Uplink Scheduling for MIMO-OFDMA Systems with Rate Constraints by Deep Learning.

This paper studies the uplink scheduling for multiinput multi-output orthogonal frequency division multiple access (MIMO-OFDMA) systems, aimed at minimizing the number of occupied resource blocks (RBs) subject to per-user rate constraints. This is a combinatorial optimization problem, whose global optimal solution is generally difficult to find due to the prohibitively large search space. To tackle the difficulty, we propose a deep learning based scheduling method, where two deep neural networks (DNNs), namely FeaNet and SchNet, are designed, respectively. FeaNet aims to judge the feasibility of the rate requirements of users given the maximal transmit power and RB resources, while at the same time learn an initial feasible scheduling decision if the scheduling problem is feasible. SchNet aims to learn the optimal scheduling decision by finding a path direction from the initial scheduling decision to the optimal decision. Simulation results demonstrate the performance gain of the proposed method over baseline scheduling methods.