High-performance Decoder for Convolutional Code with Deep Neural Network

Abstract

The use of deep neural network for decoding error control code will encounter two problems, namely, the high-precision requirements of the error control code and the complexity of the neural network due to the long code. In this paper, a deep neural network decoder is proposed to solve the decoding problem of long code by using the nature of convolutional code window decoding. A deep neural network decoder is utilized as a weak classifier, and an integrated decoder is proposed to improve the decoding performance greatly. The Viterbi decoder is improved by approximately 2 db at a bit error rate of 10−5. Both decoder methods proposed in this paper can be decoded in parallel and are suitable for high-bitrate applications. This study reveals that the accuracy of neural networks can reach 10−8 or more.