IMPROVEMENT OF CONVERGENCE SPEED IN ADAPTIVE FILTER BY MULTIRATE PROCESSING

Exact estimation of frequencies of sinusoidal signals corrupted by noi se is an important problem in a variety of areas, and various kinds of methods for the solution of this problem have been proposed. Among th ese methods, the adaptive spectral estimation method, which processes data recursively, needs consideration of the convergence property of t he adaptive filter as well as the precision of the spectral estimation . In this paper, we propose a method which improves the technique of a nalytic signal processing used to improve estimation precision slowing the convergence speed of the adaptive filters. This method makes good use of the data discarded by decimation and processes two series of t he same statistics in parallel. By this operation, the number of times the coefficients are updated in the adaptive filter is doubled and, t herefore, the rate of convergence is much improved. Furthermore, by in creasing the number of data by interpolation of the observed data, fas ter convergence speed than that of the conventional method is realized . This method has an advantage of simple structure. It can be also eas ily applied to other adaptive filtering using subband filters.