long mail! re: GSM audio level and mic sensitivity, echo issue, alsa statefiles

[Joerg Reisenweber]

Am Di  5. Mai 2009 schrieb Joerg Reisenweber:
> Dear distro maintainers and developers
> please recheck 
> http://docs.openmoko.org/trac/ticket/2121#comment:3
> 
> cheers
> jOERG
> 

for your convenience (beware, long IRC log, only partially cleaned to topic)




[2009-05-05 23:35:48] <lindi-> DocScrutinizer: now the volume level is ok but 
people complain about odd distortion :)
[2009-05-06 00:01:19] <DocScrutinizer> lindi-: if distortion happens I 
recommend reducing #48 by one step, increasing #5 accordingly to get same 
level at far end, and see if distortion is away 
[2009-05-06 00:02:50] <DocScrutinizer> lindi-: think of #48 like the record 
level dial for a tape recorder, and #5 the playback volume dial
[2009-05-06 00:04:36] <DocScrutinizer> lindi-: if you set #48 very odd, you 
either get distortion one way, or noise the other way. You need to find 
correct setting for #49 close to clipping (distortion), then adjust actual 
level to your preferences with #5
[2009-05-06 00:05:06] <DocScrutinizer> s/#49/#48/
[2009-05-06 00:05:08] <bzzbot> DocScrutinizer meant: lindi-: if you set #48 
very odd, you either get distortion one way, or noise the other way. You need 
to find correct setting for #48 close to clipping (distortion), then adjust 
actual level to your preferences with #5
[2009-05-06 00:06:39] <lindi-> DocScrutinizer: it seems to be 
non-deterministic
[2009-05-06 00:07:05] <lindi-> DocScrutinizer: it can be heard only on some 
calls, and during those calls it persists during the whole call
[2009-05-06 00:07:55] <DocScrutinizer> ny recommendation is: call to a quite 
place far end, set #5 to a rather low level (~10..15 below usual), then 
adjust #48 to get low but clear sound (no distortion) when *shouting* at the 
device. Then level up #5 to get good volume level when speaking gently to 
device
[2009-05-06 00:09:15] <DocScrutinizer> try to level up #48 bat much as 
possible in procedure mentioned above
[2009-05-06 00:11:08] <lindi-> DocScrutinizer: still i don't understand where 
the non-determinism comes from
[2009-05-06 00:11:14] <ptitjes> mickey|tv: but it seems the same missing 
throws clause also
[2009-05-06 00:11:37] <DocScrutinizer> lindi-: btw - is distortion a worse 
problem than people not hearing you?
[2009-05-06 00:12:12] <lindi-> DocScrutinizer: two out of two people say yes
[2009-05-06 00:12:20] <mickey|tv> k
[2009-05-06 00:12:24] <DocScrutinizer> lindi-: there's plenty of sources for 
non-determinism in GSM
[2009-05-06 00:13:34] <DocScrutinizer> lindi-: (2 of 2) That would bother me, 
honestly ;-) If people prefer to not hear you at all, rather than get some 
distortion
[2009-05-06 00:13:51] <lindi-> DocScrutinizer: remember voice distortion 
caused by AT%CBHZ=1 ?
[2009-05-06 00:13:52] <DocScrutinizer> XD
[2009-05-06 00:14:04] <DocScrutinizer> yup, cloudy
[2009-05-06 00:14:23] <lindi-> I've got recorded calls now :)
[2009-05-06 00:14:45] <lindi-> DocScrutinizer: so I'm somewhat suspecting 
something similar
[2009-05-06 00:14:48] <DocScrutinizer> lindi-: I closed your ticket for that 
behalf
[2009-05-06 00:15:04] <lindi-> DocScrutinizer: the one about -a7.state?
[2009-05-06 00:18:32] <DocScrutinizer> lindi-: 
http://docs.openmoko.org/trac/ticket/2282#comment:1
[2009-05-06 00:19:26] <lindi-> DocScrutinizer: yep that is all fine
[2009-05-06 00:20:29] <DocScrutinizer> lindi-: so publish your new results and 
we'll see how to tackle this
[2009-05-06 00:21:00] <DocScrutinizer> lindi-: either change topic of #2282 or 
open a new one
[2009-05-06 00:21:48] <DocScrutinizer> muahaha, *bad* for python
[2009-05-06 00:21:53] <lindi-> DocScrutinizer: yep i'll open a new one
[2009-05-06 00:22:00] <lindi-> DocScrutinizer: the non-determinism scares
[2009-05-06 00:22:02] <lindi-> DocScrutinizer: the non-determinism scares me
[2009-05-06 00:22:05] <mickey|tv> s/python/patch/
[2009-05-06 00:22:19] <DocScrutinizer> lindi-: me too, sometimes 
[2009-05-06 00:22:21] <lindi-> DocScrutinizer: i'm thinking about setting up 
automated daily 60-second test calls that get recorded :)
[2009-05-06 00:22:22] <DocScrutinizer> :-o
[2009-05-06 00:22:38] <lindi-> since my voip provider gives me 3000 minutes 
free per month
[2009-05-06 00:22:48] <DocScrutinizer> duh
[2009-05-06 00:22:56] <DocScrutinizer> which one???
[2009-05-06 00:22:59] <lindi-> (if I call their GSM numbers of course)
[2009-05-06 00:23:05] <DocScrutinizer> aaaah
[2009-05-06 00:23:11] <DocScrutinizer> so nebermind
[2009-05-06 00:23:22] <DocScrutinizer> never even
[2009-05-06 00:24:19] <DocScrutinizer> hehe
[2009-05-06 00:25:49] <lindi-> DocScrutinizer: is there any docs on what 
happens inside calypso vrt. audio and how to control it?
[2009-05-06 00:26:52] <DocScrutinizer> honestly not even linebreak should make 
any diff
[2009-05-06 00:27:22] <DocScrutinizer> lindi-: sparse and poor
[2009-05-06 00:27:24] <DocScrutinizer> :-/
[2009-05-06 00:28:36] <DocScrutinizer> mickey|tv: no intetion to argue. U 
know: real hackers can code FORTRAN in *any* language ;)
[2009-05-06 00:29:32] <mickey|tv> yo
[2009-05-06 00:29:49] <DocScrutinizer> 
http://www.phttp://www.pbm.com/~lindahl/real.programmers.htmlbm.com/~lindahl/real.programmers.html
[2009-05-06 00:31:46] <DocScrutinizer> lindi-: nevertheless you should contact 
alphaone|gone for the "nondeterministic" topic, it seems there's something 
going on that needs investigation
[2009-05-06 00:32:56] <DocScrutinizer> lindi-: alphaone|gone thought it's 
related to moko-FW version, but this theory seems to not pan out as well
[2009-05-06 00:33:41] <lindi-> DocScrutinizer: hmm
[2009-05-06 00:33:50] <DocScrutinizer> neverheless *something* obviously is 
odd there
[2009-05-06 00:35:04] <DocScrutinizer> maybe calypso ALC is locking up on 
extreme conditions or sth. No bright idea what might be the root cause, but 
its too much complaints same general direction to simply ignore them
[2009-05-06 00:35:12] <lindi-> DocScrutinizer: check 
http://iki.fi/lindi/openmoko/call-quality/recordings/
[2009-05-06 00:35:25] <lindi-> DocScrutinizer: openmoko-correct-state2.wav 
shows the problem, openmoko-correct-state3.wav and 
openmoko-correct-state4.wav don't
[2009-05-06 00:35:51] <DocScrutinizer> lindi-: check your permissions ;-)
[2009-05-06 00:36:31] <DocScrutinizer> bbiab
[2009-05-06 00:37:00] <lindi-> DocScrutinizer: should work now
[2009-05-06 00:38:08] <lindi-> DocScrutinizer: how about restarting wolfson in 
the middle of the call just to make sure it is not causing this?
[2009-05-06 00:38:48] <DocScrutinizer> WTF?! sounds like backwards
[2009-05-06 00:39:01] <lindi-> DocScrutinizer: do we know that wolfson is 
innocent? ;)
[2009-05-06 00:39:20] <DocScrutinizer> wolfson isn't mangeling audio either 
way whatsoever
[2009-05-06 00:39:31] <DocScrutinizer> yup
[2009-05-06 00:39:35] <DocScrutinizer> we know
[2009-05-06 00:39:47] <DocScrutinizer> if you set #63 to mic2
[2009-05-06 00:39:49] <lindi-> DocScrutinizer: but i thought those gain 
controls in alsa were part of wolfson?
[2009-05-06 00:40:01] <DocScrutinizer> they are
[2009-05-06 00:40:20] <DocScrutinizer> but that can't cause the mess I heard 
10sec ago ;-)
[2009-05-06 00:41:20] <DocScrutinizer> I'm not trained to listen to finish(?) 
but THAT sounds RATHER STRAAAANGE
[2009-05-06 00:42:13] <DocScrutinizer> never heard such a mess of echo back 
and forth and background noise even with my worst RTP-SIP-commections
[2009-05-06 00:42:36] <lindi-> DocScrutinizer: hmm, was there really echo?
[2009-05-06 00:42:48] <lindi-> DocScrutinizer: mostly just everything I say 
gets clipped
[2009-05-06 00:43:04] <DocScrutinizer> half of the sound feels like playback 
backwards echo to me
[2009-05-06 00:43:32] <lindi-> DocScrutinizer: are those other recordings ok 
by your standards?
[2009-05-06 00:44:08] <DocScrutinizer> yeah I noticed some clipping. but thats 
not really bad and just indicates we got reserve in amplification and you 
should probably go down on #48 from 3 to 2
[2009-05-06 00:44:37] <DocScrutinizer> (other) was so buffled I hadn't even a 
test on them
[2009-05-06 00:45:17] <lindi-> DocScrutinizer: hmm, what do you mean?
[2009-05-06 00:45:40] <lindi-> DocScrutinizer: (re the other recordings)
[2009-05-06 00:45:42] <DocScrutinizer> lindi-: (3 to 2) though you should 
check first if it gets better on adjusting #5
[2009-05-06 00:46:02] <DocScrutinizer> a minute please
[2009-05-06 00:46:21] <lindi-> i'll be away for 45 minutes or so now
[2009-05-06 00:49:01] <DocScrutinizer> lindi-: ok, it's pretty clear now those 
recordings have a strong slow dynamic compression+noisegate which makes the 
start of a word ramp up in volume - this leads to other non-finish feeling 
like it's backwards
[2009-05-06 00:49:19] <DocScrutinizer> the second recording also has a little 
clipping
[2009-05-06 00:49:45] <DocScrutinizer> you should do as adviced: go down #48 1 
step, readjust #5
[2009-05-06 00:50:14] <DocScrutinizer> and you should try other AT%Nxxxx 
values, urgently!
[2009-05-06 00:52:57] <DocScrutinizer> I never was aware ther could be such a 
bad compression/noisegate (such a slow attack time) anywhere in our device. I 
wonder if it might be introduced by some other ghost cooking the audio in 
amoniak
[2009-05-06 00:54:23] <DocScrutinizer> to realy rely on these recordings we 
had to verify all the involved components to ensure what is Freerunner and 
what is recording equipment
[2009-05-06 00:55:40] <DocScrutinizer> anyway, I'm rather scared now, as I 
never before heard such effects on OM devices
[2009-05-06 00:57:54] <DocScrutinizer> a standard "one two, one two, popp! 
eeeek, pop, one two three" probably would help a lot to take out oddity of 
finish language which nobody except lindi- knows how it should sound in 
original.
[2009-05-06 00:58:30] <DocScrutinizer> an plain recording done by laptop 
microphone and e.g. krecord would also help a lot, to compare
[2009-05-06 01:02:05] <DocScrutinizer> basics of noisegate: low volumes get 
even lower, high volumes are boosted. the ramp up and ramp down is a 
timeconstant that's much too long in the samples I heard, also the noisegate 
effect is too strong (too much boost for loud words, low words a too much 
attenuated)
[2009-05-06 01:04:05] <DocScrutinizer> now I suggest to c&p all this mumbling 
of mine of he last 30min to some appropriate place in wiki or where-ever, and 
I'll go to watch TV for a while
[2009-05-06 01:05:40] <tmzt> how much of this is caused by using an anolg link 
from baseband to wm codec?
[2009-05-06 01:05:50] <tmzt> analog
[2009-05-06 01:06:06] <DocScrutinizer> there's no codec in wm
[2009-05-06 01:06:32] <DocScrutinizer> wm8753 is a mere audio mixer for 
calypso
[2009-05-06 01:06:49] <tmzt> it doesn't have one or your not using it in this 
design?
[2009-05-06 01:07:13] <DocScrutinizer> we are not using it, and it doesn't 
have a GSM codec
[2009-05-06 01:07:24] <DocScrutinizer> it has a PCM codec
[2009-05-06 01:07:28] <tmzt> pcm codec
[2009-05-06 01:07:29] <DocScrutinizer> aka ADC
[2009-05-06 01:07:32] <tmzt> right
[2009-05-06 01:07:50] <tmzt> many phones use ssp or other pcm link
[2009-05-06 01:07:54] <DocScrutinizer> it's not used for gsmhandset.state
[2009-05-06 01:08:03] <tmzt> between baseband and codec
[2009-05-06 01:08:06] <DocScrutinizer> we use analog audio
[2009-05-06 01:08:24] <DocScrutinizer> our baseband does PCM as well
[2009-05-06 01:08:37] <DocScrutinizer> see schematics
[2009-05-06 01:08:40] <tmzt> which is what I'm asking, is this a major reason 
for these problems?
[2009-05-06 01:08:48] <DocScrutinizer> nope
[2009-05-06 01:08:57] <tmzt> ok
[2009-05-06 01:09:57] <DocScrutinizer> calypso is the major problem (heard 
this before? ;D )
[2009-05-06 01:10:41] <DocScrutinizer> ahh sorry. calypso == baseband
[2009-05-06 01:12:00] <DocScrutinizer> (for the 119 who might wonder what I'm 
talking about)
[2009-05-06 01:12:08] <tmzt> how can volume in bp affect pcm though?
[2009-05-06 01:12:23] <DocScrutinizer> err, please rephrase
[2009-05-06 01:12:32] <DocScrutinizer> elaborate
[2009-05-06 01:12:43] <tmzt> the AT commands for setting volume
[2009-05-06 01:12:48] <DocScrutinizer> yup
[2009-05-06 01:13:10] <DocScrutinizer> these are for playback (earpiece) 
mainly
[2009-05-06 01:13:10] <tmzt> since part of the solution seems to set that to 
yhe highest setting
[2009-05-06 01:13:43] <DocScrutinizer> yup, as we got too low volume on 
earpiece sometimes aka on some devices
[2009-05-06 01:14:19] <tmzt> earpiece isn't connected directly to bp though?
[2009-05-06 01:14:30] <DocScrutinizer> so we need decent level on WM8753 
*input* so we can juggle with mixer settings
[2009-05-06 01:14:47] <DocScrutinizer> nope, is thru WM8753 as well as mic
[2009-05-06 01:15:59] <DocScrutinizer> 
http://wiki.openmoko.org/wiki/Neo_1973_audio_subsystem#using_phone-internal_microphone_and_speaker
[2009-05-06 01:16:07] <DocScrutinizer> tho this is incorrect
[2009-05-06 01:16:51] <tmzt> I'm asking because the gta03 list said mc71 would 
work the same way, with analog rather than pcm routing
[2009-05-06 01:17:04] <DocScrutinizer> this is correct one
[2009-05-06 01:17:07] <DocScrutinizer> 
http://people.openmoko.org/joerg/ALSA/doc/WM8753_control_diag_gsmhandset_mic_std.png
[2009-05-06 01:17:28] <DocScrutinizer> mc75i, yup correct
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part.
Url : http://lists.openmoko.org/pipermail/devel/attachments/20090506/2b73832b/attachment.pgp