Estimation of X-Mode Reflectometry First Fringe Frequency Using Neural Networks

Abstract

One of the main challenges in X-mode reflectometry is the correct determination of the group delay measurement used for density profile reconstruction. The X-mode upper cutoff group delay measurement can be used to reconstruct the electron density profiles from the near zero density. However, due to the broad operational conditions of experimental fusion devices, the start of the upper cutoff region can occur at any probing frequency. The first fringe (FF) of the interference signal measured by reflectometry that corresponds to the start of the upper cutoff reflection is used together with the magnetic field profile to determine vacuum distance between the reflectometer antenna and the start of the plasma. An incorrect estimation of the FF probing frequency not only introduces a radial error but also a group delay error, affecting the shape of the resulting density profile. In this paper, we present the new developments in the automatic FF estimation required for the reliable reconstruction of density profiles, used in the multichannel X-mode density profile reflectometry diagnostic recently installed on ASDEX Upgrade. An improved algorithm to estimate and track the frequency of the FF along a discharge is introduced. Tests show that it is able to correctly determine the FF for most discharges. However, for a number of unanticipated cases, the algorithm provides jitter and imprecise results, introducing errors in the reconstructed density profiles. We also present a novel neural network (NN) approach for the first time for the estimation of the FF frequency. A comprehensive training set was carefully selected by experienced reflectometry diagnosticians and used to train the NN model using the open source software libraries TensorFlow and Keras. The resulting NN is able to provide more precise FF estimations than the previous algorithm. The reconstructed density profiles, using both algorithms, are presented and compared.