Statistical normalization of spherically invariant non-Gaussian clutter

摘要: Conventional detection in active sonar involves comparing the normalized matched filter output power to a fixed preset threshold. Threshold crossings from contacts of interest are labeled as detections and those undesired clutter echoes false alarms. To maintain constant false-alarm rate (CFAR) presence strong transient clutter, system can either increase threshold or apply some function that suppresses this background down an acceptable level. The latter approach leads more consistent on display, which enables operator-assisted detection. Background suppression should not come at expense contact detection; maximize probability (PD) for given alarm (PFA), likelihood ratio test (LRT) is used. However, LRT does address display issues, since achieves desired PFA varies with input distribution. Ideally, monotonically transformed using "statistical normalizer" (SN) returns CFAR without degrading optimized PD. Within radar community, proposed tuned K-distributed spherically invariant random vector (SIRV) model. model lend itself SN, closed-form expression density exist. In contrast, SIRV model, Pareto distributed power, SN readily derived. This combined Pareto-LRT/SN detector nearly matches PD performance maintains purposes.