Computational efficiency of number-theoretic-transform-implemented finite-impulse-response filters

The computational merits of using number-theoretic transforms to spectrally implement f.i.r. filters are discussed. It is demonstrated for all but low-tap filters that the spectral method can significantly reduce the multiply load and modestly reduce the add load by comparison with the usual time-domain method and the recursive `moving-window¿ finite-field spectral method.