Research And Implementation Of Pulsar Phase Characteristic Search Method For Heterogeneous Computing Platform

In this paper,we use the existing frequency domain dispersion processing algorithm to search for pulsar signals in the frequency domain,to address the problems of computationally intensive and complex pulsar search algorithms.Displayed the pulsar signal characteristics with phase information under the frequency domain data to reduce the overall computational effort.And,a fast and efficient pulsar search algorithm can be implemented on the CPU-GPU heterogeneous computing platform.The details of the work in this paper are as followed.First,we present two phase features that can be employed for pulsar search.Based on the frequency-domain dedispersion processing algorithm,the frequency-domain data processed by the sub-observation channels are calculated.Then,the calculation of characteristic images related to dispersion is analyzed and broadband features commonly used in existing pulsar search methods and a dispersion characteristic display curve that uses multi-channel frequency domain data are given.And the peak width of the dispersion characteristic curve and the characteristics of the curve with the amount of dispersion are discussed.Furthermore,the relationship between amplitude-frequency curves and broadband characteristics is analyzed,and the broadband characteristics of pulsars are effectively characterized using the multichannel amplitude-frequency curve data.On such basis,a multi-channel phase alignment feature is given by combining the dispersion characteristics,and the frequency domain representation of the dispersion characteristics is utilized to demonstrate the broadband characteristics of the pulsar signal.With such characteristics,a pulsar signal search method can be devised using empirical screening and feature filtering techniques.This method utilizes data such as the Parkes Multibeam Pulsed Survey(PMPS)to obtain empirical values for the primary screening.The secondary filtering algorithm chooses signals with Peak-to-Mean ratio based on the large amount of dispersion curve data generated by simulation.The whole algorithm will be implemented on a heterogeneous computing platform by CUDA.After the implementation of the algorithm,the corresponding phase-featured candidate identification model was designed to improve the overall search processing process and reduce manual involvement.To evaluate the algorithm,this study conducted testing and comparison using data from the Galactic Plane Pulsar Survey(GPPS)and the globular cluster survey conducted by the 500-meter Aperture Spherical Radio Telescope(FAST).The experimental results demonstrate that the phase feature search processing algorithm implemented on a heterogeneous computing platform can achieve tens of times of processing speed improvement while achieving similar processing accuracy with traditional search algorithms.The utilization of phase features allows for the avoidance of the fragmented computing architecture commonly found in traditional search algorithms,and makes efficient use of the high floating-point capabilities offered by heterogeneous computing platforms,the overall search algorithm exhibits promising practical application potential.