Cynthia Pleach wrote:
> I don't know about all of you, but I am sick of arguing with
> hardware engineering managers about what constitutes
> a passing EMC test and what does not.

Hear! Hear!

> I have seen companies impose a 6dB margin requirement,
> some 5 dB, and some 3dB.
> 
> I have made the recommendation that the minimum here
> should be 3dB.  I thought that this was fair.  Now comes
> the product that has 2.8dB margin.  Being a stick by the
> rules type of person, I listened to the engineer explain
> that it was the lousy PC his card was in not his card.  So
> I suggested that we prove his theory, purchase a new
> machine and check the old machine versus the new
> one and if there is an improvement, that I would let
> the 2.8dB stand.
> 
> Of course the manager of the group tells me that as
> a recommendation 3dB is good, but as a rule it
> is IRRESPONSIBLE.  Thus I end up with the "be
> a b___ch" option of imposing the retest for the .2dB
> or starting a precedence of well if .2 is okay, is .3 etc.
> etc.
> 
> What I would like to do is take a pole.  (must  be election
> year in the USA!!!!!!!!!!!)  I would like all of you to respond
> as to what you feel is appropriate. Then I'll run the
> stats and let you what the results are.  This way I can
> go back to the manager with the number of certification
> experts that responded and what they thought was right.
> 
> He or I  will have a hard time arguing against stats.
> (I am open to the possibility that I am wrong and
> 0 margin is acceptable).
> 
> Just reply with a number and I'll let you know what happens.
> 
> Thanks
> Cynthia
> 
> [email protected]

Cynthia, 

6dB would be comfortable.  
Unfortunately, I have been in a lab that passed at only 0.5dB. 

To throw another angle on the subject, try this on for size. 
The 80/80 rule (80% of the products pass 80% of the time) 
falls under the vail of a Binomial Distribution: 

  1.  Two outcomes during the trials: pass or fail. 
  2.  Probability of success is constant throughout the trials. 
  3.  Number of trials is constant. 
  4.  Each trial is independent. 

Try running a binomial distribution calculation on 
N = 100 things, 
where x = 80 things succeed 
with a p = 0.8 succes rate. 
Cumulative answer here.  In other words, calculate "at most". 

I think you'll be very surprised at the outcome. 

*******************************************************
Doug McKean
[email protected]
-------------------------------------------------------
The comments and opinions stated herein are mine alone,
and do not reflect those of my employer.
-------------------------------------------------------
*******************************************************

Reply via email to