Problem: incorrect phase measurements with UMIK-1 (different USB ports)

[onto_log]

Greetings to John and other forum participants.

To begin with, I want to express my deep gratitude to John for creating and supporting REW, the world's best acoustic measurement instrument (in my opinion).

A month ago, I purchased UMIK-1 directly from miniDSP, pleased with the quality of the microphone, a long cable in the kit and two calibration files. I used it for a while to measure room resonances and for this purpose the microphone works fine.
But I wanted more - the correct measurement of the phase of the speaker system with its subsequent correction in rePhase.
For phase measurement, I use the "Use acoustic timing reference" mode with a measurement length of 1M in REW. The mixer in Windows 10 is configured according to the instructions (48/24 output), no other programs work in the background. The microphone stands on a tripod, aimed exactly at the speaker system, a distance of 1.14 m. The room is quite quiet, the background noise level is about -60 dB. All conditions for correct measurements are met.

The audio path is quite simple: a desktop computer, via USB it is connected to the Audiolab M-DAC DAC, then the Genelec G Three active speaker system is connected via the balanced output. The computer has 3 different USB controllers: a built-in Intel chipset (USB 2.0), an ASMedia controller (USB 3.0) soldered on the motherboard, and a PCI-E card on a NEC controller (USB 3.0). All USB ports on all controllers are fully operational (any devices in them work correctly).

But when measuring the phase in REW, I came across the fact that the result very much depends on which USB controller I turn on UMIK-1. This manifests itself in a very different form of the impulse response and in the form of a phase curve. Repeatedly rechecked all software settings, the quality of cables, connections, and more. The difference in measurements appears only when changing the microphone controller, M-DAC does not react in any way to changing the controller.
Here are graphs of the impulse response and phase curve depending on the controller in which I include UMIK-1. For a correct comparison, I manually aligned t = 0.

Intel:

ASMedia:

NEC:

To finally make sure that the problem is not in a specific computer, I installed REW on a Lenovo X1 laptop and connected to it via USB both UMIK-1 and M-DAC. The result is similar to NEC, but still significantly different:

Based on the analysis of the pulse shape, you can reject the option with the ASMedia controller - there the pulse shape and phase curve are frankly incorrect, but the other three options are similar to the truth. The problem is that I can’t determine which of the measurements corresponds to reality - the difference in the length of the phase curve differs by hundreds of degrees, which, when you scan the curve in rePhase and generate the pulse file for the convolver, significantly affects the sound.
And I would like to understand why there is such a difference or what I am doing wrong.

I ask for help in solving the problem, I am enclosing the measurement files.

 

[John Mulcahy]

Looking at the ASMedia IR it appears the choice of controller is affecting the clock rate somehow, which seems bizarre. Try turning on the "Adjust clock with acoustic ref" Analysis option to have REW try and compensate for the variations.

 

[John Mulcahy]

As an aside, the effect will be reduced if you measure with shorter sweeps.

 

[onto_log]

John, you are a genius!
I could not answer earlier - I was busy. I checked the measurements with the “Adjust clock with acoustic ref” option turned on - and it really works, now all the measurements from any USB controller give an identical result.
All my previous measurements turned out to be wrong, but everywhere in different ways. The closest to the correct IR / phase were measurements with a separate card on the NEC controller, but still the difference with the correct reference measurements is significant.
If interested, I can publish the correct graphs and measurements file. Filters built on these measurements also give the right sound.
And I think it is necessary to introduce the situation described by me somewhere in the FAQ - I do not think that I alone encountered such a problem.

 

[John Mulcahy]

I think it would be worth attaching the final results, interesting to see.

 

[onto_log]

I took 10 measurements in a row from one microphone position and then averaged vectorly. All IR were previously aligned with t = 0.

 

[John Mulcahy]

Do you have any of the individual measurements? Curious to know how much the clock adjustment was.

 

[onto_log]

Yes, sure. Two is enough or need more?

 

[John Mulcahy]

That's fine, thanks. Only 12 ppm clock adjustment for those, within the variation one could expect when output and input are on different devices. Quite a strong reflection at 470 us by the way.

 

[onto_log]

Only 12 ppm clock adjustment for those, within the variation one could expect when output and input are on different devices.

I took these measurements when the microphone was plugged into the Intel controller, and the output to a separate card on the NEC controller.
For the final filters, I changed the connections a little - both the microphone and the output turned on to one NEC controller, the sample rate is set to 44.1, measurements in ASIO mode (ASIO4ALL), I myself am sitting in a chair next to the microphone. There are slightly different results - if interested, I can publish them.

The room is very asymmetrical, there is no acoustic preparation - both reflections and resonances are very large. In a few months I plan to move to another room, which will be built for audio.

 

[John Mulcahy]

The UMIK is a 48k device, if you set any other rate resampling will be happening somewhere.

 

[onto_log]

I know and understand that it is better to do the input and output of the same frequency at 48k, but the DAC handles the input stream at 44.1k and 48k differently. All my music library in 44.1k and the impulse file for the convolver are also done at this frequency. By ear, the filters made in 44.1k measurements are for some reason better than those made in 48k. Perhaps it is better to take measurements in 88.2-96k and do upsampling of the original audio files to this frequency?
And there is a difference in the measurements between Java and ASIO in SPL. In Java mode, about 1 dB SPL is lost - should it be?

 

[John Mulcahy]

OK, as long as you know. ASIO4All is affected by the OS volume controls, same as the Java drivers, so the SPL should be the same provided the cal files used for the ASIO4All input entry and the Java input entry are the same.

 

[onto_log]

Ok, thanks for your answers. I checked today with the latest version of beta37 - measurements in Java and Asio mode are still different. Cal file is the same, all measurement modes are identical. Green - Asio, blue - Java.

The result is repeatable with some slight differences. This is normal?

 

[John Mulcahy]

Those look like typical measurement-to-measurement differences seen for acoustic measurements. Looking at the impulse responses there are differences in the early reflections. Any change in the environment, including your body position, will result in small changes in the measurement.