A. L. Bulyanitsa, V. E. Kurochkin
Institute for Analytical Instrumentation RAS, Saint-Petersburg
The paper presents the analysis
of Ya.Z. Tsypkin's modification of the Robbins—Monro stochastic approximation
algorithm to evaluate constant signals in the presence of additive noise
with an a priori unknown distribution law, including statistical
outliers. The unbiasness property and the efficiency of estimation are
evaluated and economical hardware realization of the algorithm is considered.
The software version of the algorithm offered allows, in addition to signal
estimation, algorithm parameter adjustment and initial approximation of
the signal estimate, as well as detection of disorders in the measurement
sequence.