On sequential classification of autoregression processes with unknown noise variance

The problem of guaranteed distinguishing of a finite hypothesis number in regard to an autoregression process by direct and indirect observations (with additive noise) is considered. For every hypothesis the autoregression parameters are assumed to be known and variances of a noise process and interferences in the observation channel may take arbitrary values. Sequential classification procedures with guaranteed probability of correct solution are suggested. Asymptotic formulas are obtained for procedure duration.