DISTRIBUTED TWO-DIMENSIONAL FOURIER TRANSFORMS ON DSPs: WITH AN APPLICATIONS FOR PHASE RETRIEVAL
DISTRIBUTED TWO-DIMENSIONAL FOURIER TRANSFORMS ON DSPs: WITH AN APPLICATIONS FOR PHASE RETRIEVAL
Files
Publication or External Link
Date
2006-12-08
Authors
Smith, Jeffrey Scott
Advisor
Jacob, Bruce L
Yeung, Donald
Yeung, Donald
Citation
DRUM DOI
Abstract
Many applications of two-dimensional Fourier Transforms require fixed timing as defined by system specifications. One example is image-based wavefront sensing. The image-based approach has many benefits, yet it is a computational intensive solution for adaptive optic correction, where optical adjustments are made in real-time to correct for external (atmospheric turbulence) and internal (stability) aberrations, which cause image degradation.
For phase retrieval, a type of image-based wavefront sensing, numerous two-dimensional Fast Fourier Transforms (FFTs) are used. To meet the required real time specifications, a distributed system is needed, and thus, the 2-D FFT necessitates an all-to-all communication among the computational nodes. The 1-D floating point FFT is very efficient on a digital signal processor (DSP). For this study, several architectures and analysis of such are presented which address the all-to-all communication with DSPs. Emphasis of this research is on a 64-node cluster of Analog Devices TigerSharc TS-101 DSPs.