Searching for Pulsars with FFTs

From Master Projects
Revision as of 13:18, 28 November 2014 by Schrofer (talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search


About Searching for Pulsars with FFTs


Description

Periodicity search with FFTs

Pulsars are highly-magnetized and rapidly rotating neutron stars, and they are very interesting and elusive astronomical objects. A fundamental step in the process of finding them is the periodicity search: pulsars rotate with a precise period and this needs to be identified in order to distinguish them from other kinds of transient objects. There are different techniques for period analysis and one of them is based on the use of a Fast Fourier Transform (FFT). By using a FFT a time series is transformed from time to frequency domain, as in the frequency domain it's easier to look for a peak in the signal at a specific frequency. This frequency is the period of our pulsar candidate. However, it is not yet possible to use this technique with Graphics Processing Units (GPUs) because the memory required to store a typical input time series is many times the maximum available memory on today's top-tier GPUs. This is a huge drawback for radio astronomy as GPUs can provide the speedup needed for high-resolution searches in a power and cost efficient way.

The goal of this masters project is to research techniques to compute large FFTs (i.e. cases in which the input is far larger than the available memory) and implement them with OpenCL. This FFT OpenCL implementation needs to achieve high-performance and be tunable in order to be used on different platforms (e.g. GPUs from both AMD and NVIDIA).

Another goal of this project is to compare the performance of different many-core implementations of the FFT and study their suitability for big-data analysis.