Approaches to Cleaning Gas Response Signals from Metal Oxide Sensors: Optimisation and Generalizabilty

This paper reports on a comparative analysis of techniques – from simple polynomial curve fitting to digital filters, local regression and wavelet denoising – for cleaning thin film composite metal oxide gas sensor response signals. This research expands and extends a preliminary investigation of simple methods for smoothing metal oxide gas sensor response signals. As part of the analysis an extensive series of systematic experiments were conducted in order to tune the parameters, including span or frame sizes and degrees of polynomial as appropriate, for each of the digital filters and to select the appropriate mother wavelet and threshold chooser for the wavelet approach. The signal processing challenge of maintaining a balance between the measured signal variation and the disparity variation in the smoothed signal is outlined and considered in comparing the performance of the signal cleaning methods. The results indicate that a Savitsky Golay filter with a polynomial degree of 3 and a frame size of 9% of a signal’s width provides a practical solution for denoising metal oxide gas sensor signals because it was found to consistently give a cleaned signal that is suitable for further processing (feature extraction and pattern recognition). This work provides support for the premise that a generalized method for cleaning metal oxide gas sensor signals, regardless of sensor composition, is possible and suggests that a Savitsky Golay filter is a suitable candidate.