Adaptive cascade filter for speech analysis

Two techniques are described for sequentially adapting a finite impulse response digital filter which is constructed as a cascade of 2nd-order sections. It is shown that the radius and angle co-ordinates of each zero pair may be separately adapted. The first method uses a modified LMS algorithm, while the second, by employing prequantised coefficients, avoids the need for a control parameter. The performance, when used for linear prediction of speech, is discussed.