Joint Estimation and Detection for MIMO-STBC System Based on Deep Neural Network

Abstract

Motivated by the recent advances in deep neural network (DNN), we propose a general DNN scheme to joint channel estimation and signal detection for multi-input multioutput (MIMO) system with space-time block coding (STBC). Different from the maximum likelihood (ML) detection with imperfect channel state information (CSI) which consists of channel estimator, combiner and detector, the proposed DNN scheme is designed to replace the three modules jointly, estimate CSI implicitly and recover the transmitted symbols directly. Specifically, a four-layer DNN is constructed for MIMO-Alamouti system and is simulated under the condition of quasi-static rayleigh channel and QPSK modulation. The simulation results demonstrate that, at the same bit error ratio (BER), the signalto-noise ratio (SNR) of the DNN-based detection has a loss of 1dB compared with ML detection with perfect CSI, and has a gain of 3dB compared with ML detection with imperfect CSI in MIMO-Alamouti system.