Unsupervised Learning for Scalable Downlink Power Control in Cell-Free Massive MIMO
The paper proposes an unsupervised, physics-informed deep learning framework to control downlink power in cell-free massive MIMO systems. This method directly optimizes max-min fairness without needing user positions or labels. It achieves near doubling of worst-user spectral efficiency compared to existing scalable schemes while scaling linearly with the number of users.