"Bob Cain" wrote in message
...


Arny Krueger wrote:


This issue relates to something as basic as the assumption of how much
the
bus output increases as operating channels are added. If you presume
uncorrelated signals then its 3 dB for the first channel added. If you
presume correlated signals then its 6 dB for the first channel added and
proportionately more per successive channel that is added. It's the
difference between a geometric sum (square root of sum of squares)
versus
and sum in regular linear arithmetic.

Arny, doesn't this apply to RMS levels rather than peak
levels?

Good point. I just did a simulation with 60 seconds of 44/32 high density
white noise in CEP, and two independent waves summed to a wave whose average
was 3 dB higher, but the peak was 5.4 dB higher.

I obtained the same results for 10 seconds of noise except that the peak was
5 dB higher. Note, both numbers are less than theory, but show results that
are probably asymptotic to theory or 6 dB.

I guess that means that if you wait long enough, sooner or later you'll get
all the peaks to come together.

If one is concerned about peak levels then it would
seem to me that the 6 dB criterion would apply independently
of correlation considerations.
simpler."

That appears to be the case. Thanks for the memory jog/correction. Hey, I
knew all this stuff cold 30 years ago...

;-)

However, the simulation seems to show you might wait a while for that one
big peak.

I think this explains while most recordings have infrequent peaks that are
pretty wild compared to the rest of the file. Everything lines up at those
rare instants.