Reconstruction of Drifting Sensor Responses Based on Papoulis–Gerchberg Method

This paper presents a method to reconstruct the drifting sensor responses for electronic-nose (E-nose) systems. The sensor drift often exhibits low-frequency behaviors. Assume that the drift is a kind of missing data during the vapor exposure period, the drifting signal can then be recovered by interpolation or extrapolation using the Papoulis-Gerchberg method under the measurement boundary condition. Using a Lorentzian model to describe the sensor response, a reconstruction algorithm for the sensor response is developed by minimizing the drift recovery and modeling errors. We further adopt the principal component analysis technique and the squared prediction error index to detect the noninformative sensors so that the subsequent classification and estimation performance will not be deteriorated. Using experimental E-nose data, the proposed method is shown to be effective in reconstructing the sensor responses.