A Machine-Learning-Enhanced System Leveraging Spread-Spectrum Excitation of CNT-Based Textile Sensors for Ammonia Detection

This paper presents a novel system for ammonia detection, integrating a carbon-nanotube-based textile sensor, an electronic board optimized for rapid measurements, and a machine learning model for accurate concentration prediction. The system employs spread-spectrum excitation signals, enabling ultra-fast measurements with a response time of 8.2 milliseconds, making it highly suitable for real-time applications. The textilebased sensor provides flexibility, allowing for potential integration into wearable devices or portable sensing platforms. Experimental validation was conducted using 6 ammonia concentrations ($0 \%, 0.1 \%, 1 \%, 10 \%, 20 \%$, and 30 %), with 50 consecutive measurements per value to ensure statistical reliability. The collected data was analyzed using multiple regression models, and results demonstrated the Gaussian Process Regression model's superiority, achieving a root mean square error of 2.08 during 10-fold cross-validation. Future research will focus on testing the system with ammonia at parts-per-million (ppm) levels, optimizing the sensor's performance under different environmental conditions, and exploring integration into wearable electronics for biomedical systems.