Past Work

Real-time Search Tool for Fast Radio Bursts Based on Deep Learning

We propose a deep learning-based search process for fast radio bursts, named DRAFTS. Commonly used search algorithms generally involve eliminating radio frequency interference, using a series of dispersion grids to de-disperse the data, applying matched filters of various widths to the de-dispersed time series to calculate the signal-to-noise ratio, and finally selecting candidate signals based on a certain threshold. This method generally suffers from issues such as redundant computation, low computational efficiency, high false positive rates, and incomplete results. For example, in a set of data observed by the Arecibo telescope for FRB 20121102A, the number of signals found by different people using different algorithms varied from dozens to hundreds, showing an order of magnitude difference. Therefore, developing a new search algorithm is highly necessary. In DRAFTS, we use CUDA acceleration to convert raw time-frequency data into time-dispersion data, then use a trained object detection model to identify the arrival time and dispersion value of the bursts, and subsequently extract the burst data segments from the original data. Finally, a binary classification model is used to determine whether it is a real fast radio burst. This approach significantly improves efficiency and sensitivity, can process large amounts of data in a short time, reduces false positive rates, and addresses many issues inherent in traditional algorithms.

Date: September 13, 2024Deep Learning, Transient Sources, Digital Signal Processing

Stochastic Nature of Repeating Fast Radio Bursts

Through in-depth analysis of FRB 20121102A and FRB 20190520B, we did not find significant short-periodicity, which may be due to the random radiation nature of fast radio bursts. We used the 'Pincus Index' and 'Lyapunov Exponent' to quantify the randomness and chaos of fast radio bursts and compared them with common physical phenomena such as pulsars, earthquakes, and solar flares. We found that fast radio bursts differ significantly from earthquakes and solar flares, showing stronger randomness and less chaos. Additionally, we found no clustering in the waiting times and energies between bursts. These results suggest that fast radio bursts may originate from a high-entropy single source or a combination of different emission mechanisms/locations. The mathematical and physical methods involved also have broad application potential, potentially playing important roles in other fields beyond astronomy.

Date: April 30, 2024Fast Radio Bursts, Statistical Analysis, Mathematical Physics Methods

FAST Observations and Polarization Distribution of FRB20220912A

From September 28, 2022, to December 31, 2022, FAST observed FRB20201124A. We detected over 1000 bursts, with the highest event rate reaching 390 bursts per hour. The polarization analysis of this FRB found that its RM value was close to 0 and remained stable over two months, indicating it is in a relatively clean environment. We also discovered high circularly polarized bursts from this FRB, with rapid changes in circular polarization over time and frequency, and an inverse relationship between the proportion of circularly polarized bursts and RM, suggesting that the circular polarization of FRBs may be intrinsic to the radiation and the environment causes absorption of circular polarization.

Date: September 27, 2023Fast Radio Bursts, Radio Observations, Statistical Analysis

FAST Observations and Energy Distribution of FRB20201124A

From September 26, 2021, to October 17, 2021, FAST observed FRB20201124A. We discovered an exponentially rising burst event rate in the observational data, which suddenly disappeared after reaching its peak. The highest event rate reached 542 bursts per hour, the highest ever recorded for fast radio bursts, and the burst energy on that day exceeded 14% of the magnetic energy of a magnetar, significantly constraining the radiation mechanism of FRBs. Further analysis of the energy distribution showed that FRB20201124A exhibited a bimodal distribution. We found that the choice of burst bandwidth did not affect the energy distribution, but the burst definition did. Finally, we discovered that different repeating bursts had different energy distributions, indicating different radiation characteristics.

Date: November 30, 2022Fast Radio Bursts, Radio Observations, Statistical Analysis

Detection of Circular Polarization in Two Repeating Fast Radio Bursts from FAST Data

We systematically analyzed the FAST observational data of FRB20121102A and 20190520B and found that both repeating bursts had less than 5% of their bursts exhibiting circular polarization characteristics, with the highest circular polarization degree reaching 64%. Such high circular polarization limits the possibility of it being due to multi-path propagation. Possible mechanisms include Faraday conversion in extreme magnetic environments or intrinsic characteristics of the FRB source. Currently, the probability of circular polarization appearing in non-repeating fast radio burst pulses is higher than in repeating bursts. This finding indicates that the conditions for producing circular polarization in repeating bursts are more stringent. This discovery increases the number of circularly polarized repeating bursts from 1 to 3. As one of the few active repeating bursts, the detection of circular polarization in FRB20121102A, 20190520B, and 20201124A suggests that circular polarization may be a common feature of repeating fast radio bursts.

Date: December 15, 2022Fast Radio Bursts, Radio Observations, Polarization