Learning-based dynamic channel selection for Opportunistic Spectrum Access

Opportunistic Spectrum Access (OSA) has recently been proposed to enhance wireless spectrum utilization. The challenge of OSA is that each user can search a channel that could provide a higher throughput by probing more channels, while it also decreases the opportunity of channel access. Even though many studies tent to make a best trade-off between channel probing and transmission opportunity, it is still a waste to spend time for channel probing without gaining any throughput. Therefore, in this paper, we propose a novel concept to cope with such inefficiency by letting secondary users hop among different channels and learn channel quality from experience without the overhead of channel probing. We first apply the Lagrangian relaxation technique to approximate the solution of optimal channel selection. Next, a distributed subgradient-based algorithm is proposed to enable each user to adapt its channel selection to the variation of channel conditions. The simulation results demonstrate that the proposed algorithm allows each user to exploit only local information to select a suitable channel efficiently in a distributed manner.