You are on page 1of 4

2010 Interational Symposium on Intelligent Signal Processing ad Communication Systems (lSP ACS 2010) December 6-8, 2010

PERFORMANCE ANALYSIS OF LLR COMBINING HARQ FOR MIMO


SYSTEMS IN MOBILE WIMAX
Rahmat Mulyawan1, Fin Nugroho2, Riris Nov/, Felis Dwiyasa2, Trio Adiono1
lInstitut Teknologi Bandung (ITB), Indonesia. Tel/Fax: +62-22-250-6280
rmulyawan@students.ee.itb.ac.id, tadiono@paume.itb.ac.id
2Xirka Silicon Technology, Indonesia. PhonelFax: +62-22-2014189/+62-22-2014253
{ffn.nugroho, riris.novi, felis.dwiyasa}@xirkachipset.com
ABSTRACT
The use of hybrid automatic-repeat-request (HARQ) with
multi-antenna systems (MIMO) promises high throughput
with high reliability in broadband wireless communication
systems. One of the combining methods in MIMO-HARQ
is by directly optimizing the log-likelihood ratio (LLR)
values instead of compensating the multiple signal-to
interference-and noise power ratios (SINRs) as in
conventional combining method. The problem of this
approach is that the receiver complexity is signifcantly
increased along with the order of the modulation used.
This paper proposed simplifed implementation of the
MIMO-HARQ with LLR combining method under the
IEEE 802.l6e Mobile WiMA system. Proposed scheme
are verifed using ITU-B Pedestrian and ITU-A Vehicular
channel model with various modulation order. Simulated
packet error rate (PER) results show that the simplifed
method paired with CTC has greater performance, yet
lower complexity, compared to original direct method
paired with CC.
1. INTRODUCTION
Hybrid ARQ (HARQ) systems incorporate both
channel coding and an ARQ system to achieve high
throughput and high reliability, which are the key
properties of channel coding and an ARQ systems,
respectively [1]. Instead of discarding the previously
received signals that are detected to contain errors as in
ARQ systems, HARQ systems further enhance their
performances by combining all the received signals to
decode the transmitted message. There are two popular
types of HARQ: HARQ with Chase combining (HARQ
Chase) [2] and HARQ with incremental redundancy
(HARQ-IR) [1]. In HARQ-Chase, the transmitter sends a
message that is coded by both error-detection coding, such
as Cyclic Redundancy Check (CRC), and channel coding.
978-1-4244-7371-7/10/$26.00 2010 IEEE
Then, the receiver decodes the message and detects any
error in the CRC. If an error is detected, the receiver
requests the transmitter via feedback channel to retransmit
the message using the same modulation-and-coding
scheme. Although the same signal vector is retransmitted,
it difers fom the received signal vector fom the previous
one because of time diversity. Afer combining the new
received signal vectors with the old one, the receiver
decodes the combined transmitted message. This
procedure repeats until no eror is detected afer
combining all the received signal vectors.
Retransm|ss|on
(sameacket0ata)
I
Retransm|ss|on
+

|_a,etogtter g 3ott
CRCta|'ed CRCscceded
Fig. 1. HRQ-Chase
I
ACK
HARQ-IR works in a similar way to HARQ-Chase
except that HARQ-IR uses different modulation and
coding schemes for retransmissions, providing fexibility
and more robust throughput, but the receiver complexity
2010 Interational Symposium on Intelligent Signal Processing ad Communication Systems (lSP ACS 2010) December 6-8, 2010
becomes higher in retur [3]. Therefore, HARQ-Chase is
more widely used in wireless communications systems,
including systems based on the IEEE 802.16e Mobile
WiMA [4] because it can be easily implemented and
provides a good operating point in the tradeof between
throughput and reliability.
Combining schemes for MIMO with HARQ-Chase
proposed in [5] works in a fndamentally different way
than conventional receiver design. Instead of utilizing
receive flter to maximize the SINRs, it directly optimize
the parameter most-closely related to the decoding
performance, i.e., the log-likelihood ratio (LLR), which is
widely-used sof-bit information metric. The decoding
perforances are then analyzed by comparing the LLR
values, which makes this method also known as HARQ
with LLR combining.
This paper proposed simplifed implementation of the
MIMO-HARQ with LLR combining method under the
Mobile WiMA system. Proposed scheme are verifed
using ITU-B Pedestrian and ITU-A Vehicular channel
model in various modulation order. Simulated packet error
rate (PER) results will be presented for each scenario.
2. SYSTEM ARCHITECTURE
We employ IEEE 802.l6e MIMO system that employs
NT transmit antennas and NR receive antennas, thereby
using NT spatial streams. In this paper we use NT = NR =
2 as in Mobile WiMA System Profle [6], but this
scheme can be extended to arbitrar NT and NR.
C
0
C
+

C
C

r
. 8
C
!

~~ |adIng Cbanne|+ AWGN ~
Fig. 2. MIMO-HARQ in IEEE 802.16e Mobile WiMAX
We will focus the analysis in the receiver, mainly
MIMO HARQ Decoder. If N transmissions has occurred
for the same transmit message, then the relationship
between the transmitted signal vector and the received
signal vector at the i-th transmission is

i=
!b+
1
i
'
i=I
,
... ,N
(1)
and its conditional probability distribution fnction is
P (
i
1
!,
b)
=
n!
R exp
( - 11

i
-
!b11
2
)
(2)
where

i denotes the NR x 1 received signal vector,


!
is
NR X NT channel response matrix,
b
is NT X 1
transmitted signal vector, and 1i denotes NR x 1 additive
white Gaussian noise (A WGN) vector at time i which is
assumed to be Li.d. and zero-mean circularly symmetric
complex Gaussian (ZMCSCG) with covariance b
R

3. MIMO-HARQ DECODER
Our system uses MIMO Spatial Multiplexing (Matrix
B) with HARQ-Chase with Maximum Likelihood (ML)
decoder. The receiver can be extended to be used with
linear equalizer such as ZF or MMSE [7] but will results
in suboptimal performance [8].
A. HARQ Combining Scheme
Afer the (N
-
1) transmission, the receiver stores the
values of Euclidean distance IIN _l!112 in the buffer
for every received vector estimation g, and combine these
values with the new Euclidean distance calculated in the N
transmission. DLC scheme acts as a big ML decoder, flly
using all the relevant information, i.e.,

i and
!
for all
i = 1, ... , N. Therefore, the DLC scheme has the optimal
decoding performance. However, directly implementing
the original DLC scheme proposed by [5] imposes
prohibitively high computational complexity because it
involves a division, summations of exponential fnctions,
and logarithmic operations per each LLR value.
(3)
The complexity can be reduced by using max-Iog
MAP approximation [10]: 10gLi
exp
ai maxiai , at the
expense of some degradation in decoding performance.
The resulting LLR for hi is
(4)
Combining of the LLR can also be done in the symbol
level by using extended version of MRC scheme in
MIMO to HARQ, known as Symbol Level Combining
(SLC) [9]. Similar to DLC scheme, we can use
approximation to calculate the LLR,
(5)
2010 Interational Symposium on Intelligent Signal Processing ad Communication Systems (lSP ACS 2010) December 6-8, 2010
I
- N [ , -
- 2
( N [ )
where ! = L
i
=
I
!!
and

N =
! L
i
=l! i .
Furthermore, this approach not only simplifes LLR
calculation but also reduces the complexity of the SLC
_ I
scheme by removing the needs to calculate H
- 2
, the
square root inverse of combined channel matrix [5].
Another method to combine LLR is done in bit level or
Bit Level Combining (BLC). This method is practically
simpler to implement and provides fexibility to be used in
HARQ-IR. But it sufers fom slight performance
degradation because it neglects the fact that the same
transmit signal vector is repeatedly reused, contrary to
DLC schemes. We can derived the BLC scheme fom
DLC scheme by interchanging the order of summation
and minimization and frther simplif the algorithm using
the approximation that we have used before. The LLR of
BLC then becomes
LLRprox
)
. L7=1 [ m
i
ng(o)e x
O
)
{11i
-
!
s(O
)
11
2}
m
i
ng(l
)e x
[t
)
{
l
1i
-
!
S(
I)
In ]
(6)
B. Performance Gap
As stated previously, the simplifcation of LLR
combining algorithm can lead to some performance
degradation compared to the direct method. We measure
this gap in CTC in various modulation to verif that the
performance gap can be negligible compared to CTC's
coding gain. The results is shown in Table 1.
Table 1. Proposed vs Direct LLR calculation (CTC-1I2)
Modulation Avg. Performance Degradation
4-QM 0.1 dB
16-QM 0.15 dB
64-QM 0.2 dB
C Computational Compleit
The conventional LLR combining scheme directly
calculate LLR fom (3) and use Convolutional Coding
(CC), which is popular channel coding scheme with low
complexity implementation. The use of approximation as
we described in previous analysis will cause performance
degradation, especially in the channel decoder output. To
overcome this, we applied Convolutional Turbo Coding
(CTC) which has higher complexity but offers better
decoding performance and more robust to suboptimal
decoding implementation [10]. CTC is also a mandatory
feature for WiMAX System Profle Certifcation Test
[11]. The comparison of direct implementation with CC
and simplifed implementation with CTC, both using
memory order M = 2 [13], is presented below.
Table 2. Complexity of LLR combining (direct) with CC
Convolutional Coding (CC) with Viterbi
Process P Equivalent Addition
Branch Metric Calc 6(2AM)
Path Metric Calc 4(2AM)
Hard Decision O
Overall Complexity 10(2JM)+3
HARQ LLR Combining (direct) per Nt Symbol processed
Process P Equivalent Addition
Distance Calculator 6O
Normalization 2
Constellation LUT 12
QR-Decomposition 45 (SLC only)
Direct LLR Combining with CC M = 2
Total Operation 43 +l17 + 10K)Nt
4-QM; 6 = 4; Nt = 2 DJ
16-QAM; 6 = 16; Nt = 2 l OJ
64-QM; 6 = 64; Nt = 2

ODJ

Table 3. Complexity of LLR combining (proposed) with CTC


Convolutional Turbo Coding (CTC) with Max-Log-MAP
Process P Equivalent Addition
Branch Metric Calc 12(2AM)
Path Metric Calc (BWD) 4(2AM)
Path Metric Calc (FWD) 4(2AM)
Soft Decision 8(2AM)-3
Overall Complexity 28(2AM)-3
HARQ LLR Combining (simplified)' per Nt Symbol processed
Process P Equivalent Addition
Distance Calculator 16
Normalization 2
Constellation LUT 12
QR-Decomposition 45 (SLC only)
Simplified LLR Combining with CTC M - 2
-
Total Operation I5 +l3QNt
4-QAM; b = 4; Nt = 2 169
16-QAM; b = 16; Nt = 2 169
64-QAM; b = 64; Nt = 2 l 169
4. SIMULATION RESULTS
In this chapter, we will compare the performance of
direct and simplifed implementation of the LLR
combining HARQ. We consider parameters as in Table 3
for our simulations, assuming perfect channel estimation
and synchronization in the receiver.
Table 5. IEEE 802.16e Simulation Parameter
BfBDBIBf NBUB5
FFT Size 1024
Cyclic Prefix 1/8
Frame Duration (TOO) 5ms
Sampling Factor 28/25
Subcarrier Spacing 102.86I s (10.93 kHz)
Carrier FreQuencv 2.3 GHz
Packet Size 84 Byte
Channel Speed
ITU-B Pedestrian 3 km/hr
ITU-A Vehicular 60 km/hr
Fig. 3 depicts packet error rate (PER) against SNR
performance in low mobility (Pedestrian-B) for 64-QAM
2010 Interational Symposium on Intelligent Signal Processing and Communication Systems (lSP ACS 2010) December 6-8, 2010
and high mobility (Vehicular-A) channel for 4-QAM and
16-QAM modulation. The simplifed implementation
results will be shown in solid lines and direct
implementation results in dotted lines.
Q

O
e
w

-
u
m
O
4P| V6PbK0I
T
0
..
T
,
,

,
,

1
,
,

Z J 4 5 b
C [0C]
!b| V6P bK0I
T

dt

'
'
*
\
\
\
\
\
E
T
3
^
! TZ T4 !b
C [CL]
b4P| |60L JKI
Tb T Z ZZ
C [CL]
-CIml6 LLL*L1L
--CImpl6 CLL*L1L
--CImpl6 CLL*L1L
--- LI|6C! LLL*LL
--- LI|6C! LL*LL
-+- LI|6C! CLL*LL
Fig. 3. Comparison of HRQ Implementation Results
In 4-QAM modulation, PER around 1 % or 10-
2
is
achieved by the simplifed DLC scheme when SNR is 3
dB in high mobility. To achieve the same PER, the direct
DLC with CC needs 5 dB SNR. The trends continue in
higher modulation, for example 16-QAM, when PER less
than 10-
2
is achieved when SNR is 11 dB and 16 dB by
simplifed and direct method, respectively. The SNR
margin is mainly created by CTC implementation as
channel decoder. Not only compensate the performance
gap of the simplifed scheme, CTC also give better
performance, although increase complexity in ret. This
paradox is careflly exploited by the proposed simplifed
scheme, which has signifcant advantage along with
increasement in modulation order.
By using direct implementation scheme, the DLC and
SLC can achieve same performance results, while BLC
has signifcant performance gap as previously analyzed in
chapter III. However, the case is slightly diferent in
simplifed implementation scheme. Because of the
approximation, the performance of DLC and SLC is not
exactly the same, while the performance gap of BLC is
smaller than when direct implementation is used. In all
cases, the best performance is given by the DLC scheme,
followed by the SLC and BLC scheme. Nevertheless, the
SLC still has advantage in terms of buffer size [5] and
BLC can also be used for HARQ-IR. Therefore, the best
implementation scheme still depends on how the systems
will be deployed.
5. CONCLUSION
This paper proposed simplifed implementation of
MIMO-HARQ with LLR combining method. With simple
approximation and by using Convolutional Turbo Coding
(CTC) as channel coding scheme, the proposed scheme
can achieve better results in term of packet error rate than
direct implementation scheme using Convolutional
Coding (CC). The proposed scheme also has lower
complexity, especially in large modulation size such as
16-QAM and 64-QAM. Better results theoretically can be
achieved with pairing CTC with direct implementation of
the LLR combining. But it is not practical because the
complexity of receiver will be increased signifcantly, thus
making the proposed scheme becomes better choice for
implementation.
REFERENCES
[1] S. Lin, D. J. Costello. Jr., and M. 1. Miller, "Automatic
repeat-request error-control schemes," IEEE Commun.
Mag., vol. 22, Dec. 1984.
[2] D. Chase, "Code combining - a maimum-likelihood
decoding approach for combining an arbitrary number of
noisy packets," IEEE Trans. Commun., May 1985.
[3] D. Toumpakaris, J. Lee, A. Matache, and H. Lou,
"Performance of MIMO HARQ under receiver complexity
constraints," IEEE GLOBECOM, 2008.
[4] IEEE Std 802.l6-2004/Corl-2005,Amendment 2, Feb. 2006.
[5] E. Jang, 1. Lee, H. Lou, and J. M. Ciof, "On the combining
schemes for MIMO systems with hybrid ARQ," IEEE
Transactions on Wireless Communications, Feb. 2009.
[6] WiMA Forum Mobile System Profle, 2008.
[7] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space
Time Wireless Communications, Cambridge Press. 2003.
[8] E. N. Onggosanusi, A. G. Dabak, Y. Hui, and G. Jeong,
"Hybrid ARQ transmission and combining for MIMO
systems," in Proc. IEEE International Conrence on
Communications 2003, May 2003.
[9] J. Lee, D. Toumpakaris, E. W. Jang, H. Lou, and 1. M.
Ciof, "Transceiver design for wireless systems via MIMO
Hybrid ARQ," IEEE Communications Magazine, vol. 47,
pp. 32-40, Jan. 2009.
[10] A. Salbiyono, et. aI., "CTC Decoder for Mobile WiMA
with HARQ Support," Proc. The 5th International
Conference TSSA, 2009.
[11] A. Vasquez, E. Antelo, "Implementation of Exponential
Function in Floating Point Unit," Journal of VLSI Sig.
Processing, Springer, 2003.

You might also like