SPIDERweb: a Neural Network approach to spectral phase interferometry

TL;DR

This paper introduces a neural network-based method for spectral phase interferometry (SPIDER) pulse characterization, reducing calibration needs and improving analysis under experimental constraints using a cascade of convolutional networks.

Contribution

It presents a novel neural network approach for SPIDER data analysis, enabling more flexible and less calibration-dependent pulse characterization.

Findings

Neural networks can effectively analyze SPIDER interferograms.

The proposed method reduces the need for precalibration.

It demonstrates reliable pulse characterization with reasonable computational resources.

Abstract

Reliably characterised pulses are the starting point of any application of ultrafast techniques. Unfortunately, experimental constraints do not always allow optimising the characterisation conditions. This dictates the need for refined analysis methods. Here we show that neutral networks can provide a viable characterisation when applied to data from SPIDER. We have adopted a cascade of convolutional networks, addressing the multiparameter structure of the interferogram with a reasonable computing power. In particular, the necessity of precalibration is reduced, thus pointing towards the introduction of neural networks in more generic arrangements.