The neural decoder for convolutional codes on the mobile channel

This paper analyses the possibility of decoding convolutional codes (CC) by means of the multilayer perception neural network (NN) trained with the error backpropagation (EBP) algorithm. The result of the simulation shows: the best performance of the NN decoder is obtained when the reference codeword position is at the center of the working window and the performance does not improve considering larger window sizes. The random strategy success is encouraging because it offers a way to overcome the training procedure's complexity. The better performance of the adaptivity and redundancy fault tolerance of the NN is shown on the larger NN scale. The system is widely recommended, because its scale is minimum for a window width equal to 3. The Viterbi decoding does not suit the mobile channel.