New normalization procedure to improve signal pattern discrimination power in case of different concentration dependencies of the gas sensors in an array

Sample recognition with a metal oxide gas sensor array is often affected by a signal pattern dependence on the gas concentration, that blurs the class representation. The reason are differences of the sensor elements in the exponent of the power law that governs the concentration dependence of signals of the array elements. Under the power law the standard pattern normalization procedure using signal ratios eliminates the concentration dependence only if the two signals have the same power law exponent. A simple signal preprocessing algorithm is presented, which is able to suppress the differences in the non-linear concentration dependence of the array elements whereby concentration independent signal patterns can be obtained. Thereby the resolution of the pattern discrimination can be significantly improved resulting in improved gas recognition appropriate for a broad concentration range.