Adjusting filter coefficients

The weighting coefficients in the coefficient array (3) of a transversal filter (1) are corrected in accordance with the least mean squares algorithm by adding to each coefficient a correction derived from the product of the content of a stage of the filter shift register (2) and the difference between a signal at the filter output (5) and the output signal of a decision element (15). Each coefficient is corrected only once for each n sample periods of a signal applied to the filter input (7), where n is greater than unity, in order to reduce the amount of computation required during each sample period. Each correction is derived from the output of the filter by inserting n of said differences into the coefficient array as respective weighting coefficients and setting any remaining coefficients to zero.