1

Turbo Packet Combining for Relaying Schemes

over Multi-Antenna Broadband Channels
Houda Chafnaji

, Tarik Ait-Idir

, Halim Yanikomeroglu
+
, and Samir Saoudi

Communications Systems Department, INPT, Madinat Al Irfane, Rabat, Morocco

INSTITUT TELECOM/ TELECOM Bretagne, Signal and Communications Department, CS 83818, 29238 Brest Cedex, France.
+
Broadband Communications and Wireless Systems (BCWS) Centre,
Department of Systems and Computer Engineering, Carleton University, Ottawa, Canada
Emails: houda.chafnaji,samir.saoudi@telecom-bretagne.eu, aitidir@ieee.org, halim@sce.carleton.ca
AbstractThis paper focuses on turbo packet combining
strategies for multi-relay systems operating over multiple-
inputmultiple-output (MIMO) broadband channel. The work
presented in this paper is valid for both amplify-and-forward
(AF) and decode-and-forward (DF) relaying schemes. We derive
the optimal maximum a posteriori (MAP) turbo packet combiner
that combines multiple copies of the transmitted packet received
over multiple slots, and analyze the outage performance. We also
introduce a low-complexity minimum mean square (MMSE)-
based turbo packet combining scheme and show that it offers
BLER performance close to that of the outage probability,
while it clearly outperforms conventional soft information-based
combining.
Index TermsCooperative relaying, multiple-antenna systems,
turbo equalization, packet combining, outage probability.
I. INTRODUCTION
In wireless networks, the presence of diversity between the
source and the destination is a key requirement to combat
channel fading and enable communication at high spectral
efciencies. In this paper, we concentrate on temporal and
spatial diversity. Automatic repeat request (ARQ) combined
with forward error correction (FEC) is a popular mechanism
to exploit temporal diversity. This mechanism has been studied
for many years and is still receiving considerable attention [1],
[2]. However, it suffers from temporal diversity limitations
especially in slow fading environments. In [3], the authors
have proposed cooperative relaying transmission as a solution
for mitigating these diversity limitations. This cooperative
diversity version exploits the broadcast nature of the wireless
channel and adds spatial diversity by incorporating relays in
the network. The relays play the role of packet retransmitters
instead of the source, thereby creating an independent channel
to increase the diversity order. Cooperative relaying presents a
good alternative to classical ARQ and is becoming an area of
wide interest for many researchers (see for instance [4], [5]). In
this new transmission mechanism, one or more relays assist the
communication between the source and destination to form a
multiple- inputmultiple-output (MIMO) system and therefore
build up spacetime diversity branches that are exploited at the
destination.
Several interesting relaying schemes have been proposed,
among which are two basic modes: amplify-and-forward (AF),
and decode-and-forward (DF). The AF strategy represents the
simplest way that a relay may cooperate with the source and
the destination. Under this scheme, the relay simply amplies
the received signal and forwards it towards the destination.
However, in the DF scheme, the relay rst decodes the signal
received from the source, re-encodes and retransmits it to
the destination. This approach suffers from error propagation
when the relay transmits an erroneously decoded data block
[6], [7]. Selective DF, where the relay only transmits when it
can reliably decode the data packet, has been introduced as
an efcient method to reduce error propagation [8]. However,
the unsuccessful detection of the data packet by one or more
relays can limit the benet of relaying. In fact, for each
incorrectly detected packet, there is a waste of one time slot.
To prevent the incurring silence, a modied selective DF
scheme has been proposed in [6], [10]. In this scheme, when
the relay fails to correctly decode the packet, it sends back a
negative acknowledgment (NACK) message to the source that
directly transmits the packet to the destination during the slot
allocated to the relay. In this paper, we refer to this scheme
as ACK/NACK-aided DF.
To improve spatial diversity of a relaying system, signals
received over the sourcedestination and the relaydestination
links are combined at the receiver side. Most of the research
work in this area has focused on at fading channels. However,
in practical systems, channels connecting the source, the relay
and the destination may suffer from inter-symbol interference
(ISI) caused by frequency-selective fading. In [11], the authors
have introduced a maximum ratio combining (MRC)-aided
strategy for AF scheme operating under the so-called protocol
I, where the source broadcasts the data packet to both the relay
and the destination during the rst slot, and both the source
and the relay re-send the packet to the destination during
the second slot [12]. Block equalization has been proposed
in [13] for protocol III, where the broadcast nature of the
channel is not considered, i.e., the source sends to the relay
during the rst slot, and both the source and the relay send
to the destination in the second slot. In [14], a frequency-
domain equalizer with diversity combining has been proposed
for single antenna cooperative systems with demodulate-and-
forward relaying and using the so-called protocol II. In this
protocol, the operation mode during the rst slot is similar to
that of protocol I, while during the second slot only the relay
sends the packet to the destination.
2
In this paper, we consider a broadband relaying system
operating under the framework of protocol II as it is widely re-
garded as an efcient relaying scheme for increasing the over-
all throughput. We focus on cooperative ARQ communication
where the feedback from the destination is exploited and the
packet repetition is activated only if the destination fails to de-
code the data packet [15], [16]. We introduce a communication
framework that is valid for AF, selective DF, and ACK/NACK-
aided DF schemes. We then derive the optimal maximum
a posteriori (MAP) turbo packet combiner that performs, at
the destination side, the combination of packets received over
multiple slots. We also examine the outage probability of the
different relaying schemes, and show that they outperform
each other depending on the location of the relays. We then
introduce a low-complexity minimum mean square (MMSE)
packet combiner and show that it remarkably outperforms clas-
sical soft information-based combining. Throughout the paper
we use the following notation: (.)

and (.)
H
are the transpose
and the transpose conjugate of the argument, respectively.
diag {x} and diag {X
1
, , X
m
} denote the diagonal matrix
and the block diagonal matrix constructed from x C
n
and
from X
1
, , X
m
C
n1n2
, respectively. For x C
TN
,
x
f
denotes the discrete Fourier transform (DFT) of x, i.e.
x
f
= U
T,N
x, with U
T,N
= U
T
I
N
, where I
N
is the
N N identity matrix, U
T
is a unitary T T matrix whose
(m, n)th element is (U
T
)
m,n
=
1

T
e
j(2mn/T)
, j =

1,
and denotes the Kronecker product.
The remainder of the paper is organized as follows: In
Section II, we introduce the relay system model together with
the multi-slot block communication model, and the structure
of the iterative turbo receiver scheme we consider in this
paper. In Section III, we analyze the outage performance of the
considered relaying schemes. Section IV details the structure
of the proposed low-complexity packet combining technique.
The performance analysis is provided in Section V. Finally,
the paper is concluded in Section VI.
II. RELAY SYSTEM MODEL AND ITERATIVE TURBO
RECEPTION
A. Relay System Model
We consider a relay-assisted wireless communication sys-
tem where the M
S
antenna source terminal denoted as S
transmits information blocks to the M
D
antenna destination
terminal denoted as D with the assistance of K1 dedicated
relays denoted as R
2
, , R
k
, , R
K
. Each relay R
k
is
equipped with M
R
k
transmit and receive antennas. The source
relay (S R
k
), sourcedestination (S D), and relay
destination (R
k
D) links are assumed to be frequency
selective. The channel matrices corresponding to the A B
link are H
(AB)
0
, , H
(AB)
LAB1
C
MBMA
, where L
AB
denotes the number of symbol-spaced taps, and A {S, R
k
},
and B {R
k
, D}. Their entries are zero-mean circularly
symmetric complex Gaussian random variables. Cyclic prex
(CP)-aided transmission is assumed for all links. The average
energies of the different links are E
SR
k
, E
SD
and E
R
k
D
, and
take into account the path-loss and shadowing effects of each
link.
In this subsection, we introduce a communication model
that is valid for three relaying schemes, AF, selective DF, and
ACK/NACK-aided DF schemes. The three schemes use up
to K time slots for sending one information block from the
source to the destination, where each slot spans T channel
use. First, the source encodes its data blocks using a space
time bit interleaved coded modulation (STBICM) encoder. The
resulting symbol vector is given by,
s
_
s

0
, , s

T1

S
MST
, (1)
where s
i

_
s

1,i
, , s

t,i
, , s

MS,i
_

S
MS
is the symbol
vector at channel use i = 0, , T 1, and S is the symbol
constellation set. During the rst slot, the source inserts a CP
symbol word of length T
SD
CP
max
k=2, ,K
(L
SR
k
, L
SD
), then
broadcasts the resulting symbol frame to the K1 relays and
the destination. After CP deletion, the baseband M
D
1 signal
vector obtained at the destination is given by,
y
(1)
i
=
_
E
SD
LSD1

l=0
H
(1)
l
s
(il) mod T
+n
(1)
i
, (2)
where H
(1)
l
= H
(SD)
l
, and n
(1)
i
N
_
0
MD1
,
2
I
MD
_
is the thermal noise at the destination receiver. During the
following K 1 slots, the transmission strategy depends on
the considered relaying scheme. After CP deletion, The M
D
1
received signal vector at the destination side, at each slot
k = 2, . . . , K can be expressed as
y
(k)
i
=
_
E
k
L
k
1

l=0
H
(k)
l
s
(il) modT
+n
(k)
i
. (3)
This expression is valid for both AF and DF schemes:
In AF scheme, each relay R
k
amplies and sends the
block of received signals to the destination during the
allocated slot. In this case, H
(k)
l
in (3) is the lth equivalent
tap channel matrix corresponding to link S R
k
D.
It is the convolution of the two channels corresponding
to links S R
k
and R
k
D. L
k
= L
SR
k
+ L
R
k
D
1
is the equivalent channel order, and E
k
=
ER
k
DESR
k
MSESR
k
+
2
.
In DF schemes, the relays transmit using M
S
antennas.
In the case of selective DF, H
(k)
l
= H
(R
k
D)
l
C
MDMS
,
L
k
= L
R
k
D
, and E
k
= E
R
k
D
if relay k can correctly
decode the packet, otherwise, the packet retransmission
is not activated during slot k. When ACK/NACK-aided
DF is considered, both the relay and the source are
involved during the following K 1 slots. In fact, at
each relay, an acknowledgment message is generated after
the information block decoding. If the decoding outcome
is erroneous, the relay broadcasts a NACK message
to both the destination and the source to indicate that
during the allocated slot the source is going to directly
send the symbol frame to the destination. In this case,
H
(k)
l
= H
(SD)
l
, L
k
= L
SD
, and E
k
= E
SD
. The
reception of an ACK message indicates that during the
allocated slot, the source will keep silent while the relay
will transmit the symbol frame to the destination, which
means H
(k)
l
= H
(R
k
D)
l
C
MDMS
, L
k
= L
R
k
D
, and
3
E
k
= E
R
k
D
. In this work, we suppose perfect packet
error detection at both the relay and the destination, and
assume that the one bit ACK/NACK feedback message
is error-free.
B. Multi-Slot Block Communication Model
After k time slots, the system (source, k 1 relays, and
destination) can be viewed as a point to point MIMO link with
M
S
transmit and kM
D
receive antennas. Using the virtual
antennas concept, we construct the kM
D
T 1 block received
signal vector y
(k)
after k slots as,
y
(k)

_
y
(k)

0
, , y
(k)

T1
_

C
kMDT
, (4)
where
y
(k)
i

_
y
(1)

i
, , y
(k)

i
_

C
kMD
(5)
is the signal received over the kM
D
virtual antennas cor-
responding to reception over k consecutive slots using M
D
receive antennas. The block communication model corre-
sponding to this k-slot scheme is given by,
y
(k)
= H
(k)
s +n
(k)
, (6)
where H
(k)
C
kMDTMST
is a block circulant matrix whose
rst kM
D
T M
S
block column matrix is
_
H
(k)

0
, , H
(k)

L1
, 0
(TL)kMDMS
_

, (7)
with
_
_
_
L = max
k=1, ,K
(L
k
),
H
(k)
l

_

E
1
H
(1)

l
, ,

E
k
H
(k)

l
_

C
kMDMS
,
(8)
correspond to the order of the virtual MIMO channel, and
H
(k)
l
is the channel matrix of the lth virtual tap. Vector
n
(k)

_
n
(k)

0
, , n
(k)

T1
_

C
kMDT
, (9)
where
n
(k)
i

_
n
(1)

i
, , n
(k)

i
_

(10)
N
_
0
kMD1
,
2
I
kMD
_
denotes the thermal noise present in
the k-slot equivalent MIMO communication system. Note that
the block circulant matrix H
(k)
can be block diagonalized in
a Fourier basis as
H
(k)
= U
H
T,kMD

(k)
U
T,MS
. (11)
Therefore, applying the DFT U
T,kMD
on the k-slot received
block signal vector (4) yields the following frequency domain
block communication model,
y
(k)
f
=
(k)
s
f
+n
(k)
f
, (12)
where
_

(k)
diag
_

(k)
0
, ,
(k)
T1
_
C
kMDMST
,

(k)
i
=

L1
l=0
H
(k)
l
e
j(2il/T)
C
kMDMS
.
(13)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Slot # k
LLRs a priori
.
.
.
D
F
T
C
P
D
F
T
C
P
interleaver
SISO
CRC
check.
deinterleaver
d
e
l
e
t
i
o
n
d
e
l
e
t
i
o
n
decoder
Slot # 1
s
o
f
t
p
a
c
k
e
t

c
o
m
b
i
n
i
n
g
(To source and relays)
ACK/NACK
Figure 1. The block diagram of the turbo packet combining receiver scheme
at slot k.
C. Iterative Receiver Scheme
In this subsection, we introduce an iterative turbo packet
combining receiver scheme for decoding the data packet using
signals received during slots 1, , K. After k slots, the
receiver constructs the frequency domain block signal vector
y
(k)
f
. The decoding of the information frame is iteratively
performed by exchanging soft information between the MAP
signal combiner and the soft input soft output (SISO) de-
coder. First, the soft MAP combiner computes extrinsic log-
likelihood ratio (LLR) values about coded and interleaved
bits using a-priori information. Second, the generated soft
output is de-interleaved, and transferred to the SISO decoder
for computing a-posteriori LLR on useful bits and extrinsic
information on coded bits. After a preset number of iterations,
the decision about the data packet is performed. If the packet
is incorrectly decoded, a NACK message is sent to relay k+1
to start the packet retransmission process depending on the
relaying scheme in use. If the packet is correctly decoded,
the destination broadcasts an ACK message to the source
and relays to stop relaying and move on to the next data
packet during the next time slot. The general block diagram
of the iterative receiver is presented in Fig. 1. Let N denote
the preset number of turbo iterations performed between
the soft combiner and the decoder (index n = 1, , N),
and
(a)
t,i,j,n
(k) denotes the a-priori LLR for coded bit b
t,i,j
available at the input of the soft combiner at iteration n of slot
k. Extrinsic LLR about coded and interleaved bit b
t,i,j
at the
output of the optimal MAP combiner is expressed according
to (14), where
m
{s
t,i
} is an operator extracting the mth bit
labeling the symbol s
t,i
S, and S
m,t,i

is the set of symbol

vectors having the mth bit in symbol s
t,i
set to {0, 1},
i.e., S
m,t,i

=
_
s S
MST
:
m
{s
t,i
} =
_
.
III. OUTAGE PROBABILITY PERFORMANCE
In this section, we examine the outage probability of the
relaying schemes we study in this paper. Using simulations,
we show that these schemes outperform each other depending
on the location of relays.
A. Outage Probability
The outage probability is a meaningful tool for evaluating
the performance of non-ergodic channels. For a given SNR ,
the outage probability of the direct transmission link S D
during slot 1 is the probability that the mutual information
4

(e)
t,i,m,n
(k) = log

sS
m,t,i
1
exp
_
_
_

1
2
2
_
_
_y
(k)
f

(k)
s
f
_
_
_
2
+

(m

,t

,i

)=(m,t,i)

(a)
t

,i

,m

,n

m
{s
t

,i
}
_
_
_

sS
m,t,i
0
exp
_
_
_

1
2
2
_
_
_y
(k)
f

(k)
s
f
_
_
_
2
+

(m

,t

,i

)=(m,t,i)

(a)
t

,i

,m

,n

m
{s
t

,i
}
_
_
_
. (14)
I between the transmitter alphabet and the received signal is
below a certain transmission rate R, and can be expressed in
the case of frequency selective fading channels as,
P
Direct
out
(R, , 1) = Pr
_
I(s
f
, y
(1)
f
|
(1)
, ) < R
_
. (15)
Packet combining starts from k 2 when the direct link is
in outage. For AF scheme, at each slot k, k copies of the
transmitted packet are available at the destination side, one
from the direct link and k1 from the equivalent S R D
links. Then the outage probability for AF scheme is given by,
P
AF
out
(R, , k) =
Pr
_
1
k
I(s
f
, y
(k)
f
|
(k)
, ) < R, A
1
, , A
k1
_
, (16)
where A
u
denotes the event that the destination sends back
a NACK message at slot u = 1, , k 1. However, for
DF schemes, the received signals, after k slots, depend on
the quality of S R links. Let E
k
denote the event that
the relaying system {Source, k 1 relays, destination} is
in outage at slot k, and C
k,
the event that S R links
among the k1 links S R
2
, ,S R
k
are in outage. The
outage probability of the DF relaying system at slot k can be
expressed as,
P
DF
out
(R, , k) =
k1

=0
Pr {E
k
, C
k,
, A
1
, , A
k1
} . (17)
Note that in the case of selective DF relaying, packet trans-
mission does not occur at slot u = 2, , k 1 if link
S R
u
is in outage. This generates an event A
u
at slot u. Let
R
k,
{R
2
, , R
k1
} denote a set of relays involved in
event C
k,
, and R
k,
{R
2
, , R
k1
} \ R
k,
. The outage
probability (17) can now be expressed as,
P
DF
out
(R, , k) =
k1

=0

R
k,
P
S,R
k,
D
out

RR
k,
P
SR
out

R
k,
_
1 P
SR

out
_
, (18)
where P
SR
out
(resp. P
SR

out
) denotes the outage probability
of the S R (resp. S R

) link and is computed using (15),

while P
S,R
k,
D
out
denotes the outage probability of the system
after combining at slot k using relays in R
k,
. It is computed
similarly to (16) but using only the channel matrices of R D
links whose relays are in set R
k,
.
B. Outage Analysis
In this subsection, we compare the outage probability of the
studied relaying schemes using Monte Carlo simulations. We
consider a relaying system with the same number of transmit
and receive antennas M
S
= M
R
= M
D
= 2 and one relay
(K = 2). For simplicity, we consider a homogeneous case
in which the distance between the source and the relay l
SR
,
the relay and the destination l
RD
, and the source and the
destination l
SD
are normalized as l
SR
+ l
RD
= l
SD
= 1.
All links have the same frequency-selective fading channel
prole, i.e., L = 3 equal power paths with the same path
loss exponent = 3. The link average energy is assumed
to be E
AB
= (l
AB
)

with A = S or R, and B = R
or D. In Fig. 2, we report the outage probability versus
l
SR
for three signal-to-noise ratio SNR
SD
values, 0dB, 3dB,
and 5dB. In the legend, ACK/NACK-DF (Slow Ch) and
ACK/NACK-DF (fast Ch) denote the ACK/NACK-aided DF
scheme operating over an S D channel constant over K
consecutive slots, and an S D channel that independently
changes from slot to slot, respectively. The results show that
AF and DF relaying schemes outperform each other depending
on the relay location. In fact, AF scheme seems to be more
suitable for locations close to the destination, i.e., l
SR
> 0.6.
For locations close to the source, the relay experiences better
radio conditions, and the probability to correctly decode the
data packet becomes higher. In this case, DF schemes are
more suitable. From the gure we can see that the studied
DF schemes have similar performances when l
SR
< 0.6.
When l
SR
> 0.6, ACK/NACK-aided DF scheme clearly
outperforms selective DF scheme. However, when the S D
link experiences slow fading the gap becomes too small. In
this case, selective DF is the best as it does not involve the
source during the relaying slots.
IV. SUB-OPTIMAL FREQUENCY DOMAIN TURBO PACKET
COMBINING
It is obvious that using the MAP criterion for performing
turbo packet combining translates into very high complexity
requirements. In this section, inspired by the packet combining
concept introduced in [17], [18] for ARQ, we propose an
MMSE-based turbo packet combining scheme for broadband
relaying systems. The concept performs, in the frequency
domain, interference cancellation jointly for all slots followed
by MMSE ltering. Let s denote the conditional estimate of s,
and
2
t,i
the conditional variance of s
t,i
. With the aid of multi-
slot block communication model (12), the MMSE estimate
z
(k)
f
about s
f
, at time slot k, can be expressed according to
5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
2
10
1
10
0
SNR
SD
=0dB
l
sr
O
u
t
a
g
e
P
r
o
b
a
b
ilit
y
ACK/NACKDF (Fast Ch)
ACK/NACKDF (Slow Ch)
Selective DF
AF
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
5
10
4
10
3
10
2
10
1
10
0
SNR
SD
=3dB
l
sr
O
u
t
a
g
e
P
r
o
b
a
b
ilit
y
ACK/NACKDF (Fast Ch)
ACK/NACKDF (Slow Ch)
Selective DF
AF
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
4
10
3
10
2
10
1
10
0
SNR
SD
=5dB
l
sr
O
u
t
a
g
e
P
r
o
b
a
b
ilit
y
ACK/NACKDF (Fast Ch)
ACK/NACKDF (Slow Ch)
Selective DF
AF
Figure 2. Outage probability versus l
SR
for M
S
= M
R
= M
D
= 2, K = 2, L = 3, and the path loss exponent = 3.
the following forwardbackward ltering structure,
z
(k)
f
=
(k)
y
(k)
f

(k)
s
f
, (19)
where
(k)
= diag
_

(k)
0
, ,
(k)
T1
_
is the multi-slot
forward MMSE lter given by,
_

(k)
i

(k)
H
i
B
(k)
1
i
,
B
(k)
i
=
2
I
kMD
+
(k)
i

(k)
H
i
.
(20)

is an unconditional covariance computed as the time average

of conditional covariance matrices
i
dened as,

i
diag
_

2
1,i
, ,
2
MS,i
_
, (21)
and
(k)
= diag
_

(k)
0
, ,
(k)
T1
_
is the multi-slot back-
ward MMSE lter given by,
_

(k)
i

(k)
i

(k)
i

(k)
,

(k)
=
1
T
T1

i=0

(k)
i

(k)
i
.
(22)
Note that the proposed MMSE packet combining approach
requires the computation of matrix inverses B
1
0
, , B
1
T1

C
kMDkMD
at each turbo iteration. Furthermore, the signals
received at slots 1, , k and their corresponding channel
responses have to be stored in the receiver. The implementation
of such a receiver is feasible for downlink transmissions, where
the number of receive antennas is less than that of transmit
antennas, with limited number of relay, i.e., one or two relays.
However, for transmissions where KM
D
M
S
, this approach
becomes unfeasible in practice since the receiver will require
a huge memory as well as the inversion of large size matrices.
In this case, to prevent the computation of KM
D
KM
D
matrix inversions, we apply the matrix inversion lemma [19]
which allows us to express the inverse B
1
i
as
B
1
i
=
1

2
_
I
KMD

i
C
1
i

H
i
_
, (23)
where C
i
=
2

1
+
H
i

i
C
MSMS
. After computing
z
(k)
f
, the inverse DFT (IDFT) is then applied to obtain the
equalized time domain sequence, z
(k)
= U
H
T,MS
z
(k)
f
. The
MMSE estimate z
(k)
t,i
corresponding to antenna t and channel
use i can simply be extracted from z
(k)
as z
(k)
t,i
= e
H
t,i
z
(k)
,
with e
t,i
denotes the (M
S
i + t)th vector of the canonical
basis. At the nth iteration of time slot k, the extrinsic LLR

(e)
t,i,m,n
(k) corresponding to coded and interleaved bit b
t,i,m
is then obtained as,

(e)
t,i,m,n
(k) =
log

sS
m
1
exp
_

z
(k)
t,i
g
t,i
s

2
t,i
+

=m

(a)
t,i,m

,n

m
{s}
_

sS
m
0
exp
_

z
(k)
t,i
g
t,i
s

2
t,i
+

=m

(a)
t,i,m

,n

m
{s}
_

_
,
(24)
where g
t,i
and
t,i
denote, respectively, the equivalent channel
gain at the output of the equalizer and the residual interference
variance corresponding to channel use i, and transmit antenna
t. S
m

is the set of symbols having the mth bit set to

{0, 1}, i.e., S
m

= {s :
m
{s} = }.
V. PERFORMANCE EVALUATION
In this section, we evaluate the block error rate (BLER)
performance of the proposed packet combining strategy. We
use conventional LLR-level packet combining as a reference.
In this basic combining strategy, the turbo equalization is
performed separately for each slot, and before SISO decoding,
the extrinsic LLRs generated by the soft demapper are simply
added together with those obtained at the last iteration of
previous slot. We also use the outage probability to evalu-
ate the diversity gain achieved by the introduced combining
strategy. For simulations, we consider a system conguration
similar to that presented in III-B, and using a quadrature phase
shift keying (QPSK) modulation, and a 16 state convolutional
encoder with polynomial generators (35, 23)
8
. The length of
the code frame is 2048 bits including tails, and the CP length
is T
CP
= 3. We use the Max-Log-MAP algorithm for SISO
decoding. The iterative MMSE receiver at the destination runs
three turbo iterations.
Our main concern is to show the superior performance
of the proposed combining strategy compared to LLR-level
combining. First, we consider in Fig. 3 a relaying system
with the same number of transmit and receive antennas M
S
=
M
R
= M
D
= 2. We observe that the proposed combining
6
12 10 8 6 4 2 0 2 4 6
10
3
10
2
10
1
10
0
SNR
SD
(dB)
B
L
E
R
Direct transmission
AmplifyandForward
Selective DF
ACK/NACKDF (Fast Ch)
ACK/NACKDF (Slow Ch)
LLRlevel packet combining
Low complexity packet combining
Outage Probability
Figure 3. BLER performance for CC (35, 23)
8
, QPSK, M
S
= M
R
=
M
D
= 2, K = 2, L = 3 equal energy paths, l
SR
= 0.3 and the path loss
exponent = 3.
8 6 4 2 0 2 4 6 8 10 12
10
3
10
2
10
1
10
0
SNR
SD
(dB)
B
L
E
R
Direct transmission
AmplifyandForward
Selective DF
ACK/NACKDF (Fast Ch)
ACK/NACKDF (Slow Ch)
LLRlevel packet combining
Low complexity packet combining
Outage Probability
Figure 4. BLER performance for CC (35, 23)
8
, QPSK, M
S
= M
R
= 2,
M
D
= 1, K = 2, L = 3 equal energy paths, l
SR
= 0.7 and the path loss
exponent = 3.
strategy clearly outperforms LLR-level combining. However,
the loss compared with the outage, at 10
2
BLER, is less
than 1dB for all relaying schemes: AF, selective DF, and
ACK/NACK-aided DF. Note that both combining strategies
fail to achieve the diversity order of the relay system. Fig. 4
shows the performance of an overloaded system where M
S
=
M
R
= 2 and M
D
= 1. In this case, the proposed combining
strategy signicantly outperforms LLR-level combining, i.e.,
the performance gap is more than 2dB at 10
2
BLER. Also,
note that the proposed combining strategy almost achieves the
asymptotic slope of the outage probability, while LLR-level
combining fails to do so.
VI. CONCLUSION
In this paper, we investigated the issue of relay com-
munication over broadband MIMO channels. First of all,
we introduced the optimal MAP turbo packet combiner that
exploits all the diversities available in the relaying system.
We then examined the outage probability of different relaying
schemes, i.e., AF, selective DF, and ACK/NACK-aided DF,
and showed that they outperform each other depending on
the location of the relays. In the second part of this work,
we proposed a low complexity MMSE-based turbo packet
combiner. Using simulations, we showed that the proposed
combining strategy provides better BLER performance than
conventional LLR-level combining.
REFERENCES
[1] S. Lin, D. J. Costello, and M. J. Miller, Automatic repeat-request error
control schemes, IEEE Communications Magazine, vol. 12, pp. 5-17,
Dec. 1984.
[2] G. Caire, and D. Tuninetti, ARQ protocols for the Gaussian collision
channel, IEEE Trans. Inf. Theory, vol. 47, no. 4, pp. 19711988, Jul.
2001.
[3] E. Zimmermann, P. Herhold, and G. Fettweis, The impact of coopera-
tion on diversity-exploiting protocols, IEEE VTC-spring, Milan, Italy,
May 2004.
[4] J. N. Laneman, G. W. Wornell, and D. N. C. Tse, An efcient
protocol for realizing cooperative diversity in wireless networks, IEEE
International Symposium on Information Theory (ISIT), Washington,
DC, June 2001.
[5] A. Sendonaris, E. Erkip, and B. Aazhang, User cooperation diversity
Part I & Part II, IEEE Trans. Commun., vol. 51, pp. 1927-1948, Nov.
2003.
[6] J. N. Laneman, D. Tse, and G. W. Wornell, Cooperative diversity in
wireless networks: Efcient protocols and outage behavior, IEEE Trans.
Inform. Theory, vol. 50, no. 12, pp. 3062-3080, Dec. 2004.
[7] J. Boyer, D. D. Falconer, and H. Yanikomeroglu, Multihop diversity in
wireless relaying channels, IEEE Trans. on Comm., vol. 52, pp. 1820-
1830, Oct. 2004.
[8] F. Atay Onat, H. Yanikomeroglu, and S. Periyalwar, Relay-assisted spa-
tial multiplexing in wireless xed relay networks,, IEEE GLOBECOM,
San Francisco, USA, Nov.- Dec. 2006.
[9] P. Herhold, E. Zimmermann, and G. Fettweis, A simple cooperative
extension to wireless relaying, Int. Zurich Seminar on Commun.,
Zurich, Switzerland, Feb. 2004.
[10] G. Yu, Z. Zhang, and P. Qiu, Cooperative ARQ in wireless networks:
Protocols description and performance analysis, IEEE International
Conference on Communications (ICC), Istanbul, Turkey, June 2006.
[11] Q. Jia, T. Lv, and G. Ping, An efcient scheme for joint equaliza-
tion and interference cancellation in distributed cooperative diversity
networks, Communication Networks and Services Research (CNSR),
Halifax, Canada, May 2008.
[12] R. U. Nabar, F. W. Kneubiihler, and H. Boelcskei, Performance limits of
amplify-and-forward based fading relay channels, IEEE International
Conference on Acoustics, Speech, and Signal Processing (ICASSP),
Montreal, Canada, May 2004.
[13] H. Mheidat, M. Uysal, and N. Al-Dhahir, "Equalization techniques for
distributed space-time block codes with amplify-and-forward relaying,"
IEEE Trans. Signal Process., vol. 55, pp. 1839-1852, 2007.
[14] H. Xiong and J. X. P. Wang, Frequency-domain equalization and
diversity combining for demodulate-and-forward cooperative systems,
IEEE ICASSP, Las Vegas, Nevada, USA, March-April 2008.
[15] Y. Zhang, H. H. Chen, and M. Guizani, Cooperative Wireless Commu-
nications, Auerbach Publications, 2009.
[16] B. Zhao and M. C. Valenti, Practical relay networks: A generalization
of hybrid-ARQ, IEEE J. Select. Areas. Comm., vol. 23, no. 1, Jan.
2005.
[17] T. Ait-Idir, and S. Saoudi, Turbo packet combining strategies for the
MIMO-ISI ARQ channel, IEEE Trans. Commun., In Press.
[18] T. Ait-Idir, H. Chafnaji, and S. Saoudi, Turbo packet combining for
broadband space-time BICM ARQ systems with co-channel interfer-
ence, Conditionally accepted, IEEE Trans. Wireless Commun.
[19] S. Haykin, Adaptive Filter Theory, 3rd Ed. Upper Saddle River, NJ:
Prentice-Hall, 1996.