Jump to content

Using REW to Determine Time Delays Between Drivers


Chris A

Recommended Posts

16 minutes ago, Chris A said:

Looks like you need to either increase the gain on the HF channel by 4 dB, decrease the gain on the LF channel by 4 dB, or split the difference.

 

By the way, the spectrogram looks much better.

 

Chris

 

That was my thought exactly, so I tried to do it in baby steps. I upped the hf by 2dB and the spectrogram got a sharp kink to the right. I put it back where it was and SGram good again. That confused me. One of the input PEQs was -13dB and large Q so I might cut that back some and see what effect that has

Link to comment
Share on other sites

For me, it's really simple: don't move the microphone until you get a loudspeaker dialed in.  Certainly, I don't change the height above floor.  Changing the height above floor will invalidate the measurements, I've found.  I'm talking plus-minus within an inch of height.

 

I've found that I can move the microphone from say the right loudspeaker to the left, and since both are in full corners in a symmetric room, the differences are minimal--measuring Jubilees in corners, center K-402-MEH, and surround AMT-1/Belle bass bins. 

 

I used to measure the distance between the microphone capsule and loudspeaker with a measuring tape to make sure the distance was 1 metre, but I can use the tip of my right fingers to the inside seam of my (left side) T-shirt pocket as a ready measure that's good within a cm or two of being exactly 1 m.  In days passed, I also used tape on the floor (carpet).

 

The biggest factor of all besides not changing the height of the microphone, is making sure that the entire area between the loudspeaker and the microphone is covered with a foam rubber pad (a old mattress topper) laid out sideways.  If you don't use one of those, the phase and group delay information in the measurements is garbage.

 

Chris

Link to comment
Share on other sites

11 minutes ago, Chris A said:

For me, it's really simple: don't move the microphone until you get a loudspeaker dialed in.  Certainly, I don't change the height above floor.  Changing the height above floor will invalidate the measurements, I've found.  I'm talking plus-minus within an inch of height.

 

Of course the whole process of dialing in the speaker system entails more than getting "perfect" results in just one mike location.  But the other side of that coin is that direct comparisons can only be drawn between two or more sets of tweaks made and measured when the tweaks are the only variables.

 

I surmise that, ultimately, compromises must be decided-upon after taking measurements at several mike locations, and that a dedicated room with precision (CNC-type) placement equipment would offer the best repeatability.  Especially so in the case of a manufacturer.

 

So you get everything set just so via measurement.  At what point do you put away that equipment, and do you do any final "massaging" by ear afterward?

Link to comment
Share on other sites

I have nothing against listening window measurements or full polar-coverage measurements.  The problem is that it multiplies the effort by at least a factor of 10 (since that's the minimum number of measurements that must be taken in each axis after initially dialing-in in the horizontal and vertical directions for on-axis SPL.  I've done those measurements outside (i.e., in quasi-anechoic conditions) for a handful of horns/drivers and a full-up loudspeaker.  These are typically time-intensive affairs that yield polar sonograms such as these:

 

K-402-MEH horizonal normalized sonogram.jpg

 

CP25 vertical polar spectrogram.JPG

 

CP25 horizontal polar spectrogram.JPG

 

...etc.

 

12 hours ago, glens said:

At what point do you put away that equipment, and do you do any final "massaging" by ear afterward?

When I reach the end of the measurements/DSP crossover settings updates for a morning or afternoon, or simply the end of the day, I sit back and listen. 

 

I've found that the concept of "salt and pepper" personalization of EQ settings after measurements are unnecessary once I dial-in the loudspeakers--all loudspeakers dialed in for multichannel recordings, and at least the Jubilees for stereo recordings.  The reason for this is two-fold:

  1. the front three loudspeakers (L, C, R) have full-range polar control to well below the Schroeder frequency of the room so EQing to affect the timbre of the loudspeakers is unnecessary.
  2. I fix my stereo recordings at the source of the problem.  Demastering my stereo recordings I've found completely eliminates the need for salt-and-pepper EQ.

I've also found that after a certain threshold in dialing in has been achieved, the listening portion of the process is a real delight.  I'm no longer worried about how the loudspeakers sound, but rather how the recordings sound--the quality of the recordings themselves.

 

After the latest round of dialing in all the loudspeakers in the array this spring and summer--including resetting relative channel gains and carefully resetting loudspeaker channel delays in the array--the results are mesmerizing, demanding your listening attention. 

 

Chris

Link to comment
Share on other sites

I'd say I want to help you in your instructional efforts but it might be better to say that I don't want to hinder them.  Think of the audience member who's familiar-enough with the subject matter and asks questions for the other audience members who may not think to ask them at the time.  To that end, would you say that you've found situations where placing the mike differently, but not necessarily very differently, yields results which would suggest different settings?  And if so, how have you handled it?  Maybe room treatments, or rather, compromises in settings?

 

Also, if the whole package of room and speaker alignment needs attention, where do you start?

 

 

Link to comment
Share on other sites

23 hours ago, Chris A said:

is making sure that the entire area between the loudspeaker and the microphone is covered with a foam rubber pad (a old mattress topper) laid out sideways.

I'm sorry what? The floor area, the ceiling, the walls? All? Having a hard time picturing this.

Link to comment
Share on other sites

Just the floor...a single row of absorption squares or a pinned up comforter or blanket on each adjacent wall to the K-402 and bass bin mouths will be enough (about 2 feet wide and the height of the loudspeaker).  Using 2'x2' absorption squares will also likely improve the phantom center channel imaging, too.

 

Chris

 

Link to comment
Share on other sites

1 hour ago, glens said:

I'd say I want to help you in your instructional efforts but it might be better to say that I don't want to hinder them.

Fair enough.  I certainly need that sort of thing for a task like this.  I see this thing as evolutionary, so understand that it will take multiple revisions over time to get into a complete enough shape to be a one-stop-shopping tutorial.  Not that I'm not trying to make the effort larger than is absolutely necessary, but in the end the product has to be good enough and complete enough for those new to the subject to use to dial in their systems without high levels of angst being generated.

 

Chris

Link to comment
Share on other sites

I'm not sure that I'm an expert (famously defined by Malcolm Gladwell's book Blink as "10,000 hours of exposure to the subject matter"), but I have been vocal on audio forums about exploring the possibilities of using DSP crossovers to improve loudspeaker performance in-room.

 

I've also found little-known and perhaps surprising results using REW measurements and tweaking DSP crossover settings, mostly having to do with phase/group delay response improvement and the subjective response to the improvements, and in particular the use of crossover filtering schemas other than those typically provided in the DSP crossovers.  These are based on comments that Tom Danley has made elsewhere on audio forums which have yielded extremely interesting results of re-creating those performance capabilities in loudspeakers. Some of my measurement observations are based on benchmarking measurements of an SH-50 and looking at the extensive passive crossover network included with it. 

 

Some observations have been based on reading Toole's papers and book and tweaking, measuring and listening to those improvements, notably to the comments that as the SPL response is further flattened, the better the listening performance.  I've found Toole's comments to be quite true down to ±1 to ±1.5 dB flatness on axis or in the listening window, beyond which the room furnishings and horn/direct radiator performance off-axis become the controlling factors.

 

I'm making good progress on the promised tutorial.  Currently, it's up to ~40 pages and still seems to be growing.  The information in it is based primarily on previous threads that I've posted with additions of an outline based on two newly developed QFD matrices.  I will start on filling in some of the holes in the structure of the tutorial probably in the next couple of days.  When I get tired of working on it, I'll publish what I have and let others ask questions and make requests for further information to continue its development.

 

Chris

  • Thanks 2
Link to comment
Share on other sites

I've continued to measure, adjust, listen, repeat several times. So far I've had the best luck with really getting the group delay a lot closer to zero on many parts of the spectrum. However I seem to have ended up with a slightly "worse" spectrogram @Chris A is that progress or not?

 

I'm finding the Phase hard to understand how I actually influence it, any pointers? I can see the wrapped phase jumping quickly in parts between -180 and +180 but I can't/don't understand how negate or minimize that.

 

Sometimes when I go back and adjust the PEQs to flatten the peaks/valleys it in turn effects the Group Delay and/or the Spectrogram and other times it does not. Is that expected behavior?

Link to comment
Share on other sites

41 minutes ago, rplace said:

However I seem to have ended up with a slightly "worse" spectrogram @Chris A is that progress or not?  I'm finding the Phase hard to understand how I actually influence it, any pointers? I can see the wrapped phase jumping quickly in parts between -180 and +180 but I can't/don't understand how negate or minimize that.  Sometimes when I go back and adjust the PEQs to flatten the peaks/valleys it in turn effects the Group Delay and/or the Spectrogram and other times it does not. Is that expected behavior?

These are all areas that I intend to fill in more within the tutorial outline that I've created, but have yet to complete those areas.  In the meantime, some answers to your specific questions: 

  • Generally, it's better to smooth the spectrogram AND approximate a vertical response simultaneously.  At the point where you begin to see pre-ringing on the step response plot (shown in the impulse response view...and the source of this is generally due to overcompensating the bass bin delay) and seeing the group delay also grow due to that, you've generally reached the end of the road on tweaking delays and crossover frequencies, etc.  You have to look to different drivers to get significantly different results.
     
  • Phase is very touchy, but once you get the spectrogram results to be more-or-less impulse spikes (i.e., no right-hand tails on the spectrogram), the phase plot will begin to respond constructively.  REW's phase calculations based on the measurements are very sensitive to early reflection noise, so this is the reason why so much absorption is needed to get good phase plots.  The trade-off is always rising group delay spikes above and below the center crossover frequency.  The ear is sensitive to any spikes above ~1-2 ms, anywhere on the group delay plot above ~200 Hz (and perhaps at lower frequencies, too).  This is the challenge. 
     
  • All the "knobs" that you turn to affect phase, group delay and delay all interact with the SPL vs. frequency (a.k.a., frequency response).  This is because of the lobing behavior of the HF horn interacting with the bass bin, etc.  Change the delay, and the phase, group delay and frequency response all change, too, and the in-room acoustic lobes will shift and rotate based on those tweaks.  So there is a dialing in process that is iterative that's not avoidable in my experience. 
     
  • Perhaps using FIR filters is the solution to most of these ills, but my experience using them have been that I've been able to achieve the same subjective results without FIR filters using the drivers/horns that I've been using, so I'll continue to use IIR filters (PEQs, etc.) to dial everything in.  You may find that it's time to make the shift to FIR filtering, but note that it takes considerably more computational power to do this, plus hardware capable of handing FIR filters.  I found the miniDSP 2x4 HD does not have sufficient tap lengths to be useful to correct for the phase growth outside of the crossover region(s).

Chris

  • Like 1
  • Thanks 1
Link to comment
Share on other sites

So I have spent the better part of my free time the past week manually changing PEQ values to flatten. Lots of iterations. If the SPL looks flatter but the Spectrogram or GD changes a lot I go back to the original, if it help or does not hurt I leave it. Lots and lots of PEQ changes for all aspects like amount of attenuation, frequency and Q all over and over and over. I don't see how in my mind now it would be practical to use the EQ wizard. With each change to PEQ I then play with delay to see if I can improve

 

Here are some comparisons of last weeks quitting point and where I am Friday of this week.

 

Group delay IMO looks better just about everywhere. Any room for improvement via Xilica or am I at the mercy of my drivers now?

 

LastWeekVThisWeekGroupDelay.jpg.c8cb56c156e8a35dc4d1e3796be56c7c.jpg

 

 

Here is the SPL. Still some work to do, but I think if  you draw a line where the cursor is I'm not too bad up or down. I can make if flatter in places but it drastically changes the GD/Spectrogram

 

LastWeekVThisWeekSPL.jpg.807afd126f0c45f2b4088cbdb6871b8e.jpg

 

Phase is still a mystery but I spent a good deal of time making changes to only improve Phase, then altered as I saw fit for helping GD/SGram. Reasonable? Googling and reading I gather you want less ups and downs close together. I feel like I have done that between 200 and 2K. How do I get it "fixed" at 3K and above?

 

LastWeekVThisWeekPhase.jpg.3b9809b7bc3665353277fc913f05a81b.jpg

 

 

Now a couple of spectrograms. Still nothing quite as vertical as I've seen from @Chris A Is this acceptable? I'm not really sure if the 2nd is an improvement. It is more vertical form 200 down to 600 but it has that sharp cutback to the left at 200 where the original tails off nicely. Which is better.

 

Original

LastWeekSGram.jpg.201c1c0494e44161264770f85643d297.jpg

 

New

ThisWeekSGram.jpg.71f0098af919a0baea2cbc5a1bae42eb.jpg

 

 

 

With enough effort  you can make any one look pretty good...but the others suffer. This seems like a pretty good compromise all around, but I'm still not sure of phase. It looks nothing like Chris'

 

Thanks for playing, any comments?

 

 

Link to comment
Share on other sites

On ‎9‎/‎24‎/‎2019 at 4:22 PM, Chris A said:

The trade-off is always rising group delay spikes above and below the center crossover frequency.  The ear is sensitive to any spikes above ~1-2 ms, anywhere on the group delay plot above ~200 Hz (and perhaps at lower frequencies, too).  This is the challenge. 

 

Rereading your post I now feel like perhaps my GDelay is not that good. I zoomed in and put the cursor at 200Hz/2ms. Am I correct that above this threshold is hard on the ears?

 

LastWeekVThisWeekGroupDelay2.jpg.f812effaf958f45d3af1e59e9e0fd5d8.jpg

Link to comment
Share on other sites

1 hour ago, rplace said:

 

Rereading your post I now feel like perhaps my GDelay is not that good. I zoomed in and put the cursor at 200Hz/2ms. Am I correct that above this threshold is hard on the ears?

 

LastWeekVThisWeekGroupDelay2.jpg.f812effaf958f45d3af1e59e9e0fd5d8.jpg

Yes.  Group delay above about 1-2 ms (the lower limit being true around upper midrange frequencies below 4 kHz) is audible in well-treated rooms with low levels of early reflections as making the resulting sound fuzzy/woolly, harsh, and opaque...just like modulation distortion.  In fact, it is modulation distortion itself that is the cause of the detection of phase and group delay distortion for typical loudspeakers.  See Psychoacoustics: Facts and Models by Fastl & Zwicker (2006), section 7.3 (pg. 188). 

 

In my experience, you really want to aim for a smoothly varying phase response with correspondingly low group delay values (i.e., group delay is nothing but a plot of the instantaneous slope of phase response curve).  After you achieve a smoothly varying phase and group curves, you can begin to decrease the bass frequency delays by increasing the HF delay, until you reach the point that the group delay curve develops a discontinuity around the crossover band center frequency.

 

You show spectrograms where the peak energy time curve jumps to the right as frequency decreases.  It's this curve that you don't want discontinuities, and the places where there are discontinuities are also where the group delay curve jumps upwards.  So you start at a large right-hand tail on the spectrogram, then begin increasing the HF delay until you see a polarity shift as you get close to a 180 degrees of relative phase alignment between the HF and bass bin.  It is at this point that, if you decide to continue increasing the HF delay, you'll need to flop the polarity of a channel.  If you continue to increase the HF channel delay to correct for the right-hand tail on the spectrogram, you'll need to flop the polarity back to normal phase at some point once again.

 

Chris

Link to comment
Share on other sites

So the weekend is coming up, time to get back to important things like making things sound better. On Wednesday I had to some time to play and think I made some good progress on Group Delay and Phase. I'll hold off on the details of what I actually did and concentrate one single goal for now. Just for the sake of argument lets assume everything else is exactly the way we want it.

 

How would I address the large spike in GD around 230Hz?

 

My gut tells me I should not have multiple PEQs in the same region for both input and output and HF and LF. In my mind this is where the problem is....I know in the past @Chris A said (paraphrasing) that as the GD and Spectrogram get better eventually you are at the mercy of the drivers. Given my phase plots, GD and SPL below I think the green plot is pretty good at just +/- 2dB when drawing a horizontal line around 82dB. The phase and GD are very similar and in my opinion better than my previous posts.

 

Here are the various PEQs that I think are the problem or at least contributing to the problem. What would be a good strategy to perhaps combine them into one PEQ? Is that even a reasonable train of thought. Am I reasonable to think that if the Xilica is doing all the attenuation and addition around 200Hz and I have a "problem" around 200 they are related?

 

 

PEQs.JPG.d11e151365894d05e4cc2de184373ea4.JPG

 

 

1018721042_WednesdaySPL.jpg.0db96a04ed685ccc9b9539b5787b5b74.jpg

 

1662169557_WednesdayPhase.jpg.cec47a9a6415ed449ef00ac135bcf5c0.jpg

 

 

GDWednesday.jpg.282f9a8e00f601f6e22ffffe88c668e0.jpg

 

 

 

I know I said only one question for this post, but I have to add extra credit. See the 4K spike in GD? from time to time that will jump WAY up. I can take a few measurements changing nothing and it will stay up around 4-5ms. After waiting a bit and again changing nothing it will drop back below 2ms. I don't see that behavior in any of my measurements except at 4K. What is up with 4K

 

 

 

Link to comment
Share on other sites

19 hours ago, rplace said:

My gut tells me I should not have multiple PEQs in the same region for both input and output and HF and LF.

It is possible to simplify PEQ overlaps in the output channels with the input channels, but note that in the crossover interference bands, you will likely use more input channel PEQs if you decide to "simplify".  Just zero out the PEQs of interest, re-run REW measurement on combined HF/LF, then use the EQ facility to optimize to number of (input channel) PEQs needed. 

 

If you run separate measurements of the HF and LF drivers first and then combined measurements later, you'll probably end up using basically the same PEQs as before you started with. 

 

Also note that a lot of frequency response issues around 200 Hz are not "minimum phase" due to the position of the microphone above floor and from the front wall.  In other words, you're dealing with lobing between the two ways and 1/2 wave cancellations from room boundaries.  All that you're doing is pushing around the lobes in the crossover interference region, and trying to correct for boundary reflection cancellations.

 

Chris

  • Thanks 1
Link to comment
Share on other sites

18 hours ago, rplace said:

See the 4K spike in GD? from time to time that will jump WAY up. I can take a few measurements changing nothing and it will stay up around 4-5ms. After waiting a bit and again changing nothing it will drop back below 2ms. I don't see that behavior in any of my measurements except at 4K. What is up with 4K

Not knowing the exact nature of your HF drivers, I would guess that you are seeing non-pistonic motion of the diaphragm(s) starting at 4 kHz.  The calculation of phase vs. frequency within REW is reflecting the fact that the diaphragm is thus acting like two diaphragms, with some uncertainty in the relative phase between them at the break-point frequency of 4 kHz where the single diaphragm is starting to act like two diaphragms that are coupled to the voice coil assembly.

 

Another possibility is that there is an extremely effective nearfield reflector near the diaphragm (within 0.9 inches from the diaphragm itself) that is putting out a strong reflection at 4 kHz that REW is trying to calculate in the phase vs. frequency calculations.  This could also be a reflection off the spider of the driver assembly. 

 

Chris

  • Thanks 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...