An Unsupervised Learning Method for Noise Detection in Photoacoustic Gas Measurements

In this work, the authors propose a digital noise detection technique based on an artificial intelligence signal processing method. The proposed technique is intended to be used to detect and identify acoustic interferences such as ambient noise that influence the photoacoustic signal in resonance-based photoacoustic gas sensing systems. Particularly, an unsupervised classification method based on iterative data clustering is implemented to identify and discard samples affected by noisy interferences. Leveraging the inherent characteristics of the acquired photoacoustic gas signal, the algorithm identifies noise events, including glitches and other forms of acoustic interference, from the relevant gas signal. The method employs an iterative clustering technique, where the number of clusters increases with each iteration while adjusting the clustering threshold. As iterations proceed, the threshold decreases, thereby enhancing the system ability to detect noise by excluding more examples from the noise-free class in each step. Experimental validation confirms the feasibility of the proposed algorithm, yielding promising results. This approach presents a viable solution for improving the performance of photoacoustic gas measurement systems in noisy environments, thus contributing to enhanced measurement system resolution.