Vela pulsar: single pulses analysis with machine learning techniques

We study individual pulses of Vela (PSR\ B0833-45\,/\,J0835-4510) from daily observations over three hours (around 120,000 per observation), performed simultaneously with the two radio telescopes at Argentine Institute Radioastronomy. select 4 days in January-March 2021 and their statistical properties machine learning techniques. first use density based DBSCAN clustering techniques, associating mainly by amplitudes, find a correlation between higher amplitudes earlier arrival times. also weaker (polarization dependent) mean width pulses. identify clusters so-called mini-giant pulses, $\sim10\times$ average pulse amplitude. then perform an independent study, Self-Organizing Maps (SOM) Variational AutoEncoder (VAE) reconstruction to separate them clearly noise one observation train VAE apply it thre rest observations. SOM determine day telescope conclude that our main results are robust self-consistent. These support models for emitting regions different heights (separated each roughly hundred km) pulsar magnetosphere. model amplitude distribution interstellar scintillation patterns inter-pulses time-scale finding characterizing exponent $n_{\mathrm{ISS}}\sim7-10$. In appendices we discuss checks hardware systematics simultaneous one-polarization / two-polarizations configurations. provide detailed analysis processes radio-interferences cleaning folding.