On Rate-Splitting With Non-unique Decoding In Multi-cell Massive MIMO Systems

TL;DR

This paper explores rate-splitting with non-unique decoding in multi-cell massive MIMO systems, demonstrating significant spectral efficiency gains over traditional interference treatment methods under various channel conditions.

Contribution

It introduces a novel rate-splitting scheme with non-unique decoding for pilot contamination interference, providing an achievable region and showing improved performance over existing methods.

Findings

RS achieves higher spectral efficiency than TIN, SD, and SND.

RS maintains gains across different channel correlations and fading conditions.

The scheme is effective with a practical number of antennas.

Abstract

We consider the downlink of a multi-cell massive MIMO system suffering from asymptotic rate saturation due to pilot contamination. As opposed to treating pilot contamination interference as noise (TIN), we study the performance of decoding the pilot contamination interference. We model pilot-sharing users as an interference channel (IC) and study the performance of schemes that decode this interference partially based on rate-splitting (RS), and compare the performance to schemes that decode the interference in its entirety based on simultaneous unique decoding (SD) or non-unique decoding (SND). For RS, we non-uniquely decode each layer of the pilot contamination interference and use one common power splitting coefficient per IC. Additionally, we establish an achievable region for this RS scheme. Solving a maximum symmetric rate allocation problem based on linear programming (LP), we show that for zero-forcing (ZF) with spatially correlated/uncorrelated channels and with a practical number of BS antennas, RS achieves significantly higher spectral efficiencies than TIN, SD and SND. Furthermore, we numerically examine the impact of increasing the correlation of the channel across antennas, the number of users as well as the degree of shadow fading. In all cases, we show that RS maintains significant gain over TIN, SD and SND.