Manually time-aligning subwoofer(s) to mains - how to

[neRok]

Here's an example of what a imperfectly overlapped IR can achieve;



It means the summed result would have a slight phase shift, and reduced SPL. In 2 spots it has reduced peaks, but in another spot it may have lost a bit much SPL. But this was achieved without EQ, so EQ might be applied over the top for even different results.

 

[ernestcarl]

Here's an example of what a imperfectly overlapped IR can achieve;

View attachment 308472

It means the summed result would have a slight phase shift, and reduced SPL. In 2 spots it has reduced peaks, but in another spot it may have lost a bit much SPL. But this was achieved without EQ, so EQ might be applied over the top for even different results.

The filtered IR is still useful, of course, yet gives no indication why the magnitude response slips in very specific places unless the phase response were visible simultaneously.

You also cannot visualize the filters necessary to improve the actual magnitude and phase alignment — e.g. all pass and HPF

 

[neRok]

yet gives no indication why the magnitude response slips in very specific places

It's just the way they "sum" - the wave interference. If you make "Aligned sum" and compare the IR's, you can see the resulting differences (you are best use the dBFS plot with normalised=off, because normalised=on doesn't reveal the difference). I've done that below, where purple is the imperfect overlay with weaker SPL, but you can see it's lost power in the decay, so is sort of better.

You also cannot visualize the filters necessary to improve the actual magnitude and phase alignment — e.g. all pass and HPF

It would be nice if x-over could be applied on the same screen. But you can use EQ screen to do the EQ first, and then come back to the align screen second. I agree with your suggestion of sweeping the speakers without HP, and sub without LP (only to 200Hz say), and then "simulating" the different x-over options (frequency, order, type, etc).

 

[neRok]

what happens when we have to increase levels after this procedure (e.g. to increase sub amplitude ~10db to) dial in the Harman curve or house curve of choice.

Actually, you have to be careful. Possibly it can work if after you've aligned the speakers and sub, you then EQ the full system. But if you try EQ 1 part of the system (eg, you raise the gain on the subs), you will throw off the alignment. I made the following to demonstrate;

In that screenshot I took an 80dB dirac impulse (perfect flat impulse = red line), and applied an 80Hz LP filter (green slope). But if I applied a shelf to the dirac impulse (the blue line), and then apply the same LPF, the result is very different (the brown slope). If I wanted to maintain the same slope, I would need to apply a 93Hz LPF (the purple line). And as you can see on the overlay screen, every version has a different GD plot (so the phase slope will be different).

I would have to test the order of effects when EQ'ing after alignment to be sure how it actually works. Often it seems that SPL*EQ*LP is the same as SPL*LP*EQ (eg 1*2*3=1*3*2). Regardless, you might be better off using your target curve as part of the alignment process. So instead of aligning to a flat curve, you should align with the gain already set so that the result looks like the target. Then you might be able to match the target without any additional filters.

 

[ernestcarl]

I took an 80dB dirac impulse (perfect flat impulse = red line), and applied an 80Hz LP filter (green slope).

You’re showing a HP not a LPF’d response

There’s another video by Meyer covering all pass filters as well in this webinar series which may be of interest.

 

[neRok]

You’re showing a HP not a LPF’d response

Yep, my mistake. I get H/L pass nomenclature mixed up all the time. I think of it like "I want to cut the lows", so I would have called it a low-cut filter instead of high pass. It's back-to-front in my mind.

I'm getting through this one at the moment. It's doing a good job covering at showing the phase differences.

I have seen before that L-R 4th order is good for crossover because it sums to 0dB (-6's adding together to make 0), whereas BW sums to +3dB for even order filters (-3's adding together to make +3), but I didn't register that BW odd orders sum by 3dB (-3's adding together to make 0 because they are phase shifted). So if a flat summation is the goal, there's different options. But if you want to sum the sub with higher gain (bass boost), then any filter could work to get the desired curve at the x-over area, it just depends what curve you want. And if your speakers already have a HPF, then chucking another L-R 4th over the top doesn't mean the result will look like a L-R 4th slope.

 

[neRok]

Near the end of the video quoted in my previous post (~1:18 mark) he decides that instead of delaying the speaker to match the sub, he will add an extra highpass to the speaker to delay its bass further. So for anyone that has limited delay in their DSP, that can be another option to consider. It works because both high and low pass filters manifest as group delay in the low frequencies. So actually thinking about it just now, perhaps you could low pass at like 22kHz, and that would delay the whole spectrum?

 

[ernestcarl]

Near the end of the video quoted in my previous post (~1:18 mark) he decides that instead of delaying the speaker to match the sub, he will add an extra highpass to the speaker to delay its bass further. So for anyone that has limited delay in their DSP, that can be another option to consider. It works because both high and low pass filters manifest as group delay in the low frequencies. So actually thinking about it just now, perhaps you could low pass at like 22kHz, and that would delay the whole spectrum?

By how many microseconds? Eh, I really don't get the point of it.

 

[neRok]

By how many microseconds? Eh, I really don't get the point of it.

My LP idea didn't work, so I guess I still don't understand phase fully. But HP does work. Here I just kept stacking HP BW 20Hz 8o filters, which barely touched the speakers SPL, yet constantly altered the GD. Someone earlier in the thread said they could only set up to 9ms delay. In my screenshot, the lowest phase alignment was 11.68ms. After applying a single BW HP, I got that number under 9ms. There is a new problem at 20Hz, so maybe lower HP would have worked better, or maybe an odd order BW. I don't have the problem of limited delays, so I won't bother investigating further, but this might be what some people need to do. It's probably worse than just delaying the whole speaker, but it's probably better than not being aligned.

Edit: Like the video said: delay = latency. Maybe in home theater you don't want that latency, so stacked filters could be the better option. But also for people with limited delay, they might find using the "Impulse alignment" tool/method better because they could at least find a delay that gets the waves overlapping constructively, even if it does mean the peak energy moments are offset (thus the bass is smeared).

 

[mrlawng9]

Took me a minute to absorb all the follow up.

I was able to take some time today to perform some measurements. I think I have nailed phase alignment. Here's what I did in the event it is helpful to others:

• I took multiple measurements - Left Sat, Right Sat, Left Sub, Right Sub (shooting for a dual mono sub configuration)
• At multiple locations (groups) - MLP, Front, Back, Left, and Right
• I cross-correlated each measurement group (e.g. MLP, Front, Back, Left, and Right) being sure to use the MLP measurement as the one the others get cross-correlated to
• After cross-correlation I vector averaged each group. So now I have a vector average for L SAT, R SAT, L SUB, and R SUB across all the mic positions
• To phase-align the subs, I aligned L SUB Vector Average with R SUB Vector Average at the sub / satellite crossover point. This gave me a slight 1.02ms delay on one of my subs (they are stacked on top of each).
• I created an aligned copy and called it LR SUB Aligned copy
• Next, I vector averaged L SAT Vector Average with R SAT Vector Average and called it LR SAT Vector Average
• Last, I aligned LR SAT Vector Average with LR SUB Aligned copy at the crossover point. This gave me a 4.28ms delay value for my satellites.

It sounds very well integrated and the predicted step response looked normal.

Thanks to all for their contributions and help getting me over the hump on this - it's very much appreciated.

 

[neRok]

Took me a minute to absorb all the follow up.

I was able to take some time today to perform some measurements. I think I have nailed phase alignment. Here's what I did in the event it is helpful to others:

• I took multiple measurements - Left Sat, Right Sat, Left Sub, Right Sub (shooting for a dual mono sub configuration)
• At multiple locations (groups) - MLP, Front, Back, Left, and Right
• I cross-correlated each measurement group (e.g. MLP, Front, Back, Left, and Right) being sure to use the MLP measurement as the one the others get cross-correlated to
• After cross-correlation I vector averaged each group. So now I have a vector average for L SAT, R SAT, L SUB, and R SUB across all the mic positions
• To phase-align the subs, I aligned L SUB Vector Average with R SUB Vector Average at the sub / satellite crossover point. This gave me a slight 1.02ms delay on one of my subs (they are stacked on top of each).
• I created an aligned copy and called it LR SUB Aligned copy
• Next, I vector averaged L SAT Vector Average with R SAT Vector Average and called it LR SAT Vector Average
• Last, I aligned LR SAT Vector Average with LR SUB Aligned copy at the crossover point. This gave me a 4.28ms delay value for my satellites.

It sounds very well integrated and the predicted step response looked normal.

Thanks to all for their contributions and help getting me over the hump on this - it's very much appreciated.

Did you align the lot to the main speakers though?

 

[mrlawng9]

Did you align the lot to the main speakers though?

Yep but to get a delay value that would be the same for both left and right satellite, I vector averaged the Left and Right Satellites (after they each were cross-correlated) and aligned that to the phase-aligned subs.

 

[dualazmak]

I know and follow your continuing very interesting and intensive discussions here on this thread, and I am learning a lot; I sincerely appreciate for that.

I feel more and more strongly, however, the primitive and straightforward "tone bust sound" recording (at listening position) and measurement/analysis using independent audio analysis software, Adobe Audition 3.0.1 in my case (or you may use free Audacity), would be really needed and indispensable for our "real world" better-to-best "compromised" tuning of our audio system, as I pointed in my above post #152.

I believe we need to carefully compare such "real world" primitive recorded data with REW's theoretical targets and air sound measurements.

And, of course, we always should not forget about each of our home listening rooms could never to be anechoic, and never to be physically symmetrical...

Furthermore, let me emphasize again that we always should not over-trust objective measurements and tuning; for the final decision on any of our tuning goals, we should trust our ears and brain based on your hearing and music preferences, as I wrote in the last paragraphs of my same post.

The best tuned audio system based on objective measurements would not always well fit for our (or I should say "your") music listening preferences...

 

[neRok]

I believe we need to carefully compare such "real world" primitive recorded data with REW's theoretical targets and air sound measurements.

I have not read your linked threads yet, so will today. What you describe here sounds like the same "problem" that I noticed in the thread REW SPL plot can be a bit of a liar (is only correct for 1 cycle tone). I still have not fully understood the implications of JohnPM's posts. He said;

"The responses of a system to various tone bursts is not the system's transfer function. REW's SPL plot is an accurate view of the magnitude response of the system's transfer function."​

If the magnitude of the REW's measured impulse isn't the SPL heard under all conditions, then what good is it?

 

[dualazmak]

I have not read your linked threads yet, so will today. What you describe here sounds like the same "problem" that I noticed in the thread REW SPL plot can be a bit of a liar (is only correct for 1 cycle tone). I still have not fully understood the implications of JohnPM's posts. He said;

"The responses of a system to various tone bursts is not the system's transfer function. REW's SPL plot is an accurate view of the magnitude response of the system's transfer function."​

If the magnitude of the REW's measured impulse isn't the SPL heard under all conditions, then what good is it?

OK, I too cannot fully understand the implications of JohnPM's point.

But I clearly shared in my post #152 one typical third-party example case for "advanced measurement/analysis software would not always give you proper/correct tuning information" (ref. here).

I am rather belonging to the league of "real world objective measurements by my own validated simple tools/measurements and final subjective tuning".

After somewhat intensively tested and utilized REW and/or EqualizerAPO as well as other "advanced" software tools, I always remind myself;
"do not over trust advanced DSP tools",
"trust your (my) own primitive but well validated well understandable tools/measurements",
and of course,
"trust your (my) ears and brain based on my sound and music preferences in my own listening room environments".

EDIT:
Please understand that I do not deny the great "significance" of intensive back-and-forth shuttle between objective measurement/tuning and subjective confirmation/tuning.

The final tuning decision, however, should be based on subjective tuning fit for your own listening preference, and the final "measurement" should be an objective verification/observation/record of your best tuning thus determined.

I believe this "policy" is critically important not only for time alignment tuning but also for total Fq response tuning over 20 Hz to 22 kHz.


I do hope much good lucks in your further tuning of you own audio system in your own listening environments.

 

[ernestcarl]

My LP idea didn't work, so I guess I still don't understand phase fully. But HP does work. Here I just kept stacking HP BW 20Hz 8o filters, which barely touched the speakers SPL, yet constantly altered the GD. Someone earlier in the thread said they could only set up to 9ms delay. In my screenshot, the lowest phase alignment was 11.68ms. After applying a single BW HP, I got that number under 9ms. There is a new problem at 20Hz, so maybe lower HP would have worked better, or maybe an odd order BW. I don't have the problem of limited delays, so I won't bother investigating further, but this might be what some people need to do. It's probably worse than just delaying the whole speaker, but it's probably better than not being aligned.
View attachment 308627

Edit: Like the video said: delay = latency. Maybe in home theater you don't want that latency, so stacked filters could be the better option. But also for people with limited delay, they might find using the "Impulse alignment" tool/method better because they could at least find a delay that gets the waves overlapping constructively, even if it does mean the peak energy moments are offset (thus the bass is smeared).

Subs will sometimes have a variable "phase delay" which is really just an analog all pass filter; plus, mains can be aligned with the sub with an APF as well:

*used x2 stacked 85Hz 2nd order APF on the front LR mains -- alternative would be to use a 24dB/oct HPF


Or use FIR EQ correction applied to the mains and sub:

Few more plots:


341 ms FIR processing delay used here is a bit much for real-time online streaming of video content... so I do have a different DSP preset that only adds ~47ms instead which can be used for watching Youtube, Amazon prime video stuff and the like causing no lip-sync annoyance issues.

 

[neRok]

Hi,

I'm having difficulty aligning my main speakers / center to my sub. The sub is located next to the 'R' speaker. The phase and 80 Hz filtered IR doesn't seem "too bad", but if I try the alignment tool at 80 Hz with my 'R' speaker it's waaay off.

Could someone have a look? I've attached the measurements using 1M sweeps with acoustic reference 'L'.

Best regards,

Peter

Was just taking a look at your file. It looks like your R speaker has a reflection based issue = areas of excess phase. If you turn off FDW, turn off Filtered IR, and use 1/12 No smoothing on the SPL screen, you can then see the 2 problems of GD that dip down. I believe 80Hz 1/3octave filter is from 80/6*5=67Hz to 80/3*4=107Hz, so it captures both of those problems. I noticed the issue on the Filtered IR screen because the main hump is very late, like I noticed on someone else's room in another thread.



The issue is a reflection (possibly a "room mode"), but the fact it is happening to the R only suggests there is something unsymmetrical about your room? If the modal frequencies shown are from your Room Sim, then it's showing that 72Hz is ceiling related. Perhaps you have a sloped ceiling?

You said the sub is next to the right speaker. If I compare them, I see the same issues. So it very much seems to be the room.


The phase alignment tool isn't working because REW's reported phase seems to be that of the peak energy wave, and the peak energy wave of your R and Sub is very late because of the room. So in the following screenshot I tried to align the R with the L, and it suggested ~37ms. Actually I can see from the previous screenshots that R and L were already very closely timed (~1ms difference, not worth fixing, probably just a slight difference to LP), and so when I switch the alignment tool to impulse mode, you can see the absurdity of its suggestion;



With the impulse alignment tool, if I align at 80Hz, it suggests a better ~18ms delay. But like I said before, the L and R were already basically perfectly timed, so this is not a good suggestion either. You can see this by moving the cursor to another frequency (like 56Hz I used) and "Filter IR at Cursor", and now you can see that 18ms would have made things very bad.


Even if you just look to align the Sub with L it doesn't look great;


Here's some random ideas you can try;
1) Can you change your room around?
2) Can you move the sub to the left, so it doesn't have bad problems like the right? Then I would look at having a ~120Hz HP on the right speaker, to keep its range above its bad room modes.
2.1) Maybe the left will have to have the same HP, but maybe it can be lower (like 80Hz), and xover to the sub there? I don't know how that would go.
3) If you have to keep the sub where it is, then you might be best letting the sub only do all the bass, even with its problems, because at least that way it doesn't comb filter with the left speakers bass.
3.1) Or like 2.1, maybe have the right speaker play down to 120Hz, and the sub play under 60Hz only. Then the left speaker can handle down to 60Hz?

 

[neRok]

- Precision measurement and adjustment of time alignment for speaker (SP) units: Part-1_ Precision pulse wave matching method: #493
- Precision measurement and adjustment of time alignment for speaker (SP) units: Part-2_ Energy peak matching method: #494
- Precision measurement and adjustment of time alignment for speaker (SP) units: Part-3_ Precision single sine wave matching method in 0.1 msec accuracy: #504, #507

I keep getting distracted, so I am only up to post 504. I think what you have been doing is the same as I was doing in the VBA thread by convolving the IR from REW (the system transfer function) with different tones. I have been meaning to do exactly what you have done - play a tone, and look at the recording. I think the recording should be the same as convolving, because the IR "knows all". Have you tried convolving a tone with your measured IR, to see if the results are the same? Here's a post with a video I made showing how I convolve the IR in Audacity.

In your post 504 you play a 1 cycle wave and the 2 drivers have different results. I'm not yet sure if the extra waves before and after are because of the cone, reflections, or if it's a "wave packet".

Any signal of a limited width in time or space requires many frequency components around a center frequency within a bandwidth inversely proportional to that width; even a gaussian function is considered a wave packet because its Fourier transform is a "packet" of waves of frequencies clustered around a central frequency.​

I think that means that before and after the main wave there are "build up" and "let down" waves that make up the wave packet? The question is then why does your SQ driver have 7 strong peaks and your WO driver only 5, and which ones are for the "main wave"? I actually wonder if it would be better aligned like this (I have inverted the WO), because with this delay the 2 strongest peaks are aligned;



Edit: Actually, I think I do know why the SQ driver has more peaks, and it's because of the effects of its highpass filter. The HP introduces a "3rd peak" at the start, and even a very weak 4th peak. Applying a LP on the other hand only delays (phase shifts) the peaks.
And this is where it gets tricky, because I think technically those 2 samples have the same "magnitude" of -6dB -3dB, even though one has 3 (or 4) peaks?!

Edit2: Sorry, wrong again. I applied single -12dB/oct filters, which is a butterworth filter that gives -3dB. Your have L-R filters, which are 2 BW in a row, so -6dB/oct followed by another -6dB/oct. That's what I did in the following screenshot, and it does sum up properly when phase inversion is applied (like was told is required in the previously posted video "Filters in the Real World" at the 1:00:00 mark).

Edit3: Now that I go back and look at my suggestion of inverting+delaying your WO driver, it makes sense;

 

[dualazmak]

Well, thank you for your kind and intensive insights for my post #504 on the project thread.

First of all, I am sorry for giving you a little bit of confusion of air sound start direction, kick-down-first or kick-up-first, since you look at post #504 before looking at post #493. In all of the posts #493, #495, #497, #498, #503, #507, I always use the kick-down-first sine-wave signals for excitation of SP drivers.

You would please understand the specific diagram of your interest in #504 can be seen/observed in this way;

As you can clearly see, the kick-down started at 3.340 sec time point in SQ room air sound, and in WO room air sound. Consequently, my placement of vertical guidelines and the reading of the relative delay of WO sound's bottoms and peaks are correctly shown in the above diagram.

And, please be careful enough that the horizontal (X-axis) scale is the actual real world clock-timing, and therefore the X-axis should be precisely synchronized using the red vertical marker line at 3.340 sec and the green vertical marker line at 3.352 sec. We need to "observe and discuss" within the precisely synchronized 0.012 sec (12 msec!) time window in this specific case.
(In this specific case, I time-shifted the "whole" of the recorded sound including/using the 10 kHz time zero marker monitored by tweeter so that the kick-down time point is seen at arbitrary 3.340 sec. Of course I precisely confirmed that the kick-down start timing was exactly identical, i.e. identical absolute time distance from the time-zero marker, for both of SQ sound and WO sound in the intact recorded air sound data.)
This clearly tells, therefore, that your suggested WO inverted diagram/simulation (in staggered X-scale) is not correct, is not reflecting the real world time sequence.

Also please note that I always have 10 kHz very "sharp" time-zero marker which can be "heard/recorded" by tweeter (or super-tweeter) for exact precise time-scale synchronization for all the independently recorded real world sounds. You can easily understand this "mechanism" by reading my post #493.

Now I know your intensive insights on the reasons for two-wave aftershock in SQ sound. At present, I myself do not fully know/determine such aftershock would be due to "simple physical inertia aftershock" or "wave packet" (which you are speculating first) or "-12dB/Oct LR filters I use" or "some room reflection", or any others.

As for the two-wave aftershock in SQ sound, I found your speculations would be interesting and highly possible for which I greatly appreciate.

I believe "some room reflection" can be excluded from the reason(s) judging from the clean shape and time-sequence of the aftershock pattern, and the "simple physical inertia aftershock" would be highly possible for WO sound. (Furthermore, the identical only one-wave aftershock with WO was clearly seen by 8-wave excitation; please refer to my post #493.)

And we can see, even by 500 Hz stimulation, second and third aftershocks in WO sound are well suppressed into almost negligible level thanks to the rigid-heavy sealed well damped SP cabinet and the directly-driving (no passive LCR-network) excellent amplifier YAMAHA A-S3000 having nice damping power.


In any way, the most important observation here, however, is that "we are looking at the real world air sound wave shapes and their timings", whatever would be the patterns and reasons for the aftershocks, whatever (how much) would be the relative delay of the air sound waves, and whichever would be the main bottom(s) and main peak(s).

And you would be please reminded that the ultimate purpose of "this measurement" is to quantitatively assess the relative delay of WO sound against SQ sound giving the useful and important clue for further intensive objective "time alignment tuning" which described in detail in the rest of my post #504.

And, as you may fully agree, we can easily "validate" the accuracy and reproducibility of this measurement approach by giving intentionally larger group delay in upstream DSP.

EDIT:
One of other excellent pros of this measurement approach is that we can easily "retrospectively" analyze all the "recorded" intact data having the time-zero marker anytime after the date of the measurement. Of course I have been keeping all the test tone signals and all of the intact recorded room air sound tracks in my PC-SSD for possible present (just like I looked the data again yesterday-today) and future re-analysis and re-evaluations.

 

[neRok]

And, please be careful enough that the horizontal (X-axis) scale is the actual real world clock-timing, and therefore the X-axis should be precisely synchronized using the red vertical marker line at 3.340 sec and the green vertical marker line at 3.352 sec. We need to "observe and discuss" within the precisely synchronized 0.012 sec (12 msec!) time window.
This clearly tells, therefore, that your suggested WO inverted diagram/simulation (in staggered X-scale) is not correct, is not reflecting the real world time sequence.

I shifted the graph to visualise the required delay. Delay can come from filters on the electrical side, but also from the physical driver differences (domes curve out, but cones curve in, so their surfaces are different distances away from the mic). You are presuming that your woofer has no physical delay, but I am suggesting it does.

Perhaps you could test it by doing a range limited sweep of the WO and SQ without any filters applied? So for 500Hz it might make sense to sweep 1/3 octave = sweep 416Hz to 667Hz. Then you will be able to see the physical delay without it being "mixed" with an electrical delay.

If you did that sweep, you would see the physical delay. You would also realise the waves have the same phase ("kick" in the same order). When you apply L-R 12dB/oct filters, you must invert one of the drivers to get them to sum properly (after accounting for the required physical delay). This is because of the phase shift induced by the filters, and was explained in the Meyers Audio video. You can also simulate the same thing in your audio program (like I did in Audacity), or even in REW;