Using REW to Determine Time Delays Between Drivers

[PrestonTom]

8 hours ago, Chris A said:

 

".........

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

 Chris, is it possible that the "reflector" is not near the diaphragm, rather it might be near the measurement microphone (perhaps some structure or hardware that is holding the microphone)? I am just guessing, but I have made a similar mistake in the past.

-Tom

[Chris A]

Yes, something pretty reflective around the microphone could do that,  but I discounted that because of the characteristic distance (about .9 inches quarter wavelength boundary reflection distance).  Most microphones stick out in front of the microphone stand clamp by at least twice that distance, and I've not placed a microphone that close to the floor before--it's better to use a PZM or boundary microphone for that duty.

 

Chris

[ClaudeJ1]

On 9/24/2019 at 3:15 PM, Chris A said:

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

Great work Chris. If posted here as a PDF, this will be a great contribution to our "Audio Society" here on this Forum. I encourage you to continue.

[rplace]

On ‎4‎/‎6‎/‎2019 at 7:09 AM, Chris A said:

The step response (the plot is found under the Impulse Response plot window) should look something like this:

 

 

and not like this:

 

 

 

@Chris A, can you speak to this a little bit and how impact the desired changes? What is it or what does it represent? Do you use Psychoacoustic smoothing for all visual analysis? I noticed in some of the initial posts you have 1/48 smoothing in the title. In the two examples above where do the -0.1% and 0.8% numbers come from? Suggestions on limits to set?

[Chris A]

2 hours ago, rplace said:

...can you speak to this a little bit and how impact the desired changes? What is it or what does it represent?

 

I think I'll post Joseph D'Appolito's comments on this first, since he did such a good job in explaining the usefulness of step response (instead of using impulse response):

 

Quote

STEP RESPONSE

 

(Taken from Testing Loudspeakers: Which Measurements Matter, Part 2)

 

By Joseph D’Appolito, audioXpress 2009

 

 

...Up to this point we have looked at loudspeaker performance solely in the frequency domain. Let’s turn now to the time domain for additional performance insight. We could examine the impulse response in more detail, but it is not easily interpreted. It is dominated by the tweeter response in the first few milliseconds. It doesn’t tell us much about the woofer, or the midrange if there is one, because all the low-frequency information is in the impulse response tail, which is at a very low signal level. The step response is a much more useful tool.

 

The step input is a signal that rises instantaneously from zero to a fixed level. This is basically a DC input starting at time zero. Mathematically, the step response is the time integral of the impulse response.

 

Figure 17 shows the response of an ideal loudspeaker to a step input. Loudspeakers are high-pass devices that cannot produce a static (i.e., DC) acoustic output. Therefore, the step response must drop below zero for a sufficient time to produce a net output of zero over time. The ideal step response is an exponentially decaying cosine wave oscillating at the fundamental resonant frequency of the loudspeaker.

 

 

FIGURE 17: Step input and ideal step response.

 

Figure 18 shows the step response for my example loudspeaker on an expanded time scale. The oscillatory portion of the response is not shown. This plot is actually a combination of two step responses: the initial sharp rise of the tweeter followed by the much slower broader rise of the woofer. This is shown more clearly in Fig. 19, where the tweeter and woofer step responses are plotted separately.

 

FIGURE 18: Step response of example loudspeaker.

FIGURE 19: Individual driver step responses.

 

What can you tell from these plots? First, you see that both the tweeter and woofer are connected with positive polarity. Both initially rise in the positive direction. Next you see a smooth hand-off from the tweeter to the woofer at roughly 3.1msec. This speaks well of the crossover design. Finally, from Fig. 18 you see that the speaker is not time coherent. Comparing rise times, the woofer is approximately 250 µs behind the tweeter.

 

If you reverse the polarity of the tweeter, you get the step response shown in Fig. 20. There is now no longer a smooth transition from the tweeter to the woofer. The frequency response (Fig. 21) shows a null in the crossover region of about 12dB due to the tweeter polarity inversion. The response curves shown have been 1/6th octave smoothed. The raw curve shows a notch of greater than 20dB, which is a strong indication that the drivers are in-phase at crossover. (I will discuss this condition in more detail in the section on phase response.)

 

 

FIGURE 20: Step response-tweeter polarity reversed.

 

 

FIGURE 21: Example loudspeaker response with reversed tweeter polarity.

 

Determining driver polarity can be of great value in home theater setups. For example, you may be using a center channel speaker from a different manufacturer than those of the right and left channels. If the center channel tweeter is connected out of phase to get flat frequency response while the left and right channel speakers use in-phase tweeters, you will degrade the imaging of the full system. The effect with woofers can be even more dramatic...

 

When using DSP crossovers, there is an opportunity to move the woofer response to lay on top of the tweeter response (which is highly desirable and audible if you can achieve it), but that discussion gets into "zero phase shift crossover filters".  Suffice it to say that the use of the right sort of crossover filters and delay can erase the two separate "humps" in the step response, but note that there are limitations of how far the woofer/bass bin horn mouth(s) can be in the vertical plane from the tweeter or midrange drivers/horn mouths.

 

2 hours ago, rplace said:

Do you use Psychoacoustic smoothing for all visual analysis?

Yes.  There is a good reason for this, and it has to do with how the human hearing system works.  It's much better to use how the ears work in our loudspeaker corrections than to use our eyes and guess if this is what we hear also.

 

2 hours ago, rplace said:

I noticed in some of the initial posts you have 1/48 smoothing in the title.

There are some instances where it is interesting to look at 1/48th octave to see the underlying structure of the problem, but in general when making changes to the PEQs, polarity, delay, and crossover filter types, it's always better to use psychoacoustic smoothing to do it.  This is especially true within the REW EQ facility, which tries to optimize the PEQs to flatten response, and failing to use psychoacoustic smoothing will always result in a less-than-desirable set of PEQs that are generated.

 

2 hours ago, rplace said:

In the two examples above where do the -0.1% and 0.8% numbers come from? Suggestions on limits to set?

That's where my cursor was when the plots were taken within REW.  Since I suppressed the cursor within REW when I plotted the curves, you don't see the cursor, but the values of where the cursor was are still there (unfortunately).

 

Chris

[rplace]

On ‎10‎/‎7‎/‎2019 at 4:10 PM, Chris A said:

I think I'll post Joseph D'Appolito's comments on this first, since he did such a good job in explaining the usefulness of step response (instead of using impulse response):

 

That was very helpful, thanks. I've been concentrating on the Impulse now and finally made some progress. I have some very different XO points and corresponding Phase/SPL/GD/SGrams. While I like the sound of them all I for sure have one I prefer. However, I want to make sure my judgement is not clouded by looking at graphs and guessing at which one I think should sound better. Below I'll have 3 different settings with 4 graphs each. I'd be curious what you all think should sound better purely on a visual basis, looking at the plots. I'll tell you which one I like better after I get some feedback.

 

Some have really nice Spectrograms and GDs, others have better Phase (I think) only one has good impulse.

 

QUESTION: If you are going to have turn over 8 times is it better to have them equally spread out or have very long plots in the middle of the spectrum and quick changes at one end, the other or both? I seem to recall that phase at higher frequencies is not as hard on the ears....so I would think having the phase changes spread out over the middle would be better. Am I right?

 

Here are 4 plots for September 29th. I say good GD, Good Spectrogram, terrible impulse and not sure about Phase, thus my question above.

 

 

 

 

 

 

[rplace]

Now a day later on September 30. Similar Spectrogram, Worse GD than above, better phase I think, but not really sure. I thin the 3 cycles from 60-2.5K are better but not by much, are they? And again terrible impulse

 

 

 

 

 

 

[rplace]

Finally this morning. I got some good impulse, the only one. The Spectrogram suffers a lot, the GD not so great at that one big spike but overall decent. The Phase in my mind is the best because of the long stretches between 70 and 2.5K, is it?

 

As for the phase on this one, nothing I can do will get rid of the quick zig-zag at 5-60K or ~2.7K. Attenuating that SPL hump at 8-9K really messes up the phase and GD from 8K out to 15K.

 

 

 

 

[Chris A]

11 minutes ago, rplace said:

While I like the sound of them all I for sure have one I prefer.

Which one is that?  Can you identify the crossover frequency and relative placement of the HF and bass bins?

 

15 minutes ago, rplace said:

QUESTION: If you are going to have turn over 8 times is it better to have them [the phase trace 180 degree jumps] equally spread out or have very long plots in the middle of the spectrum and quick changes at one end, the other or both?

Generally, the hearing system is most sensitive to midrange frequencies from ~400-500 Hz to 3-5 kHz, so having a smooth group delay curve (the derivative of the phase curve) is nice to have.  The ear can hear group delay growth, but is fairly tolerant of GD growth above 3-5 kHz and below 100-200 Hz.  Having a flat phase curve is audibly a very nice thing to have, but if you can't achieve that, having a fairly smooth group delay curve with smoothly rising phase curve is the next best thing, in my experience. 

 

If you look at the phase/group delay curves of Klipsch Heritage products (Khorns, Belles, La Scalas, Cornwalls, and Heresies), you'll see really big jumps in the GD curve and huge chasms in the phase curve at the crossover interference bands--where the drivers/horns are not time-aligned and the phase jumps by 720 degrees or more.  These are generally centered at 400 (Khorn and La Scala)-700 Hz (Heresy) and 5-6 kHz (all midrange-tweeter crossovers, some with extremely wide interference bands).  So if PWK found that he could get away with these big phase/group delay discontinuities, he chose those frequencies to do it.  The GD spikes that you have on your third-day trial are minuscule compared to the Heritage line GD spikes.  If it were me, I'd try to push the lower frequency of the phase/GD jumps downward toward 200 Hz (if possible) since 440 Hz is in the middle of the midrange band (tuning fork frequency). 

 

Beyond that, I think that you're presently coaxing out the best compromise that you'll be able to get without using FIR filtering.

 

Chris

[rplace]

On ‎10‎/‎11‎/‎2019 at 11:47 AM, Chris A said:

Which one is that?  Can you identify the crossover frequency and relative placement of the HF and bass bins?

 

Prior to the three above I had been doing many iterations of a 800 XO with Bessel 24dB on the HF and Bessel 18dB on the LF. I went back and listened to it this morning and still like it, but I think I like at least two of the 3 above better than the 800. I've learned some things in the past 2-3 weeks so I need to go back and mess with the 800 settings a bit to see if I can optimize a bit more.

 

As for the three above I posted last Friday here are the details:

• First Group, blue plots, September 29th are 250 XO Bessel 6dB on the HF and 12dB on the LF
• 2nd group, orange plots, September 30th are 400 XO  Bessel 6dB on the HF and 12dB on the LF
• 3rd group, green plots, October 11th are 200 XO Bessel 6dB on the HF and 12dB on the LF. Additionally I put a much higher delay on the HF section and inverted the phase. I also moved the mouth of the large Oris 150 Horn to be more inline with the driver in the LF's H-Frame. See pictures and explanation below.

 

I was reading a bit (can't remember the post) and looking at pictures of 402 Horns moved forward and overhanging the bass bin of a 2-way Jubilee set up. I got the impression that having the mouth of the horn closer to the exit of the bass bins helped. So I put the front edge of my horn inline with the bass bins driver. Kind of hard to photograph, but if you can imagine prior to this the front edge of the Oris horn was just slightly behind the zebrawood box of the bass bin. Now it is on the same vertical plane as the actual 15" bass driver. The open baffle H-Frames I'm using have the front of the driver centered front to back in the bass bin. That is, there is 9 inches of wood both in front of and behind the mounting surface. If you can conceptualize the capitol letter "H" as if you were looking down on it from above the left/right vertical lines of the H are the sides of the speaker and the horizontal part connecting the left/right is my driver's front edge.

 

Speaker from above and to the left. Might not look like the horn is in the middle of the box but it is.

 

Originally they were offset more like below.

 

Construction of H frame so you can see how the driver is centered in the box on it's front edge.

 

 

I'm now wondering if the horn should be at the front of the bass bin's box, Not centered on the driver. I think the point of the Open Baffel H is the provide some sort of pressure/wave/loading....not really sure of the term I'm looking for but different from a traditional open baffle to provide some sort of direction for the sound waves I think. If that is the case the front edge would be similar to the exit of the Jube's bass bin, no?

 

At any rate I like the 3rd one the best. But now that I have the horn/driver lined up differently I think I need to measure/adjust the first two....and my 800 XO from a few weeks back.

 

Thoughts?

 

[Chris A]

1 hour ago, rplace said:

I'm now wondering if the horn should be at the front of the bass bin's box, Not centered on the driver.

It doesn't matter much as long as the difference in front wave arrival times is compensated by delay.  If the HF driver is a dipole radiator, then having its back wave pretty much lined up with the bass bin back wave will be to your advantage in terms of midbass/midrange "apparent soundstage depth". 

 

1 hour ago, rplace said:

At any rate I like the 3rd one the best.

The impulse response looks the best to my eyes. 

 

1 hour ago, rplace said:

Thoughts?

The 800 crossover frequency will avoid the stretching of the midrange impulse energy (spectrogram plot) and potentially clean up the spectrogram a bit to increase the crispness of the transients. 

 

The reason why I place my K-402s slightly in front of the Jub bass bins is so that I can lower then K-402 centerline to be closer to the KPT-KHJ-LF bass bin centroid ("W" section bass bin with dual mouths).  Lowering the K-402 allow me to significantly flatten the the resulting phase plot and the group delay plot.

 

Chris

[Rudy81]

@rplace, I just now stumbled onto this thread. Where are you placing your mic for these measurements? @Chris A, where is the recommended mic placement? Aimed at the ceiling or at the drivers?

 

I may have to dive back into aligning my speakers.  Just for grins I picked up a set of Fostex Fe206En drivers.  Very inexpensive and they have great driver specs for the Oris horn. So far I really like the sound.....and you can't beat the price. 

[Chris A]

9 minutes ago, Rudy81 said:

@rplace, I just now stumbled onto this thread. Where are you placing your mic for these measurements? @Chris A, where is the recommended mic placement? Aimed at the ceiling or at the drivers?

 

I may have to dive back into aligning my speakers.  Just for grins I picked up a set of Fostex Fe206En drivers.  Very inexpensive and they have great driver specs for the Oris horn. So far I really like the sound.....and you can't beat the price. 

...One metre in front of the frontal plane of the loudspeaker, on axis with the HF/midrange horn (very slightly off-axis by an inch or two to avoid dead-on mouth bounce issues from the horn).  Put down a lot of absorption material the width of the loudspeaker all the way to the microphone, and make it thick absorption--the thicker/fuzzier, the better (if you have excess absorption panels available, just use those).  Keep all other acoustically reflective objects away from the loudspeakers and microphone, with a clean area of at 3-4 feet all around if you can.  These actions will significantly clean up your phase and group delay measurements so that you can correctly set good time delays.

 

Hope things are going well for you and yours, Rudy.

 

Chris

[Rudy81]

Chris.

All well sir, thank you.....just working and looking forward to retirement in a few years.  You know I'm bored when I start delving back into speaker time alignment threads.

 

I'll add a picture of my latest speaker build and poly installation.  I recently also changed my wall treatments to poly's and they are working out great. 

 

Are you saying that I should make a 'tunnel' type of absorbent surround from the speaker to the mic? 

 

 

[Rudy81]

Chris, more questions for when you have time.

 

In step 1 you talk about flattening the SPLs.  I have never done that and just let Audyssey Pro handle that job.  I know from taking measurements outdoors that my LF and HF drivers are actually pretty darn flat if not affected by room boundaries.  I am wondering if I should flatten the driver responses as you indicate, and leaving those settings in my crossover permanently....thus leaving less work for Audyssey to do. Opinion?

 

Next, I wonder if the readings should be taken at the 1 meter point vs. the listening position.  I am curious about this since it is the room that seems to affect the response curve as compared to a very flat response curve when readings are taken outdoors.  Certainly not questioning your expertise, but wondering which would be most effective, and what is the downside to an LP measurement. Once the timings are set you will be listening from much farther away than the measured 1 meter.  This topic has always confused me.....

 

I use an Ashly xover, which not an option listed in REW, so I will need to find a similar xover to the Ashly.

 

Previously, I used this method to align my drivers: https://www.hometheatershack.com/forums/rew-forum/38617-aligning-driver-phase-rew-v5-example.html

[etc6849]

On 4/6/2019 at 3:17 PM, Chris A said:

I set the microphone at listening height--just below the top of  the bass bin, one metre away from the front baffle of the Jub bass bin--in order to minimize in-room reflections which show up in the group delay plot. For loudspeakers that are over 6 feet tall, you will have to move back to the listening position and use as much absorption and time gating as you can to suppress first reflections obliterating the phase/group delay data.

 

I have a small room, so I just did all measurements at the listening position, with the mic pointed toward the ceiling and the tip at ear height.

 

For the time alignment, I also did something different.  My thinking was for time alignment, the primary concern would be around the crossover region.  So I performed band limited measurements for the drivers over a reasonable overlapping region, then compared the impulse/ETC plots.

 

For example, to align a sub to a woofer with a 110Hz crossover, I would do measurements covering 20-200Hz for both my sub and the woofer.  I then compare the two ETC plots, and get the time delay that way.  I did similar method for the woofer and midrange, etc...

 

What is interesting is that this seemed to produce decent results after looking at the peak energy curve on the spectrogram (something I didn't think about using for time alignment).

 

[etc6849]

@Chris A  My step response looks very different than yours.  I think mine might look more ideal, but honestly, I have never looked at my step response so I am curious what you think.

 

 

I have never used psy smoothing either, but seems like a good idea so here is what things look like for SPL and phase.  Note I intentionally tilted levels when I set them up, and I'm not using any PEQ or Dirac.  Perhaps if I point the mic at the speaker as you suggest it would look nicer.  I'm definitely capturing more of the room effects the way I've been measuring.  One thing that is confusing is the phase shift that appears at 6kHz and 13kHz when I unwrap phase.  This only appears in my front two speakers and not my rear speakers.

 

 

 

[Rudy81]

1 hour ago, etc6849 said:

 

I have a small room, so I just did all measurements at the listening position, with the mic pointed toward the ceiling and the tip at ear height.

 

For the time alignment, I also did something different.  My thinking was for time alignment, the primary concern would be around the crossover region.  So I performed band limited measurements for the drivers over a reasonable overlapping region, then compared the impulse/ETC plots.

 

For example, to align a sub to a woofer with a 110Hz crossover, I would do measurements covering 20-200Hz for both my sub and the woofer.  I then compare the two ETC plots, and get the time delay that way.  I did similar method for the woofer and midrange, etc...

 

What is interesting is that this seemed to produce decent results after looking at the peak energy curve on the spectrogram (something I didn't think about using for time alignment).

 

 

Your technique is more of what I have always done.  However, I seem to recall that once you add the xovers it changes the phase and the timing.  I must be dense because I have read countless articles and books on the subject and it is still confusing on 'how to do it'.  That's why I am very interested in Chris' cookbook methodology.

 

[Rudy81]

Last night, just coincidentally, I worked on trying to 'time' align my 4 subwoofers.  I have 4 DIY identical subs placed in the corners of the room.  I ran sweeps for all four and decided that both rear subs needed a 3.93ms. delay.  Measurement was made with the mic at ear height, room sweet spot, with mic pointing up at the ceiling.  It is a calibrated mic BTW.

 

My added delays made sense from the viewpoint that my LP is closer to the back subs than the front subs.  Based on Chris' idea, I ran the spectrogram plots and here are the results.  As you can see, the delay helped things quite a bit.  But....I think things can be improved.  In pondering how to do that, I looked at the phase plots and it is clear that the rear subs, with the delay are still a bit out of phase with the front subs.  Tomorrow I will work on aligning the phases and see if that improves the SPL plots. 

 

Clearly there is something to be said for aligning drivers, or subs in this case. I am curious to see if phase aligning all four subs will improve the SPL between 100-200hz.  I am motivated to really getting a handle on this time/phase alignment stuff. 

 

 

[etc6849]

1 hour ago, Rudy81 said:

 

Your technique is more of what I have always done.  However, I seem to recall that once you add the xovers it changes the phase and the timing.  I must be dense because I have read countless articles and books on the subject and it is still confusing on 'how to do it'.  That's why I am very interested in Chris' cookbook methodology.

 

 

I do level, then set the crossovers, then do time alignment as I described.  Things should be done in this order.  Even setting level later will impact the crossover point (it sounds like you already are aware of this, but I'm putting it here for others).  The slope at the crossover region is seldom a brick wall, so there is plenty of overlap for the time alignment method I describe.

 

Technically, FIR filters don't impact phase (using them for woofer to midrange and midrange to tweeter), so really I could go back and adjust things out of order, but for the low order bessel filter I use for my woofers and subs I definitely can't do that obviously.