CNN and RNN-based Deep Learning Methods for Digital Signal Demodulation
Tian Bao Wu · 2019
In this paper we presented a neural network-based method to demodulate digital signals. After training with different modulation schemes, the learning-based receiver can perform demodulation without changing receiver hardware by loading certain parameters based on the modulation scheme. Combining CNN's ability to extract local features and RNN's time series modeling ability, we designed a mixed neural network model with parallel architecture and simulate FSK, PSK, QAM demodulation over AWGN and Raleigh-faded channels. The results show that the mixed neural network model can equal or even exceed the performance of the conventional demodulation method (matched filter or correlation-based demodulation). With this kind of receiver, we can intelligently process multiple types of digital modulated signals with flexibility.