神经网络曲线拟合在温补晶振上的应用  

Application of neural network curve fitting in TCXO

作　　者：高古鹏 陈向东[1] 谢冰青[2] 李建龙[1] 

机构地区：[1]西南交通大学,四川成都610031 [2]西华大学,四川成都610039

出　　处：《现代电子技术》2014年第20期84-86,90,共4页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(61171050)

摘　　要：石英晶振作为重要的频率源器件，其频率稳定度至关重要。但温度对石英晶振的影响很大。传统的微处理器温度补偿晶振中在拟合曲线时算法简单，导致软件引起的误差较大。利用了神经网络算法在曲线拟合上的应用，拟合出补偿电压与温度之间的函数关系，微处理器根据温度传感器采集的温度控制AD芯片产生补偿电压，从而使压控振荡电路输出稳定的频率的目的。实验结果表明：温度在-10～80℃时，频率稳定度达到±0.35 ppm，比未补偿时提高了近20倍，比其他曲线拟合方法得出的效果要好。The quartz crystal oscillator is an important frequency source device. Its stability is crucial. However there is a great influence of temperature on the quartz crystal oscillator. The traditional MCU quartz crystal oscillator of temperature com-pensation may result in large error caused by software because its algorithm for curves fitting is too simple. In this paper,a func-tion relation between compensation voltage and temperature was fitted by using a neural network algorithm for curves fitting. The MCU controls AD chip to generate a compensation voltage according to temperature collected by temperature sensors,so as to make the VOC circuit output a stable frequency. The experimental results show that,when the temperature is from -10oC to 80oC,frequency stability gets 0.35 ppm,which increases nearly 20 times greater than the frequency stability without compensa-tion,and the results got from the proposed method is better than those of other curve fitting methods.

关 键 词：石英晶振 微处理器 温度补偿 神经网络算法 曲线拟合 

分 类 号：TN752.34[电子电信—电路与系统]