The application of encoder-decoder neural networks in high accuracy and efficiency slit-scan emittance measurements

TL;DR

This paper presents an advanced slit-scan emittance measurement method using encoder-decoder neural networks and machine learning to significantly improve speed and accuracy in high-brightness electron beams.

Contribution

It introduces a novel application of encoder-decoder neural networks combined with machine learning to enhance the efficiency and precision of emittance measurements.

Findings

Measurement time reduced from 15 minutes to 90 seconds.

Machine learning effectively reduces beamlet image noise.

Analysis of error sources improves measurement reliability.

Abstract

A superconducting radio-frequency (SRF) photo injector is in operation at the electron linac for beams with high brilliance and low emittance (ELBE) radiation center and generates continuous wave (CW) electron beams with high average current and high brightness for user operation since 2018. The speed of emittance measurement at the SRF gun beamline can be increased by improving the slit-scan system, thus the measurement time for one phase space mapping can be shortened from about 15 minutes to 90 seconds. A parallel algorithm and machine learning have been used to reduce the beamlet image noise. In order to estimate the uncertainty in the calculation of normalized emittance, we analyze the main error contributions such as slit position uncertainty, image noise, space charge effects and energy measurement inaccuracy.