You are on page 1of 4

Analysis of HARQ Schemes Using Reed-Solomon Codes

K. Kotuliaková and J. Polec


Department of Telecommunications
Slovak University of Technology
Ilkovičova 3, 812 19 Bratislava, Slovak Republic
Phone: (+421) 2 68279-604 Fax: (+421) 2 68279-601 E-mail: kkotul@ktl.elf.stuba.sk

Keywords: Automatic-repeat-request (ARQ), throughput, forward-error-correction (FEC), hybrid ARQ,


Reed-Solomon code (RS), Land Mobile Satellite channel (LMS).

Abstract - Presented paper deals with the problem of successfully _ received _ blocks
ηB = (3)
using ARQ (Automatic Repeat Request) schemes. In all _ sent _ blocks
particular, we focus on the subset of Hybrid ARQ
methods. We analyse the throughput performance of And the relation between bit and block throughput is:
pure and hybrid ARQ techniques, further on we
m
present the throughput analysis of hybrid ARQ ηb = ⋅η B (4)
schemes in Binary Symmetric Channel (BSC) and n
Land Mobile Satellite (LMS) channel environment
based on simulation results. Forward error correction where m is length of information field and n is whole
(FEC) used in hybrid ARQ schemes is provided by length of transmission block.
Reed-Solomon (RS) codes.

2. ERROR CONTROL TECHNIQUES


1. CHANNEL MODEL
2.1 ARQ (Automatic repeat request)
Digital transmission channel is the most erroneous part
of the system. But in most cases the precise analytical ARQ schemes belong to feedback error control
description of the causes which lead to error occurence is techniques because they use feedback channel to ensure
not available. Usually the transmission channel is the error-free transmission. Basic outline of ARQ schemes
accepted as entity and the process of error generation is could be: transmitter sends coded block and awaits ACK
modelled statistically. from receiver, if receiver detects error in received block
By using the channel model with independent error asks for retransmission, whole block is retransmitted, only
occurence we assume that every error is generated detection codes are used to keep system complexity low.
independently, in other words the channel has no memory In literature ARQ schemes are divided into three groups
and every binary symbol is received corrupted with equal [1],[2],[3]:
probability Pb. One of these models is Binary Symmetric
Channel (BSC). The block error rate can be expressed as • Send-and-wait (SW) scheme
[2]:
• Go-back-N (GBN) scheme
Pe = 1 − (1 − Pb ) n
(1) • (ideal) Selective repeat (ISR) scheme
where n is block length in symbols. Likewise we can Relative throughput of three pure ARQ schemes is defined
express the probability of event that block contains m - as [1],[3]:
errors [2]:
1 − Pe
⎛n ⎞ η B − SW = (5)
m
P (m, n) = ⎜⎜ ⎟⎟ ⋅ Pb ⋅ (1 − Pb ) n − m (2) 1+ S
m
⎝ ⎠
1 − Pe
Channel models for satellite and mobile systems (i.e. η B −GBN = (6)
Land Mobile Satellite channels) belong to category of 1 + S ⋅ Pe
models with independent error occurence.
η B−SR = 1 − Pe (7)
1.1 Relative throughput where S stands for RTD (round trip delay) - delay of
acknowledgements in number of blocks that can be sent
We will use relative throughput as a key measure to until ACK is received.
compare different ARQ methods and to describe their
efficiency. According to its physical nature is block
throughput defined as [3]:
2.2 FEC (forward error correction) N
⎛N⎞ i
PT = ∑ ⎜⎜ i
i =t +1⎝
⎟⎟ PS (1 − Ps ) N −i (8)
Basic facts about FEC techniques are: ⎠
• Transmitter sends coded data blocks continuously. where t is error-correction ability and PS is probability of
reception of corrupted symbol.
• Receiver tries to discover and repair errors that
incurred on transmission path. P s = 1 – ( 1 – P b )b (9)
Which means that ideal FEC techniques don’t require Average number of bits transmitted until data block is
retransmissions of blocks and therefore their throughput correctly received can be expressed:
remains constant independently from the state of the
E{N} = L + PR . E{N} (10)
transmission channel. Although the FEC techniques seem
to be ideal for data transmission, they don’t provide where L is number of bits in block and PR is probability of
satisfactory results especially in systems that require error- retransmission. Assuming that L = N.b and PR = PT :
free transmission. Their failure is based mainly on
N ⋅b (11)
following facts: firstly, no matter how strong error E{N } =
correction code we use it can still repair only finite number 1 − PT
of errors and secondly, we can’t ensure the stability and
Performance (throughput) of RS codes can be expressed
quality of transmission path (mobile and satellite systems
as:
in particular).
Hybrid ARQ techniques, which combine the preferences K ⋅ b K ⋅ b ⋅ (1 − PT )
of both types ARQ and FEC offer solution to these η RS = = (12)
problems. E{N } L
and using previous statements converted into:
2.3 Reed-Solomon Codes
K ⎡ N
⎛N⎞ ⎤
They are considered the best block error correcting codes η RS = ⋅ ⎢1 − ∑ ⎜⎜ ⎟⎟ PSi (1 − Ps ) N −i ⎥ (13)
N ⎣ i =t +1⎝ i ⎠ ⎦
for given N and K (where such code exists), in other words
they have the best error correcting capability for given Fig.1 shows how code parameters affect the
code rate [4],[5]. N is codeword length in symbols and K performance of RS code. The more powerful code we use
stands for number of information symbols in codeword. (lower code rates) the longer the throughput stays constant
RS codes are subset of non-binary BCH codes and are with rising error rate. On the other hand by using powerful
defined in Galois field GF(q), where q is prime number or correction code the maximum throughput falls down.
power of prime number. Codeword length is N = q - 1
symbols whereas one symbol is q-nary and mapped into b
– bits, q = 2b. RS codes are described with (N, K, q) and
code rate is rc = K / N. It can be proven [4] that:
• Minimal codeword distance is dmin = N – K + 1.
• Number of repairable errors is t = ( dmin – 1 ) / 2
= ( N – K ) / 2.
• For given K, N there is no other code with higher t.
It is also known [4] that RS codes are especially suitable
for bursts of errors, which means that given code can
repair t bit errors if there is only single-bit error in each
from t symbols, but it can also repair up to t.b bit errors if
every bit from t symbols is corrupted.
Disadvantage of RS codes (block codes) is that if more
that t errors occur in codeword, decoder either detects it
Fig. 1. Performance (throughput) comparison of RS codes.
and declares decoding failure, or codeword is so corrupted
RS1– ( 1023, 921, 210 ) rc=0,9 t=51; RS2 – ( 511, 307, 29 )
that it converts to space of another codeword and decoder rc=0,6 t=102; RS3 – ( 255, 77, 28 ) rc=0,3 t=89.
”repairs” it into wrong codeword (decoding error).
Therefore it is recommended to secure every codeword 2.5 HARQ ( hybrid ARQ )
with additional protection (checksum or CRC).
There are two main types of HARQ schemes: type I and
2.4 Performance of RS codes type II. HARQ type I contains additional symbols for error
detection and correction in transmission block. If
Probability of RS decoder anomaly (PT) including both corrupted symbol count is lower or equal to maximum
decoding failure and decoding error [4]: count of repairable errors t, block is repaired and accepted
by receiver as correct one. If error count is higher than t,
block is rejected by receiver and retransmission is
demanded. Properly chosen error-correcting code
significantly reduces the number of retransmissions
compared to pure ARQ mode. Disadvantage of this
technique is that non-repairable blocks are discarded by
decoder although they still can contain some useful
information.
The HARQ scheme type II, unlike type I, doesn’t
discard corrupted blocks but leaves them for further
processing. During retransmission only stronger protection
part of previous corrupted block is sent, not whole block
is retransmitted. Decoder can use these protection symbols
to repair the stored corrupted block. So it preserves system
resources such as bandwidth and processing time.
However type II techniques require difficult and
sophisticated implementation.
Fig. 2. Comparsion of ARQ & HARQ schemes, block
length = N = 255 symbols, RS - ( 255, 229, 28 ), rc = 0.9, t
= 13. RTD (S) = 1block.
3. THROUGHPUT ANALYSIS OF HARQ
On the basis of previous results we focus on HGBN and
We analyze the theoretical relative throughput of HARQ HSR schemes. Fig.3 reflects more realistic scenario
using the knowledge of throughput of pure ARQ and because we used RTD = 20 blocks. At this state of delay
error-correcting capability of RS codes. the throughput of (H)SW falls too low so its further
analysis is not interesting. Fig.3 shows that throughput of
3.1 Throughput in BSC pure GBN falls down rapidly with rising delay, on the
other hand rising delay has only minimal influence on
We use the formulas for pure ARQ schemes (5), (6), (7) throughput of hybrid GBN and throughput is similar to
and as probability of wrong block reception Pe (1) we use HSR. These observations implicate that in praxis the most
the probability of RS decoder anomaly PT (8). As a result efficient solution would be using Go-Back-N technique
we get the formulas of theoretical relative bit throughput combined with suitable error-correcting code (i.e. RS). We
of three basic hybrid ARQ schemes ( type I ) in channel receive relatively high throughput in wide range of BER
with independent errors occurrence (BSC): while maintaining the affordable processing complexity of
GBN (unlike SR, which has high buffer memory
requirements).
N N
⎛ ⎞
1− ∑⎜⎜ ⎟⎟PSi (1− Ps ) N −i
i K
ηb−HSW = i=t+1⎝ ⎠ ⋅ (14)
1+ S N

N
⎛N⎞
1 − ∑⎜⎜ ⎟⎟PSi (1 − Ps ) N −i
i =t +1⎝ i ⎠ K
ηb−HGBN = ⋅ (15)
N
⎛N⎞ N
1 + S ⋅ ∑ ⎜⎜ ⎟⎟PSi (1 − Ps ) N −i
i =t +1⎝ i ⎠

⎡ N
⎛N⎞ ⎤ K
η b− HSR = ⎢1 − ∑ ⎜⎜ ⎟⎟ PSi (1 − Ps ) N −i ⎥ ⋅ (16)
⎣ i =t +1⎝ i ⎠ ⎦ N
Fig. 3. Comparsion of ARQ & HARQ schemes, block
length = N = 511 symbols, RS - ( 511, 307, 29 ), rc = 0.6, t
Fig.2 shows the throughput of HARQ schemes. The
= 102. RTD (S) = 20 blocks.
(H)SW scheme’s performance is much lower than
(H)GBN and (H)SR. The interesting part is that
performance of hybrid GBN is very similar to hybrid SR
In further analysis we focus on hybrid Go-Back-N
scheme’s while throughput of pure GBN is much lower
scheme.
than throughput of pure SR scheme.
3.2 Throughput in LMS channel 4. CONCLUSION

We aim on the worst case of LMS channel and that is its The scope of this paper was to offer basic view on how
urban version. Model is based on the fact that in urban throughput of main ARQ schemes combined with error-
zones direct path between transmitter and receiver is correcting codes is affected. As a result of analysis and
blocked, therefore transmitted signal is diffused. Direct simulation it can be stated that carefully chosen error-
wave is completely attenuated and receiver receives correcting code can strongly improve the performance of
reflected signals from all directions and phases of these ARQ technique by reducing the number of
signals are randomly shifted. It is called Rayleigh fading retransmissions. Go-Back-N scheme combined with RS
code was found to be the most attractive having considered
channel too.
its satisfactory throughput performance and low
Simulations were accomplished using the simulation
complexity. As an example we can take HGBN and RS
program described in [2], enriched with new module code with rc = 0.6 in BSC channel. When using this code
allowing simulation of HARQ schemes using RS codes. the transmission is at BER=4% still very well possible
Simulation assumes: compared to situation without RS code when at
• Feedback channel is error-free, ACK are always BER=0.1% transmission already becomes impossible. In
LMS-urban channel the SNR gain using discussed scheme
received correctly.
is at least 10 dB and in LMS-open channel SNR gain is
• BPSK modulation is used for transmission. 6dB.
• Ideal detection code is used. If there is an error in
transmitted block it is always detected.
5. ACKNOWLEDGEMENT
• LMS channel model with independent error
occurrence is used. Research described in the paper was financially
supported by the Slovak Research Grant Agency (VEGA)
Tab.1 shows the parameters of three RS codes used with under grant No. 1/0883/08 and grant No. 1/4084/08.
GBN in simulation.

REFERENCES
N K rc t [errors] block size
[bit] [1] J. Polec, T. Karlubíková, Stochastické modely
511 459 0,9 26 4599 v telekomunikáciách I., FABER, Bratislava, 1999.
[2] P. Košút, "Analýza priepustností adaptívnych Go-Back-N
511 307 0,6 102 4599
schém", PhD. thesis, Slovak University of Technology,
511 153 0,3 179 4599 Bratislava, 2001.
[3] K. Kotuliaková, "Hybridné ARQ metódy v bezdrôtových
Tab. 1. Parameters of RS codes used in simulation.
komunikačných kanáloch", PhD. thesis, Slovak University
Fig.4 shows the results of the simulation. It is obvious of Technology, Bratislava, 2005.
that the SNR needed for successful transmission is [4] S. Choi, K.G. Shin, "A Class of Adaptive Hybrid ARQ
significantly higher when using pure GBN scheme Schemes for Wireless Links", IEEE Trans. on Vehicular
compared to hybrid GBN schemes. Techn., vol. 50, no.3, pp. 777-790, 2001.
[5] M. Purser, Introduction to Error-Correcting Codes, Artech
House, 1995.
0.9
GBN (pure)
[6] J. Pavlovičová, M. Oravec, M. Mokoš, M. Keleši, "Error
H-GBN (rc = 0.9)
H-GBN (rc = 0.6)
Concealment Using Neural Networks for Block-Based
0.8
H-GBN (rc = 0.3)
Image Coding", Proc. of int. conference EC-SIP-M 2005,
0.7 5th EURASIP, Smolenice, ISBN 80-227-2257-X, pp. 52-57,
0.6
July 2005.
relative bit throughput

0.5

0.4

0.3

0.2

0.1

0
5 10 15 20 25 30 35 40
SNR [dB]

Fig. 4. Comparsion of ARQ & HARQ schemes using RS


codes in LMS channel (urban), RTD = 20 blocks.

You might also like