Exploration vs exploitation for distributed channel access in cognitive radio networks: A multi-user case study

In this paper, we consider the design of efficient strategies that allow multiple secondary users to choose channels with unknown idle probabilities to sense and access without centralized control. Due to the limited sensing capability of CR, the secondary user cannot sense all the channels simultaneously. How to design intelligent sensing strategy is crucial to track the varying spectrum opportunities. However, the availability probability of each channel is not known a prior. The need to learning the information creates a fundamental trade-off between exploration and exploitation. First, the scenario in which a single cognitive user wishes to opportunistically exploit the availability of idle spectrum is considered. An index based strategy from the classical multi-armed bandit problem is efficient to achieve asymptotically optimal performance. Then, the multiuser case is considered. However, the index based strategy for the single user cannot be applied to multiuser scenario directly. If so, collisions among secondary users degrade the overall network performance greatly. We find that randomized selection of the channel to sense is essential to avoid collisions. We extend the exploration and exploitation idea to the multiuser scenario and come up with a randomization based mixed strategy which takes the activity of other secondary users into consideration in the learning process and at the same time achieves exploration and exploitation tradeoff. Numerical simulation results show that the proposed scheme can achieve near optimal in terms of total network performance as the centralized scenario without any information exchange among cognitive users.