FSV overview
Motivation
EMC design has often relied on the development of a knowledge base in a limited number of internal experts within an organization, perhaps providing an ‘internal consultancy’ function to other parts of the company. Often, the other parts of the company may have had little direct expertise on EMC. However, this picture is changing for a number of reasons:
Leaner companies mean that more staff need to
have a broader awareness of technical issues but associated with this is the
limitations in developing a ‘group’ knowledge base.
Industry consolidation, partnerships and
collaborations has resulted in more distributed design teams.
Higher frequency circuit operation and increased
space limitations means that circuits need to operate in more taxing
environments.
Design cycle times are much reduced compared
with a few years ago, with the resulting need to invest in, and rely on,
virtual prototyping tools.
Graphical representations of important
parameters, such as cross-talk are becoming visually more complex.
The increase in general complexity of systems
results in less clear-cut technical requirements during design and for
purchasing decisions.
While all measurements will be subject to errors
and measurement artifacts, the fact that distributed teams may be measuring the
same product in several sites and comparing their results with third party
measurements, the question remains: just how similar are these results?
There is a clear need for a knowledge technology that can assist in the comparison of visually complex data in order to quantify an ostensibly subjective assessment and to provide some insight into the underlying ‘physics’ and providing those involved in discussing the results with an objective framework.
One such knowledge technology is
the Feature Selective Validation (FSV) method.
The FSV method is an heuristic technique that was developed to mimic the
process by which a group of engineers look at visually presented data in order
to judge the quality of comparison. It
is based on separate comparisons of the data envelopes and the data’s
individual features. It appears to work
well for a broad range of situations.
Outline of FSV (From IEEE EMC Symposium 2005)
The Feature
Selective Validation (FSV) method is a candidate technique for the quantified
validation of CEM. It is based on the
decomposition of the original data into two parts: amplitude (trend / envelope)
data and feature data. The former will
account for the slowly varying data across the data set and the latter will
account for the sharp peaks and troughs often found in EMC data.
The thrust
for the original development of the FSV was that correlation lacked sensitivity
for the visually complex data typical to the field and was not able to provide
sufficient supporting data to validate the CEM simulations in the way described
earlier. Reliability factors, originally
designed for surface crystallography, showed more promise, but did not satisfy
the acid test of providing an output which is broadly in agreement with the
overall response of a group of engineers – the “group mean”. Hence, the FSV was developed using a similar
approach to the development of reliability factors but recognizing that the
engineers’ decision making is based largely on a
- ADM (Amplitude Difference Measure) and FDM (Feature Difference Measure). These are similarly available as a numerical value or converted to a natural language descriptor. These single figure goodness-of-fit values, combine to give the GDM.
- GDM (Global Difference Measure). An overall single figure goodness-of-fit between the two data sets being compared. This allows a simple decision to be made about the quality of a comparison. This may be numerical or converted to a natural language descriptor (excellent, very good, good, fair, poor, very poor).
- GDMi, ADMi and FDMi. These are point by point comparisons of the amplitude differences, the feature differences and the global difference. This allows a user to analyze the resulting data in some detail, probably with the aim of understanding the origin of the contributors to poor comparisons.
GDMc, ADMc, FDMc. These give a probability density function which shows the proportion of the point-by-point analyses of each of the components that falls into the six natural language descriptor categories). This provides a measure of confidence in the single figure comparison.
Visual rating scale
Throughout this discussion, there is an implication that
visually compared data can be categorized.
This is the case using the visual rating scale developed by Dr
