Existing beam diagnostics are invasive, and oftentimes cannot operate at the required resolution. In this work we present a Machine learning-based Virtual Diagnostic (VD) tool to accurately predict the Longitudinal phase space (LPS) for every shot using spectral information collected non-destructively from the radiation of a relativistic electron beam. VD is a computational tool based on deep learning that can be used to predict a diagnostic output. VDs are especially useful in systems where measuring the output is invasive, limited, costly or runs the risk of altering the output. We show a few applications (experimental or simulated data) for high repetition-rate machine (LCLS-II) or a high-current, ultra-short bunch facility (FACET-II). Then, given a prediction, we relay how reliable that prediction is, i.e., quantify the uncertainty of the prediction. Finally, we show how VD can be used for machine optimization as aberration corrector tuning with ML-based emittance measurements.

Recent breakthroughs in laser wakefield accelerator (LWFA) technology have allowed them to drive free-electron lasers [1]. This is a significant accomplishment as LWFA electron beams are not as well controlled as beams from conventional accelerators. However, longitudinal structure in LWFA beams could be harnessed to accelerate the self-amplified spontaneous emission (SASE) process. Pre-bunched beams have been shown to achieve gain with shorter saturation length than conventional beams [2]. Because of the nature of the LWFA process, electron beams from LWFAs emerge from the plasma with preformed microstructures. The parameters of the accelerator dictate the shape, size and coherence of these features. Coherent optical transition radiation (COTR) can diagnose features in microbunched portions of the electron beam. We present experimental results across three different LWFA regimes demonstrating extreme visible microbunching (up to 10%), as well as sub mm-mrad emittance substructures in LWFA electron beams. In each regime we examined the near field COTR at eight different wavelengths from a foil directly after the end of the accelerator. Depending on the LWFA operating regime, we observe different levels of bunch substructure. How this structure evolves across optical wavelengths is also LWFA-regime dependent. The COTR point spread function model enables the annular shapes observed in the near field to be remapped as the actual 2D beam distributions [3]. We have also used COTR interferometry to measure sub mm-mrad divergence of the microbunched portion of the beam. In addition, we employed a multi-octave spectrometer to measure the spatially averaged TR spectrum from IR to near-UV wavelengths to characterize longitudinal beam shape. Wavelength-dependent variations in the size and radial distribution of the TR images can be correlated with features in the reconstructed longitudinal profile. Combining the longitudinal information acquired by the multi-octave spectrometer with multi-wavelength images of the foil, we observe features in the 3D beam that are unresolvable using other techniques. Moreover, with the aid of physically reasonable assumptions about the bunch profile, reconstructions of the 3D electron bunch distribution will be presented.

Temporal diagnostics of FEL pulses are generally of great benefit to FEL facilites, in particular to provide information to users and for the setup of special modes such as fresh-slice schemes. In this contribution we present FEL power profile measurements with femtosecond resolution at SwissFEL. The FEL temporal profiles are obtained from the longitudinal phase-space of the electrons after the undulator section. We use the transverse wakefields of a corrugated structure to horizontally streak the electron beam, and vertical dispersion to access the energy information. The advantages of this approach, in comparison to the standard streaking using transverse deflecting rf structures, are cost-effectiveness and stability against arrival time jitter.

Recent advances in bunch compression and FEL schemes have enabled ultrashort sub-fs electron and X-ray pulses. The timing jitter is, at best, one order of magnitude larger that the pulse duration. This can be handled by high precision pump-probe delay measurements and data sorting. However, only a small fraction of the pulses will be in the relevant time window. The acceleration and compression in non-linear achromat bunch compressors enables cancellation of the energy and timing jitter caused by modulator high voltage (HV) ripple. The cancellation works at a specific off-crest acceleration phase, the so-called magic angle. We present experimental data showing the current performance at the MAX IV linac, and the benefit of operating at the magic angle. Another major contribution to energy and arrival time jitter is lasers, both for the electron guns and the experiment, and how they are synchronized to the reference RF field. The RF distribution can either be optical or electrical. By extracting the reference RF directly from the gun laser, we have eliminated the relative jitter between the gun laser pulses and the reference field. We show data of the improved performance in our optical master oscillator scheme. A full synchronization system that includes the experimental lasers is under development. Our current plan is to base the synchronization system on a continuous wave reference laser to take advantage of the high frequency of optical waves, instead of relying on the envelope of pulsed lasers. Combining acceleration at or around the magic angle with the high-precision synchronization system we aim at a timing jitter on the order of 1 fs at the end of the linac.