On Normalization of Matched Filter Belief in GaBP for Large MIMO Detection

Abstract

This paper proposes a normalized matched filter (MF) belief in Gaussian belief propagation (GaBP) detection especially for a large multiple-input multiple-output (L-MIMO) configuration where a base station (BS) has tens of antennas. In a massive MIMO channel where the BS has hundreds of antennas, damped GaBP is known to be an effective detector in terms of low computational complexity and its detection capability. However, in L- MIMO channels, GaBP is subject to ill convergence behavior of iterative detection due to lack of channel hardening effects obtained by massive number of receive antennas. To improve the convergence property, we investigate the MF belief, instead of a traditional log likelihood ratio (LLR) belief. Then, we propose the novel normalized MF belief according to instantaneous channel state. As a side effect of the normalization, a noise variance estimator is not necessary. Finally, we demonstrate the validity of the normalized MF belief with the aid of damped processing, in terms of suppression of bit error rate (BER) floor as well as approach to maximum likelihood detection (MLD) limit.