>I also cannot place any trust in figures that conclude that the units get even
>worse than the actual units tested from production. i.e. the worse production
>unit was 43 dB and the calculations come up with 43.41 dB.
This is fundamental to statistical analysis. If the units do not come
off the production line with uniform characteristics, why should we
assume that the 10 samples we pick from a 1000-size lot include the
worst ? We can't ... we must assume that they actually follow a bell-
curve, and the computation of mean and standard deviation represents
our attempt at fitting the samples to the bell curve template in order
to determine the shape of that bell curve. Having selected a sample of
8, and fitted a distribution that that, you may find that after drawing
2 more samples and finding that they have BETTER data than any of the
previous samples, your standard deviation increased enough that the
projected worst case got worse, maybe enough to make you fail where
you passed before.
Here's an example:
Assume legal limit 50 dB with 6dB margin.
Samples 1-6 43 dB
samples 7-8 41 dB
samples 9-10 34 dB
The first 8 samples yield a mean of 42.25, standard eviation 1.035
and a 95% confidence interval of 41.53-43.29, i.e. 6.71 dB margin.
Adding the last 2 samples moves the mean to 40.60, but the standard
deviation jumps to 3.60, yielding a 95% confidence interval of
38.37-44.20, i.e. the margin dropped to 5.80 dB.
/ Lars Poulsen [email protected] +1-805-562-3158
OSICOM Technologies (Internet Business Unit) (formerly RNS)
7402 Hollister Avenue Manager of Remote Access Engineering
Goleta, CA 93117 Internets designed while you wait
