Abstract
In many engineering applications, the need for signal detection often arises. In many such applications, it is advantageous to employ a detection scheme which is relatively insensitive to possible variations in the nominal noise model. That is, it is often desirable to design a "robust" detector. We present a new approach toward robust signal detection which is based on geometric concepts. As opposed to the commonly employed classical saddlepoint techniques, our methods readily admit a quantitative measure of robustness. Initially, we develop robustness measures which are applicable to a parameterized noise family. We then extend these techniques to obtain a measure of robustness which allows for quite general classes of admissible distribution functions. Our approach thus is seen to make possible investigations of the quantitative tradeoff between detector performance and robustness. Throughout our discussions, we illustrate the application of this geometric approach via various specific examples.
Thompson, Michael Wayne (1987). Applications of geometric concepts toward robust signal detection. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -747649.