一种基于广义回归神经网络的微波功率器件X参数建模方法  

An X-parameter modeling method for microwave power devices based on generalized regression neural networks

作　　者：林倩 王晓政 杨姝玥 邬海峰 王静[1] LIN Qian;WANG Xiaozheng;YANG Shuyue;WU Haifeng;WANG Jing(School of Physics and Electronic Information Engineering,Qinghai Minzu University,Xining810007,China;Chengdu Ganide Technology Co.,Ltd.,Chengdu610000,China)

机构地区：[1]青海民族大学物理与电子信息工程学院,青海西宁810007 [2]成都嘉纳海威科技有限责任公司,四川成都610000

出　　处：《电子元件与材料》2023年第12期1490-1497,共8页Electronic Components And Materials

基　　金：国家自然科学基金(62161046);青海省重点研发与转化专项科技援青项目(2022-QY-212)。

摘　　要：为了解决微波功率器件大信号参数提取复杂的问题,提出了一种基于广义回归神经网络(GRNN)对微波功率器件进行建模的方法。与此同时,采用双隐藏层BP神经网络进行建模。为了验证两种模型的建模效果,采用谐波平衡实验得到预测数据与期望数据的三次谐波及其模值。最后通过对比得出,双隐藏层BP神经网络模型的三次谐波误差分别为5.287,3.320和4.483 dBm。GRNN模型的三次谐波误差分别为0.130,0.001和1.235 dBm。此外,双隐藏层BP神经网络模型的三次谐波模值误差分别为0.003,0.521×10^(-4)和0.683×10^(-6)。GRNN模型的三次谐波模值误差分别为0.001,0.235×10^(-4)和0.304×10^(-6)。通过以上实验证明了所提出的GRNN模型可以有效地对GaN高电子迁移率晶体管进行大信号建模。To solve the complex problem of extracting large signal parameters of microwave power devices,a modeling method was proposed for microwave power devices,which was based on generalized regression neural network(GRNN).Meanwhile,a double hidden layer BP neural network was used for modeling as well for comparison.To verify the two modeling methods,harmonic balance experiments were conducted to obtain the three harmonics and modulus of the predicted and expected data.Finally,the three harmonic errors are 5.287 dBm,3.320 dBm and 4.483 dBm by using the double hidden layer BP neural network model,which are 0.130 dBm,0.001 dBm and 1.235 dBm,respectively,for the GRNN model.The three harmonic modulus errors are 0.003,0.521×10^(-4) and 0.683×10^(-6) for the double hidden layer BP neural network model,which are 0.001,0.235×10^(-4) and 0.304×10^(-6),respectively,for the GRNN model.These results demonstrate that the GRNN model can effectively simulate the GaN high electron mobility transistors with large signals.

关 键 词：GRNN BP神经网络 X参数 微波功率器件 

分 类 号：TN61[电子电信—电路与系统]