5th International ICST Conference on Cognitive Radio Oriented Wireless Networks and Communications

Research Article

High performance FFT on multicore processors

Download263 downloads
  • @INPROCEEDINGS{10.4108/ICST.CROWNCOM2010.9283,
        author={J. Barhen and C. Kotas and T.S. Humble and P. Mitra and N. Imam and M.A. Buckner and M.R. Moore},
        title={High performance FFT on multicore processors},
        proceedings={5th International ICST Conference on Cognitive Radio Oriented Wireless Networks and Communications},
        publisher={IEEE},
        proceedings_a={CROWNCOM},
        year={2010},
        month={9},
        keywords={FFT  cognitive radio  multicore procesors  optical core processors  transverse vectorization},
        doi={10.4108/ICST.CROWNCOM2010.9283}
    }
    
  • J. Barhen
    C. Kotas
    T.S. Humble
    P. Mitra
    N. Imam
    M.A. Buckner
    M.R. Moore
    Year: 2010
    High performance FFT on multicore processors
    CROWNCOM
    IEEE
    DOI: 10.4108/ICST.CROWNCOM2010.9283
J. Barhen1,*, C. Kotas1, T.S. Humble1, P. Mitra1, N. Imam1, M.A. Buckner2, M.R. Moore2
  • 1: Center for Engineering Science Advanced Research, Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States of America
  • 2: RF and Microwave Systems Group, Measurement Science & Systems Eng. Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States of America
*Contact email: barhenj@ornl.gov

Abstract

Importance of achieving high performance Fourier transforms for Cognitive Radio applications can not be over-emphasized. This includes signal detection in the presence of noise power uncertainty, multi-resolution spectrum sensing, minimization of subcarriers' side lobes in OFDM modulators, multi-stream processing, or spectrum loading, for example. With the emergence of advanced multicore processors, there is a remarkable opportunity to develop novel, massively parallel implementations of the FFT. This paper reviews recent advances in the area, and presents results for three classes of devices: the IBM Cell multi-SIMD processor, the Nvidia Tesla SIMT processor, and the EnLight digital optical core device.