Pattern recognition using AC treatment for semiconductor gas sensor array

Semiconductor gas sensor using tin oxide as sensing material has been used to detect gases based on the fact that impedance of the sensing material varies when the gas sensor is exposed to the gases. This variation comprises of two parts. The first one is variation in resistance of the sensing material and the other is expressed in terms of the sensor capacitance variation. Normally, only variation of the sensor resistance is considered. In this paper, using AC measurement with a capacitor-coupled inverting amplifier circuit, both changes in the sensor resistance and variations in the sensor capacitance were investigated. These characteristics were represented as magnitude gain and phase shift of AC signal at a specific frequency after passing it through the sensor and the designed circuit. A two-stage artificial neural network, which utilized the information above, was employed to identify and quantify three combustible gases: methane, propane and butane. The network outputs were approximately proportional to concentrations of test gases with reasonable level of error.