Stable Multiuser Channel Allocations in Opportunistic Spectrum Access

We consider the distributed channel allocation problem in an asymmetrical opportunistic spectrum access (OSA) system where each secondary user possibly has different channel reward even in the same channel due to geographic dispersion. We formulate this problem as a Gale-Shapley stable theorem using game theory to optimize the sum reward of all secondary users. It is challenging to achieve the stable matching of user-channel pairs without centralized control and prior knowledge of channel availability statistics. In this paper, we present a novel decentralized order-optimal learning Gale-Shapley scheme (OLGS) in which secondary users learn from their local history data and individually adjust their behaviors in a time-varying OSA system. The proposed scheme eliminates collisions among secondary users by a one-to-one user-channel matching policy. It also achieves stable spectral allocations using learning method without assuming known channel parameters and independent of information exchange among secondary users. Simulation results show that the system regret of the OLGS solution grows with time at the logarithmic order with low complexity.