Speaker
Description
Turn-by-turn (TBT) BPM data in the LHC is often affected by noise, limiting the extraction of resonant driving terms (RDTs) and reducing the precision of nonlinear optics studies. We developed a denoising autoencoder trained on simulated tracking data to reconstruct clean transverse oscillations and suppress noise directly in the time domain. The method produces cleaner frequency spectra and significantly improves RDT visibility compared to established methods such as singular value decomposition, even when trained on fewer turns. In its current form, the autoencoder performs well on data that resemble the training set. However, when applied to new conditions—different noise levels, excitation amplitudes, tunes, or beam configurations—its ability to generalise decreases. These results demonstrate that autoencoders can substantially improve TBT data quality. Establishing broader and more diverse training datasets is a promising next step toward applying this technique to real LHC measurements.
Funding Agency
CERN
| In which format do you inted to submit your paper? | LaTeX |
|---|