T.A. Lepekhina, V.I. Nikolaev, M.A. Semyonov, E.F. Tolstov

Enlargement of SAR digital radiohologram volumes leads to the need of data processing rate improvement. Radiohologram processing in near-real time requires the ability of data processing time pre-estimation. The most computationally intensive operation is Fast Fourier Transform (FFT) calculation, FFT classic version performance having a complex dependence on a variety of factors. The FFT algorithm for general-purpose processors presented in this paper is designed to optimize the use of basic computer resources, ensuring high and stable processing speed. The algorithm is based on two concepts: special input/output data representation with the object to optimize utilization of arithmetic logic unit and cache memory data transfer rate, and predetermination of cache memory usage for maximum cache filling and minimum RAM using. After optimization, the most limiting factor is RAM memory bandwidth and at the same time there are at most three actual memory calls for strings of 2k elements, where k≤21. The developed algorithm can be used both in the imaging software for SAR ground segment and as part of the software for flight tests and validation.