Deep Learning Based Modulation for Quantized SIMO Communications
Abbas Khalili, Elza Erkip · 2022
High resolution analog to digital converters (ADCs) are major sources of power consumption in conventional fully digital receivers. To reduce power consumption, it is suggested to use a small number of low resolution ADCs. However, for a given number of low resolution ADCs, the best modulation scheme and corresponding quantization strategy that maximizes the spectral efficiency are still unknown. In this paper, a point-to-point singleinput multiple-output system with a hybrid receiver using low resolution ADCs is considered. A novel deep learning based modulation scheme and the associated quantization strategy are developed to maximize the spectral efficiency. It is observed that the derived modulation scheme can achieve optimal high SNR rates and perform near Shannon capacity at low and intermediate SNRs. Our results provide new insights into the optimal utilization of low resolution ADCs and optimal power-performance trade-off in terms of the spectral efficiency.