Performance Analysis of Massive MIMO Multi-Way Relay Networks with Low-Resolution ADCs

TL;DR

This paper analyzes the impact of low-resolution ADCs on the performance of massive MIMO multi-way relay networks, providing closed-form rate approximations and quantifying the effects of quantization and channel estimation errors.

Contribution

It introduces a novel analytical method for rate analysis in mixed-ADC massive MIMO relay systems using SVD and Haar matrix properties, accounting for imperfect CSI.

Findings

Achievable rate approximations are derived for various ADC resolutions.

Performance degradation due to low-resolution ADCs is quantified.

Rate performance is nearly linear with the square of quantization coefficients.

Abstract

High power consumption and hardware cost are two barriers for practical massive multiple-input multiple-output (mMIMO) systems. A promising solution is to employ low-resolution analog-to-digital converters (ADCs). In this paper, we consider a general mMIMO multi-way relaying system with a multi-level mixed-ADC architecture, in which each antenna is connected to an ADC pair of an arbitrary resolution. By leveraging on Bussgang's decomposition theorem and Lloyd-Max algorithm for quantization, tight closed-form approximations are derived for the average achievable rates of zero-forcing (ZF) relaying considering both perfect and imperfect channel state information (CSI). To conquer the challenges caused by multi-way relaying, the complicated ZF beam-forming matrix, and the general mixed-ADC structure, we develop a novel method for the achievable rate analysis using the singular-value decomposition (SVD) for Gaussian matrices, distributions of the singular values of Gaussian matrices, and properties of Haar matrices. The results explicitly show the achievable rate behavior in terms of the user and relay transmit powers and the numbers of relay antennas and users. Most importantly, it quantifies the performance degradation caused by low-resolution ADCs and channel estimation error. We demonstrate that the average achievable rate has an almost linear relation with the square of the average of quantization coefficients pertaining to the ADC resolution profile.