Bayesian inference using JET’s microwave diagnostic system

Abstract

At the JET tokamak, three electron cyclotron emission (ECE) diagnostics (two Martin-Puplett interferometers and a heterodyne radiometer) and a reflectometer form the basic microwave diagnostic system. The standard analysis approaches deduce electron density and temperature profiles independently of each diagnostic measurement. Via the Bayesian framework Minerva, the microwave diagnostic system is modelled, and electron temperature and density profiles are inferred jointly for an Ohmic JET plasma. Furthermore, profile length-scales for different plasma domains, a magnetic field correction, distinct reflection properties of the high-field side and low-field side walls, and radiometer sensitivities are estimated together. This inference scheme can use one of two models to predict broadband ECE spectra; one is less accurate but fast for a single prediction, and a more accurate model, relying on the ray-tracer SPECE parallelised via web services. The faster model allows the investigation of correlations between parameters and the execution of a numerical marginalisation, i.e. an uncertainty propagation.