LPM-Turbo Decoding Structure based on Simulink

Li Xiangling, Wang Ke, Xu Zhao

College of Communications and Engineering
Jilin University
Changchun, China
Email: [email protected]

Abstract—Turbo decoding model with the linear prediction

model is developed in Simulink based on the turbo decoding
theory to replace one of iterations and reduce the decoding
delay time. The factors are presented in the results of simulation,
which are the linear prediction model, the length of interleave
and the different decoding arithmetics that influence on the
error-correcting performance of the modified structure of turbo
decoder. The error-rate performance of turbo decoding structure
based on the linear prediction model is only lower by 0.0∼0.2dB
than the traditional one, and the modified model can reduce the
decoding delay time. As the length of interleave increases, the
performance of the modified decoding structure also increases, Figure 1. Turbo decoder block diagram
but the length of interleave should be chosen properly because
the decoding delay time is proportional to the length of interleave. B. Turbo Codes Developed in Simulink Currently
Log-MAP algorithm has the best performance, the performance
of which is 0.5dB higher than MAX-Log-MAP algorithm when I. S. Raad designed the implementation of a turbo codes
BER is 10−2 . test bed in Simulink, and analyzed the factors, which are
Index Terms—Simulink; turbo codes; the linear prediction the iterative numbers, the constraint length and the length of
model; the iterative decoding structure.
interleave as well as different decoding arithmetics that influ-
I. INTRODUCTION ence on the error-correcting performance [3]. Fig.2 shows the
turbo decoder developed in Simulink. According to the coding
The iterative decoding method is used to Turbo decoder to
theory of turbo codes in CDMA2000, Chen Zhao designed the
complete the exchange of the soft information between the
model of turbo encoder [4]. Zhou Xianwei designed the model
decoders. But with the long delay caused by iterations, turbo
of turbo decoder by S-function, and analyzed the performance
codes can not be used to the communication system which has
of turbo codes on different channels [5]. Xie Wei designed
high requirements in real-time data transmission. Turbo de-
the pipelining decoder by modifying the decoding model of
coder based on the linear prediction control algorithm replaces
I. S. Raad [6]. Zhong Xiaochun analyzed the performance of
one of iterations, and the decoding delay time is reduced, but
turbo coding, the conventional coding and non-channel coding
the error-rate performance of the modified decoder is only
[7].
lower by 0.1∼0.2dB than the traditional one [2].
On the theoretical basis of turbo decoder based on the linear
prediction control algorithm, this paper designs the implemen-
tation of a turbo encoder and decoder test bed in Simulink,
and the factors are presented in the results of simulation, which
influence on the error-correcting performance of the modified
structure of turbo decoder, such as the linear prediction model
and the length of interleave as well as the different decoding
arithmetics.
Figure 2. Turbo decoder developed in Simulink
II. ARCHITECTURE OF TURBO DECODER
A. Turbo Decoder Block Diagram C. Turbo Decoder Based on the Linear Prediction Control
Turbo decoder consists of decoders, interleavers and de- Algorithm
interleavers. Fig.1 shows the turbo decoder block diagram. Lc The polynomial curve fitting method is used to the linear
is the channel confidence coefficient. With a few iterations, the prediction control algorithm. Fig.3 shows the decoder block
decoding process is stopped when either the iterative stopping diagram. Buffers (M1 and M2) store the extrinsic information
criterion or the fixed number of iterations is satisfied. At the and the priori information which are related to the second
end of the decoding process a hard decision is carried out on decoder. P arithmetic model (PAM) outputs the polynomial
the soft output of the second decoder. coefficient. Prediction and calculation model (FCM) estimates

978-1-4244-3709-2/10/$25.00 ©2010 IEEE

the external information of the number n+1 iteration, and
replaces the number n+1 iteration of the decoding process.

Figure 5. LPM-Turbo decoder developed in Simulink

The Linear Pre model is the linear prediction model (LPM).

Figure 3. Decoder block diagram based on the prediction control algorithm
Fig.6 shows the LPM structure developed in Simulink. The
rate transition module recovers the sample time amplified by
III. DESIGN IN SIMULINK the buffer module.

A. Turbo Encoder

The encoder adopts the Recursive Systematic Convolutional

Code (RSC), whose generator matrix is [111; 101]. The en-
coder, the interleaver and the puncture are completed by S-
function.
The input is generated by the Bernoulli Binary Gener-
ator,and transmitted into the first encoder and the second
encoder after it has gone through an interleaver. The puncture
Figure 6. Linear Prediction Model developed in Simulink
keeps the systematic bits, the odd number of bits from code1
and the even number of bits from code2. Then, the encoded
bits are transmitted into the Additive White Gaussian Noise IV. RESULTS
(AWGN) channel modulated by Binary Phase Shift Keying A. The Influence of the Different Length of Interleaves
(BPSK) modulation. Fig.4 is turbo encoder developed in Fig.7 shows the performance of LPM-Turbo decoder with
Simulink. the different length of interleaves. It can be seen that, compar-
ing the decoding performance of the length of 570 and 378, the
former is 0.0∼0.2dB better than the latter when SNR is greater
than 1.2dB. And comparing the performance of the length of
1146 and 570, the former is 0.0∼0.3dB better than the latter
when SNR is greater than 0.6dB. As the SNR increases, the
differences of three curves increase. In all, when the length of
interleave is longer, the decoding performance is better, which
Figure 4. Turbo encoder developed in Simulink is more obvious when SNR is bigger.
B. The Influence of Different Algorithms
B. Turbo Decoder Based on the Linear Prediction Model Fig.8 shows the performance of LPM-Turbo decoder with
(LPM-Turbo) different algorithms. It can be seen that, the performances of
MAX-Log-MAP algorithm and SOVA algorithm are almost
Fig.5 shows LPM-Turbo decoder developed in Simulink. the same when SNR is lower than 1.4dB, and there are the
The information sequence passed through the AWGN channel differences of 0.1∼0.4dB when SNR is higher than 1.4dB.
is received by the first decoder. The Le2 Initialization model, The performances of MAP algorithm and Log-MAP algorithm
which includes a Matlab Fcn module and a delay module, nearly coincide expect when SNR is 0.6dB and 1.4dB. As
ensures that the initial value of the priori information is zero, bit error rate (BER) is 10−2 , the performances of Log-MAP
and the priori information of the first decoder comes from the algorithm and MAX-Log-MAP algorithm have the difference
second decoder of the last iteration. The iterative number is of 0.5dB. In all, the error-correcting performance of Log-MAP
controlled by the counter limited module, the switch1 module algorithm is best.
is a selector, and the nforpre module can generate a constant
n. Switch1 chooses the information from the linear prediction C. The Possibility of the Linear Prediction Model
model when the iterative number is n+1, otherwise it chooses Fig.9 shows the extrinsic information of LPM-Turbo de-
the information from the second decoder. coder and turbo decoder, and the iterative number is six
 and Fig.9(b), the extrinsic information begins to converge
before the fifth iteration so that the predicted information
is precise and the differences of the extrinsic information in
the last three iterations are zero. From Fig.9(c) and Fig.9(d),
although the trend of the extrinsic information is instability
and the differences of the extrinsic information in the last
three iterations are about ±0.1, the decoding performance is
unaffected. From Fig.9(e) and Fig.9(f), although the change
of the extrinsic information is in trend and instability in the
end, the decoding performance is also unaffected. In all, the
method with the linear prediction model is possible.
Fig.10 shows the decoding delay time of the p arithmetic
Figure 7. Comparison of varying length of interleaver
model and the first decoder when SNR is 2, 1 and 0 separately.
The length of interleave is 1146, 570 and 378 from left to right.
It can be seen that, the decoding delay time is proportional to
the length of interleave, and the decoding delay time of the
p arithmetic model is two times higher than the first decoder,
so that LPM-Turbo decoder can save time compared with the
traditional one.

Figure 8. Comparison of different decoding algorithms

when the extrinsic information is replaced. From Fig.9(a) Figure 10. Comparison of the decoding delay time

(a) (b) (c)

(d) (e) (f)

Figure 9. Prediction of the extrinsic information
 (a) (b)

(c) (d)
Figure 11. BER performance of turbo codes and LPM-Turbo codes

D. The Performance of LPM-Turbo Codes formance of LPM-Turbo decoder is only lower by 0.0∼0.2dB
Fig.11 shows the error-correcting performance of LPM- than the traditional one, and it can reduce the decoding delay
time. When the length of interleave is longer, the performance
Turbo codes and turbo codes. From Fig.11(a), when the
of LPM-Turbo decoder is better, but the length of interleave
iterative number choosing the predicted information is five,
should be chosen properly because the decoding delay time is
the difference caused by the prediction is 0.2dB and two
proportional to the length of interleave. Log-MAP algorithm
curves are almost the same when SNR is 0.0∼0.4dB and
has the best performance, the performance of which has 0.5dB
0.8dB. From Fig.11(b), when the iterative number choosing
higher than MAX-Log-MAP algorithm when BER is 10−2 .
the predicted information is six, two curves nearly coincide
between 0.0∼0.8dB, and the difference caused by the pre- R EFERENCES
diction is 0.1∼0.2dB between 1.0∼2.0dB. From Fig.11(c), [1] C. Berrou, A. Glavieux and P. Thitimajshima, Near Shannonlimit error-
the difference caused by the prediction is 0.0∼0.1dB, and correction coding and decoding: Turbo-codes, in Proc. ICC’93, Geneva,
the performance is the worst when SNR is 1.0∼1.2dB and Switzerland, pp. 1064-1070, May. 1993.
[2] Xu Zhao, Wang Ke and Li Zhuo, Iterative Decoding Structure Design of
1.6∼2.0dB. From Fig.11(d), the difference caused by the LPCA-Turbo, The Second International Conference on Information and
prediction is 0.0∼0.1dB, and the performance is the worst Computer Science, in press.
when SNR is 1.0∼1.2dB. In all, the difference caused by the [3] I. S. Raad and M. Yakan, Implementation of a Turbo Codes Test Bed
in the Simulink Environment, in Signal Processing and Its Applications,
prediction is 0.0∼0.2dB, and as the iterative number choosing 2005. Proceedings of the Eighth International Symposium on, vol. 2,
the predicted information increases, the performance of LPM- 28-31, pp. 847 - 850, Aug. 2005.
Turbo codes also increases. [4] Chen Zhao and Chen Fang, The Simulation of a Code Algorithm of Turbo
in CDMA 2000, Journal of EEE, vol. 27, no. 3, Jun. 2005.
V. CONCLUSION [5] Zhou Xianwei, Zhao Xin and Wang Lina, A simulation system of Turbo
Code using SIMULINK, Control and Automation, vol. 22, no. 5-3, 2006.
This paper developed a model of LPM-Turbo decoder in [6] Xie Wei, Hu Guijun, Li Gongyu, Deng Qing and Shi Xinliang, Modeling
Simulink and the factors presented are the linear prediction in Simulink and Performance Test of Turbo Codes, Journal of Jinlin
University, vol. 25, no. 5, Sept. 2007.
model, the length of interleave and different decoding arith- [7] Zhong Xiaochun, Han Jiajia and Liang Nawei, Simulation of Turbo Codes
metics, which influence on the error-correcting performance. System Based on Simulink, Journal of Southwest Jiaotong University, vol.
As is shown in the results of simulation, the error-rate per- 43, no.6, Dec. 2008.