.
[239] A. L. Chua and M. Motani, Improved Channel Estimation
for OFDM Based Systems with Transmitter Diversity, Proc.
Asilomar Conf. Signals, Systems and Computers, vol. 1, Monterey, CA, Nov. 2003, pp. 114246.
[240] L. Shao and S. Roy, A Rate-One Non-Orthogonal SpaceTime Coded OFDM System with Estimation for Frequency
Selective Channels, Proc. IEEE Globecom Conf., vol. 1, Taipei,
Taiwan, Nov. 2002, pp. 67680.
[241] H. Y. Liu et al., Combining Adaptive Smoothing and DecisionDirected Channel Estimation Schemes for OFDM Wlan
Systems, Proc. Intl. Symp. Circuits and Systems, vol. 2,
Bangkok, Thailand, May 2003, pp. 14952.

BIOGRAPHIES
KEMAL OZDEMIR (kemal.ozdemir@gmail.com )received the B.S. and
M.S. degrees in electrical engineering from Middle East Technical
University, Ankara, Turkey in 1996 and 1998, respectively, and the
Ph.D. degree in electrical engineering from Syracuse University,
Syracuse, NY in 2005. He has worked for Philips between 2000
and 2003 and Triverity Inc., between 2004-2005. He was a visiting
scholar at University of South Florida between 2003-2004. Currently, he is with Logus Broadband Wireless Solutions working on
fixed/mobile WiMAX base stations. His research interest are the
development of Signal Processing algorithms and their efficient
implementation on FPGA's, development of MAC algorithms, and
the signal integrity issues for the next generation wireless systems.
HUSEYIN ARSLAN [SM] (arslan@eng.usf.edu) has received his PhD.
degree in 1998 from Southern Methodist University (SMU), Dallas,
Tx. From January 1998 to August 2002, he was with the research
group of Ericsson Inc., NC, USA, where he was involved with several project related to 2G and 3G wireless cellular communication
systems. Since August 2002, he has been with the Electrical Engineering Dept. of University of South Florida. He has alsobeen
working for Anritsu Company, Morgan Hill, CA (as a visiting professor during the summers of 2005 and 2006) as a part-time consulting since August 2005. His research interests are related to
advanced Signal Processing techniques at the physical layer, with
cross-layer design for networking adaptivity and Quality of Service
(QoS) control. He is interested in many forms of wireless technologies including cellular, wireless PAN/LAN/MANs, fixed wireless
access, and specialized wirelessdata networks like wireless sensors
networks and wireless telemetry. The current research interests
are on UWB, OFDM based wireless technologies with emphasis on
WIMAX, and cognitive and software defined radio. He has served
as technical program committee member, session and symposium
organizer in several IEEE conferences. He is editorial board member for Wireless Communication and Mobile Computing journal,
and was technical program co-chair of IEEE wireless and
microwave conference 2004.

IEEE Communications Surveys & Tutorials 2nd Quarter 2007