To reduce a computation amount, or the like, in the convolution computation of the discrete-time signal x[n] of a length N and the filter coefficient h[m] of a length M.
N and M satisfy the condition: M-1≤(P-N)/2. The discrete-time signal x[n] is divided into x1[n] and x2[n]; the length of the x1[n] is set to P/2-(M-1); the length of the x2[n] is set to the remaining length; zero padding is executed to the x1[n] and x2[n]; and the zero padded data x1'[n] and x2'[n] of the length P/2 and Q are obtained. Q is a minimum multiplier of 2, equal to or larger than N2+M-1. Zero padding is applied to the h[m] and zero padded data h1'[n] and h2'[n], corresponding to the x1'[n] and x2'[n] are obtained. FFT is applied to the respective zero-padded data; multiplication is applied to the frequency domain; IFFT is applied to the multiplication result, to obtain two discrete-time signals y1[n] and y2[n]; and they are added through an overlap-add method, to obtain an output discrete-time signal y[n]. Since Q≤P/2 is obtained, computation amount, or the like, is reduced by a non-dividing system.
Eiji Yamada
Eiji Sasaki
Toshio Sawada
Daido Patent Office