Power Control in Cellular Massive MIMO with Varying User Activity: A Deep Learning Solution

This paper demonstrates how neural networks can be used to perform efficient joint pilot and data power control in multi-cell Massive MIMO systems, by exploiting the problem structure. We first consider the sum spectral efficiency (SE) optimization problem for systems with a dynamically varying number of active users. Since this problem is non-convex, an iterative algorithm is first derived to obtain a stationary point in polynomial time. We then use this algorithm together with deep learning to achieve an implementation that provable can be used in real-time applications. The proposed neural network, PowerNet, only uses the large-scale fading information to predict both pilot and data powers. One key feature is that PowerNet can manage a dynamically changing number of users per cell without requiring retraining, which is not the case in prior works and thus makes PowerNet an important step towards a practically useful solution. Numerical results demonstrate that PowerNet only loses $2\%$ in sum SE, compared to the iterative algorithm, in a nine-cell system with up to $90$ active users per in each coherence interval, and the runtime was only $0.03$ ms on a graphics processing unit (GPU).