Discrete Fourier transform processor based on the prime-factor algorithm

An architecture for a fast DFT processor is presented, which is well suited for implementation using custom or semicustom LSI techniques. A new indexing scheme, which results in an in-place and in-order prime-factor algorithm is first developed. An implementation of a set of four low-order DFT modules based on Winograd algorithms is then proposed. Serial arithmetic, parallel data streams and pipelined computations are used in these modules to give simple circuit configurations with high throughput rates. The low-order DFT modules can be combined without any arithmetic operations to compute DFTs with a wide range of transform lengths. For moderate transform lengths, these modules can be directly cascaded to give high throughput rates. In the case of long transform lengths, they can be combined using the new indexing scheme and intermediate memory. An efficient method of computing DFTs of real-valued data using this processor is also presented.