Tensor-Based Downlink Channel Reconstruction for FDD Massive MIMO

The acquisition of downlink channel state information (CSI) at the base station is a prerequisite for various applications in frequency division duplex (FDD) massive multipleinput multiple-output (MIMO) systems. To obtain accurate CSI, conventional approaches employ downlink training and feedback with a considerable overhead. In this paper, we exploit the partial reciprocity between FDD uplink and downlink channels to propose a tensor-based method for downlink channel reconstruction. According to this partial reciprocity, the tensor-based method efficiently estimates angle and delay parameters of the downlink channel from the uplink channel. Then, downlink training and feedback are incorporated by exploiting the sparse scattering property of the channel, so as to estimate the gain parameters of the multipath components with a small amount of overhead. Downlink channel reconstruction is achieved according to the estimated gain, angle and delay parameters. Experimental results on a Ray-tracing dataset demonstrate the effectiveness of the proposed downlink training and feedback scheme, and the superior channel estimation of the proposed tensor-based method compared with several alternative methods.

I. Introduction

Frequency division duplex (FDD) is widely-used as a key duplexing mode in massive multiple-input multiple-output (MIMO) systems [1]–[9]. Massive MIMO deploys large-scale antenna arrays at the base station (BS) to ensure high spectral and energy efficiencies. However, due to the configuration of large-scale antenna arrays, it is challenging for FDD massive MIMO to acquire accurate downlink channel state information (CSI) at the BS [1]–[4].