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

[pozz]

Hey sorry to resurrect an old thread, but I'm at my wits end trying to get this figured out. I've made some great progress over the last week by researching, but I'm still having trouble figuring out if the how to make the correction and if it's an improvement or not. I'm trying to align my sub (SVS SB 2000) and mains (Genelec 8030s) for a small home studio.

I followed all the steps in the first post exactly by the OP and now have FDW plots of step and SPL/Phase. I've attached some plots, but the phase plots all look very offset from each other. All alignment has been to the L monitor with the mic at my listening position (UMIK-1).

PHASE
View attachment 133719


SPL
View attachment 133720

STEP
View attachment 133721


I watched a JBL video (See below) and the presenter mentioned we should be aligning the INITAL START of the peaks if we have dissimilar frequency response - not aligning the peaks due to cancellation problems that can be caused from that. Does this limited sweep and windowing account for this? I'm concerned because my sub can go to about 15 hz, and my mains only go to about 50 hz.

If anyone could give me a nudge in the right direction this weekend, I would be forever grateful. I'll buy you a virtual beer Cheers!

Using IR will be hit or miss (you might get lucky because of your room, speaker/sub placement) because it is a linear presentation of frequency content, not the logarithmic presentation we're used to for the frequency scale. That means that the IR will be heavily biased toward higher frequencies.

If the measured impulse contains energy at higher frequencies, as when you're measuring your mains, its peak will appear earlier than peak of the impulse measured for the sub. It's the energy content between high and low frequencies that affects the shape and positioning of the peaks. So it would be wise to use a frequency restricted signal for the measurement, ideally a range where the sub and mains are outputting the same amplitudes, if you want to align with peak IR, but this is of course won't work if you've already applied high pass and low pass filters. Using the very onset of the IR to align may work, but it's hard to see it for low frequencies.

Next thing to note is that time alignment applies to only one position when subs and mains are in physically different places in the room. If you move, the relative arrival times change. If you have multiple seats, the relative arrival times for each will be different.

So the thing to do is phase align rather than time align, especially if you have multiple subs. That's as far as I've gotten. If the phase alignment is within x range of degrees across all seats, you will get y range of dB difference at the crossover, theoretically. The lower the better, obviously. So you won't have perfect phase alignment ever. Just a range which you have to deem acceptable by looking at the frequency response.

I don't know all the relative strengths and weaknesses of using a frequency dependent window, but using one does bias measurements toward initial arrival of the wavefront rather than the steady state response.

@KSTR do you have advice here? He has said before you are in the speaker designer's shoes when dealing with these problems.

Right now I'm of the opinion that the best way to do phase alignment is to use pseudorandom pink noise with the microphone placed at the best seat, roughly equidistance from mains and subs. Then using a program like Open Sound Meter or SMAART to measure the real time coherence and phase and make phase adjustments on the fly. Goal is to match the slope of each trace.

 

[abdo123]

After measuring main and the sub you should apply FDW of 6 cycles to eliminate reflections that blurr the time domain measurements but you shouldn't use "Estimate IR delay".

Can someone please explain to me how to do this particular step in REW?

 

[JStewart]

You can use the IR Windows button.

https://www.roomeqwizard.com/help/help_en-GB/html/impulseresponse.html#top

 

[abdo123]

You can use the IR Windows button.

https://www.roomeqwizard.com/help/help_en-GB/html/impulseresponse.html#top

thank you!

a little bit on topic now, for people running two subwoofers, it's clear that if you're running stereo sub-bass that it's enough to align each sub to the main that they're closest to.

However for mono sub-bass, what should you prioritize? what is the best plan of action?

Should we align the mains to each other, the subs to each other and then all 4 to each other?

 

[ernestcarl]

Hey sorry to resurrect an old thread, but I'm at my wits end trying to get this figured out. I've made some great progress over the last week by researching, but I'm still having trouble figuring out if the how to make the correction and if it's an improvement or not. I'm trying to align my sub (SVS SB 2000) and mains (Genelec 8030s) for a small home studio.

I followed all the steps in the first post exactly by the OP and now have FDW plots of step and SPL/Phase. I've attached some plots, but the phase plots all look very offset from each other. All alignment has been to the L monitor with the mic at my listening position (UMIK-1).

PHASE
View attachment 133726


SPL
View attachment 133727

STEP
View attachment 133728


I watched a JBL video (See below) and the presenter mentioned we should be aligning the INITAL START of the peaks if we have dissimilar frequency response - not aligning the peaks due to cancellation problems that can be caused from that. Does this limited sweep and windowing account for this? I'm concerned because my sub can go to about 15 hz, and my mains only go to about 50 hz.

If anyone could give me a nudge in the right direction this weekend, I would be forever grateful. I'll buy you a virtual beer Cheers!

I would largely skip the impulse graphs and work directly with aligning the phases as best as possible. Vector average the left and right monitors and then use REW's "Alignment Tool" under the ALL SPL view to sum the single sub and (L+R) mains responses -- IMPORTANT: assuming the acoustic timing reference is correct -- for the sub's reference time, either use the left or right monitor or sum from the two to get the average timing.

You can virtually adjust the gains, delays, and polarity between the two responses to be summed without having to take multiple measurements or using real time analysis using pink noise with software like Smaart.

You can apply FDW to clear things a bit, but at these low frequencies you aren't going get much of an improvement.

It was obvious looking at the phase traces of just the mains (in the SPL & Phase view) that polarity needed to be inverted:

And here's the summed group delay:

sorry, 4ms delay for the mains and not the sub as in the heading.

 

[fluid]

Using IR will be hit or miss (you might get lucky because of your room, speaker/sub placement) because it is a linear presentation of frequency content, not the logarithmic presentation we're used to for the frequency scale. That means that the IR will be heavily biased toward higher frequencies.

If the measured impulse contains energy at higher frequencies, as when you're measuring your mains, its peak will appear earlier than peak of the impulse measured for the sub. It's the energy content between high and low frequencies that affects the shape and positioning of the peaks. So it would be wise to use a frequency restricted signal for the measurement, ideally a range where the sub and mains are outputting the same amplitudes, if you want to align with peak IR, but this is of course won't work if you've already applied high pass and low pass filters. Using the very onset of the IR to align may work, but it's hard to see it for low frequencies.

A simple way to get around this and get a useful display looking at the impulses is to apply 1/3 oct filtering to the IR in the Filtered IR tab set to the crossover frequency or nearest available frequency. This will give a view similar to a wavelet and by aligning both central peaks by reading the relative offset between the peaks or offsetting t=0 to move the display you will get good alignment between the two. Both can be seen at the same time in the overlays impulse window. You have to zoom out to see it from when you were looking at the main pulse.

 

[ernestcarl]

A simple way to get around this and get a useful display looking at the impulses is to apply 1/3 oct filtering to the IR in the Filtered IR tab set to the crossover frequency or nearest available frequency. This will give a view similar to a wavelet and by aligning both central peaks by reading the relative offset between the peaks or offsetting t=0 to move the display you will get good alignment between the two. Both can be seen at the same time in the overlays impulse window. You have to zoom out to see it from when you were looking at the main pulse.

View attachment 134480

Hmmmn. Theoretically, this might work... I think. But the actual measured impulses in the attachment provided, unfortunately, were never ideal nor fully intact (more common than not in reality) -- I mean, just look at the magnitude and different slope of the phases in both speakers and sub. Gigantic dips around the 80Hz region, too. It would be quite hard to "align" entire length of the region of interaction by simply looking at one filtered IR region even if it were the set xo.

 

[fluid]

It works practically as well as theoretically If the attachment you are referring to is the graph I posted those are just simulated filter slopes to represent an idealized driver and show what it could look like. Filtering the impulse works pretty much the same as using a wavelet as an excitation signal.

Good article here
https://www.prosoundweb.com/the-won...ifference-between-delay-phase-shift-polarity/

 

[ernestcarl]

It works practically as well as theoretically If the attachment you are referring to is the graph I posted those are just simulated filter slopes to represent an idealized driver and show what it could look like. Filtering the impulse works pretty much the same as using a wavelet as an excitation signal.

Good article here
https://www.prosoundweb.com/the-won...ifference-between-delay-phase-shift-polarity/

Attachment is the post by @bbizzle who's the person who resurrected the old thread:

https://www.audiosciencereview.com/...subwoofer-s-to-mains-how-to.15269/post-805919

Download the compressed mdat file and go take a look.

Some of my own measured impulses at the LP would be difficult to sum optimally this way. I've tried that technique before and do not find it a practical solution I would choose to employ regularly.

 

[ernestcarl]

It works practically as well as theoretically If the attachment you are referring to is the graph I posted those are just simulated filter slopes to represent an idealized driver and show what it could look like. Filtering the impulse works pretty much the same as using a wavelet as an excitation signal.

Good article here
https://www.prosoundweb.com/the-won...ifference-between-delay-phase-shift-polarity/

Thanks for the link.

 

[abdo123]

@ernestcarl

It would be incredibly useful if you could write your own guide on the topic of aligning subwoofers. even after reading the entire discussion there are still few gaps in my knowledge. This guide assumes that you only have one subwoofer for example.

 

[ernestcarl]

@ernestcarl

It would be incredibly useful if you could write your own guide on the topic of aligning subwoofers. even after reading the entire discussion there are still few gaps in my knowledge. This guide assumes that you only have one subwoofer for example.

I only have one sub at the moment to give as an example, but I could simulate "multiple subs" in a room by moving around that one sub into multiple positions (keeping the main speakers and LP the same to keep as my timing reference). Hmmmn... I'll try that later and will get back to you.

 

[ernestcarl]

To be honest, dunno when I'll get to writing that guide... but in the meantime, here's an example comparing the use of REW's native Alignment tool and the other manual filtered IR method:

Sub Alignment Result Comparison.mdat

Crossovers are at 120Hz so we also will be using 125Hz preset filtered IRs.

8.5ms time offset to align the 125Hz filtered IRs


REW's native Alignment Tool:

15.6ms time offset to fully balance out the magnitude and phase range of interaction.


Forgot to factor in that fact that we gotta move that time delay to the sub box in this case before clicking the align button to get the correct virtual sum -- the actual acoustical time reference recorded were from the closer main monitors:

However, in reality, the time delay will be added to the main Left & Right monitors like this (in miniDSP):

The filtered IR method involves more steps since you have to manually split the views between the overlays windows and the main REW window -- not to mention one has to remember to manually enable or disable the filters before vector averaging the time corrected impulses.

In this example, the difference is not huge, sure, but we can see that REW's alignment tool (green trace) yields slightly better results (t=0):

 

[abdo123]

To be honest, dunno when I'll get to writing that guide... but in the meantime, here's an example comparing the use of REW's native Alignment tool and the other manual filtered IR method:

Sub Alignment Result Comparison.mdat

Crossovers are at 120Hz so we also will be using 125Hz preset filtered IRs.

View attachment 134694
8.5ms time offset to align the 125Hz filtered IRs


REW's native Alignment Tool:
View attachment 134700
15.6ms time offset to fully balance out the magnitude and phase range of interaction.


Forgot to factor in that fact that we gotta move that time delay to the sub box in this case before clicking the align button to get the correct virtual sum -- the actual acoustical time reference recorded were from the closer main monitors:
View attachment 134695

However, in reality, the time delay will be added to the main Left & Right monitors like this (in miniDSP):

View attachment 134697

The filtered IR method involves more steps since you have to manually split the views between the overlays windows and the main REW window -- not to mention one has to remember to manually enable or disable the filters before vector averaging the time corrected impulses.

In this example, the difference is not huge, sure, but we can see that REW's alignment tool (green trace) yields slightly better results (t=0):

View attachment 134703

it will take some processing time for this to seep in as , but thank you for the effort!

I was wondering since you're using MiniDSP, is it better to use crossover filters with dramatic slopes (48db/oct)? Theoritically this means that the sub and main phase need to align over a smaller frequency range which theoritically means it can be better optimized.

Also lower IMD from the mains since less subbass frequencies will be produced.

 

[ernestcarl]

is it better to use crossover filters with dramatic slopes (48db/oct)? Theoritically this means that the sub and main phase need to align over a smaller frequency range which theoritically means it can be better optimized.

Also lower IMD from the mains since less subbass frequencies will be produced.

I used to do that... but no longer with a sub. The more dramatic the slope, the greater change in the phase response -- digitally applied, this isn't that much of a big deal, I guess -- but it's uglier to me, aesthetically speaking.

From recollection, I noticed that the improvements in distortion were minimal to none -- but this also depends on the monitors. My experiments between 70ish to 140ish sub to mains crossovers using different slopes yielded zero improvement in distortion seen above 200Hz.

If ever I were to use higher order filters (above 24db/oct), it would be more so if I were actively trying to avoid negative phase interaction between my speakers and sub beyond a certain point. It's best to match the phases between all speakers with simple delay compensation -- or, alternatively, FIR filters to get an optimal summation however shallow or steep the slope whatsoever one chooses.

I know, miniDSP doesn't exactly have enough taps in some situations...

In my one simple (rather two) case, with a "sealed" nearfield monitor desk setup in a treated room, I could get something very close to linear in both magnitude and phase utilizing only 1363 taps:

24dB/oct slopes

Overlap applied in consideration of sub's lower overall room gain and declining SPL with rising frequency. But decreasing the speaker xo below 75Hz caused a negative interaction (lost SPL level due to cancellation) -- so best stop at 75Hz.

24db/oct slopes

SPL of the sub in the standing position had greater room gain and was not declining with increasing frequency so both xo are set at 120Hz.

We'd see many more rotations of the phase in ported systems.

Nevertheless, I think the more important point whatever the scenario is getting the placement (primary), gains, and delays optimized -- prior any EQ correction -- which in turn would give the best possible xo results. Phase optimization and FIR correction can be eschewed in most cases -- unless... for some reason the region of interaction between speaker boxes were truly bad.

 

[Ron Texas]

Is there a way to know by listening that subs are time aligned? For example does a kick drum sound real? I have stereo subs crossed over at a very high 170 Hz. Subs and mains are the same distance from the LP with the subs very close to the mains. The entire setup is symmetrical left to right. Mains are subject to a slight delay from the built in DSP of the Crown XLS 1502 amp they are connected to. Subs are Rythmik L12's which don't have much internal delay.

I'm wondering why the measured delays in the examples are so high. I suppose the physical distance to the subs is much more than to the mains and the subs are using DSP's with a lot of delay. I managed to find out the delay of an L12 is 3 ms and an SVS 3000 pro is 6 ms. Then, I wonder if there is any delay in the complex crossover in my LS50's. One other bit is typically a large driver of 12 or 15" has a 1 ms delay relative to a much smaller faster driver due to the time it takes to get that much mass and air moving.

Some measurements I made with REW (which I don't have a lot of confidence that I did right) told me the timing difference was tiny, less than 0.2 ms. Kick drums do sound right, by the way.

 

[fluid]

View attachment 134694
8.5ms time offset to align the 125Hz filtered IRs

I'm not sure what you did to get the 8.5ms value but when I read the differences between the highest peaks after the 1/3 oct filtering I got a delay of 16.76ms with your data when using the Left Only and Left Sub which gave this shape. It also produced a nice response in the alignment tool when the Left channel was offset by that amount. The alignment tool is great and you are right turning the filtering on and off isn't immediately intuitive but you only need to do it once to find the delay value. If the crossover is completely complimentary the wave shape matches really well, if the crossover is bit off either way it shows in the alignment away from the centre peak. To me it is just another way of looking at the data

 

[ernestcarl]

I'm not sure what you did to get the 8.5ms value but when I read the differences between the highest peaks after the 1/3 oct filtering I got a delay of 16.76ms with your data when using the Left Only and Left Sub which gave this shape. It also produced a nice response in the alignment tool when the Left channel was offset by that amount. The alignment tool is great and you are right turning the filtering on and off isn't immediately intuitive but you only need to do it once to find the delay value. If the crossover is completely complimentary the wave shape matches really well, if the crossover is bit off either way it shows in the alignment away from the centre peak. To me it is just another way of looking at the data

View attachment 134871

Strange... I must have clicked on 80Hz preset in one of the impulses? Not sure what happened exactly, but I entered your 16.76ms delay and it quite close in value 15.6ms -- the difference is not of any import; the aligned sums are almost the same. Though, my main point was really if the impulses were not as intact as in the sample here, then it would be difficult to eyeball which peaks to align. Personally, I use the impulses to check how large the time deviation is to get delays within the ballpark before using the alignment tool itself. Indeed, goes to show that with any method it takes practice... either way, it's also rather easy to get things wrong.

 

[ernestcarl]

Is there a way to know by listening that subs are time aligned? For example does a kick drum sound real? I have stereo subs crossed over at a very high 170 Hz. Subs and mains are the same distance from the LP with the subs very close to the mains. The entire setup is symmetrical left to right. Mains are subject to a slight delay from the built in DSP of the Crown XLS 1502 amp they are connected to. Subs are Rythmik L12's which don't have much internal delay.

I'm wondering why the measured delays in the examples are so high. I suppose the physical distance to the subs is much more than to the mains and the subs are using DSP's with a lot of delay. I managed to find out the delay of an L12 is 3 ms and an SVS 3000 pro is 6 ms. Then, I wonder if there is any delay in the complex crossover in my LS50's. One other bit is typically a large driver of 12 or 15" has a 1 ms delay relative to a much smaller faster driver due to the time it takes to get that much mass and air moving.

Some measurements I made with REW (which I don't have a lot of confidence that I did right) told me the timing difference was tiny, less than 0.2 ms. Kick drums do sound right, by the way.

What is the difference in distance between your subs and main monitors? Could be a number of reasons why delay is larger than expected... but I don't really think about that -- unless the delay is really big like around 30ms or so.

Listening to short audio clip samples do sometimes work for me -- if the deviation is big enough -- but I have to compare it with the objective measurements to make sure I'm not just imagining things as well. As mentioned, one can look at the phase curves and/or the impulse graph views... Another way to look at things would be through a wavelet spectrograph -- but as always, it will be easier to "read" how well aligned the system is if impulses measured at the listening position come out relatively clean or intact/complete, which isn't always the case.

NO EQs applied other than crossovers:

No delays

delays applied

 

[Ron Texas]

@ernestcarl subs are right next to the mains to the inside and distance from mains to LP and subs to LP is the same. I'm busy for the next few days, but will try the spectrogram soon. Thanks.