Classification of the L-, H-mode, and plasma-free state: Convolutional neural networks and variational autoencoders on the edge reflectometer for KSTAR.

Classifying and monitoring the L-, H-mode, and plasma-free state are essential for the stable operational control of tokamaks. Edge reflectometry measures plasma density profiles, but the large volume of data and complexity in reconstruction pose significant challenges. There is a need for efficient methods to analyze complex reflectometer data in real-time, which can be addressed using advanced computational techniques. Here, we show that machine learning (ML) techniques can classify discharge states using raw signal data from an edge reflectometer installed on the Korea Superconducting Tokamak Advanced Research. The deep convolutional neural network models achieved classification accuracy of up to 99% when using 2D spectrogram inputs, demonstrating a significant improvement over 1D raw signal inputs. Additionally, the variational autoencoder model effectively clustered the discharge states in the latent space without any label information, further validating the model's capability to classify discharge states. These results suggest that the ML model can effectively handle the complexity of reflectometer data and accurately classify plasma discharge states. This approach not only facilitates real-time diagnosis but also reduces the need for manual data processing.

In-situ frequency calibration of frequency modulated continuous wave reflectometry.

Frequency modulated continuous wave reflectometers have been used to measure the plasma density profiles of the KSTAR tokamak. Three reflectometers are operating in an extraordinary polarization mode in the frequency range of Q band (33-50 GHz), V band (50-75 GHz), and W band (75-110 GHz). Each full frequency band is linearly swept in 20 µs. The accuracy of the density profile measurement is dependent on how precisely the frequency is calibrated. Two new frequency measurement techniques are developed to in situ calibrate the instantaneous frequency during the sweep time of 20 µs. First the intermediate frequency (IF) of the receiver is analyzed based on wavelet transform as applying a fixed frequency signal from a synthesizer to the local oscillator (LO) port. The frequency of the transmitted microwave is simply obtained by adding or subtracting the measured IF to the given LO frequency. Once the group delay of the whole system is known, the frequency can be calibrated by measuring the IF. By measuring the reflectometer output without plasma with delay lines of different lengths and subtracting them, the frequency can be calibrated by eliminating the unknown group delay of the system except the known group delay of the delay line. These two techniques are described in detail, and the calibrated frequencies conducted by the two techniques are compared.

A frequency measurement algorithm for non-stationary signals by using wavelet transform.

Scalogram is widely used to measure instantaneous frequencies of non-stationary signals. However, the basic property of the scalogram is observed only for stationary sinusoidal functions. A property of the scalogram for non-stationary signal is analytically derived in this paper. Based on the property, a new frequency measurement algorithm is proposed. In addition, a filter that can separate two similar frequency signals is developed based on the wavelet transform.

Graphics processing unit-assisted density profile calculations in the KSTAR reflectometer.

Wavelet transform (WT) is widely used in signal processing. The frequency modulation reflectometer in the KSTAR applies this technique to get the phase information from the mixer output measurements. Since WT is a time consuming process, it is difficult to calculate the density profile in real time. The data analysis time, however, can be significantly reduced by the use of the Graphics Processing Unit (GPU), with its powerful computing capability, in WT. A bottle neck in the KSTAR data processing exists in the data input and output (IO) process between the CPU and its peripheral devices. In this paper, the details of the WT implementation assisted by a GPU in the KSTAR reflectometer are presented and the consequent performance improvement is reported. The real time density profile calculation from the reflectometer measurements is also discussed.

Design of a Doppler reflectometer for KSTAR.

A Doppler reflectometer has been designed to measure the poloidal propagation velocity on the Korea Superconducting Tokamak Advanced Research (KSTAR) tokamak. It has the operating frequency range of V-band (50-75 GHz) and the monostatic antenna configuration with extraordinary mode (X-mode). The single sideband modulation with an intermediate frequency of 50 MHz is used for the heterodyne measurement with the 200 MHz in-phase and quadrature (I/Q) phase detector. The corrugated conical horn antenna is used to approximate the Gaussian beam propagation and it is installed together with the oversized rectangular waveguides in the vacuum vessel. The first commissioning test of the Doppler reflectometer system on the KSTAR tokamak is planned in the 2014 KSTAR experimental campaign.

Development of frequency modulation reflectometer for Korea Superconducting Tokamak Advanced Research tokamak.

Frequency modulation reflectometer has been developed to measure the plasma density profile of the Korea Superconducting Tokamak Advanced Research tokamak. Three reflectometers are operating in extraordinary polarization mode in the frequency range of Q band (33.6-54 GHz), V band (48-72 GHz), and W band (72-108 GHz) to measure the density up to 7 × 10(19) m(-3) when the toroidal magnetic field is 2 T on axis. The antenna is installed inside of the vacuum vessel. A new vacuum window is developed by using 50 µm thick mica film and 0.1 mm thick gold gasket. The filter bank of low pass filter, notch filter, and Faraday isolator is used to reject the electron cyclotron heating high power at attenuation of 60 dB. The full frequency band is swept in 20 µs. The mixer output is directly digitized with sampling rate of 100 MSamples/s. The phase is obtained by using wavelet transform. The whole hardware and software system is described in detail and the measured density profile is presented as a result.

Development of frequency modulation reflectometer for KSTAR tokamak: data analysis based on Gaussian derivative wavelet.

A frequency modulation reflectometer has been developed to measure the density profile of the KSTAR tokamak. It has two channels operating in X-mode in the frequency range of Q band (33-50 GHz) and V band (50-75 GHz). The full band is swept in 20 µs. The mixer output is directly digitized at the sampling rate of 100 MSamples∕s. A new phase detection algorithm is developed to analyze both amplitude and frequency modulated signal. The algorithm is benchmarked for a synthesized amplitude modulation-frequency modulation signal. This new algorithm is applied to the data analysis of KSTAR reflectometer.

Development of a digital integrator for the KSTAR device.

The development of an integrator for magnetic diagnostics becomes more important as the pulse length of fusion devices gets longer and longer, especially for present-day superconducting fusion devices. A small offset in the signal can cause a significant drift in the integrator output for long pulse experiments. A lock-in amplifying digital integrator has been developed for Wendelstein 7-X (W7-X). It succeeds in suppressing the drift to a low value but requires about 100 ms for data processing. To shorten the data processing time, a Field Programmable Gate Array (FPGA) built in the digitizer is utilized. Since there is no need to transfer the data to an external computer, the integration can be done in real time. The microprocessor built in the digitizer directly transfers the data integrated in the internal FPGA into the reflective memory installed in the same compact Peripheral Component Interconnect chassis. These features result in a very compact system design. The design and the preliminary results of the digital integrator will be presented.

Korea Superconducting Tokamak Advanced Research vacuum and gas puffing system.

A piezoelectric valve, which has a flow rate of about 463 mbar l/s, has been installed to fuel the Korea Superconducting Tokamak Advanced Research (KSTAR) tokamak. The valve flow rate is in situ calibrated by analyzing the pressure rise curve while fueling the vessel at a constant rate. The calibration method and results are presented. In addition to the flow rate, other vacuum system parameters, such as the pumping speed and the vessel volume, were experimentally obtained. Based on these measurements, a KSTAR vacuum system simulator was developed to calculate the valve drive signal to obtain a programmed pressure trace. An arbitrarily shaped pressure trace was successfully controlled in KSTAR with this hardware and software system.