Sensors and Actuators B 142 (2009) 19–27 Contents lists available at ScienceDirect Sensors and Actuators B: Chemical journal homepage: www.elsevier.com/locate/snb Extracting discriminative information from the Padé-Z-transformed responses of a temperature-modulated chemoresistive sensor for gas recognition F. Hossein-Babaei ∗ , S.M. Hosseini-Golgoo, Amir Amini Electronic Materials Laboratory, Electrical Engineering Department, K.N. Toosi University of Technology, Tehran 1635-1355, Iran article info Article history: Received 4 August 2008 Received in revised form 16 July 2009 Accepted 27 July 2009 Available online 3 August 2009 Keywords: Artificial olfaction Pattern recognition Chemoresistive gas sensor Temperature modulation Padé-Z transform Feature extraction abstract The response patterns of a temperature-modulated chemoresistive gas sensor were transformed to multi- exponential functions which facilitated the extraction of their discriminative features for gas diagnosis. The patterns were generated for air contaminated with different concentrations of various volatile organic compounds by applying a staircase heating voltage waveform to the microheater of a tin oxide-based sensor that modulated its operating temperature in the 50–400 ◦ C range. Padé-Z transform was utilized for the transformation, and a novel heuristic procedure facilitated the extraction of the components of the feature vectors from the transformed data. These vectors were classified by the available techniques. The method differentiated the patterns generated for methanol, ethanol, 1-propanol, 1-butanol, and acetone contaminations in the wide concentration range examined. The method was also used to separately estimate the amount of the discriminative information in various steady state and transient response features; the results are anticipated to help design more elaborate temperature-modulated sensors for gas diagnosis. © 2009 Elsevier B.V. All rights reserved. 1. Introduction Their attractive combination of quality factors has rendered chemoresistive gas sensors [1–9] suitable for applications rang- ing from the simplest household CO and fire detectors [8] to sensor arrays utilized in artificial olfaction systems [10–13]. In these devices, the electrical resistance of a polycrystalline oxide semiconductor pallet varies with the composition of the surround- ing atmosphere, and the magnitude of this variation defines the response to any polluting, combustible, or poisonous contaminant [1–5]. The sensitivity as well as the response level vary from one target gas to the next, but, in practice, no selective detection is expected from a single sensor as its response to gas A at concentra- tion C A can be similar to that to gas B at a different concentration C B . Different techniques have been used to overcome the lack of selec- tivity in these sensors [4,14–18], among which the modulation of their operating temperature is well known [18–29]. Chemoresistive gas sensors operate at elevated temperatures, and their responses are strongly temperature dependent [7]. A pre- programmed variation of the operating temperature results in a complex temporal response pattern that contains information on ∗ Corresponding author. Tel.: +98 21 8873 4172; fax: +98 21 8876 8289. E-mail addresses: fhbabaei@kntu.ac.ir, fhbabaei@yahoo.com (F. Hossein-Babaei), smhosseini@ee.kntu.ac.ir (S.M. Hosseini-Golgoo), amir.amini@ee.kntu.ac.ir (A. Amini). the nature and concentration of the contaminant present [19]. It has been shown that the extraction and classification of this infor- mation can facilitate contaminant diagnosis [22–29]. The temporal modulation of the operating temperature is achieved by applying the power waveforms of different shapes to the heating element of the sensor [21–24]; an example is the staircase voltage waveform applied to the sensor heater [24,25]. Each step of the staircase brings the sensor to its corresponding temperature plateau and allows enough time for the sensor response to approach its steady state level at that temperature. The created complex temporal pattern contains a number of rises, plateaus, and falls, the details of which are related to the composition of the surrounding atmosphere of the sensor [25]. The response patterns of the temperature-modulated sensors have been processed both in time [24] and frequency domains [26] for the extraction of the diagnostic information. The tempo- ral response pattern obtained for a specific target gas amounts to a large pile of numerical data; the complications of the high dimensional calculations involved are avoided by the applica- tion of proper mathematical transformations on the generated patterns, which map them into a low-dimensional space where they can readily be classified by using appropriate classification techniques. Utilization of the fast Fourier transform for this dimen- sional reduction has resulted in a successful classification of the responses of a temperature-modulated sensor to the binary mix- tures of CO and NO 2 in a wide concentration range in air [21]. Wlodek et al. attempted feature extraction by fitting a family of 0925-4005/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.snb.2009.07.039