Intelligent Spectrum Anti-Jamming With Cognitive Software-Defined Architecture

Malicious jamming in congested, contested, and constrained wireless environments has stimulated the demand of jamming-resistant communications. Nevertheless, classical anti-jamming techniques suffer from not only an extremely low data rate but also the risk of being interfered by the recently developed high-power jammers. As a promising solution, dynamic spectrum access (DSA) with software-defined platforms and cognitive functionality has been conceptualized, while systematic design remains an open issue. To this end, in this article, we propose a cognitive software-defined network (CSDN) architecture. Based on this, spectrum semantic extraction is developed to remove irrelevant spectrum features, such that learning a frequency preconfiguration policy (for DSA) can directly utilize the spectrum occupancy information. In order to learn the policy online, we propose a novel kernel-based distributed actor-critic algorithm, where the kernel method and online feature admission are exploited to make policy learning more adaptive. Numerical results show that semantics-driven reinforcement learning (RL) makes policy learning more efficient and accurate. Moreover, the proposed RL outperforms various baseline algorithms in terms of the strength in adapting to dynamic jamming patterns as well as the convergence speed of policy learning. Finally, a testbed is prototyped to validate the CSDN-based intelligent spectrum anti-jamming communication system.