Michael,
The only such technique I know of to indicate uncertainty
is through a sampling test performed under CISPR 16.
A minimum of 3 samples can be used although the
test recommends 5 samples or more.
AT EACH EMISSION FREQUENCY each sample is measured and
the arithmetic mean and sample standard deviation of the sample
set is calculated. The following calculation is then performed:
X + kS <= L
Where: X = Arithmetic Mean
S = Sample Standard Deviation
k = a constant
L = the measurement Limit in dB
The constant, k, is derived from a non-central "t" distribution and
is a function of the sample size, the desired confidence level and the
desired % of units that can be predicted to pass the limit based on that
confidence level. The larger the sample set, the smaller is k.
As a result of this test, if the value is at or below the limit, the test
indicates that 80% of production would pass with an 80% level of
confidence. If the sum is greater than the limit, L, then the "80/80
rule" is broken so to speak.
It would seem that a similar experiment or test could be devised
to evaluate lab measurement uncertainty, but to my knowledge, nobody
does this.
The CISPR 16 test does shed some light on what is really expected
by the regulations:
1. Statistically, it is impossible to guarantee that all shipped
production measures below the legal limit because to
guarantee that, 100% sampling would be required. Cost
to do this would be prohibitive.
2. If 80% production can be predicted to pass the limit with
an 80% confidence level, that is considered reasonable
for practical considerations.
3. You would need to do such a statistical test at regular
intervals to ensure that product from your production
process meets the 80/80 rule on an ongoing basis.
The factor, k, can be changed to yield other confidence and % pass
levels by referring to an applicable t-distribution table. Therefore
you could fine-tune the test on your production to meet customer
specifications or your own internal corporate spec.
Finally, the VCCI in Japan uses the above test for their market sampling
test program in cases where a one-off production sample exceeds
emissions limits.
Good luck in getting a lab to statistically define and document its
measurement uncertainty! Let us know if you find someone that has
done it.
Regards,
tony_fredriks...@netpower.com
----------
From: Michael_Barge
To: 'emc-p...@ieee.org'
Subject: Measurement Uncertainty
List-Post: emc-pstc@listserv.ieee.org
Date: Friday, January 10, 1997 3:33PM
FROM: michael_ba...@atk.com
Item Subject: Measurement Uncertainty
Greeting Tregers;
There seems to be a requirement that, when giving a measured value, there
must be an uncertainty associated with that value describing the confidence
of that value.
(1) Do most labs report an uncertainty measurement in the test report,
on
the data sheet, on a certificate of compliance?
(2) How did you generate the measurement of uncertainty for emission
tests? For immunity tests?
AND MOST IMPORTANTLY
(3) What do you tell the customer when he is below the limit by less
than
the measurement uncertainty? When he is above by less?
J Michael Barge
Alliant Techsystems
Annapolis, MD
