Learning for Time Series Differential Phase-Shift Keying-based Non-coherent Receivers

Abstract

In this paper, using the recurrent neural network (RNN) for non-coherent differential phase-shift keying (DPSK) signal detection is proposed. While deep-learning has been utilized for estimating the channel and detecting the signal in various communication systems models, making use of the time series encoding of DPSK modulation along with the time series feature of the recurrent neural networks for Rayleigh fading communication channel has not been studied before. The use of both of deep neural network (DNN) and recurrent neural network to detect the differential encoded time series signal is considered and compared. Both of the DNN-based and RNN-based detectors are compared to conventional non-coherent DPSK detection, coherent phase-shift keying (PSK) detection, and the analytical error rate of coherent and non-coherent detection. Extensive simulation results exhibit that the proposed RNN-based signal detector outperforms both of conventional non-coherent differential detection, and the DNN-based detector in terms of symbol error rate performance. In addition, it does not need any statistical information about the channel or require complex online calculations.