In several situations it is essential to be able to measure the composition of a gas mixture, discriminate between or quantify different gases mixtures (odors), or in some cases to give a 'fingerprint', which represents the gas mixture. A gas sensor system is designed to measure certain attributes of a gas mixture. A property of crucial importance of such a system is the ability to identify/quantify attributes of the examined analytes. This property can be referred to as the selectivity of the system. In this work, the objective has been to examine and develop methods, which improve the selectivity of gas sensor systems.

In a distributed gas sensor system a gas mixture is catalytically changed in a controlled way and monitored at different stages of catalytic conversion with a gas sensor array. The acquired sensor response profile thus depends on the sensor properties, the catalyst activity, and the applied gas mixture. Such a system has been extensively examined regarding the operating conditions, e.g. flow rate, different sensors and catalysts, and various analytes in different concentrations. Major differences in response profile between different molecule types were observed. It has been shown that by using an array of identical Pd gate metal-oxidesemiconductor field-effect transistor (MOSFET) sensors and a Pd catalyst, the components in a binary mixture could be quantified. Extending the design to two different catalysts and two types of sensors, the components in quaternary mixtures could be accurately estimated. Thus, the catalyst provides information not available from only the sensor, and the selectivity of the system is consequently improved as compared to a system without catalyst.

From a multi-parameter sensor, more than one signal is acquired, each giving different information. Several approaches can be made to extract multiple parameters from one sensor. A simple method is to analyze the transient response curve from a gas sensor when exposed to a rectangular pulse of test gas. Simple parameters, e.g. responses, derivatives, and integrals, from a single MOSFET sensor, have been shown to provide different information in several applications. These parameters could, furthermore, be used to estimate properties of the analyzed samples with high accuracy. Thus, it was shown that the MOSFET sensor could be used as a multi-parameter sensor as the information content was increased compared to extracting only the steady state response value. A method to optimize these parameters (response, derivative, and integrals), utilizing the leverage from a principal component analysis was developed, further improving the information content.

If an array of multi-parameter sensors is used in a specific application, a large number of variables are acquired. A feature selection process is thus in many cases necessary. A method, based on a forward selection procedure, applying the error from a multiple linear regression as the selection criteria, was developed. Other methods were also examined, but the forward selection procedure outperformed all other tested algorithms. A practical application of gas sensor systems, estimation of wood hydrolyzate properties, was finally studied. Using a multi-sensor array, extracting several parameters from each sensor, and performing a forward selection procedure, suitable variables were found. The variables were then used to estimate interesting properties of the samples with high accuracy using pattern recognition routines.