Model-Based and Data-Driven Approaches for Downlink Massive MIMO Channel Estimation

• Published in: IEEE transactions on communications Vol. 70; no. 3; pp. 2085 - 2101
• Main Authors: Ghazanfari, Amin, Van Chien, Trinh, Bjornson, Emil, Larsson, Erik G.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.03.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Asymptotic methods; Channel estimation; Downlink; Downlink channel estimation; Downlinking; Estimation; Hardening; linear precoding; Low level; Massive MIMO; Neural networks; non-isotropic scattering; Precoding; Rayleigh channels; linear precoding; Downlink channel estimation; Precoding; non-isotropic scattering; Neural networks; Channel estimation; Estimation; Massive MIMO; Rayleigh channels; Downlink
• ISSN: 0090-6778; 1558-0857; 1558-0857
• DOI: 10.1109/TCOMM.2021.3133939

We study downlink channel estimation in a multi-cell Massive multiple-input multiple-output (MIMO) system operating in time-division duplex. The users must know their effective channel gains to decode their received downlink data. Previous works have used the mean value as the estimate, motivated by channel hardening. However, this is associated with a performance loss in non-isotropic scattering environments. We propose two novel estimation methods that can be applied without downlink pilots. The first method is model-based and asymptotic arguments are utilized to identify a connection between the effective channel gain and the average received power during a coherence interval. The second method is data-driven and trains a neural network to identify a mapping between the available information and the effective channel gain. Both methods can be utilized for any channel distribution and precoding. For the model-aided method, we derive all expressions in closed form for the case when maximum ratio or zero-forcing precoding is used. We compare the proposed methods with the state-of-the-art using the normalized mean-squared error and spectral efficiency (SE). The results suggest that the two proposed methods provide better SE than the state-of-the-art when there is a low level of channel hardening, while the performance difference is relatively small with the uncorrelated channel model.