Robust and nonparametric detection of fading narrowband signals

This thesis deals primarily with the design and analysis of algorithms for detection of narrowband slowly fading signals, embedded in noise of uncertain distribution. The signal's phase and amplitude are random, but are essentially constant over the observation interval; in addition, the Doppler frequency of the signal is unknown. This is the most common model of target echoes received by radar or sonar systems. Attention is focused on i.i.d. noise samples whose marginal distribution belongs to an epsilon-mixture family, where the scale (variance) of the nominal part of the distribution is also unknown. This is an important generalization of the model. The purpose of the derived detection algorithms is two-fold: first, to assure constant probability of false alarm, regardless of the noise distribution (DF-CFAR); and second, to achieve optimal maximin probability of detection (Pd) performance.