Liskevych R. Universal algorithms of fast ortogonal transformation for parallel information technologies

The dissertation is dedicated to the development of universal orthogonal transformation algorithms for parallel information technologies. New universal fast bicascade algorithms of direct and inverse discrete Fourier transformations of real sequence, Hartley, second-fourth form cosine and sine transformations have been derived. The author shows, that aforecited algorithms could be implemented in a universal computational system. The transformation type in such a system could be chosen by selecting of corresponding phase factors for the two-point base operations only. The process of the phase factors groups assembling is governed by the general rule. Such an approach allows to reduce the expenses on algorithms realization in 1.2-1.6 times. The new algorithmic approach to neural networks of the signal filtration synthesis has been developed. It reduces the network training time essentially (in N/log2N times).