The Fast Fourier Transform on a Reconfigurable Processor

This paper describes the implementation of the Fast Fourier Transform (FFT) on the Reconfigurable Data Path Processor (RDPP) under development by the Institute of Advanced Microelectronics and NASA Goddard Space Flight Center. The RDPP implements a multi-stage computational pipeline, optimized for systolic signal processing applications. We implement the Goertzel FFT algorithm, which is able to exploit the data path parallelism of the RDPP. The paper introduces the RDPP and the Geortzel algorithm, describes a Fourier Transform Hyperspectral Image data conversion application, and discusses scaling issues. I. THE RECONFIGURABLE DATA PATH PROCESSOR The emerging field of Reconfigurable Computing seeks to achieve the performance of dedicated hardware with the flexibility of software through architectures that can be reconfigured at run time to optimize them for different computing tasks [5]. The Reconfigurable Data Path Processor (RDPP) is a reconfigurable multiprocessing chip being developed for highthroughput, data-intensive processing tasks aboard spacecraft. The RDPP contains 16 reconfigurable processing elements, each with a multiplier, an arithmetic/logic unit, data path formatting logic, and data path selection logic, operating on a 24-bit-wide data path. In addition, there are five 24-bit bi-directional I/O ports. The RDPP implements a synchronous data flow computational model, optimized for data-intensive processing tasks such as signal processing [1]. The RDPP operates in two phases: (1) configuration, and (2) execution. In configuration, the processing elements are programmed for specific processing tasks, programmable interconnects are configured to form a processing pipeline, and a run-time program is loaded. In the execution phase, the RDPP reads and processes data from an input stream and writes to an output stream.