Jamming-Resistant Multiradio Multichannel Opportunistic Spectrum Access in Cognitive Radio Networks

For achieving optimized spectrum usage, most existing opportunistic spectrum sensing and access protocols model the spectrum sensing and access problem as a partially observed Markov decision process by assuming that the information states and/or the primary users' (PUs) traffic statistics are known a priori to the secondary users (SUs). While theoretically sound, the existing solutions may not be effective in practice due to two main concerns. First, the assumptions are not practical, as before the communication starts, PUs' traffic statistics may not be readily available to the SUs. Second and more serious, existing approaches are extremely vulnerable to malicious jamming attacks. By leveraging the same statistic information and stochastic dynamic decision-making process that the SUs would follow, a cognitive attacker with sensing capability can sense and jam the channels to be accessed by SUs, while not interfering PUs. To address these concerns, we formulate the antijamming, multichannel access problem as a nonstochastic multi-armed bandit problem. By leveraging probabilistically shared information between the sender and the receiver, our proposed protocol enables them to hop to the same set of channels with high probability while gaining resilience to jamming attacks without affecting PUs' activities. We analytically show the convergence of the learning algorithms and derive the performance bound based on regret. We further discuss the problem of tracking the best adaptive strategy and characterize the performance bound based on a new regret. Extensive simulation results show that the probabilistic spectrum sensing and access protocol can overcome the limitation of existing solutions and is highly resilient to various jamming attacks even with jammed acknowledgment (ACK) information.