Poster Abstract

P10.52 Leon Oostrum (ASTRON / UvA)

Theme: Data processing pipelines

The Apertif real-time radio transient pipeline

The Apertif upgrade has transformed the Westerbork radio telescope into a highly efficient survey instrument. The Apertif-LOFAR Exploration of the Radio Transient sky (ALERT) survey is on sky at all times. For processing the 50 GB/s of data, we have built the Apertif Radio Transient System (ARTS). This 160-GPU supercomputer searches all data in real time, providing triggers of interesting events to the outside world within only 10 seconds of the light reaching the telescope receivers.

The same radio transient detected by ARTS arrives at LOFAR a few minutes later. For the most promising candidates, ARTS automatically sends a VO event to LOFAR which then records a few seconds of raw data towards the transient detected by ARTS.

I will show the ARTS real-time pipeline, which has been developed over the past few years at ASTRON, NLeSC, and UvA. The pipeline uses a combination of OpenCL, C, and Python to process all data in real time. Inspection of the initial candidates produced by the GPU code would require over a thousand people, so instead they are further processed based on their metadata, and finally the raw data of any remaining candidates are classified by a convolutional neural network. This process reduces the number of false positives by over a thousand times, so a team of <10 people can still keep up. For the remaining candidates, LOFAR can be also be triggered and the high resolution data are stored for further analysis by the astronomers.