Centralized and Distributed Power Allocation for Max-Min Fairness in Cell-Free Massive MIMO

TL;DR

This paper introduces neural network-based methods for power allocation in cell-free Massive MIMO systems, enabling efficient, fair, and locally-informed distribution of transmission power to improve user performance.

Contribution

It generalizes max-min fairness power allocation to arbitrary precoding, and develops neural network models for both system-wide and local power allocation in cell-free Massive MIMO.

Findings

Neural networks can approximate system-wide max-min fairness power allocation.

Local neural networks outperform existing distributed methods.

Proposed methods reduce computational complexity significantly.

Abstract

Cell-free Massive MIMO systems consist of a large number of geographically distributed access points (APs) that serve users by coherent joint transmission. Downlink power allocation is important in these systems, to determine which APs should transmit to which users and with what power. If the system is implemented correctly, it can deliver a more uniform user performance than conventional cellular networks. To this end, previous works have shown how to perform system-wide max-min fairness power allocation when using maximum ratio precoding. In this paper, we first generalize this method to arbitrary precoding, and then train a neural network to perform approximately the same power allocation but with reduced computational complexity. Finally, we train one neural network per AP to mimic system-wide max-min fairness power allocation, but using only local information. By learning the structure of the local propagation environment, this method outperforms the state-of-the-art distributed power allocation method from the Cell-free Massive MIMO literature.