Belief Propagation for Distributed Downlink Beamforming in Cooperative MIMO Cellular Networks

We propose a new graphical model approach to cooperative multiple-input multiple-output (MIMO) cellular networks. The objective is to optimize downlink transmit beamforming at each BS in order to maximize the sum throughput over the entire network. While ideal centralized beamforming requires full channel state information (CSI) sharing among all BSs in the network and huge computational complexity for combinatorial optimization, the proposed graphical model enables distributed beamforming which requires only local CSI sharing between neighboring BSs and efficiently solves the optimization problem in a distributed manner. As distributed solvers for this problem, we derive message-passing algorithms which can be implemented with polynomial-time computational complexity. Furthermore, we make a slight approximation on the objective function to derive a simpler graphical model, providing further complexity saving. Simulation results indicate that the proposed distributed downlink beamforming achieves average cell throughput typically within just 2% of ideal centralized beamforming.