A comparative study of signal processing techniques for clustering microsensor data (a first step towards an artificial nose)

Microsensor technology has progressed to the point where it is now feasible to place several hundred sensors on a computer chip. Such a sensor array can potentially be used in many applications including detecting hazardous chemical emissions, food processsing, and fire detection. This paper addresses an important aspect involved in microsensor applications, namely how the sensor signals are processed. The problem treated involves classifying whether a sensed signal is generated by one of four chemicals. Two broad approaches to processing the sensor signals are discussed, one based on classical signal processing approaches, and one based on a model of how the olfactory system in animals functions. The classical approaches used include: Gram Schmidt orthogonalization, fast Fourier transforms, and Haar wavelets. For the experimental signals treated, the classical approaches give superior results compared to those produced by the olfactory model.