Beruflich Dokumente
Kultur Dokumente
, Ali Ghrayeb
, Mazen Hasna
D
[n] =
_
2R
c1
SD
[n] +R
c2
SD
[n] +A[n]R
c2
RD
[n]
_
+
,
(6)
where ()
+
denotes the bigger of the contents of the parenthe-
ses and zero, n is the discrete time index for the total period of
both stages,
SD
[n] =
E
b
N0
|h
SD
[n]|
2
,
RD
[n] =
E
b
N0
|h
RD
[n]|
2
and A[n] distinguishes a wrong decoded bit from a correct
one, formally dened as
A[n] =
_
+1, u[n] = u[n]
1, u[n] = u[n]
. (7)
Since we assumed quasi-static fading for the SD and RD
channels,
SD
[n] =
SD
,
RD
[n] =
RD
. Hence, assuming
an all-zero codeword was transmitted, the probability of the
destination decoding a codeword of weight d bits (also called
the pairwise error probability, or PEP),conditioned over the
instantaneous SNRs
SD
,
RD
, can be found as
P (d|
SD
,
RD
) = Q
__
{n}d
D
[n]
_
. (8)
We split d into d
1
+d
2
= d, where d
1
, d
2
refer to the weight of
the error event during the rst and second stage, respectively.
Since the relay is contributing only during the second stage,
we can split d
2
= d
r
+ d
, where d
r
indicates the weight of
bits(in the error event) receiving contribution from the relay
during the second stage, and d
_
(2d
1
R
c1
+d
2
R
c2
)
SD
+ (d
c
d
e
) R
c2
RD
_
+
_
. (9)
Observing the piece-wise nature of P (d|
SD
,
RD
), we
need to evaluate the conditional PEP for different values of
SD
,
RD
, d
c
, d
e
, namely
P (d|
SD
,
RD
, d
c
, d
e
) =
Q(0) =
1
2
, d
e
> d
c
and
RD
>
SD
Q
_
_
()
_
, otherwise,
(10)
where is dened as
=
2d
1
R
c1
+d
2
R
c2
d
e
R
c2
d
c
R
c2
. (11)
To obtain the average PEP, we integrate (9) over the
joint PDF of (
SD
,
RD
). Assuming that the fades ex-
perienced by source-destination and relay-destination chan-
nels are independent, p
() =
1
exp
_
_
,
SD
=
E
b
N0
E
hSD
_
|h
SD
|
2
_
,
RD
=
E
b
N0
E
hRD
_
|h
RD
|
2
_
Without going into the details of the derivation (we refer the
interested reader to [10] for the complete proof), (9) becomes,
by substituting Craigs formula for the Q function (cf. [9]),
integrating over the PDFs and some algebraic manipulation
P(d|d
c
, d
e
) =
_
/2
0
(1+s1 )(1+)
d +
1
2
1
1+
, d
c
< d
e
1
_
/2
0
_
1
1+s1
__
1
1+s2
_
d, d
c
d
e
,
(12)
where s
1
, s
2
are dened as
s
1
=
(2d
1
R
c1
+ (d
2
) R
c2
)
2 sin
2
, s
2
=
(d
c
d
e
) R
c2
2 sin
2
,
respectively. We note from (12) that the PEP can tend to a
constant, given that d
e
> d
c
, as tends to . We also observe
that in all cases where d
e
> 0 the resultant PEP will increase.
1007
Authorized licensed use limited to: QATAR UNIVERSITY. Downloaded on December 27, 2009 at 04:15 from IEEE Xplore. Restrictions apply.
Note that the PEP expression given in (12) is conditional on
d
e
, d
c
(or d
e
, d
r
: d
r
= d
c
+d
e
) which are specic to a group
of error events of weight d. Thus, we need to sum the PEP over
the probability of an error word of weight d having d
r
, d
c
, d
e
as components, as
P(d) =
d2
dr=0
dr
de=0
P(d|d
e
, d
r
= d
c
+d
e
)p
de
(d
e
)p
dr
(d
r
).
(13)
Assuming a uniform distribution of relayed/error bits over the
forwarded frame, the PDF functions for d
r
, d
e
are equal to
p
dr
(d
r
) =
_
d2
dr
__
2Nd2
dRdr
_
_
2N
dR
_ , p
de
(d
e
) =
_
dr
de
__
dRdr
dEde
_
_
dR
dE
_ , (14)
where d
R
, d
E
represent the total number of forwarded bits
from the relay and the number of which are wrong, respec-
tively. Finally, assuming maximum-likelihood (ML) decoding
of the received codeword at the destination, the resultant bit
error rate can be upper-bounded by (cf. [9])
P
b
(e) <
d=dfree
N
i=1
i
N
_
N
i
_
p(d|i)P(d), (15)
where d
free
is the free Hamming distance of the SCCC, and
p(d|i) is the input/output weight distribution function of the
SCCC code.
IV. FINDING AN OPTIMUM THRESHOLD
Having analyzed the system for a given (d
E
, d
R
) combi-
nation, we now discuss setting a threshold at the relay to
achieve minimum end-to-end bit error rate at the destination.
We study the optimal threshold under the assumption that the
relay knows the positions of errors, and then we propose a
practical threshold relying only on the knowledge of CSI (of
the source-relay channel).
A. Genie-Aided Threshold
As a benchmark for any thresholding scheme, we assume
that for any given frame the relay knows the location of
errors. Although impractical, this assumption provides us with
a limiting case of thresholding. Given that the relay knows the
LLR values of all wrong bits, a threshold can be set as the
absolute value of any of these LLRs; preventing all but the
desired number of error to be forwarded. Let L
wrong
be the set
of bits that are known to be wrong, formally dened as
L
wrong
= {|L
u
[n]|}
n: u[n]=u[n]
Assuming L
wrong
is sorted in a decreasing order, we then
formally dene genie-aided thresholds as
T
0
= L
wrong
1
, T
1
= L
wrong
2
, T
2
= L
wrong
3
, . . . (16)
where T indicates the value of the LLR threshold, and the
sub-index indicates number of errors allowed to pass. The
question of optimality here is to choose the threshold that
results in minimum end-to-end bit error rate; keeping in mind
that allowing a few errors to pass also allows correct bits that
can contribute to overall performance.
The analysis of this case is beyond the scope of this paper, by
reason of its impracticality. We nevertheless present simulation
results in [10] that illustrate that the optimal choice in this case
would be allowing no errors to pass at all.
B. CSI-Based Threshold
Observing that the instantaneous PDF of the LLR values
at the relay depends on the underlying source-relay CSI. We
propose a metric that relies only on the CSI of the source-
relay channel. Denoted by Z, this metric is equal to the mean
source-relay channel energy during the transmission of the
current frame, formally dened as
Z =
R
c1
N
N/Rc
1
n=1
|h
SR
[n]|
2
.
We then set a threshold that is linearly proportional to Z as,
T
Z
= Z +. (17)
We establish in [10] that such a threshold guarantees a constant
source-relay bit error rate (with varying channel realizations),
and hence a constant false rejection rate as well. Hence, we
can model the resultant d
E
, d
C
as jointly independent binomial
random variables. And the resultant PEP at the destination is
expressed in (18)
P(d) =
N/Rc
2
dR=0
dR
dE=0
d2
dr=0
dr
de=0
_
P(d|d
e
, d
r
= d
c
+d
e
)
p
de
(d
e
|d
E
, d
R
)p
dr
(d
r
|d
R
)p
dE
(d
E
)p
dR
(d
R
)
_
, (18)
where p
dr
(d
r
|d
R
) and p
de
(d
e
|d
R
, d
E
) are dened in (14).
p
dE
(d
E
) and p
dC
(d
C
) are binomial distributed with success
rates
dE
N/Rc
2
and
dC
N/Rc
2
, respectively and number of trials
N/R
c2
. p
dR
(d
R
= d
C
+ d
E
) = p
dC
(d
C
) p
dE
(d
E
) where
denotes discrete convolution.
The last step in deriving the optimal threshold is establishing
the relationship between T
Z
and
d
E
,
d
R
. Since such a rela-
tionship will depend on the code used between the source and
relay (in our case, the inner constituent encoder of the SCCC),
and the type of decoder used at the relay we opt to nd this
relationship empirically. To choose the optimal threshold, we
need to select the
_
d
E
,
d
R
_
pair that minimizes end-to-end
bit error rate in (15) when the corresponding P(d) found in
(18) is used. In [10], we simulate the source-relay part of
the system under different values of
SR
, and , which
resulted in possible
_
d
E
,
d
C
_
pairs. Hence, we can select the
best combination of and in terms of minimum end-to-end
bit error rate. The optimal point of operation for both values of
SR
= 6, 9 dB were found to be = 0.5, = 3 for
SR
= 6
dB, and = 0.5, = 2.5 for
SR
= 9 dB.
V. SIMULATION RESULTS
We simulated the proposed system, implementing several
relaying protocols to compare their performances. Throughout
1008
Authorized licensed use limited to: QATAR UNIVERSITY. Downloaded on December 27, 2009 at 04:15 from IEEE Xplore. Restrictions apply.
our simulations, the system encodes a frame of N = 100
information bits, with R
c1
= 1/3 and R
c2
= 1/2. The
constituent codes of the SCCC were (13, 17)
8
for the outer
code (E
1
), (27, 31)
8
for the inner code (E
2
), and the desti-
nation implemented an iterative decoder (see [8] for a block
diagram of a typical SCCC iterative decoder) running for 5
iterations. The source-relay channel was block-faded with 20
independently-faded blocks.
To illustrate the strength of our proposed system, we simulate
other protocols at the relay. Namely, employing only a CRC
check at the relay (referred to as simple CRC), forwarding ana-
log LLR values from the relay (adapting the implementation
in [6], and simple DF (without any error propagation control).
In addition, we also display the performance of the optimal
thresholds obtained previously (namely, (16) and (17)).
Simulation results are shown for
SR
= 6 dB in Fig.2. In
addition, the upper bound derived previously for the CSI-
based threshold is also plotted. Compared to simple CRC,
both threshold types display signicant gains, with the CSI-
based threshold displaying as much as 5 dB of gain over
simple CRC; albeit without much diversity gain. Also notable
is the error oor that is displayed when using analog-LLR
forwarding and simple DF, with the latter ooring at a value
an order of magnitude higher. Comparing both thresholds, we
can see a genie-aided threshold provides up to 5 dB gain over
the CSI-based threshold.
In Fig.3, we show simulated results for
SR
= 9 dB. We
note that both threshold types display higher gains compared
to simple CRC. The CSI-based threshold achieves both a
diversity gain and a coding gain over CRC. Analog LLR
relaying, however, achieves slightly better bit error rate ( 0.7
dB) but starts losing diversity at > 20 dB. Comparing
both thresholds, we can see the genie-aided threshold still
outperforms the proposed CSI-based threshold.
VI. CONCLUSION
In this paper, we have presented a novel technique to
mitigate error propagation in cooperative communications. Our
proposed system relied on soft estimates of bits, and used
them to block unreliable bits from being forwarded to the
destination. We developed upper bounds on the performance
of the system and showed that they converge to the actual
performance. We compared our technique with just using CRC
at the relay, with simple DF, and analog LLR forwarding and
displayed signicant improvement in diversity and bit error
rate. While analog LLR forwarding and simple DF caused
an error oor in the end-to-end bit error rate of the system,
CRC lost too much diversity by discarding the whole frame,
and our proposed technique was able to circumvent both
disadvantages.
REFERENCES
[1] D. Chen and J. N. Laneman, Modulation and demodulation for coop-
erative diversity in wireless systems, IEEE Trans. Wireless Commun.,
vol. 5, pp. 17851794, Jul. 2006.
[2] G. Kramer, M. Gastpar, and P. Gupta, Cooperative strategies and
capacity theorems for relay networks, IEEE Trans. Inf. Theory, vol. 51,
pp. 30373063, Sep. 2005.
0 5 10 15 20 25 30
10
6
10
5
10
4
10
3
10
2
10
1
10
0
in dB
E
n
d
-
t
o
-
e
n
d
B
E
R
simple DF
analog LLR relaying
Simple CRC
Bound on CSIbased threshold
CSIbased thresholding
genieaided Thresholding
Bound on errorfree relaying
error free relaying
Fig. 2. Bit error rate vs. , for
SR
xed at 6 dB. Analog LLR relaying
was implemented according to [6]
0 5 10 15 20 25 30
10
6
10
5
10
4
10
3
10
2
10
1
10
0
in dB
E
n
d
-
t
o
-
e
n
d
B
E
R
Simple DF
Simple CRC
analog LLR relaying
Bound on CSIbased threshold
CSIbased threshold
genieaided threshold
Bound on errorfree relaying
error free relaying
Fig. 3. Bit error rate vs. , for
SR
xed at 9 dB. Analog LLR relaying
was implemented according to [6]
[3] H. V. Khuong and H. Y. Kong, LLR-based decode-and-forward proto-
col for relay networks and closed-form ber expressions, IEICE Trans.
Fundamentals, vol. E89A, pp. 18321841, Jun. 2006.
[4] T. Hunter and A. Nosratinia, Diversity through coded cooperation,
IEEE Trans. Wireless Commun., vol. 5, pp. 283289, Feb. 2006.
[5] B. Zhao and M. Valenti, Distributed turbo coded diversity for relay
channel, Electronics Letters, vol. 39, pp. 786787, May 2003.
[6] Y. Li, B. Vucetic, T. F. Wong, and M. Dohler, Distributed turbo coding
with soft information relaying in multihop relay networks, IEEE J. Sel.
Areas Commun., vol. 24, pp. 20402050, Nov. 2006.
[7] Y. Hairej, A. Darmawan, and H. Morikawa, Cooperative diversity using
soft decision and distributed decoding, in Proc. Mobile and Wireless
Communications Summit, Nagoya, Japan, Jul. 2007, pp. 15.
[8] T. M. Duman and A. Ghrayeb, Coding for MIMO Communication
Systems. New York: Wiley, 2008.
[9] M. Elturi, W. Hamouda, and A. Ghrayeb, A convolutional-based coded
cooperation scheme for relay channels, IEEE Trans. Veh. Technol.,
accepted for publication, May 2008.
[10] G. Al-Habian, A. Ghrayeb, M. Hasna, and A. Abu-Dayya, A novel
technique for mitigating error propagation in cooperative communication
networks, submitted to IEEE Trans. on Communications. [Online].
Available: http://users.encs.concordia.ca/g alhabi/jrnl.pdf
1009
Authorized licensed use limited to: QATAR UNIVERSITY. Downloaded on December 27, 2009 at 04:15 from IEEE Xplore. Restrictions apply.