Genetic algorithm‐based neural‐network modeling approach applied to AlGaN/GaN devices

An accurate equivalent circuit large‐signal model (ECLSM) for AlGaN‐GaN high electron mobility transistor (HEMT) is presented. The model is derived from a distributed small‐signal model that efficiently describes the physics of the device. A genetic neural‐network‐based model for the gate and drain currents and charges is presented along with its parameters extraction procedure. This model is embedded in the ECLSM, which is then implemented in CAD software and validated by pulsed and continuous large‐signal measurements of on‐wafer 8 × 125‐μm GaN on SiC substrate HEMT. Pulsed IV simulations show that the model can efficiently describe the bias dependency of trapping and self‐heating effects. Single‐ and two‐tone simulation results show that the model can accurately predict the output power and its harmonics and the associated intermodulation distortion (IMD) under different input‐power and bias conditions. © 2012 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2013.     

Neural network modeling of microwave FETs based on third‐order distortion characterization

A new method for characterization of HEMT distortion parameters, which extracts the coefficents of a Taylor series expansion of I_ds(V_gs, V_ds), including all cross‐terms, is developed from low‐frequency harmonic measurements. The extracted parameters will be used either in a Volterra series model around a fixed bias point for 3^rd‐order characterization of small‐signal I_ds nonlinearity, or in a large‐signal model of I_ds characteristic, where its partial derivatives are locally characterized up to the 3^rd order in the whole bias region, using a novel neural‐network representation. The two models are verified by one‐tone and two‐tone intermodulation distortion (IMD) tests on a PHEMT device. © 2006 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2006.     

Fuzzy neural‐based approaches for efficient RF/microwave transistor modeling

In today's RF and microwave circuits, there is an ever‐increasing demand for higher level of system integration that leads to massive computational tasks during simulation, optimization, and statistical analyses, requiring efficient modeling methods so that the whole process can be achieved reliably. Since active devices such as transistors are the core of modern RF/microwave systems, the way they are modeled in terms of accuracy and flexibility will critically influence the system design, and thus, the overall system performance. In this article, the authors present neural‐ and fuzzy neural‐based computer‐aided design techniques that can efficiently characterize and model RF/microwave transistors such as field‐effect transistors and heterojunction bipolar transistors. The proposed techniques based on multilayer perceptrons neural networks and c‐means clustering algorithms are demonstrated through examples. © 2008 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2009.     

A consensual modeling of the expert systems applied to microwave devices

In this work, a consensual approach is developed for modeling RF/microwave devices. In the proposed method, multiple individual models generated by an expert system ensemble are combined by a consensus rule that results in a consistent and improved generalization outputting with the highest possible reliability and accuracy. Here, the expert system ensemble is basically constructed by the competitor and diverse regressors which in our case are back‐propagation artificial neural network (ANN), support vector (SV) regression machine, k‐nearest neighbor and least squares algorithms that perform generalization independently from each other. In the case of excessive data, to reduce the amount of the data, the expert system ensemble of regressors can be shown to be trained by a subset consisting of the SVs. Main feature of the consensual modeling can be put forward as due to diversity in generalization process of each member of the ensemble, the resulted consensus model will effectively identify and encode more aspects of the nonlinear relationship between the independent and the dependent variables than will a single model. Thus, in the consensual modeling, an enhanced single model is built by combining the most successful sides of the competitor and the diverse contributors. Finally, consensual modeling is demonstrated typically for the two devices: the first is a passive device modeling which is synthesis of the conductor‐backed coplanar waveguide with upper shielding and the second is an active device modeling which is the noise modeling of a microwave transistor. © 2010 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2010.     

Neural network methods for the modeling and control of welding processes

While welding processes are of great importance in manufacturing, their modeling and control is still subject of research. The highly nonlinear, strongly coupled, and multivariable nature of these processes renders the use of analytical tools practically impossible. In this article a novel approach is presented which employs networks of simple nonlinear units: a neural network. A widely used welding process, the Gas Tungsten Arc Welding is presented and the problem of its modeling and control is exhibited. A very brief introduction to neural networks is followed by presenting the experimental results for modeling the static and dynamic behavior of the process, as well as some practical recommendations regarding the use of the neural network techniques for controlling these processes.     

On the small signal modeling of advanced microwave FETs: A comparative study

Although many successful techniques have been proposed in the last decades for extracting the small signal equivalent circuit for microwave transistors from scattering parameter measurements, small signal modeling is still object of intense research. Further improvement and development of the proposed methods are incessantly required to take into account the continuous and rapid evolution of the transistor technology. The purpose of this article is to facilitate the choice of the most appropriate strategy for each particular case. For that, we present a brief but thorough comparative study of analytical techniques developed for modeling different types of advanced microwave transistors: GaAs HEMTs, GaN HEMTs, and FinFETs. It will be shown that a crucial step for a successful modeling is to adapt accurately the small signal equivalent circuit topology under “cold” condition to each investigated technology. © 2008 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2008.     

Neural network modeling of active devices for use in MMIC design

This work can be classified into three parts: The first part is a multidimensional signal–noise neural network model for a microwave small-signal transistor. Here the device is modeled by a black box, whose small signal and noise parameters are evaluated through a neural network, based upon the fitting of both these parameters for multiple bias and configuration with their target values. The second part is the computer simulation of the possible performance (F,V_i,G_tmax) triplets. In the final part, which is the combination of the first two parts, the performance curves are obtained using the relationships among operation conditions f, V_CE, and I_CE; the noise figure, input VSWR and maximum stable transducer gain.     

Novel neural network modeling method and applications

Neural networks play an important role for designing the parametric model of electromagnetic structures. The current neural network methods are unfit for a circuit model with many input variables because it is costly to extract a large number of the training data and test data to complete the highly nonlinear mapping approximation. This article proposes a new neural network modeling method—the multidimensional neural network model, which can be used to solve the issue of multivariable radiofrequency and microwave passive device modeling. The entire multidimensional neural network modeling problem is simplified into a set of neural network submodels through decomposition method. Then the submodels are combined into an equivalent model, and the final entire model is produced through the neural‐network mapping model developed with the submodels and equivalent model. A microstrip hairpin filter model is developed using the proposed method. The simulation results show the correctness and the effectivity of the proposed method. © 2015 Wiley Periodicals, Inc. Int J RF and Microwave CAE 25:769–779, 2015.     

Neural network approach to trajectory synthesis for robotic manipulators

The paper deals with the collision free trajectory synthesis for industrial robotic manipulators. A new efficient method is proposed that is based on a neural network collision model. The developed iterative transformation procedure provides small computing times for the C-space synthesis and yields sufficiently precise configuration space map for the manipulators with many degrees of freedom. A topologically ordered neural network model is proposed to find the path in the configuration space. The stability of this model is proved using the Lyapunov function technique. To generate the collision model, a modification of the Radial Basis Function Network (RBFN) is used. The developed technique is illustrated by an application example of designing a robotic manufacturing cell for the automotive industry.     

Modeling power and intermodulation behavior of microwave transistors with unified small‐signal/large‐signal neural network models

This article presents a detailed procedure to learn a nonlinear model and its derivatives to as many orders as desired with multilayer perceptron (MLP) neural networks. A modular neural network modeling a nonlinear function and its derivatives is introduced. The method has been used for the extraction of the large‐signal model of a power MESFET device, modeling the nonlinear relationship of drain‐source current I_ds as well as gate and drain charge Q_g and Q_d with respect to intrinsic voltages V_gs and V_ds over the whole operational bias region. The neural models have been implemented into a user‐defined nonlinear model of a commercial microwave simulator to predict output power performance as well as intermodulation distortion. The accuracy of the device model is verified by harmonic load‐pull measurements. This neural network approach has demonstrated to predict nonlinear behavior with enough accuracy even if based only on first‐order derivative information. © 2003 Wiley Periodicals, Inc. Int J RF and Microwave CAE 13: 276–284, 2003.     

Application of data mining methods to efficient microwave active device modeling

In this work, the signal and noise behaviors of a microwave transistor within its operation domain (voltage drain to source–V_DS, current of drain to source—I_DS, frequency—f) are modeled by data mining techniques (DMT) without using any information on the microwave circuit theory. The device is modeled by a black box whose small signal (S) and noise parameters are evaluated through data mining techniques, based on the fitting of both of these parameters for multiple bias and configuration. It has been shown that DMT have a high potential of faithful and efficient device modeling. © 2012 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2013.     

Small signal behavioral modeling technique of GaN high electron mobility transistor using artificial neural network: An accurate,fast, and reliable approach

This article reports a comparative study of two artificial neural network structures and associated variants used to describe and predict the behavior of 2 × 200 μm² GaN high electron mobility transistors (HEMTs), utilizing radiofrequency characterization. Two architectures namely multilayer perceptron and cascade feedforward, have been investigated in this work to develop the behavioral model. A study is conducted utilizing the two architectures, all trained using Levenberg‐Marquardt, in terms of accuracy, convergence rate, and generalization capability to develop the behavioral model of GaN HEMT. However, to ensure the robustness of the model, accuracy, convergence rate, time elapsed, and generalization capability of the proposed model is also tested under couple of training algorithms, activation functions, number of hidden layers and neuron embedded inside it, methods for initialization of weights and bias and certain other vital parameters playing vital role in influencing the model accuracy and effectiveness. An excellent agreement found between measured S‐parameters and the proposed model proves the effectiveness of the proposed approach and excellent prediction ability for a sweeping multibias set and broad frequency range of 1 to 18 GHz. Moreover, a very good generalization capability is also recorded under variation of crucial parameters of GaN HEMT‐based neural model.     

Bias‐dependent small‐signal modeling based on neuro‐space mapping for MOSFET

In this article, bias‐dependent small‐signal modeling approach based on neuro‐space mapping is proposed for MOSFET. Good agreement is obtained between the simulated and measured results for a 130 nm MOSFET in the frequency range of 100 MHz–40 GHz confirming the validity and effectiveness of our approach. In addition, higher accuracy is achieved by our approach in contrast to conventional empirical model. © 2011 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2011.     

Microwave noise modeling for PHEMT using artificial neural network technique

An improved noise model for pseudomorphic high electron mobility transistors (PHEMT) based on the combination of the artificial neural network (ANN) and conventional equivalent circuit modeling technique is presented. The frequency dispersion of the gate noise model parameter P, drain noise model parameter R, and the correlation coefficient C have been taken into account by using an ANN model. The influence of the gate leakage current can be accommodated by using the proposed noise model. The noise model parameters are determined directly from on wafer noise parameters measurement based on the noise correlation matrix technique. Good prediction for noise parameters and significant improvements of the accuracy of noise parameters are obtained up to 26 GHz for 2 × 40 μm gate width (number of gate fingers × unit gate width) 0.25 μm Double Heterojunction δ‐doped PHEMTs over a wide range of bias points. © 2008 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2009.     

MRNN:一种新的基于改进型递归神经网络的WSN动态建模方法:应用于故障检测

提出了一种适用于无线传感器网络WSN的故障检测方法,该方法运用改进的递归神经网络MRNN为WSN的节点、节点的动态特性以及节点间的关系建立相关模型,对WSN节点进行识别和故障检测。MRNN的输入选择建模节点的先前输出值及其邻居节点的当前及先前输出值,模型基于一种新的改进的反向传播型神经网络,该神经网络的输入以及传感器网络的拓扑结构基于通用的非线性传感器模型。仿真实验将MRNN方法与卡尔曼滤波法进行了全面的比较。实验表明,MRNN在置信因子较小的情况下与卡尔曼滤波方法相比有较高的故障检测精度。     

溶解氧浓度的前馈神经网络建模控制方法

针对污水处理过程溶解氧(DO)浓度控制问题,提出了一种基于前馈神经网络的建模控制方法(FNNMC).本文构造了神经网络建模控制系统,通过对建模神经网络和控制神经网络隐含层学习率的分析,证明了学习算法的收敛性以及整个系统的稳定性.最后,本文基于国际基准的Benchmark Simulation Model No.1 (BSMl)进行了仿真实验,验证了合理选取学习率的重要性,并通过与PID和模型预测控制(MPC)等已有控制方法的比较,验证了神经网络建模控制方法针对污水处理过程溶解氧浓度控制具有良好的建模能力,更高的控制精度以及更好的动态响应能力.     

Neural networks implementation for modeling and control design of manufacturing systems

This paper presents a dynamic neural network implementation for the modeling and control design of a class of manufacturing systems. The evolution of the considered systems is supposed to be continuous and non-stochastic. A separate implementation of the system elements is detailed. These elements are then connected together in order to obtain a global net that simulates the behavior of the real system. The obtained model is modular and can be adapted easily for any modification of the system. Permanent correction rules are developed to control the speed of the machines according to a desired profile and to take into consideration the buffers limited capacities. The convergence of the control design is proved. The proposed approach is applied on an exhaust valves assembly workshop.     

水下机器人的神经网络自适应控制

研究了水下机器人神经网络直接自适应控制方法,采用Lyapunov稳定性理论,证明了存在有界外界干扰和有界神经网络逼近误差条件下,水下机器人控制系统的跟踪误差一致稳定有界.为了进一步验证该水控制方法的正确性和稳定性,利用水下机器人实验平台进行了动力定位实验、单自由度跟踪实验和水平面跟踪实验等验证实验.     

Trust region‐based optimization of PKI neural models for RF/microwave devices

This article proposes a new trust region‐based optimization technique for Radio Frequency (RF)/microwave devices. The proposed approach is apt for modeling scenarios, where standard ANN multilayer perceptron (MLP) and Prior Knowledge Input (PKI) models fail to deliver a satisfactory model. This approach feeds output of standard ANN model as knowledge input to PKI model. The ANN model and the PKI model form a symbiotic pair to yield accurate results. In this paper, the dogleg routine is exploited in the process of optimization to obtain valid trust region steps. The proposed method is compared with sensitivity technique via several RF/microwave components. © 2012 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2013.