On quick change detection in finite state spaces based on sequential decisions

The aim of the paper is to adapt some basic algorithms of detection theory used in communication systems, as solutions in the process of change detection and diagnosis based on statistical signal processing paradigm. Two methods are considered: a classical one based on maximum a posteriori probability detection rule and, respectively, a sequential detection procedure. Both methods are studied in the case of multiple hypotheses, i.e. a finite multidimensional decision space. Three base signals are used as state descriptors: constant levels, harmonic signals, and a weighted sum of cosines based signals. The observation noise is band limited and the signal to noise ratio is varying from -10 to + 20 dB. Theoretical results for average probability errors are established. Computer based simulation shows better results for sequential based decisions that open the gate for direct application in the process of quick change detection, for cases with known (a priori or- learned) states.