Sie sind auf Seite 1von 12

702 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO.

2, FEBRUARY 2013

Decoding and Performance Bound of


Demodulate-and-Forward Based
Distributed Alamouti STBC
Ankur Bansal, Member, IEEE, Manav R. Bhatnagar, Member, IEEE, and Are Hjørungnes, Senior Member, IEEE

Abstract—In a demodulate-and-forward (DF) based coopera- the distributed nodes provide benefits of a co-located multiple
tive communication system, erroneous relaying of the data leads antenna system, such as diversity gain [1], [2].
to degradation in the performance of the destination receiver.
The relays can simply demodulate-and-forward (DF) [3]–
However, a maximum likelihood (ML) decoder in the destination
can improve the receiver performance. For achieving a diversity [5] or amplify-and-forward (AF) [6] the data of the source in
gain, the Alamouti space-time block code (STBC) can be used a symbol-wise manner. The AF protocol is a non-regenerative
in the DF based cooperative system in a distributed manner. protocol, where the relays merely scale the signals received
In this paper, we derive an ML decoder of the distributed from the source by an analog value. In the DF or regenerative
Alamouti STBC for the DF based cooperative system with two protocol, the source broadcasts uncoded 1 data to the relay and
imperfect relaying nodes. We also consider a DF cooperative
communication system in which one out of two relays is in the destination. The relay demodulates the symbols transmit-
outage. A piece-wise linear (PL) decoder for the DF cooperative ted by the source and forwards them to the destination in
system with the distributed Alamouti code and one relay in uncoded form; hence, the destination has two received replicas
outage is proposed. The PL decoder provides approximately of the transmitted data. However, the relay cannot always
the same performance as that of the ML decoder with reduced demodulate the symbols, sent by the source, perfectly and
decoding complexity. We derive the pairwise error probability
(PEP) of the proposed ML decoder with binary phase-shift it can relay erroneous data to the destination. Hence, the DF
keying constellation. An optimized transmit power allocation for protocol cannot achieve diversity in its pure form [4]. The
the relays is performed by minimizing an upper bound of the demodulate-and-forward protocol requires a low-level func-
PEP. It is shown by simulations that the proposed ML decoder tionality of symbol-wise demodulation in the relay and reduces
enables the DF protocol based cooperative system to outperform the complexity of the hardware and the energy consumption
the same rate amplify-and-forward protocol based cooperative
system when both systems utilize the distributed Alamouti STBC. at the relay; nevertheless, it can be extended to combine
with coding techniques [7], [8]. A more complicated form
of the DF protocol is the decode-and-forward protocol [9],
Index Terms—Distributed space-time block code, demodulate-
and-forward protocol, maximum likelihood decoding, pairwise where the source and relays utilize forward error correction
error probability. (FEC) coding. In the decode-and-forward protocol, the source
transmits the coded data and the relay attempts to fully decode
the source codeword; the relay re-encodes and transmits the
I. I NTRODUCTION decoded codeword to the destination upon successful decod-
ing. The relays in the decode-and-forward protocol require
OOPERATIVE communication is a potential technology
C for future wireless communication systems. It is very
useful for the nodes aspiring for better throughput, despite
more complex hardware and battery power.
It is shown in [3], [5] that the performance of the DF based
simple uncoded cooperative communication system with a
their poor links to the destination. In cooperative communica-
single pair of the source and destination node and a single
tion, a source node transmits the data to the neighboring nodes
relay can be improved by using a maximum-likelihood (ML)
(relays) which have better links to the destination and agree
decoder in the destination receiver. The ML decoder considers
to cooperate with the source node. The cooperation among
the possibility of erroneous transmission at the relay terminal
Manuscript received January 30, 2012; revised July 30, 2012; accepted
and maximizes the probability density function (p.d.f.) of
October 14, 2012. The associate editor coordinating the review of this paper the received data in the destination terminal [3]. An ML
and approving it for publication was H. Yousefiźadeh. decoder of the uncoded binary phase-shift keying (BPSK)
A. Bansal and M. R. Bhatnagar are with the Department of Electrical
Engineering, Indian Institute of Technology Delhi, Hauz Khas, New Delhi
data in a multiple antenna based DF cooperative system,
110016, India (e-mail: {ankur.bansal, manav}@ee.iitd.ac.in). M. R. Bhatnagar where the relays utilize orthogonal transmissions, is given
is the corresponding author. in [10]. In [11], an ML decoder and a low complexity
A. Hjørungnes was with UNIK - University Graduate Center, University
of Oslo, NO-2027, Kjeller, Norway (e-mail: arehj@unik.no).
piecewise linear (PL) decoder of the single antenna based
This work was supported in part by the Department of Science and DF cooperative system are derived, which require the average
Technology, Government of India under SERC scheme for the Project error probability of the source-relay links for decoding of
"Interference Cancelation in MAC Based Multiuser MIMO Communication
Systems" (Project Ref. No. SR/S3/EECE/0089/2009). The material in this
the source’s data in the destination. In [12], [13], estimate-
paper was presented in part at IEEE Vehicular Technology Conference (VTC)
Fall, Sep. 2012, Quebec City, Canada. 1 Here from uncoded we mean that without forward error correction (FEC)
Digital Object Identifier 10.1109/TWC.2012.121412.120145 coding.
1536-1276/13$31.00 
c 2013 IEEE
BANSAL et al.: DECODING AND PERFORMANCE BOUND OF DEMODULATE-AND-FORWARD BASED DISTRIBUTED ALAMOUTI STBC 703

and-forward (EF) protocol is discussed for a cooperative


system utilizing uncoded transmissions. In the EF protocol, the
relay forwards an unconstrained minimum mean-square error Rଵ
estimate that maximizes the generalized signal-to-noise ratio
(SNR) at the destination over additive white Gaussian noise
(AWGN) channel [13, Eq. (12)]. It is shown in [13, Section IV-
B and Fig. 4] that for M -point constellations, M > 2, the EF
S D

protocol tends to approach the DF protocol.


In order to achieve better diversity in the cooperative com-
munication system with multiple relays, distributed space-time
block codes (STBCs) have been proposed mainly for the AF Rଶ
protocol in the literature [14]–[18]. In such systems, each relay
transmits a single row of the STBC designed for a co-located Fig. 1. Block diagram of cooperative communication system with two relays
multiple-input multiple-output (MIMO) system. Moreover, no and a single source-destination pair.
orthogonal transmission is utilized by the relays. Therefore,
the distributed STBC based cooperative communication sys-
tem realizes a virtual MIMO system with distributed antennas. II. S YSTEM M ODEL
In [9], distributed STBC for DF protocol with relay selection
is discussed by utilizing uncoded transmissions. In order to Let us consider a cooperative system containing one source
reduce the chances of wrong relaying, the relays, having high S, one destination D, and two relays R1 and R2 as shown
SNR links to the source, are selected for transmission of the in Fig. 1. Each node contains one antenna and it can either
distributed STBC in [9]. However, it is shown in [17] that the transmit or receive the data at a time. It is assumed that
distributed STBC utilizing uncoded DF protocol with relay the channel between the source and the destination is very
selection performs much worse than the AF based distributed poor; therefore, direct transmission between the source and
STBC. the destination is not possible. The source utilizes two relaying
In this paper, our main contributions are as follows. 1) We nodes for transmission of its uncoded data to the destination.
derive an ML decoder of the distributed Alamouti STBC [19] The transmission from the source to the destination can be
in a DF based cooperative communication system with two decoupled into two orthogonal phases. In Phase I, the source
relays and M -ary constellation. 2) A low complexity PL transmits the uncoded data sequentially to both relays. The
decoder for the case when one of the two relays is in outage is relays symbol-wise demodulate the data of the source and
derived. 3) An upper bound of the pair-wise error probability transmit the Alamouti STBC in a distributed manner to the
(PEP) is derived for the proposed ML decoder of the DF destination in Phase II. In this phase, the source remains silent.
based distributed Alamouti code with BPSK constellation. The destination decodes the data by utilizing an ML decoder.
4) It is shown by simulation that the DF based distributed It is assumed that the channels of all links involved in the
Alamouti STBC with the proposed ML decoder outperforms cooperation follow the Rayleigh block fading model [20] and
the distributed Alamouti code with AF protocol. stay constant for a block of at least two consecutive time
The main differences between the present paper and [11] intervals.
are as follows. In [11], a simple cooperative relay network Let s1 , s2 ∈ A, where A is a complex-valued M -point
with a single relay is considered; whereas, a cooperative set- constellation, be two uncoded symbols to be transmitted by
up with two relays and no direct link is presented in this paper. the source to the destination in a block2 . The transmission of
The proposed ML decoder in this paper is applicable to a DF s = [s1 , s2 ] to the destination is performed in two decoupled
cooperative system where the data of the source reaches the phases. In Phase I, the source sequentially transmits s1 and
destination through two relays in the form of Alamouti STBC. s2 to the relays; hence, the data received in the m-th relay,
A suboptimal PL decoder for a simple DF cooperative system m ∈ {1, 2}, can be written as
is derived in [11], which closely approximates the ML decoder
in [11]; whereas, in the present paper, a low-complexity PL y m = h m s + em , (1)
decoder of the distributed Alamouti STBC is derived for the
case when one of the two relays is in outage. The conditional where y m = [ym,1 , ym,2 ] represents the 1 × 2 received data
symbol error rate (SER) of the PL decoder is derived in [11]; vector, ym,n , n = 1, 2, denotes the signal received in the
however, we find the conditional PEP of the proposed ML m-th relay in n-th time interval, hm denotes the channel
decoder with BPSK constellation. gain of the link between the source and the m-th relay, and
The rest of this paper is organized as follows: In Section II, em = [em,1 , em,2 ] is the 1 × 2 complex-valued additive white
the system model is introduced. The derivation of the ML Gaussian noise vector with each element having zero mean and
decoder and an upper bound of the PEP of the DF based dis- N0 variance. It is assumed that the channel hm is a circular
tributed Alamouti STBC is given in Section III. In Section IV, 2
complex Gaussian random variable with zero mean and σm
the ML and PL decoders of the distributed Alamouti code variance, and remains constant over transmission of s1 and s2 .
with a single relay in outage are derived. Simulation results
are discussed in Section V. Section VI concludes the article. 2 Since each block is independent of other because of the block fading
This article contains two appendices. model, we skip the notation for a block in the article.
704 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 2, FEBRUARY 2013

The Alamouti STBC is given as [19] arbitrary complex-valued M -point constellation (A) is given
  as
s1 −s∗2 ⎛ ⎞
X= . (2) 1 (1 − 2 )a1 + (1 − 1 )2 a2
s2 s∗1
⎜ +1 2 a3 + (1 − 1 )(1 − 2 )a4 ⎟
In a co-located antenna system with two transmit antennas, Λdp,q = ln ⎜
⎝ 1 (1 − 2 )b1 + (1 − 1 )2 b2 ⎠

each antenna transmits a distinct row of (2) in two consecutive +1 2 b3 + (1 − 1 )(1 − 2 )b4
time intervals. However, in the DF cooperative system with ⎛ ⎞
2 (1 − 1 )c1 + (1 − 2 )1 c2
two relays, both relays act as distributed spatial dimensions;
⎜ +1 2 c3 + (1 − 1 )(1 − 2 )c4 ⎟
hence, they can transmit distinct rows of Alamouti STBC in +ln ⎜ ⎟
⎝ 2 (1 − 1 )d1 + (1 − 2 )1 d2 ⎠ , (6)
two consecutive time intervals. Therefore, in Phase II, the
relays demodulate the symbols transmitted by the source by +1 2 d3 + (1 − 1 )(1 − 2 )d4
using an ML demodulator and transmit the Alamouti STBC
where 1 and 2 denote the uncoded instantaneous probability
in two consecutive time intervals by using the the estimates
of errors in decoding a symbol belonging to A in R1 and
of s1 and s2 . It is further assumed that the transmissions from
R2 , respectively; p = [p1 , p2 ], q = [q1 , q2 ], p1 , p2 , q1 , q2 =
the relays are perfectly synchronized.
1, 2, ..., M , p = q;
Let x̂n,m be the estimated symbol by the m-th relay in
the n-th time interval, then we can write the data received at

M
− N1 |y1 −f1 xl −f2 xp2 |
2
the destination in the two consecutive time-intervals, due to a1 = (M − 1) e 1 ,
transmission of Alamouti code from the relays, as l=1,
l=p1

y1 = f1 x̂1,1 + f2 x̂2,2 + z1 ,

M 2
e− N1 |y1 −f1 xp1 −f2 xl | ,
1
a2 = (M − 1)
y2 = −f1 x̂∗2,1 + f2 x̂∗1,2 + z2 , (3) l=1,
l=p2

where fm is the channel gain of the m-th relay-destination

M
1 2

link and zm denotes the complex-valued zero mean AWGN a3 = e− N1 |y1 −f1 xl −f2 xk | ,
l=1, k=1,
noise with N1 variance. It is assumed that the channel fm is l=p1 k=p2

circular complex Gaussian with zero mean and Ω2m variance, 2


a4 = (M − 1)2 e− N1 |y1 −f1 xp1 −f2 xp2 | ,
1
(7)
and remains constant over the transmission period of one
distributed Alamouti block. We can rewrite (3) as
 

M
x̂1,1 −x̂∗2,1 c1 = (M − 1) e− N1 |y2 +f1 xp2 −f2 xl | ,
1 ∗ ∗ 2

y d = [f1 , f2 ] + zd , (4)
x̂2,2 x̂∗1,2 l=1,
l=p1

where y d = [y1 , y2 ] represents the 1 × 2 received data vector

M 2
e− N1 |y2 +f1 xl −f2 xp1 | ,
1 ∗ ∗
and z d = [z1 , z2 ] denotes the 1 × 2 AWGN noise vector. c2 = (M − 1)
From (4), we can observe that the distributed Alamouti STBC, l=1,
l=p2
transmitted by both relays, will be

M
− N1 |y2 +f1 x∗ ∗ 2
k −f2 xl |
  c3 = e 1 ,
x̂1,1 −x̂∗2,1
S= . (5) l=1, k=1,
x̂2,2 x̂∗1,2 l=p1 k=p2
2
− N1 |y2 +f1 x∗
p2 −f2 xp1 |

c4 = (M − 1)2 e 1 ; (8)
It can be seen from (5), that when x̂n,i = x̂n,j , where
n, i, j ∈ {1, 2}, and i = j, then SS H is not necessarily and bi and di , i = 1, 2, 3, 4, can be obtained by substituting
proportional to the identity matrix. Therefore, S is not neces- {q1 , q2 } in place of {p1 , p2 } in (7) and (8), respectively; and
sarily an orthogonal STBC when the relays commit error in xp1 , xp2 , xq1 , xq2 ∈ A.
demodulation of the symbols transmitted by the source.
Proof: Refer Appendix A for the proof.
The proposed LLR decoder, given in (6), is used to decide
III. ML D ECODER OF THE D ISTRIBUTED A LAMOUTI between two symbol vectors xp = [xp1 , xp2 ] and xq =
xp
STBC AND ITS P ERFORMANCE B OUND [xq1 , xq2 ], xp = xq , as follows: Λdp,q ≷ 0. The proposed
xq
We assume that the relays utilize an ML decoder for decoder can be applied to all possible dissimilar vector pairs
demodulation of the symbols transmitted by the source. The containing the symbols belonging to the M -point constella-
ML demodulator of an arbitrary constellation over co-located tion, A, for a final decision of the transmitted symbol vector.
MIMO links is well known in the literature [21]. In this Remark 1: It can be seen from (6) that the ML decoding
section, we will explain the ML decoding of the distributed of the two symbols transmitted by the source is performed
Alamouti code in the destination node. jointly. Since the distributed Alamouti STBC S, in DF based
Theorem 1: For the DF based cooperative relay network cooperative system is not necessarily an orthogonal design
with no direct link and two relays, a log likelihood ratio (LLR) under the estimation errors in relays, decoupled decoding of
based ML decoder of the distributed Alamouti STBC using an the symbols is not possible.
BANSAL et al.: DECODING AND PERFORMANCE BOUND OF DEMODULATE-AND-FORWARD BASED DISTRIBUTED ALAMOUTI STBC 705

TABLE I
Further, it can be noticed from (6) that for decoding the data VALUES OF P R (El |h1 , h2 ).
transmitted by the source, the destination requires the knowl-
edge of the uncoded instantaneous error probabilities of R1 El x̂1,1 x̂2,1 x̂1,2 x̂2,2 Pr (El |h1 , h2 )
and R2 . Since the instantaneous probability of error depends E1 xq1 xq2 xq1 xq2 (1 − 1 )2 (1 − 2 )2
upon the instantaneous channel coefficients [22], in practice, E2 x̄q1 xq2 xq1 xq2 1 (1 − 1 )(1 − 2 )2
the values of h1 and h2 can be forwarded by the relays to E3 xq1 x̄q2 xq1 xq2 1 (1 − 1 )(1 − 2 )2
the destination for calculation of the values of 1 and 2 . In E4 xq1 xq2 x̄q1 xq2 2 (1 − 1 )2 (1 − 2 )
order to provide the channel state information (CSI) of the E5 xq1 xq2 xq1 x̄q2 2 (1 − 1 )2 (1 − 2 )
source-relay and the relay-destination links in the destination, E6 x̄q1 x̄q2 xq1 xq2 21 (1 − 2 )2
we can use a three phase training protocol as follows. In the E7 xq1 x̄q2 x̄q1 xq2 1 2 (1 − 1 )(1 − 2 )
first phase, the source transmits training data to the relays E8 xq1 xq2 x̄q1 x̄q2 22 (1 − 1 )2
such that the relays estimate the source-relay channels. In the E9 x̄q1 xq2 xq1 x̄q2 1 2 (1 − 1 )(1 − 2 )
second phase, the destination estimates the relay-destination E10 x̄q1 xq2 x̄q1 xq2 1 2 (1 − 1 )(1 − 2 )
channels by using the training data transmitted by the relays. E11 xq1 x̄q2 xq1 x̄q2 1 2 (1 − 1 )(1 − 2 )
The acquired CSI of the source-relay links is forwarded by E12 x̄q1 x̄q2 x̄q1 xq2 21 2 (1 − 2 )
the relays to the destination over some reliable and secure E13 xq1 x̄q2 x̄q1 x̄q2 1 22 (1 − 1 )
channels in the third phase. In Subsection V-A, we will discuss E14 x̄q1 xq2 x̄q1 x̄q2 1 22 (1 − 1 )
effect of imperfect values of 1 and 2 over the performance E15 x̄q1 x̄q2 xq1 x̄q2 21 2 (1 − 2 )
of the proposed ML decoder. E16 x̄q1 x̄q2 x̄q1 x̄q2 21 22
Remark 2: Let us define γ̄1  P0 σ12 /N0 , γ̄2  P0 σ22 /N0 ,
γ̄3  P1 Ω21 /N1 , and γ̄4  P2 Ω22 /N1 as the average SNRs
of S-R1 , S-R2 , R1 -D, and R2 -D links, respectively, where P0
decoder of the distributed Alamouti code is given by
and Pm are the average transmitted powers of the source
and the m-th relay, respectively. If the channel between the Pr{xq → xp |s = xq } =
source and the relays is very good, i.e., γ̄1 , γ̄2 → +∞, such 16 ⎛ ⎞

that 1 , 2 → 0+ , then after some algebra we get a simplified
μ +λ −μ −λ 
Eh1,h2,f1,f2 Q⎝
p p q q ⎠

decoder from (6) as l=1 σp2 +δp2 +σq2 +δq2 +2σp σq +2δp δq 
El

Λdp,q= 2Re{(y1∗ f1 +y2 f2∗ )(xp1−xq1)+(y1∗ f2 −y2 f1∗ )(xp2−xq2)}.
× Pr {El |h1, h2 } , (12)
(9)

The decoder of (9) is equivalent to the ML decoder of the where xp = xq , s denotes the transmitted symbol vector,
Alamouti code in a co-located antenna system [19, Eq. (13)] Pr{·} represents the probability, Eh1 ,h2 ,f1 ,f2 [·] denotes the
which provides decoupled decoding of the transmitted sym- expectation over h1 , h2 , f1 , and f2 ; and Q (·) |El denotes q-
bols. function [21, Eq. (2.3.10)] evaluated at El , where El , l =
1, 2, ..., 16 is the l-th mutually exclusive event depending
For BPSK (i.e., M = 2), the ML decoder of (6) gets
upon the correct and erroneous decoding by both relays. For
simplified into
example, the event E1 will occur if both relays decode s1
and s2 correctly, and E2 corresponds to an event when R1
Ap Bp
Λdp,q = ln + ln , (10) commits error in decoding s1 but it decodes s2 correct, and
Aq Bq R2 decodes s1 and s2 correctly. All such events are listed
in Table I. The probability Pr {El |h1 , h2 } is the conditional
where
probability of occurrence of the event El given that the channel
2 gains h1 , h2 are perfectly known. The terms μ , λ , σ2 , and δ2 ,
Ap = 1 (1 − 2 )e− N1 |y1 −f1 x̄p1 −f2 xp2 | + (1 − 1 )2
1

 ∈ {p, q} are given in (33).


2 2
× e N1 | 1 1 p1 2 p2 | +   e N1 | 1 1 p1 2 p2 |
− 1 y −f x −f x̄ − 1 y −f x̄ −f x̄ Proof: Refer Appendix B for the proof.
1 2
2 It is difficult to analytically find the average PEP from (12),
− N1 |y1 −f1 xp1 −f2 xp2 |
+ (1 − 1 )(1 − 2 )e 1 , therefore, we can numerically calculate the average PEP by
2
Bp = 2 (1 − 1 )e N1 | 2 1 p2 2 p1 | + (1 − 2 )1
− 1
y +f x −f x̄ using (12). From (12), an upper bound of PEP (UBPEP) for
2 2
the proposed ML decoder of distributed Alamouti code can
× e− N1 |y2 +f1 x̄p2 −f2 xp1 | +   e− N1 |y2 +f1 x̄p2 −f2 x̄p1 |
1 1
1 2 be given as
2
− N1 |y2 +f1 xp2 −f2 xp1 | Pr{xq → xp |s = xq} ≤ Pr{xq → xp |s = xq}.
+ (1 − 1 )(1 − 2 )e 1 , (11) max
{xp1,xp2,xq1,xq2}∈A4
p=q

Aq and Bq can be obtained by replacing the variables {p1 , p2 } (13)


with {q1 , q2 } in the expressions of Ap and Bp , respectively; The UBPEP given in (13) provides the probability of the
p1 , p2 , q1 , q2 = 1, 2; xp1 , xp2 , xq1 , xq2 ∈ [1, −1]; and x̄k = worst possible error, which contributes significantly in the
−xk , for k ∈ {p1 , p2 , q1 , q2 }. PEP; therefore, for finding an optimized power distribution,
Theorem 2: The uncoded average PEP of the proposed ML we minimize the UBPEP of (13).
706 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 2, FEBRUARY 2013

⎛ − N1 |y1 −f1 xq1 −f2 xp2 |


2 ⎞
20
αp,q + 1
(M−1) e
1
⎜ 2 ⎟
ε1=0 ⎜ +(1 − N1 |y1 −f1 xp1 −f2 xp2 | ⎟
15 ⎜ − 1 )e 1 ⎟
ε =10−6 Λd(1)
p,q = ln ⎜ ⎟,
1 ⎜ α + 1 − N1 |y1 −f1 xp1 −f2 xq2 |
2

10 ε =10−5 ⎝ p,q (M−1) e
1

2
− N1 |y1 −f1 xq1 −f2 xq2 |
1
ε =10−4 +(1 − 1 )e 1
1
5
ε =10 −3 (15)
1
φ(t )

ε =10−2 
1

M 2
e− N1 |y1 −f1 xl −f2 xp2 | . It is diffi-
0 1
1
where αp,q = 1
(M−1)
l=1
−5 l=p1 ,q1
cult to simplify (15) further. However, let us neglect the term
−10 αp,q in the numerator and denominator in (15) to obtain a
suboptimal decoder. It will be shown by using the quadrature
−15 phase shift keying (QPSK) signaling scheme in Fig. 8 that
neglecting these terms does not degrade the performance of
−20
−20 −15 −10 −5 0 5 10 15 20
the decoder significantly at all SNRs. Therefore, we get the
t1 following approximate LLR from (15):

1 + (M − 1)(1 − 1 )et1
Fig. 2. Plots of the function φ(t1 ) given in (16) versus the variable t1 Λp,q ≈ φ(t1 ) = ln
d(1)
, (16)
for different values of the probability of error in the relay using QPSK 1 et1 + (M − 1)(1 − 1 )
constellation.
where, for M -PSK constellation
2  ∗
TABLE II t1 = Re (y1 − f2 xp2 )f1∗ (xp1 −xq1 ) . (17)
VALUES OF T1 FOR DIFFERENT VALUES OF 1 USED IN F IG . 2. N1
 10−2 10−3 10−4 10−5 10−6 In [10], a PL combiner is derived for DF based MIMO
T1 ±5.6937 ±8.0054 ±10.3089 ±12.6115 ±14.9141 relay network using the binary phase shift keying (BPSK)
constellation. It can be seen from (16) that when 1 = 0,
φ(t1 ) = t1 ; and for very large and very small values of
t1 , φ(t1 ) is clipped to T1 = ±ln [(M − 1)(1 − 1 )/1 ]. We
IV. ML AND PL D ECODERS IN D ESTINATION WITH A have plotted φ(t1 ) for different values of 1 and t1 for QPSK
S INGLE R ELAY IN O UTAGE constellation (M = 4) in Fig. 2. The values of T1 for different
values of 1 are also listed in Table II. It can be seen from
Fig. 2 and Table II that when t1 > T1 , φ(t1 ) ≈ T1 and when
Let us assume that one of the two relays is in outage. This t1 < −T1 , φ(t1 ) ≈ −T1 . Moreover, for −T1 ≤ t1 ≤ T1 ,
scenario exists when one of the two relays is very close to φ(t1 ) ≈ t1 . Therefore, we can approximate φ(t1 ) by a PL
the source such that the channel between the source and this function as follows:
relay is very good; whereas, another relay is relatively very ⎧
⎨ −T1 , if t1 < −T1 ,
far from the source, hence, it experiences its channel from the φ(t1 ) ≈ φPL (t1 )  t1 , if −T1 ≤ t1 ≤ T1 , (18)
source in outage. If R1 is the relay in outage, then 2 = 0, ⎩
T1 , if t1 > T 1 .
and we get the following ML decoder for M -ary constellation
from (6): Following a similar procedure as stated above, we can get an
approximation for the second term on the R.H.S. of (14) for
⎛ M 2 ⎞ xp2 = xq2 and xp1 = xq1 as follows:
e− N1 |y1 −f1 xl −f2 xp2 |
1
1
1 + (M − 1)(1 − 1 )et2
l=1,
⎜ (M−1) l=p1 ⎟
⎜ − N1 |y1 −f1 xp1 −f2 xp2 |
2 ⎟ Λp,q ≈ φ(t2 ) = ln
d(2)
, (19)
⎜ +(1 − 1 )e 1 ⎟ 1 et2 + (M − 1)(1 − 1 )
Λdp,q = ln ⎜
⎜ M 2


e− N1 |y1 −f1 xl −f2 xq2 |
1
⎜ 1 ⎟ where, for M -PSK constellation, t2 =
⎝ ⎠ ∗
l=1,
(M−1) 2 ∗ ∗
l=q1
2 Re (y 2 − f 2 x )f 1 (x −x ) . We can approximate
+(1 − 1 )e − N1 |y1 −f1 xq1 −f2 xq2 | N1 p1 q2 p 2
1
φ(t2 ) by a PL function similar to (18) with t1 replaced by t2 .
⎛ M 2 ⎞
e− N1 |y2 +f1 xl −f2 xp1 |
1 ∗ ∗
1 Therefore, we get a low complexity and approximate LLR
l=1,
⎜ (M−1) l=p2 ⎟ decoder as follows:
⎜ 2 ⎟ ⎧
+(1 − 1 )e− N1 |y2 +f1 xp2 −f2 xp1 |
1 ∗ ∗
⎜ ⎟ ⎪ d(2)
+ln ⎜
⎜ M ∗ 2
⎟ . (14)
⎟ ⎨φPL (t1 ) + Λp,q , if xp1 = xq1 and xp2 = xq2 ,
⎜ − N1 |y2 +f1 x∗
l −f2 xq1 | ⎟
1
l=1, e 1 Λp,q ≈ Λp,q + φPL (t2 ), if xp1 = xq1 and xp2 = xq2 , (20)
d d(1)
⎝ (M−1) l=q2 ⎠ ⎪
⎩ d(1) d(2)
− N1 |y2 +f1 x∗
q2 −f2 xq1 |
∗ 2
Λp,q + Λp,q , if xp1 = xq1 and xp2 = xq2 .
+(1 − 1 )e 1

If p = q such that xp1 = xq1 and xp2 = xq2 , then we can V. S IMULATION R ESULTS
rewrite the first term in the right hand side (R.H.S.) of the We have considered BPSK, QPSK, and 16-QAM constel-
LLR decoder of (14) as lation and Rayleigh fading channels in the simulations.
BANSAL et al.: DECODING AND PERFORMANCE BOUND OF DEMODULATE-AND-FORWARD BASED DISTRIBUTED ALAMOUTI STBC 707

0
10
γ̄1 = γ̄2 = γ̄3 = γ̄4
−1
10
γ̄1 = γ̄2 = γ̄3 = γ̄4 −1
10

−2
10
−2
10
−3
10

SER
BER

−3
γ̄1 = γ̄2 = 10γ̄3 = 10γ̄4
−4 10
10

−5
10
γ̄1 = γ̄2 = 10γ̄3 = 10γ̄4
−4
10
−6
10

0 5 10 15 20 25 30 35 0 5 10 15 20 25
SNR [dB] SNR [dB]

Fig. 3. Performance of the proposed ML decoder , sub-optimal decoder Fig. 4. Performance of the proposed ML decoder , sub-optimal decoder
◦ [19, Eq. (13)], and distributed Alamouti code with no decoding error in the ◦ [19, Eq. (13)], distributed Alamouti code with no decoding error in the
relays ∗ under BPSK modulation. relays ∗ under QPSK modulation.

A. Performance of the Proposed ML Decoder and A Sub-


optimal Decoder of Distributed Alamouti Code
−1
We have simulated the performance of the distributed 10
Alamouti code with BPSK and QPSK constellations, and two
erroneous relays. The simulations are performed under the
following scenarios: 1) γ̄1 = γ̄2 = γ̄3 = γ̄4 , i.e., all links
have same SNR. 2) γ̄1 = γ̄2 = 10γ̄3 = 10γ̄4 , i.e., the SNR 10
−2
SER

of the S-Rm link is 10 dB higher than that of the Rm -D link. Sub−optimal decoder
We apply the proposed ML decoder and existing ML decoder MSE=100%
MSE=50%
of the co-located Alamouti code [19, Eq. (13)] for decoding
−3 MSE=40%
of the distributed Alamouti STBC in the destination node. 10 MSE=30%
The existing ML decoder of [19, Eq. (13)] is a sub-optimal MSE=20%
decoder of the distributed Alamouti STBC which assumes MSE=10%
that the relays are error free or perfectly know the data of MSE=0
−4
source. It can be seen from Figs. 3 and 4 that the proposed 10
0 5 10 15 20 25
ML decoder significantly outperforms the sub-optimal decoder SNR [dB]
of the distributed Alamouti code at all SNRs for BPSK and
QPSK constellations. For example, a gain of approximately Fig. 5. Performance of the proposed ML decoder with perfect and imperfect
5 dB is obtained at SER=10−2 by the proposed ML decoder as values of hm known in the destination.
compared to the sub-optimal decoder [19, Eq. (13)] for BPSK
and QPSK constellations when the SNRs of all links are same. have plotted SER versus SNR plots for QPSK constellation
Moreover, the proposed ML decoder provides better diversity in Fig. 5 under the following assumption: γ̄1 = γ̄2 = γ̄3 = γ̄4
than the sub-optimal decoder of the distributed Alamouti and for different values of the mean square error (MSE) in the
STBC. The performance of the distributed Alamouti STBC values of the channel gains forwarded by
the relays. The
 MSE
with no decoding error in the relays, i.e. that of the co-located  2 2
is defined in percentage as MSE  E hm − ĥm  /σm ×
Alamouti code, is also plotted in Figs. 3 and 4. It can be seen
from Figs. 3 and 4 that if we increase the SNR of the source- 100%, where ĥm denotes the estimate of hm available in the
relay links, the distributed Alamouti STBC with the proposed destination. It is assumed that the channel estimates of both
ML decoder performs close to the co-located Alamouti STBC source-relay links received by the destination from the relays
at all SNRs considered in the figures. It can be observed from lead to the same amount of MSE. Due to this estimation error,
Fig 3 that for γ̄1 = γ̄2 = γ̄3 = γ̄4 , the proposed ML decoder the values of 1 and 2 calculated by the destination will be
provides a bit error rate (BER) of 1.4×10−4 and 1.4×10−6 at erroneous. It can be seen from Fig. 5 that for small values
20 dB and 30 dB SNR, respectively. Therefore, the proposed of MSE, i.e., MSE=10%, the proposed ML decoder performs
ML decoder achieves full diversity of two. very close to the ML decoding with MSE=0. However, as
In order to demonstrate the sensitivity of the proposed ML the value of MSE increases, the performance of the proposed
decoder with respect to the error in the values of 1 and 2 ML decoder approaches the performance of the sub-optimal
(the instantaneous probability of error of S-Rm links), we decoder [19, Eq. (13)] of the distributed Alamouti STBC.
708 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 2, FEBRUARY 2013

−1
Algebraic code 10
Golden code
Alamouti code
−1
10
−2
10

−2
10

PEP
SER

−3
10

−3
10
γ̄1 = γ̄2 = 32γ̄3 = 32γ̄4
−4
10 γ̄1 = γ̄2 = 100γ̄3 = 100γ̄4
PEP of co-located Alamouti Code
−4
10

0 5 10 15 20 25 30 0 5 10 15 20
SNR [dB] SNR [dB]

Fig. 6. Comparison of the performance of the distributed Golden [23], the Fig. 7. PEP versus SNR performance of the proposed ML decoder with
distributed algebraic [24], and the proposed distributed Alamouti STBC. BPSK constellation.

Nevertheless, from low to moderate values of the MSE, i.e.,


0 ≤MSE≤ 50%, the proposed ML decoder performs better −1
10
than the sub-optimal decoder of the distributed Alamouti
STBC. For example, for MSE=50% and SER=10−2 , an SNR
−2
gain of approximately 3 dB is achieved by the proposed ML 10
decoder as compared to the sub-optimal decoder. For acquiring
the partial/imperfect channel knowledge of the source-relay −3
10
links, the destination needs less number of feed-forward bits
SER

as compared to those required for perfect CSI. This saves the


−4
useful bandwidth. 10
We have shown the performance of the distributed Golden
Proposed ML decoder with γ̄1=γ̄2 =γ̄3
code [23] and distributed algebraic STBC [24] in an uncoded −5
10 Proposed PL decoder with γ̄1=γ̄2 =γ̄3
cooperative system utilizing the DF protocol in Fig. 6. It is
assumed that γ̄1 = γ̄2 = 32γ̄3 = 32γ̄4 . Since both STBCs are Perfect relaying at both relays
−6
non-orthogonal STBCs, the destination receiver utilizes joint 10
decoding based ML decoder under the assumption that the 0 5 10 15 20 25 30 35
relays are error free. We have also shown the performance SNR [dB]
of the distributed Alamouti STBC with the proposed ML
decoder in Fig. 6. The constellations for all three schemes Fig. 8. SER versus SNR Performance of the proposed ML and PL decoders
with QPSK constellation.
are chosen to provide a data-rate of 2 bits/sec/Hz in the DF
based uncoded cooperative system with two relays. It can
be seen from Fig. 6 that the proposed ML decoder based with increasing value of the average SNR of S-Rm links.
distributed Alamouti STBC significantly outperforms the other
distributed non-orthogonal STBCs. For example, the proposed
ML decoder for SER=10−3 achieves a SNR gain of approxi-
mately 3 dB and 6 dB as compared to the distributed Golden B. Performance of the Proposed PL Decoder
code and distributed algebraic code, respectively. Moreover,
the distributed Alamouti STBC with the proposed ML decoder In Fig. 8, we have shown the performance of the proposed
achieves full diversity of two in contrast to the other distributed ML and PL decoders for the uncoded cooperative commu-
non-orthogonal STBCs. nication system with one of the two relays in outage. The
In Fig. 7, we have plotted the PEP for the proposed simulations are performed with the uncoded symbols based
ML decoder of the distributed Alamouti STBC with BPSK distributed Alamouti STBC and QPSK constellation. It is
constellation under the following scenarios: 1) γ̄1 = γ̄2 = assumed that γ̄2 → ∞, i.e., R2 has perfect knowledge of the
32γ̄3 = 32γ̄4 . 2) γ̄1 = γ̄2 = 100γ̄3 = 100γ̄4 . The value of data transmitted by the source and γ̄1 = γ̄3 = γ̄4 . It can be
the PEP at each SNR is calculated by using (12). We have seen from Fig. 8 that the proposed PL decoder works similar
also plotted the exact PEP of the co-located Alamouti STBC to the proposed ML decoder at all SNRs considered in the
in Fig. 7. It can be observed from Fig. 7 that the PEP of figure. Further, it can be seen from Fig. 8 that the proposed
the distributed Alamouti code with the proposed ML decoder ML and PL decoders achieve full diversity of two under the
gets closer to the exact PEP of co-located Alamouti STBC considered scenario.
BANSAL et al.: DECODING AND PERFORMANCE BOUND OF DEMODULATE-AND-FORWARD BASED DISTRIBUTED ALAMOUTI STBC 709

−1
10

10γ̄1 = 10γ̄2 = γ̄3 = γ̄4


−2
10 −2
BER

10

SER
γ̄1 = γ̄2 = γ̄3 = γ̄4

−3
10
Uniform ML−AF with poor S−R links
Optimized Proposed ML−DF with poor S−R links
−3 ML−AF
10 Proposed ML−DF
−4
10 ML−DF with perfect knowledge at relays
0 5 10 15 20 0 5 10 15 20 25 30
SNR [dB] SNR [dB]

Fig. 9. Performance of the proposed ML decoder with uniform and PEP Fig. 10. Comparison of the DF and AF based distributed Alamouti code
optimized power distributions, and BPSK constellation. with 16-QAM constellation and different SNR conditions of S-R links.
0
C. UBPEP Optimized Power Distribution 10

In this subsection, we find an optimized transmit power



2
distribution of both relays under a constraint C1: Pm =
m=1
2Pmax , where Pmax is the maximum power transmitted by −1
10
both relays. Another distribution of the transmit powers of
both relays is obtained under a different constraint (other
than C1) defined as C2: 0 ≤ Pm ≤ Pmax . For the above
SER

two optimization problems, we consider 20 dB SNR of the


relay-destination links and BPSK constellation. It is assumed 10
−2
ML−AF with 80% estimation error
that the S-R1 channel is poor with γ̄1 = 100.5 and SNR ML−AF with 20% estimation error
of the S-R2 link is given as follows: γ̄2 = 100γ̄ P1
3
= 100γ̄
P2 .
4
ML−AF with perfect CSI
Both optimized distributions are obtained by minimizing the Proposed ML−DF with 80% estimation error
Proposed ML−DF with 20% estimation error
UBPEP in (13). Under the constraint C1, the optimized
Proposed ML−DF with perfect CSI
distribution is P1 = 0.7Pmax and P2 = 1.3Pmax . However, the 10
−3

optimized power distribution results into uniform distribution 0 5 10 15 20 25 30


under constraint C2. Intuitively, this optimized distribution SNR [dB]

makes sense because the destination has perfect knowledge


Fig. 11. Comparison of the DF and AF based distributed Alamouti code
of the channel gain h1 , and it utilizes it in an ML sense with imperfect channel knowledge of the source-relay link in the destination
for decoding of the data. Therefore, it is not recommended utilizing 16-QAM constellation.
to reduce the power of a poor relay if this reduction cannot
be compensated by increment in the transmit power of other use half-duplex relaying nodes. We have plotted the SER
relay. It can be seen from Fig. 9 that the proposed ML decoder versus SNR performance of the proposed ML decoder (6) of
provides improvement in the performance of the distributed the DF based distributed Alamouti STBC and an existing ML
Alamouti STBC with optimized power distributions under decoder [17, Section II] of the AF based distributed Alamouti
constraint C1. STBC in Fig. 10 for 16-QAM constellation. Further, it is
assumed that both decoders have perfect knowledge of the
D. Comparison of the DF and AF Based Distributed Alamouti channel coefficients of all links involved in cooperation. It
STBC can be seen from Fig.10 that the DF protocol based distributed
We compare the performance of the uncoded transmissions Alamouti STBC with the proposed ML decoder significantly
based two cooperative communication systems with a single outperforms the AF protocol based distributed Alamouti STBC
source-destination pair and two relays for 16-QAM constella- for SNR > 5 dB. For example, a SNR gain of approximately
tion by assuming that all links are similar, i.e., having the same 5 dB is achieved at SER=10−2 by the DF based distributed
average SNR. One of the cooperative communication systems Alamouti STBC with the proposed ML decoder as compared
utilizes distributed Alamouti STBC with uncoded DF protocol to the AF based distributed Alamouti STBC of [17]. Moreover,
and the proposed ML decoder in the destination; whereas, we have compared the SER versus SNR performance of the
another cooperative system uses uncoded transmissions based AF and DF systems with S-Rm links being ten times poorer
distributed Alamouti STBC with AF protocol [17]. It is as- than the Rm -D links in Fig. 10. It can be noticed from Fig. 10
sumed that the DF and AF based distributed Alamouti STBCs that the DF based Alamouti code outperforms the AF based
710 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 2, FEBRUARY 2013

Alamouti STBC in this case as well. We have also shown the y1 and y2 are independent when hm , fm , sn , and x̂n,m are
performance of the proposed ML decoder when the relays are known for all m, n. Therefore,
error free in Fig. 10. By comparing different plots in Fig. 10,
it can be deduced that the proposed ML decoder enables the pyd |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,1 ,x̂1,2 ,x̂2,1 ,x̂2,2 }∈A4
distributed Alamouti STBC based DF system to achieve the = py1 |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,1 ,x̂2,2 }∈A2
second order diversity. It can also be noticed from Fig. 10 that ×py2 |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,2 ,x̂2,1 }∈A2 . (22)
the SER of the DF based distributed Alamouti code decays at
the same rate as regular co-located Alamouti code. Hence, Depending upon the channel quality of the S-R1 and S-
unlike distributed STBC based AF system, no log factor is R2 links, there exist the following four possibilities: 1) R1
involved in the asymptotic behavior of the proposed distributed demodulates the data erroneously, and R2 takes a correct
Alamouti code based DF cooperative system. decision. 2) R2 demodulates the data erroneously, and R1 takes
Fig. 11 shows the comparison of the distributed Alamouti a correct decision. 3) R1 and R2 both take wrong decisions.
code utilizing DF and AF protocols with imperfect channel 4) R1 and R2 both demodulate the data correctly. Considering
knowledge of the source-relay links in the destination. The these four cases, we can write
SER versus SNR performance curves are plotted for MSE of
20% and 80% in the CSI of the source-relay links. It can be py1 |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,1 ,x̂2,2 }∈A2
observed from Fig. 11 that the performance of the AF based = 1 (1−2) py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2
Alamouti code [17, Section II] is more severely affected due
to the imperfect CSI of the source-relay links as compared + (1 − 1 ) 2 py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2
to the DF based distributed Alamouti code with the proposed + 1 2 py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2
ML decoder. Moreover, it can be noticed from Fig. 11 that + (1 − 1 ) (1 − 2 ) py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 . (23)
the proposed ML decoder with imperfect CSI of the source-
relay links outperforms the AF based distributed Alamouti Similarly,
code with perfect CSI of the source-relay links, for the MSE
and SNR values considered in the figure. py2 |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,2 ,x̂2,1 }∈A2
= 1 (1 − 2 ) py2 |f1 ,f2 ,x̂1,2 =xp1 ,x̂2,1 =xp2
VI. C ONCLUSIONS + (1 − 1 ) 2 py2 |f1 ,f2 ,x̂1,2 =xp1 ,x̂2,1 =xp2
We have derived an ML decoder of the DF based dis- + 1 2 py2 |f1 ,f2 ,x̂1,2 =xp1 ,x̂2,1 =xp2
tributed Alamouti STBC in a cooperative system which works + (1 − 1 ) (1 − 2 ) py2 |f1 ,f2 ,x̂1,2 =xp1 ,x̂2,1 =xp2 . (24)
well for the complex-valued M -ary constellations. Moreover,
the proposed ML decoder enables the DF based distributed Since z1 and z2 ∼ CN (0, N1 ), we have
Alamouti STBC to significantly outperform an existing AF
based distributed Alamouti STBC. We have also derived the 1 − N1 |y1 −f1 xp1 −f2 xp2 |2
py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 = e 1 ,
conditional PEP of the proposed ML decoder applied to the πN1
distributed Alamouti STBC with BPSK constellation. 1 − N1 |y2 +f1 x∗p −f2 x∗p |2
py2 |f1 ,f2 ,x̂1,2 =xp1 ,x̂2,1 =xp2 = e 1 2 1 .
πN1
(25)
A PPENDIX A
P ROOF OF T HEOREM 1 From [26, Section III], it can be deduced that the
An ML decoder of the symbols s1 , s2 can be obtained by p.d.f.s py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 , py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 ,
maximizing the conditional joint p.d.f. of received data vector and py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 denote the p.d.f.s of a Gaussian
y d (given in (4)) in the destination over two consecutive time mixture random variable. With these observations, it follows
intervals. It is equivalent to maximize a likelihood ratio for that
decoding the data [25]. It can be shown by using the analysis

M 2
e− N1 |y1 −f1 xl −f2 xp2 | ,
1
given in [25, Section 2.3] that the destination needs to find py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 = k0
the following LLR to decide between two symbol vectors xp l=1,
l=p1
and xq :

M 2
e− N1 |y1 −f1 xp1 −f2 xl | ,
1
pyd |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,1 ,x̂1,2 ,x̂2,1 ,x̂2,2 }∈A4 py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 = k0
Λp,q = ln
d
, l=1,
pyd |f1 ,f2 ,h1 ,h2 ,s=xq ,{x̂1,1 ,x̂1,2 ,x̂2,1 ,x̂2,2 }∈A4 l=p2

(21) k0
M
M
py1 |f1 ,f2 ,x̂1,1 =xp1 ,x̂2,2 =xp2 =
where pyd |f1 ,f2 ,h1 ,h2 ,s=xp ,{x̂1,1 ,x̂1,2 ,x̂2,1 ,x̂2,2 }∈A4 is the con- (M − 1) l=1, k=1,
ditional joint p.d.f. of the received data vector y d given l=p1 k=p2

− N1 |y1 −f1 xl −f2 xk |2


that channels of the S-R1 , S-R2 , R1 -D, and R2 -D links, the ×e 1 , (26)
symbols transmitted by the source s1 , s2 , and the symbols
transmitted by two relays x̂1,1 , x̂1,2 , x̂2,1 , x̂2,2 are perfectly where k0 = 1/[πN1 (M − 1)]. Similarly, conditional p.d.f.s of
known in the destination. Since the AWGN noises z1 and z2 y2 can be obtained. Using (21)-(26) and the conditional p.d.f.s
are independent of each other, it can be observed from (3) that of y2 , we get (6).
BANSAL et al.: DECODING AND PERFORMANCE BOUND OF DEMODULATE-AND-FORWARD BASED DISTRIBUTED ALAMOUTI STBC 711

A PPENDIX B and variances of wp,i , i = 1, 2, 3, 4 are given as


P ROOF OF T HEOREM 2
1 $ 2
Let us state the following definition and results before μp,1 = ln (1 (1 − 2 )) − |f1 x̄p1 + f2 xp2 |
 N 1 %
providing a proof of Theorem  2.    ∗
− 2Re (f1 x̂1,1 + f2 x̂2,2 ) (f1 x̄p1 + f2 xp2 ) ,
Definition 1: If X ∼ N μ, σ 2 , where N μ, σ 2 denotes
1 $
the real-valued normal distributionwith mean
 μ and variance μp,2 = ln (2 (1 − 1 )) − |f1 xp1 + f2 x̄p2 |2
σ 2 , then Y = exp {X} ∼ LogN μ, σ 2 , where LogN (·, ·)  N 1 %

denotes the log-normal distribution [27, Eq. (5.30)]. − 2Re (f1 x̂1,1 + f2 x̂2,2 ) (f1 xp1 + f2 x̄p2 ) ,
 
Lemma 1: If Yi = exp {Xi } ∼ LogN μi , σi2 , i = 1 $ 2
μp,3 = ln (1 2 ) − |f1 x̄p1 + f2 x̄p2 |
1, 2, .., N , where Xi ∼ N μi , σi2 are correlated normal  N 1 %

random variables (RVs) with correlation coefficient ρi,j , i, j = − 2Re (f1 x̂1,1 + f2 x̂2,2 ) (f1 x̄p1 + f2 x̄p2 ) ,

N
1$ 2
1, 2, ..., N, i = j, then Y = Yi approximately3 follows μp,4 = ln ((1− 1 )(1− 2 ))− |f1 xp1 +f2 xp2 |
i=1
 N 1 %
the distribution
 of a log-normal distributed random variable ∗
− 2Re (f1 x̂1,1 + f2 x̂2,2 ) (f1 xp1 +f2 xp2 ) ,
Z ∼ LogN μz , σz2 , where
N 

σi2
σ2 2 2 2 2
μz = ln eμi + 2 − z , (27) 2
σp,1 = 2
|f1 x̄p1 +f2 xp2 | , σp,2 = |f1 xp1 +f2 x̄p2 | ,
i=1
2 N1 N1
2 2 2 2 2 2
and σp,3 = |f1 x̄p1 +f2 x̄p2 | , σp,4 = |f1 xp1 +f2 xp2 | ,
⎡ ⎤ N1 N1

N  2 N
−1 
N (30)
2μi +σi2
⎢ e eσi − 1 + 2 ⎥
⎢ i=1 i=1 j=i+1 ⎥ and the means and variances of vp,i , i = 1, 2, 3, 4 are given as
⎢ 2 2
μi +σi /2 μj +σj /2 ⎥
⎢ ×e e (e ρi,j σi σj
−1) ⎥
σz2 = ln ⎢
⎢ N +1 ⎥. (28)
⎥ 1 &
⎢  μ + σi2
2
⎥ λp,1 = ln (1 (1 − 2 ))− |f1 x̄p2 − f2 xp1 |2
⎢ e i 2 ⎥ N1 '
⎣ i=1 ⎦
+ 2Re {(f2∗ x̂1,2 − f1∗ x̂2,1 ) (f1 x̄p2 − f2 xp1 )} ,
1 & 2
Refer [28] for a proof.   λp,2 = ln (2 (1 − 1 ))− |f1 xp2 − f2 x̄p1 |
N1
Lemma
 2: If  Z ∼ LogN μz , σz2 , then 1/Z ∼ '
LogN −μz , σz2 . + 2Re {(f2∗ x̂1,2 − f1∗ x̂2,1 ) (f1 xp2 − f2 x̄p1 )} ,
 
Proof: Let Z = eX , where X  ∼ N μz , σz2 , then 1 & 2

1/Z = e−X = eX and X  ∼ N −μz , σz2 . Therefore, λp,3 = ln (1 2 )− |f1 x̄p2 − f2 x̄p1 |
N1 '
1/Z ∼ LogN −μz , σz2 .
+ 2Re {(f2∗ x̂1,2 − f1∗ x̂2,1 ) (f1 x̄p2 − f2 x̄p1 )} ,
 If Z1 = e  and Z22  = e , where X1 ∼
X1 X2
Lemma
 3:
2
N μz1 , σz1 and X2 ∼ N μz2 , σz2 are correlated normal 1 &
λp,4 = ln ((1 − 1 )(1 − 2 ))− |f1 xp2 −f2 xp1 |2
random variables with correlation coefficient  ρ, then Z1 /Z2 ∼ N1 '
LogN (μz1 −μz2 , σz21 + σz22 + 2ρσz1σz2 .  + 2Re{(f2∗ x̂1,2 − f1∗ x̂2,1 ) (f1 xp2 −f2 xp1 )} ,
Proof: Since 1/Z2 ∼ LogN −μz2 , σz22 , From the
properties of the sum of two correlated Normal ran-
dom variables [21], we  get Z1 /Z2 ∼ LogN (μz1 − μz2 , 2
δp,1 =
2 2 2
|f1 x̄p2 −f2 xp1 | , δp,2 =
2 2
|f1 xp2 −f2 x̄p1 | ,
σz21 + σz22 + 2ρσz1 σz2 .   N1 N1
 4: If Y =
Lemma  exp {X} ∼ LogN μ, σ 2 , then aY ∼ 2 2 2 2 2 2
LogN ln(a) + μ, σ 2 , where a is a constant. δp,3 = |f1 x̄p2 −f2 x̄p1 | , δp,4 = |f1 xp2 −f2 xp1 | , (31)
N1 N1
Proof: We can write aY = eln(a) Y = eln(a)+X =
X
e , where ln(·) denotes the natural logarithm. Since X  ∼ and the statistics (mean and variance) of wq,i and vq,i , i =
N ln(a) + μ, σ 2 , therefore, it proves the above result. 1, 2, 3, 4 can be found by substituting {xq1 , xq2 } in place of
Using (3) and Lemma 4 in (10), we get {xp1 , xp2 } in (30) and (31), respectively.
It can be observed from (29) that the random variables wp,i ,
wp,1 + wp,2 + wp,3 + wp,4 i = 1, 2, 3, 4, are fully correlated with unity correlation coeffi-
Λdp,q = ln
w + wq,2 + wq,3 + wq,4 cient. Hence, by using Lemma 1 in (29), we can approximate
q,1
vp,1 + vp,2 + vp,3 + vp,4 the LLR decoder as
+ln , (29)
vq,1 + vq,2 + vq,3 + vq,4 Wp Vp
 Λdp,q = ln + ln , (32)
where w,i ∼ 2
LogN μ,i , σ,i , and v,i ∼ Wq Vq

LogN λ,i , δ,i2
,  ∈ {p, q}, i = 1, 2, 3, 4. The means 
4 
4
where W = w,i and V = v,i ,  ∈ {p, q}. Following
i=1 i=1
3 Only matches the mean and variance of the log-normal sum and the Lemma 1, it can be observed that W and V ,  ∈ {p, q}
approximating log-normal RV. are log-normal distributed random variables. The mean and
712 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 2, FEBRUARY 2013

variance of W are given as [5] M. Ju and I.-M. Kim, “ML performance analysis of the decode-
4  and-forward protocol in cooperative diversity networks,” IEEE Trans.

2
σ,i 2
σ,i Wireless Commun., vol. 8, no. 7, pp. 3855–3867, July 2009.
μ = ln eμ,i + 2 − , [6] J. N. Laneman and G. W. Wornell, “Exploiting distributed spatial
i=1
2 diversity in wireless networks,” in Proc. 2000 Allerton Conf. Commun.,
⎡  2 3
4 ⎤ Contr., Comput., pp. 1–10.

4 2

[7] B. Zhao and M. C. Valenti, “Distributed turbo coded diversity for the
2μ,i +σ,i
⎢ e e −1 +2
σ,i
⎥ relay channel,” IEE Electron. Lett., vol. 39, no. 10, pp. 786–787, May
⎢ i=1 ⎥
⎢ i=1 j=i+1 ⎥ 2003.
⎢ ×eμ,i +μ,j + 2 σ,i + 2 σ,j(eσ,i σ,j −1)
1 2 1 2
⎥ [8] M. Janani, A. Hedayat, T. E. Hunter, and A. Norsatinia, “Coded
2 ⎢ ⎥
σ = ln ⎢ 2 +1 ⎥, cooperation in wireless communications: space-time transmission and
⎢ 4 σ2 ⎥ iterative decoding,” IEEE Trans. Signal Process., vol. 52, no. 2, pp.
⎢ μ + ,i

⎢ e ,i 2
⎥ 362–371, Feb. 2004.
⎣ i=1 ⎦ [9] J. N. Laneman and G. W. Wornell, “Distributed space-time-coded
protocols for exploiting cooperative diversity in wireless networks,”
IEEE Trans. Inf. Theory, vol. 49, no. 10, pp. 2415–2425, Oct. 2003.
(33) [10] G. V. V. Sharma, V. Ganwani, U. B. Desai, and S. N. Merchant,
“Performance analysis of maximum likelihood detection for decode
2
where are given in (30). Similarly for V ,  ∈ {p, q},
μ,i , σ,i and forward MIMO relay channels in Rayleigh fading,” IEEE Trans.
the mean (λ ) and variance (δ2 ) can be obtained by chang- Wireless Commun., vol. 9, no. 9, pp. 2880–2889, Sept. 2010.
[11] M. R. Bhatnagar and A. Hjørungnes, “Ml decoder for decode-and-
2 2
ing the parameters {μ,i , σ,i } in (33) with {λ,i , δ,i } given forward based cooperative communication system,” IEEE Trans. Wire-
in (31). Moreover, we can notice from (32) that Wp and Wq , less Commun., vol. 10, no. 12, pp. 4080–4090, Dec. 2011.
[12] K. S. Gomadam and S. A. Jafar, “On the capacity of memoryless relay
and Vp and Vq , are fully correlated RVs with unity correlation networks,” in Proc. 2006 IEEE International Conf. Commun., pp. 1580–
coefficient. Applying Lemmas 1- 3, and the above results 1585.
in (32), it can be concluded that the LLR decoder of (10) [13] K. M. Gomadam and S. A. Jafar, “Optimal relay functionality for
SNR maximization in memoryless relay networks,” IEEE J. Sel. Areas
approximately follows the distribution of a real-valued normal Commun., vol. 25, no. 2, pp. 390–401, Feb. 2007.
distributed RV Z with mean (μp + λp − μq − λq ) and variance [14] Y. Jing and B. Hassibi, “Cooperative diversity in wireless relay networks
(σp2 + δp2 + σq2 + δq2 + 2σp σq + 2δp δq ). with multiple-antenna nodes,” in Proc. 2005 Int. Symp. Inf. Theory, pp.
815–819.
Therefore, the conditional probability of error of decoding [15] ——, “Distributed space-time coding in wireless relay networks,” IEEE
xp in place of xq , such that p = q, using the LLR de- Trans. Wireless Commun., vol. 5, no. 12, pp. 3524–3536, Dec. 2006.
coder (10) for BPSK constellation will be given as [16] F. Oggier and B. Hassibi, “An algebraic family of distributed space-time
codes for wireless relay networks,” in Proc. 2006 IEEE Int. Symp. Inf.
Pr{xq → xp |h1 , h2 , f1 , f2 , s = xq ,x̂1,1 ,x̂2,1 ,x̂1,2 ,x̂2,2} Theory, pp. 538–541.
[17] Y. Jing and H. Jafarkhani, “Using orthogonal and quasi-orthogonal
= Pr {Z > 0|h1 ,h2 ,f1 ,f2 , s = xq ,x̂1,1 ,x̂2,1 ,x̂1,2 ,x̂2,2} , designs in wireless relay networks,” IEEE Trans. Inf. Theory, vol. 53,
⎛ ⎞ no. 11, pp. 4106–4118, Nov. 2007.
μ + λ − μ − λ [18] F. Oggier and B. Hassibi, “An algebraic coding scheme for wireless relay
= Q ⎝ ⎠ . (34)
p p q q
networks with multiple-antenna nodes,” IEEE Trans. Signal Process.,
σp2 + δp2 + σq2 + δq2 + 2σp σq + 2δp δq vol. 56, no. 7, pp. 2957–2966, July 2008.
[19] S. M. Alamouti, “A simple transmit diversity technique for wireless
The PEP of the ML decoder given that the channel gains of communications,” IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1451–
1458, Oct. 1998.
all involved links and the data transmitted by the source are [20] T. Marzetta and B. Hochwald, “Capacity of a mobile multiple-antenna
known in the destination will depend upon sixteen mutually communication link in Rayleigh flat fading,” IEEE Trans. Inf. Theory,
exclusive events of correct or erroneous decoding by both vol. 45, no. 1, pp. 139–157, Jan. 1999.
[21] J. G. Proakis and M. Salehi, Digital Communications, 5th edition.
relays, i.e., El , given in Table I. Hence, the PEP of the McGraw-Hill Book Company, 2008.
proposed ML decoder conditioned on El can be expressed as [22] M. K. Simon and M. S. Alouini, Digital Communication over Fading
Channels. John Wiley & Sons, Inc., 2005.
Pr {xq → xp |h1 , h2 , f1 , f2 , s = xq } = [23] J. C. Belfiore, G. Rekaya, and E. Viterbo, “The Golden code: a 2 × 2
full-rate space-time code with non-vanishing determinants,” IEEE Trans.
16

Inf. Theory, vol. 51, no. 4, pp. 1432–1436, Apr. 2005.
Pr {Z > 0|f1 , f2 , s = xq , El } Pr {El |h1 , h2 } . (35) [24] B. A. Sethuraman, B. S. Rajan, and V. Shashidhar, “Full-diversity, high-
l=1 rate space-time block codes from division algebras,” IEEE Trans. Inf.
Theory, vol. 49, no. 10, pp. 2596–2616, Oct. 2003.
The values of Pr {El |h1 , h2 } for each El are also listed in [25] H. L. V. Trees, Detection, Estimation, and Modulation Theory: Part I.
Table I. The average PEP given in (12) can be obtained by Detection, Esimation, and Linear Modulation Theory. John Willey &
Sons, Inc., 2001.
using (34) and (35). [26] L. Trailovic and L. Y. Pao, “Variance estimation and ranking of target
tracking position errors modeled using Gaussian mixture distributions,”
Automatica, vol. 41, no. 8, pp. 1433–1438, Aug. 2005.
R EFERENCES [27] A. Papoulis and S. U. Pillai, Probability, Random Variables and Stochas-
[1] A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diversity— tic Processes, 4th edition. McGraw-Hill Book Company, 2002.
part-I: system description,” IEEE Trans. Commun., vol. 51, no. 11, pp. [28] L. F. Fenton, “The sum of lognormal probability distributions in scatter
1927–1938, Nov. 2003. transmission systems,” IRE Trans. Commun. Syst., vol. CS-8, pp. 57–67,
[2] A. Nosratinia, T. E. Hunter, and A. Hedayat, “Cooperative communica- 1960.
tion in wireless networks,” IEEE Commun. Mag., vol. 42, no. 10, pp.
74–80, Oct. 2004.
[3] J. N. Laneman and G. W. Wornell, “Energy-efficient antenna sharing and
relaying for wireless networks,” in Proc. 2000 IEEE Wireless Commun.
Netw. Conf., pp. 7–12.
[4] D. Chen and J. N. Laneman, “Modulation and demodulation for coop-
erative diversity in wireless systems,” IEEE Trans. Wireless Commun.,
vol. 5, no. 7, pp. 1785–1794, July 2006.
BANSAL et al.: DECODING AND PERFORMANCE BOUND OF DEMODULATE-AND-FORWARD BASED DISTRIBUTED ALAMOUTI STBC 713

Ankur Bansal received the B.Tech. degree in Elec- Are Hjorungnes worked as a Professor at the
tronics & Communication Engineering from Uttar Faculty of Mathematics and Natural Sciences at the
Pradesh Technical University (UPTU), Lucknow, in University of Oslo, Norway with office located at
2007 and M.Tech. in Signal Processing from Netaji UNIK - University Graduate Center. He obtained
Subhas Institute of Technology (NSIT), New Delhi, his Sivilingeniør (M.Sc.) degree (with honors) in
India in 2009. He is currently pursuing the PhD de- 1995 from the Department of Telecommunications
gree from Indian Institute of Technology Delhi,New at the Norwegian Institute of Technology in Trond-
Delhi, India. His research interests include multiple- heim, Norway, and his Doktor ingeniør (Ph.D.)
input multiple-output (MIMO) systems, cooperative- degree in 2000 from the Signal Processing Group at
communications, distributed space-time block code the Norwegian University of Science and Technol-
(DSTBC) based communication systems, and mul- ogy. He authored the book Complex-Valued Matrix
tiuser communications. He worked as a lecturer in Raj Kumar Goel Institute Derivatives: With Applications in Signal Processing and Communications
of Technology, Ghaziabad, India, from 2009-2010. (Cambridge University Press, 2011).
From August 2000 to December 2000, he worked as a researcher at Tam-
Manav R Bhatnagar received the M.Tech. degree pere University of Technology, in Finland, within the Tampere International
in communications engineering from the Indian In- Center for Signal Processing. From March 2001 to July 2002, he worked
stitute of Technology Delhi, New Delhi, India, in as a postdoctoral fellow at the Federal University of Rio de Janeiro in
2005 and the Ph.D. degree from the University of Brazil, within the Signal Processing Laboratory. From September 2002 to
Oslo, Oslo, Norway, in 2008. August 2003, he worked as a postdoctoral fellow at the Helsinki University
From 2008 to 2009, he was a Post Doctoral Re- of Technology in Finland, within the Signal Processing Laboratory. From
search Fellow with UNIK-University Graduate Cen- September 2003 to August 2004, he was working as a postdoctoral fellow
ter, Kjeller, Norway. He held visiting appointments at the University of Oslo in Norway, at the Department of Informatics,
with the Wireless Research Group at the Indian within the Digital Signal Processing and Image Analysis Group. He held
Institute of Technology Delhi, India; the SPINCOM visiting appointments at the Image and Signal Processing Laboratory at the
Group of the University of Minnesota Twin Cities, University of California, Santa Barbara, the Signal Processing Laboratory
Minneapolis; the Alcatel-Lucent Chair at SUPÉLEC in France; the ECE of the Federal University of Rio de Janeiro, the Mobile Communications
Department of the Indian Institute of Science at Bangalore, Bangalore, India; Department at Eurecom Institute in France, the University of Manitoba in
UNIK - University Graduate Center of the University of Oslo, Norway; and Canada, the Alcatel-Lucent Chair at SUPÉLEC in France, the Department of
the Department of Communications and Networking of Aalto University, Fin- Electrical and Computer Engineering at the University of Houston in USA, the
land. He is currently an Assistant Professor with the Department of Electrical Electrical and Computer Engineering Department at University of California,
Engineering, Indian Institute of Technology Delhi, New Delhi, India. His San Diego, USA, and the Department of Electrical Engineering, University
research interests include signal processing for multiple-input multiple-output of Hawaii at Manoa, USA.
(MIMO) systems, cooperative-communications, non-coherent communication He served as an Editor for IEEE T RANSACTIONS ON W IRELESS C OMMU -
systems, distributed signal processing for cooperative networks, multiuser NICATIONS from March 2007 to May 2011. In 2010 and 2011, he was a Guest
communications, ultra-wideband (UWB) based communications, free space Editor for IEEE J OURNAL OF S ELECTED T OPICS IN S IGNAL P ROCESSING
optical (FSO) communication, and cognitive radio. and IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS, in the
Dr. Bhatnagar was selected as an “Exemplary Reviewer” of the IEEE C OM - special issues on “Model Order Selection in Signal Processing Systems” and
MUNICATIONS L ETTERS for the year 2010. Since August 2011, he has been “Cooperative Networking - Challenges and Applications,” respectively. He
an Editor for the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS. co-authored the papers winning the best paper awards at IEEE International
Conference on Wireless Communications, Networking and Mobile Computing
(WiCOM 2007), 7th International Symposium on Modeling and Optimization
in Mobile, Ad Hoc, and Wireless Networks (WiOpt 2009), and 5th Interna-
tional Conference on Internet Monitoring and Protection (ICIMP 2010).

Das könnte Ihnen auch gefallen