Learning Algorithms for Minimizing Queue Length Regret

We consider a system consisting of single transmitter/receiver pair and N channels over which they may communicate. Packets randomly arrive to the transmitter’s queue wait be successfully sent receiver. The transmitter attempt frame transmission on one channel at time, where each includes packet if is in queue. For channel, an attempted successful with unknown probability. objective quickly identify best minimize number packets T time slots. To analyze performance, we introduce length regret, expected difference between total learning policy controller that knows rates, priori. One approach designing would apply algorithms from literature solve closely-related stochastic multi-armed bandit problem. These policies focus maximizing transmissions time. However, show these methods have $\Omega (\log {{T}})$ regret. On other hand, there exists set queue-length based can obtain order optimal ${O}(1)$ use our theoretical analysis devise heuristic are shown perform well simulation.