A deep neural networks based demodulator for PD-SCMA-VLC

Abstract

Power domain sparse code multiple access (PD-SCMA) scheme-based visible light communications (VLC) can support higher spectrum efficiency and increased number of users in wireless access networks. In PD-SCMA-VLC, successive interference cancellation (SIC) and message passing algorithm (MPA)-based receivers are used to recover the transmitted signals in the power and code domains, respectively, which are complex requiring higher signal to noise ratio levels to deliver the quality of services. We propose a deep neural networks (DNN) based scheme to recover the PD-SCMA signal effected by the mixed noise, multipath distortions, and the nonlinearity due to the light emitting diode, channel and photodiode. We show by experiment that, the proposed DNN significantly outperforms the conventional SIC-MPA in terms of the achievable data rate and bit error rate performance. For DNN-based PD-SCMA-VLC, the achievable data rates are more than 100 and 120 Mbps for the user groups 2 and 1, respectively.