Belief Propagation Decoding of Polar Codes using Intelligent Post-processing

Polar codes, as only one provable capacity-achieving channel code, have become one of the 5G standards. Belief propagation (BP) algorithm has become one of popular decoding polar codes because of unique advantage of high parallelism. But compared to the successive cancellation list (SCL) algorithm, BP algorithm still has a performance gap. This paper exploits the design of polar BP decoder using intelligent post-processing to improve error performance. The post-processing is used to find the error bits and flip it as the messages of the initial frozen positions. We also use deep neural networks to improve the accuracy of finding error bits. Simulation shows that compared with the original BP decoding, the approach proposed in this work can achieve a performance improvement of about 0.3 dB with slight complexity increase.

I. Introduction

Polar Codes, as the first provable capacity-achieving channel codes that are strictly proved in binary discrete memoryless channel, have attracted extensive attention because of their low coding and decoding complexity. The decoder design of polar codes has been the focus of the current research. In general, decoding algorithms of polar codes have successive cancellation (SC) algorithm [1], successive cancellation list (SCL) algorithm [2] and belief propagation (BP) algorithm [3]. But in the case of finite length, the performance of successive cancellation (SC) algorithm is not ideal. Compared to the inherently serial SC algorithm, SCL algorithm improves the error performance of polar code by sacrificing computation complexity and storage space. However, due to the inheritance of the SC algorithm, SCL algorithm suffers from high delay. Under low delay conditions, BP algorithm can achieve better error performance than SC algorithm. But compared to SCL decoding, BP algorithm still has a certain performance gap.