Key points are not available for this paper at this time.
... We present a singlepulse search method, improving on commonly used neural network classifiers thanks to the filtering of radio frequency interference based on its spectral variance and the magnetar's rotation. With this approach, we were able to lower the signal to noise ratio (S/N) detection threshold from 8 to 5. This allowed us to find over 115,000 spiky single pulses - compared to 56,000 from the neutral network approach. Here, we present the temporal variation of the overall profile and single pulses. Two distinct phases of different single pulse activity can be identified: phase 1 from December 2018 to mid-2019, with a few single pulses per hour, and phase 2 from September 2020 with hundreds of single pulses per hour (with a comparable average flux density). We find that the single pulse properties and folded profile in phase 2 exhibit a change around mid-March 2021. Before this date, the folded profile consists of a single peak and single pulses, with fluences of up to 1000 Jyms and a single-peaked width distribution at around 10 ms. After mid-March 2021, the profile consists of a two peaks and the single pulse population shows a bimodal width distribution with a second peak at 1 ms and fluences of up to 500 Jyms. We also present asymmetries in the phase-resolved single pulse width distributions beginning to appear in 2020, where the pulses arriving earlier in the rotational phase appear wider than those appearing later. This asymmetry persists despite the temporal evolution of the other single pulse and emission properties. We argue that a drift in the emission region in the magnetosphere may explain this observed behaviour. Additionally, we find that the fluence of the detected single pulses depends on the rotational phase and the highest fluence is found in the centre of the peaks in the profile. ...
Bause et al. (Wed,) studied this question.