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

[Ron Texas]

@ernestcarl I measured and made a wavelet spectrograph. The dotted line is smooth and flat in the crossover frequency range. Thank you for showing me this tip.

 

[Spark!]

Hello everyone, as a learning exercise I am trying to align my sub (Presonus Eris Sub 8) with my left main (Presonus Eris E3.5). I watched the JBL webinar posted here and the presented approach is similar to what @fluid suggested in #46. Following those guidelines, I measured my left channel and my sub channel from 20 Hz to 20000 Hz (using the right channel as acoustic timing reference) one at a time with no digital crossovers (the sub has a xover knob which I set to the max frequency available) and I filtered both of the signals with 100 Hz 1/3 octave filter in the "Filtered IR" section of REW, since I would like to apply later a 100 Hz LR crossover filter. Now, if I try to align the highest peaks of the two filtered signals, I get a huge delay of about 38 ms that I need to apply to the sub, as you can see from here:

The real distance between sub and left main is about 1 meter. As you can see from the .mdat file, my room is not treated and has a lot of big peaks and notches in the frequency response. Did I make some mistakes during the procedure? This is my first time using REW.

 

[fluid]

I had a look at your mdat I think the fact that both measurements have a lot of output above and below the crossover point is causing the filtered IR view to be as you show and not be very helpful.

If you Use the Alignment tool to align phase between the two at 100Hz then it gives a figure of 7.39ms or anywhere up about 11ms depending on which options you choose. That seems a much more reasonable figure.

 

[Spark!]

Thanks for the reply. I tried to compare the frequency response of the Alignment Tool (~7 ms of delay, blue line) with the one with ~38 ms of delay and it seems that the latter is a little bit higher near the crossover point. Should I still use the one suggested by REW?

 

[fluid]

I would go with the shorter one as it seems that there is a room issue with your left speaker that is causing a delay in the energy at 80 to 100Hz which is what the wavelet is showing

 

[ernestcarl]

I agree with Fluid to go with ~7.3+ ms delay for the mains. Maybe reduce the sub's gain if you perceive it as too strong.

Phase traces from the impulses you provided (not particularly pristine) are difficult to read without applying FDW:

One can generate an equivalent aligned sum response by adding -7.3ms delay to the sub in REW's alignment tool.

Wavelet spectrogram

1/6 resolution, linear peak 100% scale, Normalized


All you need to do is confirm with actual measurements.

 

[Spark!]

Thanks for your help.
I redid all the measurements in the same listening spot and I have measured the result of the alignment after applying the delay suggested by REW.

In the updated .mdat you can find:
1) RR left new: the new left channel measure;
2) RR sub new: the new sub measure;
3) Aligned sum from AL: the simulated aligned sum from the Alignment Tool, obtained after applying 7 cycle FDW on 1) and 2). The delay found by REW is 7.95 ms (to apply to the left channel) at 100 Hz;
4) RR left dly 7.95ms: the left channel measure after applying the 7.95 ms delay;
5) RR left dly 7.95ms + sub: the final measure with left aligned and sub.

If I plot the wavelet spectrogram of the final measure 5) I have a strange behavior at about 130 Hz which I can't explain:

Have I made any mistakes during the procedure?

 

[ernestcarl]

Thanks for your help.
I redid all the measurements in the same listening spot and I have measured the result of the alignment after applying the delay suggested by REW.

In the updated .mdat you can find:
1) RR left new: the new left channel measure;
2) RR sub new: the new sub measure;
3) Aligned sum from AL: the simulated aligned sum from the Alignment Tool, obtained after applying 7 cycle FDW on 1) and 2). The delay found by REW is 7.95 ms (to apply to the left channel) at 100 Hz;
4) RR left dly 7.95ms: the left channel measure after applying the 7.95 ms delay;
5) RR left dly 7.95ms + sub: the final measure with left aligned and sub.

If I plot the wavelet spectrogram of the final measure 5) I have a strange behavior at about 130 Hz which I can't explain:
View attachment 136780

Have I made any mistakes during the procedure?

When I generate the 'simulated' aligned sum, I usually disable FDW and change smoothing back to 1/48.

And if your sub is close to symmetrical in distance between left and right monitors, I'd prefer to measure the left monitor minus sub with the same left monitor as reference so that the impulse peak would be set close to t=0 by default. But it's fine setting it to the right monitor because one can shift the IR peak after summing the responses later on e.g. you could apply FDW 7 to the 'aligned sum' and in the SPL & Phase view click on "Estimate IR delay" > Shift IR.

Oh, yeah, your wavelet graph looks correct. It's only the new set of measured responses that has changed from the previous one. Other automated DRC software would take multiple measurements to create an average. We are only summing single measurements here taken at different conditions so small random variations are sort of expected. And as you probably are aware, it is largely the room that is causing havoc with all the numerous reflection and cancellation issues.

*I generated a couple more for comparison

 

[Daverz]

Tried the measurement procedure in the OP, and it shows that the needed delay is just a bit larger than the 9 ms maximum my nanoDIGI allows.

The sub (measured to the driver dustcap) is only about 1 meter further away from the listening position, so the remaining difference must be due to using two different model DACs for the 3 channels of the 2.1 system (and maybe the sub, an SVS SB1000 Pro, is adding some delay). I'll try swapping the DACs later (they are an Auralic Vega and a Topping E30).

 

[fluid]

If I plot the wavelet spectrogram of the final measure 5) I have a strange behaviour at about 130 Hz which I can't explain:

The behaviour isn't strange it is what happens when there is boundary interference. There is a null in time which then goes away as other parts of the room fill it in so when you look at a long time frequency plot it looks good but when you look at the spectrogram or use a Frequency dependent window it looks terrible and lumpy.

If you can try and move your left speaker position it may only take 10 to 20cm of movement for it to be in a much better position for interaction with the room in this region. If your room is rectangular then you can use the room simulator in REW to virtually move the speakers around in the room and see where you get the best response. It exaggerates the peaks and nulls but look for where you get the least deep nulls in the bass, peaks can be equalised but nulls are likely there to stay without drastic action. This is all based on distances so it is a surprisingly accurate prediction.

 

[fluid]

The sub (measured to the driver dustcap) is only about 1 meter further away from the listening position, so the remaining difference must be due to using two different model DACs for the 3 channels of the 2.1 system (and maybe the sub, an SVS SB1000 Pro, is adding some delay).

The difference in propagation delays due to DAC's is tiny and changing them won't help you. The lowpass filters can add a lot of apparent delay over and above the physical distances. Trying using the alignment tool or filtering the impulses that will give you a much better idea if the delay number is correct.

 

[Daverz]

The difference in propagation delays due to DAC's is tiny and changing them won't help you. The lowpass filters can add a lot of apparent delay over and above the physical distances. Trying using the alignment tool or filtering the impulses that will give you a much better idea if the delay number is correct.

Yeah, several milliseconds of latency does seem odd from a DAC. Which lowpass filters do you mean? These measurements are made with the XO bypassed in the nanoDIGI and LFE setting used on the sub (no lowpass).

If I use Estimate IR Delay on each impulse response, I get a difference between the two estimated delays that is closer to the physical distance between mains and subs. But at this point I have to admit I don't know how to use this method.

 

[fluid]

I was meaning the crossover filters that will be needed unless they are linear phase they can add a lot of apparent delay, if they were switched off then they wouldn't be the cause. The best delay setting will be found with the crossovers in place. The delay could be room interference as in the last example. Look at the spectrogram to see what is going on it's very hard to tell much at all from the graph you posted, posting a link to an mdat makes it easier for others to see what you have and offer their thoughts.

 

[Daverz]

I was meaning the crossover filters that will be needed unless they are linear phase they can add a lot of apparent delay, if they were switched off then they wouldn't be the cause. The best delay setting will be found with the crossovers in place. The delay could be room interference as in the last example. Look at the spectrogram to see what is going on it's very hard to tell much at all from the graph you posted, posting a link to an mdat makes it easier for others to see what you have and offer their thoughts.

Here are the mdat files:

https://drive.google.com/file/d/1ktfSEJax__ib-ySQOL1_NfZ8-r5QdJSO/view?usp=sharing

 

[fluid]

You can get a reasonable phase alignment between 75 and 100Hz with 6.92ms on the main speakers if a 6 cycle FDW is used. If the alignment point is moved up to 100Hz with no windowing the time blows out to 17ms due to the increased room interaction. The energy peak is delayed in the spectrogram at some point both are suffering boundary interference but at different frequencies.

 

[Spark!]

Thank you very much @fluid and @ernestcarl, I learned a lot of new things! Do you have any recommended reading about all of these topics?

In the mean time I have another doubt with the above spectrogram plots below 50 Hz: as you can see, the peak energy arrives ~60 ms later than the energy above 50 Hz. Reading about the wavelet spectrogram, I saw that in an ideal plot the peak energy needs to arrive at about the same time for all the frequencies, is it correct? In my <50 Hz region the sub is 'soloist', so this 60 ms delay is due to a not ideal interaction between the room and the sub? Are our ears sensitive to these different delays in the spectrum? I also tried to do some 'blind' tests between the aligned system and the not aligned one but I couldn't hear any differences, maybe my ears are just bad (or maybe I didn't test it with the appropriate audio materials)

 

[Daverz]

You can get a reasonable phase alignment between 75 and 100Hz with 6.92ms on the main speakers if a 6 cycle FDW is used. If the alignment point is moved up to 100Hz with no windowing the time blows out to 17ms due to the increased room interaction. The energy peak is delayed in the spectrogram at some point both are suffering boundary interference but at different frequencies.

I should have moved a leather chair and a coffee table out of the way.

It turns out a lot of the delay is due to the latency of the subwoofer:

https://www.audiosciencereview.com/forum/index.php?threads/svs-subwoofers.17378/#post-563096

 

[dasdoing]

imo people are overcomplicating.
play around with rough delay jumps until you get excess group delay close to zero (use dirac impulse for audible pre-delays).
from there play around with finer delay steps until you get the flattest in the crossover region

 

[ernestcarl]

Thank you very much @fluid and @ernestcarl, I learned a lot of new things! Do you have any recommended reading about all of these topics?

In the mean time I have another doubt with the above spectrogram plots below 50 Hz: as you can see, the peak energy arrives ~60 ms later than the energy above 50 Hz. Reading about the wavelet spectrogram, I saw that in an ideal plot the peak energy needs to arrive at about the same time for all the frequencies, is it correct? In my <50 Hz region the sub is 'soloist', so this 60 ms delay is due to a not ideal interaction between the room and the sub? Are our ears sensitive to these different delays in the spectrum? I also tried to do some 'blind' tests between the aligned system and the not aligned one but I couldn't hear any differences, maybe my ears are just bad (or maybe I didn't test it with the appropriate audio materials)

Could be caused by one or more number of things e.g. room boundary interference, or maybe it could also partly be delay from the sub's port (?) as well -- if it has one, though, I'm just throwing out gueses here.

I'm not really sure if there is an all-in-one authoritative source reference to point to... I mean, there are several books about small room acoustics, speaker measurements etc. But, much of what I know about REW measurements is from what I've read in the REW help documentation, as well as various threads in this and other forums. Oh, and lots, and lots of experimentation in the wee hours of the night to ward of boredom.

BTW, with regards to the imperfect wavelet spectrogram... an ideal line is just that: an ideal.

I could probably only count with one or two hands the number of times I've seen 'idealish' spectrogram plots. In a real room with very real constraints, you're bound to see all kinds of weird imperfections.

Theoretically, though, you might be able to 'audibly' gain improvements with better measurements using FIR convolution e.g. Audiolense, rePhase and the like. However, it still isn't going to magically turn a 'cheapish' under $500 sub into a 'better' $1000 sub -- or turn a really bad room, into great room e.g. is the decay good or bad?

In terms of the question "is it even possible to hear the difference with better systems? ... with even less group delay in the sub bass?" It is possible (I myself thought I couldn't hear gross differences -- I actually can), but you would need to find systems that are objectively better (with actual measurements as proof) than what you currently already have -- and really listen to them. Or if you are able to use advanced convolution to significantly reduce any excess group delay -- you could then do an AB or ABX tests to hear for yourself. But again, convolution to correct stuff will not necessarily make a cheap sub sound like a more expensive and (objectively better) measuring one -- esp. with regards to SPL headroom. And, as I may have mentioned more than once now, room acoustics and getting the ideal placement perfectly right is really the greater obstacle for most people.

 

[ernestcarl]

imo people are overcomplicating.
play around with rough delay jumps until you get excess group delay close to zero (use dirac impulse for audible pre-delays).
from there play around with finer delay steps until you get the flattest in the crossover region

Probably, but I'd like to think spreading this type of information out there is far more beneficial than not.