Joint Congestion Control and Routing Subject to Dynamic Interruptions in Cognitive Radio Networks
Cognitive radio networks suffer from dynamic interruptions from primary users. The joint congestion control and routing are tackled using stochastic control techniques. Centralized dynamic programming is applied for the primal optimization, which provides a performance upper bound. Q-learning is applied when the primary user knowledge is unknown. Dual optimization based decomposition is used to decentralize the stochastic control. A heuristic scheme based on the Limited Look-a-head Policy (LLP) and binary pricing is proposed to tackle the prohibitive difficulty in the dual optimization. Numerical simulation shows that the proposed algorithms achieve the optimal or near-optimal performance.