[App_rpt-users] announcements

[Kevin Custer]

> On 9/8/2013 8:37 PM, Buddy Brannan wrote:
>> Rewrite of a codec? Hmm, I'm intrigued.
> On 9/8/2013 8:47 PM, Bryan D. Boyle wrote:
>
> Should have said rewrite of tx/rx bandpass filtering rules...We talked
> about it a little bit while I was working for him a month ago while on
> the bench.

Actually - you are correct Bryan.....

When Jeff DePolo was first getting into this technology, he discovered 
problems with the receiver input filter.  Jeff is a perfectionist, and 
like me is an audio purist.

There is a brief reference to his work (detailed below) in the link I 
shared:
http://www.wanrepeater.com/xipar_urd_filters.doc

Here is an email Jeff wrote while testing:

/I spent most of today on a hair-pulling excursion to try to find the cause of some really hellacious aliasing distortion through the repeater.  I could give you the long story some other time, but the short version is that I found a bug in the rx frontend FIR code where the history buffer is shifted to add a new sample.  The line:

memmove(x+1, x, nx-1)

should be

memmove(x+1, x, (nx-1)*2)

Gotta move 16-bit ints not bytes.

With that corrected, the aliasing distortion dropped immensely as you might imagine.  Aliasing-related distortion products are now below -40 dBr as compared to as little as -6 dBr previously!

However, I'm still not happy with the audio.  I'm seeing 1-2% THD through Asterisk from dongle Rx port to Tx port.  That led me down yet another path to quantify the performance of the hardware (DMK Engineering dongle) via my Audio Precision test set, analysis software in Windows (had the dongle connected to an XP machine for testing), etc..  For the most part, it passes, with the basic specs coming fairly close to the specs for the CM108.
It still has other issues, but none that would account for the THD that I'm seeing.  There's also a problem somewhere that manifests as excruciatingly high distortion below about 450 Hz when the input amplitude is over maybe -6
dBFS (I haven't dug into that deeply yet, but I'm guessing it's an overflow/underflow problem in 16 bit integer math somewhere).  That's where I'm stopping for the night.

Anyway, thought you'd want to know about the FIR bug if nothing else.

Please keep in mind that I'm NOT trying to find faults in your code.  I have some perfectionist tendencies, and they really come through when it comes to audio.  If you're open to constructive criticism I'll pass along whatever else I find.  If not, just tell me to go away and I will.

				--- Jeff/

Jeff also wrote a followup:

/I re-did all of the FIR filters, found a few other bugs along the way, etc.  There is still a problem in the preemphasis/deemphasis routines that can result in integer overflows under certain conditions, and I have some other things I want to fix, but I'm going to stop for now until I have some more time to spend on it.//
/
Ultimately, Jeff went through everything and got it to a state where he was satisfied.  I'm fairly sure all of these improvements exist in ACID as well as XIPAR, maybe Dude can verify?

Thanks,
Kevin


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.keekles.org/pipermail/app_rpt-users/attachments/20130908/6807965a/attachment.html>