Performance evaluation of the sequential neural network decoder with improved bit error rate

Abstract Due to the dimensionality of the neural network channel code decoder's training complexity increases exponentially with the length of the code term. While the computational capacity has advanced considerably, it is still difficult to manage the word code for long blocks. In this work, we proposed a decoder based on a neural network called the Sequential Neural Decoder (SNND). The SNND consists of several Sub-Models, and transfers the final state of the present Sub-Model as initial state to the following model. The output of the SNND's bit error rate (BER) continues unchanged as submodels increase, and the performance of Viterbi soft decisions in the white additive Gaussian noise (AWGN) channel is near. The output of the SNND, however has decreased along with the increased order of modulation.