Bayesian Non-Parametric Active Learning for Information Asymmetric ECCM in Cognitive Radar Electronic Warfare

In this paper, we develop a Bayesian non-parametric active learning algorithm for electronic counter-countermeasure in cognitive radar electronic warfare. An adversarial target employs electronic countermeasure through a jammer to degrade the tracking performance of a cognitive radar. We consider an information asymmetric scenario where the jammer has complete knowledge of the cognitive capabilities of the radar, tracking the target. We model the radar-jammer interaction using the Principal-Agent Problem (PAP) framework. The response of the jammer is modeled through Gaussian process (GP), a Bayesian non-parametric learning framework. Using the GP model, we develop an active learning algorithm (GP-UT-ECCM) for efficiently learning the jammer response. Numerical results illustrate that GP-UT-ECCM outperforms existing methodology in terms of cumulative regret and SINR performance criteria, while simultaneously reducing radar-jammer interactions. GP-UT-ECCM is ideal for electronic warfare scenarios requiring fast response time and shorter radar-jammer interactions.