Research On Array Technology Of Surface Acoustic Wave Gas Sensor

As much attention has been paid to chemical weapons in all over the world, many countries such as USA and the Netherlands have put a lot of resource in researching the toxic agents sensors and even have put the sensors in anti-terrorism. However, the domestic research of surface acoustic wave (SAW) based chemical sensors to detect chemical agents just starts from the very beginning. Particularly, there is a lot work to do in the field of SAW sensor array.The paper firstly explains the principles of SAW gas sensor and its array, introduces two different structure SAW devices, discusses the adsorption mechanism of the sensitive membrane and the impact of sensitive membrane to sensors sensitive principle as well as the advantages of hyperbranched polymer membrane as sensitive coating. The concept of distribution coefficient is given to show the gas absorption of polymer coating. The Linear Salvation Energy Relationship(LSER) is also given to indicate the gas-polymer distribution coefficient, which provides the principles of the selection of polymer film.The paper describes in details two kinds of detection circuits: the traditional detection circuit and a Direct Digital Synthesizer (DDS) combind with Phase Locked Loop (PLL) detection circuit. Compared with the traditional solution, DDS + PLL solution has the features of good stability, convenience in measurement as well as high accuracy. Moreover it does not need the frequency counter circuit.Finally, pattern recognition algorithm is researched in two parts of the signal pre-processing and artificial neural networks. The simulation and analysis are implemented and the results show that the normalized algorithm for array data eliminates the interference due to changes in concentration. So it is an effective way to pre-process the signal data when we focus on exact type of gas without the concentration factor. The simulation of BP, LVQ, and PNN neural network is also done and the results show that PNN and LVQ have the identification rate of 100%, while BP network can only reach 80%. PNN has the advantage on training speed. But in the face of a large number of training samples, PNN has difficulties in hardware realization.