Language-based cost functions for fully adaptive radar under imprecise performance standards

There is a push to create radar systems that optimize their own performance in real-time. In order to accomplish this, the performance standards of the radar system must be known. Often, this is given in imprecise, natural language statements describing the goals and requirements of the radar system by individuals in charge of the “big picture.” In this paper, we use a previously developed method called language-based cost functions (LBCFs) to optimize the performance of a fully adaptive radar target tracker when performance standards are given by vague statements in natural language. This is followed by a catch-all stochastic algorithm to find relaxed, time efficient solutions. The full method is then tested on a basic simulation of a human running. Results show that the combination of LBCFs and the optimization algorithm allow the radar to far outperform a classical non-adaptive radar. Results also show that the specific optimization algorithm used impacts the radar's ability to find satisfactory solutions which in turn dramatically alters the tracker's performance.