Sequential neural network decoder for convolutional code with large block sizes

Abstract

Due to the curse of dimensionality, the training complexity of the neural network based channel-code decoder increases exponentially with the codeword length. Although computation power has made significant progresses, it is still hard to deal with convolutional code with large block sizes. In this paper, we proposed a neural network based decoder termed Sequential Neural Network Decoder (SNND). The SNND consists of multiple sub models, and it passes the last state of the current sub model to the following model as the initial state. The bit error rate (BER) performance of the SNND remains unchanged as the number of sub models increases. It achieves a performance close to that of the soft decision Viterbi decoder under Additive white Gaussian noise (AWGN) channel. However, the SNND’s performance is found to decrease when the modulation order increases.