Representation and Analysis of Signals. Part XXX. Robust Multi-Input Processors.
JOHNS HOPKINS UNIV BALTIMORE MD DEPT OF ELECTRICAL ENGINEERING
Pagination or Media Count:
A statistical procedure which maintains high performance over a class of noise distributions is called robust. The dissertation establishes the identity between multi-input signal processors which are asymptotically optimal in a density minimizing Fisher Information and those which are robust in the sense of maximizing the minimum performance over the class of noise distributions. The problem considered is more general than that treated by previous investigators. Maximin robust detectors of coherent and noncoherent location shift in p-variate noise are derived. A multi-variate scale-related problem, equivalent in the univariate case to the non-coherent location shift situation, is also treated. A generalized soft limiter, shown to be maximin robust against a contaminated normal distribution, is analyzed detail and a performance table is given. Author