On Optimal Minimax Jamming and Detection of Radar Signals

A game between an intelligent jammer J and decision maker DM is considered. DM seeks to detect a coherent slowly fading narrowband signal under a Neyman-Pearson criterion. His observations are corrupted with additive narrowband noise, the source of which is J's jamming with a power constraint, but otherwise almost arbitrary statistics. DM knows J's action but the converse is not true. When the number of samples increases asymptotically, a minimax solution for the game exists where the jamming is Gaussian, independent of the desired signal amplitude level and probability distribution. The same result also holds for detection of a nonrandom baseband signal.