Feature Extraction for Machine Learning-Based Alignment of W-band Higher-Order Mode Generator

This paper proposes methods of extracting features from simulation data for use in machine learning models to optimize the alignment of the W-band mode generator. We introduce the Scalar Correlation Factor (SCF) and Mean Square Error (MSE) as methods for comparing fields to quantify the field data used as features. These methods are validated using Deep Neural Networks (DNN) to predict mode generator misalignments for effective optimization. The two feature extraction methods are compared based on simulation and experimental results, revealing that SCF demonstrates outstanding accuracy in predictions.