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

[levimax]

I disagree with what you say about «difficult» and «unreliable». I have done this alignment myself without difficulty, and with reliable results.

What manual method do you use? REW time align tool doesn't seem to work at LF. Impulse offset methods can be difficult to interpret and LF in room measurements can be difficult to repeat and interpret. I have OK results with loopback nearfield measurements and a tape measure.

 

[DonH56]

For a quick audible alignment I play a test tone at the crossover frequency and adjust the sub's phase for the loudest sound (sub plus mains) at the listening position. Easier to do with a cheap (~$30 USD) SPL meter.

 

[janbth]

What manual method do you use?

Like I said in post #218 above, I look at the frequency response and tweak the delay of the sub(s) to flatten the response at the crossover frequency.

To be a little more specific: Choose an initial delay for the sub. This could simply be 0 ms. (I was using a miniDSP 2x4 HD with this setup.) Do a sweep with REW. Set to 10 ms. Do a sweep. Set to 20 ms. Another sweep. Etc. etc. You will then get a family of curves, and the one that is smoothest at the crossover provides the best value for the delay.

As an extra step, you can compute the excess group delay in REW (see REW documentation) to check that the sub doesn’t fire too early or too late.

 

[levimax]

Like I said in post #218 above, I look at the frequency response and tweak the delay of the sub(s) to flatten the response at the crossover frequency.

To be a little more specific: Choose an initial delay for the sub. This could simply be 0 ms. (I was using a miniDSP 2x4 HD with this setup.) Do a sweep with REW. Set to 10 ms. Do a sweep. Set to 20 ms. Another sweep. Etc. etc. You will then get a family of curves, and the one that is smoothest at the crossover provides the best value for the delay.

As an extra step, you can compute the excess group delay in REW (see REW documentation) to check that the sub doesn’t fire too early or too late.

OK that makes sense. You are not relying on accurate LF timing measurements rather you are looking at the magnitude response. It is like a "manual MIMO" in a way.

 

[Keith_W]

What manual method do you use? REW time align tool doesn't seem to work at LF. Impulse offset methods can be difficult to interpret and LF in room measurements can be difficult to repeat and interpret. I have OK results with loopback nearfield measurements and a tape measure.

The REW alignment tool works very well? This is what I do:

1. Take a sub measurement with XO in place with a tweeter or loopback timing ref.
2. Take a sweep of both main speakers (with HPF in place, if any) with the same timing ref.
3. Open the alignment tool, load both graphs, then fiddle with phase alignment until there are as few dips as possible (ignore the peaks). Also pay attention to the impulse graph, you want the sub to the right of the main impulse. You also don't want any delay greater than half a period. If you need more, consider inverting the polarity.
4. Click on the "Make aligned copy" on each measurement, then vector sum them.
4. Dial those settings in and take a verification measurement. Compare it to your vector sum. If it's the same, you have successfully completed the alignment procedure. You may proceed to EQ.

 

[SoundsGood64]

The REW alignment tool works very well? This is what I do:

1. Take a sub measurement with XO in place with a tweeter or loopback timing ref.
2. Take a sweep of both main speakers (with HPF in place, if any) with the same timing ref.
3. Open the alignment tool, load both graphs, then fiddle with phase alignment until there are as few dips as possible (ignore the peaks). Also pay attention to the impulse graph, you want the sub to the right of the main impulse. You also don't want any delay greater than half a period. If you need more, consider inverting the polarity.
4. Click on the "Make aligned copy" on each measurement, then vector sum them.
4. Dial those settings in and take a verification measurement. Compare it to your vector sum. If it's the same, you have successfully completed the alignment procedure. You may proceed to EQ.

Going to have to try this. Two questions:

When you sweep the main speakers is the sub turned on or off?

Would it make more sense to deal with polarity first, in a step before determining sub delay? It seems like polarity would have a huge impact on phase and there are already many moving parts. The way I've been dealing with polarity is, with an 80hz crossover already set, take a 20-200hz sweep of L and R main speakers with sub, all 3 together with normal sub polarity, then the same sweep with the 3 together but with inverted sub polarity. Then I overlay the two and see which one has higher SPL in the crossover range, 60-100hz. A deep dip probably isn't good. So looking for higher SPL and no big dips in the crossover range, and then pick normal or inverted based on which looks better. Not sure if this method makes sense to folks, so putting it for consideration.

 

[Keith_W]

Going to have to try this. Two questions:

When you sweep the main speakers is the sub turned on or off?

If you are trying to align the subwoofer to the main speakers, it makes sense to take a measurement of the main speakers alone, wouldn't it? So subwoofer has to be off for that measurement.

Would it make more sense to deal with polarity first, in a step before determining sub delay? It seems like polarity would have a huge impact on phase and there are already many moving parts.

You could, if you want. As long as the same result is reached, it doesn't matter how you get there.

 

[gnarly]

Ime, the problem with in-room sub measurements is you rarely know exactly what is being measured....
a. the direct sub
b. a reflection stronger than the direct sub
c. or mostly likely, a summed vector of the sub and its stronger reflections.

We may be able to make repeatable in-room measurements at any specific location....but which of the above is being measured?

I've found Coherence, the statistical measurement of 'likeness' between stimulus signal and DUT (sub), to be a good indicator of how strong the reflections are in the measurement. (I wish REW had coherence, like all the dual-channels do.)

One thing for sure ime, when coherence is low, meaning multiple arrivals are being summed together,
you can throw the measurement out the window, for the purpose of measuring or EQing the sub alone.

 

[levimax]

The REW alignment tool works very well? This is what I do:

1. Take a sub measurement with XO in place with a tweeter or loopback timing ref.
2. Take a sweep of both main speakers (with HPF in place, if any) with the same timing ref.
3. Open the alignment tool, load both graphs, then fiddle with phase alignment until there are as few dips as possible (ignore the peaks). Also pay attention to the impulse graph, you want the sub to the right of the main impulse. You also don't want any delay greater than half a period. If you need more, consider inverting the polarity.
4. Click on the "Make aligned copy" on each measurement, then vector sum them.
4. Dial those settings in and take a verification measurement. Compare it to your vector sum. If it's the same, you have successfully completed the alignment procedure. You may proceed to EQ.

I find the REW alignment tool to work well for higher frequencies but I have had no luck with it for LF sub integration. Maybe in some rooms it will work but not mine. When I try to use it I get nonsensical delays and polarity switches. I believe this is because LF in room measurements are often compromised for the reasons given by @gnarly one post up.

 

[Keith_W]

Ime, the problem with in-room sub measurements is you rarely know exactly what is being measured....
a. the direct sub
b. a reflection stronger than the direct sub
c. or mostly likely, a summed vector of the sub and its stronger reflections.

We may be able to make repeatable in-room measurements at any specific location....but which of the above is being measured?

When in doubt, make a simulation to understand what is the minimum-phase behaviour of the sub, and the non-minimum phase contribution of the room. As it happens, I was helping my friend do exactly that today. He couldn't understand the squiggly lines of his sub's step response.

First step is to make the simulation. I looked at his curve and guessed: bandpass filter; HPF 20Hz (Butterworth 4), LPF 100Hz (Butterworth 4). As you can see, it's a pretty good approximation of his sub.

Then use the simulation to guide your eyes when you read the step response. You can clearly see what the shape should be, and how it deviates from the ideal response. That's his left sub, it's clearly worse than his right sub which is below.

 

[kifeep]

When in doubt, make a simulation to understand what is the minimum-phase behaviour of the sub, and the non-minimum phase contribution of the room. As it happens, I was helping my friend do exactly that today. He couldn't understand the squiggly lines of his sub's step response.

View attachment 548166

First step is to make the simulation. I looked at his curve and guessed: bandpass filter; HPF 20Hz (Butterworth 4), LPF 100Hz (Butterworth 4). As you can see, it's a pretty good approximation of his sub.

View attachment 548168

Then use the simulation to guide your eyes when you read the step response. You can clearly see what the shape should be, and how it deviates from the ideal response. That's his left sub, it's clearly worse than his right sub which is below.

View attachment 548167

That's a nice trick. How does this compare to making a minimum phase copy of the sub response? I don't have access to REW at the moment, but will try to remember to test it myself later.

 

[levimax]

When in doubt, make a simulation to understand what is the minimum-phase behaviour of the sub, and the non-minimum phase contribution of the room. As it happens, I was helping my friend do exactly that today. He couldn't understand the squiggly lines of his sub's step response.

View attachment 548166

First step is to make the simulation. I looked at his curve and guessed: bandpass filter; HPF 20Hz (Butterworth 4), LPF 100Hz (Butterworth 4). As you can see, it's a pretty good approximation of his sub.

View attachment 548168

Then use the simulation to guide your eyes when you read the step response. You can clearly see what the shape should be, and how it deviates from the ideal response. That's his left sub, it's clearly worse than his right sub which is below.

View attachment 548167

This is a good way to see how much the room changes the measurement. How do you use this to help with crossover alignment or EQ?

 

[Keith_W]

This is a good way to see how much the room changes the measurement. How do you use this to help with crossover alignment or EQ?

You don't. My friend had a hard time understanding what all those wiggles meant, so I made the simulation for him. That's all it's for, it's easier on the eyes.

Crossover alignment is very simple, all you do is shift the delay until you get as few dips as possible. Then you EQ all the peaks out.

 

[Keith_W]

That's a nice trick. How does this compare to making a minimum phase copy of the sub response?

Raw measurement vs. simulation.

Minphase copy vs. simulation.

 

[janbth]

Ime, the problem with in-room sub measurements is you rarely know exactly what is being measured....
a. the direct sub
b. a reflection stronger than the direct sub
c. or mostly likely, a summed vector of the sub and its stronger reflections.

We may be able to make repeatable in-room measurements at any specific location....but which of the above is being measured?

None of these. It seems you are thinking in terms of traveling waves, but that applies only above the Schroeder frequency, not below. For subs, we are in the modal region, where the room behaves like a collective system. Direct, reflected, 2nd reflected, 3rd reflected, … they are all added together in a steady state frequency response. Which is why impulse response based methods for integrating subs are flawed. Look at the frequency response instead.

 

[SoundsGood64]

Here are the steps:

1. With speaker delays, speaker gain trims, sub gain, sub polarity and crossover at 80hz in CamillaDSP, proceed to determine sub delay.

2. In REW, set acoustic timing reference to timing reference speaker
(the speaker physically closest to the MLP, example Surround Right).
Decide which front speaker to use to time align with sub, the one that
is closer to the sub is a good choice. Let's assume that is the R speaker.

3. Measure R front (sub turned off, XO active), L front (sub turned off, XO active) and LFE (XO active) using:
- Sweep: 40-120Hz, 512k length (I saw advice calling for 2million but it seemed risky for speakers based on REW documentation)
- Level: -9dBFS
- AVR: -9dB
- Single sweep mode (required when using acoustic timing reference)

4. For each measurement, IR Windows, apply FDW (Frequency Dependent Windowing)
of 6 cycles, and apply to all SPL selections.

5. Go to overlay and choose phase. Under actions, Unwrap phase for both measurements.

6. In the phase overlay view, find the sub delay such that the phase of R and LFE are equivalent at 80hz. (sub_delay6_7.5ms.jpg; L blue, R red, LFE with no delay light green, LFE with 7.5ms delay dark green). Here a sub delay of 7.5ms got the phase of LFE and R roughly equivalent at 80hz.

View attachment 547809

Some questions for discussion:

1 Should I determine sub delay with 80hz crossover active? I saw conflicting advice on this. It made sense to me that I should do it with crossover active because that is the state in which the sub delay will be used when listening to music and I think the 4th order Linkwitz Riley filter used for crossover may affect the phase (also I noticed that if the LPF for LFE on the Denon AVR is set to 120hz as is common practice, that also affects the phase, so I leave it at 250hz). However, I have some doubt about this. Post 1 here from QMuse measured without crossover but later posts in this thread indicated to perform measurements with crossover active.

2 The sub delay determination must be done after the sub polarity is already set, correct (because the polarity will affect the phase)?

3 Notice in figure that L and R speakers, although previously matched for gain and delay (at a higher frequency than sub range) are still not completely in phase with each other. How to deal with that? To keep it simple, I chose to set sub delay such that the phase curves of R and LFE overlap at the 80hz (crossover point). But what about instead simply setting the sub delay to zero given that the LFE phase sits roughly at the midpoint between L and R at 80hz? Or should I work to get L and R in phase so they are overlapping at 80hz, but if I do that that will throw off the speaker delays (that were determined using a higher frequency sweep)?

Thanks for considering.

I've found answers here to #1 and #2 above, but #3 is still a question. In post #225, Keith_W stated "2. Take a sweep of both main speakers (with HPF in place, if any) with the same timing ref." In this method both L and R speakers are measured at the same time, which is similar to phase aligning the subwoofer to the mid-point between L and R speakers at 80hz when they were measured separately. So instead of phase aligning the sub to an average of the L and R speakers, how about instead using an all pass filter to phase align the L speaker to the R speaker at 80hz, and then with the crossover and all pass filters active, determine the sub delay to phase align the subwoofer to the R speaker? How does the method of using an all pass filter to phase align L and R at the crossover point followed by phase aligning the sub compare to the method of phase aligning the sub to an average of the L and R speaker?

 

[janbth]

I've found answers here to #1 and #2 above, but #3 is still a question. In post #225, Keith_W stated "2. Take a sweep of both main speakers (with HPF in place, if any) with the same timing ref." In this method both L and R speakers are measured at the same time, which is similar to phase aligning the subwoofer to the mid-point between L and R speakers at 80hz when they were measured separately. So instead of phase aligning the sub to an average of the L and R speakers, how about instead using an all pass filter to phase align the L speaker to the R speaker at 80hz, and then with the crossover and all pass filters active, determine the sub delay to phase align the subwoofer to the R speaker? How does the method of using an all pass filter to phase align L and R at the crossover point followed by phase aligning the sub compare to the method of phase aligning the sub to an average of the L and R speaker?

If you set proper time delays on L and R (from #1), then you should already have L and R phase aligned, shouldn't you?

 

[gnarly]

When in doubt, make a simulation to understand what is the minimum-phase behaviour of the sub, and the non-minimum phase contribution of the room. As it happens, I was helping my friend do exactly that today. He couldn't understand the squiggly lines of his sub's step response.

View attachment 548166

First step is to make the simulation. I looked at his curve and guessed: bandpass filter; HPF 20Hz (Butterworth 4), LPF 100Hz (Butterworth 4). As you can see, it's a pretty good approximation of his sub.

View attachment 548168

Then use the simulation to guide your eyes when you read the step response. You can clearly see what the shape should be, and how it deviates from the ideal response. That's his left sub, it's clearly worse than his right sub which is below.

View attachment 548167

Thanks Keith, but not for me. I'm an old goat who's made too many measurements for too long, and have become a bit cynical as to how well they reflect reality (in the manner we hope). So not trusting measurements all that far, I'm just not willing to peer down deep into them. I used to do that, lord knows I used to....

 

[john61ct]

Easier to do with a cheap (~$30 USD) SPL meter.

Can you recommend one that stores X session results for later retrieval? Not pricey.

Alternatively, reco software that makes use of my UMIK-2, ideally via Android, but windoze or linux otherwise?

 

[gnarly]

None of these. It seems you are thinking in terms of traveling waves, but that applies only above the Schroeder frequency, not below. For subs, we are in the modal region, where the room behaves like a collective system. Direct, reflected, 2nd reflected, 3rd reflected, … they are all added together in a steady state frequency response. Which is why impulse response based methods for integrating subs are flawed. Look at the frequency response instead.

Hi, must disagree. All sound is traveling waves...some just happen to be big enough to cause in-room resonances, but they are still traveling waves.

What I do agree with you say is the room measurement everyone takes is almost always the sum of direct, reflected, 2nd reflected, 3rd reflected....
Which is exactly the problem !!! Treat it as a lumped sum for correction, and risk moving towards a lump of mud.

Imo, room correction via DSP might be legitimately viable, when and if we take a rooms impulse response, and invert that.
Not using our speakers to provide the stimulus, which are directional, bu with a made for room measurements omni speaker.
Hoping to try this. Or probably for me to avoid buying an omni, is just record a balloon pop and invert the impulse.