Сокращение времени вычисления операции секционной свертки с помощью учета практической производительности быстрого преобразования Фурье

Abstract

In digital signal processing the question of the speed of the algorithms speed used is becoming more and more relevant. The convolution and correlation operations used are most often based on standard function libraries, which are focused on reducing data processing time by splitting the source data into sections. With small amounts of data, these algorithms work quite efficiently. However, in practice, with a significant increase in the dimension 
 of the input data, the methods lose quite a lot in the speed of data processing. A method for calculating the convolution of large signals based on the practical performance of the fast Fourier transform is proposed. The optimal size of the section is analyzed, in which the practical performance of existing algorithms remained at a sufficiently high level. Based on the experimental calculations carried out, the optimal dimension of the section used in the convolution calculation formulas was chosen. The proposed method has been tested on published data from various studies. The significant advantages of the proposed method in solving a number of problems are the reduction of the convolution calculation time for long signals by tens of percent and the possibility of fine-tuning the method for specific computing platforms when using preliminary run-time testing on a fast Fourier transform platform of various sizes