Pulsar periodicity searching in CUDA

Abstract

A Pulsar, first discovered by Jocelyn Bell in 1967, is a rotating neutron star that can be observed from afar as a pulsating light. The next breakthrough in Pulsar Astronomy came in the form of discovering two pulsars orbiting each other in a stable system. The first binary pulsar, discovered by Russell Hulse and Joseph Taylor in 1974, confirmed the existence of gravitational waves - waves that transport gravitational energy or radiation. In turn, such findings enabled scientists to test Einstein's Theory of Relativity. Pulsars emit electromagnetic radiation that can be recorded as a radio signal. The process of Radio Pulsar Searching involves recording the sky signal over some bandwidth, dividing it into a number of frequency channels and sampling it at a constant time interval. A standard pulsar signal is characterised by extremely periodic dispersed pulsations. As the pulsations are often too faint to be seen individually, one must rely on the integrated signal of many hundreds to millions of single pulses. The Fast Folding Algorithm (FFA) is an integrating function that folds a signal on itself for a number of times at a position P, assumed to be the period of the pulsation. If the peaks are indeed distanced by a period of P, then the folding process at that period will intensify the peaks and make the pulsations a lot more noticeable. The FFA introduces a shift in the phase of the signal as it folds, thereby checking for different periods at one go. By varying the folding period, multiple periods can be searched at one go. The aim of this project is to implement a pulsar periodicity searching tool using Fast Folding Analysis and Concurrent Computing techniques to improve the search speed, thereby producing a fast and accurate module that can be integrated into a pulsar searching package. Fast Folding Analysis consists of a Fast Folding algorithm to fold the signal and a rudimentary signal-to-noise analysis algorithm to check the folded candidates. General Purpose Computation on Graphics Processing Units using CUDA is the technology chosen to Implement this parallel approach with. Using the GPU as a high performance multi-processor is a relatively new approach to supercomputing that, one that presents itself as a far cheaper alternative to the traditional supercomputing architecture.