Comparison of parallel discrete fourier transform implementations for modelling of optical communication systems

Discrete Fourier Transform (DFT) is a well-known mathematical tool used for scientific and engineering purposes. One of the areas, where DFT is frequently used is digital signal processing (DSP). The application of DSP techniques is used in many diverse areas, from telecommunication sector through medical data processing up to modern home entertainment systems. Many of the DSP applications require real time data processing. In the scientific area the DFT is frequently used for modelling of electrical circuits, communication systems and other areas, where the direct (from time domain to frequency domain) and inverse transformations are performed. This contribution presents two available libraries to accelerate DFT using parallelization. Presented libraries are implemented for use on recent computer processors (Central Processing Unit, CPU), as well as for the calculation of the graphics cards (Graphic Processing Unit, GPU). We briefly introduce the libraries and then present the comparison of their performance.