Bruno Stuber and Dino Zardet
Xilinx and University of Applied Sciences Northwestern Switzerland

A fast Fourier transform (FFT) is the ultimate way to accomplish spectral analysis in digital signal processing. Today, many standard solutions are offered to perform real-time FFTs, either with DSPs or FPGAs. However, to go beyond established limits, a thorough analysis of the system architecture is necessary, as is a design approach that focuses on timing.

In this article, we will discuss the implementation of a 32k-point FFT (32,768 points) with power spectrum accumulation. The 8-bit input data flows continuously at a rate of 2 gigasamples per second (GSPS), therefore providing an analog bandwidth of 1 GHz. The spectrometer is implemented in a single Xilinx Virtex-II Pro XC2VP70 FPGA operating at an overall clock speed of 125 MHz.