Anthony Martin

Feature selective validation (FSV) for validation of computational electromagnetics (CEM). part I-the FSV method

A goal for the validation of computational electromagnetics (CEM) is to provide the community with a simple computational method that can be used to predict the assessment of electromagnetic compatibility (EMC) data as it would be undertaken by individuals or teams of engineers. The benefits of being able to do this include quantifying the comparison of data that has hitherto only been assessed qualitatively, to provide the ability to track differences between model iterations, and to provide a means of capturing the variability and range of opinions of groups and teams of workers. The feature selective validation (FSV) technique shows great promise for achieving this goal. This paper presents a detailed analysis of the FSV method, setting it firmly in the context of previous comparison techniques; it suggests the relationship between validation of graphically presented data and the psychology of visual perception. A set of applicability tests to judge the effectiveness of computer-based CEM validation techniques is also proposed. This paper is followed by a detailed comparison with visual assessment, which is presented in Part II

The feature selective validation (FSV) method

The feature selective validation (FSV) method is one of the candidate techniques for the quantitative validation of computational electromagnetics (CEM), particularly within electromagnetic compatibility (EMC) and signal integrity (SI). In applications so far, it has demonstrated significant promise and is likely to be a central feature of the growing research interest in CEM validation. This paper presents a detailed review of the FSV method and implementation. It concludes with a discussion of some of the current research topics.

Progress in quantifying validation data

There are a number of reasons why a numerical value is a desirable outcome for the validation of numerical models: these include the desire to report objectively and succinctly on the comparison with a benchmark result, and the need to rank order implementations of a canonical problem. The key challenge is to develop techniques which provide a numerical output that do not challenge excessively the interpretations of experienced engineers and which presents both sufficient discrimination between comparisons and consistency across the various sub-disciplines of EMC. While this paper concentrates on the subject of numerical code generation, the rationale and associated concepts are just as relevant to other aspects of EMC such as quantifying experimental repeatability. The purpose of this paper is to present some of the work done to date on generating a numerical value representing the overall quality of comparison for two sets of data. This paper reviews and contrasts various techniques currently available to produce this numerical overview. The techniques include correlograms, feature selective validation (FSV), integrated error against log frequency (IELF) and reliability factors. The paper concludes that the most taxing aspect of the work required to develop such techniques is benchmarking against human interpretation, a rating scale is discussed which can assist in obtaining this information.

Issues in validation of complex-valued simulations for signal integrity analysis

Signal integrity is concerned with such design issues as signal distortions caused by the effects of unmatched impedances, designed-in or parasitic capacitance/inductance and current return paths. It is of increasing importance because of the widespread use of high-speed digital circuitry and it is, therefore, a natural ally to EMC engineering. Circuit simulators, such as SPICE are widely used to analyze circuit performance, with significant success. More interest is being generated in mixed circuit/field models and full wave simulators in order to extend the applicability of signal integrity investigations. Like more established EMC studies, the validation of numerical models for this purpose is an issue to be addressed. This is particularly important, as the models may be required to compare complex valued data as opposed to only magnitude (or single valued data) as in the case of electric network characterization or equivalent circuit extraction. This paper considers how the feature selective validation (FSV) scheme can be applied to quantifying complex valued data.