The identification of turbulence sources would drive to a deeper understanding of confinement dynamics in tokamak plasmas. Turbulence results from a mixture of instabilities corresponding to sources at different timescales and spatial scales. Using poloidal correlation reflectometry and multi-pin Langmuir probe, it was shown in the T-10 and the Tokamak Experiment for Technology Oriented Research (TEXTOR) tokamaks that the reflectometry frequency spectrum is the superposition of several components: broadband component, quasi-coherent (QC) modes and low-frequency components. The relevance of QC modes is associated with their link with the trapped electron mode instability. This link was exhibited in the transition from the linear ohmic confinement (LOC) to the saturated ohmic confinement (SOC) regime. A method is presented in this paper to extract the QC mode component from the reflectometry data, enabling its separation from the broadband component and the study of its time evolution. It is a first step toward the discrimination of turbulence sources. The central idea explores a way to combine the approach of signal processing and machine learning. The continuous wavelet transform on the basis of complex Morlet wavelet has proved to be efficient in providing a decomposition of a signal at different scales over time for fluctuation tackling; clustering techniques, such as the mini-batch K-means, are able to tackle clusters at different scales. The method was applied to Tore Supra and TEXTOR reflectometry data. In Tore Supra, the amplitude of the extracted QC mode component decreases during the LOC–SOC transition. In TEXTOR, the amplitude of the coherent spectra of the extracted QC mode component is similar to the experimental coherent spectra obtained through correlation reflectometry. The developed method permits the extraction of components, preserving their physical and statistical properties.
Read full abstract