A new gradient approximation method for the design of minmax FIR filters with powers of two coefficients

A new local search algorithm for the design of FIR linear phase digital filters with sum or difference of two powers of two coefficients is presented. The order of coefficient selection is based on the gradient approximations of the minmax cost function. The aim is to confine the search along the low gradient directions. The performance test of the algorithm is done against a similar local search method which has no specific coefficient ordering method.