mpv/DOCS/tech/hwac3.txt

mails by A'rpi and Marcus Blomenkamp <Marcus.Blomenkamp@epost.de>
describing how this ac3-passtrough hack work under linux and mplayer...
-----------------------------------------------------------------------
Hi,

> I received the following patch from Steven Brookes <stevenjb@mda.co.uk>.
> He is working on fixing the digital audio output of the dxr3 driver and
> told me he fixed some bugs in mplayer along the way. I don't know shit
> about hwac3 output so all I did was to make sure the patch applied
> against latest cvs.
> This is from his e-mail to me:
>
> "Secondly there is a patch to dec_audio.c and
> ac3-iec958 to fix the -ac hwac3 codec stuff and to use liba52 to sync it.

> Seems to work for everything I've thrown at and maintains sync for all audio
> types through the DXR3."

patch applied (with some comments added and an unwanted change (in software
a52 decoder) removed)

now i understand how this whole hwac3 mess work.
it's very very tricky. it virtually decodes ac3 to LPCM packets, but really
it keeps the original compressed data padded by zeros. this way it's
constant bitrate, and sync is calculated just like for stereo PCM.
(so it bypass LPCM-capable media converters...)

so, every ac3 frame is translated to 6144 byte long tricky LPCM packet.
6144 = 4*(6*256) = 4 * samples_per_ac3_frame = LPCM size of uncompressed ac3
frame.

i wanna know if it works for sblive and other ac3-capable cards too?
(i can't test it, lack of ac3 decoder)

A'rpi / Astral & ESP-team

-----------------------------------------------------------------------
Hi folks.
I spend some time fiddling with ac3 passthrough in mplayer. The
traditional way of setting the output format to AFMT_AC3 was no ideal
solution since not all digital io cards/drivers supported this format or
honoured it to set the spdif non-audio bit. To make it short, it only
worked with oss sblive driver IIRC.

Inspired by alsa's ac3dec program I found an alternative way by
inspecting to which format the alsa device had been set. Suprise: it was
simple 16bit_le 2_channel pcm. So setting the non-audio bit doesn't
necessarily mean the point. The only important thing seems to be
bit-identical output at the correct samplerate. Modern AV-Receivers seem
to be quite tolerant/compatible.

So I changed the output format of hwac3 from

AFMT_AC3 channels=1
   to
AFMT_S16_LE channels=2

and corrected the absolute time calculation. That was all to get it
running for me.

-----------------------------------------------------------------------
Hi there.

Perhaps I can clear up some mystification about AC3 passthrough in 
general and mplayer in special:

To get the external decoder solution working, it must be fed with data 
which is bitidentical to the chunks in the source ac3 file (compressed 
data is very picky about bit errors). Additionally - or better to say 
'historically' - the non-audio bit should be set in the spdif status 
fields to prevent old spdif hardware from reproducing ugly scratchy 
noise. Note: for current decoders (probably those with DTS capability) 
this safety bit isn't needed anymore. At least I can state that for my 
Sherwood RVD-6095RDS. I think it is due to DTS because DTS sound can 
reside on a ordinary AudioCD and an ordinary AudioCD-Player will always 
have it's audio-bit set.

The sample format of the data must be 2channel 16bit (little endian 
IIRC). Samplerates are 48kHz - although my receiver also accepts 
44100Hz. I do not know if this is due to an over-compatability of my 
receiver or if 44100 is also possible in the ac3 specs. For safety's 
sake lets keep this at 48000Hz. AC3 data chunks are inserted into the 
stream every 0x1600 bytes (don't bite me on that, look into 
'ac3-iec958.c': 'ac3_iec958_build_burst').

To come back to the problem: data must be played bit-identically through 
the soundcard at the correct samplerate and should optionally have it's 
non-audio bit set. There are two ways to accomplish this:

1) Some OSS guy invented the format AFMT_AC3. Soundcard drivers 
implementing this format should therefore adjust it's mixers and 
switches to produce the desired output. Unfortunately some soundcard 
drivers do not support this format correctly and most do not even 
support it at all (including ALSA).

2) The alternative approach currently in mplayer CVS is to simply set 
the output format to 48kHz16bitLE and rely on the user to have the 
soundcard mixers adjusted properly.

I do have two soundcards with digital IO facilities (CMI8738 and 
Trident4DWaveNX based) plus the mentioned decoder. I'm currently running 
Linux-2.4.17. Following configurations are happily running here:

1. Trident with ALSA drivers (OSS does not support Hoontech's dig. IO)
2. CMI with ALSA drivers
3. CMI with OSS drivers

For Linux I'd suggest using ALSA because of it's cleaner architecture 
and more consitent user interface. Not to mention that it'll be the 
standard sound support in Linux soon.

For those who want to stick to OSS drivers: The CMI8738 drivers works 
out-of-the-box, if the PCM/Wave mixer is set to 100%.

For ALSA I'd suggest using its OSS emulation. More on that later. 
ALSA-0.9 invented the idea of cards, devices and dubdevices. You can 
reach the digital interface of all supported cards consitently by using 
the device 'hw:x,2' (x counting from 0 is the number of your soundcard). 
So most people would end up at 'hw:0,2'. This device can only be opened 
in sample formats and rates which are directly supported in hardware 
hence no samplerate conversion is done keeping the stream as-is. However 
most consumer soundcards do not support 44kHz so it would definitively 
be a bad idea to use this as your standard device if you wanted to 
listen to some mp3s (most of them are 44kHz due to CD source). Here the 
OSS comes to play again. You can configure which OSS device (/dev/dsp 
and /dev/adsp) uses which ALSA device. So I'd suggest pointing the 
standard '/dev/dsp' to standard 'hw:0,0' which suports mixing and 
samplerate conversion. No further reconfiguration would be needed for 
your sound apps. For movies I'd point '/dev/adsp' to 'hw:0,2' and 
configure mplayer to use adsp instead of dsp. The samplerate constrain 
is no big deal here since movies usually are in 48Khz anyway. The 
configuration in '/etc/modules.conf' is no big deal also:

alias snd-card-0 snd-card-cmipci	# insert your card here
alias snd-card-1 snd-pcm-oss		# load OSS emulation
options snd-pcm-oss snd_dsp_map=0 snd_adsp_map=2	# do the mapping

This works flawlessly in combination with alsa's native 
SysVrc-init-script 'alsasound'. Be sure to disable any distribution 
dependant script (e.g. Mandrake-8.1 has an 'alsa' script which depends 
on ALSA-0.5).

Sorry for you *BSD'lers out there. I have no grasp on sound support there.

HTH Marcus