Beruflich Dokumente
Kultur Dokumente
i=0
c
i
X
ki
, k = N/2, . . . , N/2 1 (1)
where N is the total number of OFDM subcarriers, and to normalize
the power of the output signals, the encoding weights satisfy
K1
i=0
c
2
i
= 1. (2)
Then, an inverse discrete Fourier transform (DFT) operation is applied
to {S
k
} to transform the signals into the time domain, yielding
s
n
=
1
N
N/21
k=N/2
S
k
e
j
2kn
N
, n = 0, . . . , N 1 (3)
where the sampling time is T
s
. Before being transmitted, a cyclic
prex (CP) longer than the maximum delay spread (L 1)T
s
, with
L being the length of the channel taps, is appended to the transformed
signal. Assuming perfect synchronization, the received signal at the
receiver after the removal of CP can be expressed as
y
n
=
L1
l=0
s
nl
h
(l)
(n) +w
n
, n = 0, . . . , N 1 (4)
where h
(l)
(n) is the lth tap of the channel impulse response at the nth
time instant, and w
n
is the additive white Gaussian noise (AWGN)
with variance N
0
. In this paper, we assume that identical statistics for
each channel tap and channels over different taps are uncorrelated, i.e.,
E
_
h
(l)
(t)h
(l
(t )
_
=
_
2
l
r(), l = l
0, l = l
(5)
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 2, FEBRUARY 2014 989
where ()
L1
l=0
2
l
=
2
H
, and P(f) is the power spectral density of h
(l)
(t).
Although the shape of the Doppler spectrum is not restricted in our
analysis, for conciseness, we will only consider two special Doppler
spectra of great interest in the simulations. One accounts for the
classical U-shaped spectrum (Jakes model), i.e.,
P(f) =
_
1
f
d
1
_
1
_
f
f
d
_
2
, |f| < |f
d
|
0, otherwise
(6)
and the other one is
P(f) =
1
2
[(f +f
d
) +(f f
d
)] . (7)
Notice that (7) corresponds to the scenario that the OFDM system is
only impaired by CFO.
At the receiver, by applying the DFT to the received signals in (4)
and after the removal of the cumulative phase shift common to all
subcarriers [19], the signal at the mth subcarrier can be readily derived
as (see, e.g., [20])
S
m
=S
m
H
m
(0) +
N/21
k=N/2,k=m
S
k
H
m
(mk) +W
m
m = N/2, . . . , N/2 1 (8)
where H
m
(mk) =
L1
l=0
G
l
(mk)e
j2/Nkl
, G
l
(z) =
N1
n=0
h
(l)
(n)e
j2/Nzn
, and W
m
is the DFT of the AWGN. From
(8), it is clear that the rst term is the desired signal, and the second
term is the ICI generated from other subcarriers due to the frequency
offset. Finally, the transmit bit sequence is estimated from the signals
on all subcarriers in one OFDM block by using a maximum-likelihood
sequence detector [21] in conjunction with a demodulator.
III. OPTIMAL CORRELATIVE CODING STRATEGY
Here, we derive the optimal coding weights for the bandlimited
discrete-time OFDM system. To do so, the analytical expressions for
the power of the desired signal and the ICI, after adopting CCunder the
DTSM,
1
are rst derived, followed by the derivation of the analytical
CIR and our proposed optimal CC strategy.
A. CIR Analysis
CIR is a commonly used ICI indicator that simply neglects the
AWGN [2]. To evaluate CIR, one needs to calculate the average power
of the desired signal and the ICI separately. Considering the periodical
property, without loss of generality, we assume that the target signal is
allocated at the (N/2)th subcarrier. Therefore, the average power of
the desired signal can be derived from (8) as
P
s
= E
_
S
N/2
H
N/2
(0)
2
_
2
X
2
H
(9)
where E{} denotes the expectation, sinc(x) = sin(x)/x,
f
=
fNT
s
is the normalized frequency offset with respect to the subcarrier
spacing, and = 2
_
f
d
0
P(f)sinc
2
(
f
)df.
1
For brevity, in the sequel, unless otherwise specied, the term DTSM is
tied with the assumption of a nite number of subcarriers and, similarly, the
term CTSM with the assumption of an innite number of subcarriers.
Proof: See Appendix A.
On the other hand, the average power of the ICI can be calculated as
P
ICI
=E
_
_
_
N/21
z=N/2+1
S
z
H
N/2
(N/2 z)
2
_
_
_
=P
(1)
ICI
+P
(2)
ICI
(10)
where
P
(1)
ICI
=
2
X
N1
z=1
E
_
H
N/2
(z)
2
_
2
X
2
H
(1 ) (11)
P
(2)
ICI
=
2
X
K1
k=1
K1k
i=0
c
i
c
i+k
G
k
8
2
X
2
_
_
f
d
_
0
P(f) sin
2
(
f
) df
_
_
K1
k=1
K1k
i=0
c
i
c
i+k
_
k
z=1
1
z
_
_
P
tap
(k)e
j
N k
_
k
. (12)
Here, {} represents the real part, G
k
= 2
N1
z=1+k
{{H
N/2
(z)
H
N/2
(z+k)}}, and P
tap
(k)=
L1
l=0
2
l
e
j2./Nlk
denotes the
DFT of the power delay prole (PDP), which reects the channel
correlation in the frequency domain. It is not surprising to observe that
(1) is the same as the one calculated from the CTSM in [11, Eq. (3.5)]
without CC and in [17, Eq. (9)] with CC, wherein both equations
2
X
and
2
H
have been normalized to unity.
2
Note that the approximation
in (12) is tight, provided that
2
f
1.
Proof: By analogy with the steps in Appendix A, (1) can be
readily obtained. The derivation of G
k
and the proof of (12) are
presented in Appendix B.
From (10) to (12), it can be concluded that P
(1)
ICI
is the ICI power
due to the frequency offset without CC, which is related to the power of
transmit signals, channel gains, and the shape of the Doppler spectrum,
and P
(2)
ICI
is the power reduction due to the introduction of CC, which
depends on the coding coefcients.
Consequently, given the average power of the desired signal and
the ICI in (9) and (10), respectively, the CIR with CC can be readily
derived as
CIR
CC
=
1 +
1
2
H
K1
k=1
K1k
i=0
c
i
c
i+k
G
k
. (13)
Note that, for conventional OFDM systems, the average power of the
desired signal is always given by (9), whereas the ICI power termP
(2)
ICI
in (10) vanishes. Then, the total ICI power becomes
P
ICI
2
X
2
H
(1 ) (14)
leading to the CIR in the form of
CIR
OFDM
=
1
. (15)
2
This again veries the claim in [17] that, when CC is not adopted, the
application of the result derived under the assumption of an innite number of
subcarriers directly to a scenario of a system with a nite number of subcarriers
is valid.
990 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 2, FEBRUARY 2014
B. Optimal Coding Coefcients
We aim to select a group of coding coefcients from the available
coefcient set subject to (2) such that the CIR achieves its maximum in
(13). Clearly, this is equivalent to search for the optimal coding weights
in minimizing P
(2)
ICI
since P
s
and P
(1)
ICI
do not depend on the coding
weights. To this end, let us rst dene
Q
k
=
1
k
_
k
z=1
1
z
_
_
P
tap
(k)e
j
N
k
_
(16)
and rewrite P
(2)
ICI
in a matrix representation as
P
(2)
ICI
=
8
2
X
2
H
2
f
d
_
0
P(f) sin
2
(
f
) df c
T
K
R
K
c
K
(17)
where c
K
= [c
0
, c
1
, . . . , c
K1
]
T
, and
R
K
=
1
2
2
H
_
_
0 Q
1
Q
K1
Q
1
0 Q
K2
.
.
.
.
.
.
.
.
.
.
.
.
Q
K1
Q
K2
0
_
_
. (18)
Then, after applying the theorem about the RayleighRitz ratio, it is
easy to see that
P
(2)
ICI
P
(2)
ICI,min
=
8
2
X
2
H
K, min
2
f
d
_
0
P(f) sin
2
(
f
) df (19)
where
K, min
is the smallest eigenvalue of R
K
.
3
Since R
K
is a
symmetric matrix with all diagonal entries being zeros, we always have
K, min
0, and the maximum CIR
CC
becomes
CIR
CC, max
=
1 +
8
K, min
2
_
f
d
0
P(f) sin
2
(
f
) df
(20)
provided that the optimal coding vector c
K, opt
is chosen to be the
normalized eigenvector corresponding to the eigenvalue
K, min
.
As a nal remark, if
f
1, then it follows that sin
2
(
f
)
(
f
)
2
, and as a consequence, (20) can be approximated as
CIR
CC, max
1 + 4
K, min
2
f
d
(21)
where is dened as
=
2
f
2
d
f
d
_
0
f
2
P(f) df (22)
which equals to 1/2 for Jakes model and is 1 in the presence of only
CFO [17].
So far, it has been clear that the optimal coding coefcient sequence
c
K, opt
is in fact the normalized eigenvector corresponding to the
smallest eigenvalue of R
K
, whose entries, excluding the diagonal
entries, are calculated from (16), which depends on the shape of
the PDP only. Since, for both the AWGN channel and the at fast
3
Comparing the derivations of optimal coding coefcients in this paper with
those in [18], one can easily see that the main difference lies in the calculation
of (18) and [18, Eq. (11)], in which the former needs about K multiplications,
whereas the latter needs about N multiplications to formulate each nonzero
entry of the matrix. Moreover, the effect of the frequency offset has been
separated in this paper, whereas it was not separated in [18]. Since K N,
our proposal is much more efcient.
fading channel, it follows that P
tap
(k) = 1 for all k, from the given
analysis, we can see that c
K, opt
is unique. For a doubly selective
fading channel, since it usually holds that L N, P
tap
(k) can be
approximated as a constant for a small K, which is always assumed
for decoding complex considerations. These imply that the optimal
coding coefcients derived from the AWGN channel or the at fast
fading channel with small K are also applicable to a doubly selective
fading channel.
By comparing the denominator of (21) with [11, Eq. (10)] and [11,
Eq. (18)], it is again not surprising to observe a similar mathematical
expression for the reduced ICI power as we assume the same ICI
introduction mechanism and ICI reduction approach. However, we
note that the main difference for the adoption of either the CTSM
or the DTSM lies in the computation of the minimum eigenvalue
of R
K
, whose nonzero entries under the CTSM are similar to
[11, Eq. (24)], whereas under the DTSM, it becomes (16). One can
readily conclude by comparing these two equations that they share
nearly the same computational complexity. However, intuitively, we
will expect a performance gap between two proposals for practical
use since a small perturbation in the entries of a matrix may lead to a
nonignorable change in the eigenvalues of the matrix. In the following,
we will examine the exact inuence of this change in terms of CIR
performance and see whether the optimal coding coefcients derived
fromthe CTSMalso maximize the CIRof bandlimited OFDM systems
under Nyquist sampling.
IV. SIMULATIONS AND DISCUSSIONS
To validate our analysis, here, simulations are performed in which
2
X
=
2
H
= 1, and N = 1024.
A. Comparison Between Two Coding Coefcient Sequences
Table I lists the optimal coding coefcients and analytical maximum
coding gains for different coding lengths under the CTSM and the
DTSM over the AWGN channel, respectively. For the CTSM, these
two parameters are, respectively, denoted c
K, opt
and
K, min
, which
are calculated as in [11], whereas for the DTSM, these parameters
are denoted c
K, opt
and
K, min
, which are calculated by (18)(20).
The column c
T
K, opt
R
K
c
K, opt
refers to the real achievable coding
gain in a bandlimited OFDM system by adopting the optimal coding
strategy based on the CTSM, where one should note that R
K
is
calculated by using (18). From Table I, we observe that the optimal
coding coefcients c
K, opt
derived from the DTSM differ from c
K, opt
in the CTSM when K > 2. Note that, when K = 1, CC is essentially
absent, and when K = 2, the equality between two coding coefcient
sequence is due to the symmetry property of R
K
. We observe that the
coding gain
K, min
, which is calculated from the CTSM in [11],
highly overestimates the real achievable gain c
T
K, opt
R
K
c
K, opt
,
particularly for a larger coding length. For example, the coding gain
in practice is limited to 0.5822, whereas it is estimated as 0.8225
in [11]. From Table I, it is expected that the real achievable coding
gain c
T
K, opt
R
K
c
K, opt
is smaller than the maximum coding gain
K, min
derived from the DTSM. Note that, as the coding length
grows, the gap between c
T
K, opt
R
K
c
K, opt
and
K, min
shrinks,
leading to a nearly equal coding gain for a sufciently large coding
length (e.g., see the case for innite coding length).
B. Performance Comparison Between Two Coding Strategies
Fig. 2 shows the improved CIRs of OFDM systems impaired by the
CFO of value
f
d
= 3% versus different coding lengths over AWGN
channels. The real CIR performance, when employing the coding
strategy proposed in [11] to a bandlimited OFDM system and the
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 2, FEBRUARY 2014 991
TABLE I
COMPARISON OF THE TWO OPTIMAL CODING SEQUENCES AND TWO CODING GAINS FOR DIFFERENT CODING LENGTHS UNDER THE
CTSM AND UNDER THE DTSM, RESPECTIVELY
Fig. 2. CIR performance of OFDM systems with CC of different coding
lengths when CFO of value
f
d
= 3%exists over AWGN channels.
optimal one analyzed in (21), are also compared. For reference, the
analytical CIRs based on the CTSM in [11] are given in the subgraph.
As shown in Fig. 2, except for the case of K = 2, the analytical CIRs in
[11] always overestimate their real achievable values in a bandlimited
OFDM system, and this discrepancy becomes much more serious for
larger K. Moreover, it is clear that the coding strategy, which optimizes
the CIR in the CTSM, in fact provides a suboptimal solution in the
DTSM. For example, there exists a performance gap of about 0.2 dB
between the suboptimal solution and the proposed optimal solution
when K = 3. However, since the approximation in the calculation of
the nonzero entries of R
K
in the CTSM is randomized, uctuations
on the performance gaps for different coding lengths are observed.
Finally, the close agreement between the analytical results and sim-
ulation results validates our derivations.
We remark that the aforementioned performance gap is also ob-
served for other CFO and Doppler effects, as shown in Fig. 3. As
for the case in which OFDM systems are incurred by the Doppler
shift of the U-shaped spectrum over at fast fading channels, a similar
phenomenon to a CFO case is encountered, with the exception of a
higher CIR for a specic
f
d
= f
d
NT
s
. This CIR difference can be
easily accounted for the Doppler spectrum adopted in (6) and (7),
where the U-shaped spectrum can be viewed as a sum of several
single frequency offsets less than
f
d
with a normalized energy. AS
shown in Fig. 3, the suboptimal coding strategy in [11] may result
in a nonnegligible performance penalty compared with the optimal
coding strategy when
f
d
becomes large in a bandlimited OFDM
system. For example, a gap of a nearly 1% achievable CIR exists
for OFDM systems by adopting the suboptimal coding strategy when
the CFO of value
f
d
= 5% is considered. This is, as analyzed in
Section III, because the approximation of the R
K
entries according to
the assumption of an innite number of subcarriers leads to a coding
gain difference with the optimal one derived in this paper.
Consider a doubly selective fading channel, in which the Doppler
spectrum follows the Jakes model and where the PDP is expo-
Fig. 3. CIR performance of OFDM systems with length-three CC in the
presence of either CFO or the Doppler shift of the U-shaped spectrum.
Fig. 4. CIR performance of OFDM systems with CC of different coding
lengths over doubly selective fading channels with the Doppler shift of the
U-shaped spectrum and the exponentially decaying PDP.
nentially decaying according to
2
l
=
1
C
e
Dl/L
, where the factor
C =
L1
l=0
e
Dl/L
is chosen to normalize the PDP as
2
H
=
L1
l=0
2
l
= 1, and the parameter D governs the decay rate. Note that,
when D = 0, the PDP reduces to a uniform type. Fig. 4 shows the
performance comparison results in terms of CIR between the two pro-
posals derived under the CTSM and the DTSM in practical use versus
different coding lengths when L = 64 and
f
d
= 5%. As expected,
under a frequency-selective fading channel, performance gains of our
proposed coding strategy over the previous one in practical use are
also observed. It is shown that, for a normalized power, a uniform PDP
results in a worse CIR performance than an exponential decaying one.
However, we again note that the shape of the PDP scarcely affects the
optimal coding coefcients. One can readily verify that the optimal
992 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 2, FEBRUARY 2014
coding coefcients of both proposals for a doubly selective fading
channel are the same as their counterparts derived from the AWGN
or the at fast fading channel, some of which are listed in Table I.
V. CONCLUSION
The frequency-domain CC coefcients, which are optimal in the
sense of CIR maximization, have been derived for discrete-time
OFDM systems. Simulation results revealed the performance gap
between the existing coding strategy, which is optimal in continuous-
time OFDM systems under an innite-bandwidth assumption, and our
proposal for a practical communication system.
APPENDIX G
PROOF OF THE AVERAGE POWER OF THE DESIRED SIGNAL
By resorting to (2) and the independence among the transmit
symbols, we have
P
s
=
2
X
K1
i=0
K1
k=0
c
i
c
k
E
_
X
N/2i
X
N/2k
_
E
_
H
N/2
(0)
2
_
=
2
X
E
_
H
N/2
(0)
2
_
. (23)
From the denition in (5), the expectation in the last equation in (23)
can be calculated as
E
_
H
N/2
(0)
2
_
=
1
N
2
L1
l=0
L1
=0
E{G
l
(0)G
l
(0)} e
j(ll
)
=
1
N
2
L1
l=0
L1
=0
N1
n=0
N1
=0
E
_
h
(l)
(n)h
(l
)
(n
)
_
e
j(ll
)
=
1
N
2
L1
l=0
2
l
N1
n=0
N1
=0
r ((n n
)T
s
)
=
2
H
N
2
f
d
_
f
d
P(f)
N1
n=0
N1
=0
e
j2f(nn
)T
s
df
= 2
f
d
_
0
P(f)
sin
2
(
f
)
N
2
sin
2
_
f
N
_df . (24)
The substitution of (24) into (23) yields (9).
APPENDIX H
DERIVATION OF G
k
AND PROOF OF (12)
We can readily obtain G
0
2
2
H
(1 ) from the denition of G
k
and the result derived in (11). For k = 0, by using the identity
1
N
N1
n=0
e
j
2mn
N
=
_
1, m = 0
0, m = 0
(25)
one can show that
G
k
=
2
N
2
_
N1
z=1+k
L1
l=0
2
l
e
j
2
N
lk
N1
n=0
N1
=0
r (n n
)T
s
)
e
j2z(nn
)
N
e
j
2n
k
N
_
=
2
N
2
_
P
tap
(k)
N1
z=0
N1
n=0
N1
=0
r ((n n
)T
s
)
e
j
2z(nn
)
N
e
j
2n
k
N
_
2
N
2
_
P
tap
(k)
k
z=0
N1
n=0
N1
=0
r ((n n
)T
s
)
e
j
2z(nn
)
N
e
j
2n
k
N
_
= 2
f
d
_
f
d
P(f) sin
2
(
f
)
z=0
_
P
tap
(k)e
j
N k
_
N
2
sin
_
N
(
f
+z)
_
sin
_
N
(
f
+z k)
_ df.
(26)
Since k < K, it follows that
G
k
2
f
d
_
f
d
P(f)
sin
2
(
f
)
2
k
z=0
1
(
f
+z)(
f
+z k)
df
_
P
tap
(k)e
j
N
k
_
=4
f
d
_
f
d
P(f)
sin
2
(
f
)
k
2
k
z=0
z
z
2
2
f
df
_
P
tap
(k)e
j
N
k
_
. (27)
When
2
f
1, we can further approximate G
k
as
G
k
8
2
f
d
_
0
P(f) sin
2
(
f
) df
1
k
_
k
z=1
1
z
_
_
P
tap
(k)e
j
N
k
_
. (28)
which, as a substitution into the rst equation in (12), makes the second
equation true.
REFERENCES
[1] Y. Zhao and S. G. Hggman, Intercarrier interference self-cancellation
scheme for OFDM mobile communication systems, IEEE Trans. Com-
mun., vol. 49, no. 7, pp. 11851191, Jul. 2001.
[2] M. Wen, X. Cheng, X. Wei, B. Ai, and B. Jiao, A novel effective ICI self-
cancellation method, in Proc. IEEE GLOBECOM, Houston, TX, USA,
Dec. 2011, pp. 15.
[3] M.-X. Chang, A novel algorithm of inter-subchannel interference self-
cancellation for OFDM systems, IEEE Trans. Wireless Commun., vol. 6,
no. 8, pp. 28812893, Aug. 2007.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 2, FEBRUARY 2014 993
[4] H. G. Yeh, Y. K. Chang, and B. Hassibi, A scheme for cancelling inter-
carrier interference using conjugate transmission in multicarrier commu-
nication systems, IEEE Trans. Wireless Commun., vol. 6, no. 1, pp. 37,
Jan. 2007.
[5] C. L. Wang and Y. C. Huang, Intercarrier interference cancelling using
general phase rotated conjugate transmission for OFDM systems, IEEE
Trans. Commun., vol. 58, no. 3, pp. 812819, Mar. 2010.
[6] C. L. Wang, P. C. Shen, Y. C. Lin, and J. H. Huang, An adaptive receiver
design for OFDM symstem using conjugate transmission, IEEE Trans.
Commun., vol. 61, no. 2, pp. 599608, Feb. 2013.
[7] X. Cheng, Q. Yao, M. Wen, C.-X. Wang, L. Song, and B. Jiao, Wideband
channel modeling and ICI cancellation for vehicle-to-vehicle communica-
tion systems, IEEE J. Sel. Areas Commun., vol. 31, no. 8, Aug. 2013.
[8] J. Ma, P. V. Orilik, J. Zhang, and G. Y. Li, Reduced-rate OFDM
transmission for inter-subchannel interference self-cancellation over high-
mobility fading channels, IEEE Trans. Wireless Commun., vol. 11, no. 6,
pp. 20132023, Jun. 2012.
[9] Y. Zhao and S. G. Hggman, Intercarrier interference compression in
OFDM communication systems by using correlative coding, IEEE Com-
mun. Lett., vol. 2, no. 8, pp. 214216, Aug. 1998.
[10] W. Bai and Y. Kim, Effects of correlative coding on OFDM systems for
intercarrier-interference suppression, IEEE Trans. Veh. Technol., vol. 56,
no. 6, pp. 37323738, Nov. 2007.
[11] H. Zhang and Y. G. Li, Optimum frequency-domain partial response
encoding in OFDM system, IEEE Trans. Commun., vol. 51, no. 7,
pp. 10641068, Jul. 2003.
[12] Y. Zhang and H. Liu, Frequency-domain correlative coding for MIMO-
OFDM systems over fast fading channels, IEEE Commun. Lett., vol. 10,
no. 5, pp. 347349, May 2006.
[13] J. M. Choi and J. H. Lee, Frequency-domain partial response coding
for Alamouti SFBC-OFDM system in doubly selective channels, IEICE
Trans. Commun., vol. E92-B, no. 6, pp. 22982302, Jun. 2009.
[14] K.-H. Kim and H.-M. Kim, An ICI suppression scheme based on
the correlative coding for Alamouti SFBC-OFDM system with phase
noise, IEEE Trans. Wireless Commun., vol. 10, no. 7, pp. 20232027,
Jul. 2011.
[15] C.-D. Chung, Correlatively coded OFDM, IEEE Trans. Wireless Com-
mun., vol. 5, no. 8, pp. 20442049, Aug. 2006.
[16] Y.-P. Lin and S.-M. Phoong, OFDM transmitters: Analog representation
and DFT-based implementation, IEEE Trans. Signal Process., vol. 51,
no. 9, pp. 24502453, Sep. 2003.
[17] Y. Li and L. J. Cimini, Bounds on the interchannel interference of OFDM
in time-varying impairments, IEEE Trans. Commun., vol. 49, no. 3,
pp. 401404, Mar. 2001.
[18] K.-H. Kim and H.-M. Kim, A suppression scheme of the ICI using
the general correlative coding in OFDM systems, in Proc. IEEE ICCT,
Guilin, China, Nov. 2006, pp. 14.
[19] L. Rugini and P. Banelli, BER of OFDM systems impaired by carrier
frequency offset in multipath fading channels, IEEE Trans. Wireless
Commun., vol. 4, no. 5, pp. 22792288, Sep. 2005.
[20] Y. Mosto and D. C. Cox, ICI mitigation for pilot-aided OFDM mobile
systems, IEEE Trans. Wireless Commun., vol. 4, no. 2, pp. 765774,
Mar. 2005.
[21] G. D. Forney, Jr., Maximum-likelihood sequence estimation of digital
sequences in the presence of intersymbol interference, IEEE Trans. Inf.
Theory, vol. IT-18, no. 3, pp. 363378, May 1972.
[22] X. Cheng, Q. Yao, C.-X. Wang, B. Ai, G. L. Stuber, D. Yuan, and B. Jiao,
An improved parameter computation method for a MIMO V2V Rayleigh
fading channel simulator under non-isotropic scattering environments,
IEEE Commun. Lett., vol. 17, no. 2, pp. 265268, Feb. 2013.
[23] X. Cheng, C.-X. Wang, D. Laurenson, S. Salous, and A. Vasilakos,
An adaptive geometry-based stochastic model for non-isotropic MIMO
mobile-to-mobile channels, IEEE Trans. Wireless Commun., vol. 8, no. 9,
pp. 48244835, Sep. 2009.