Adaptive Nonparametric Detection

The aim of an adaptive detector is to vary the characteristics of the device so that nearly optimal sensitivity is obtained among classes of signals from targets or noises perturbing the measurements. The objective is to reduce the losses inherent in using Bayes average statistics to describe the signals and noise. In order to obtain this functional independence, two sensors are used, only one of which can contain the signal. The technique then is to choose a processing system that is inherently distribution free (nonparametric) and vary the available free parameters of the processor in accordance with the system inputs, to bring the sensitivity of the detector to nearly that which could be obtained if the statistics of both the signal and noise were known. It can be shown that if an optimal nonparametric detector exists it must be of the rank‐order type. For weak signals, the optimum detector uses linear processing of the rankings. In the adaptive detector described in this paper, the linear processor coefficients are varied intermittently in accordance with the results of each successive detection test. Theoretical calculations verified by computer simulation indicate that this type of device has a definitive lower threshold so that no detection can occur at all with signals below the threshold, while for signals above threshold, essentially perfect detection can take place given sufficient data.