Optimizing non-recursive digital filters to non-liner phase characteristics

Various methods based on optimization have been used to design linear phase filters. One such method has been to use a general-purpose optimization program to minimize some error criterion, a function of the filter coefficients and of the error between the specified and achieved gain responses. However, if this were to be used with arbitrary phase designs, the error criterion would have to be formulated as a function that combines the gain and phase errors in a meaningful way. It is shaown here that this particular difficulty can be avoided by regarding the phase specification as a deviation from the linear phase and splitting the characteristic into real and imaginary components, rather than gain and phase, and optimizing separately.