Simulation of CFAR detection algorithms for arbitrary clutter distributions

Simulation and performance estimation methodologies are developed for constant false-alarm rate (CFAR) detection algorithms based on the powerful concept of importance sampling (IS). Such algorithms involve crossings of a random threshold. Compression of the threshold density function produces the required biasing to implement IS procedures. Adaptive optimisation of simulation estimators and estimation of detector threshold multipliers are described. Easily computable approximations for false-alarm probabilities (FAPs) of cell averaging (CA)-CFAR detectors are derived. Fast simulation results are described for examples with known clutter distributions. The practically important situation when clutter densities are unknown is dealt with. Algorithms blind to the density and having appreciable gains over conventional Monte Carlo simulation are demonstrated. In a limited experiment these are shown to track a step change in clutter distribution. It is argued, albeit tentatively, that the procedures could point the way to implementation of estimators for FAPs and their control through threshold adaptation