A de‐embedding method with matrix rectification and influences of residual errors on model parameters extraction of InP HEMTs

As the cutoff frequency of InP HEMTs enters the terahertz band, high frequency measurement and modeling techniques in hundreds of gigahertz become urgent needs for further millimeter monolithic integrated circuits design. We proposed a new de‐embedding method linking device measurements and modeling based on full EM simulation data acquired from HFSS and advanced design system (ADS). The simulation results for passive dummy structures are well consistent with experiments, and the de‐embedding method is proved very effective for a resistive passive device with high distributed embedding surroundings in frequency range below 40 GHz. Based on these experimental facts, the EM simulations were extended up to 300 GHz and corresponding de‐embedding deviation was further investigated. Results show that the proposed de‐embedding method has very high accuracy in the whole frequency region with a maximum S‐parameters deviation of only 2.58%. However, further analysis proves that the small residual errors still significantly affect extracted small signal model parameters of InP HEMTs especially for transit time τ. Thus, further improvements on de‐embedding accuracy or careful considerations of more error functions in modeling process are necessary for obtaining physically meaningful model parameters.     

Bayesian inference‐based small‐signal modeling technique for GaN HEMTs

A new modeling methodology for gallium nitride (GaN) high‐electron‐mobility transistors (HEMTs) based on Bayesian inference theory, a core method of machine learning, is presented in this article. Gaussian distribution kernel functions are utilized for the Bayesian‐based modeling technique. A new small‐signal model of a GaN HEMT device is proposed based on combining a machine learning technique with a conventional equivalent circuit model topology. This new modeling approach takes advantage of machine learning methods while retaining the physical interpretation inherent in the equivalent circuit topology. The new small‐signal model is tested and validated in this article, and excellent agreement is obtained between the extracted model and the experimental data in the form of dc I–V curves and S‐parameters. This verification is carried out on an 8 × 125 μm GaN HEMT with a 0.25 μm gate feature size, over a wide range of operating conditions. The dc I–V curves from an artificial neural network (ANN) model are also provided and compared with the proposed new model, with the latter displaying a more accurate prediction benefiting, in particular, from the absence of overfitting that may be observed in the ANN‐derived I–V curves.     

Microwave frequency small‐signal equivalent circuit parameter extraction for AlInN/GaN MOSHEMT

This article presents an accurate and efficient extraction procedure for microwave frequency small‐signal equivalent circuit parameters of AlInN/GaN metal‐oxide‐semiconductor high electron mobility transistor (MOSHEMT). The parameter extraction technique is based on the combination of conventional and optimization methods using the computer‐aided modeling approach. The S‐, Y‐, and Z‐ parameters of the model are extracted from extensive dynamic AC simulation of the proposed device. From the extracted Y‐ and Z‐ parameters the pad capacitances, parasitic inductances and resistances are extracted by operating the device at low and high frequency pinch‐off condition depending upon requirement. Then, the intrinsic elements are extracted quasi analytically by de‐embedding the extrinsic parameters. S‐parameter simulation of the developed small‐signal equivalent circuit model is carried out and is compared with TCAD device simulation results to validate the model. The gradient based optimization approach is used to optimize the small‐signal parameters to minimize the error between developed SSEC model and device simulation based s‐parameters. The microwave characteristics of optimized SSEC model is carried out (f_T = 169 GHz and f_max = 182 GHz) and compared with experimental data available from literature to validate the model.     

Differential evolution based small signal modeling of GaN HEMT

In this article differential evolution based method of small signal modeling of GAN HEMT has been investigated. The method uses a unique search space exploration strategy to obtain optimized values of intrinsic and extrinsic elements pertaining to compact small signal model from extracted equivalent circuit elements and measured S‐parameter data. Effectiveness of the method has been illustrated by comparing the measured S‐parameter data of a 4 × 0.1 × 75 μm² GaN/SiC HEMT in the frequency range of 1 to 30 GHz wherein modeled and measured data are in good agreement.     

基于Stacking模型融合的光伏发电功率预测

为了提高光伏发电输出功率的预测精度和可靠性,本文提出一种基于Stacking模型融合的光伏发电功率预测方法.选取某光伏电站温度、湿度、辐照度等历史实测数据为研究对象,在将光伏发电功率数据进行特征交叉以及基于模型的递归特征消除法进行预处理和特征选择的基础上,以XGBoost、LightGBM、RandomForest 3种机器学习算法作为Stacking集成学习的第一层基学习器,以LinearRegression作为第二层元学习器,构建了多个机器学习算法嵌入的Stacking模型融合的光伏发电功率预测模型.预测结果表明,该方法的R²、MSE分别达到了0.9874和0.1056,相较于单一的机器学习模型,预测精度显著提升.     

基于k个标记样本的弱监督学习框架

聚类是机器学习领域中的一个研究热点,弱监督学习是半监督学习中一个重要的研究方向,有广泛的应用场景.在对聚类与弱监督学习的研究中,提出了一种基于k个标记样本的弱监督学习框架.该框架首先用聚类及聚类置信度实现了标记样本的扩展.其次,对受限玻尔兹曼机的能量函数进行改进,提出了基于k个标记样本的受限玻尔兹曼机学习模型.最后,完成了对该模型的推理并设计相关算法.为了完成对该框架和模型的检验,选择公开的数据集进行对比实验,实验结果表明,基于k个标记样本的弱监督学习框架实验效果较好.     

Lp范数约束的多核半监督支持向量机学习方法

在机器学习领域,核方法是解决非线性模式识别问题的一种有效手段.目前,用多核学习方法代替传统的单核学习已经成为一个新的研究热点,它在处理异构、不规则和分布不平坦的样本数据情况下,表现出了更好的灵活性、可解释性以及更优异的泛化性能.结合有监督学习中的多核学习方法,提出了基于L_p范数约束的多核半监督支持向量机(semi-supervised support vector machine,简称S³VM)的优化模型.该模型的待优化参数包括高维空间的决策函数f_m和核组合权系数θm.同时,该模型继承了单核半监督支持向量机的非凸非平滑特性.采用双层优化过程来优化这两组参数,并采用改进的拟牛顿法和基于成对标签交换的局部搜索算法分别解决模型关于f_m的非平滑及非凸问题,以得到模型近似最优解.在多核框架中同时加入基本核和流形核,以充分利用数据的几何性质.实验结果验证了算法的有效性及较好的泛化性能.     

Incorporating domain knowledge in machine learning for soccer outcome prediction

The task of the 2017 Soccer Prediction Challenge was to use machine learning to predict the outcome of future soccer matches based on a data set describing the match outcomes of 216,743 past soccer matches. One of the goals of the Challenge was to gauge where the limits of predictability lie with this type of commonly available data. Another goal was to pose a real-world machine learning challenge with a fixed time line, involving the prediction of real future events. Here, we present two novel ideas for integrating soccer domain knowledge into the modeling process. Based on these ideas, we developed two new feature engineering methods for match outcome prediction, which we denote as recency feature extraction and rating feature learning. Using these methods, we constructed two learning sets from the Challenge data. The top-ranking model of the 2017 Soccer Prediction Challenge was our k-nearest neighbor model trained on the rating feature learning set. In further experiments, we could slightly improve on this performance with an ensemble of extreme gradient boosted trees (XGBoost). Our study suggests that a key factor in soccer match outcome prediction lies in the successful incorporation of domain knowledge into the machine learning modeling process.

     

基于随机森林的卫星快变遥测数据建模

现代卫星已逐渐成为国家重大基础设施,为了解其在轨运行状态,需要对遥测数据进行分析;其中快变遥测数据包含了大量卫星服务情况信息,对该数据进行基于机器学习算法的分析建模,可以更好利用特征维度高、数据量大的快变遥测数据,为人工智能在卫星数据建模、运维方面提供一种可能方案;提出一种基于随机森林算法对在轨卫星快变遥测数据进行建模的方法,并引入改进的二次网格搜索方法对模型参数进行调优;使用模型对某频点功率测量值进行预测,结果显示R2值达到0.98以上,预测值误差较小,建立了效果较好的快变遥测数据模型,为实现基于机器学习的快变遥测数据分析提供了一种可能的方案;     

High‐order internal model‐based iterative learning control design for nonlinear distributed parameter systems

This article deals with the problem of iterative learning control algorithm for a class of nonlinear parabolic distributed parameter systems (DPSs) with iteration‐varying desired trajectories. Here, the variation of the desired trajectories in the iteration domain is described by a high‐order internal model. According to the characteristics of the systems, the high‐order internal model‐based P‐type learning algorithm is constructed for such nonlinear DPSs, and furthermore, the corresponding convergence theorem of the presented algorithm is established. It is shown that the output trajectory can converge to the desired trajectory in the sense of (L²,λ) ‐norm along the iteration axis within arbitrarily small error. Finally, a simulation example is given to illustrate the effectiveness of the proposed method.     

B model: A browsing behavior model based on High-Level Petri Nets to generate behavioral patterns for e-learning

As Internet use has proliferated, e-learning systems have become increasingly popular. Many researchers have taken a great deal of effort to promote high quality e-learning environments, such as adaptive learning environments, personalized/adaptive guidance mechanisms, and so on. These researches need to collect large amounts of behavioral patterns for the verification and/or experimentation. However, collecting sufficient behavioral patterns usually takes a great deal of time and effort. To solve this problem, this paper proposes a browsing behavior model (B² model) based on High-Level Petri Nets (HLPNs) to model and generate students’ behavioral patterns. The adopted HLPN contains (1) Colored Petri Nets (CPNs), in which colored tokens can be used to identify and separate student, learning content and assessment, and (2) Timed Petri Nets (TPNs), in which time variable can be used to represent the time at which a student reads learning content. Besides, to validate the viability of the B² model, this paper implements a B² modeling tool to generate behavioral patterns. The generated behavioral patterns are compared with actual behavioral patterns collected from elementary school students. The results confirm that the generated behavioral patterns are analogous to actual behavioral patterns.     

An improved millimeter‐wave small‐signal modeling approach for HEMTs

An improved method to determine the small‐signal equivalent circuit model for HEMTs is presented in this study, which is combination of the analytical approach and empirical optimization procedure. The parasitic inductances and resistances are extracted under pinch‐off condition. The initial intrinsic elements are determined by conventional analytical method. Advanced design system (agilent commercial circuit simulator) is used to optimize the whole model parameters with small deviation of initial values. An excellent agreement between measured and simulated S‐parameters is obtained for 2 × 20 μm² gate width HEMT up to 40 GHz. © 2013 Wiley Periodicals, Inc. Int J RF and Microwave CAE 24:464–469, 2014.     

18‐31 GHz GaN wideband low noise amplifier (LNA) using a 0.1 μm T‐gate high electron mobility transistor (HEMT) process

GaN technology has attracted main attention towards its application to high‐power amplifier. Most recently, noise performance of GaN device has also won acceptance. Compared with GaAs low noise amplifier (LNA), GaN LNA has a unique superiority on power handling. In this work, we report a wideband Silicon‐substrate GaN MMIC LNA operating in 18‐31 GHz frequency range using a commercial 0.1 μm T‐Gate high electron mobility transistor process (OMMIC D01GH). The GaN MMIC LNA has an average noise figure of 1.43 dB over the band and a minimum value of 1.27 dB at 23.2 GHz, which can compete with GaAs and InP MMIC LNA. The small‐signal gain is between 22 and 25 dB across the band, the input and output return losses of the MMIC are less than ?10 dB. The P_1dB and OIP3 are at 17 dBm and 28 dBm level. The four‐stage MMIC is 2.3 × 1.0 mm² in area and consumes 280 mW DC power. Compared with GaAs and InP LNA, the GaN MMIC LNA in this work exhibits a comparative noise figure with higher linearity and power handling ability.     

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.     

Frequency‐dependent behavioral‐model‐based nonlinear analysis and design of a low‐profile transmit active phased array antenna

Design and behavioral‐model‐based nonlinear analysis of a 3‐GHz active‐phased array antenna (APAA) are presented. Four nonlinear power amplifiers are employed in the output ports of the feeding network (FN) and analyzed based on a 5‐order polynomial model with frequency‐dependent coefficients. The FN is based on 4‐port new Gysel power dividers and combiners arranged in such a way to feed the array with Gaussian‐like amplitude and in‐phase distributions. Beam steering capability is obtained in 2 directions by a new technique including a phase shifter and an amplitude controller (AC). The features result in a low‐profile APAA whose design and fabrication complexity and cost are reduced. Single and 2‐tone power tests are performed to develop analytical expressions in nonlinear region for array factor as a function of the model, FN and the phase and ACs. A similar system with frequency‐independent model is also analyzed for comparison in terms of scan loss, beamwidth, side‐lobe level, beam position, and gain. A microstrip array antenna including the power amplifiers, pre‐amplifiers, AC, delay‐line‐based phase shifters and Gysels is fabricated and measured. The simulation results at the single and dual tones and the intermodulation products are presented which have a good agreement with the measurements.     

Reliability estimation using a genetic algorithm-based artificial neural network: An application to a load-haul-dump machine

In this study, a neural network-based model for forecasting reliability was developed. A genetic algorithm was applied for selecting neural network parameters like learning rate (η) and momentum (μ). The input variables of the neural network model were selected by maximizing the mean entropy value. The developed model was validated by applying two benchmark data sets. A comparative study reveals that the proposed method performs better than existing methods on benchmark data sets. A case study was conducted on a load-haul-dump (LHD) machine operated at a coal mine in Alaska, USA. Past time-to-failure data for the LHD machine were collected, and cumulative time-to-failure was calculated for reliability modeling. The results demonstrate that the developed model performs well with high accuracy (R² = 0.94) in the failure prediction of a LHD machine.     

Identifying mortality risk factors amongst acute coronary syndrome patients admitted to Arabian Gulf hospitals using machine‐learning methods

Acute coronary syndrome (ACS) is a leading cause of mortality and morbidity in the Arabian Gulf. In this study, the in‐hospital mortality amongst patients admitted with ACS to Arabian Gulf hospitals is predicted using a comprehensive modelling framework that combines powerful machine‐learning methods such as support‐vector machine (SVM), Naïve Bayes (NB), artificial neural networks (NN), and decision trees (DT). The performance of the machine‐learning methods is compared with that of the performance of a commonly used statistical method, namely, logistic regression (LR). The study follows the current practise of computing mortality risk using risk scores such as the Global Registry of Acute Coronary Events (GRACE) score, which has not been validated for Arabian Gulf patients. Cardiac registry data of 7,000 patients from 65 hospitals located in Arabian Gulf countries are used for the study. This study is unique as it uses a contemporary data analytics framework. A k‐fold (k = 10) cross‐validation is utilized to generate training and validation samples from the GRACE dataset. The machine‐learning‐based predictive models often incur prejudgments for imbalanced training data patterns. To mitigate the data imbalance due to scarce observations for in‐hospital mortalities, we have utilized specialized methods such as random undersampling (RUS) and synthetic minority over sampling technique (SMOTE). A detailed simulation experimentation is carried out to build models with each of the five predictive methods (LR, NN, NB, SVM, and DT) for the each of the three datasets k‐fold subsamples generated. The predictive models are developed under three schemes of the k‐fold samples that include no data imbalance, RUS, and SMOTE. We have implemented an information fusion method rooted in computing weighted impact scores obtain for an individual medical history attributes from each of the predictive models simulated for a collective recommendation based on an impact score specific to a predictor. Finally, we grouped the predictors using fuzzy c‐mean clustering method into three categories, high‐, medium‐, and low‐risk factors for in‐hospital mortality due to ACS. Our study revealed that patients with medical history related to the presences of peripheral artery disease, congestive heart failure, cardiovascular transient ischemic attack valvular disease, and coronary artery bypass grafting amongst others have the most risk for in‐hospital mortality.     

Novel configuration of ultra‐wide band uni‐planar configuration of complementary split ring resonator composite right/left‐handed based band pass filter with sharp roll off and good stopband attenuation

In this paper, a compact novel simple design of ultra‐wide bandpass filter with high out of band attenuation is presented. The filter configuration is based on combining an ultra‐wide band composite right/left‐handed (CRLH) band pass filter (BPF) with simple uni‐planar configuration of complementary split ring resonator (UP‐CSRR). By integrating two UP‐CSRR cells, the ultra‐wideband CRLH filter roll‐off and wide stopband attenuation are enhanced. The filter has 3 dB cutoff frequencies at 3.1 GHz and 10.6 GHz with insertion loss equals 0.7 dB in average and minimum and maximum values of 0.48 dB and 1.05 dB, respectively over the filter passband. Within the passband. The transition band attenuation from 3 dB to 20 dB is achieved within the frequency band 1.9 GHz to 3.1 GHz (48%) at lower cutoff and the frequency band 10.6 GHz to 11.4 GHz (7%) at upper stopband. Moreover, the filter has a wide stopband attenuation >20 dB in frequencies 11 GHz to 13.6 GHz (21%) and ends with 3 dB cutoff frequency at 14.8 GHz. Furthermore, the designed filter size is very compact (23 × 12 mm²) whose length is only about 0.17 λ_g at 6.85 GHz. The filter performance is examined using circuit modeling, full‐wave simulations, and experimental measurements with good matching between all of them.     

Stability investigation for InP DHBT mm‐wave power amplifier

In this article, we discuss stability issues for mm‐wave monolithic integrated power amplifiers using InP double heterojunction bipolar transistor (DHBT) technology targeting E‐band applications at 71–76 GHz and 81–86 GHz. Different stability detection methods based on the classical two‐port K‐Δ_s pair, linear three‐port graphical analysis, system identifications, circuit modal analysis, and normalized determinant function are all reviewed. The corresponding techniques are employed to predict the occurrence of instability at 15 GHz observed during measurements on a fabricated monolithic microwave integrated circuit power amplifier. Experimental results from a redesigned power amplifier with improved stability are presented to confirm that the previously detected oscillation loop is removed using odd‐mode stabilization resistors with the correct choice of values and locations. © 2012 Wiley Periodicals, Inc. Int J RF and Microwave CAE 23: 662–674, 2013.