Speaker
Description
A high energy particle interacting in ice will induce a particle cascade. The Radar Echo Telescope (RET) aims to detect this cascade by means of the radar echo method. In our simulations, it is possible to transmit a radio signal to scatter off the cascade, and a return radio signal will be observed at a number of user-defined receivers, in ice. Several properties in the radio signals have been observed when systematically varying the direction of the cascade, such as patterns in the power, peak frequency and arrival time at a particular receiver. Given these patterns, it is possible to train machine-learning algorithms to reconstruct the cascade direction. As such, we simulate a setup of multiple receivers and vary the position, direction and energy of the cascade, and show the reconstruction accuracy that can be currently achieved. We train lower-level machine learning algorithms such as Gradient Boosted Machines (GBMs), simple neural networks and linear regression on the signal properties, as well as more complex models (convolutional neural networks) on the signal spectrograms.
