You are on page 1of 4

th

50 International Symposium ELMAR-2008, 10-12 September 2008, Zadar, Croatia

HARQ schemes for HSDPA – analysis and simulation

Vladimír Hájovský, Kvetoslava Kotuliaková, Ivan Kotuliak


SLOVAK UNIVERSITY OF TECHNOLOGY
Ilkoviþova 3, 812 19 Bratislava
Slovak Republic
E-mail: hajovsky@ktl.elf.stuba.sk

Abstract - Presented paper concerns with HSDPA (High Speed Packet Data Access) concept, especially
describes HARQ (hybrid ARQ) schemes used to achieve high data throughput in 3G networks. Simulation is
done in AWGN channel using Soft decision Viterbi decoder for decoding convolutionally encoded data.
Comparative simulation of Stop&Wait protocol using Chase combining and Incremental redundancy is shown.

Keywords - HSDPA, HARQ, Chase combining, Incremental redundancy

1. INTRODUCTION versions separately. In other case, versions are


combined according to their puncturing patterns and
3GGP standardized three HARQ scheme types then decoded.
for HSDPA: In this article, we compare Chase Combining and
x Chase combining (CC) Incremental Redundancy HARQ schemes in terms
x Incremental redundancy (IR) of relative bit throughput. The rest of article is
x Self decodable IR organized as follows: Section 2 describes basic
principle of N-channel Stop&Wait protocol, in the
In [1], Chase showed that sequence gained by section 3, we show models of CC and IR, section 4
combination of identical data block versions comprises description and parameters of our
provides lower BER (bit error rate) than separate simulations and finally, section 5 depicts results and
sequences do. This leads to idea of storing not explanation of our simulations.
decodable copy in UE (User equipment) buffer. As
soon as next identical copy arrives, it is soft
combined with stored one. This technique of soft 2. N–CHANNEL STOP&WAIT
combining was named „Chase combining“. Stated
concept, evolved within 3GPP Rel. 5 specifications, 3GPP Rel. 99 deploys Selective Repeat (SR)
reduces delays in comparison with HARQ schemes ARQ scheme. As it is known about this scheme,
used in Rel. 99 for UMTS (Universal mobile only incorrectly received blocks are retransmitted.
telecommunications system). Transmission is ensured by Radio link control
Main disadvantage of this scheme is the (RLC) layer in Radio network controller (RNC).
requirement for the same modulation and coding There is need for identifying of each data block by
scheme (MCS) within retransmission. This fact its own sequence number. This and many other
reduces advantages gained by using adaptive requirements which HSDPA carries are out of the
modulation and coding (AMC) and fast link scope of this work and can be found in [3]. But just
adaptation. them and increasing complexity of system lead to
Subsequently, there were introduced new ARQ scheme replacement. Protocol Stop&Wait is
algorithms based on incremental redundancy (IR). easy to implement but due to its low efficiency there
IR scheme is based on achieving coding gain while was compromise between SR and S&W
adding redundant information into retransmitted data standardized called N-channel Stop&Wait scheme.
block instead of repeating same block. Before
decoding in receiver, these blocks are combined The 3 channel example is given in Fig. 1. N-
together with first sent block. This procedure channel S&W can be described as activation of N
ensures better data protection and more efficient use parallel HARQ processes, while each of them
of AMC. Disadvantage of this scheme is inability to employs separate single S&W protocol. Each HARQ
decode retransmitted versions separately. process transmits data in downlink direction (High
To soften IR disadvantage, self decodable IR speed downlink shared channel), while other
scheme has been proposed. In this case, redundant instances wait for acknowledgement (ACK) in
information is added to the first data block and uplink direction (Uplink dedicated physical control
single versions are produced by employing puncture channel)
patterns. In case of good channel conditions during
retransmission, receiver is able to decode all

557

uthorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on August 03,2010 at 12:27:28 UTC from IEEE Xplore. Restrictions apply
th
50 International Symposium ELMAR-2008, 10-12 September 2008, Zadar, Croatia

UE2
Info Encoded Transmitted Received Combined Recovered
ULDPCC A X
bits packet
R
UE1 R`
ULDPCC A N A A A 1 Error detection
NACK R` (<R)
Node-B

HS-DSCH 2
UE1 UE1 UE2 UE1 UE1 UE2 UE1 UE1 UE2

Fig. 3. Principle of Incremental redundancy


UE1 – channel 1 UE1 – channel 2 UE2 – channel 3 ACK NACK

Fig. 1. 3-channel S&W 4. SIMULATION ENVIRONMENT

As a result, N-channel S&W achieves better Our simulations are based on simulator of ARQ
performance than SR in UMTS. Simple advantage schemes and its enhancements for simulating
against SR is making transmission robust over convolutional codes. It was created and developed at
repeating errors in one of N channels. If an error KTL FEI STU in Bratislava.
occurs in one channel, other channels may keep on This simulator was primarily intended for
transmitting data. Another reason for employing simulation of pure ARQ schemes [6]. Its
S&W protocol is its simplicity, which allows using modifications enabled it for simulations of hybrid
schemes like Chase combining and Incremental ARQ schemes using Reed-Solomon codes,
redundancy. Even more MAC-hs (Medium access convolutional codes and punctured convolutional
control-high speed) layer is able to allocate all codes. We are using module SOFT-DECISION
channels for one user when radio conditions are VITERBI DECODER which works for AWGN
good. This strategy puts strict criteria on acceptable channel, codes with base rate ½ and codes derived
delays. Transmitter must be ready to resend packet from this mother code applying puncture patterns.
just N-1 intervals after previous transmission. Our modification is inserted before decoding
module while applying combining mechanism of
different puncture patterns in IR and after decoding
3. CHASE COMBINING & IR MODELS module while combining same not decodable
versions in CC.
Fundamental principles of CC and IR are
depicted in Fig. 2, Fig. 3. Original version of data 4.1 Channel model with independent error
block is numbered 1, retransmitted is marked 2. In occurrence
CC, retransmitted version is identical to original and
both of them are combined after receiving (in case This model assumes that errors origination in
when second version was not decodable without transport channel is not dependent i.e. each binary
combination). As a result SNR (Signal to Noise symbol is received in error with the same error
Ratio) is increased and final throughput is higher [2]. probability Pb, without influence of previously
received symbols. In this simulation, we are using
AWGN channel model because of its simplicity.
Information Transmitted Received Combined Recovered
bits packet
R
4.2 Relative bit throughput
1 Error detection
NACK Key value, which is used for comparison of
+
single HARQ techniques, is relative bit throughput.
2
On the physical basis, block throughput is defined
as:

Fig. 2. Chase combining principle succesfully _ received _ blocks (1)


KB
all _ transmitted _ blocks
In IR, puncturing pattern is typically different
from original version, i.e. next version carries And relationship between block and bit
redundant information. That is after receiving throughput is:
combined with original version according to agreed m (2)
Kb ˜K B
puncturing pattern. This leads to more powerful n
forward error correcting code with lower code rate
(R). Increasing of coding gain improves throughput where m stands for length of information field in the
as a result [2]. data block and n stands for entire data block length.

558

uthorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on August 03,2010 at 12:27:28 UTC from IEEE Xplore. Restrictions apply
th
50 International Symposium ELMAR-2008, 10-12 September 2008, Zadar, Croatia

4.3 FEC – used codes all bits in second version in case of unsuccessful
decoding of first one. This is the mean we created
In HSDPA convolution and turbo codes are used. 1-2-restart combining for IR. 1-2-3-restart
Base rate of turbo encoder for HSDPA is 1/3. We combining was simulated like application of CC
have available convolution encoder with base rate ½. scheme on data stored in receiver after incoming and
Codes with various rates are obtained by applying decoding of first two versions in 1-2-restart
puncturing patterns. combining scheme for IR.

4.4. Enumeration of errors 5. SIMULATION RESULTS

As stated above, we are using SOFT-DECISION In our simulations, we used S&W scheme with
VITERBI DECODER. By using it, we have access block length 3319 bits and Roundtrip delay (RTD) =
to bit level of decoded packet. If an error occurs in 4. First we applied CC strategy on pure ARQ
decoded packet, it is declared as wrong, its decoding schemes. We acknowledged dependency of RTD
is stopped and we are waiting for next packet. For and block length on relative bit throughput. Later,
simulation requirements, in case of CC, we need to we simulated SR and S&W scheme. We showed that
decode entire packet to store information about bit N-channel S&W fully replaces SR in terms of
error probability. In the next cycle, this probability is throughput.
compared bit by bit. Bit, which was decoded Primary goal of this paper is to cover simulation
correctly, is chosen. Packets are not discarded until results of CC and IR in comparison to base rate code
number of allowed versions reaches limit. and punctured code without combining.
In following simulations we will use
4.5 Modulations convolutional encoder with base rate ½. In Fig. 4,
there is depicted result of CC simulation applied on
Deployed modulations are important factor, convolutional codes (n, k, L):
which has influence on resulting channel throughput. (2, 1, 3), (2, 1, 5) and (2, 1, 7) with appropriate
In HSDPA, QPSK is used primarily. For achieving generation polynomials:
higher bit rates 16QAM is proposed. Available (7, 5 oct.), (35, 23 oct.) and (171, 133 oct.).
simulator uses BPSK modulation. For investigation where n = output bits, k = input bits, L =
of principles of CC and IR, this modulation and constraint length of code and oct. is octal value.
AWGN channel model is fully sufficient.
0.1
H-S&W, L=3
H-S&W, L=3, CC - 2x
4.6 Redundant data 0.09
H-S&W, L=3, CC - 3x
H-S&W, L=5
0.08
H-S&W, L=5, CC - 2x
Except information and parity bits, there is also H-S&W, L=5, CC - 3x
0.07 H-S&W, L=7
24 bit CRC (Cyclic redundancy check) added to
relative bit throughput

H-S&W, L=7, CC - 2x
each transmitted block. This redundancy is not taken 0.06 H-S&W, L=7, CC - 3x

in account while computing throughput. 0.05

0.04
4.7 Chase Combining
0.03

We implemented this strategy like soft decision 0.02

about block integrity by means of comparing bit by


0.01
bit. We have proposed 1-2-restart combining and 1-
2-3-restart combining scheme. Their names tell how 0
-2 -1 0 1 2 3 4 5 6 7

many versions are stored before the buffer in UE is SNR [dB]

flushed. Disadvantage of our simulator is that it is Fig. 4. Chase combining on hybrid S&W
unable to change simulated SNR while delivering
versions of the same packet. We can only rely on Blue colour depicts simulations without CC, i.e.
random simulator of Gaussian noise. with only one version for decoding in the receiver.
Red and black colour shows combinations of
4.8 Incremental Redundancy identical copies of the same data block with two or
three used versions.
Principle of this strategy is repeating differently We can see that after CC was applied, schemes
punctured versions of blocks encoded by which had originally lower correcting ability (lower
convolution code and consecutive combination in constraint length L) start to transmit data at lower
the receiver before decoding. We implemented self SNR than schemes with originally higher correcting
decodable IR. ability.
Software implementation of this strategy was Most notable is example of code (2, 1, 3) with
done by using puncture pattern in first version and three versions. Coding gain in dB for CC is depicted
by arrows with appropriate line styles. We can see

559

uthorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on August 03,2010 at 12:27:28 UTC from IEEE Xplore. Restrictions apply
th
50 International Symposium ELMAR-2008, 10-12 September 2008, Zadar, Croatia

that coding gain is decreasing with increasing violet colour shows IR applied with significant shift
constraint length. However, it is natural that more in SNR.
powerful coding schemes offer their advantages in
earlier raise of higher throughput that achieves
approximately 50% of maximal value. 6. CONCLUSION
Fig. 5 depicts behaviour of IR scheme while
using different puncturing patterns. We used In this paper, we focused on Chase combining
following patterns: {1, 0, 1, 0, 1, 0, 1}, {1, 1, 0, 1, 0, and Incremental redundancy schemes for HSDPA.
1, 0} and {1, 0}, {1, 1}. Our goal was to describe them in terms of relative
Move of margin SNR, which ensures possible bit throughput with use of different codes,
transmission is the same among differently puncturing schemes and allowed number of
punctured schemes. The only difference is retransmitted copies.
throughput, which is proportionally lower at first Our results acknowledge undoubted contribution
version and at both combinations. of CC & IR. Simulations showed that the use of the
IR compensates disadvantage of puncturing which is
0.18
increasing SNR needed for transmission. By the
H-S&W, (8, 7, 7) - punct.
0.16 H-S&W, (8, 7, 7) - punct. IR - 2x means of the IR, we can exploit advantage of high
H-S&W, (8, 7, 7) - punct. IR - 3x
H-S&W, (3, 2, 7) - punct.
bit throughput when the channel condition is good –
0.14
H-S&W, (3, 2, 7) - punct. IR - 2x
H-S&W, (3, 2, 7) - punct. IR - 3x
this is given by puncturing characteristics and also
0.12 advanteges of punctured versions combining that
relative bit throughput

carries adequate throughput during bad channel


0.1
conditions.
0.08

0.06
ACKNOWLEDGEMENT
0.04

This work has been supported by Slovak national


0.02
grant agency under the Grant No. 1/4084/07, No.
0 1/0883/08 and applied research grant.
-1 0 1 2 3 4 5 6 7 8 9 10
SNR [dB]

Fig. 5. Incremental redundancy on hybrid S&W


REFERENCES
When applying CC and IR with three versions,
IR achieves better results in terms of throughput in [1] Chase, D. 1985. Code Combining—A
comparison to results within 1-2-restart combing. Maximum-Likelihood Decoding Approach for
In the Fig. 6, comparison of best achieved results Combining an Arbitrary Number of Noisy
is depicted. Packets. IEEE Transactions on
Communications 33, no. 5:385–93 (May).
0.18
[2] Mohamad Assaad, Djamal Zeghlache TCP
H-S&W, (2, 1, 7)
0.16 H-S&W, (2, 1, 7), CC - 3x Performance over UMTS-HSDPA Systems,
H-S&W, (8, 7, 7), punct.
H-S&W, (8, 7, 7), punct. IR - 3x
2007
0.14
[3] 3GPP TS 25.308 V5.7.0 (2004-12) 3rd
0.12 Generation Partnership Project; Technical
relative bit throughput

Specification Group Radio Access Network;


0.1
High Speed Downlink Packet Access
0.08 (HSDPA); Overall description; Stage 2
(Release 5)
0.06
[4] Kacur, J., Korosi, J. An Accuracy Optimization
0.04 of a Dialog ASR System Utilizing Evolutional
Strategies, ISPA 2007, Istanbul, Turkye, Sept.
0.02
2007
0
-2 0 2 4 6 8 10
[5] Trúchly, P., Urbanoviþ, M.: MPLS Throughput
SNR [dB] over GEO satellites, ELMAR-2006, 48th
Fig. 6. CC, IR, base rate code and punctured code on International Symposium on Multimedia
hybrid S&W Signal Processing and Communications, 7-9
June 2006, Zadar, Croatia, str. 305-308
Blue colour depicts original base rate code with [6] Košút, P., Polec, J.: Investigation into optimum
maximal throughput about 0.1, red colour shows CC go-back-N ARQ strategy of Bruneel and
applied on this code with shift in SNR. Black colour Moeneclaey, Electronics Letters, Vol.36, No.4
shows punctured code without any combination, February 2000, ISSN: 0013-5194 381-382

560

uthorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on August 03,2010 at 12:27:28 UTC from IEEE Xplore. Restrictions apply

You might also like