Scalable Cell-Free Massive MIMO Systems: Impact of Hardware Impairments

TL;DR

This paper evaluates the impact of hardware impairments, including phase noise, on scalable cell-free massive MIMO systems, deriving capacity bounds and proposing hardware-aware signal processing techniques.

Contribution

It introduces a comprehensive model for hardware impairments in SCF mMIMO, including phase noise, and derives capacity bounds and optimal combiners considering these impairments.

Findings

Separate LOs outperform common LOs in SCF mMIMO.

Additive transmit distortion has a greater negative impact than receive distortion.

Closed-form lower bounds on capacity are derived using deterministic equivalents.

Abstract

Standard cell-free (CF) multiple-input-multiple-output (mMIMO) systems is a promising technology to cover the demands for higher data rates in fifth-generation (5G) networks and beyond. These systems assume a large number of distributed access points (APs) using joint coherent transmission to communicate with the users. However, CF mMIMO systems present an increasing computational complexity as the number of users increases. Scalable cell-free CF (SCF) systems have been proposed to face this challenge. Given that the cost-efficient deployment of such large networks requires low-cost transceivers, which are prone to unavoidable hardware imperfections, realistic evaluations of SCF mMIMO systems should take them into account before implementation. Hence, in this work, we focus on the impact of hardware impairments (HWIs) on the SCF mMIMO systems through a general model accounting for both additive and multiplicative impairments. Notably, there is no other work in the literature studying the impact of phase noise (PN) in the local oscillators (LOs) of CF mMIMO systems or in general the impact of any HWIs in SCF mMIMO systems. In particular, we derive upper and lower bounds on the uplink capacity accounting for HWIs. Moreover, we obtain the optimal hardware-aware (HA) partial minimum mean-squared error (PMMSE) combiner. Especially, the lower bound is derived in closed-form using the theory of deterministic equivalents (DEs). Among the interesting findings, we observe that separate LOs (SLOs) outperform a common LO (CLO), and the additive transmit distortion degrades more the performance than the additive receive distortion.