Joint Learning and Optimization-Based Resource Management in Hybrid Network of Cooperative and Non-Cooperative Massive MIMO Systems

Multi-cellmultiple-input multiple-output (MIMO) systems have the potential to significantly improve wireless network throughput. However, the cooperative MIMO network experiences the limited backhaul capacity, while the non-cooperative MIMO network requires high fronthaul bandwidth. To overcome these issues, we consider the hybrid MIMO networks to take advantages of both non-cooperative and cooperative MIMO network. In such hybrid system, the decision of whether cells operate in non-cooperative or cooperative mode, as well as the allocated bandwidth for cells are challenging due to the unavailability of channel state information (CSI). Furthermore, cells operating in the cooperative mode should optimize the quantized level selection and power allocation when the backhaul capacity is limited. To address this problem, we propose a joint learning-based and optimization-based framework. Particularly, we employ both large-scale and small-scale fading models to formulate the optimization problems, and then use the obtained solutions of these problems to determine fixed-size features in dynamic networks for the learner. Our numerical results show when the placement decision and the bandwidth assignment for cells are determined, the proposed algorithm can compute the quantized level and allocate power to significantly improve the system performance. Moreover, the proposed hybrid network model controlled by our algorithms outperforms both the non-cooperative and the cooperative MIMO systems in term of net throughput.