摘要 |
An adaptive filter has the filter weights adjusted by progressively smaller steps as the final condition is approached, thereby combining rapid convergence with low residual error.
<??>The signal may be analog, and converted to digital form. A reference signal is obtained by delaying the input signal, and this si filtered in filter (700), subtracted from the undelayed signal in a subtractor (703), and the error signal, representing the difference, is used for computing the changes in filter weights.
<??>When the filter is initialised a set of scaling factors mu l is set and the increments DELTA W l in the weights are computed in known manner. When these increments repeatedly change sign, an updating circuit (708-L) reduces the scaling factors so that the increments are reduced in magnitude as the filter approaches its final setting. If the increments are successively of the same sign, the scaling factor is increased.
<??>An error threshold detector circuit (709) may be provided which detects a condition in which the error signal e(k) suddenly increases, and this may be used to reset the scaling factors mu to their initial values.
|