Jump to content

Using REW to Determine Time Delays Between Drivers


Chris A

Recommended Posts

So here is the measurements at 600, 800 and 1000. Nothing too earth shattering I can see. I think the Spectrograms/Group Delays are much more interesting for discussion. I'll post them in different posts so we can comment individually on what each means

 

NOTE Red = 800 not 8000

 

6h8h1K.jpg.7a1c53ff7b714d0b33621734b0e602ee.jpg

Link to comment
Share on other sites

Here are the 3 Spectrograms for the sweeps above. Note that on my 350XO point on the previous page they are not "smooth" before applying any delay. That is none of these have delay yet. Just full sweeps with various crossover points all Bessel 6db. They look more like what Chris and others have posted in the past. They also look a lot alike.

 

Spectro600.jpg.d4110c2508f76a057030f775d7eafb61.jpg

 

Spectro800.jpg.fa061d99eac370bbda429dcb7efe6308.jpg

 

Spectro1000.jpg.8c0c001d1bf460111a3f2e40d6e65364.jpg

 

 

Link to comment
Share on other sites

Now the group delays for those same sweeps/Spectrograms above 600/800/1000.

 

GD600.jpg.2151c8e89f77dcddf1a11fce2de90f3a.jpg

 

GD800.jpg.d2866314b7a7219f1641237c6fd5b789.jpg

 

GD1000.jpg.d2dff0224f07949de118fafeaac95e31.jpg

 

 

To my eyes I say 600 and 1000 are decent. Looks like only need to figure and apply the delay of around 5ms occurring around those two 300Hz humps.

 

What is the funky stuff happening on the 800 sweep between 150 and 600?

 

Curious to hear what others think for both Spectrograms and Group Delays. Again no delay on any just moved my XO from 350 earlier this morning to 600/800/1000

Link to comment
Share on other sites

You probably need to further iterate the delays and perhaps polarity, but the 800 Hz crossover shows a nearly flat excess group delay from 150-700 Hz.  That's the one that I'd focus on  first.  It looks like you can add another couple of ms of delay to the HF channel to bring the excess group delay back down toward the zero level. I would also zoom in on the group delay plots--making 10 ms full scale.  Then you can see the effect of your tweaking delay more clearly.  You'll also need to readjust your PEQs to accommodate the new delay value, once you find the "sweet spot".

 

EDIT: The spectrogram plots also show that you need to move the delay value up by about 2 to 2.5 ms.

 

Chris

Link to comment
Share on other sites

2 hours ago, Chris A said:

You probably need to further iterate the delays and perhaps polarity, but the 800 Hz crossover shows a nearly flat excess group delay from 150-700 Hz.  That's the one that I'd focus on  first.  It looks like you can add another couple of ms of delay to the HF channel to bring the excess group delay back down toward the zero level. I would also zoom in on the group delay plots--making 10 ms full scale.  Then you can see the effect of your tweaking delay more clearly.  You'll also need to readjust your PEQs to accommodate the new delay value, once you find the "sweet spot".

 

EDIT: The spectrogram plots also show that you need to move the delay value up by about 2 to 2.5 ms.

 

Chris

 

 

Now I'm stumped. I thought on the Spectrogram you looked for the sharp change in direction on the Peak Energy curve near the XO point. That would make me think about 1.5ms (first blue line) or perhaps ~3.6ms (2nd blue line). How do I read below and come up with 2-2.5ms (the range of the two black lines).

 

Spectro800Questions.jpg.ef29900f556a11f026d8d2397a3beb7c.jpg

 

 

As for the Group Delay, I'm still dumb as a rock missing the basic concept. I zoomed in on where I think the issue should be. Are we looking for the difference between red and white lines where they depart? If so moving from left to right which orange circle shows us the roughly 2ms issue?

 

150-700 seems like the problem to me you say "nearly flat" so I am obviously not getting it. I thought the closer the white line (excess) tracked to the sweep (red line) the better because there was no gap/delay between them.

 

GD800Question.jpg.cff11fad2c26e61a748f6bd7da489a1b.jpg

Link to comment
Share on other sites

You want the white line (excess group delay) to be smooth and minimized, i.e., as close to zero as possible.  The red lines are part of the "minimum phase" behavior of the driver itself, not your dialing in process, i.e., you can't materially change the red trace unless you use different drivers/horns.  The peaks that you have circled are not something that you can change--it's part of the driver/horn combination itself (even though the white line plot shows them to be excess group delay).  Zero group delay is a horizontal line lying on zero ms on the vertical axis.

 

893462326_Groupdelayplot.jpg.98afb6e5af292a06d99343566473dbdf.jpg

 

You also want the spectrogram to be an impulse--straight up and down, NOT spread out to the right above ~200 Hz, i.e., something like this:

 

Spectrogram View Settings.jpg

 

You also need to look at the SPL response and phase plot, but if you achieve a minimized, flat and smooth excess group delay curve, you will have done almost all of your due diligence.

 

Chris

 

Link to comment
Share on other sites

I can see that there needs to be a tutorial on how to use REW's plots with a DSP crossover to achieve optimal results.  Mitch Barnett's book (Accurate Sound Reproduction Using DSP) just doesn't tell the novice what they need to know and how to read each plot to glean information to achieve optimal results.

 

Toole wrote a small section on group delay (sections 4.8.1 and 4.8.2 in his 3rd Edition), and really nothing on excess group delay in particular.  And it's Toole's book that's been doing most of the communicating, unfortunately.  Same thing for reading phase plots and doing something about phase issues.  There is absolutely nothing on Danley-style crossover filters (i.e., achieve near zero phase shifts through the crossover regions), and the audible effect of doing all these improvements, some of which you'll find in the Subconscious Auditory Effects of Quasi-Linear Phase Loudspeakers thread.  There is no discussion of the issues that you can run into while dialing everything in, or how to look for the source of problems that you hear--but in the measurement plots, etc.

 

Perhaps I can cobble something together based on the threads that I've posted.  It's really odd that this hasn't been done by someone else up to this point.  Just reading the REW tutorials really doesn't get you there, especially when dialing in high performance horn-loaded (i.e., full range directivity) loudspeakers.  To me, it should be clear and the users easily understand the "why" of each part of the process.
 

Chris

Link to comment
Share on other sites

13 hours ago, Chris A said:

You want the white line (excess group delay) to be smooth and minimized, i.e., as close to zero as possible.  The red lines are part of the "minimum phase" behavior of the driver itself, not your dialing in process, i.e., you can't materially change the red trace unless you use different drivers/horns

 

That helps a lot, thanks! I thought we were looking for the differences between the two lines at a given point (the XO point most likely). Knowing the goal in a perfect speaker would be a horizontal white makes sense. Zero on the graph, zero all across, no delay anywhere. Now I get it.

 

13 hours ago, Chris A said:

You also want the spectrogram to be an impulse--straight up and down, NOT spread out to the right above ~200 Hz, i.e., something like this:

 

Again, very helpful to know the goal. In an earlier post in one of these threads I asked if a perfect speaker would have a vertical line at zero, I don't believe anyone ever answered that question. Making me think now that Chris is the only one that would actually know that answer. My take away was that we didn't need it vertically at zero we just wanted smooth transitions moving to the right, not jagged. Obviously that was the wrong conclusion. Learning more every day, thanks.

 

I'm getting the feeling that not many here actually know what they are doing, but are not as willing as me to admit they don't know how to use these tools or "why" behind what the changes do. That is fine, I personally just like to understand the why and know what I am actually doing, not just blindly follow instructions.

 

13 hours ago, Chris A said:

I can see that there needs to be a tutorial on how to use REW's plots with a DSP crossover to achieve optimal results.  Mitch Barnett's book (Accurate Sound Reproduction Using DSP) just doesn't tell the novice what they need to know and how to read each plot to glean information to achieve optimal results.

This is very much the case for me. I don't really have any sort of background to understand what I am looking at. I'm pretty methodical and pretty good at following directions. I just need to know WHAT the goal is and how to implement the changes and I'll be on it like a dog on a bone. So far it has been a lot of trial and error and shooting in the dark.

 

Link to comment
Share on other sites

16 hours ago, Chris A said:

You'll also need to readjust your PEQs to accommodate the new delay value, once you find the "sweet spot".

 

I ran lots of sweeps with various delays in tiny increments/decrements and think I have a better Spectrogram and Group Delay. Maybe not a sweet spot but better spot, as in a "not so bitter spot".;)

 

I have some questions on this second round of PEQs. Since I only put delay on the HF do I run a HF only sweep and optimize with REW Wizard; or do I run a full (HF/LF combined) sweep? No matter the answer above I'll get say 4 to 6 new PEQ values. Surely they won't be anywhere near the existing ones I started with. How do I apply the new values?

  • Do I simply combine them to the existing PEQs going from say 4 to 8
  • Do I try and find ones that are close to each other in Hz value and +/- dB value and combine them
  • Do I replace my existing ones with only the new ones  - I don't think so since the new ones are based in the existing ones

If the answer is simply add them so you end up with more, what happens if I run out of PEQs? Do I need to make a best guess on winners and losers?

Link to comment
Share on other sites

5 minutes ago, rplace said:

My take away was that we didn't need it vertically at zero we just wanted smooth transitions moving to the right, not jagged.

Well, you do want a smooth maximum energy curve with no discontinuities, but you also want the spectrogram to be as vertical as you can get it, even to the point of "backing up a little"  to push the overall response closer to zero overall. 

 

One of the reasons why I use QFD matrices (shown in first post in this thread) is to show the connectivity of the "whats" to the "hows", and to show that there is oftentimes contention between the "whats" which leads to compromises in the "hows". Tradeoffs are often encountered.  One of my friends that I worked with for many years used to say that "to teach, you not only have to tell them 'what' and 'how', you must also tell them 'why'".

 

Chris

Link to comment
Share on other sites

2 hours ago, rplace said:

Since I only put delay on the HF do I run a HF only sweep and optimize with REW Wizard; or do I run a full (HF/LF combined) sweep?

Since changing the delay value(s) only changes the crossover interference region between the two "ways", changing the delay values doesn't change the flatness of the individual drivers. So to answer your question: no you don't need to run individual driver sweeps again, only combined sweeps. 

 

In this instance, you need to use the input channel PEQs instead of the output channel PEQs, since you're only dealing with combined response.  If you've set the delay value in a bad spot, you'll see big dips in response where the two drivers are cancelling each other at some frequency or frequencies.  In this instance, you need to continue tweaking the delay until these SPL nulls are minimized, and the spectrogram is still as vertical as you can get it, and the excess group delay curve shows the minimum number and intensity of peaks.

 

I use REW's EQ facility to help me optimize the frequency response in the crossover region using the input channel PEQs.  If you've still got cancellation notches in the measurement, then you need to go back and keep  working on the delay tweaking (usually in 10ths or 100ths of a millisecond) to find a better value.  Remember that phase goes in 360 degree cycles, even if you're more than one wavelength out of sync, so you can have good looking crossover region with no nulls and still be one or more complete 360 cycles out of phase. 

 

If you run out of PEQs, then you need to step back and take a look at the number of PEQs that you're using early in the process, before you're combining the two ways in one sweep.  I usually bump the flatness target up by one dB to ±3 dB or even higher so when doing the individual driver flattening before combining the drivers together:

 

1282543722_FlatnessTgtSetting.JPG.a74011f28b369e9b44106eb2f16d8d6e.JPG

 

I also bump the target level (under the "Target Settings" drop-down menu just above the "Filter Tasks" menu) up or down a little to see if there is a better "happy medium" that results in fewer PEQs being used to get the job done.  Then you can afterwards readjust the overall channel gain within the crossover to compensate for the relative channel gain difference

 

Make sure that you're also using something like psychoacoustic smoothing on the frequency response traces before running the EQ optimizer, otherwise you'll run out of PEQs trying to correct every little hill and valley.  Smoothing is needed to correct to what the ear hears, not what the eye sees.

 

Chris

 

Link to comment
Share on other sites

Sometimes (like in your case), you'll do everything you can to minimize the use of PEQs, but still run out of output channel PEQs and sometimes input PEQs, too.  In this case, you can clear all the PEQs in both the input and output channels (after you've got the delays and the relative channel gains dialed in), then run a combined sweep without PEQs, and then run the EQ optimizer again using the output PEQs first--noting where the crossover frequency is so that you're not trying to correct one driver using output PEQs when two drivers are responsible for the response, then start on the input PEQs when you run out of output channel PEQs.

 

Sometimes your eye is better than the PEQ optimizer within REW, so don't be afraid to manually edit and combine PEQs where you think it would be better.  You can immediately see the effect on the combined response within the EQ optimizer window to your manual edits (a strong reason to use REW to do the PEQ optimizations), so you can iterate based on the predicted response in the EQ facility.

 

Input PEQs in the crossover are always used in the crossover regions to correct the combined response in the crossover interference bands.

 

Chris

Link to comment
Share on other sites

1 hour ago, Chris A said:

In this instance, you need to use the input channel PEQs instead of the output channel PEQs, since you're only dealing with combined response.  If you've set the delay value in a bad spot, you'll see big dips in response where the two drivers are cancelling each other at some frequency or frequencies.  In this instance, you need to continue tweaking the delay until these SPL nulls are minimized, and the spectrogram is still as vertical as you can get it, and the excess group delay curve shows the minimum number and intensity of peaks.

 

That makes sense. So to recap PEQs on output of HF and LF till you set delay then take a combined sweep (with delay) and apply REW's EQ to the input side. Correct?

 

Now more questions

I'm on the fence between two delay numbers based on Spectrogram/GD I see merits in both (I think) and they are quite a bit apart in time. Would it be better to have the vertical line closer to zero but it dropping off at a higher point or have it slightly to the right of zero but further down the spectrum? I'll post Spectrogram and GD for both.

 

First the spectrograms so you can visually line them up. First one is 2.0ms (1.9 was very close) it has a more vertical zero line till it drops off. Second is 2.4ms It has the energy line slightly more right of zero but further down and it has a part that comes back around 400 till dropping off. The ones between 2.0 and 2.4 were visually "worse" than these two.

 

2ms800.jpg.40b6444f415ccdb57b0300d6aab529f0.jpg

 

2.4ms800.jpg.3f49097ee764aa39a9f2f26a701d1b48.jpg

 

Now the Group delays for those same two above 2.0 first then 2.4. To me 2.4 seems better because visually there seems to be more closer to zero (the 800 stuff)...but on the 2.0 there is a lot more in one chunk near zero 400-650.

 

2ms800GD.jpg.314b631568d7688f58d957db3cb5e6be.jpg

 

2.4ms800GD.jpg.44dc2988d61790ba87e5f92dd0522ee4.jpg

 

 

 

 

Link to comment
Share on other sites

20 minutes ago, rplace said:

So to recap PEQs on output of HF and LF till you set delay then take a combined sweep (with delay) and apply REW's EQ to the input side. Correct?

Yes.

23 minutes ago, rplace said:

Would it be better to have the vertical line closer to zero but it dropping off at a higher point or have it slightly to the right of zero but further down the spectrum?

You can look at the phase curve to see what's happening there.  Generally, you're at the point where listening to the results may be the deciding factor.  I'd also experiment with using a steeper high pass filter or a bigger attenuating PEQ on the HF, centered at the crossover frequency, in order to curtail that extra energy after the initial pulse from 200-800 Hz--something like the following:

 

   Gain = -15

   Frequency = 800 Hz (or slightly lower)

   BW = 0.33 (and you can iterate this value from 0.2 to 1.0 while adjusting the center frequency). 

 

The objective is to attenuate that late energy coming from your HF driver in this area to clean up the impulse response.

 

Then you might see that you don't need as much delay, i.e. try 1.25 ms delay after you use a steeper HF filter/attenuating PEQ. 

 

Chris

Link to comment
Share on other sites

2 hours ago, Chris A said:

You can look at the phase curve to see what's happening there

 

Any guidance there? I see I can wrap or unwrap the phase. Wrapped it goes from 180 to -180 along the sweep....I think. Unwrapped it shows you all the iterations or times around 360. I think this is what is talked about in changing phase multiple times....but not really sure beyond you want the phase to be in sync.

Link to comment
Share on other sites

It depends on whether or not you can get the whole unwrapped phase on the screen after zooming out on the phase vertical scale.  If you see a substantial increase in phase growth in the crossover region or a sudden jump in phase (i.e., a discontinuity), you can iterate the delay value until the phase is as flat as you can manage. 

 

I've found that phase is a "jumpy" variable, and careful delay value manipulation when getting close to an optimal delay is usually required.  It's usually a lot easier to see the effect in the spectrogram view.

 

Chris

 

 

Link to comment
Share on other sites

38 minutes ago, Chris A said:

It depends on whether or not you can get the whole unwrapped phase on the screen after zooming out on the phase vertical scale.  If you see a substantial increase in phase growth in the crossover region or a sudden jump in phase (i.e., a discontinuity), you can iterate the delay value until the phase is as flat as you can manage. 

 

I've found that phase is a "jumpy" variable, and careful delay value manipulation when getting close to an optimal delay is usually required.  It's usually a lot easier to see the effect in the spectrogram view.

 

Chris

 

 

 

 

So all just different ways of looking at the same thing or do we need to know a bit from each? I see big changed in the Spectrogram but not really in the Group Delay. I'm going to ignore phase for the time being until I can devote some time to understanding that plot better. I think I have my best Spectrogram yet. While I thought the whole idea behind a first order XO (6dB slope) was to get rid of phase problems by making the LF a 12dB slope and the HF an 18dB slope I got a much more vertical line. Am I using those terms correctly???

 

Can't seem to remove the slight shift to the left of zero with out making a huge, sharp jump left.

 

Is this good? Or at least better?

 

NOTE that this includes the 6 PEQs REW suggested on the input side. It did include some ones with boost. Prior to this all of mine on output had ONLY attenuation.

206793696_800HF18dBLF12dB2.35msDelay.jpg.33b0616b08493dac3af7b6c9fa825420.jpg

 

Corresponding GD

 

699186985_800HF18dBLF12dB2.35msDelayGD.jpg.25986c94adbc5f673ec853151eb22850.jpg

Link to comment
Share on other sites

While focusing on Spectrogram and Group Delay to set the delay then putting in the Input's PEQs I've really not paid much attention to the All SPL plot. It looks pretty good at 2dB flat from 30 to a tad beyond 1K then it really goes down. Is that ok/normal? I was successful at bringing that up with Gain, but it made a mess out of the Spectrogram once again so I changed it back.

 

627284105_800HF18dBLF12dB2.35msAllSPL.jpg.adea821f72f87157887e927fd7294fd3.jpg

 

Link to comment
Share on other sites

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

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...