Signal conditioning for semiconductor gas sensors being used as detectors in gas-chromatographs and similar applications

Abstract The ability of feedforward neural networks to approximate the dynamic behavior of linear and non-linear sensors was investigated. The networks were trained with data from simulated and real systems to compute the time varying sensor input from the measurable sensor output. Characteristic curves describing the static and dynamic network properties were calculated. The network performance was tested and visualized in time and frequency domain (Bode diagram). It is shown that feedforward neural networks are an appropriate tool to approximate the dynamic behavior of non-linear gas sensors.