Abstract
Designing a cognitive radar system capable of adapting its parameters ischallenging, particularly when tasked with tracking a ballistic missilethroughout its entire flight. In this work, we focus on proposing adaptivealgorithms that select waveform parameters in an online fashion. Our noveltylies in formulating the learning problem using domain knowledge derived fromthe characteristics of ballistic trajectories. We propose three reinforcementlearning algorithms: bandwidth scaling, Q-learning, and Q-learning lookahead.These algorithms dynamically choose the bandwidth for each transmission basedon received feedback. Through experiments on synthetically generated ballistictrajectories, we demonstrate that our proposed algorithms achieve the dualobjectives of minimizing range error and maintaining continuous trackingwithout losing the target.