2025ApJ...984..111S

Traditional pulsar timing techniques involve averaging large numbers of single pulses to obtain a high signal-to-noise ratio (S/N) profile, which is cross-correlated with a template to measure a time of arrival (TOA). However, the morphology of individual single pulses varies greatly due to pulse jitter, and pulses of different fluence contribute differently to the S/N of the pulse average. Our study proposes a method that accounts for these variations by identifying a range of “states” and timing each separately. We selected two 1 hr observations of PSR J2145-0750, each in a different frequency band, with the Green Bank Telescope. We normalized the pulse amplitudes using the dynamic spectrum to account for scintillation effects and probed different excision algorithms to reduce radio-frequency interference. We then measured four pulse parameters (amplitude, position, width, and energy) to classify the single pulses using automated clustering algorithms. For each cluster, we calculated an average pulse profile and template and used both to obtain a TOA and TOA error. Finally, we computed the weighted average TOA and TOA error, the latter as a metric of the total timing precision for the epoch. The TOA is shifted relative to the one obtained without clustering, and we can estimate the shift with this weighting using the same data. For the 820 and 1400 MHz bands, we obtained TOA uncertainties of 0.057 and 0.46 μs, compared to 0.066 and 0.74 μs when no clustering is applied. We conclude that tailoring this method could help improve the timing precision for certain bright pulsars in NANOGrav's data set.