On Mon, Aug 23, 2010 at 04:34:50PM -0400, J. Gordon Rankin wrote:
Part of the problem could be that everything is working synchronously. The reading of the USB drive and the outputting to the USB converter are both happening in sync because the application would work this way.
True. But for the USB side of the story, there is a reserved part of the bandwidth for isochronous (real-time) data which is guaranteed for exacty such scenarios, so bulk data bursts wont't kill the stream.
The question is whether this reserved bandwidth is large enough for the application that causes trouble here, and if the driver does the right thing to let the core know about its needs.
High End Audio is usually nuts... but I have experienced that this type of combination results in poor output even when it works 100%. I have yet had time to figure out why... same reason that FLAC never sounds as good as a WAV file in my experience, though there I can verify that the CPU usage does go up higher than most people and engineers would suspect.
That shouldn't be a problem as long as the application reads its bulk data is chunks which are large enough. As far as I'm concerned, I've never seen such issues on Linux or Mac OS X, unless for very small data blocks, resulting in too tight scheduling. In other words, it might be an issue for any kind of real-time audio processing which requires minimal latency, but not for an audio player which can operate on huge blocks.
Daniel
Daniel Mack wrote:
Hi,
FWIW, I currently have no clue what could be the reason for this issue. I copied the linux-usb mailing list, maybe anyone over there has an idea.
Summary is: Demian is trying to play back an audio file over an USB connected soundcard, and the file itself is also stored on a media connected via USB. The transfer alone seems to be reasonably fast (tested with 'dd'), and the card itself also works fine (tested with a file stored on a different media), but the combination of them both fails. At least for high sample rates - iow, high data throughput.
Could there be anything wrong with the isochronous bandwith reservation?
Thanks, Daniel
On Sun, Aug 22, 2010 at 04:51:55PM -0700, Demian Martin wrote:
Daniel: I'll try to provide all the info you want. I'll need some prompting for some details, please be patient with me.
First the system. I have tested on both an intel D945GSEJT atom based MB and an Alis 3.3 AMD Geode based mb. OS: Linux auraliti-player 2.6.32auraliti-2.0 #1 PREEMPT Fri Jul 30 06:48:14 GMT 2010 i686 GNU/Linux Distro Voyage Linux 0.6.1 It's a dedicated headless 2 channel audio player, sort of a high performance squeezebox. Details here: www.auraliti.com
When I play a file from a network source (connected through nfs) I have no problems with either .wav or .flac.
I don't know how to tell with certainty if the USB devices are connecting through a common or separate root hubs.
USB stick speed- It took 6.2 seconds to pass a 1 minute 19 second 192/24 file to dev/null so USB speed is not a problem.
It seems there is a collision when there is a high speed USB storage device and the USB class 2 audio on the same host port both moving data at the same time.
All of the following were done with the Alix board (Geode 500 MHz processor). The Intel does similar but seems more likely to work.
Under some conditions the system loses connection to the USB audio device: auraliti-player:~# aplay -v -Dplughw:1 /media/usb0/Trio-Mytek8X192ADDA-192k.WAV Playing WAVE '/media/usb0/Trio-Mytek8X192ADDA-192k.WAV' : Signed 24 bit Little Endian in 3bytes, Rate 192000 Hz, Stereo aplay: set_params:1116: Unable to install hw params: ACCESS: RW_INTERLEAVED FORMAT: S24_3LE SUBFORMAT: STD SAMPLE_BITS: 24 FRAME_BITS: 48 CHANNELS: 2 RATE: 192000 PERIOD_TIME: 125000 PERIOD_SIZE: 24000 PERIOD_BYTES: 144000 PERIODS: 4 BUFFER_TIME: 500000 BUFFER_SIZE: 96000 BUFFER_BYTES: 576000 TICK_TIME: 0 auraliti-player:~#
Fixed by disconnecting and reconnecting the device.
Here is what I get playing a 44.1 file, after about 15 seconds:
Playing WAVE '/media/usb0/Piano_44_1.wav' : Signed 24 bit Little Endian in 3bytes, Rate 44100 Hz, Stereo Plug PCM: Linear conversion PCM (S32_LE) Its setup is: stream : PLAYBACK access : RW_INTERLEAVED format : S24_3LE subformat : STD channels : 2 rate : 44100 exact rate : 44100 (44100/1) msbits : 24 buffer_size : 22050 period_size : 5513 period_time : 125011 tstamp_mode : NONE period_step : 1 avail_min : 5513 period_event : 0 start_threshold : 22050 stop_threshold : 22050 silence_threshold: 0 silence_size : 0 boundary : 1445068800 Slave: Hardware PCM card 1 'WaveLink HS SPDIF' device 0 subdevice 0 Its setup is: stream : PLAYBACK access : MMAP_INTERLEAVED format : S32_LE subformat : STD channels : 2 rate : 44100 exact rate : 44100 (44100/1) msbits : 32 buffer_size : 22050 period_size : 5513 period_time : 125011 tstamp_mode : NONE period_step : 1 avail_min : 5513 period_event : 0 start_threshold : 22050 stop_threshold : 22050 silence_threshold: 0 silence_size : 0 boundary : 1445068800 appl_ptr : 0 hw_ptr : 0 underrun!!! (at least 0.015 ms long) Status: state : XRUN trigger_time: 903.806908643 tstamp : 903.807038997 delay : 0 avail : 0 avail_max : 11028 aplay: xrun:1259: xrun: prepare error: File descriptor in bad state auraliti-player:~#
And the system stops playing.
And very quickly on a 176.4 file:
Playing WAVE '/media/usb0/01_Rimsky-Korsakov Dance of the Tumblers.wav' : Signed 24 bit Little Endian in 3bytes, Rate 176400 Hz, Stereo Plug PCM: Linear conversion PCM (S32_LE) Its setup is: stream : PLAYBACK access : RW_INTERLEAVED format : S24_3LE subformat : STD channels : 2 rate : 176400 exact rate : 176400 (176400/1) msbits : 24 buffer_size : 88200 period_size : 22050 period_time : 125000 tstamp_mode : NONE period_step : 1 avail_min : 22050 period_event : 0 start_threshold : 88200 stop_threshold : 88200 silence_threshold: 0 silence_size : 0 boundary : 1445068800 Slave: Hardware PCM card 1 'WaveLink HS SPDIF' device 0 subdevice 0 Its setup is: stream : PLAYBACK access : MMAP_INTERLEAVED format : S32_LE subformat : STD channels : 2 rate : 176400 exact rate : 176400 (176400/1) msbits : 32 buffer_size : 88200 period_size : 22050 period_time : 125000 tstamp_mode : NONE period_step : 1 avail_min : 22050 period_event : 0 start_threshold : 88200 stop_threshold : 88200 silence_threshold: 0 silence_size : 0 boundary : 1445068800 appl_ptr : 0 hw_ptr : 0 underrun!!! (at least 0.010 ms long) Status: state : XRUN trigger_time: 1050.313984810 tstamp : 1050.314067000 delay : 0 avail : 1413 avail_max : 22071 aplay: xrun:1259: xrun: prepare error: File descriptor in bad state auraliti-player:~#
If in interpose a USB 1 hub on the usb storage I get this:
auraliti-player:~# aplay -v -Dplughw:1 /media/usb0/01_Rimsky-Korsakov\ Dance\ of\ the\ Tumblers.wav Playing WAVE '/media/usb0/01_Rimsky-Korsakov Dance of the Tumblers.wav' : Signed 24 bit Little Endian in 3bytes, Rate 176400 Hz, Stereo Plug PCM: Linear conversion PCM (S32_LE) Its setup is: stream : PLAYBACK access : RW_INTERLEAVED format : S24_3LE subformat : STD channels : 2 rate : 176400 exact rate : 176400 (176400/1) msbits : 24 buffer_size : 88200 period_size : 22050 period_time : 125000 tstamp_mode : NONE period_step : 1 avail_min : 22050 period_event : 0 start_threshold : 88200 stop_threshold : 88200 silence_threshold: 0 silence_size : 0 boundary : 1445068800 Slave: Hardware PCM card 1 'WaveLink HS SPDIF' device 0 subdevice 0 Its setup is: stream : PLAYBACK access : MMAP_INTERLEAVED format : S32_LE subformat : STD channels : 2 rate : 176400 exact rate : 176400 (176400/1) msbits : 32 buffer_size : 88200 period_size : 22050 period_time : 125000 tstamp_mode : NONE period_step : 1 avail_min : 22050 period_event : 0 start_threshold : 88200 stop_threshold : 88200 silence_threshold: 0 silence_size : 0 boundary : 1445068800 appl_ptr : 0 hw_ptr : 0 underrun!!! (at least 1.537 ms long) Status: state : XRUN trigger_time: 1219.804905467 tstamp : 1219.820257912 delay : 0 avail : 88220 avail_max : 88220 underrun!!! (at least 13.900 ms long) Status: state : XRUN trigger_time: 1222.763695807 tstamp : 1222.902684537 delay : 0 avail : 66169 avail_max : 88219 underrun!!! (at least 1.279 ms long) Status: state : XRUN trigger_time: 1225.972436940 tstamp : 1225.985216250 delay : 0 avail : 88221 avail_max : 88221 underrun!!! (at least 14.141 ms long)
etc. until I stop the playback.
44.1 playback is fine under these conditions (but then why USB audio class 2?)
Further, using MPD for playback I have discovered that the high bit rate flac files play fine but the same files as wave have the same type of stall/stop problems.
I tried Windows on the Intel platform with similar problems using windows media player but no problems with VLC. However with Windows it is not easy to know what is really going on inside.
Is it possible that the essentially synchronous nature of playing a wave file from USB source to USB Audio device causes a collision since the data is going both ways at essentially the same rate and timing?
Let me know what other info you need.
Thanks for looking at this. -Demian
-----Original Message----- From: Daniel Mack [mailto:daniel@caiaq.de] Sent: Saturday, August 21, 2010 2:27 AM To: Demian Martin Cc: alsa-devel@alsa-project.org Subject: Re: Problem with USB Class 2 Audio Driver
On Fri, Aug 20, 2010 at 05:40:22PM -0700, Demian Martin wrote:
I'm testing an early sample of the Wavelength Wavelink, a USB to S/PDIF adapter that supports 44.1-192 KHz sample rates and uses async usb to talk to ALSA. It works and plays all of the sample rates correctly with the git version of Alsa from 7/30/2010 whenthe files are sourced from the network.
What do you mean by "from the network"? How does your test setup look like?
However if I try to play from a usb source to the usb dac it doesn't work and the whole system gets unstable.
You could also be more precise here :) What doesn't work, how are you testing, and in which regard does the system get unstable?
The platforms I have tested it on seem to have a single USB host interface but with USB 2 there should be enough bandwidth to pass the data from a USB stick to the cpu and back. If I use
an
older 96 only usb dac on the same system it works (unless the down conversion isn't a direct divide, which overloads the CPU, but that is a different issue). The problem seems to be sample rate independent and hits the moment I try to access the file. This is using MPD as a player.
Ah, so your data file is stored on a media which is also connected to USB? Did you connect the two devices to different USB ports or do they share one uplink with a hub?
What additional info do you need to troubleshoot this? Is it an intrinsic limitation to the interface? What additional tests should I do? It's possible it's a hardware issue but how do I divide them so I can go back
to
the hardware guy if it's his issue?
I'm not aware of any limitation, but there could be such issues as exceeded bandwith on the bus and the like. How many audio channels are we talking about?
You could measure the speed of your USB media by using something like this:
$ time dd if=/path/to/192khz.file of=/dev/null
This should take significantly less time than - let's say - half the real-time audio playback time of the file, so there's enough headroom to transport the audio data back to the USB DAC.
What kind of system is this, after all? Did you try other OS on the same hardware for comparsion?
Daniel
-- J. Gordon Rankin Owner and Chief Scientist ====== Wavelength Audio, ltd ====== High-End Audio since 1981 SET Tube Amplifiers, DACS & Preamps http://www.WavelengthAudio.com =================================== Computer USB DACS http://www.USBDacs.com =================================== SET Tube Guitar Amplifiers NAMM Member since 1998 http://www.Guitar-Engines.com =================================== 3703 Petoskey Avenue Cincinnati, Ohio 45227 USA mailto:waudio@cinti.net mailto:wavelength@fuse.net (513) 271-4186 phone/voicemail