Performance and Architectural Tradeoffs in Scalable Cell-Free Massive MIMO

Abstract

The massive number of APs is often perceived as a complexity bottleneck for the scalable deployment of Cell-free (CF) systems. In this context, we propose various system-level results and valuable insights on some of the multidimensional design challenges: the energy efficiency (EE) gap between cellular and the worst-case scenario of scalable CF systems, the interplay between different split processing options, the fronthauling bandwidth, and the offered low-resolution hardware implementations. We discuss the need for physical layer optimizations to trade off performance versus complexity. As a first result, we reveal significant fronthaul bandwidth savings through joint power control and access point scheduling, and proper dimensioning of resolutions for the converters and fronthaul data. In the context of the EE gap analysis, we first provide a novel generalized system model framework that depicts the possibility of all levels of processing with a single system model and allows multiple transmit antennas at each AP and user in the pool of M APs and K users and multistream transmission per user. We formulate a novel optimization framework to maximize the EE where the non-convex fractions corresponding to user performance are coupled with the log-sum function due to the necessity of selecting the optimal number of data streams for each user. To solve this problem, we propose a solution to determine the optimal number of data streams, power allocations, and transmit/receive digital filters. Based on these solutions, we introduce a novel four-step alternating optimization algorithm. Regarding the EE gap analysis, in the worst-case scenario of scalable CF networks, which is of practical interest, CF remains roughly twice as energy-efficient as cellular networks. To facilitate future comparative studies, we also provide a detailed complexity analysis.