Channel Tracking and Detection Based on Long-Short Term Memory in Millimeter Wave System

Abstract

Millimeter wave communication is one of the most promising technology for 5G and beyond in the future. Massive MIMO technology and beamforming technology are deployed to compensate for the severe path loss. However, millimeter wave channels still exist some problems such as susceptibility to channel abrupt changes (CAC) due to environmental impacts. Therefore, detecting the CAC of the millimeter wave channel effectively is one of the key issues in keeping high service quality. This paper proposes a Long-Short Term Memory (LSTM) algorithm for the detection of CAC. Specifically, extended kalman filter (EKF) is exploited for channel tracking, and the obtained channel state information (CSI) is collected to train the LSTM network in the offline training phase. Then, the trained LSTM network would detect CAC consecutively in the online learning phase. The key of this algorithm is to make full use of the effective information in different slots to further improve the detection performance. The results prove that, compared with traditional algorithms, the proposed algorithms decrease the false detection rate (FDR) by 47% while the missed detection rate (MDR) can be maintained at a stable level.