Learning-based adaptive design for dynamic spectrum access in cognitive radio networks.

Abstract

This thesis is concerned with dynamic spectrum access in cognitive radio networks. The main objective is designing online learning and access policies which maximize the total throughput of the secondary users in a cognitive radio network. As the first approach, we consider the auction-based formulation in design of dynamic spectrum access mechanisms where it is assumed that primary channels are heterogeneous with distinct availability statistics unknown to each secondary user (SU). Considering this approach, we first apply a unit demand (UD) auction which is called DGS (Demange- Gale-Sotomayor) auction. Applying the DGS auction, we explore the instantaneous link condition of each SU for its throughput maximization. To tackle the issues of this UD auction, we propose a learning-based unit demand (LBUD) auction. Our proposed auction mechanism incorporates a distributed learning of the primary channels into the auction mechanism to explore both primary channel availability statistics and instantaneous link gains of the SUs for their throughput maximization. This new mechanism substantially improves the communication overhead and also the SUs’ throughputs where the primary channels have dissimilar availability statistics. The proposed LBUD auction preserves the strong property of the UD auction, i.e., it is dominant strategy incentive compatible. To improve convergence speed of the iterative procedure used in the auction, we further propose an adaptive price increment algorithm. Simulation results show the effectiveness of the proposed LBUD auction mechanism in terms of the throughput gain.

As our second approach, we model the problem of designing decentralized dynamic spectrum access policies as a decentralized multi-armed bandit(DMAB) problem. Using DMAB formulation, we first propose a truly decentralized online learning and access policy where in addition to channel availability statistics, the secondary user population is also assumed to be unknown to the SUs. To reduce collision events at different learning stages, we then improve an existing access policy by exploiting a ”perceived population” by each secondary user. We also develop a distributed learning and access policy which is effective in a wide range of primary channel conditions. As our last approach, we investigate designing of a decentralized online learning and channel access in a cognitive radio network with M secondary users. We formulate the distributed channel selection problem in a cognitive network as a strategic game which is proved to be an exact potential game. Applying stochastic learning automata, we propose an adaptive decentralized access policy where each SU probabilistically selects one of the M-best channels to access. Based on collision events, we update the channel selection probability. In our proposed adaptive policy, two underlying distributed learning algorithms are utilized in parallel: i) Learning from sensing history on the primary channel availability, and ii) Learning from collision history on channel selections among SUs to avoid further collision. Simulation results show the effectiveness of our proposed adaptive policy in various distributions of mean channel availabilities across primary channels, as compared with other existing policies.