Escolar Documentos
Profissional Documentos
Cultura Documentos
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.
p p p T
where XmP = [x(m−1)Γ+1 x(m−1)Γ+2 · · · xmΓ ] , m = 1, 2, · · · , Mt , and all signal-to-noise ratio (SNR) at the destination node. εm =
fm T is
p
xk , k = 1, 2, · · · , ΓMt , are complex symbols and are mapped from an the normalized CFO where
fm is the CFO between the mth relay
information subvector S p = [s1p , s2p , · · · , sΓM p
t
]T by a rotation opera- and the destination node, and T is the OFDM symbol duration. In
tion (6), θεbm = 2πεm (bN + bLcp + Lcp )/N + θ0,m where Lcp is the length
T T T
[X1p , X2p , · · · , X Mp t ]T = ΘS p , (3) of cyclic prefix, and θ0,m is the phase rotation between the phase of
the destination node local oscillator and the carrier phase of the mth
where slp is ((p − 1)ΓMt + l − 1)th entry
of S for 1 ≤ l ≤ ΓMt and relay at the start of the received signal. Uεm is the ICI matrix induced
satisfies the energy constraint E |slp |2 = 1. Θ is an Mt Γ × Mt Γ by εm and is given by
rotation matrix.
One property of the SF code is that each subcarrier is only used Uεm = FN Ωεm FH
N, (7)
by one transmit antenna. From (2), the mth column of C, denoted as where Ωεm = diag[1, e j2πεm /N
,··· ,e j2πεm (N−1)/N
]. From the definition
Cm , can be written as of Uεm , clearly Uεm is a unitary matrix and the element at its lth row
and kth column is
Cm = Mt Pm (IP ⊗ Θ) S = Mt Pm X, (4)
sin[π((k − l)N + εm )]e jπ((k−l)N +εm )(N−1)/N
where X = (IP ⊗ Θ) S and Pm is a diagonal matrix corresponding to Uεm (l, k) = uεm ,(k−l)N = , (8)
Nsin[π((k − l)N + εm )/N]
the mth column of C. For the SF code in [5], Pm has the following
form where 0 ≤ l, k ≤ N − 1.
From (7) and (8) we can see that if εm 0, the matrix Uεm is no
1, if l = l = (t − 1)ΓMt + (m − 1)Γ + i longer diagonal and that (8) shows the effect of ICI. Furthermore, if
Pm (l, l ) = , (5)
0, else more than one CFO exists, which is possible in a cooperative com-
munication system, it is difficult for the receiver to compensate all
where 0 ≤ i ≤ Γ − 1 and 1 ≤ t ≤ P. the CFOs at the same time.
Furthermore, if the signal constellations are QAM and PAM, it
can be shown that the optimum rotation matrix Θ is a unitary ma-
5. ICI MITIGATION AND SIGNAL DETECTION IN THE
trix [5]. Based on this property, it is not difficult to verify that the
PRESENCE OF MULTIPLE CFOS
elements of X are uncorrelated.
At the receiver, each SF submatrix G p is decoded by ML or
It is shown in [7] that CFO may not decrease the diversity order of SF
sphere decoding independently and the maximum diversity order
codes and the performance degradation can be regarded as that due
ΓMt Mr can be achieved. Here Mr is the number of receive anten-
to the loss of equivalent SNR, i.e., signal-to-interference-plus-noise
nas.
ratio (SINR). Based on this observation, we propose to deal with ICI
mitigation due to multiple CFOs by first increasing the SINR , and
3. COOPERATIVE PROTOCOL then decoding the SF code by ML or sphere decoding.
Substituting (4) into (6), we get the following signal model:
The cooperative communication system we use in this work includes √
one source node, one destination node, and a number of relay nodes. Zb = ρDb Xb + Wb , (9)
Here, the decode-and-forward cooperative protocol is adopted. In Mt jθεb
where D = m=1 e m Dm Pm for Dm = Uεm Hm and the elements
b
the first phase, the source node broadcasts the information while the
of Xb are uncorrelated from one another. For (9), the linear MMSE
relays receive the same information. In the second phase, the Mt
filter can maximize the output SINR [8]. Thus we can first maximize
relays, which have detected the received information symbols cor-
the SINR by a linear filtering operation, and then decode the SF
rectly, will help the source to transmit. The detected symbols are
code by ML or sphere decoding. We call this detection method the
parsed into blocks of size N and the bth block, b = 0, 1, · · · , is en-
MMSE-F method. Note that when we decode the SF codes, we use
coded to an N × Mt SF codeword matrix C b in a distributed fash-
two approximations. The first is that the filtered Gaussian noise is
ion [1]. Here N is equal to the number of subcarriers in one OFDM
treated as if it were white. The second is that the residual ICI terms
symbol. Finally, the mth relay transmits Cmb , the mth column of C b ,
are regarded as additional Gaussian noise.
by the standard OFDM technology. At the destination node, the re-
The major drawback of the MMSE-F method is its computa-
ceived OFDM symbols are used to decode. Although in a coopera-
tional complexity since calculating MMSE filter coefficient is of the
tive communication system, each column of an SF code corresponds
order of O(N 3 ). It could be a burden when N is large. The Q taps
to one relay, to be consistent with the notation used in the SF coding
(Q-T) method is a common way to reduce computational complexity
literature we will use the term antenna in place of relay.
for the OFDM system, e.g., [6], where for the concerned subcarrier
only its L circular left and right subcarriers are considered. That
4. RECEIVE SIGNAL MODEL means we use Q (Q = 2L + 1) taps filter instead of the full N-taps
filter. Roughly speaking, if NQ3 ≤ N 3 , the computational complex-
At the destination node, after standard steps, the bth received OFDM ity can be reduced at the expense of a certain performance loss. To
symbol Zb in the frequency domain is given by partially tackle the above problem, in the remainder of this section,
we propose a simple and efficient ICI mitigation method, the M-FFT
ρ t jθεb
M
Zb = e m Uεm Hm Cmb + Wb , (6) method.
Mt m=1 First, at the transmitter, we re-arrange the SF code C in (1)-(2)
by some row permutations and denote this row-wisely permuted SF
where Hm is an N × N diagonal matrix whose diagonal elements code by C which has the following structure:
are the channel frequency response from the mth relay to destina- T
T T 1 T 2 T P T T
tion, Wb is an N × 1 vector with each entry being a zero mean unit C = diag [X11 , X12 , · · · , X1P ]T , · · · , [X M t
, XM t
, · · · , XMt
] ,
variance complex Gaussian random variable, and ρ stands for the (10)
3254
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.
where Xmp is defined in (2). From (10) we can see that after row per- Method Number of multiplications
mutations, the non-zero entries of the mth column of C are grouped.
Note that the subcarriers from PΓ(m − 1) to PΓm − 1 are used by the MMSE-F O(14N 3 )
mth transmit antenna. We call these subcarriers the mth group. As- Q-T O(13NQ3 )
sume that after permutations the ((m − 1)Γ + i)th row, 0 ≤ i ≤ Γ − 1,
p
of G p , is located at the n(m−1)Γ+i p
th row, 0 ≤ n(m−1)Γ+i ≤ N − 1, of M-FFT Mt N + (Mt − 1) N2 log2 N
p p
C , i.e., x(m−1)Γ+i+1 will be transmitted at the n(m−1)Γ+i th subcarrier
p Table 1. Comparison of computational complexity.
by the mth transmit antenna. Then, for (10) we have n(m−1)Γ+i =
(m − 1)ΓP + (p − 1)Γ + i.
In [5], it has been shown that the diversity order and coding ad- b
time domain by −εl is equivalent to multiplying Z b by (e jθεl Uεl )H .
vantage of this kind of SF codes depends only on the relative posi- Mt jθεb
tions of the permuted rows corresponding to the entries of Xmp with Let Ub = l=1 e l Uεl Pl . Then, Z̄b can be expressed as
p
respect to the position n(m−1)Γ . Since we do not change the relative √ bH b b H
positions for the entries of Xmp , i.e., n(m−1)Γ+i
p p
− n(m−1)Γ = i for each Z̄b = ρU D X + Ub Wb (14)
⎛ Mt ⎞ ⎛ Mt ⎞
pair of m and p, C should have the same diversity order and coding ⎜
√ b H ⎜⎜⎜ ⎟
⎟⎜ ⎜ ⎟
⎟
ρU ⎜⎝ e jθεl Uεl Pl ⎟⎟⎟⎠ ⎜⎜⎜⎝ Pl Hl ⎟⎟⎟⎠ X b + Ub Wb ,
b H
advantage as that of C. When we use C , we can get the following =
frequency domain SISO-OFDM model from (9) as l=1 l=1
3255
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.
demonstrate that the grouping operation of subcarriers plays an im- −2
10
portant role in the M-FFT method, we also simulate a scheme that we
directly apply multiple FS and FFT operations without the grouping
operation and it is denoted by M-FFT-WG . From Fig.1 we can see
that the performance of the M-FFT-WG method is worse than that of −3
10
the directly decoding. This is because without the grouping step, in
the FS operation, the increased ICI power from other antennas can
SER
even exceed the canceled ICI power.
MMSE−F
Fig.2 shows some extensions of the M-FFT method. For M-FFT- −4
10 M−FFT−Z4
Z4, the 0th to 29th subcarriers are used by the first transmit antenna, M−FFT−Q−T, Q=4
the 32nd to 61th subcarriers are used by the second transmit antenna M−FFT−Z4 with SIC
and all the other subcarriers are not used. We can see that the perfor- M−FFT−Q−T−Z4, Q=7
mance of the M-FFT method is greatly increased and is comparable Q−T, Q=15
with that of the Q-T method with Q = 15. The cost is that the fre- −5
10
16 17 18 19 20 21 22 23 24
quency efficiency is decreased by 6.25%. The effect of zero padding SNR(dB)
10
−4 Q−T, Q=5, εMax=0.2 [7] D. N. Dao and C. Tellambura, “Intercarrier interference self-
MMSE−F, εMax=0.2 cancellation space-frequency codes for MIMO-OFDM,” IEEE
Trans. Veh. Technol., vol. 54, pp. 1729–1738, Sep. 2005.
−5
10
4 6 8 10 12 14 16 18 20 22 24 [8] D. Tse and P. Viswanath, Fundamentals of Wireless Communi-
cation, Cambridge University Press, 2005.
[9] J. Benesty, Y. A. Huang, and J. Chen, “A fast recursive algorithm
Fig. 1. SER performance of the MMSE-F, Q-T, and M-FFT meth- for optimum sequential signal detection in a BLAST system,”
ods. IEEE Trans. Signal Processing, vol. 51, pp. 1722–1730, July
2003.
3256
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.