REW - Umik 2 (timing drift)

[sublover]

Hey guys,



I'm new here, my name is Danny and im from the Netherlands.



I really hope someone can help me out because i have no clue anymore



My setup

Nad 399 bluos Dirac full

LS50 meta speakers

Perlisten R8S sub



I always used the Umik - 1 for REW and Dirac

I have read (and still read it, very info dense) the amazing tutorial about REW from Keith. And here i discovered also that Umik 1 has it limitations.



I have noticed when doing multiple measurements from the same speaker with timing acoustic ref on, i never get consistend readings. And i can see this with Impulse response. I have read that John (rew) stated the umik 1 doesnt have an internal clock. And the umik2 does have and that its super consistent.



Today i received the UMIK 2 and i couldnt wait to try I was hoping to see consistent timing.



I didnt see it.. . Its still inconsistent?!

do i need to adjust a setting?

there must be going something wrong on my part? please i hope someone can help me. Thanks in advance

 

[staticV3]

Hi @sublover! Welcome to ASR.

What inconsistencies are you seeing, what is the complete signal chain, and how are you sending audio to/from REW?

 

[sublover]

Hi StaticV3,

Thanks for your reply.

Attachad a mdat file. That will give you some info you need.

Laptop Samsung Galaxy book 3 WIndows 11
USB C to Jack adapter (with dac) RCA to NAD C399 LINE OUT

Inconsistencies in delay. I select acoustic timing ref ,

Delay -0,0183 ms (-6,3 mm, -0,25 in)
using estimated IR delay relative to Acoustic reference played from L with no timing offset
Clock adjustment: -12,1 ppm

Delay -0,0034 ms (-1,2 mm, -0,05 in)
using estimated IR delay relative to Acoustic reference played from L with no timing offset
Clock adjustment: -11,4 ppm


I get reading with big differences, thats why i thought buying umik 2 would solve it.

 

[DVDdoug]

I'm not an REW expert...

There is ALWAYS latency through the computer (related to buffering). I don't know why it's not consistent. I would expect REW to have a way of subtracting it out.

I have read that John (rew) stated the umik 1 doesnt have an internal clock. And the umik2 does have and that its super consistent.

All USB audio devices have a clock/oscillator. Soundcard too.

 

[sublover]

I'm not an REW expert...

There is ALWAYS latency through the computer (related to buffering). I don't know why it's not consistent. I would expect REW to have a way of subtracting it out.


All USB audio devices have a clock/oscillator. Soundcard too.

Hi Doug,

yeah and thats no problem for me. Dirac ads 7ms latency, sub dsp 4ms, bluesoumd 50ms. But its consistent.

if it is not consistent there is no way Dirac can set levels gain delay and phase right.

And if i want to align Left and right speaker with thats also impossible. Because the values are not correct.

 

[staticV3]

Inconsistencies in delay. I select acoustic timing ref ,

Delay -0,0183 ms (-6,3 mm, -0,25 in)
using estimated IR delay relative to Acoustic reference played from L with no timing offset
Clock adjustment: -12,1 ppm

Delay -0,0034 ms (-1,2 mm, -0,05 in)
using estimated IR delay relative to Acoustic reference played from L with no timing offset
Clock adjustment: -11,4 ppm


I get reading with big differences, thats why i thought buying umik 2 would solve it.

I cannot find/see any inconsistencies in your data:


All peaks are aligned to within 50µs.

 

[Keith_W]

One of the differences between UMIK-1/2 is that UMIK-1 relies on the external USB host clock, whilst the UMIK-2 has an internal clock. However, BOTH microphones have clocks that are independent of the DAC. This means that both will suffer from latency jitter, but the UMIK-2 is significantly less prone to this than the UMIK-1.

This is the impulse response of all four of your measurements. I aligned the SPL but not t=0. All your measurements were taken of the left speaker, and using the left speaker as acoustic timing reference. I have to admit that I don't have much faith in REW's acoustic timing measurement. As you can see, the measurements are jittered with the green measurement a significant outlier.

What is happening: you have two devices - the ADC (microphone) and the DAC. Both are on separate clocks. REW has to talk to both devices via WASAPI. When REW wants to take a measurement, it has to tell WASAPI to start both devices. Without getting into a discussion about USB bus scheduling, polling intervals, operating system scheduling, CPU loads, etc., the result is that the devices are started at different times every time you take a measurement.

If you had an XLR microphone and an interface, REW only needs to talk to one device, and it can do it via ASIO which has less latency jitter than WASAPI.

I use an XLR microphone and an interface. Here is a close-up of the left speaker, with the left speaker as the acoustic timing reference. You can see there is virtually no latency jitter.

Your UMIK-2 has a loopback feature. I suggest that you use it to take loopback measurements. That should eliminate latency jitter. I don't own a UMIK-2, so I can't talk you through the process. I should try to borrow one and do some experiments. This is incorrect.

I have read the REW manual a few times to try to figure out how the acoustic timing measurement works. IF the timing chirp was part of the impulse response, then there would be no latency jitter. But for the life of me, I can't figure out why you wouldn't make it part of the impulse response. At the moment, what I THINK is happening is that REW starts the ADC/DAC when it takes the timing chirp, and it does it again when it measures the impulse response. Why not simply fix the timing chirp a known quantity (say 700ms and not "about 700ms" in the manual) and start the ADC/DAC only one? That would eliminate all problems with latency jitter. Because I don't understand why REW implemented the timing chirp this way, I don't use it - I don't know what it is telling me.

And BTW, even though I wrote that book, I would STILL hardly call myself an REW expert. The only reason I wrote that book is because SOMEONE has to answer the same questions which are asked over and over, and nobody wanted to do it. There are a lot of things about REW that I don't understand, and how and why REW implements acoustic timing measurements is one of them. I use Acourate for all my timing measurements, and Acourate forces you to create your own tweeter reference and read it manually. I KNOW what the tweeter reference is because I made it myself.

BTW, we should keep a sense of proportion here. As @staticV3 mentions, the latency jitter is only about 30us in your measurement. For most timing measurements, this tiny amount is insignificant. Yes it is better if it's not there, but it also mostly does not matter.

 

[kifeep]

Your UMIK-2 has a loopback feature

I wasn't aware of this and am having trouble finding more information. Do you have a link or reference?

 

[Keith_W]

I wasn't aware of this and am having trouble finding more information. Do you have a link or reference?

No, I don't sorry. The UMIK-2 loopback feature has been talked about many times here and elsewhere. So while I have not personally verified that it has a loopback feature, enough people say it exists for me to believe it

 

[kifeep]

No, I don't sorry. The UMIK-2 loopback feature has been talked about many times here and elsewhere. So while I have not personally verified that it has a loopback feature, enough people say it exists for me to believe it

Hm. UMIK-2 has better timing than UMIK-1 (as you mention above), but the only mention I found of UMIK-2 being capable of loopback is another one of your posts:

Do some speakers play better with DSP? (Recommendations?)

I'm looking at updating my listening setup with better speakers (budget: ~$500 USD). My question is if there are speaker styles (like KEF's coaxial) that take to DSP better, vs being able to bi-amp and use the DSP as an active crossover (like the Klipsch RP-600m II's). At present, my system...

www.audiosciencereview.com

Most of what I'm seeing here is along the lines of "If UMIK-1 isn't accurate enough, get a UMIK-2 or use loopback":

Search results for query: UMIK-2 loopback

www.audiosciencereview.com

I also just checked the UMIK-2 manual and the minidsp forum and didn't find anything there either. I think this might not actually be true.

 

[Keith_W]

I think this might not actually be true.

I did a search too. I think you are right, I gave incorrect information. Thanks for that, i'll go edit the post.

 

[sam_adams]

I think you are right,

I took the hit for you by asking John about this since it seems to show up for the UMIK-1, also. There's nothing in the change logs from the last two years for the current beta builds about this, so, either this was added as a feature in an earlier version or it's a bug. We'll see tomorrow.

 

[3ll3d00d]

IF the timing chirp was part of the impulse response, then there would be no latency jitter. But for the life of me, I can't figure out why you wouldn't make it part of the impulse response. At the moment, what I THINK is happening is that REW starts the ADC/DAC when it takes the timing chirp, and it does it again when it measures the impulse response. Why not simply fix the timing chirp a known quantity (say 700ms and not "about 700ms" in the manual) and start the ADC/DAC only one? That would eliminate all problems with latency jitter. Because I don't understand why REW implemented the timing chirp this way, I don't use it - I don't know what it is telling me.

I don't understand what you mean by making it part of the main ir, they are two entirely separate responses with different bandwidths.

As far as I know the acoustic timing reference works the same as the way acourate uses the tweeter to align measurements except it's that early timing chirp used to align. Ie a single block of audio data is recorded containing 2 or 3 sweeps (the before chirp, the main sweep, the optional after chirp). The early chirp peak position is identified for alignment then the main sweep measurement portion is subject to analysis and delay is reported based on that position relative to the timing chirp.

Finally if https://www.roomeqwizard.com/help/help_en-GB/html/analysis.html#adjustclockacoustic is set then the timing between the before and after chirp is used to further adjust for any clock difference. I guess this one is basically like a form of resampling and is perhaps missing in these examples?

In my experience this is reliable enough to design a passive crossover, which is a fairly high bar, hence it's definitely possible to get good results.

 

[sublover]

I cannot find/see any inconsistencies in your data:
View attachment 511530

All peaks are aligned to within 50µs.

Hi Static, i see what you mean. But lets say if i sweep 10x left channel with acoustic timing reference, than every measurement is different. I want to 100% perfect time align my left and right speaker before Dirac. And if all the measurements ar 2mm or 9mm off all the time than time aligning is not possible. And i was under the impression that UMIK 2 would solve this. Jogn tested the umik 1 and 2 and he stated that the umik 2 does not have such inconsistencies. For FR its of course no problem.

Today i tested the umik 1 and 2 and i see identical timing differences when i do multiple sweeps.

 

[OCA]

Hey guys,



I'm new here, my name is Danny and im from the Netherlands.



I really hope someone can help me out because i have no clue anymore



My setup

Nad 399 bluos Dirac full

LS50 meta speakers

Perlisten R8S sub



I always used the Umik - 1 for REW and Dirac

I have read (and still read it, very info dense) the amazing tutorial about REW from Keith. And here i discovered also that Umik 1 has it limitations.



I have noticed when doing multiple measurements from the same speaker with timing acoustic ref on, i never get consistend readings. And i can see this with Impulse response. I have read that John (rew) stated the umik 1 doesnt have an internal clock. And the umik2 does have and that its super consistent.



Today i received the UMIK 2 and i couldnt wait to try I was hoping to see consistent timing.



I didnt see it.. . Its still inconsistent?!

do i need to adjust a setting?

there must be going something wrong on my part? please i hope someone can help me. Thanks in advance

Did you tick "Adjust clock with acoustic timing reference" in preferences?

 

[sublover]

Did you tick "Adjust clock with acoustic timing reference" in preferences?

Hi OCA!!, this forum is really amazing. All the big names, watched so many youtube tutorials. And Keith i have his book. Ok i look like a groupie now lol. Wish i joined earlier. So many nice and helpfull people

Yeah i did, but i do read somewhere John said this should not be chosen when the mic had internal clock. I just have to find where i read this.

To give you guys an insight why im so obsessed with accuracy is because im at battle with Dirac, i cant get the image center locked. Vocals always tend to lean right with Dirac. Dirac off the vocals centers 99%. So i want the best possible gear to get consistent readings.

 

[JohnPM]

I have read the REW manual a few times to try to figure out how the acoustic timing measurement works. IF the timing chirp was part of the impulse response, then there would be no latency jitter. But for the life of me, I can't figure out why you wouldn't make it part of the impulse response. At the moment, what I THINK is happening is that REW starts the ADC/DAC when it takes the timing chirp, and it does it again when it measures the impulse response. Why not simply fix the timing chirp a known quantity (say 700ms and not "about 700ms" in the manual) and start the ADC/DAC only one? That would eliminate all problems with latency jitter. Because I don't understand why REW implemented the timing chirp this way, I don't use it - I don't know what it is telling me.

That's not what REW does. When using the acoustic timing reference (with clock correction) the stimulus consists of a timing reference, the sweep and then a repeat of the timing reference, all as one continuous signal. When REW analyses the captured data it searches for the first timing reference signal (using cross-correlation with the timing signal data) and then the second timing reference. The number of input samples between those two sequences (with fractional sample precision) is compared to the number of samples between them in the stimulus and used to determine the clock rate correction to apply to the input data. Once the input data has been resampled to correct its clock rate the sweep data is extracted (using the position of the first timing signal as a reference) and processed to generate the IR.

The variations seen in the OP's measurements may be due to replay clock variations (the clock adjustments are different for each measurement) and/or difficulties in reliably extracting the timing signal location. Pointing the mic towards the speakers (e.g. straight ahead) rather than at 90 degrees might help, since that would improve the capture of the direct HF content.

 

[JohnPM]

For comparison, 4 UMIK-2 measurements with acoustic timing ref made at my desk just now:

 

[sublover]

For comparison, 4 UMIK-2 measurements with acoustic timing ref made at my desk just now:

View attachment 511644

Exactly, here we are, spot on! this is what i mean. John could reproduce this with is UMIK-2.
What on earth am i doing different? Never ever what i try i can get it like this.

What could it be? at the moment the umik -1 and 2 perform the same with REW at my place.

 

[JohnPM]

You could place your mic a metre or so from a speaker and pointing directly at it and make some measurements to see if the variation is mainly due to replay chain clock variations (if the variation is still there) or environmental reflections at the measurement position (if the variation goes away). It's all probably academic as far as your imaging concern goes however, since the total variation in the measurements you posted corresponds to less than half an inch of movement. If there aren't gross arrival time differences between left and right at your listening position any imaging issues are more likely to be due to differences between the left and right channel frequency responses.