程旭瀚 王軍



摘 要 ???:本文提出了一種基于MLP-ELM的GaN HEMT小信號特性的建模方法,首先基于GWO建立了一種混合參數(shù)提取法,解決20元等效電路參數(shù)提取不精確的問題;然后利用等效電路模型獲得的S參數(shù)結(jié)合MLP-ELM建立了一種精確的經(jīng)驗(yàn)?zāi)P停行Ы鉀Q等效電路模型無法在多偏置范圍內(nèi)表征小信號特性的問題;最后利用MLP-ELM建立了一種基于經(jīng)驗(yàn)的小信號模型.經(jīng)過仿真分析得出,本文所建模型精度高,在整個偏置范圍內(nèi)有效且具備等效電路模型不具有的泛化能力.
關(guān)鍵詞 :等效電路模型; 灰狼優(yōu)化算法; S參數(shù); 多層感知器; 極限學(xué)習(xí)機(jī)
中圖分類號 :TN386 文獻(xiàn)標(biāo)識碼 :A DOI : ?10.19907/j.0490-6756.2023.044003
收稿日期: ?2022-11-08
基金項(xiàng)目: ?國家自然科學(xué)基金(69901003); 四川省教育廳科研基金(18ZA0502)
作者簡介: ??程旭瀚(1996-), 男, 研究方向?yàn)槲⒉ㄆ骷<皯?yīng)用. E-mail: chengxh702020@163.com
通訊作者: ?王軍. E-mail: wangjun197008@163.com
An MLP-ELM-based modeling method for the ?small-signal properties of GaN HEMT
CHENG Xu-Han, WANG Jun
(Southwest University of Science and Technology College of Information Engineering, Mianyang 621010, China)
An MLP-ELM-based modeling method for GaN HEMT small-signal properties is proposed. First, a hybrid parameter extraction method is developed based on the GWO to solve the problem of inaccurate extraction of 20-element equivalent circuit parameters; then the S-parameters obtained from the model are combined with MLP-ELM to establish an accurate empirical model, which effectively solves the problem that the equivalent circuit model cannot represent the small signal properties in the multi-bias range; finally, an empirical-based small signal model is developed with MLP-ELM. The simulation results show that the proposed model has high precision, is effective in the whole bias range and has the generalization ability that the equivalent circuit model lacks.
Equivalent circuit model; Gray wolf optimization algorithm; Scattering parameters; Multilayer perceptron; Extreme learning machine
1 引 言 GaN HEMT具有高電子飽和速度和寬帶隙等特點(diǎn),廣泛應(yīng)用于如功率放大器和低噪聲放大器等電路模塊.設(shè)計這些模塊需要可靠、準(zhǔn)確的小信號特性建模方法,常用的方法是等效電路模型. 該模型的準(zhǔn)確性依賴于模型拓?fù)渑c參數(shù)提取的方法 ?[1-3].由于20元等效電路結(jié)構(gòu)復(fù)雜,直接提取法并不適用,因此基于優(yōu)化的提取過程是非常必要的 ?[2,3].這種將優(yōu)化方法與直接提取法相結(jié)合的方法被稱為混合參數(shù)提取法.在模型優(yōu)化問題上群智能算法是可行且有效的 ?[4]. 針對器件建模問題,研究者基于群智能算法提出了眾多的混合參數(shù)提取法 ?[5-7],其中灰狼優(yōu)化算法(Gray Wolf Optimization,GWO)具備魯棒性強(qiáng)……