LSTM based Receiver Design for Baseband Signal Demodulation

Abstract

This paper summarizes an intelligent correlation receiver design for baseband demodulation with Long Short Term Memory based Deep Learning technique. We explain the generation of training data, bipolar signaling, consideration of channel noise. i.e, Additive White Gaussian Noise, training of model, fine tuning of dense layers to properly fit the communication application. Better Bit Error Rate performance over conventional correlation receiver is observed by DL model subjected to a AWGN noise channel with a noise in Signal to Noise Range of −50 dB to 20 dB. We validate this by applying the customized model to reconstruct successfully a Pulse Code Modulated 255 $\mu$-law companded audio signal immersed in AWGN noise channel.