Adaptive Link Rate Selection for Throughput Maximization with Batched Thompson Sampling

In this paper, we investigate the link rate selection for point-to-point throughput maximization under unknown channel statistics. The transmitter has limited policy switch times due to the constrained deployment opportunities. In order to maximize the expected system throughput, the wireless system must identify the optimal link rate as soon as possible. To meet this demand, we regard the time between two consecutive policy switching slots as a batch, and formulate the problem into the batched online sequential decision making framework. In particular, we propose two algorithms called the Modified/Constrained Batched Thompson Sampling (MBTS/CBTS), and show the expected regret is order-optimal with only logarithmic batches. Simulation results are provided to validate that the proposed algorithms can achieve a better regret and policy deployment trade-off compared with current state-of-the-art methods, i.e., greatly reducing the policy switch times with little regret growth.