A novel sequential design strategy for global surrogate modeling

Crombecq, Karel; Tommasi, Luciano De; Gorissen, Dirk; Dhaene, Tom

doi:10.1109/wsc.2009.5429687

Cited by 84 publications

(61 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In successive steps, additional sampling points are selected one by one in a sequential way until the overall variation of the NF pattern is characterized. In order to sample the NF pattern as efficiently as possible, the robust sampling strategy from [10]- [12] is applied to determine optimal coordinates of the sampling points in a sequential way [13]- [15]. The sampling algorithm makes a balanced trade-off between exploration and exploitation criteria: • Exploitation ensures that additional scans are performed in regions of the design space where the amplitude of the near-field component that is being measured is changing more rapidly.…”

Section: Sequential Sampling Algorithm For Two Simultaneous Modelsmentioning

confidence: 99%

Sequential sampling algorithm for simultaneous near-field scanning of amplitude and phase

Claeys

Pissoort

Deschrijver

et al. 2014

2014 International Symposium on Electromagnetic Compatibility

Self Cite

View full text Add to dashboard Cite

Abstract-This paper describes an automated sequential sampling algorithm for EMI near-field scanning of electronic systems which allows to measure both magnitude and phase of the electromagnetic near-fields simultaneously. The main goal of the sequential sampling algorithm is to drastically reduce the total measurement time to obtain a complete model of the electronic system's near-field distribution. Measuring both magnitude and phase is important for predicting the far-field emission from the near-field or for building equivalent radiation models of the device under test. Previous work described such a sequential sampling algorithm for amplitude-only measurements. The extension towards both amplitude and phase poses two challenges. First, the underlying sampling and modelling techniques have to be adapted such that they can handle building up two separate models at the same time using a common set of optimal sampling points and without significant increase of the measurement time. Second, a good choice has to be made with respect to which components will be sampled and modelled. It is shown that the most advantageous choice is to sample and model the real and imaginary components of the near-fields instead of the amplitude and phase directly.

show abstract

Section: Sequential Sampling Algorithm For Two Simultaneous Modelsmentioning

confidence: 99%

Sequential sampling algorithm for simultaneous near-field scanning of amplitude and phase

Claeys

Pissoort

Deschrijver

et al. 2014

2014 International Symposium on Electromagnetic Compatibility

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, in this method one has to assure that the scores from different candidate rankers are properly normalized. This approach can be applied to combine LOLA-Voronoi candidate rankers, but not the one based on the model error measure [11], [12]. To include the model errorbased candidate ranker in the ranker combination, one can use a random selection.…”

Section: B Adaptive Sampling Algorithmmentioning

confidence: 99%

Nonlinear behavioral models of HEMTs using response surface methodology

Barmuta

Gibiino

Ferranti

et al. 2014

2014 International Conference on Numerical Electromagnetic Modeling and Optimization for RF, Microwave, and Terahertz Applicati

View full text Add to dashboard Cite

Abstract-In this paper, the response surface methodology is proposed to model nonlinear microwave devices using different sampling techniques. Each of the methods represents a distinct approach: exploration-oriented (Voronoi tessellation), nonlinearity-exploitation-oriented (LOcal Linear Approximation) and model-error-minimization-oriented. This allows to build accurate and compact global behavioral models of drain voltage at different harmonics of a 0.15 µm GaAs HEMT transistor with only few hundreds of samples. After choosing the best sampling technique, two types of global models are compared: Radial Basis Function and Kriging. It is shown that the modeling convergence depends on the model type, and better results are obtained using the Kriging model.

show abstract

“…The toolbox provides a number of sequential sampling strategies (Voronoi, Kriging, LOcal Linear Approximation (LOLA), etc.). We use the LOLA-Voronoi sampling strategy [4], which implements a trade-off between exploration (filling up the space to sample as equally as possible) and exploitation (selecting data points in highly nonlinear regions). The LOLA method identifies non-linear regions by comparing the gradients at the neighboring samples, while the Voronoi tessellation maximizes the distances among the samples.…”

Section: A Modeling Proceduresmentioning

confidence: 99%

“…It is used to construct Xparameters behavioral models with a small number of samples, and at the same time of handling errors related to lack of simulator convergence and violation of the linearity assumptions. The technique is based on sequential sampling and behavioral modeling using Radial Basis Function (RBF) models [4]. First, we investigate two common empirical GaAs HEMT models (i.e., Chalmers [5] and Curtice3 [6] models) with respect to their validity region using the sequential sampling.…”

Section: Introductionmentioning

confidence: 99%

Efficient Generation of <formula formulatype="inline"> <tex Notation="TeX">${\rm X}$</tex></formula>-Parameters Transistor Models by Sequential Sampling

Barmuta

Ferranti

Lewandowski

et al. 2014

IEEE Microw. Wireless Compon. Lett.

View full text Add to dashboard Cite

Abstract-This paper proposes a sequential sampling technique to generate efficiently multidimensional X-parameters models for microwave transistors, while guaranteeing X-parameters' validity and overcoming simulator convergence issues. The sequential sampling process selects a set of samples that are subsequently used to construct behavioral models with radial basis functions. The proposed method was compared with a tabular Xparameters model with cubic spline interpolation. The radial basis function models demonstrate very fast convergence and greater accuracy already for a few tens of samples. The proposed technique is illustrated for a GaAs HEMT using Curtice3 and Chalmers empirical model simulations as the data source.

show abstract

A novel sequential design strategy for global surrogate modeling

Cited by 84 publications

References 21 publications

Sequential sampling algorithm for simultaneous near-field scanning of amplitude and phase

Sequential sampling algorithm for simultaneous near-field scanning of amplitude and phase

Nonlinear behavioral models of HEMTs using response surface methodology

Efficient Generation of <formula formulatype="inline"> <tex Notation="TeX">${\rm X}$</tex></formula>-Parameters Transistor Models by Sequential Sampling

Contact Info

Product

Resources

About