Ultralite AVB

[fritz_distran]

Hi everyone,

we've got an UltraLite AVB as part of a test setup connected to a computer running Ubuntu 20 (ALSA driver version k5.4.0-39-generic, library version 1.2.2) and are observing the same channel hopping and signal modulation issues as described in this thread. When playing a stereo signal at 192 kHz, roughly every 10 seconds the channels switch (order: 1/2 -> 17/18 -> 9/10 -> 1/2, etc).

Sometimes, more than one channel pair is active even though only a single stereo signal is played:



And that seems to be exactly the times when apparently a modulation with a harmonic signal with a fundamental frequency of 8 kHz occurs. For example, when I play a sweep signal, I see multiple copies of the sweep every 8 kHz, including aliased versions. This happens about 30% of the time and at least once every 10 second period:



Did anyone find out what exactly causes this problem? Is it an ALSA bug or an issue with the UltraLite AVB? I've been in touch with MOTU and got the somewhat weird explanation that they "only support class compliancy on Mac and Windows" (as if the USB standard was different for Windows, MacOS, and Linux ). The UltraLite AVB page on the MOTU website (as of today) states something different though: "The UltraLite AVB connects to your computer with audio class compliant hi-speed USB 2.0 (compatible with USB 3.0 and iOS) for across-the-board compatibility now and in the future."

An issue we did not investigate yet is USB cable length (we're currently using a 5m cable). The longer the cable, the more likely are data transmission issues... Did anyone here play around with this yet?

Best regards,
Fritz

[bluebell]

It's a change in the AVB's firmware that's not handled correctly by the Linux kernel (implicit feedback point). Drumfix made patch, I think you find it earlier in the thread. The patch doesn't fix everything. You may encounter a modulo-8 output channel hopping after some hours but that can be worked around when you connect system:playback_1,9,17 to the left channel and 2,10,18 to the right channel.

If cou can't compile a kernel then go back to firmware 1.3.2+520 or 1.3.4+558.

There will be still issues when you use jack_bufsize to change jackd's buffer size on the fly (input channel hopping) but you can work.

Maybe Drumfix will find the time to fix that hopping issue some day.

[fritz_distran]

Hi bluebell,

thanks for the explanations.

We've installed the patch by Drumfix from the post on Mon May 04, 2020 12:24 am. However, now when connecting the UltraLite AVB, the computer crashed badly (not even the num lock light on the keyboard reacted anymore). Then I saw the patch posted on Fri May 08, 2020 9:15 pm ("one more try") but that one (and also both of them together) just lead to a crash a few seconds after the UltraLite AVB is plugged in.

Typically the last log entries before the crash are things like:

Code: Select all

systemd[992]: Reached target Sound Card.
systemd[1505]: Reached target Sound Card.
systemd-udevd[1674]: controlC1: Process '/usr/sbin/alsactl -E HOME=/run/alsa -E XDG_RUNTIME_DIR=/run/alsa/runtime restore 1' failed with exit code 99.

Sometimes it's just the "Reached target Sound Card".

First of all, which patch is the right one? Or do I need to apply both of them (even though they have some overlap, meaning that some manual merging needs to be done)?

Second, what is the correct way to get all the configurations right? For now we just copied over a .config file from another Linux source directory, which at least produces a bootable kernel on this particular machine (the default settings apparently don't do that).

Best regards,
Fritz

[bluebell]

Yeah, the first version led to crashes. I attched the patch that works with linux-5.7-rc4.

To configure it after the patch just type
make oldconfig

That takes the options of your running kernel. Answer all questions with ENTER.

make
sudo make modules_install install

If you are short on disk space in the partition where your kernel and initramdisk reside then type
sudo make INSTALL_MOD_STRIP=1 modules_install install

[Curly-Mo]

Would anyone be able to share the 1.3.4+558 firmware?

I've been following this thread and fighting with my Ultralite AVB for over a year now.

I haven't yet been able to successfully get the patches to work on Archlinux.

I've tried some older firmware versions, but still get the channel hopping issue. Sounds like some people have had luck with 1.3.4+558, but I can't find that version anywhere.

Thanks!

[root2]

Would anyone be able to share the 1.3.4+558 firmware?

I've been following this thread and fighting with my Ultralite AVB for over a year now.

I haven't yet been able to successfully get the patches to work on Archlinux.

I've tried some older firmware versions, but still get the channel hopping issue. Sounds like some people have had luck with 1.3.4+558, but I can't find that version anywhere.

Thanks!

I am running the latest firmware with kernel patches and it's working fine for weeks now!
I attached my PKGBUILD, just download it and run the following commands:

Code: Select all

unzip linux-motu.zip
makepkg -si

If you have any problems building the kernel, feel free to ask

[fritz_distran]

Yeah, the first version led to crashes. I attched the patch that works with linux-5.7-rc4.

To configure it after the patch just type
make oldconfig

That takes the options of your running kernel. Answer all questions with ENTER.

make
sudo make modules_install install

If you are short on disk space in the partition where your kernel and initramdisk reside then type
sudo make INSTALL_MOD_STRIP=1 modules_install install

Hi bluebell,

thank you very much for the detailed explanation. This definitely improved the situation a lot. I could compile the 5.7-rc4 kernel with the new patch and it works more often and for longer periods than before.

However, I do still get channel shifts, and now they are not necessarily multiples of 8 anymore (I've had shifts by 15, 20 or 23 channels already). Also, before I had mostly modulation type artifacts with some additional noise (see spectrogram in my first post here), but now I also have that additional noise alone.

In any case, the behavior seems to depend on the buffer size and the number of channels I use. As a tendency I'd say there are less artifacts when using large buffers and ALSA devices with less channels (e.g. "Front speakers" or "IEC958 (S/PDIF) Digital Audio Output").

Best regards,
Fritz

P.S.: Is there a reason to use the 5.7-rc4 kernel and not a more recent one?

[bluebell]

P.S.: Is there a reason to use the 5.7-rc4 kernel and not a more recent one?

That version was the recent one when Drumfix wrote the patch and I tested it.

I asked Drumfix to adapt the patch to 5.7.7 but got no response yet.

[fritz_distran]

Just a quick update: with the latest patch and a shorter USB cable it seems to work flawlessly (no channel hopping, no artifacts so far). Just one or the other is not enough (we've previously tried a shorter USB cable without the patch and it didn't solve the issues).

[ahellquist]

Just a quick update: with the latest patch and a shorter USB cable it seems to work flawlessly (no channel hopping, no artifacts so far). Just one or the other is not enough (we've previously tried a shorter USB cable without the patch and it didn't solve the issues).

Hello !

With what firmware is this ?

[fritz_distran]

Hi ahellquist

It's the latest firmware, 1.3.5+626.

Fritz

[ahellquist]

Hi ahellquist

It's the latest firmware, 1.3.5+626.

Fritz

Ok. Great to know.
I will build a patched kernel and upgrade one of my devices to check how it works.

Thanks

/Anders

[bluebell]

Just a quick update: with the latest patch and a shorter USB cable it seems to work flawlessly (no channel hopping, no artifacts so far). Just one or the other is not enough (we've previously tried a shorter USB cable without the patch and it didn't solve the issues).

No slow modulo 8 output channel hopping after some hours of operation (one hop after many hours)?

I think that this hopping occurs from time to time when there is a small error in communication, and it's a Linux bug.

[fritz_distran]

Indeed, after playing sound for many hours, there were some issues (8 kHz modulation artifacts). I could get back to normal operation by stopping and starting the playback though (BTW, changing the buffer size / latency in the settings page of the JUCE DemoRunner application seems to get the communication back to normal, too). In any case it's already much better than without the patch or with the patch and a 5 m USB cable.

[mxxxl]

Dear all,

First of all: Thank you for your work! It was very frustrating to try out my brand new Ultralite AVB ES FW 1.3.5+637 I bought after reading some postive comments about the Ultralite AVB and Linux on the internet. Unfortunately theese postings dated back to 2018 and I did not expect that things would take a turn for the worse with the newer version.

On my Lenovo T430 with Debian Buster bitcrushing and channelhopping started instantly after playback, on a T410 with a similar system it worked for about 10 minutes before going wild.

I just installed the new 5.7.0-rc4 kernel with the patch bluebell posted before. Bitcrushing is gone now, but there is one modulo 8 channel hop almost instantly after starting playback, but then it seems to stay stable 8 channels above...
After about one hour 8khz artefacts though...

Hope that we will find a solution soon,
all the best,
Mathias