Resource minimization driven spectrum sensing policy

In this paper a reinforcement learning-based distributed sensing policy is proposed for cognitive radio networks. The proposed sensing policy is controlled by a fusion center that employs action-value learning to focus the search for idle frequencies to those parts of the spectrum that persistently provide a high data rate. The fusion center learns the local sensing performances of the secondary users and attempts to minimize the number of assigned users for sensing under a constraint on the global detection probability. A heuristic polynomial time algorithm iteratively employing the Hungarian method is proposed for finding a feasible assignment that minimizes the number of active sensors. Simulation results show that the proposed algorithm is able to find near-optimal solutions in practise significantly faster than an exact branch-and-bound search.