Learning Algorithms for Channel Allocation in Multi-Transceiver Systems

In this work, we aim to design algorithms that learn the unknown service rates of the available channels and allocate them to different transceivers in the system. The objective is to minimize queue-length regret, defined as the difference between the cumulative queue length under a learning algorithm and that under the optimal (genie) policy, which knows the channel service rates. Towards this end, we propose the LUCB (Limited Upper Confidence Bound) algorithm, which carefully combines the traditional UCB algorithm with greedy exploitation and carefree exploration strategies. We conduct a detailed simulation study to demonstrate the efficacy of the proposed LUCB algorithm compared to the Traditional UCB (TUCB) algorithm commonly used in Multi-Armed Bandit (MAB) settings.