Long USB cable causing measurable differences in impulse response

[Keith_W]

I was helping someone with REW measurements in another forum when I came across something I have never seen before.

Here are five measurement's of a left speaker showing exactly the same frequency response.

And here are the same five measurements showing different impulse responses. I was absolutely baffled by this, how can the FR be the same but the impulse responses be different!?!? I had to ask John Mulcahy for help. He noted:

The culprit turned out to be a long USB cable. Sure enough, when the OP swapped his USB cable for a shorter one, the problem went away.

Here is the evidence, from his new set of measurements.

Even then, I am still baffled. The OP did not state if the long USB cable was to the microphone (UMIK-1) or to the DAC. The UMIK-1 has the ADC is in the microphone. How can a long USB cable affect the measurement, when the measurement has already been digitized? And conversely, how would a long USB cable to the DAC change the IR but not the frequency response? I'm not questioning the phenomenon (clearly it exists), but I do not understand why it happened. Any opinions?

 

[xanalog]

And conversely, how would a long USB cable to the DAC change the IR but not the frequency response?

Could it be that IR response REW is measuring is greatly above the audio frequency (>1MHz?), thus not impacting our old-n-trusted 20Hz-20kHz FR?

 

[Devnull]

Long cable length and marginal termination impedance might cause an increase in the BER.

I guess the question for John should be is how is the impulse response calculated?

 

[sam_adams]

I was helping someone with REW measurements in another forum

You are not going to be able to help someone that does not understand how to use the software. Half of measurements that he posted have no timing reference. In the case where there is an acoustic reference used, some have the right speaker as as the acoustic reference and some have the left speaker as an acoustic reference. You cannot compare the timing of the measurements if there is no consistent timing reference.

Another thing is that his measurements are band-limited. He's measuring from 35 Hz to 22.059 KHZ. His measurements need to be from at least 10 HZ—better to go to 2 Hz but the default MiniDSP .cal files don't correct below 10 Hz so the correction defaults to 0dB below that point—to half the sample rate, 24 KHz.

He's also using a Windows PC. Often, Intel-based computers will have multiple USB base hub devices in them that have the ports on the left side of the computer on one hub and the ports on the right side of the computer on another hub. This can cause slight differences in clock timing—on the order of a few µs—between devices connected on one hub and devices connected on the other. Uwe Seiber's USB Device Tree Viewer is an invaluable piece of software to determine the layout of USB devices on Windows PCs. Unfortunately, there is nothing equivalent for Mac users.

Until that person has the discipline to develop a setup and workflow when using REW, there is no way that you will be able to solve his self-inflicted problems.

 

[Keith_W]

You are not going to be able to help someone that does not understand how to use the software. Half of measurements that he posted have no timing reference. In the case where there is an acoustic reference used, some have the right speaker as as the acoustic reference and some have the left speaker as an acoustic reference. You cannot compare the timing of the measurements if there is no consistent timing reference.

Don't be so hard on him, he's learning very quickly. That's more than can be said for some other people i've tried to help. All his latest measurements have a timing reference. And FWIW - whether there is a timing reference or not does not change the shape of the impulse response, only the position. That is what is under discussion here.

 

[DWPress]

I believe that USB cables have a limit of around 10' (not sure which USB spec they use) for signal integrity, they sell longer "booster" USB cables that have active electronics. I ran into this problem with my Umik 1, but also with my laser cutter 16' from my computer.

 

[NTK]

I did an experiment. REW computes the impulse response from the transfer function which is obtained from sine sweep. The transfer function is calculated from the FFT of the mic measurement divided by the FFT of the sine sweep. However, REW doesn't measure the sine sweep (the test signal to the speaker). It uses the theoretically calculated numbers of the sweep (which are also sent to the DAC to generate the signal).

I simulated what would happen if some of the samples coming back from the mic are lost. I created a sine sweep, and filtered it with a high pass and low pass filter. The blue curves are for no missing sample. For the red curves, I removed 10 samples when the sweep was during the high frequency roll off.

 

[Keith_W]

I simulated what would happen if some of the samples coming back from the mic are lost. I created a sine sweep, and filtered it with a high pass and low pass filter. The blue curves are for no missing sample. For the red curves, I removed 10 samples when the sweep was during the high frequency roll off.

That shows what we see then. The frequency response does not change, but the shape of the impulse response changes. Very interesting.

 

[amirm]

The frequency response does not change,

Frequency response does change. Removed samples create high frequency energy which shows up in that impulse but not the band limited frequency response.

 

[Sokel]

If the cable is longer than the specified standard of USB 2 or not certified any error becomes insignificant as we're talking about something that is not suppose to work this way.
People complain about that usually at the edge of the use, like using interfaces with them which traditionally consume more energy or trying to play high-rate DSD, etc.

 

[KSTR]

Even then, I am still baffled. The OP did not state if the long USB cable was to the microphone (UMIK-1) or to the DAC. The UMIK-1 has the ADC is in the microphone. How can a long USB cable affect the impulse response, when the impulse response has already been digitized? And conversely, how would a long USB cable to the DAC change the IR but not the frequency response? I'm not questioning the phenomenon (clearly it exists), but I do not understand why it happened. Any opinions?

An USB cable actually can change audio stream data (which has no error correction or packet resend strategies) but I've never had a single case where the corruption was stable with just a few changed/dropped samples now and then. Either the cable works or doesn't. I've tried really hard to experimentally modify a cable electrically to land exactly on the trip point where it still works for like 99.9% of the data and only corrupts 0.1%, in a digital loopback scenario. I failed, and from the technical POV it is clear that the transition region between go and no-go is extremely narrow. Exceeding the specified length (which I did not examine) might be one corner case where the transition region might be a bit broader, though. Simple adding a few cm of extension, or using another USB socket on the PC, or inserting a hub, things like that, would still have major impact and would bias the pass/fail rate strongly to one direction. This is what I would suggest to try. If the symptoms remain the same, transmission errors from the cable are not the issue.

I'm inclined to think we are seeing a secondary effect but I have no clue what exactly, maybe stuff like power supply issues from the long cable making the bus-powered DAC or mic operate on the edge.