Design of Parallel Concatenated Convolution Turbo

code using Harmony Search Algorithm

Subhabrata Banerjee
Sudipta Chattopadhyay
Department of Electronics & Communication Engineering
Department of Electronics & Telecommunication Engineering
Future Institute of Engineering & Management
Jadavpur University
Kolkata,India
Kolkata,India
[email protected]
[email protected]

Abstract–In modern communication system error-correction coding Now prophecy of convergence behavior using EXIT chart
pattern boosts sufficient redundancy which in turn upsurges the method based on mutual information has been described by
immunity to pact with the channel impairments. Turbo code can Stephan ten Brink [4] efficiently, especially in low SNR region.
deliver restrained progress in BER performance in an AWGN
The above task has been carried out by estimating value of bit
channel by varying interleaver size and free distance. In this paper,
error probability and turbo cliff location.
an optimized Parallel Concatenated Convolution Turbo code has
been proposed to improve BER performance ominously. The
As no comprehensive notional assessment has been achieved so
identical Recursive Systematic Convolutional (RSC) encoder has far, an analytical derivation [5] of EXIT charts associated with
been used to prove the enhancement in BER performance. The the convolutional code for AWGN channel has been executed
simulation result shows that the proposed Harmony-Search (HS) here. By analyzing the effect of implanted dummy variable, the
optimized Parallel Concatenated Convolution Turbo code is performance gain has been effortlessly appraised by the projected
superior in performance to the conventional Parallel Concatenated scheme.
Convolution Turbo Code. Enhancement of overall performance of Parallel Concatenated
Convolution code [6] has been done using diversity of coded
Keywords- BER, Free distance, Harmony search, Interleaver, Parallel
cooperation scheme. Enrichment of total enactment for L-relay
Concatenated Convolution Turbo code.
channel has been accomplished by transmitting the Maximal
I. INTRODUCTION Ratio Combining (MRC) sequence consisting of systematic and
To recover the preferred signal with high quality at the parity bits through sovereign fading track. Decoding has been
receiver end, redundant bits are added in decoder with the signal, performed by Maximum A-Posteriori (MAP) algorithm.
which make it more invulnerable to channel impairment. For Motivated by above research trends and challenges, an optimized
performance enhancement, numerous optimization techniques Parallel Concatenated Convolution Turbo code has been
have been introduced to redesign the system by adjusting the proposed in this paper to improve the BER performance. The
different system parameters. proposed scheme consists of identical type of component code in
Lot of research has been carried out in the field of Turbo code. parallel mode, separated by an interleaver. The new technique is
Parallel concatenated Turbo code presented by Berrou et, al [1] developed with similar type of either primitive or non-primitive
provides BER improvement either in “waterfall” or “error-floor” polynomials, optimized by harmony search algorithm [11].
region by proper arrangement of two Recursive Systematic The paper is organized in the following manner: section II
Convolution codes associated with decoder driven by feedback pronounces the system model comprising of Parallel
decoding statute. In order to achieve further up-gradation in BER Concatenated Convolution Turbo encoder and decoder; section
performance, a new parameter called effective free distance [2] III recites the problem formulation; development of the proposed
has been introduced. algorithm based on formulated fitness function is offered in
In an attempt to minimize bit error probability, encoding and section IV; simulation results and its critical analysis are abridged
decoding has been done by grouping [3] of N number of in section V; section VI includes the concluding remarks as well
interleaved information bits, preferably N=2 or 3. as some of the scopes for future work related to this paper.

978-1-4799-5912-9/14/$31.00 ©2014 IEEE

 II. SYSTEM MODEL III. PROBLEM FORMULATION
Parallel Concatenated Convolution Turbo code is constituted by The upper bound of a convolution code for AWGN channel using
two identical RSC encoders separated by a random interleaver. maximum likelihood decoding is defined by [7] [8]
Fig1a shows the general structure of Parallel Concatenated
2N
wk d 2R E
Convolution Turbo encoder with input bit Ik. The output bit
Pb (e) ≤ ∑ Q k c b (1)
consists of Z1k where output sequence consist of coded plus k =1 L N0
parity sequence and Z2k where output frame consist of permuted
version of coded plus parity sequence and I1k where coded Where d k and wk denotes the whole Hamming weight and
sequence is exactly equal to message sequence. Finally the information weight respectively.
punctured coded sequence is transmitted over AWGN channel.
By choosing all the code word with same Hamming weight, the
average information weight per code word can be defined as

Wf
wf = (2)
Nf
Here N f denotes the total number of code words and

W f represents the total information weight of all code words. So

Fig.1a. Block diagram of Parallel Concatenated Convolution the upper bound of turbo encoder using maximum likelihood
Turbo code encoder decoding is represented by
At receiver section as shown in Fig 1b., de-multiplexing has been 2m N f wf d 2 Rc Eb
done on received sequence R into sequences R1, R2 and R3
corresponding to I1k, Z1k, Z2k.The incoming signals of
Pb (e) ≤ ∑
f = f free L
Q
N0
(3)

Decoder-A are R1, R2. By observing the data, Decoder-A Now for increasing length of information sequence and for large
generates bits probabilities called extrinsic probabilities. Now value of N
output of Decoder-A is passed through interleaver and
interleaved output fed into the Decoder-B. De-interleaving Wf Δ

operation is done on extrinsic probabilities of Decoder-B and

lim = w′f (4)
N →∞ Nf
returned back to become prior probabilities to Decoder-A. The
Now the upper bound can be embodied by
process goes on till maximum number of iterations is reached or
convergence criteria is met. Decoding operation is done using 2m
d 2 Rc Eb
BCJR i.e. forward-backward algorithm which reduces decoding Pb (e) ≤ ∑ N ′ w′ Q f f
N0
(5)
complexity. f = f free

Using pseudo random interleave, mapping has been done in turbo

code from y to y/. As N f w f << N for low weight code so
N f wf
<< 1.
L
Hence for high SNR, the asymptotic recital of turbo code
approaches
Fig.1b. Block diagram of Parallel Concatenated Convolution
Turbo code decoder N free w free d free 2 Rc E b
Pb ≈ Q (6)
L N0
Where d free , N free , W free , Rc , L are the free distance, free
Define harmony memory, harmony memory
distance multiplicity associated to input frames, information bit accepting rate, pitch limit, pitch adjusting rate
multiplicity, code rate and interleaver length respectively.
Now as the BER performance of turbo code may be pretentious while (n<maximum generations)
by different parameters of component code like interleaver size, Generate new harmonics of
information bit multiplicity free distance of the code and weight
spectrum, so a fitness function[10][11] need to be suggested by d free , N free , w free by accepting best one
choosing the parameters form equation(6).The fitness function Generate new harmonics by pitch adjustment
can be defined as
if
f = c1 * d free − c 2 * N free − c3 * w free (7) random >r (accept)

IV. PROPOSED ALGORITHM select existing one randomly

The BER performance of turbo code may be affected by different else if
parameters of component code. They mainly depend on random >r (pitch adjusting)
interleaver length, free distance of the code and weight spectrum.
tune the pitch randomly within limit
To improve BER performance, the PCCCP has been optimized
else
using a newly developed music inspired harmony search
algorithm proposed by Geem in 2001 [11]. In the Harmony generate fresh harmonics randomly
Search (HS) algorithm, each musician plays a sweet note for end if
finding a best harmony all together. [11].Then this algorithm has Accept novel harmonic
been improved and converted to hybrid adaptive Harmony
end while
Search (HS) algorithm [12] to optimize various numerical
problems by employing global information and gives superior Discover the best solution
performances in terms of heftiness and adeptness. end
The methodology for generation of optimized Parallel
Concatenated Convolution code using Harmony Search V. SIMULATION RESULTS
Algorithm has been described here. The pseudo codes Simulation results have been presented in this section. Simulation
corresponding to this proposed algorithm has been described has been carried out using MATLAB 2013a. During simulation,
subsequent parameters have been fixed related to Harmony-
below.
Search algorithm as shown in Table1. Moreover, different
Input: BER of the Parallel Concatenated Convolution Turbo parameters of Turbo code have been set according to Table2.
code for different interleaver length that needs to be controlled. During simulation, the values of C1, C2 and C3 have been set
as c1 = 100, c 2 = 10, c3 = 1 . [10][11]
Output: Optimized values of d free , N free , w free used as a
TABLE1. PARAMETERS SETTING OF HS ALGORITHM
parameter in system function for better system response by
Parameter Value
improving BER performance.begin No. Of iterations 1000
Harmony memory size 6
Objective Function: Harmony consideration rate 0.9
Maximum and minimum pitch 0.9 and 0.4
f ( x ) = c1 * d free − c 2 * N free − c3 * w free adjusting rate

Minimum and maximum 0.001 to 1.00

Initialize values of (c1, c2, c3) bandwidth

Generate initial population of harmonics

( d free , N free , w free )
TABLE2. SETTING OF PARAMETERS BASED ON TURBO CODE
Parameter Value
Code Rate 1/3,1/2
Interleaver Length 128,256,512,1024
Channel AWGN

The usefulness of the proposed technique has been assessed in

terms of the BER performance of the optimized Parallel
Concatenated Convolution Turbo code. For this purpose, the
BER curves obtained with optimized codes have been presented
in Fig 2 for different interleaver lengths with code rate 1/3.

(d) 1024-bit Interleaver

Fig 2. Comparison of BER performance between
optimized and conventional PCCCs using
AWGN channel (code rate = 1/3)
Fig 2 shows that the optimized Parallel Concatenated
Convolution Turbo code exhibits better BER performance than
conventional Parallel Concatenated Convolution Turbo code in
the “waterfall” and “error-floor” segment. It can also be said that
the optimized Parallel Concatenated Convolution Turbo code can
hold its dominance in BER performance even at lower SNR
value for different lengths of the interleaver. Moreover, the BER
(a) 128-bit Interleaver
performance of the optimized Parallel Concatenated Convolution
Turbo code has been depicted in Fig 3 for another code rate 1/2.
Results similar to the previous case are obtained here. A close
inspection between there Figures shows that the BER value
obtained with code rate 1/3 and interleaver length 128 is found to
be 2.227e-05at an SNR value of 1dB. Whereas, code rate of 1/2
provides a BER value of 9.954e-08, which is much lower than
the earlier one as expected theoretically.

(b) 256-bit Interleaver

(a) 128-bit Interleaver

(c) 512-bit Interleaver

(b) 256-bit Interleaver (d) 1024-bit Interleaver

Fig 3. Comparison of BER performance between

Optimized and conventional PCCCs using AWGN
channel(code rate=1/2).

The critical investigation of the above observations has been

abridged in Table 3 for different values of interleaver length as
well as code rates.

(c) 512-bit Interleaver

TABLE 3. BER VALUES FOR DIFFERENT INTERLEAVER LENGTHS (IL) WITH CODE RATES (CR1= 1/3 AND CR2=1/2)
SNR (dB) 1 2 3
IL CR Conventional Optimized Conventional Optimized Conventional Optimized

128 CR1 3.182 e-01 2.227e-05 4.605e-02 4.69e-10 7.338e-03 1.126e-14

CR2 1.188 e-01 9.954e-08 7.338e-03 1.126e-14 5.04e-04 1.456e-21
256 CR1 1.591e-01 1.113e-05 2.302e-02 2.345e-10 3.669e-03 5.628e-15
CR2 5.941e-02 4.977e-08 3.669e-3 5.628e-15 2.52e-04 7.279e-24

512 CR1 7.956e-02 5.567e-06 1.151e-02 1.172e-10 1.834e-03 2.814e-15

CR2 2.970e-02 2.489e-08 1.834e-03 2.814e-15 1.260e-04 3.64e-22

1024 CR1 3.978e-02 2.783e-06 5.756e-03 5.862e-11 9.172e-04 1.407e-15

CR2 1.485e-02 1.244e-08 9.172e-04 1.407e-15 6.299e-05 1.82e-22
From Table 3 it is noted that the BER performance of the [3]Mark Bingeman, Amir Kayvan Khandani, Farideh Khaleghi, “Parallel
Concatenated Convolution coding”, United States Patent, No.Us 6,298,463B1,
conventional Parallel Concatenated Convolution Turbo code
Oct 2, 2001.
improves considerably when optimized values of [4] Stephan ten Brink, “Convergence Behavior of Iteratively Decoded Parallel
d free , N free , w free are used. Obviously, for SNR value of 2 dB, Concatenated Codes”, IEEE Transactions on Communications, Vol. 49, NO. 10,
Oct 2001.
the percentage improvement in BER using optimized Parallel [5] Tobias Breddermann, Peter Vary, “EXIT Functions for Parallel Concatenated
Concatenated Convolution Turbo code with interleaver length Insertion Convolutional Codes”, GLOBECOM, December 2011, Houston, TX,
1024 is found to be 99.9999999998% over the conventional USA.
[6] Haithem Ben Chikha, Slim Chaoui, Iyad Dayoub,Jean-Michel Rouvaen,
Parallel Concatenated Convolution Turbo code with code rate Rabah Attia, “A Parallel Concatenated Convolutional-Based Distributed Coded
1/2. Whereas, the percentage improvement in BER using Cooperation Scheme for Relay Channels”, Published online: 15 November 2011,
optimized code is observed as 99.999998% when code rate is 1/3. Springer Science and Business Media, LLC. 2011.
By changing the code rate from 1/2 to 1/3 the percentage [7] Perez L.C., Seghers J., Costello D.J.: ‘A distance spectrum interpretation of
turbo codes’, IEEE Trans. Inf. Theory, 1996, 42, pp. 1698–1709
improvement in BER performance is about 99.553% for an [8] Patrick Robertson, “Illuminating the Structure of Code and
interleaver length of 128. of Parallel Concatenated Recursive Systematic (Turbo) Codes”,
inProc.GLOBECOM’.94(San Francisco, CA, NOV 1994). Vol 3, pp. 1298-1303.
[9] Václav Snášel, Jan Platoš, Pavel Krömer , Nabil Ouddane, “Genetic
VI.CONCLUSION Algorithms Searching for Turbo Code Interleaver, and Solving Linear Ordering
Problem” , 978-0-7695-3184-7/08, 2008. IEEE.
In this paper, an optimized Parallel Concatenated Convolution [10] Vac1av Snasel, Jan Platos, Pavel Kromer, Nabil Ouddane, Ajith Abraham,“
turbo code has been suggested for the improvement of BER in Interleaver Optimization by Population Based Metaheuristics” 2010 Second
low, moderate as well as high values of SNR. Performance World Congress on Nature and Biologically Inspired Computing Dec. 15-
assessment complexity will be increased by increasing the length 17,2010 in Kitakyushu, Fukuoka, Japan .
[11] Z. W. Geem, J. H. Kim and G. V. Loganathan, “A new heuristic
of interleaver. But using a well- known meta-heuristic Harmony optimization algorithm: Harmony search, Simulation”, SIMULATION vol.76,
search algorithm it can be handled effectively. Significant no.2, pp.60-68, 2001, ©2001, Simulation Councils Inc. ISSN 0037-5497/01.
enhancement in BER has been noticed by optimizing [12] ] Zhi Kong, Lifu Wang, Zhaoxia Wu, Shiqing Qi, Dexuan Zou, “An
Improved Adaptive Harmony Search Algorithm”, Proceedings of the 10th World
d free , N free , w free using harmony search algorithm. The Congress on Intelligent Control and Automation ,July 6-8, 2012, Beijing, China,
noteworthy impact of the exhaustive search procedure on the 978-1-4673-1398-8/12 ©2012 IEEE.
performance of Parallel Concatenated Convolution Turbo code is
quite impressive. The high grade of percentage improvement in
BER values of the proposed code over the existing work has
correctly recognized its supremacy in this context. One of the
imminent choices of this work could be the formulation of new
objective function in order to achieve better performance.
Moreover, the work described in this paper could further be
extended by proper adaptation of different parameters associated
with harmony search algorithm. Moreover, new objective
function may be formulated in this regard. The efficient use of
recently developed upgraded version of Harmony Search (HS)
algorithm could be identified as one of the major scopes in
future.

REFERENCES:
[1] Berrou C., Glavieux A., Thitimajshima P.: ‘Near Shannon limit error
correcting coding and decoding: turbo codes’.Proc. Int. Conf. Communications,
May 1993, pp. 1064–1070
[2]Sergio Benedetto, Guido Montorsi, “Design of Parallel Concatenated
Convolutional Codes”, IEEE Transactions on Communications, Vol. 44, NO. 5,
May 1996.