Detecting Pulsars with Neural Networks: A Proof of Concept

TL;DR

This paper introduces a neural network pipeline that efficiently detects pulsars in radio telescope data, suppresses interference, and corrects dispersion effects, enabling real-time analysis on standard hardware.

Contribution

The authors develop a neural network-based method that improves pulsar detection by suppressing RFI and correcting dispersion, achieving real-time performance on consumer-grade GPUs.

Findings

Capable of identifying faint pulsars with low false positives.

Produces data quality comparable to traditional dispersion measures.

Operates 200 times faster than real-time on test data.

Abstract

Pulsar searches are computationally demanding efforts to discover dispersed periodic signals in time- and frequency-resolved data from radio telescopes. The complexity and computational expense of simultaneously determining the frequency-dependent delay (dispersion) and the periodicity of the signal is further exacerbated by the presence of various types of radio-frequency interference (RFI) and observing-system effects. New observing systems with wider bandwidths, higher bit rates and greater overall sensitivity (also to RFI) further enhance these challenges. We present a novel approach to the analysis of pulsar search data. Specifically, we present a neural-network-based pipeline that efficiently suppresses a wide range of RFI signals and instrumental instabilities and furthermore corrects for (a priori unknown) interstellar dispersion. After initial training of the network, our analysis can be run in real time on a standard desktop computer with a commonly available, consumer-grade GPU. We complement our neural network with standard algorithms for periodicity searches. In particular with the Fast Fourier Transform (FFT) and the Fast Folding Algorithm (FFA) and demonstrate that with these straightforward extensions, our method is capable of identifying even faint pulsars, while maintaining an extremely low number of false positives. We furthermore apply our analysis to a subset of the PALFA survey and demonstrate that in most cases the automated dispersion removal of our network produces a time series of similar quality as dedispersing using the actual dispersion measure of the pulsar in question. On our test data we are able to make predictions whether a pulsar is present in the data or not 200 times faster than real time.