User Pairing in Network-Coded Cooperative Wireless Communications Using Combinatorial Multi-Armed Bandit Framework

Today, network-coded cooperative wireless communications have found vast functionality. User cooperation provides spatial diversity in the network without forcing any additional hardware cost. User pairing and cooperation of pairs for transmitting information to one another have contributed to the realization of this fact. Choosing the proper pair for each user has a direct effect on the performance and utilization of these networks. Thus, selecting the appropriate pair is one of the most important challenges in these kinds of networks which requires an efficient method to achieve maximum functionality. Assignment problem of users can be formulated as a weighted matching problem in a graph, in which the weight of each edge between nodes shows the achievable throughput of assigning those two users to each other. In this study since the edge weights are unknown and stochastic, the pairing problem is addressed using a learning-based method. This method creates an appropriate contrast between exploration and exploitation and gradually converges to the optimum matching solution. Combinatorial Multi-Armed Bandit framework has been considered distinctly to solve this online weighted matching problem. This algorithm with low sampling complexity can learn optimal user assignment by taking a dynamic approach and also obtains maximum average throughput in long term. Simulation results show that the regret of the proposed algorithm will be next to nothing in the long term.