Latency of USB-Interface of SQ?

[Tobi]

Hi,
I had the chance for a test-run of SQ5 at my local dealer. I streamed 30 channels from my laptop via the integrated USB-Interface and was wondering about the latency. I needed to set a buffer size of 1024 to get a playback without drop-out. That is more than twice the expected buffer size from experience with my RME USB Interface.

Can you give some experiences what is possible?

Thanks and Best Regards,
Tobi

[DanTarbill]

Giving this topic a bump because I’d like to know the answer too!

[Tobi]

seems, we are the only ones :-/?

[Anonymous]

No you wont be the only ones watching this column

[Wes]

Hey Guys:
I’ve been meaning to reply on this thread but haven’t been out to the forum in a while. I have an SQ-5 that I use as an interface to my DAW (Reaper and/or Sonar Pro). I did initially have quite a few problems with some “bit crusher” type artifacts and drop outs but the latest firmware has fixed all that. My buffer is set at 16 samples and is rock solid. My typical latency is 6 msec. I wasn’t particularly scientific about my measurement…recording a single track into the DAW with the output coming back to the SQ and patching to another input recording a second track.The delay between to two recorded tracks was roughly 6 msec. As a disclaimer let me say that my DAW machine is a monster (8 core CPU with 32g RAM and an SSD). I’m doing a lot of drum tracking and latency just hasn’t been an issue. Your results may vary!

[Tobi]

Hi Wes,
thanks for your answer. I am a little bit confused:

When using 96kHz Sampling Rate, then I get:

1/96000Hz = 0,01ms
0,01ms * 16 Samples = 0,16ms Delay

So I am confused: On 16 Samples you should have 0,16ms Delay — why is it 6ms?

Thanks and Best Regards,
Tobias

[SteffenR]

@Tobi you only calculated the buffer delay, you totally missed the A/D conversation, the latency introduced to the desk intself, the latency from the USB bus and the latency of the DAW system

6ms is very good for such a system

[Tobi]

Thanks — thats right. Steffen, do you also use the SQ USB interface? Maybe you have a more “normal” computer and can give me a valid buffer size for those? 16 Samples is very low, I dont get that even with RME-Cards (Driver does not allow such values).

[SteffenR]

16 samples is the latency of the SQ, without the latency for the USB inteface and the PCI buffers and so on…

RME interfaces have a FPGA for internal mixing, matrix switching and processing
it depends on what generation but these have comparable latency times…

the problem here is always the computer system and it’s interface, it will always add some ms

[Tobi]

Back to my first question: on a short test at my dealer SQ USB Interface would only run at 1024 Samples Buffer Size without dropout?

🙂

[peterlanders]

The issue isn’t the SQ having 1024 samples of latency, it’s the computer it was connected to not being able to keep up with the SQ’s data stream with anything less than a 1024 sample buffer. CPU, hard disk or SSD, USB chipset, how much stuff is running in the background, whether the SQ is alone on the USB bus, the recording software: all those things can contribute. But the SQ is going to deliver its data stream at its own speed, regardless of the computer’s ability to keep up.

[jmole]

Hi, I recently acquired a SQ-5, upgraded the firmware to 1.3.2, and discovered that the latency over USB is terrible. Much worse than most dedicated USB audio interfaces.

This test was done on macOS 10.12.6, using Ableton Live. Delay compensation was turned off, and a sound was played out through the USB interface, and then returned via a Tie-Line back over the USB interface.

from pc —- USB output channel 1 —-> mixer tie-line —-> USB input channel 1 —> to pc.

We would expect the minimum round trip latency to be as follows:
( buffer_size * 2 ) * 1/96kHz + mixer_latency.

However, this is not the case. There’s an extra ~10ms of latency in the system, likely from A&H’s poor USB implementation. I also noticed in the firmware release notes that they fixed some bugs related to dropped samples over USB. My guess is they just increased the buffer size on the hardware side which added a bunch of latency, but I didn’t measure an earlier firmware version, so I can’t say for sure.

Check out the results I attached to this post. RTL was measured at:

32 samples: ~10ms RTL
64 samples: ~10ms RTL
128 samples: ~12ms RTL
256 samples: ~14ms RTL
512 samples: ~20ms RTL

For comparison, I ran this same test with a MOTU 624 thunderbolt interface, and the RTL at 64 samples/96kHz was ~2ms. I’m starting to regret returning the MOTU…

Attachments:

You must be logged in to view attached files.

[jmole]

Just checked again on firmware 1.2.2, and the latency is actually a little worse, by some hundreds of microseconds. Not sure what’s going on here.

I really don’t want to have to buy a SQ Dante, RedNet PCIeR, and thunderbolt PCIe cage, just to use my computer for effects.

The other way to do it would be to buy a low latency audio interface and use some analog patching between the interface and the mixer to pull the channels I want into the computer, and patch the outputs into the mixer.

Both these solutions seem like a huge hack considering the mixer has a USB port and should be more than capable of ~3ms RTL @ a 64 sample buffer with 96kHz sampling. A&H, please get your act together and fix this problem.

[jmole]

I repeated my latency testing process on windows. Same setup as before, but using the ASIO drivers. I found the oddest thing. The sample delay between playback and loopback recording was non-deterministic.

I made several recordings in a row, without changing the ASIO buffer size. Each recording had a different round trip latency.

Again, the signal flow is:

pc -> usb channel 1 OUT -> tie line on SQ -> usb channel 1 IN -> pc

It seems that when playback commences, the driver tries to establish sync with the SQ. This “point of sync” varies from recording to recording. That means, even with delay compensation enabled in Live, you will never be able to get phase coherent recording and playback, because the round-trip latency is non-deterministic from session to session.

What a crock!

[jmole]

Just did more testing on macOS, with this handy little program: https://www.oblique-audio.com/tmp/beta.html

Here’s a table of results:

Note there are 2 samples taken at each buffer setting (32 & 128). One sample was taken via tie-line connection, the other sample was taken /w an XLR connecting input and output together.

So the path for one is:
pc -> USB out -> tie-line -> USB in -> pc
and the other is:
pc -> USB out -> XLR out -> XLR in -> USB in -> pc

The rows with the (-102.4) noise floor show the XLR connection, which adds about 50 samples to the latency for AD/DA conversion.

Anyway, this just goes to show that the best case RTL you will get will never be less than 10 ms on macOS.

[Author]

Posts