计算机工程

• •    

一种 基于循环神经网络的极化码BP译码算法

  

  • 发布日期:2021-01-11

A Recurrent Neural Network Based Belief Propagation Decoding Algorithm for Polar Codes

  • Published:2021-01-11

摘要: 置信传播(BP)算法作为极化码最常用的软判决输出译码算法之一,具有并行传输、高吞吐量等优点,但存在收敛较 慢,运算复杂度较高等缺陷。因此,本文针对以上缺陷,提出了一种基于循环神经网络的偏移最小和近似置信传播 (RNN-OMS-BP)译码算法。仿真结果表明,相比于传统 BP 译码算法,该译码算法在提升 BER 性能的前提下,减少约 75%的 加法运算且收敛速度大幅提升。相比于基于深度神经网络(DNN)的 BP 译码算法,在确保 BER 性能无显著下降的前提下,本 文译码算法使用加法运算替代了乘法运算,同时节省了约 80%的存储空间开销。

Abstract: Belief Propagation(BP) is one the most commonly used soft decision decoding algorithm in polar codes, which has the advantages of parallelism and high throughput, however, suffering from slow convergence and high computation complexity. In order to solve these problems, this paper proposes a recurrent neural network based BP decoding algorithm for polar codes with offset min-sum approximation(RNN-OMS-BP). Simulation results show that compared with traditional BP algorithm, the proposed decoding algorithm can reduce the addition operation by about 75% and greatly improve the convergence speed while improve the BER performance. Compared with DNN-BP decoding algorithm, On the premise of ensuring no significant decline in BER performance, the proposed decoding algorithm in this paper uses addition operation to replace multiplication operation, and at the same time saves about 80% storage space overhead.