Integration of gpn in the RTA
Milestone ID: 2058
The real-time pipeline of the LST collaboration is currently running with a standard pipeline, using a combination of Hillas technique with random forests for the classification (gammaness) and the reconstruction of energy and incoming direction of the event. The γ-PhysNet method is particularly interesting for real time analysis because it allows for an adaptation of the realtime pipeline for the observation conditions (zenith, azimuth, night sky background and stars in the field of view). This is very important to deal with such conditions since the realtime pipeline is used to real time feedback, which means that no data selection can be applied to clean the samples, in the contrary of the offline analysis. moreover, the low energy are the most affected by these effects, since the amount Cherenkov light is near the detection threshold, and so more easily overshooted by noise. Given the type of sources that realtime pipeline have to deal with (flaring AGNs, GRBs, electromagnetic gravitational wave counterparts) it is very important to lower the low energy frontier as explained in the previous sections. It is as well important to note that the flux variability precision is mostly driven by the statistics in IACTs, which means that every increase of the low energy effective area will lead to the possibility to explore a finer time binning in realtime lightcurve.
Otherwise, the modularity of the software allows to add different type of algorithms for the different stage of the analysis. For the inclusion of γ-PhysNet in the real-time pipeline, it is foreseen to add the algorithm after the first stage of production of data, providing data with Hillas parameters and images. The pipeline can be updated in order to perform a simultaneous production of high level data, providing a reference pipeline to check the impact of the γ-PhysNet . In addition to the quality of the physical reconstruction, the computing time is a strong requirement for the realtime analysis and need to be tested with the GPUs available in the LST onsite cluster.