Using REW to Determine Time Delays Between Drivers

[Rudy81]

Disregard the last question...you already noted that the filters should be in place.

 

If I understand you correctly, what I should do is to run a sweep with xovers in place and  both drivers operating. I am guessing initially NO delay to either so that you can get an idea of where the delay should be based on spectrogram and GD plots.  Is that how you would do it?

[Chris A]

19 minutes ago, Rudy81 said:

I am guessing initially NO delay to either so that you can get an idea of where the delay should be based on spectrogram and GD plots.  Is that how you would do it?

Yes.

[Rudy81]

What do you make of this?  20-5khz sweep.  NO time delay of any kind.  Xover on both drivers L/R 24dB at 300hz. What can you deduce concerning time delay based on the plots?

 

 

 

 

 

[Rudy81]

Chris, so you can illustrate things better to us dummies, I took the above plots with NO delays in either driver. It occurs to me that it might be helpful to compare graphs with NO delay in either driver.  Delay of 0.7 ms woofer only.  Delay of 0.7ms HF Oris only. That should allow us dummies to better understand what the graphs are telling us.

 

Here is the SPL comparison and I am aware the HF section needs gain reduced. Red is No delay in either driver, xovers set at 300Hz. Orange is the woofer section delayed 0.7ms. Green is the Oris delayed 0.7ms. This is an exaggerated time for illustration.

 

 

[Rudy81]

Here are the GD plots.

 

[Rudy81]

Spectrograms.

 

[Rudy81]

Phase comparison near the crossover looks like this, and a closeup near the xover region.

 

 

 

[Rudy81]

Just for grins, I figured the best SPL was to delay the HF 0.13ms.  Spectrogram and inverted Spectrogram for comparisson at 0.88ms delay and the spectrogram at HF delay 0.13.

 

[Chris A]

4 hours ago, Rudy81 said:

What do you make of this?  20-5khz sweep.  NO time delay of any kind.  Xover on both drivers L/R 24dB at 300hz. What can you deduce concerning time delay based on the plots?

 

 

I would recommend opening up the sweep to 10-20,000 Hz first (and always when doing these sort of adjustments).  It's the absence of information above 5 kHz which is the issue here, not that below 20 Hz.  I'd also plot the group delay with a 1ms minor division scale on the vertical axis.  One really can't see very much at the resolution of 5ms minor scale.

 

If I had to choose based only on what I see here, I'd choose 0.5 ms delay on the HF channel, then rerun the sweep to see if there are any issues with lobing/cancellations around the 300 Hz crossover point...notably check at 900 Hz where the peak energy time curve moves horizontally at that frequency to see if the dashed line closes up toward a vertical line to any degree.

 

The issue that I see is the 5ms jump in group delay at 5 kHz: this is a fairly large jump that needs exploration as to the cause...corresponding to a 68 inch jump in the acoustic center backward away from the microphone.  Extending the sweep to 20 kHz and increasing the magnification of the vertical axis scale is a start at understanding this, along with "generating minimum phase" and turning on the "excess group delay" curve.

 

Here is a recent measurement that shows a typical compression driver/horn-loaded bass bin group delay plot, after the HF delay is dialed in.  Note especially the "excess group delay" curve that is added to the plot by selecting it after the "generate minimum phase" button is pushed from the Controls menu (the Controls icon is far to the right on the plot, above the plot area):

 

 

What can be seen here is that there is about a 0.4 ms rise in the excess group delay curve near the 550 Hz crossover point (using first order filters), then the excess group delay again subsides to zero at 100 Hz.  This is a good compromise on where to set the overall delay (1.53ms HF channel delay) and results in the excellent GD performance overall that you see here.  The little 0.5 ms rise at 13.6 kHz is due to the non-pistonic transition of the K-691 driver at that frequency (also visible in the spectrogram below as very mild chattering above 10 kHz).  This is an extremely well-behaving measurement example:

 

 

I offer this example so that you might iterate your measurement plots to see what is needed to see, rather than me guessing what the next step might need to be.  It's a lot easier when you've got REW up and running and you can make the changes in plots and measurement sweeps in real time.

 

Alternatively, if you can run a sweep from 10 Hz--20 kHz, then package up the .mdat REW measurement, you can send it to my email address, then I might be able to respond more effectively to this specific case.  You can PM me for my email address if you don't still have it in your email address book.  My email address hasn't changed in the ~9 years since you were here.

 

Chris

[Rudy81]

Ok, just sent you my files.  Here is what I think you need in terms of proper graphs.

 

[Rudy81]

I guess I don't understand this whole GD plot.  All my readings are around  the 5ms. line.  You would think a single driver from 300hz to 20khz would eventually settle at the 0ms line....no?

[Chris A]

I can now see what the issue is...calibrating the impulse response zero time in each plot.  Below you will see an SPL/phase measurement just after being taken, with the Controls menu turned on to the right.  There is a button on the Controls menu marked "Estimate IR Delay" ("IR" = impulse response).  You need to push that button, then a window will pop up called "Delay estimation complete", and you push the "Shift IR" button on that menu.

 

 

Then you return to the Controls menu and push the "Generate minimum phase" ("3")  button, then the "Unwrap Phase" button ("4").  Then you will see something like this:

 

 

Notice that the SPL vs. frequency curve is unaffected, but the phase curve (in this example) has been transformed.  If you then save the measurement, these settings (all except "Generate minimum phase") will be saved with the measurement so you don't have to push these buttons again.  The "Generate minimum phase" button will have to be pushed after every time that you reopen a saved measurement.

 

Now your phase and group delay plots will be useful for setting channel delay (time alignment). 

 

Chris

[Chris A]

After accomplishing the above calibration of the measurement's IR datum, calculating the minimum phase (and at the same time, the minimum group delay, excess phase and group delay curves) and upwrapping the phase curve, the following plot will show the correct value for HF channel delay (see cursor position at the 300 Hz crossover frequency in this example):

 

 

So the correct value of channel delay to use in your next measurement is shown by the vertical axis value of the excess group delay curve at the crossover frequency (300 Hz): 1.344ms in this case.

 

Chris

[Chris A]

By the way, the above shift in delay also required a polarity reversal on the bass bin so that it would be in-phase at the crossover interference band.  The reason for this is due to the 1.4 ms delay, which is just about 1/2 a wavelength at the 300 Hz crossover point.

 

Here is the group delay plot (including excess group delay) without the polarity flop on the bass bin, but also showing the correction of the excess group delay curve back towards zero. The obvious polarity issue at the 300 Hz crossover point yet to be addressed, as shown by the spike of the group delay curve at that point.

 

 

And below you will find the spectrogram after the bass bin polarity is flopped:

 

 

EDIT: Also note that the HF driver itself experiences a lagging impulse (and phase) response below 1 kHz, which shows up as the broader portion of the impulse column between ~180-1000 Hz.  This is inherent in the driver itself.  However, the overall impulse response has been moved toward the left

 

Chris

[rplace]

23 hours ago, Chris A said:

So the correct value of channel delay to use in your next measurement is shown by the vertical axis value of the excess group delay curve at the crossover frequency (300 Hz): 1.344ms in this case.

 

If all of the above was done (Generate Minimum Phase, Shift IR, Unwrap, etc.) on a measurement that already had some, but not the correct delay, could you just add that value to the existing delay?

 

38 minutes ago, Chris A said:

By the way, the above shift in delay also required a polarity reversal on the bass bin so that it would be in-phase at the crossover interference band.  The reason for this is due to the 1.4 ms delay, which is just about 1/2 a wavelength at the 300 Hz crossover point.

 

Can we "see" that anywhere or is it just the fact that you know 1.4ms at 300Hz = ~half a wavelength so *you* know (I would not) to reverse the phase.

[Chris A]

I would probably take a measurement sweep without the polarity reversal just to convince myself that the polarity reversal will be needed, then take another sweep with the polarity reversal...

 

...but yes, I often try to move the channel delay in multiples of half the crossover point period (i.e., 1/frequency) to avoid having the search around for the fine-tuned delay value that will give me a good SPL response across the crossover interference band.

 

14 minutes ago, rplace said:

If all of the above was done (Generate Minimum Phase, Shift IR, Unwrap, etc.) on a measurement that already had some, but not the correct delay, could you just add that value to the existing delay?

Yes.

 

Chris

[rplace]

4 minutes ago, Chris A said:

Why wouldn't you want to correct the IR datum time?  It creates all kinds of offset issues otherwise, and tomorrow or the next day, how will you remember what the correct mental offsets should be to offset the phase and group delay plots?

 

I'm must not be asking my question correctly or it is not an appropriate question to ask. I was just wondering if somebody had arrived at the point where they SPL flattening, gain setting and some delay set (say 2ms for the sake of example) if they cold just add the 1.344ms to the existing 2ms to get to 3.344ms or if they first needed to zero out the delay that was present, take new sweeps then preform the estimate IR/Shift IR/etc.

 

Until a few minutes ago I never knew to perform those 1,2,3,4 steps above on the SPL & Phase screen. I've always looked at the phase wrapped to see how many times it traveled up and down between 180 and -180 thinking the less the better. Then I used the GD screen and generated minimum phase there hoping to find everything under 2ms. If not under 2ms I've been looking for places the Min Group Delay is flat as potential places to EQ the GD above 2ms down. So I've been tweaking EQ and Delay hoping to get better GD, Impulse and Spectrogram. I've been doing that iteratively. Wondering now if I could go to your post 5 above this one and use it to dial in some more appropriate delay.

 

I've learned a lot from you in the last couple of weeks and I could tell that Rudy's were not quite right, but I could not say why. I think what I've really learned is what the graphs should look like but I've not learned how to interpret them correctly to turn the proper knobs and dials to get there.

[Chris A]

Re-read my response.  I changed it when I realized that I'd answered a question you hadn't asked.

 

On 10/28/2019 at 8:03 AM, rplace said:

Until a few minutes ago I never knew to perform those 1,2,3,4 steps above on the SPL & Phase screen.

I was afraid of that.  This latest portion of the thread will be featured prominently in the tutorial.

 

On 10/28/2019 at 8:03 AM, rplace said:

Until a few minutes ago I never knew to perform those 1,2,3,4 steps above on the SPL & Phase screen.

Perhaps you know which buttons to push to get your desired results now...?  Anyway, there are a few "gotchas" that I find with virtually any application (even Word and Excel), so I just learn to automatically do certain things within each application to get what I need.

 

Chris

[rplace]

21 minutes ago, Chris A said:

Perhaps you know which buttons to push to get your desired results now...? 

 

Yea, for sure. Thanks. I went back and did those 4 steps to some measurements I had saved. Ones I thought looked pretty good for phase. On a couple it added a lot more 180 to -180 zig zags on others it just moved the phase overall to the left. Is that normal behavior. Are the "better" ones still the ones with the least total number of phase shifts/cycles/round-trips? BTW, what is the proper term for cycling through 360 degrees of phase over and over across the 20-20K range?

[Rudy81]

Chris, thank you for all your help the last few days and your patience.  As you found out, part of the problem is that some us, including me, don't know how to properly use REW to get the most information from that powerful tool.  I am slowly getting a grip on what you are trying to teach us.  I am really looking forward to implementing these ideas on my system when I get back from work. 

 

I still struggle to understand the concepts of the acoustic center vs. the physical center of the drivers.  That is one issue I am still researching.  In my case, the drivers are only physically separated by 7 inches or so.  But, the delay of 1.4ms or so shows a need for a displacement of about a foot and a half of 'electronic' distance.....so, it doesn't add up in my brain.

 

I do encourage you to write that how to .pdf you spoke of.  It would be a great 'cookbook' for us who don't totally understand all this stuff.  I think when I retire I'll look for a course on audio or subjects that will teach what I lack in audio knowledge. If I could be of any help when you get around to working on the cookbook, please let me know.  Be glad to help.

 

When I get back from work I will be back on the trail of the elusive time delay.