[Asterisk-Users] fixed point mec3

[The Traveller]

Hey Mark,

On Sun, Jun 29, 2003 at 17:08:00 -0500, Mark Spencer wrote:

> > I was just testing with MeetMe, which was one of the things where MEC3
> > went wrong for me in the past.  After around 8 channels from an
> > E100P-connected PRI joined the conference, everything became one big
> > chaos of noise.  When enough channels leave, the audio returns back to
> > normal again.  There's definately an improvement, as the previous MEC3
> > failed in about the same way with just 1 channel in the conference, as
> > far as I remember, but that test was a while back and many other
> > things have changed in the meantime.
> 
> Does it always break at precisely the same number of calls, and it is
> immediate?  Does experimenting with STEP_SHIFT and/or MAX_BETA change your
> results?

More specific tests show that it actually starts to fail at only 4
calls, at which point bursts of noise can already be heared.  Add a 5'th
one and the noise will become continuous.  I tried some different values
for "STEP_SHIFT" and "MAX_BETA" and put the results in the table below.
Note that I only tested with a max. of 14 channels in the conference.
The number of channels indicates what amount it took to start producing
audible noise, unless specified otherwise.


  STEP_SHIFT  |  MAX_BETA  |  number of channels in conference
  1           |  1024      |  4
  2           |  1024      |  4
  4           |  1024      |  8
  8           |  1024      |  8
  16          |  1024      |  14, with no noise.

"MAX_BETA" didn't really seem to influence the amount of channels at which
noise became audible, but it did influence the amount and intensity of noise
produced after you exceed the amount of channels at which it becomes
audible, where lowering it means less noise.  With it set to "256" and
"STEP_SHIFT" set to "2", I could get 8 channels in and still distinguish
the audio of the channels from the audio of the noise, while with "MAX_BETA"
set to "1024", the noise will completely fill the conference after adding the
5'th channel.

If you want me to do any more tests with specific values or test possible
fixes, let me know.


   Grtz,

     Oliver