Anda di halaman 1dari 4

WK,(((,QWHUQDWLRQDO&RQIHUHQFHRQ&RPPXQLFDWLRQ7HFKQRORJ\3URFHHGLQJV

Prediction of Hybrid-ARQ Based on Mutual


Information Model for LDPC coded OFDM system

Xiang Chen, Zesong Fei, Jingming Kuang, Wei Sun


RCDCT, Dept. of electronic engineering
Beijing Institute of Technology (BIT)
Beijing, P. R. China
xiaxiang@bit.edu.cn, feizesong@bit.edu.cn

Abstract—An accurate and simple link quality model is of utmost parity bits is predicted by MI model, and the proper coding rate
importance in link adaptation (LA) design. Our previous work in is chosen to carry out the retransmission process.
[1] has verified the good accuracy of the mutual information (MI)
model for rate compatible (RC) LDPC coded OFDM system. In The rest of the paper is organized as follows. The MI model
this paper, the MI model is used to predict the retransmission is described in detail in section II. We present the new RC-
length T of two hybrid ARQ schemes, partial chase combining LDPC family and look-up table (LUT) of the MI model. We
(PCC) and incremental redundancy (IR). Simulation results also propose the method to predict the retransmission length T
verified that the one time retransmission satisfied a certain target for PCC and IR with MI model in section III. Simulation
BELR with predicted T, i.e. the needed minimum retransmission results verify the accuracy of MI model’s prediction in section
length. Thus the MI model can serve as an accurate prediction IV. Section V concludes this paper.
model for high spectrum efficiency hybrid ARQ design of LDPC
coded OFDM system.
II. MUTUAL INFORMATION MODEL
Keywords-mutual information model; Hybrid ARQ; partial The MI model serves not only as a link to system interface,
chase combining; incremental redundancy but a convenient LUT for LA. Fig. 1 shows the diagram of the
MI link quality model, including two separate sub-models: the
I. INTRODUCTION modulation sub-model and the coding sub-model. They are
independent, the modulation sub-model merely relates to the
Nowadays, data service becomes strong support of global demodulation algorithm and modulation order, while the
telecommunication industry. Link adaptation (LA) [2] is the coding model only relies on AWGN performance of the
key technology to improve system throughput and spectrum decoding algorithm, coding rate and block size.
efficiency. Taking the HSDPA system for instance, hybrid
ARQ and automatic modulation coding (AMC) are two
A. The Modulation Model
effective LA technologies to reduce delay and to raise peak
data rate. Compared with AMC, hybrid ARQ works in What can be observed in fig. 1 is the parameter passed from
relatively smaller range of channel state without changing the modulation model to coding model is modulated symbol-level
modulation and coding scheme (MCS). mutual information (SI) of all the sub-carriers, each SI
corresponds to a channel state during one coding block. For M-
Hybrid ARQ technology syncretizes ARQ and FEC to ary modulation, with modulation order m, the SI of the channel
reach higher throughput with lower delay. There are two types symbol signal-to-noise ratio (SNR) value γ is defined as
of hybrid ARQ schemes. One is chase combining (CC) [3],
with which the transmitter resends the same packet as the
previous one and the receiver adds up the soft output packets of
the demodulator to get performance gain. The other is
incremental redundancy (IR) [4], with which the transmitter
sends additional parity bits and the receiver combines them
together to obtain new codes of lower coding rate.
Note that when bit error rate is not very high, it is not
necessary for CC to retransmit the entire packet. Thus we
proposed partial CC (PCC), with which only part of the
previous packet is retransmitted. Moreover, the length of the
retransmitted block is alterable and can be predicted by mutual
information (MI) model [5, 6]. For FEC whose coding rate can
be changed flexibly by rate compatible (RC), similar strategy
could be applied to IR. In this paper, we did a little modify to Figure 1. Diagram of the MI link quality model structure
the RC-LDPC codes in [1]. The length of need additional

This work is sponsored by Sino-Swedish IMT-Advanced Cooperation Project


(No.2008DFA11780), and the national science foundation of China under
Grant (No.60502026), and Ericsson.
Contact author: Zesong Fei, feizesong@bit.edu.cn
üüüüüüüüüüüüüüüüüü
978-1-4244-2251-7/08/$25.00 ©2008 I E E E


Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.
WK,(((,QWHUQDWLRQDO&RQIHUHQFHRQ&RPPXQLFDWLRQ7HFKQRORJ\3URFHHGLQJV

RBIRMI = RBI MI / N (3)


6
BPSK
QPSK
16QAM
RBIR is mapped into such link level metrics as block error
5
64QAM probability (BLEP) and frame information (FI). BLEP could be
Symbol-level information

4
considered as instantaneous block error rate (BLER), and FI is
received decoded bit information within one coding block, i.e.
3
FI ({γ j , j = 1 ~ J }) = (1 − BLEP ) ⋅ K (4)
2
Lei Wan, S. Tsai and M. Almgren have proved in [6] that
1
for Turbo codes, the relationship between RBIR and the
decoding performance for a given codec only depends on the
0 coding rate, and the mapping between RBIR and the quality
-20 -15 -10 -5 0 5 10 15 20 25 30
Es/N0 [dB] indicators, which is independent to the channel condition, so it
can be obtained by the AWGN simulation results.
Figure 2. Symbol-level mutual information of different modulation schemes
Seen from the system-level simulator, the instantaneous
­ ½ signal to interference and noise ratios (SINRs) are sent to the
° P (Y | X , γ ) ° MI model and the BLEP performance is returned, so the
SI (γ , m) = E XY ®log 2 ¾
° ¦X P ( X ) P (Y | X , γ ) ° (1) throughput value can be obtained. Taking an OFDM system
¯ ¿ with only one MCS for instance, the throughput is
­ +∞ +∞ ½
° P (Y | X , γ ) ° N used ⋅ R ⋅ M ⋅ (1 − BLEP )
= EX ® ³ ³ P (Y | X , γ ) ⋅ log 2 dYI YR ¾ Throughput = (5)
°YR =−∞ YI =−∞ ¦X P ( X ) P (Y | X , γ ) ° TOFDM
¯ ¿
Where the modulated symbol X belongs to a certain Where N used is the number of used data sub-carriers. R is
modulation constellation, and the received symbol the coding rate. M is modulation order, TOFDM is the duration
Y = (YR + i * YI ) . We assume P ( X ) = 1/ M as the a priori of an OFDM symbol.
probability of X in this paper. P(Y | X , γ ) is the probability
density function of Y conditioned on the noise-free channel III. PREDICTION OF HYBRID ARQ USING MI MODEL
symbol X and parameterized by channel state γ .
A. RC LDPC codes and MI LUT
With regard to the modulation model, fig. 2 gives the SNR- We continue to use the extending check matrix H scheme
to-SI mapping. What can be seen is how much mutual to get the lower coding rates than the rate-0.5 mother code. Yet
information a modulated symbol carries through at different we use block puncturing [7] to get the higher coding rates. The
channel state under certain modulation order. Obviously, the RC-LDPC code family used here is listed in table I.
maximum SI of each modulation scheme is equal to its
modulation order. What’s more, at higher SNR range, e.g. Our previous work in [1] has confirmed that the MI model
above 15dB, a symbol transmitted by modulation schemes with is capable of unifying BLEP performance of various
higher modulation order will deliver more SI, namely higher modulation schemes and diverse sub-carrier states, and even
modulation order is more spectral efficient. On the other hand, different coding block length that the IEEE802.16e standard
at lower SNR range, e.g. below 0dB, a modulated symbol sent specified. Thus it will be accurate enough to implement LA
by modulation modes with lower modulation order will with two LUTs, one is built according to the BLEP vs. RBIR
transmit the same SI as higher modulation scheme, thus each performance of BPSK modulation and different coding rates
coded bit will carry more normalized-SI, defined as SI divided over AWGN channel, as fig. 3 shows. The other is the RBIR-
by the modulation order, which implies that lower modulation to-SNR mapping table based on (1), (2) and (3).
order induces robustness. Thus higher modulation order has
more spectral efficiency at the cost of loss in robustness. TABLE I. RATE COMPATIBLE LDPC CODE FAMILAY

B. The Coding Model Mother Rate compatible schemes


code (N, K) Extending H (N,K) Block puncturing (N,K)
The coding model consists of two parts: the SI
rate 0.4, ( 720, 288) rate 0.55,(524,288), rate 0.6,(480,288)
collection/correction unit and the quality mapping unit. The SI rate 0.5, rate 0.3, ( 960, 288) rate 0.65,(443,288), rate 0.7,(412,288)
collection/correction unit first collects the received coded bit (576,288) rate 0.2, (1440, 288) rate 0.75,(384,288), rate 0.8,(360,288)
information (RBI) among the J symbols, whose SNR values rate 0.1, (2880, 288) rate 0.85,(339,288), rate 0.9,(320,288)
are {γ1, γ 2 ,..., γ J }, with the modulation order {m1 , m2 ,..., mJ } :
J
RBI MI = ¦ SI (γ j , m j ) (2)
j =1

It is normalized by the number of total coded bits N to the


received coded bit information rate (RBIR) by Figure 3. Look up table of the MI model



Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.
WK,(((,QWHUQDWLRQDO&RQIHUHQFHRQ&RPPXQLFDWLRQ7HFKQRORJ\3URFHHGLQJV

Figure 4. Hybrid ARQ prediction based on the MI link quality model

Figure 3 illustrates the prediction of retransmission length T


of hybrid ARQ based on the MI model. If the first transmission
is failed, the RBI value RBI1 is calculated by (1) and (2) based
on the sub-channel states Ȗ first . The needed RBI RBI _ com for
a certain target BLEP can be found in the LUT as shown in fig.
3. Retransmission is to compensate for the lacking RBI,
ΔRBI = RBI _ com − RBI1 for target BLEP. Accordingly,

T = Δ RBI / RBIRretrans (6)

Where RBIRretrans can be found in LUT based on the Figure 6. Retransmission rate R_need prediction of IR using the MI model

channel states of retransmission Ȗ retrans .


IV. SIMULATION RESULTS
B. Partial Chase Combining To evaluation the accuracy of MI prediction model for the
Retransmitted block is a part of first transmitted block. As two hybrid ARQ schemes, PCC and IR, we built a simple link
shown in fig. 5, N is the total block length. The first T bits are level simulation platform. The simulation parameters are listed
transmitted two times on different channel states. From the in table II. Notice that even the actual BLER of first
point of receiver, the T bits are equivalent to be send one time transmission is lower than the target BLER, the first
on at an effective SNR, at which the transmitted SI transmission may still be failed, but the calculated T will be a
SI _ com = SI1 + SI 2 , SI1 and SI 2 are corresponding SI delivered minus value. To avoid this case, a minimum value of T is set.
at SNR1 and SNR2 respectively. The other (N-T) bits Figure 7 and 8 presents the result of MI prediction for PCC.
transferred SI_1. The transmitted RBI is RBI _ com computed When SNR is lower than -3.8dB, the target BLER still cannot
by (2). T is big enough as long as RBI _ com ≥ RBI _ target , be reach even after retransmission. That is because the
retransmission length has reached 576 and the maximum
RBI _ target is the needed RBI for decoding and is found by LUT. retransmission time is set as one. More retransmission times
are needed to get better performance. When SNR is between
C. Incremental Redundancy -3.8dB and -1.4dB, the actual BLER is always below and next
For IR, the coding rate cannot change continuously due to to target BLER. And T is changing from 576 to 100. That is to
the limitation of RC-LDPC family. However, the proper say, the minimum needed retransmission length predicted by
candidate coding rate can be found which is lower and nearest the MI model satisfied the BELR requirement of the system,
to the target coding rate. As fig. 6 illustrates, RBIR _ need is thus system throughput is enhanced as well. When SNR is
higher than -0.9dB, T is equal to 100, the set minimum
found in LUT for target BLEP. If the first transmission is
retransmission length. That means the BLER of original coding
failed, the nearest candidate coding rate that can offer
rate is below target BLER and no retransmission is needed in
RBIR _ com ≥ RBIR _ need will be chosen as the actual rate after fact.
retransmission, T = K /( R _ com − R first ) , R first is the initial rate.
TABLE II. SIMULATION PARAMETERS

Aspect Working Assumption


Type of Hybrid ARQ PCC, IR
Maximum Retransmission Times one
Minimum Retransmission Length 100 bits
Target BLER 0.01
Candidate LDPC Coding Rate 0.1, 0.2, 0.3, 0.4, 0.5, 0.55, 0.6, 0.65,
0.7, 0.75, 0.8, 0.85, 0.9
Modulation Schemes BPSK, QPSK, 16QAM, 64QAM
Figure 5. Retransmission length T prediction of PCC using the MI model Channel Type AWGN



Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.
WK,(((,QWHUQDWLRQDO&RQIHUHQFHRQ&RPPXQLFDWLRQ7HFKQRORJ\3URFHHGLQJV

Figure 7. BLER vs. SNR curve of the MI model predicted PCC, Figure 9. BLER vs. SNR curve of the MI model predicted IR,
rate-0.5 LDPC code, BPSK, AWGN rate-0.5 LDPC code, BPSK, AWGN

600

550

500
the number of retransmitted bits T

450

400

350

300

250

200

150

100
-6 -5 -4 -3 -2 -1 0
Es /N0 (dB)

Figure 8. Predicted retransmission length of the MI model for PCC, Figure 10. Predicted predicted coding rate of the MI model for IR,
rate-0.5 LDPC code, BPSK, AWGN rate-0.5 LDPC code, BPSK, AWGN

Figure 9 and 10 illustrates the result of MI prediction for IR.


The first transmission uses rate-0.9 LDPC code. When SNR is REFERENCES
between -0.9dB and 5.5dB, one time retransmission is enough [1] X. Chen, L. Wan, Z. Fei, et al "The Application of the EESM and MI-
and different coding rate is chosen for finer SNR range within based Link Quality Models for Rate Compatible LDPC Coded OFDM
[-0.9dB, 5.5dB], as fig. 10 shows. The corresponding BLER System" IEEE VTC 2007-fall, IEEE 66th, page(s):1288-1292, 30 Spet.-
3 Oct. 2007.
performance is shown with solid curve in fig. 9, which is
[2] Wang Hai, Wan Lei, Liao Min. HSDPA Link Adaptation Based on
zigzag and is below target BLER within [-0.9dB, 5.5dB]. It Novel Quality Model. Vehicular Technology Conference, 2005. VTC
implies that the sub-optimal retransmission length T is obtained 2005-Spring. 2005 IEEE 61st, Volume 1, 30 May-1 June 2005, Page(s):
rather than the optimal one. That’s because the coding rate gap 334 - 338.
is not fine enough. If the coding rate gap is much smaller than [3] P. Frenger, S. Parkvall, E. Dahlman, "Performance comparison of
0.05, the solid curve will be flat and slightly below the target HARQ with Chase combining and incremental redundancy for
BLER within [-0.9dB, 5.5dB], just like the star-marked curve HSDPA", IEEE VTC 2001 Fall. Volume 3, Page(s):1829-1833, Oct.
2001
within [-3.8dB, -1.4dB] in fig. 7. When the minimum coding
[4] Jung-Fu Cheng, "On the coding gain of incremental redundancy over
rate is reached, one time retransmission is not enough, e.g. the chase combining", IEEE GLOBECOM '03. Vo. 1, page(s): 107 - 112,
SNR range below -0.9dB in fig.9, the actual BLER is always Dec. 2003.
higher than target BLER. [5] S. Tsai and A. Soong, "Effective-SNR Mapping for Modeling Frame
Error Rates in Multiplestate Channels," 3GPP2-C30-20030429-010, Apr
2003.
V. CONCLUSIONS
[6] L. Wan, S. Tsai, and M. Almergn, "A fading-Insensitive Performance
The MI model is used to predict the minimum Metric for a Unified Link Quality Model," IEEE WCNC, Vol.4, pp.
retransmission length T of RC-LDPC codes for partial CC and 2110-2114, Apr 2006.
IR schemes. The predicted minimum retransmission length [7] ZTE. Structured LDPC coding with rate matching. 3GPP TSG RAN
satisfies the target BLER requirement, which confirms that the WG1 #44 R1-060499. Denver, USA, February 13-17, 2006
MI model provides accurate and simple look up table for
Hybrid-ARQ design of LDPC coded OFDM system.



Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 06:53 from IEEE Xplore. Restrictions apply.

Anda mungkin juga menyukai