Novel use of channel information in a neural convolutional decoder

A neural convolutional decoder which exploits the channel information is introduced. The method uses a recurrent neural network, tailored to the used convolutional code and the channel model. No supervision-besides possible channel estimation-is required. Also, no distinct equalizer is needed. As an example, we show the structure of the neural decoder for 1/2 rate code with constraint length 3 in a two-path channel environment. For testing, the 1/2 rate code with constraint length 5 is used in two-path fading channels. The simulation results show that the proposed decoder works well compared to the traditional way of using some equalizer and the Viterbi decoder. The hardware implementation of the neural decoder seems feasible and its complexity increases only polynomially while in Viterbi algorithm the complexity increases exponentially as a function of the constraint length