The Implementation of Large FFT on Homogeneous Multi-core Systems

Large FFT is widely applied to high throughput and high real-time applications such as image processing, radar signal processing, satellite communications and biomedicine. At the same time, the development of multi-core system has been the mainstream in the past few years. So it is valuable to study the implementation of large FFT on the multi-core systems which is high efficient and low resource consumption Extremely. This paper presents one implementation of large FFT on homogeneous multi-core system and achieves a good balance between the on-chip memory consumption and the payload in the process elements (PE). This paper adopts the 2-diamension FFT principle and radix-2 decimation-in-time (DIT) FFT method, which divides the large FFT into several shorter ones and computes each part with the process elements (PE) on the homogeneous multi-core system. In order to reduce the time of processor execution and reduce programming tasks, this paper make the source data and the results of data of each butterfly access to the same storage area skillfully. In addition, a pipeline structure is introduced for multiple FFTs computing at the same time aiming at improving the parallelism.