Robust continuous-time detection in linear process noise
Rao, Srinivasa Patibandla
Johnson, Don H.
Doctor of Philosophy
Linear processes are suitable for modeling the random received waveforms in a scattering channel, which represents radar, sonar and multipath communication channels. We address the continuous-time detection problem where both the noise and signal-plus-noise waveforms are modeled as linear processes. Uncertainty in the nominal model is considered in the form of classes of probability distributions induced on the function space $L\sb2\lbrack 0,T\rbrack$ by the processes under the two hypotheses. By embedding the linear processes in the larger class of infinitely divisible processes, and using an integral representation for the latter class, we identify the pair of distributions that are least favorable for the discrimination of the two linear processes; an optimal detector designed for these distributions is robust for the uncertainty classes considered. An investigation of a suboptimal performance criterion also leads us to the same robust detector.
Electronics; Electrical engineering