Convergence performance of binary adaptive detectors

Convergence performance results associated with the transient of the performance measures (probabilities of false alarm and detection) for three multiple-observation binary adaptive detectors are presented. The multiple-observation binary adaptive detector consists of a selected single-observation adaptive detector followed by a binary integrator/detector (J out of M detector). Three types of single-observation adaptive detectors are considered: nonconcurrent mean level adaptive detection (MLAD), concurrent MLAD, and the generalized likelihood ratio test (GLRT). The desired input signal is modeled as a Swerling II target and the input noises as Gaussian random variables (RVs). Detection performance P _D of each binary adaptive detector is evaluated as a function of the number of input channels N, the number of independent input sample vectors-per-channel used to estimate the unknown input covariance matrix, the order of the binary detector M, the desired probability of false alarm P _F, and the matched filter output signal-to-noise(SIN) power ratio. Tables of detection performance are provided that will aid in specifying the optimal J* for the J out of M detector and finding the number of input samples-per-channel K* necessary to achieve a 3 dB loss in optimal performance for a given P _D, P _F, M, N, and single-observation detector configuration. Significantly, it was found that K* and J* are relatively invariant of the single-observation detector configuration and the chosen P _D.<>