Learning methods for temporal-spatial opportunistic spectrum access in cognitive radio networks
Date of Issue2017-02-27
University of Technology of Troyes
As a natural resource, the radio spectrum is usually regulated by government agencies and static spectrum allocation policies are widely adopted by most countries. With the increasing popularity of mobile devices and need for high speed data transmission, static spectrum allocation policies can no longer satisfy all demands for spectrum. In cognitive radio networks (CRNs), opportunistic spectrum access (OSA) alleviates the spectrum under-utilization problem by allowing unlicensed secondary users (SUs) to identify and exploit the unused spectrum owned by primary users (PUs) temporally and spatially while limiting the interference to PUs below a predefined threshold. Designing effective methods for temporal-spatial OSA is thus crucial for improving spectrum utilization nowadays. We first consider the problem of estimating the no-talk region of the PU for temporal-spatial OSA, i.e., the region outside which SUs may utilize the PU's spectrum opportunistically regardless of whether the PU is transmitting or not. Based on a distributed learning framework, we propose a distributed boundary estimation algorithm that allows SUs to determine the boundary of the no-talk region collaboratively through message passing between SUs. We analyze the trade-offs between estimation error, communication cost, setup complexity, throughput and robustness. Simulation results suggest that our proposed algorithm have lower estimation errors and better robustness compared to various other methods. Within the no-talk region of the PU, SUs who do not interfere with each other can make use of the same PU channel. We then formulate and study a multi-user multi-armed bandit (MAB) problem that exploits the temporal-spatial OSA of PU channels for these SUs located inside the region. We first propose a centralized channel allocation policy for finding an optimal channel allocation and learning the statistics of the channels. We show that this policy is order-optimal with logarithmic regret, but requires solving a NP-complete optimization problem at exponentially increasing time intervals. To overcome the high computation complexity at the central processor, we also propose heuristic distributed learning policies that however have linear regrets. We compare the performance of our proposed policies with other distributed policies recently proposed for temporal (but not spatial) OSA. Simulation results suggest that our policies perform significantly better in terms of average regret than the benchmark policies. Finally, we also propose three collaborative channel learning policies for temporal-spatial OSA, which embed collaboration in the channel statistics learning process. We identify spectrum access opportunities via information exchange among neighboring SUs by applying consensus algorithms on their channel sensing observations, empirical estimates of channel idle probabilities and estimated channel ranks. We compare the performance of these policies with a distributed channel allocation policy. Simulation results suggest that our proposed collaborative policies outperform the distributed access rank learning policy which does not consider collaborations in the learning process.
DRNTU::Engineering::Electrical and electronic engineering