Ongoing work towards real-time reflectometry density profile measurements on ASDEX Upgrade

Previous studies performed in ASDEX Upgrade ELMy H-mode regimes (Santos et al. [2003]), ITER reference scenario, have shown that O-mode FM-CW reflectometry should meet the accuracy required on ITER for using reflectometry for position control purposes. This application, however, requires the development of robust unsupervised methods to evaluate the density profiles in real time. To allow a statistical evaluation of the performance of these methods, an extensive H-mode mode density profile database was built using O-mode reflectometry experimental data. An 1D full wave FDTD code (Hacquin et al. [2001]) was used to generate O-mode raw data adapted to the ASDEX Upgrade reflectometer configuration. To benchmark the robustness of the data processing techniques to the effects of plasma turbulence, several levels of radial density fluctuations (Heuraux et al. [2003]) were applied to this database. Preliminary studies using neural networks for density profile inversion gave promising results and this approach is now being pursued. The adopted method not only proved to be quite robust to density fluctuations but it also intrinsically compensated the lack of density profile data below ne = 3.6× 1018m−3, essential for the standard Abel inversion procedure. This robustness provides a relief on the requirements and sophistication of the procedures used for group delay evaluation, allowing the use of more real-time friendly signal processing techniques. In this paper are shown ASDEX Upgrade H-mode edge density profiles, reconstructed using an highly optimized real-time data processing code, based on the proposed neural network approach. The implementation strategy of the real-time upgrade to the ASDEX Upgrade O-mode reflectometer is also discussed.