The System Of FPGA QCM Humidity Sensor Temperature Compensation Based On Neural Networks

Humidity played a decisive role in the survival and development of natural biological,which is an important physical quantity.The measurement of humidity is also particularly important. Quartz Crystal Microbalance (QCM) sensor is a research hot spot in recent years,which has many advantages such as high sensitivity, selectivity, good stability, wide measurement range.etc.With the development of new humidity sensitive material, QCM sensor also has a great breakthrough in the measurement of humidity.General speaking, QCM humidity sensor is sensitive element which has a quartz crystal thin film surface humidity, As a quartz crystal transducer elements, the use of a frequency variation of the quality of the quartz crystal, the humidity signal into a frequency signal output, enabling humidity detection. But the inherent characteristics of the quartz crystal humidity and temperature characteristics of the material will affect the output of the sensor, and cause the measurement error. In order to reduce the influence of temperature on the QCM humidity sensor, we must design a temperature compensation system. The system use the hardware and software compensation methods.This article is designed to test the humidity of hardware systems. Design of the humidity sensor and the reference QCM crystal oscillation circuit, the resonance frequency of the signal generated after shaping filter, into the differential frequency signal circuit to obtain two frequency difference between the two frequencies, The subtract frequency offset circuit portion due to temperature drift. Then,FPGA to control the temperature sensor on the other hand to get the scene temperature. FPGA based on the detected temperature signal and the frequency signal obtained humidity values by the algorithm of neural network, and displayed it on the LCD.Software compensation is achieved by an FPGA chip from BP neural network algorithm. The whole process can be divided into two steps:1, Using MATLAB toolbox and the measured data to establish BP neural network model. By changing the learning rate and the number of hidden layer neurons get the best BP neural network model.2,The FPGA can achieve optimum propagation stage before BP neural network model. The core stage of prior to spread is achieved by Sigmoid transfer function. Since the Sigmoid function is a nonlinear function, FPGA is difficult to directly achieve,this design achieved by piecewise linear methods to the look-up table.QCM humidity sensor for temperature and humidity related experiments to verify the effect of the temperature compensation system.