An Efficient ADMM-Aided Deep Learning-Based Signal Detector for Uplink Massive MIMO

Abstract

This paper provides an efficient Alternating Direction Method of Multipliers (ADMM)-aided deep neural network (DNN) framework for signal detection based on uplink Massive Multiple Input Multiple Output (Massive MIMO) systems. Since the ADMM is a good candidate for distributed computing and solving complicated problems, it is integrated into a DNN to enhance the learning performance of the DNN. Not only can the developed DNN emulate the ADMM operations with fast speed and good performance, but also it fully exploits the characteristics of the samples for signal detection. Specifically, the DNN comprises two parts: the data-processing module is adopted for making the samples sparse and iteration-mapping module is for reconstructing transmit signals with the aid of the ADMM. Then, a learning policy is proposed, which is composed of an offline learning method and a lifelong learning-based online method. Simulation results show that the proposed scheme is capable of reducing signal-detection error compared with previous work. Also, it is corroborated that the proposed scheme only requires low computational complexity.