任意波形の正逆フーリエ変換ニューラルネットワーク

概要

論文の詳細を見る
A new method for construction of multi-layered neural networks capable of direct and inverse Fourier transformations of an arbitrary wave form is developed. In this method, a direct or inverse Fourier transformation neural network with one point input is arranged individually at every point of the sampled wave, or at the real or imaginary part of every complex spectral line, respectively. Each network is then individually trained using a certain number of input data for expressing an arbitrary wave form or spectrum. The direct and inverse Fourier transformations concerning the entirety of an arbitrary wave form or spectrum are obtained by superposing the transformation results by the individual point networks already learnt enough. The neural networks constructed in this way successfully achieved the direct and inverse Fourier transformations of a wave sampled at 16 points.
社団法人人工知能学会の論文
1991-09-01