Time Alignment of Speaker Drivers

[gnarly]

Yes. Don’t confuse phase shift with time delay. Not so easy if the crossover filter is IIR Biquad and phase of each driver not known precisely. With FIR filters, one can completely compensate the phase first to be flat and the same for each driver - assuming you that extract the minimum phase of the driver correctly first. Then setting the delay so that the signal from each driver reaches the microphone at the same time is straight-forward. But in this process ^ you have to do the phase compensation before time alignment first.

Yes indeed
Time delays and phase are simply not substitutes for each other.
And as you say, for aligning drivers, get the phase of each correct first, then time align..

Hopefully this example of that finds some interest.......here's the acoustic measurements acoustic of a 5-way's passbands.
Each passband was first individually tuned as a linear-phase passband. So each has flat phase at zero, throughout its passband and summations.

First without time alignment.

With linear-phase passbands, time alignment requires impulse peaks align. The physical design of the speaker happended such that the blue low driver and green mid driver need very little delay between them, as their impulse peaks are close together. The purple sub needs the most delay added. Orange and yellow HF and VHF need less, but it's clear they need slightly different amounts of delay due to small null at xover.

The solid black line is the summed acoustic response. The black dots are what summation could be with proper time alignment....which follows blows.

Proper time alignment of 5-sections. All impulse peaks aligned.
Note passbands' responses don't change, just the summations change.

Hope twas of interest...

 

[gnarly]

Maybe this can help you:

https://www.nature.com/articles/s43588-021-00183-z.pdf

Thx OCA. Good paper. Talk about measurement bandwidth needed...neuro-science... yikes. 2-3 Hz to who knows how high.
Wavelets have been a good learning tool for me.

 

[Waveform Fidelity]

Yes indeed
Time delays and phase are simply not substitutes for each other.
And as you say, for aligning drivers, get the phase of each correct first, then time align..

Hopefully this example of that finds some interest.......here's the acoustic measurements acoustic of a 5-way's passbands.
Each passband was first individually tuned as a linear-phase passband. So each has flat phase at zero, throughout its passband and summations.

First without time alignment.

With linear-phase passbands, time alignment requires impulse peaks align. The physical design of the speaker happended such that the blue low driver and green mid driver need very little delay between them, as their impulse peaks are close together. The purple sub needs the most delay added. Orange and yellow HF and VHF need less, but it's clear they need slightly different amounts of delay due to small null at xover.

The solid black line is the summed acoustic response. The black dots are what summation could be with proper time alignment....which follows blows.
View attachment 503658

Proper time alignment of 5-sections. All impulse peaks aligned.
Note passbands' responses don't change, just the summations change.
View attachment 503659
Hope twas of interest...

Very interesting - thanks for sharing. What type of test signal are you using to get the impulse responses? When you measure the EQ acoustic response of each passband you get the combined driver plus the filter response, plus the delay from the FIR filter, so how do you get the combined phase response to verify? What I do is to measure the minimum phase of the driver and then measure the FIR filter response (subtracting the DSP delay) from only the DSP filter and combine them as a check point that filter response+driver response = desired acoustic target response.

Is you 5 driver system used for AV applications? I see that the overall latency is around 92 milliseconds getting close to ITU recommended limits for “lip sync”

 

[gnarly]

Very interesting - thanks for sharing. What type of test signal are you using to get the impulse responses? When you measure the EQ acoustic response of each passband you get the combined driver plus the filter response, plus the delay from the FIR filter, so how do you get the combined phase response to verify?

Thanks. I use either sine sweeps or time-averaged pink. Prefer pink, but use sweeps when i want to see harmonics.
After measuring the raw response of a drivers passband, I use the measurement software's FIR generator to auto target match the specified xover curves, and flatten in-band mag and phase. Then put filter in place and verify acoustic mag, phase and impulse..
Here is the mid sections raw response,

followed by the corrected response which has lin-phase 96dB/oct hpf and lpf.

It's pretty easy to see all is well with response and impulse, and with flat phase thru passband and summations.
When each section has flat phase like that, I know the entire speaker will have flat phase with nothing more needed than time alignment.

So after setting delays as indicated by impulse peaks' arrivals, I just measure the entire speaker to confirm.

From your depiction, I gather we both do basically the same process.

Yeah, 90ms too much for lip sync. I'm not into home-theatre, but I do set up speakers for live sound. Then I hold latency to 15ms, and use IIR for all sub related work.

 

[Waveform Fidelity]

Thanks. I use either sine sweeps or time-averaged pink. Prefer pink, but use sweeps when i want to see harmonics.
After measuring the raw response of a drivers passband, I use the measurement software's FIR generator to auto target match the specified xover curves, and flatten in-band mag and phase. Then put filter in place and verify acoustic mag, phase and impulse..
Here is the mid sections raw response,
View attachment 503694

followed by the corrected response which has lin-phase 96dB/oct hpf and lpf.
View attachment 503696

It's pretty easy to see all is well with response and impulse, and with flat phase thru passband and summations.
When each section has flat phase like that, I know the entire speaker will have flat phase with nothing more needed than time alignment.

So after setting delays as indicated by impulse peaks' arrivals, I just measure the entire speaker to confirm.
View attachment 503701

From your depiction, I gather we both do basically the same process.

Yeah, 90ms too much for lip sync. I'm not into home-theatre, but I do set up speakers for live sound. Then I hold latency to 15ms, and use IIR for all sub related work.

Do you hear any pre-ringing artifacts? I am using LR48 slopes with nothing audible - but then I am using a coincident driver. Also are you designing the FIR filters in Matlab or some other tool? My DSP is limited to 512 taps so I can only flatten the phase to about 15kHz, but I figure thats OK. The UNIQ tweeter has a strong resonance at 24kHz also making it difficult to get a smooth top end above 15KHz anyway.

Once you build a speaker system like this, its hard to go back to conventional designs that have large group delay caused by analog or IIR equivalent crossovers.

 

[gnarly]

Do you hear any pre-ringing artifacts? I am using LR48 slopes with nothing audible - but then I am using a coincident driver.

No pre-ring I can hear. That said I do try to make sure the linear-phase xovers are fully acoustically complementary, and drivers are pretty coincident, speaker being a synergy/MEH. (I tend to believe the hullabaloo about pre-ring is mostly parrot speak, not first hand experience.)

Also are you designing the FIR filters in Matlab or some other tool?

The measurement program I've been showing has a FIR generator. Using the same mid section as above, the FIR filter section is now shown.
I simply fill in the yellow fields: desired high-pass and low-pass, smoothing, # taps, and window choice. Peak Norm adjusts output level.
Right clicking on the coefficents gives export file options. That's it.
One time for laughs I timed how fast I could do a 5-way from scratch, starting with making raw measurements for each section.
22 minutes to finished speaker! Gotta love modern tools.

Once you build a speaker system like this, its hard to go back to conventional designs that have large group delay caused by analog or IIR equivalent crossovers.

I agree.
Although that said, I do still experiment with traditional min-phase /IIR tunings. I like to A/B processing strategies, learning what each does best.

 

[Digital_Thor]

Could be interesting one day, to hear what FIR does in direct comparison to IIR. Because I do understand some of the advantages of FIR, but I haven't heard them yet in contrast to a well designed active IIR or passive speaker.
I don't exactly know how this experiment should be carried out, since so many things change as you shift between the two, that I find it difficult to truly compare them.
Any experiment I've heard, where you took a 'normal' speaker and added FIR, it always just sound dull and flat, like you just give it -3dB with a Q ranging from 1-6kHz.
Maybe I just can't hear it

 

[gnarly]

Could be interesting one day, to hear what FIR does in direct comparison to IIR. Because I do understand some of the advantages of FIR, but I haven't heard them yet in contrast to a well designed active IIR or passive speaker.
I don't exactly know how this experiment should be carried out, since so many things change as you shift between the two, that I find it difficult to truly compare them.
Any experiment I've heard, where you took a 'normal' speaker and added FIR, it always just sound dull and flat, like you just give it -3dB with a Q ranging from 1-6kHz.
Maybe I just can't hear it

Yep, I've had the same experience, when hearing 'normal' speakers when FIR was layered on top of the existing passive or active design.
Most of the attempts I've seen/heard like that, have been about trying to linear the existing speaker's phase. Dull and flat are good descriptors.

As opposed to that kind of global FIR correction, I have taken existing speakers, both passive and multi-way active, and completely converted them multiway FIR by tuning each driver section individually, like in the example I've been posting.
The passives were easy to improve. One good active wasn't too hard to improve, but one already excellent active was too hard to tell if any improvement occurred..
And there was the big problem of making fair comparisons, as there was no certainty when I converted one speaker and compared it to an original, that the two were the same to start.

For my FIR vs IIR experiments on a single speaker, it's pretty easy to make fair comparisons. Prosound open-architecture processors allow complete on-the-fly switching between about every filter, parameter, every setting in them, even full sets of FIR files..
So for example, on the 5-way I've been showing, I can easily switch change between full linear-phase processing and regular IIR processing. Preset changes are instant and silent. Makes A/B easy.

 

[Waveform Fidelity]

Very interesting implementation. I would like to double check my filter implementation with a different measurement tool to validate my designs. I currently use a mix of Soundeasy and Arta. What is the name of the tool you use please?

 

[Keith_W]

Very interesting implementation. I would like to double check my filter implementation with a different measurement tool to validate my designs. I currently use a mix of Soundeasy and Arta. What is the name of the tool you use please?

You are using Bodzio SoundEasy? Very interesting. I have meant to get in contact with that guy for a while, given that he's in the same city as me.

 

[Waveform Fidelity]

You are using Bodzio SoundEasy? Very interesting. I have meant to get in contact with that guy for a while, given that he's in the same city as me.

Yes, Bohdan has been developing Soundeasy since at least 2014, when I first started using it. He developed Ultimate Equalizer a WIN 7 programe that will equalize your complete speaker system to produce flat amplitude and phase as he demonstrates by showing acoustic reproduction of square waves from his multi-driver speakers.

For me, the audio work is purely a hobby which I have picked up again after many years. I retired recently being a communications engineer working on 5G/Satellite systems particularly RF antennas and propagation - you would be amazed as how much of that theory and practice can be applied to acoustics and digital signal processing.

I would recommend making contact with him - ask him about linear phase, pre-ringing, his new HBT/IHBT routines to precisely extract the minimum phase of a driver and take a look at the latest version of Soundeasy - very feature rich and advanced.

 

[gnarly]

Very interesting implementation. I would like to double check my filter implementation with a different measurement tool to validate my designs. I currently use a mix of Soundeasy and Arta. What is the name of the tool you use please?

The software I've been using/showing is Crosslite+. It is a combination of a base measurement and filter simulation program, with a separately purchase FIR add on module.
It's strongly geared for live sound technicians, who already have good experience with other commercial software such as Smaart or Systune.
As the base measurement/simulation module has a pretty steep learning curve to it. The FIR generator add-on is comparatively child's play to master.

A nifty combined measurement/simulation and FIR generator has shown up on ASR recently if you haven't seen it, called LinFIR.
It's freeware for now under development, other than an optional directivity mapping module.
It has a bit of a learning curve too, but I think more so on the filter generation side, and less so on the measurement side.

In ways, I often think REW is the maybe most advanced/capable measuring program around, if you're willing to confine work to single channel sine sweeps.
I like your combo of Soundeasy and Arta. Arta I have used a bit. No experience with Soundeasy, just always appreciated Bodhan's articles.

 

[Digital_Thor]

Thanks. I use either sine sweeps or time-averaged pink. Prefer pink, but use sweeps when i want to see harmonics.
After measuring the raw response of a drivers passband, I use the measurement software's FIR generator to auto target match the specified xover curves, and flatten in-band mag and phase. Then put filter in place and verify acoustic mag, phase and impulse..
Here is the mid sections raw response,
View attachment 503694

followed by the corrected response which has lin-phase 96dB/oct hpf and lpf.
View attachment 503696

It's pretty easy to see all is well with response and impulse, and with flat phase thru passband and summations.
When each section has flat phase like that, I know the entire speaker will have flat phase with nothing more needed than time alignment.

So after setting delays as indicated by impulse peaks' arrivals, I just measure the entire speaker to confirm.
View attachment 503701

From your depiction, I gather we both do basically the same process.

Yeah, 90ms too much for lip sync. I'm not into home-theatre, but I do set up speakers for live sound. Then I hold latency to 15ms, and use IIR for all sub related work.

So, how does it look off-axis?

 

[gnarly]

So, how does it look off-axis?

Here ya go...easier to just link to a different post on it.
The first set of off-axis traces are for main only, without sub.

A Broad Discussion of Speakers with Major Audio Luminaries

MLP=main listening position, the sweet spot, mine is about in the inner corner of a 2.8 meter triangle. And no EQ means no EQ, just the speaker (a semi-active big-ish 3-way with analog el. x-over) and my room (also biggish, about 80m² and treated though) Thx. Again, very nice. Larger...

www.audiosciencereview.com

 

[Keith_W]

Yes, Bohdan has been developing Soundeasy since at least 2014, when I first started using it. He developed Ultimate Equalizer a WIN 7 programe that will equalize your complete speaker system to produce flat amplitude and phase as he demonstrates by showing acoustic reproduction of square waves from his multi-driver speakers.

For me, the audio work is purely a hobby which I have picked up again after many years. I retired recently being a communications engineer working on 5G/Satellite systems particularly RF antennas and propagation - you would be amazed as how much of that theory and practice can be applied to acoustics and digital signal processing.

I would recommend making contact with him - ask him about linear phase, pre-ringing, his new HBT/IHBT routines to precisely extract the minimum phase of a driver and take a look at the latest version of Soundeasy - very feature rich and advanced.

Thanks, I already use Acourate. But I have spent some time on his website and read his thoughts on linear phase. Interestingly, his webpage contains some pictures of a gathering held at my house, more than 10 years ago, which was just before I began my DSP journey. He has never been over here, but it does appear we may have friends in common!