JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 10, NO.
2, JUNE 2008
Power Allocation for OFDM-Based Cooperative Relay
Systems
Victor K. Y. Wu, Ye (Geoffrey) Li, Marilynn P. Wylie-Green, Tony Reid, and Peter S. S. Wang
Abstract: Cooperative relays can provide spatial diversity and im-
prove performance of wireless communications. In this paper,
we study subcarrier power allocation at the relays for orthogonal
frequency division multiplexing (OFDM)-based wireless systems.
For cooperative relay with amplify-and-forward (AF) and decode-
and-forward (DF) algorithms, we investigate the impact of power
allocation to the mutual information between the source and des-
tination. From our simulation results on word-error-rate (WER)
performance, we find that the DF algorithm with power allocation
provides better performance than that of AF algorithm in a single
path relay network because the former is able to eliminate chan-
nel noise at each relay. For the multiple path relay network, how-
ever, the network structure is already resistant to noise and channel
distortion, and AF approach is a more attractive choice due to its
lower complexity.
Index Terms: Cooperative diversity, cooperative relay, power allo-
cation.
I. INTRODUCTION
Recently, relays are being exploited to improve performance
in wireless communications systems. The relays are a network
of transceiver nodes between the transmitter and receiver that fa-
cilitate the transfer of information. This type of scheme is known
as cooperation or cooperative communications in the literature
because the relay network is cooperating with the transmitter
and receiver to improve performance. One application example
of such technologies is the MIT-initiated One Laptop per Child
(OLPC) project [1], which aims to provide affordable laptops
equipped with meshed networking functionality to children in
the developing world. Since cellular and Internet connectivity
is sparse and sporadic in these regions, such laptops can coop-
erate to make the best use of available bandwidth. In this paper,
we restrict ourselves to a single path relay network and a mul-
tiple path relay network in the context of orthogonal frequency
division multiplexing (OFDM) systems with power allocation.
The authors in [2]–[4] have provided several physical layer
relay algorithms. These include amplify-and-forward (AF) and
decode-and-forward (DF). In AF, a node amplifies its received
symbols, subject to a power constraint, before forwarding them
to the next node. This algorithm is obviously with low com-
plexity. In DF, a node fully decodes the received symbols,
re-encodes them and then forwards them. In other words, this
scheme attempts to eliminate channel distortion and noise at
Manuscript received November 7, 2007.
Wu and Li are with the School of Electrical and Computer Engineering, Geor-
gia Institute of Technology, Atlanta, Georgia, USA, email: vwu3@illinois.edu,
liye@ece.gatech.edu.
Green, Reid, and Wang are with the Nokia Siemens Networks, Dallas, Texas,
USA, email:{marilynn.green, tony.reid, peter.wang}@nsn.com.
1229-2370/08/$10.00
c 2008 KICS
1
each node by means of the redundancy of error-correct coding.
The authors in [5] and [6] have investigated cooperation for
a single path of relays connected in series. The motivation for
this network structure is that broader wireless coverage can be
achieved while still maintaining a low power constraint at the
transmitter. Analog relaying and digital relaying are considered
as two possible relay algorithms. These are equivalent to the AF
and DF algorithms, respectively. Each node has a certain trans-
mit power limit. The outage probability is then minimized by
allocating power among the relay network under these power
constraints. This power allocation accounts for the channel con-
ditions in the network in order to achieve the optimal outage
probability. Simulations indicate that 2 dB of total power can
be saved for 5 relays by using optimal power allocation instead
of uniform power allocation. This is for the DF case. However,
at high signal-to-noise ratio (SNR) values, the DF and the AF
cases are almost the same.
The authors in [7] have investigated cooperation for multi-
ple paths of relays connected in parallel. In the conventional
scheme, all relays use AF scheme. This is called all-participate
AF (AP-AF). The authors also consider an algorithm where only
one relay is selected in the transmission to maximize the mutual
information. This is called selection AF (S-AF). S-AF selects
the relay which results in the maximum mutual information be-
tween transmitter and receiver. Simulations of outage probabil-
ity indicate that 5 dB of SNR can be saved for 3 relays by using
S-AF instead of AP-AF. The authors in [8] derive symbol error
probabilities for multiple paths of relays.
In this paper, we continue to investigate the series and parallel
cooperative relay networks using OFDM signals. We consider a
single path relay network and a multiple path relay network. Us-
ing the AF relay algorithm, we derive the input-output relations
and the mutual informations for both networks. Using a power
constraint at each relay, we consider two relay power allocation
schemes: Constant gain allocation and equal power allocation.
Using the DF relay algorithm, we derive input-output relations
for both networks. We also compare word-error-rates (WERs)
for the two networks using the AF and DF relay algorithms. The
rest of this paper is organized as follows. We study power allo-
cation for the single path relay network in [5], [6] and the multi-
ple path relay network in [7] in Sections II and III, respectively.
Finally, Section IV concludes the paper and provides future re-
search directions.
II. SINGLE PATH RELAY NETWORK
In this section, we consider the single path relay network. We
first study the impact of power allocation to the mutual infor-
mation for the AF and the DF relay networks, respectively, and
then present their WER performance from computer simulation.
2 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 10, NO. 2, JUNE 2008

(0)
nk r1 nk
(m)
rm+1 Now, consider the variance of wk . Using (3) and (4), we have
(m−1)
nk
(1)

(0)
hk
(m) Receiver Rwk wk = E [wk wk∗ ]
(m−1) hk
r0 hk hk rm  
m m
N0 B X  Y (i) (i) 
Transmitter = b p (6)
N j=0 i=j+1 k k
Fig. 1. Single path relay network.
∗
where E [·] is the expectation operator, (·) is the complex con-
 (i) 2
 
(i)
A. Amplify-and-Forward jugate operator for a scalar, and bk = hk  , for i = 0, · · ·, m.
Rwk wk is positive for a nonzero N0 . We define a normalized
Fig. 1 shows the single path relay network. In the figure, r0
version of the system in (5)
is the transmitter, rm+1 is the receiver, and r1 , · · ·, rm are m
relay nodes connected in series forming a single path link be- ỹk = h̃k xk + w̃k (7)
tween the transmitter and receiver. The relays perform AF re- p p
laying. We assume that OFDM with N subcarriers is used in the where ỹk = yk / Rwk wk , h̃k = hk / Rwk wk , and w̃k =
(0) (m)
p
system. hk , · · ·, hk are the complex subchannel gains at the wk / Rwk wk . The variances of w̃k and ỹk are
(0) (m)
kth subcarrier in the link, for k = 1, · · ·, N . nk , · · ·, nk are
E [w̃k w̃k∗ ] = 1 (8)
the corresponding noises, which are assumed to be mutually in-
dependent and circular symmetric complex Gaussians all with and
zero mean and variance N0 B/N , where N0 is the power spec- !
m
tral density of the underlying continuous time noise process and 1 Y (i) (i)
(0) E [ỹk ỹk∗ ] = b k pk + 1, (9)
B is the OFDM bandwidth of the system. Let pk = Ptot /N be Rwk wk i=0
the transmit power on q the kth subcarrier, where Ptot is the net
(l)
transmitter power and pk be the amplifying gain used in the respectively. The cross terms do not appear in (9) because h̃k ,
AF algorithm at the lth relay,q
for l = 1, · · ·, m. The kth receive w̃k , and xk are mutually independent. Note that the normalized
(l)
system has unit variance noise.
symbol at rl is amplified by pk before it is forwarded to the
(0)
next node. Let xk be the kth transmit symbol with zero mean A.1 Mutual Information
and unit variance, yk be the kth receive symbol at the receiver, To derive the mutual information, note that the differential en-
(l) (l) tropy of a circular symmetric complex Gaussian vector, v, with
and xk be the kth receive symbol at the lth relay. Note that xk
is also the kth transmit symbol at the lth relay. covariance matrix, K, is h (v) = log2 det (πeK) [9]. When the
Using Fig. 1, the input-output relation at the lth relay is circular symmetric complex Gaussian
is a scalar, v, the differ-
  ential entropy is h (v) = log2 πeσv2 , where σv2 is the variance
l−1 q ! l−1 l−1 q of v. Let Ik be the mutual information between the transmitter
(l) (i) (i) (0) (i) (i) (j)
Y X Y
xk = hk pk xk +  hk pk  nk . and receiver on the kth subcarrier
i=0 j=0 i=j+1
(1) Ik = h (ỹk ) − h (w̃k )
" m
! #
1 Y (i) (i)
The input-output relation at the receiver is = log2 b p +1 . (10)
Rwk wk i=0 k k
 
m q ! m m q
Y (i) (i) (0)
X Y (i) (i) (j) The total mutual information between the transmitter and re-
yk = hk pk xk +  hk pk  nk .
ceiver, I, is the sum of all Ik divided by N . That is, after sub-
i=0 j=0 i=j+1
stituting (6) into (10), we have
(2)
  
(0) Qm (i) (i)
 
Denote 1 XN b k i=1 b k p k
I= log2 1 + SNR  P Q  
m m N m m (i) (i)
b p
q q k=1 j=0 i=j+1 k k
(i) (i) (j) (i) (i)
Y Y
hk = hk pk , γk = hk pk , (3) (11)
i=0 i=j+1
where SNR = Ptot /N0 B.
and
m A.2 Relay Power Allocation
(l) (l)
X
wk = γk nk . (4) We assume that the net transmit power at the transmitter and
l=0 at each each relay is Ptot , that is,
Then, (2) can be written as N
( q 2 )
(l) (l)
X  
E  pk xk  = Ptot . (12)
yk = hk xk + wk . (5) k=1
WU et al.: POWER ALLOCATION FOR ODFM-BASED COOPERATIVE DIVERSITY RELAY SYSTEMS 3

0
At the transmitter, we assume a uniform power distribution, that 10
(0)
is, pk = Ptot /N . To derive the power constraint at each relay,
substitute (1) into (12) to arrive at
 ! 
N (l) l−1 l−1 Yl−1
X pk  (0) Y (i) (i) 1 X (i) (i)
bk b k pk + bk pk  = 1.
N SNR

WER
k=1 i=1 j=0 i=j+1 −1
10
(13)

Note that (13) is defined recursively. The power constraint for Hard, AF, CT
(l) (1) (l−1) (1)
pk depends on pk , · · ·, pk . pk is the base case in the re- Hard, AF, EQ
Hard, DF
cursion, which follows from (13), when l = 1. Soft, AF, CT
(l) Soft, AF, EQ
One power allocation at the lth relay is to set pk constant Soft, DF
−2
(l) 10
for all subcarriers. This results in moving pk in (13) out of the 0 3 6 9 12 15 18 21 24
summation because it is no longer a function of k SNR (dB)
(l) (l) (a)
pk,ct = pct = 0
10
N SNR
 !  . (14)
N l−1 l−1 Y
l−1
SNRb(0) (i) (i) (i) (i)
X Y X
k bk pct + bk pct 
k=1 i=1 j=0 i=j+1

We call this constant gain allocation (CT). This is in fact the AF

WER

−1
scheme as defined in [2], where a receive symbol is multiplied 10

by a constant gain to satisfy a power constraint. In our context,

this means that every subcarrier is multiplied by the same gain,
(l) Hard, AF, CT
pct . Note that this power allocation does not require each relay Hard, AF, EQ
to have any channel state information (CSI). That is, we do not Hard, DF
Soft, AF, CT
actually need to use (14). We can use (12) directly to solve for Soft, AF, EQ
(l) Soft, DF
pct . −2
10
(l) 0 3 6 9 12 15 18 21 24
A second power allocation is to choose pk such that every
SNR (dB)
subcarrier transmits the same power at the lth relay. That is,
q (l) (l) 2 (b)
(12) becomes E  p x  = Ptot /N , for k = 1, · · ·, N .
k,eq k
This is equivalent to setting every summand on the left hand side Fig. 2. WERs in a single path relay network with TU channels using AF
and DF for N = 128: (a) m = 2 and (b) m = 4.
of (13) to 1/N . We then have

(l) SNR (m+1)

pk,eq = l−1
! l−1 Y
l−1
. (15) zero mean and unit variance. Let yk be the kth receive
(0) (i) (i) (i) (i) (l)
symbol at the receiver and yk be the kth receive symbol at the
Y X
SNRbk bk pk,eq + bk pk,eq
i=1 j=0 i=j+1 lth relay. Using Fig. 1, the input-ouput relation at the lth relay is
r
We call this equal power allocation (EQ). Note that this power (l) (l−1) Ptot (l−1) (l−1)
allocation does require each relay to have the CSI of its upstream y k = hk x + nk . (17)
N k
channels.
The input-output relation at the receiver is
B. Decode-and-Forward r
(m+1) (m) Ptot (m) (m)
In DF, each relay fully recovers the information bits (with yk = hk x + nk . (18)
N k
possible errors) after receiving an OFDM symbol. It then con-
verts the information bits back into an OFDM symbol and then
transmits it. The transmitter and all the relays transmit with the C. Simulations
same uniform power distribution. That is,
We simulate WERs versus SNR for both the AF and DF cases.
(0) (l) Ptot At the transmitter (and at the transmitter structure of a relay
pk = pk = (16) using DF), each information word contains 83 bits. We use a
N
rate 1/3 convolutional encoder with generator sequences [111],
for k = 1, · · ·, N and for l = 1, · · ·, m. [111], and [110] to encode the information word into a 255-bit
(0)
Let xk be the kth transmit symbol from the transmitter and codeword. A zero bit is padded at the end to make 256 bits. The
(l)
xk be the kth transmit symbol from the lth relay, all with with bits are then interleaved and modulated onto N = 128 QPSK
4 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 10, NO. 2, JUNE 2008

(0 |1)
subcarriers to form one OFDM symbol. At the receiver (and at nk
(1| m+1) (1|m+1)
the receiver structure of a relay using DF), the codeword is re- r1 hk nk
(0 |1)
covered (with possible errors) using a matched filter and dein- hk
terleaving. A Viterbi decoder is used to decode the codeword.
r0 rm+1
Both hard decisions and soft decisions are used.
We consider the m = 2 and 4 relays cases. We assume that hk
(0|m)
Transmitter Receiver
all distances between any two adjacent transceiver nodes are the rm
hk
(m| m+1) (m|m+1)
nk
(0| m)
same. Therefore, all path loss effects are normalized to 0 dB. nk
Shadowing is assumed to be log-normally distributed. That is,
the received power gain due to shadowing in dB is a zero-mean Fig. 3. Multiple path relay network.
Gaussian with variance of 8 dB, which is typical for cellular land
mobile applications [10]. We model frequency selective fading Using Fig. 3, the input-output relation for the lth path is
as Typical Urban (TU) channels [10]. We use an OFDM band-  q q 
width of 800 kHz divided into N = 128 equal blocks. Maintain- (l) (0|l) (l|m+1)
y k = hk hk pk
(0) (l)
pk xk +
ing OFDM orthogonality, this translates into an OFDM sym-
bol period of Ts = 160 µs. The simulation results are shown
q
(l|m+1) (l) (0|l) (l|m+1)
in Fig. 2. hk p k nk + nk . (19)
As shown in the plots, there are significant error rate perfor- Denote
mance gains when using DF instead of AF. The gains are even h iT
(1) (m)
larger when we increase the distance between the transmitter yk = yk ··· yk , (20)
and receiver (and thus, add more relays). The AF error rates
suffer because more channel distortion and noise enter the sys- 
(0|1) (1|m+1)
q
(0)
q
(1)

tem. The DF error rates suffer only slightly because noise and hk hk pk pk
 
channel distortion are eliminated at each relay. This results in hk =  ..
, (21)
 
. q
the large performance gains for m = 4. In terms of power al-  q 
(0|m) (m|m+1) (0) (m)
location when using AF, CT is the preferable choice since EQ hk hk pk pk
requires CSI and only results in small performance gains over
CT. As expected, soft decisions give better performance than  q 
(1|m+1) (1)
hard decisions in Viterbi decoding. 
hk pk 0 
Γk = 
 .. 
,
 .
q Im×m 

(m|m+1) (m)
III. MULTIPLE PATH RELAY NETWORK 0 hk pk
In this section, we consider a multiple path relay network, (22)
following the same structure as Section II.
 (0|1) 
nk
A. Amplify-and-Forward
 .. 

 . 

(0|m)
Fig. 3 shows the multiple path relay network. In the figure,  nk
 
nk =  (1|m+1) , (23)

r0 is the transmitter, rm+1 is the receiver, and r1 , · · ·, rm are m  nk 
relay nodes connected in parallel forming a multiple path link 
 ..


between the transmitter and receiver. The relays perform AF re-  . 
(m|m+1)
laying. We assume that OFDM with N subcarriers is used in nk
(0|1) (0|m) (1|m+1) (m|m+1)
the system. hk , · · ·, hk , hk , · · ·, hk are the and
complex subchannel gains at the kth subcarrier in the link, for
(0|1) (0|m) (1|m+1) (m|m+1) wk = Γk nk . (24)
k = 1, · · ·, N . nk , · · ·, nk , nk , · · ·, nk are
the corresponding noises, which are assumed to be mutually in- Then, (19), for all l = 1, · · ·, m, can be written as
dependent, zero-mean, circular symmetric complex Gaussians
all with variance N0 B/N , where N0 is the power spectral den- yk = hk xk + wk . (25)
sity of the underlying continuous time noise process and B is
(0) The large boldface zeros in (22) represent zero values in the off
the OFDM bandwidth of the system. Let pk = Ptot /N be the diagonal entries in the left m × m submatrix of Γk .
transmitter power on the q kth subcarrier, where Ptot is the net Now, consider the covariance of wk . Using (21) to (24), we
(l)
transmitter power and pk be the amplifying gain used in the have
AF algorithm at the lth relay,q
for l = 1, · · ·, m. The kth receive  (1|m+1) (1) 
bk pk + 1 0
(l)
symbol at rl is amplified by pk before it is forwarded to the N0 B  ..
Rwk wk =

N
 . 
next node. Let xk be the kth transmit symbol with zero mean (m|m+1) (m)
(l)
and unit variance and yk be the kth receive symbol from the lth 0 bk pk + 1
path at the receiver. (26)
WU et al.: POWER ALLOCATION FOR ODFM-BASED COOPERATIVE DIVERSITY RELAY SYSTEMS 5

(i|j)  (i|j) 2 q
where bk = hk  , for i = 0, · · ·, m, for j = 1, · · ·, m + 1, multiply its entire OFDM receive symbol by a constant, pct ,
(l)
and i 6= j. Since the diagonal entries of Rwk wk are never zero, such that the total transmit power is Ptot , similar to constant
−1 −1 −1 −1
Rwk wk
and Rwk2wk are well defined, where Rwk2wk Rwk2wk = gain allocation in Section II-A.
1 1
−1
Rw . Also, if we define Rwk wk as Rw Equal power allocation (EQ) in this case is
k wk Rwk wk = Rwk wk ,
2 2
k wk
−1 1 1
−1
then Rwk2wk Rw 2
k wk = Rwk wk Rwk wk = I. We define a normal-
2 2
(l) SNR
pk,eq = (0|l)
. (34)
ized version of the system in (25) SNRbk +1
ỹk = h̃k xk + w̃k (27) Note that this power allocation does require each relay to have
the CSI of its upstream channel.
−1 −1
where ỹk = Rwk2wk yk , h̃k = Rwk2wk hk , and w̃k =
−1
Rwk2wk wk . The covariance matrices of w̃k and ỹk are B. Decode-and-Forward
In DF, each relay fully recovers the information bits (with
E w̃k w̃kH = I
 
(28)
possible errors) after receiving an OFDM symbol. It then con-
and verts the information bits back into an OFDM symbol and then
transmits it. The transmitter and all the relays transmit with the
E ỹk ỹkH = h̃k h̃H
 
k + I, (29) same uniform power distribution. That is,

(0) (l) Ptot

respectively. The cross terms do not appear in (29) because h̃k , pk = pk = (35)
w̃k and xk are mutually independent. Note that the normalized N
system has identity covariance noise. for k = 1, · · ·, N and for l = 1, · · ·, m.
(0)
Let xk be the kth transmit symbol from the transmitter and
A.1 Mutual Information (l)
xk be the kth transmit symbol from the lth relay, all with with
(m+1)
Let Ik be the mutual information between the transmitter and zero mean and unit variance. Let yk be the kth receive sym-
receiver on the kth subcarrier (l)
bol at the receiver and yk be the kth receive symbol at the lth
relay. Using Fig. 3, the input-ouput relation at the lth relay is
Ik = h (ỹk ) − h (w̃k )
r
−1
= log2 1 + hH Ptot (0)


k Rwk wk hk . (30) (l)
y k = hk
(0|l)
x + nk .
(0|l)
(36)
N k
The total mutual information between the transmitter and re-
ceiver, I, is the sum of all Ik divided by N . That is, after sub- The input-output relation at the receiver is
stituting (21) and (26) into (30), we have m r !
(m+1)
X (i|m+1) Ptot (i) (i|m+1)
yk = hk x + nk . (37)
N k
N
" m (0|i) (i|m+1) (i)
!#
1 X X bk bk pk i=1
I= log2 1 + SNR (i|m+1) (i)
. (31)
N b p +1
k=1 i=1 k k Note that in this situation, there are m (modified in gen-
(0)
eral) copies of the original kth transmit symbol xk , namely,
(1) (m)
A.2 Relay Power Allocation xk , · · ·, xk . Therefore, the receiver has to assume (incor-
rectly in general) that all m relays perform perfect recovery of
We assume that the net transmit power at the transmitter and (l) (0)
the information bits so that xk = xk . That is, the receiver as-
at each each relay is Ptot , as in (12). At the transmitter, we sumes the kth receive symbol is
(0)
assume a uniform power distribution, that is, pk = Ptot /N . !
To derive the power constraint at each relay and thus, possible m r m
(m+1)
X (i|m+1) Ptot (0)
X (i|m+1)
power allocations, we use a derivation similar to the one in Sec- yk = hk xk + nk . (38)
i=1
N i=1
tion II-A to arrive at
N (l)   This allows the receiver to use a filter that is matched to
X pk (0|l) 1 m
bk + = 1. (32) (i|m+1)
X
N SNR hk .
k=1 i=1

CT in this case is
C. Simulations
(l) (l) N SNR
pk,ct = pct = N 
. (33) We simulate WERs versus SNR for both the AF and DF cases.
We use exactly the same configuration in Section II-C. The sim-

(0)
X
SNRbk +1
ulation results are shown in Fig. 4.
k=1
The performance gains resulting from using DF instead of
Again, this power allocation does not require each relay to have AF diminish as we add more paths between the transmitter and
any channel state information (CSI). The lth relay only has to receiver (and thus, add more relays). This is because for m = 4
6 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 10, NO. 2, JUNE 2008

0
10 rithms, such as hybrid schemes. For example, depending on the
channel conditions, a relay can AF, DF, or even just discard sub-
carrier symbols. This in turn leads to more possibilities for relay
power allocation. In this paper, we only investigate the single
path relay network and the multiple path relay network. Other
general relay networks need to be considered in the context of
WER

−1
10
OFDM as well. This will lend more insight into developing a
general theoretical framework for OFDM in cooperative relay
networks.
Hard, AF, CT
Hard, AF, EQ
Hard, DF
Soft, AF, CT
REFERENCES
Soft, AF, EQ [1] (2005) One laptop per child. [Online]. Available: http://laptop.media.mit.
Soft, DF
−2 edu/
10
0 3 6 9 12 15 18 21 24 [2] J. N. Laneman, “Cooperative diversity in wireless networks: algorithms
and architectures,” Ph.D. dissertation, MIT, Cambridge, MA, Sept. 2002.
SNR (dB) [3] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative diversity in
(a) wireless networks: Efficient protocols and outage behavior,” IEEE Trans.
0
Inf. Theory, vol. 50, pp. 3062–3080, Dec. 2004.
10 [4] J. N. Laneman, G. Wornell, and D. N. C. Tse, “An efficient protocol for
realizing cooperative diversity in wireless networks,” in Proc. IEEE ISIT
2001, June 2001, p. 294.
[5] M. O. Hasna and M. Alouini, “Performance analysis of two-hop relayed
transmission over rayleigh-fading channels,” in Proc. IEEE VTC 2002-fall,
Sept. 2002, pp. 1992–1996.
[6] ——–, “Optimal power allocation for relayed transmissions over rayleigh-
fading channels,” IEEE Trans. Wireless Commun., vol. 3, pp. 1999–2004,
WER

−1 Nov. 2004.
10
[7] Y. Zhao, R. Adve, and T. J. Lim, “Improving amplify-and-forward relay
networks: Optimal power allocation versus selection,” in Proc. IEEE ISIT
2006, July 2006, pp. 1234–1238.
Hard, AF,
hard, AF, CT
CT [8] A. Ribeiro, X. Cai, and G. B. Giannakis, “Symbol error probabilities
hard,
Hard, AF,
AF, EQ
EQ for general cooperative links,” IEEE Trans. Wireless Commun., vol. 4,
hard,
Hard, DF
DF pp. 1264–1273, May 2005.
soft,
Soft, AF,
AF, CT
CT
[9] I. E. Telatar, “Capacity of multi-antenna gaussian channels,” Eur. Trans.
soft,
Soft, AF,
AF, EQ
soft,
Soft, DF
Telecom., vol. 10, pp. 585–596, Nov. 1999.
−2
10 [10] G. L. Stüber, Principles of Mobile Communications, 2nd ed. Norwell, MA:
0 3 6 9 12 15 18 21 24 Kluwer Academic Publishers, 2001.
SNR (dB)
(b)
Victor K. Y. Wu received his B.A.Sc. in 2005, from
Fig. 4. WERs in a multiple path relay network with TU channels using the Division of Engineering Science at the University
AF and DF for N = 128: (a) m = 2 and (b) m = 4. of Toronto, Toronto and his M.S. in 2006, from the
School of Electrical and Computer Engineering at the
Georgia Institute of Technology, Atlanta. He is cur-
relays and thus, for 4 paths, the system is already resistant to rently pursuing a Ph.D. from the Department of Elec-
trical and Computer Engineering at the University of
noise and channel distortion. Therefore, DF cannot provide any Illinois at Urbana-Champaign. His research interests
more significant improvement over AF. In such a situation, AF are wireless networks and RFID systems.
might be a more attractive choice due to its lower complexity. In
terms of power allocation when using AF, CT is the preferable
choice because it gives better performance than EQ and does not
require CSI. As expected, soft decisions give better performance Ye (Geoffrey) Li received his B.S.E. and M.S.E. de-
than hard decisions in Viterbi decoding. grees in 1983 and 1986, respectively, from the De-
partment of Wireless Engineering, Nanjing Institute
of Technology, Nanjing, China, and his Ph.D. degree
in 1994 from the Department of Electrical Engineer-
IV. CONLCUSIONS ing, Auburn University, Alabama. He was a teaching
assistant and then a lecturer with Southeast Univer-
In this paper, we exploit cooperative relays by studying two sity, Nanjing, China from 1986 to 1991, a research and
subcarrier power allocation schemes as well as the system mu- teaching assistant with Auburn University, Alabama
from 1991 to 1994, and a post-doctoral research asso-
tual information in OFDM-based wireless networks. WER sim- ciate with the University of Maryland at College Park,
ulations indicate that the DF relay algorithm provides better per- Maryland from 1994 to 1996. He was with AT&T Labs—Research at Red Bank,
formance than that of AF in a single path relay network because New Jersey as a senior and then a principal technical staff member from 1996
to 2000. Since 2000, he has been with the School of Electrical and Computer
the former is able to eliminate channel noise at each relay. For Engineering at Georgia Institute of Technology as an associate and then a full
the multiple path relay network, however, the network structure professor. He is also holding a visiting chair professor position at the University
is already resistant to noise and channel distortion, and AF is of Electronic Science and Technology of China since March 2006. His general
research interests include statistical signal processing and telecommunications,
more attractive due to its lower complexity. with emphasis on OFDM and MIMO techniques, cross-layer optimization, and
Future research includes investigating additional relay algo- signal processing issues in cognitive radios. In these areas, he has published over
WU et al.: POWER ALLOCATION FOR ODFM-BASED COOPERATIVE DIVERSITY RELAY SYSTEMS
100 papers in refereed journals or conferences and filed about 20 patents. He
also has two books, entitled, Blind Equalization and Identification (co-authored
with Z. Ding, published by Mercel Dekker, Inc. in 2000) and OFDM for Wire-
less Communications (co-authored with G. Stüber, published by Springer in
2006). He is active in professional societies. He once served or is currently
serving as an editor, a member of editorial board, and a guest editor for 6 techni-
cal journals. He organized and chaired many international conferences, includ-
ing technical program vice-chair of the IEEE 2003 International Conference on
Communications. He has been awarded an IEEE fellow for his contributions to
signal processing for wireless communications in 2005.
Marilynn P. Wylie-Green received the B.S. and
M.S.E. degrees in electrical engineering from Tem-
ple University in 1989 and 1991, respectively and the
M.S.E.E. and Ph.D. degrees in electrical engineering
from the University of Pennsylvania in 1993 and 1995,
respectively. In 1995, she worked at the Wireless In-
formation Laboratory (WINLAB) of Rutgers Univer-
sity on the research topic of mobile location estima-
tion. In 1997, she joined the Nokia Research Center
where she became involved with various aspects of ad-
vanced physical layer design for several wireless sys-
tems, including GSM, W-CDMA, UWB, and WiMAX. In 2007, she transitioned
into the Research and Technology arm of Nokia Siemens Networks. She is the
author of approximately 45 international publications and 26 granted/pending
patents. She is a senior member of the IEEE.
She is a member of several national honor societies including Eta Kappa
Nu National Engineering Honor Society, Golden Key National Honor Society,
Alpha Lambda Delta National Honor Society, and she is a former recipient of
the National Science Foundation Fellowship for graduate study. Her current re-
search interests include statistical signal processing, optimized transceiver de-
sign and cross-layer optimization.
7
Anthony Reid received his B.S.E.E. (Cum Laude)
from Rensselaer Polytechnic Institute, M.S.E.E. from
Stanford University and Ph.D. from Southern Methodist
University (SMU). He has over 3 decades of indus-
try experience in the telecommunications and military
defense arenas. He has worked in the telecomm indus-
try at Bell Labs, Nortel Networks and Nokia Research
Center (NRC) and in the defense industry at Sandia
Laboratories, TI, and Raytheon Company. Most re-
cently he is an Engineering Fellow at Raytheon work-
ing in the areas of algorithms and systems for Space
and Airborne Systems. As a Principal Scientist at NRC his work covered design,
research and standardization of physical layer for wireless mobile systems (e.g.,
WiMAX, WLAN, 4G, and GSM/EDGE). He is senior member of IEEE. He has
over 20 publications and 25 granted/pending patents. His research interests are
in mobile wireless communications and also applied optimal estimation theory
and statistical signal processing.
Peter S. S. Wang received his M.S.E. and Ph.D. de-
grees in 1986 and 1991, respectively, from the Depart-
ment of Electrical Engineering Department at the Uni-
versity of Texas at Arlington, Texas. He was a visiting
assistant professor with University of Texas at Arling-
ton from 1991 to 1994. He was a staff engineer with
Motorola Research Lab. in Texas from 1994 to 1998.
He was a senior research scientist at Nokia Research
Center in Texas from 1998 to 2006. Since 2006, he has
been work with Nokia Siemens Networks in Texas.
His work covers wireless mobile location services, RF
channel modeling, and wireless communication systems. His research interests
are mobile wireless communications and diffractive optics.