Machine learning powered multiple cross-interference suppression strategy for surface acoustic wave sensing platform

Recently, there have been significant developments in surface acoustic wave (SAW) technologies for diverse physical, chemical, and biological sensing applications, but they commonly show poor selectivity in a complex environment, primarily due to the interactions of multiple sensing parameters and the resulting interferences. Although machine learning algorithm has been tried to solve this problem, a commonly used single algorithm often exhibits limited effectiveness in addressing complex decoupling issues involving multiple environmental variables. To address this key challenge, we develop a machine learning powered strategy which utilizes scattering parameter analysis to effectively mitigate cross-interferences among various sensing parameters. We validate this approach using aluminum scandium nitride (AlScN) film-based SAW devices. Employing advanced machine learning techniques such as Multi-Layer Perceptron Regression and Random Forest Regression, we demonstrate significant reduction of influences from the non-targeted parameters. To overcome the limitations associated with employing a single machine learning model to decipher complex datasets and sensing mechanisms, we further propose a Stacking Ensemble model which integrates multiple machine-learning frameworks, enhancing the accuracy of multi-parameter predictions. Notably, for predicting ultraviolet intensities, the Stacking model achieves a substantial reduction in error, exceeding 15 % across three evaluation metrics, when compared to the single Random Forest model. The work provides a promising solution for a cross-interference suppression technology of SAW sensing platform.