Speaker
Description
The use of cold tests (low power) for Quasi-Optic (QO) mode converter performance analysis and verification has been an important topic in Gyrotron research [1]. This involves the use of low-power coaxial cavity mode generators producing higher-order TE-modes at mm-Wave and THz frequencies, which can be converted to linear polarized Gaussian-like beams [2],[3]. Misaligned mode generators produce output with low mode purity, which is not desirable for the gyrotron cold test [1]. To eradicate this problem, an individual must make manual fine-tuning for multiple alignment position adjustments, which does not guarantee perfect alignment. In return, the process consumes time for proper alignment.
In this study, deep learning is used to detect misalignment positions in a 95 GHz TE6,2 mode generator cavity to reduce alignment time and increase obtainable mode purity. The misaligned sources are simulated using Commercial Software (CST Microwave) to obtain data for the training of the detection model. The Deep Neural Networks (DNN) algorithm trains the model used to predict the misalignment positions. The obtained model is used to predict misalignment positions in experimental data, which proves to be efficient. Using the predicted misalignment positions, the mode generator can be aligned well to a perfect alignment position.
The training data consist of 21 and 31 x-axis and z-axis misalignments, respectively. To learn important features from the data, 55 frequencies are simulated. This makes the total dataset of 35,805 with 651 data points. Using scalar correlation factor (SCF) as a mode comparison method, the features are extracted from the simulated data. The DNN algorithm with 64x32x4 hidden layers is used to train the model. The new misaligned experimental data are used for verification with misalignment prediction using the above-mentioned model shown in Figure 1. The results show a linear trend in prediction with less prediction error of a maximum of 0.3 mm. Figure 2 shows a proper aligned mode using after detecting the misalignment positions, which has a mode purity of 96.8%. This method proves to be effective in direct misalignment prediction and reduces alignment time to a factor of 20 with increased mode purity.
ACKNOWLEDGMENTS
This work was supported by National R&D Program through the National Research Foundation of Korea(NRF) grant funded by the Korea government (MIST), (No. 2021M1A7A4091139).
References
[1] L. Wang, X. Niu, Y. Liu, “Higher Order Rotating Mode Generator Using Quasi-Optical Techniques,” IEEE Transactions on Plasma Science, vol. 48, no. 10, 2020.
[2] A. Sawant, M. S. Choe, M. Thumm, and E. Choi “Orbital Angular Momentum (OAM) of Rotating Modes Driven by Electrons in Electron Cyclotron Masers,” Scientific Reports, vol. 7, no. 3372, June 2017.
[3] T. Omori et al, “Overview of the ITER EC H&CD system and its capabilities,” Fusion Engineering and Design, vol. 86, pp. 951-954, 2011.
