In this paper, systematic neural network modeling techniques are presented for microwave and design using the concept of inverse where inputs to model electrical parameters outputs geometrical parameters. Training directly may become difficult due nonuniqueness input-output relationship in model. We propose a new method solve such problem by detecting multivalued solutions training data. The data containing divided into groups according derivative information forward that individual do not...

This paper proposes an advanced technique to develop combined neural network and pole-residue-based transfer function models for parametric modeling of electromagnetic (EM) behavior microwave components. In this technique, networks are trained learn the relationship between pole/residues functions geometrical parameters. The order pole-residue may vary over different regions We a tracking solve order-changing problem. After proposed process, model can be used provide accurate fast prediction...

This article introduces the deep neural network method into field of high-dimensional microwave modeling. Deep learning is nowadays highly successful in solving complex and challenging pattern recognition classification problems. investigates use networks to solve modeling problems that are much more than solved by previous shallow networks. The most commonly used activation function existing rectified linear unit (ReLU), which a piecewise hard switch function. However, such ReLU not...

This paper presents a new technique for artificial neural network (ANN) inverse modeling and applications to microwave filters. In of component, the inputs model are electrical parameters such as S-parameters, outputs geometrical or physical parameters. Since analytical formula input-output relationship does not exist, ANN becomes logical choice, because it can be trained learn from data in modeling. The main challenge is nonuniqueness problem. problem that different training samples with...

This article presents an overview of artificial neural network (ANN) techniques for a microwave computer-aided design (CAD). ANN-based are becoming useful performing forward/inverse modeling active/passive components to enhance circuit design. With measured or simulated data devices, ANNs can be trained learn relevant relationships, which are, otherwise, computationally expensive efficient analytical formulas not available. Fundamental concepts the ANN structure and training, such as...

The trend of using accurate models such as physics-based FET models, coupled with the demand for yield optimization results in a computationally challenging task. This paper presents new approach to microwave circuit and statistical design featuring neural network at either device or levels. At level, represents physics-oriented model yet without need solve physics equations repeatedly during optimization. speeds up by replacing repeated simulations. method is faster than direct original...

Neural networks have recently been introduced to the microwave area as a fast and flexible vehicle modeling, simulation optimization. In this paper, novel neural network structure, namely, knowledge-based (KBNN), is proposed where empirical or semi-analytical information incorporated into internal structure of networks. The knowledge complements capability learning generalization by providing additional which may not be adequately represented in limited set training data. Such becomes even...

For the first time, we present modeling of microwave circuits using artificial neural networks (ANN's) based on space-mapping (SM) technology, SM-based neuromodels decrease cost training, improve generalization ability, and reduce complexity ANN topology with respect to classical neuromodeling approach. Five creative techniques are proposed generate neuromodels. A frequency-sensitive neuromapping is applied overcome limitations empirical models developed under quasi-static conditions, Huber...

Modeling and computer-aided design (CAD) techniques are essential for microwave design, especially with the drive towards first-pass success. We have described neural networks modeling design. Neural suitable when a required relationship which analytical formulas hard to derive, or computational effort is too high. This can be either of IO overall model (straight network model), output-input (inverse between existing desired data (neuro-SM), subpart (knowledge based network). fast evaluate,...

Yield-driven optimization is important in microwave design due to the uncertainties introduced manufacturing process. For first time, we extend this paper use of polynomial chaos (PC) approach from electromagnetic (EM)-based yield estimation EM-based structures. We formulate a novel objective function for yield-driven EM optimization. By incorporating PC coefficients into formulation, analytically related variables, which are nominal point. then derive sensitivity formulas with respect...

Space mapping (SM) is a recognized method for speeding up electromagnetic (EM) optimization. The SM technique often requires an equivalent circuit as the coarse model. In practical cases, models are not always available. This letter addresses this situation and proposes new coarse- fine-mesh EM optimization incorporating mesh deformation. By deformation into coarse-mesh optimization, responses change continuously values of geometrical design variables change. incorporation also improves...

In this paper, we propose an efficient knowledge-based automatic model generation (KAMG) technique aimed at generating microwave neural models of the highest possible accuracy using fewest accurate data. The is comprehensively derived to integrate three distinct powerful concepts, namely, generation, knowledge networks, and space mapping. For first time, simultaneously utilize two types data generators, coarse generators that are approximate fast (e.g., two-and-one-half-dimensional...

For the first time, we propose a robust algorithm for automating neural-network-based RF/microwave model development process. Starting with zero amount of training data and then proceeding neural-network in stage-wise manner, can automatically produce neural that meets user-desired accuracy. In each stage, utilizes error criteria to determine additional training/validation samples required their location input space. The dynamically generates these new during training, by automatic driving...

Artificial neural networks (ANN) recently gained attention as a fast and flexible vehicle to microwave modeling design. Fast models trained from measured/simulated data can be used during design provide instant answers the task they have learned. We review two important aspects of neural-network-based modeling, namely, model development issues nonlinear modeling. A systematic description key in approach such generation, range distribution samples input parameter space, scaling, etc., is...

This paper presents a novel sensitivity-analysis-based adjoint neural-network (SAANN) technique to develop parametric models of microwave passive components. allows robust model development by learning not only the input-output behavior modeling problem, but also derivatives obtained from electromagnetic (EM) sensitivity analysis. A derivation is introduced allow complicated high-order be computed simple artificial (ANN) forward-back propagation procedure. New formulations are deduced for...

Neural networks are useful for developing fast and accurate parametric model of electromagnetic (EM) structures. However, existing neural-network techniques not suitable models that have many input variables because data generation training become too expensive. In this paper, we propose an efficient method EM behavior modeling microwave filters variables. The decomposition approach is used to simplify the overall high-dimensional problem into a set low-dimensional sub-neural-network...

This paper proposes a pole-residue-based adjoint neuro-transfer function (neuro-TF) technique with electromagnetic (EM) sensitivity analysis for parametric modeling of EM behavior microwave components respect to changes in geometrical parameters. The purpose is increase model accuracy by utilizing information and speed up development reducing the number training data required developing model. proposed consists original pole-residue based neuro-TF models. New formulations are derived...

Space mapping (SM) is a recognized method for speeding up electromagnetic (EM) optimization. Existing SM approaches are mostly based on sequential computation mechanism. This paper proposes parallel EM In the proposed method, surrogate model developed in each iteration trained to match fine at multiple points simultaneously. Multi-point training and enables be valid larger neighborhood than that standard SM. The formulation of multi-point inherently suited implemented through computation....

This paper proposes a novel technique to develop low-cost electromagnetic (EM) centric multiphysics parametric model for microwave components. In the proposed method, we use space mapping techniques combine computational efficiency of EM single physics (EM only) simulation with accuracy simulation. The responses respect different values geometrical parameters in nondeformed structures without considering other domains are regarded as coarse model. is developed using modeling methods such...

A new Wiener-type dynamic neural network (DNN) approach for nonlinear device modeling is proposed in this paper. The analytical formulation of DNN structure consists a cascade simplified linear part and static part. equations are obtained by vector fitting enhancing the efficiency model. sensitivity analysis technique derived to train with dc, small- large-signal data. gradient-based training algorithm also formulated speed up model can be trained accurate relative Furthermore, provides...

This article proposes a multifeature-assisted neuro-transfer function (neuro-TF) surrogate-based electromagnetic (EM) optimization technique exploiting trust-region algorithms for microwave filter design. The proposed addresses the situation where response of starting point is far away from design specifications. We propose to utilize multiple feature parameters help move passband into range pole–zero-based neuro-TF introduced in this extract when responses are not explicitly identified....

Knowledge-based neural network modeling techniques using space-mapping concept have been demonstrated in the existing literature as efficient methods to overcome accuracy limitations of empirical/equivalent circuit models when matching new electromagnetic data. For different problems, mapping structures can be different. In this paper, we propose a unified automated model generation algorithm that uses l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Feature-based electromagnetic (EM) optimization techniques can help avoid local minima in microwave design. Zeros of the transfer functions are recently used to extract features when filter responses not explicitly identifiable. This letter proposes a feature zero-adaptation approach enlarge surrogate range by overcoming problem varying orders function w.r.t. changes design variables. In this way, proposed technique allows larger step sizes for optimization, therefore, speeding up overall EM...