Distributed Active Sonar Detection in Dependent $K$-Distributed Clutter

Distributed active sonar systems exploit the simultaneous detection of targets on multiple receiver platforms to improve system performance. False alarm performance modeling for such systems typically assumes independence from sensor to sensor; however, false alarms from clutter (e.g., shipwrecks, mud volcanoes, or rock outcroppings) are expected to produce statistically dependent data. In this paper, a clutter model exhibiting intersensor dependence and having K-distributed marginal probability density functions (pdfs) is proposed and analyzed. Under the constraint of equal sensor-level performance, the m-of- n fusion processor is seen to be optimal for most cases of interest for all but the heaviest tailed clutter. The system-level probability of false alarm is derived for the m-of- n fusion processor and approximations are developed for the system-level probability of detection for common target models. In analyzing fusion performance, the and processor (m = n) is seen to perform best in heavy clutter when the target model has some measure of intersensor consistency while m/n ap 0.25 is seen to perform best for the highly variable Gaussian target model. The performance of AND, OR (m = 1) , and median fusion processors is seen to degrade as the clutter pdf tails increase, although the effect is confounded between clutter tail heaviness and intersensor dependence. The importance of accounting for intersensor dependence, however, is illustrated by an up to 10-dB overestimation of performance when heavy tailed clutter data are incorrectly assumed to be independent across sensors.