Pulscan: Binary pulsar detection using unmatched filters on NVIDIA GPUs

Jack White, Karel Adámek, Jayanta Roy, S. M. Ransom, Wesley Armour

Abstract

The Fourier Domain Acceleration Search (FDAS) and Fourier Domain Jerk Search (FDJS) are proven matched filtering techniques for detecting binary pulsar signatures in time-domain radio astronomy datasets. Next generation radio telescopes such as the SPOTLIGHT project at the GMRT produce data at rates that mandate real-time processing, as storage of the entire captured dataset for subsequent offline processing is infeasible. The computational demands of FDAS and FDJS make them challenging to implement in real-time detection pipelines, requiring costly high performance computing facilities. To address this we propose Pulscan, an unmatched filtering approach which achieves order-of-magnitude improvements in runtime performance compared to FDAS whilst being able to detect both accelerated and some jerked binary pulsars. We profile the sensitivity of Pulscan using a distribution (N = 10,955) of synthetic binary pulsars and compare its performance with FDAS and FDJS. Our implementation of Pulscan includes an OpenMP version for multicore CPU acceleration, a version for heterogeneous CPU/GPU environments such as NVIDIA Grace Hopper, and a fully optimized NVIDIA GPU implementation for integration into an AstroAccelerate pipeline, which will be deployed in the SPOTLIGHT project at the GMRT. Our results demonstrate that unmatched filtering in Pulscan can serve as an efficient data reduction step, prioritizing datasets for further analysis and focusing human and subsequent computational resources on likely binary pulsar signatures.

Introduction

Detecting binary pulsars in real-time is a significant "big data" challenge in modern radio astronomy. Pulsars in binary systems exhibit orbital motion that causes Doppler shifts, "smearing" their signals across multiple frequency bins in a radio telescope's data. Traditional detection methods like Fourier Domain Acceleration Search (FDAS) and Jerk Search (FDJS) utilize "matched filters"—complex templates designed to match the exact expected shape of these smeared signals. While highly sensitive, these methods are computationally intensive, making them difficult to implement in real-time pipelines for next-generation telescopes like the SPOTLIGHT project at the Giant Metrewave Radio Telescope (GMRT), where data rates make storage infeasible.

Methods

Pulscan addresses these computational demands by shifting from "matched" to "unmatched" filtering. Instead of complex "chirp" templates that account for phase, Pulscan employs a simplified boxcar filter on the magnitude-squared (power) spectrum.

Comparison of the FDJS matched filter and the Pulscan unmatched boxcar filter on the normalized power spectrum of PSR J1227-4853.

This approach offers several key advantages:

  • Recursive Computation: Boxcar filters are simple sums that can be calculated recursively. This allows for calculating wide filters nearly as fast as narrow ones, unlike traditional convolutions.
  • Data Reduction: By working on the real-valued power spectrum rather than complex FFT data, Pulscan immediately halves memory requirements.
  • Phase-Insensitivity: Discarding phase information allows a single boxcar filter to detect pulsars accelerating both toward and away from the observer, effectively halving the required template count compared to FDAS.
  • GPU Optimization: The algorithm is heavily optimized for NVIDIA GPUs (including Grace Hopper). To maintain vector-parallel efficiency, the authors developed a 16th-degree polynomial approximation for calculating signal significance ($\chi^2$ statistics) in a purely parallel fashion.

Evaluation and Results

The authors evaluated Pulscan using a massive distribution of 10,955 synthetic binary pulsars generated via SIGPROC, covering a wide range of orbital periods and signal-to-noise ratios.

Parameters and distributions used for generating the synthetic binary pulsar dataset.

Validation was also performed using real-world observations of PSR J1227-4853, a known 1.69 ms pulsar from the GMRT.

Key results include:

  • Performance: The fully optimized GPU implementation on an NVIDIA H100 is between 70x and 250x faster than a 24-core CPU. A typical 600-second observation can be searched in just 0.015 seconds.
  • Efficiency: Pulscan outperformed the industry-standard PRESTO software by over 10x on equivalent CPU hardware.
  • Sensitivity: Although unmatched filtering is inherently less sensitive than matched filtering, Pulscan successfully detected PSR J1227-4853 with high significance, proving its utility as a "flagger" or data reduction step.

Conclusion

Pulscan provides a high-performance solution for real-time binary pulsar discovery. By prioritizing computational economy through unmatched filtering, it enables next-generation telescopes to process massive data streams efficiently, focusing subsequent computational resources on the most promising candidates. Future work involves full integration into the AstroAccelerate pipeline for deployment at the GMRT.

Summary updated

Abstract

The Fourier Domain Acceleration Search (FDAS) and Fourier Domain Jerk Search (FDJS) are proven matched filtering techniques for detecting binary pulsar signatures in time-domain radio astronomy datasets. Next generation radio telescopes such as the SPOTLIGHT project at the GMRT produce data at rates that mandate real-time processing, as storage of the entire captured dataset for subsequent offline processing is infeasible. The computational demands of FDAS and FDJS make them challenging to implement in real-time detection pipelines, requiring costly high performance computing facilities. To address this we propose Pulscan, an unmatched filtering approach which achieves order-of-magnitude improvements in runtime performance compared to FDAS whilst being able to detect both accelerated and some jerked binary pulsars. We profile the sensitivity of Pulscan using a distribution (N = 10,955) of synthetic binary pulsars and compare its performance with FDAS and FDJS. Our implementation of Pulscan includes an OpenMP version for multicore CPU acceleration, a version for heterogeneous CPU/GPU environments such as NVIDIA Grace Hopper, and a fully optimized NVIDIA GPU implementation for integration into an AstroAccelerate pipeline, which will be deployed in the SPOTLIGHT project at the GMRT. Our results demonstrate that unmatched filtering in Pulscan can serve as an efficient data reduction step, prioritizing datasets for further analysis and focusing human and subsequent computational resources on likely binary pulsar signatures.
BibTeX
@article{white2025pulscan,
  author = {Jack White and Karel Adámek and Jayanta Roy and S. M. Ransom and Wesley Armour},
  title = {{Pulscan: Binary pulsar detection using unmatched filters on NVIDIA GPUs}},
  journal = {The Astrophysical Journal Supplement Series},
  year = {2025},
  doi = {10.3847/1538-4365/adc89e},
}
Figure