POMDP-based cross-layer power adaptation techniques in cognitive radio networks

We investigate the spectrum access and power adaptation techniques in a cognitive radio network to optimize throughput of a secondary user with specified sensing error limit. Using partially observable Markov decision process framework, we first study the optimal policies, where the primary user is assumed to be in busy, concurrent or idle state, and the secondary user either stay idle or transmits with any of the two designed power level. The collision is avoided with proper reward choices. Although the primary user's states are hidden, their activity statistics, ranges of transmission, and interference thresholds are assumed to be known. The instantaneous optimal policy for each time-slot is then obtained for the current belief of the states obtained through channel sensing. We also propose a forward algorithm based technique that updates belief using the sensor output in the first slot and then using the acknowledgment feedback in the subsequent time-slots in a frame. Simulation results show that the proposed cross-layer technique is more throughput efficient than the physical layer optimal case, specially when the primary user activity is slowly varying and/or frame size is smaller.