Learning-Based Dynamic Clustering for Coordinated Multipoint Transmission in 5G and Beyond

The rapid growth of data traffic along with the diversity of services/use cases supported by the next generation network, necessitates the requirement of coordination and flexibility in the network so as to take prompt action with the changes in the network dynamics such as time varying channel conditions. To address the issue of sudden degradation in user performance due to channel variations in the network, we focus on dynamic clustering for Coordinated Multi-point (CoMP) transmission. In this paper, we consider an optimal user-centric dynamic clustering algorithm for CoMP with the aim of maximizing the throughput subject to the constraint on the cost of transmission from the CoMP cluster i.e., coordinating set of Base Stations (BSs). An on-line dynamic clustering algorithm for CoMP is proposed based on Q-learning approach. The proposed algorithm does not require explicit knowledge of the transition probability of the channel state variation for implementation. We conducted extensive simulations and simulation results demonstrate the significant performance improvement of the proposed algorithm which converges to the optimal policy obtained from Relative Value Iteration Algorithm (RVIA).