DEVELOPMENT OF DATA PROCESSING TECHNIQUES FOR DAILY SURVEY AT 110 MHZ BY BSA LPI USING HIGH-PERFORMANCE COMPUTING METHODS

Abstract

Purpose: The goal of the work is to develop a workable technique for fast processing of large arrays of radio astronomy data to search for responses to extragalactic transient events, which a priori should have large dispersion (DM ~ 100÷2000 pc · cm-3). Records of 20-day continuous observations of the sky sphere circumpolar zone with an area of about 8 sq deg and separate files with repeating fast radio burst FRB 121102 were used as a test sample of the technique. Design/methodology/approach: The BSA (Big Scanning Antenna) LPI radio telescope has a multi-beam diagram and is capable of recording daily in the frequency range of 109–111.5 MHz in 96 beams in the declination range from – 8° to +42°. The number of frequency bands varies within 6 to 32 for the time constant varying within 0.1 to 0.0125 s, respectively. In the mode of recording 32 frequency bands with the time constant 0.0125 s, 3.4 GB of data are recorded every hour, 87 GB – daily, and 32 TB – per year. To end of August 2018, about 120 TB of data have been accumulated. Processing this amount of data for a number of scientific problems urgently requires the use of high-performance computing technologies. Findings: The paper proposes two ways to process data: using graphics processors (CPU+GPU, calculation in the C/C++ programming language using OpenCL) and using cluster computing (by using the message passing interface MPI on multiprocessor nodes). To refine the methodology, we use both the processing of pulsar data (using graphic accelerators allows us to increase the processing speed by 2–3 orders of magnitude) and search for responses to extragalactic transient events, which a priori should have large dispersion delays. Examples of such events include fast radio bursts (FRB), responses to gamma radio bursts (GRB) and, finally, possible responses to gravitational events recorded by the LIGO detectors. After processing the polar zone, 697 candidates for pulsed events with high dispersion measures were detected. About half of them is caused by man-made interference, the rest being caused both by the usual interplanetary scintillations of radio sources, and also by the possible soughtfor FRBs. It is still difficult to separate these two classes of events. Processing the data observed in the FRB 121102 zone also indicates the presence of event candidates. The results require more thorough analysis. Сonclusions: After finalizing the high-performance computing methodology, it will allow us to process data observed with the BSA LPI multipath diagram immediately after they are recorded.

Keywords

• radio observation
• big data
• processing technique