Machine-learning approaches to dispersion measure estimation for fast radio bursts 2026 Machine-learning approaches to dispersion measure estimation for fast radio bursts – 8

Hosein Rajabi, Zhejian Liu, Fereshteh Rajabi, Martin Houde · 2025

Read Machine-learning approaches to dispersion measure estimation for fast radio bursts 2026 Machine-learning approaches to dispersion measure estimation for fast radio bursts – 8 by Hosein Rajabi, Zhejian Liu, Fereshteh Rajabi, Martin Houde (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-24003.txt.

Introduction

Hosein Rajabi, Zhejian Liu, Fereshteh Rajabi, and Martin Houde automate a measurement radio astronomers need for every fast radio burst. These millisecond cosmic flashes arrive smeared by their passage through ionized gas, and the dispersion measure quantifying that smearing encodes distance and plasma conditions. Estimating it by traditional dedispersion is computationally heavy and open to human bias.

As a proof of concept, the team trained three deep learning architectures, a plain convolutional network, a fine-tuned ResNet-50. A hybrid convolutional-recurrent model, on synthetic burst spectra modeled on the CHIME telescope's specifications. The hybrid model gave the most accurate and stable dispersion estimates at low computational cost across the tested range. Because the networks were trained on simulations, the authors note they can be fine-tuned on real CHIME observations and adapted to future facilities, aiming at real-time measurement in large burst surveys.

Catalog record

CREATORHosein Rajabi, Zhejian Liu, Fereshteh Rajabi, Martin Houde
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-24003

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Machine-learning approaches to dispersion measure estimation for fast radio bursts 2026 Machine-learning approaches to dispersion measure estimation for fast radio bursts – 8 Hosein Rajabi, Zhejian Liu, Fereshteh Rajabi, Martin Houde Abstract Fast radio bursts (FRBs) are bright, mostly millisecond-duration transients of extragalactic origin whose emission mechanisms remain unknown. As FRB signals propagate through ionized media, they experience frequency-dependent delays quantified by the dispersion measure (DM), a key parameter for inferring source distances and local plasma conditions. Accurate DM estimation is therefore essential for characterizing FRB sources and testing physical models, yet current dedispersion methods can be computationally intensive and prone to human bias. In this proof-of-concept study, we develop and benchmark three deep-learning architectures, a conventional convolutional neural network (CNN), a fine-tuned ResNet-50, and a hybrid CNN–LSTM model, for automated DM estimation. All models are trained and validated on a large set of synthetic FRB dynamic spectra generated using CHIME/FRB-like specifications. The hybrid CNN–LSTM achieves the highest accuracy and stability while maintaining low computational cost across the investigated DM range. Although trained on simulated data, these models can be fine-tuned on real CHIME/FRB observations and extended to future facilities, offering a scalable pathway toward real-time, data-driven DM estimation in large FRB surveys. 1 Introduction Fast radio bursts (FRBs) are highly energetic, short-duration radio transients whose physical origin and emission mechanisms remain elusive despite significant theoretical and observational progress. Their spectra provide key diagnostics for probing both the source and its environment. FRBs exhibit several characteristic spectro-temporal features, including the well-known dispersion sweep, where lower frequencies arrive later; a tendency for shorter burst durations at higher observing frequencies; and, in many repeating sources, a systematic downward drift of emission frequency with time across sub-bursts. The observed spectra therefore encode information about both the source and the propagation path from the emission region to the observer. As the signal propagates, it interacts with intervening media and experiences dispersion, scattering, and scintillation. Disentangling these effects is essential for isolating the intrinsic properties of the source. Accurate dedispersion is particularly critical, as many parameters derived from FRB observations are sensitive to errors in the estimated DM ( Rajabi et al. 2020 ; Chamma et al. 2021 ). Precise DM estimation not only enables recovery of intrinsic temporal and spectral structures but also provides insights into their host environments. Once propagation effects are characterized, theoretical models can focus on how intrinsic source properties and dynamics produce the observed spectro-temporal behaviour. The Triggered Relativistic Dynamical Model (TRDM; Rajabi et al. 2020 ) is one such framework, quantifying how source dynamics influence spectral structures. TRDM predicted an inverse relationship between sub-burst slope and duration, known as the sub-burst slope law, a prediction subsequently confirmed across multiple repeating FRB sources ( Rajabi et al. 2020 ; Chamma et al. 2021 ; Wang et al. 2022 ; Jahns et al. 2023 ; Chamma et al. 2023 ; Brown et al. 2024 ; Chamma et al. 2024 ) . Such models establish a physical link between source dynamics and observed signatures, offering stringent constraints on theoretical descriptions of FRB emission. In parallel, considerable effort has focused on developing accurate and efficient dedispersion techniques. In FRB detection pipelines, identifying genuine bursts among false positives (e.g. noise and radio-frequency interference, RFI) involves several processing stages, including real-time dedispersion algorithms that typically maximize the signal-to-noise ratio (S/N) of the dedispersed pulse. Two main dedispersion approaches are used in real-time analysis: incoherent and coherent. The incoherent method corrects for dispersion by applying time-delay adjustments to individual frequency channels, with accuracy limited by channel bandwidth. Although less computationally demanding than its coherent counterpart, it remains expensive for blind searches that must test thousands of DM trials. Various implementations, such as brute-force ( Barsdell et al. 2010 ) , tree dedispersion ( Taylor 1974 ) , sub-band methods, and fast discrete dispersion measure transforms (FDMT), offer different trade-offs in accuracy and computational efficiency depending on the underlying hardware architecture (CPU or GPU). Coherent dedispersion ( Ait Allal et al. 2009 ; Bassa et al. 2017 ) operates directly on raw voltage data, reversing the dispersive effects of the interstellar medium (ISM) in the Fourier domain to recover intrinsic pulse shapes with high accuracy. Although particularly effective for high-DM bursts from known repeaters, it remains prohibitively expensive for blind searches. Semi-coherent approaches, which apply coherent dedispersion at selected DM trials followed by incoherent refinement, can balance sensitivity and computational cost. While these real-time algorithms enable rapid FRB detection, they are optimized primarily for S/N rather than for fine-scale structural fidelity, and their precision degrades for complex or low-S/N bursts. In contrast, several post-processing algorithms, such as DM phase ( Seymour et al. 2019 ) , the DM power method ( Lin et al. 2022 ) , and related structure-maximization techniques ( Platts et al. 2021 ; Marthi et al. 2022 ) , focus on recovering the DM that best preserves the burst substructure. These approaches can achieve sub–pc cm -3 precision but are computationally intensive and typically applied to bright, high-S/N events in offline analyses. Despite these advances, dedispersion remains computationally demanding for real-time applications and can lose precision for faint or morphologically complex bursts, as often encountered with instruments such as the Canadian Hydrogen Intensity Mapping Experiment (CHIME/FRB). Moreover, high-precision post-processing methods are impractical for large-scale or automated analyses due to their computational cost and reliance on manual intervention. These limitations motivate the development of data-driven approaches that combine the speed and automation of real-time systems with the accuracy and structural sensitivity of post-processing algorithms. In this paper, we investigate the use of machine-learning (ML) techniques to improve dedispersion and DM estimation for FRBs. While convolutional neural networks (CNNs; Connor & van Leeuwen 2018 ) have previously been proposed as substitutes for traditional dedispersion backends, most ML studies to date have focused on the classification or detection of FRBs rather than on quantitative DM estimation. Here, we extend this exploration by developing and comparing three deep-learning architectures, a baseline CNN, a fine-tuned ResNet-50, and a hybrid CNN–LSTM model, designed to recover the intrinsic DM directly from frequency–time dynamic spectra. Section 2 reviews previous ML applications to FRBs, Section 3 describes the generation of synthetic FRB data used for training and validation, and Section 4 outlines the architecture and training procedures for the three models. Our results are presented and discussed in Section 5 , and the main conclusions and future directions are summarised in Section 6 . Additional details on the simulated data, cross-validation, and test-set performance are provided in the Appendices at the end. 2 FRBs and Machine Learning As previously stated, ML techniques have increasingly been applied to FRBs, though most studies have focused on classification rather than quantitative parameter estimation. Existing applications fall into two main categories: distinguishing genuine FRB signals from radio-frequency interference (RFI) or known pulsars ( Wagstaff et al. 2016 ; Akeret et al. 2017 ; Mosiane et al. 2017 ; Czech et al. 2018a ; Agarwal et al. 2020 ) , and differentiating between repeating and non-repeating sources ( Yang et al. 2023 ; Zhu-Ge et al. 2023 ; Luo et al. 2023 ; García et al. 2024 ; Sharma & Rajpaul 2024 ; Qiang et al. 2025

Read the complete text