Robust monitoring of thermal and fast ions using collective Thomson scattering: Combining physics- and data-driven...

Isolating a signal of interest from multivariate time-series is of interest for many scientific purposes. An example is the monitoring of thermal- and fast-ion dynamics with fusion plasma diagnostics such as collective Thomson scattering (CTS), which requires careful characterization of an often highly dynamic diagnostic background. Here, we develop and evaluate new CTS background estimation methods inspired by both physics-based approaches and electroencephalogram signal processing. To apply these, we present, for the first time, CTS data from the ASDEX Upgrade (AUG) fusion device taken using rapid (∼5 kHz) on-pulse power modulation of the CTS probe gyrotron. We find that the best performing method is a physics-informed principal component analysis, yielding a typical error on the background estimate of ∼0.5 eV, a factor of ∼3 improvement compared to simpler techniques. We show that this enables CTS-based thermal- and fast-ion monitoring at AUG on the basis of individual 2-ms acquisition pulses, even in the presence of otherwise deleterious edge-localized modes. It also allows accurate tracking of the evolution of CTS spectra across changes in fast-ion heating schemes, in excellent agreement with theoretical expectations. Furthermore, we apply synthetic CTS signals to demonstrate the benefits of rapidly modulating the CTS probe gyrotron power for improved background subtraction. This predicts more accurate estimates for higher modulation amplitude, frequency, and on-pulse duration and may serve as a basis for novel recommendations for the operation of existing and future CTS diagnostics.