Nonnegative Gaussian process on quasi-periodic noise for plasma radiation tomography during a disruption

In cases where the data collected from plasmas contain high level quasi-periodic noise and it is difficult to perform tomography reconstruction based on a least square method, we have developed a method to infer a plasma state using Bayesian probability theory and Gaussian process with non-negativity priors. Since most fusion devices utilize magnetic fields to control plasma, the plasma itself and the control coils may produce electromagnetic noises that affect the integrity of the diagnostics. In KSTAR, we use fast bolometer with Absolute Extreme Ultra-Violet (AXUV) detectors to measure radiated power during plasma disruption. In this work, we introduce a method of extracting radiation signals from the raw signals, which are contaminated by quasi-coherent noise, using Bayesian inference and Gaussian process. When a solution of the least square estimation for tomography reconstruction is difficult to obtain due to lack of line integrated signals, prior knowledge is often reflected in finding solutions using a regularization technique. Here, we use Gaussian process prior as a regularization for statistical image reconstruction of AXUV signals. The non-negativity prior is utilized for tomography reconstruction as the radiation can not have negative values.