Multicell Power Control Under Rate Constraints With Deep Learning

In the paper we study a deep learning based method to solve multicell downlink power control problem for sum rate maximization subject per-user constraints and per-base station (BS) constraints. The core difficulty of this is how ensure that learned results by neural network (DNN) satisfy To tackle difficulty, propose cascade projection block after traditional DNN, which projects infeasible onto constraint set. designed on geometrical interpretation constraints, low complexity, meeting real-time requirement online applications. Explicit-form expression backpropagated gradient derived proposed block, with DNN can be trained directly maximize via unsupervised learning. Simulation demonstrate advantages over existing numerical optimization methods, show robustness model mismatch between training testing datasets.