Soft Decision Signal Detection of MIMO System Based on Deep Neural Network

Abstract

This paper proposes a multiple-input multiple-output (MIMO) soft decision signal detection method for a timevarying communication system. In this algorithm, the training samples, including system channel state information and received data, are input to a deep neural network (DNN), and then we employ cross-entropy loss function and root mean square propagation (RMSProp) descent algorithm to offline train and optimize the parameters of the DNN. Besides, the output layer of the DNN uses the sigmoid function as the activation function, and the negative value of the input value of the sigmoid function is the log-likelihood ratio (LLR). In this way, we can obtain the LLR value via removing the sigmoid function during the online testing without the complicated process of calculating the LLR value. Combining the DNN with the soft decision technology improves signal detection performance. Simulation results show that the proposed algorithm is better than the MMSE algorithm and similar to ML algorithm.