MIMO Microwave Sensor for Selective and Simultaneous Detection of Methanol and Ethanol Gases at Room Temperature

Abstract

This article presents a machine learning (ML)-assisted novel microwave sensor for the simultaneous and selective detection of multiple volatile organic compounds (VOCs) at room temperature (RT) using molecularly imprinted polymer (MIP)/carbon nanotube (CNT)-based sensing layers. The synthesized materials were characterized using Fourier-transform infrared (FT-IR) spectroscopy and subsequently used to develop two chemiresistive sensors, two single-port antenna sensors, and a two-port multiple-input-multiple-output (MIMO) antenna sensor, all systematically developed step by step. First, high-precision interdigitated electrode (IDE)-based sensors were functionalized with MIP/CNT sensing materials for individual target VOCs. These sensors were then evaluated for their electrical and gas-sensing properties and optimized to achieve the desired sensitivity. Next, antenna sensors were developed using these optimized structures, demonstrating selectivity for methanol and ethanol with a sensitivity of similar to 1.0 MHz/1000 ppm. The detection limits (DLs) were 77 ppm for ethanol and 90 ppm for methanol, both well below their safety thresholds, ensuring the sensors' suitability for practical applications. ML-based signal processing techniques were used to isolate cross-reactivity between chemical interferents and mutual coupling effects from closely spaced array elements. This enabled the simultaneous and selective detection of multiple VOCs in complex mixtures. The ML models trained on experimental data achieved an impressive F1 score of 0.9917, demonstrating accurate discrimination of VOC types. In addition, the models produced an R-squared value of 0.994 for gas concentration estimation, confirming their predictive accuracy. The developed sensors exhibited high selectivity and specificity when tested against methanol, ethanol, acetone, and isopropanol. Moreover, the antenna sensors operated within their bandwidth during gas sensing, eliminating the need for complex tuning circuits. In addition, perturbation analysis confirmed the ML model's robustness, as it maintained high accuracy despite input noise, ensuring reliable real-world performance.