Performance of Cell-Free Massive MIMO in Realistic Urban Propagation Environments

TL;DR

This work questions the practicality of UE-centric cell-free massive MIMO in realistic urban propagation by comparing raytracing-based city-wide channels with the conventional log-distance model in Amsterdam and Dresden. It uses a UE-centric CF-mMIMO framework with configurable CPU clusters (Q) and serving APs (E), MMSE downlink processing, and two channel models to evaluate spectral efficiency under realistic site-specific shadowing and urban canyon effects. The findings show substantial non-uniformity and outages for worst-case UEs under raytracing, especially in Amsterdam, and only limited revival of performance in Dresden unless very large serving sets are employed—which introduces prohibitive signaling and cost. The study highlights the need for propagation-environment-aware network architectures to realize CF-mMIMO benefits in real urban deployments.

Abstract

While UE-centric cell-free massive MIMO (CF-mMIMO) provides high and uniform throughput performance under the assumption of a uniform propagation environment modeled by the log-distance path loss channel model, the performance under a realistic urban propagation environment is not yet fully addressed. In this paper we conduct the first comparative performance study of CF-mMIMO under both the widely assumed log-distance channel model and the realistic urban propagation environment obtained via raytracing using real 3D city layouts and practical AP locations. Our results show that with the raytracing channel model, CF-mMIMO cannot achieve as high and uniform throughput performance as observed with the log-distance channel model, putting into question the attractiveness in practice of CF-mMIMO for real urban deployments.

Figures (9)

• Figure 1: Urban AP distributions according to Algorithm \ref{['aplocate']} for $M=324$ APs, with APs marked by crosses and the uniform grid in Step 4 of Algorithm \ref{['aplocate']} marked by dashed lines.
• Figure 2: UE-centric CF-mMIMO network architecture.
• Figure 3: Path loss vs. distance for the example AP in the city of Amsterdam in Fig. \ref{['snr']}, with different path loss models.
• Figure 4: Heatmap of SNR spatial distribution for different cities with different channel models, for an example AP (red cross).
• Figure 5: Spectral efficiency of CF-mMIMO with different AP selection strategies for different city areas: raytracing vs. log-distance channel model ($Q=5$).