A POMDP approach to opportunistic spectrum access with feedback

This paper considers opportunistic spectrum access (OSA) at a cognitive network. Since the sensing capacity of a user is limited, the user cannot obtain full observations of the spectrum, this system is usually formulated as a partially observable Markov decision process (POMDP) problem. However, the reward (or cost) of a user in a time slot is also unobservable and it´s replaced by the expectation in the traditional algorithms, now, a feedback called error check is added at the ending of the time slot to inform us whether the spectrum access is successful. Consequently, the accurate reward in a time slot can be given and the observations are more precise. The goal is to maximize the expected data throughput, based on the past history of sensing and selection of channels and error checks. Compared with the existing algorithms, it is illustrated through numerical evaluations that this approach performs better.