In high resolution sensor systems, impulsive clutter returns raise the tail of the background distribution. Processing such non-Gaussian data with a Gaussian detector increases the number of false alarms. One solution to this problem involves approximating the background distribution at the output of the signal processor with a Spherically Invariant Random Vector (SIRV) density function. The SIR V-based model allows for application of the likelihood ratio which correctly matches the non-Gaussian data. This paper demonstrates the utility of the SIRV model by analyzing real recorded sonar clutter. Processing these non-Gaussian returns with the SIRV likelihood ratio leads to a ten-fold reduction in the number of false alarms as compared to processing with a traditional Gaussian detector.
展开▼
