Effect of corruption within the recursive estimation of spectral parameters for LPC

Inverse filter parameters may be determined from estimates of the autocorrelation or covariance of a sampled signal by methods which admit corruption of partial results and generate subsequently optimal values for the remaining parameters. Experimental results are presented to indicate the extent to which such techniques can compensate for the effects of irregular or very gross parameter quantization. Three-way comparisons are made between the spectral responses of the inverse filters designed with the globally optimal parameters, corrupted versions of those parameters and parameters determined with the corruption introduced within the estimation algorithm.