A Super-resolution Channel Prediction Approach based on Extended Matrix Pencil Method

Abstract

This paper addresses the challenge of mobility in massive multiple-input multiple-output (MIMO) communication systems. In the deployment of 5G, this problem leads to alarmingly high performance degradation. In this paper, we propose a novel multi-dimension Matrix Pencil (MDMP) channel prediction method in order to tackle this practical challenge. More specifically, our method calculates accurate estimations of path angles, delays and Doppler simultaneously. In order to do so, we exploit the angular-frequency-domain and angular-time-domain structures of the wideband channel, and propose a path pairing procedure by exploiting the super-resolution property of the estimated angles. Our method is able to deal with the realistic constraint of time-variant path delays introduced by user movements, which have not been considered so far in literature. We prove that with only two arbitrary channel samples given, that prediction error converges to zero in the scenario with time-variant delay and arbitrary delay of channel state information (CSI), if the number of base station (BS) antennas is large enough. Unlike the existing Prony-based angular-delay domain (PAD) prediction method that assumes the CSI delay is an integral multiple of the pilot interval, our MDMP method breaks such a limitation and is therefore more general. Simulation results under the clustered delay line (CDL) model of 3GPP demonstrate that in the high-mobility scenario with time-variant path delays and a large CSI delay of 16 ms, our proposed MDMP method can approach to the performance of the stationary scenario.