Study On A Modeling Method For The Thermopile Infrared Sensor Based On Neural Network

Sensor is a kind of test device that can convert the non-electric signals into electrical signals or other signals by a certain law. It is the primary part to achieve automatic detection and control.In order to achieve circuit function simulation and characteristic analysis, Circuit simulation program is used to simulate the electronic devices and circuit modules, which are already existent or the inexistent models, which need to be structured. Because of the lack of sensor models in current circuit simulation software, most of the circuit simulation software can not simulate the circuit with a sensor. All existing sensor modeling studies are aiming at single-input and single-output sensors. But the outputs of many sensors are always affected by many factors, which will cause multiple-input and single-output characteristic. Therefore, the study on a circuit simulation modeling method of sensors with multiple-input and single-output characteristics has practical significance and application value.This paper, taking thermopile infrared sensor as example, studies a modeling method of sensor with multiple-input and single-output, which bases on neural network. Thermopile infrared sensors, the output of which are effected by temperature, chopper frequency, incidence angle and transmission of optical filter, has multiple-input and single-output characteristic. In the modeling process, neural network's nonlinear approximation and self-adaptive ability are fully used. Firstly, the neural network is used to approximate those characteristics respectively, which show the effects of different inputs on output. The network weights and the threshold values can be obtained. Then, PSpice circuit description language is used to describe each neural network architecture respectively, by which the equivalent circuit of each network can be established; Finally, based on the impact of the relationship between inputs and output , outputs of the several sub-circuits are combined and the final output is produced. In order to simulate the real device connection method in the circuit, this paper designs the conversion circuit reflecting the output characteristics of the device. After this process, the final model is achieved. The validity of the established model and the feasibility of the modeling method are proved through the actual simulation tests.The proposed modeling method in this paper has referential significance to other sensor modeling.