A history of Time Aligned Loudspeakers

[gberchin]

With Linkwitz filters this is impossible!

Any step response that returns to zero means that the overall system has a highpass characteristic. That is to say, the woofer response does not extend all the way to DC. Here is what an ideal, full-bandwidth LR4 step response looks like (square waves shown, which for our argument can be treated as two steps of opposite polarity).

 

[kimmosto]

is this in headphones?

Yes, continuous transient signal with headphones.

Also I don't understand why the numbers you have are well below the numbers in the peer reviewed studies? The sensitivity to group delay goes down with frequency not up. WRT L/R 12 and .16 cycles at 350Hz is approximately 60 usec ? The most sensitive region is 1-4k and that is well below the thresholds there.

Ca. 350 Hz is the lowest inaudible XO with textbook LR24. Not with LR12. LR24 XO @350 Hz creates GD hump of 1.54 ms at ca. 220 Hz. The peak is constant 0.45 cycles at XO frequency (350 Hz in this example).

I've not yet started new tests with LR12. Delay is much shorter than with LR24 so audible XO limit could be found somewhere below 140 Hz. Audibility is program-dependent so I don't see why different studies should give equal results. Timing error creates dynamic error with wide range transients so it's not totally unnatural that lower fundamental in a transient program and longer delay variation due to lower/steeper XOs is easier case for hearing and other body sensors than some short delay at "the most sensitive region" only.
Audibility limit as a function of frequency is also confusing presentation because group delay spectrums of different textbook XOs are not linear. They are stepping with some transition slope, with or without extra hump, and high-pass at LF has it's own peak.

 

[Trifonov Audio]

Any step response that returns to zero means that the overall system has a highpass characteristic. That is to say, the woofer response does not extend all the way to DC. Here is what an ideal, full-bandwidth LR4 step response looks like (square waves shown, which for our argument can be treated as two steps of opposite polarity).

With your Method (Technology) - things are different from "simple" passive filters which are not easy at all!

 

[gberchin]

things are different from "simple" passive filters which are not easy at all!

It is probably obvious that I am a strong advocate of matched-delay subtractive crossovers. However, having listened to a variety of different crossovers -- Gaussian-derived MDS, Bessel-derived MDS, Butterworth and LR both passive and active -- my opinion has evolved to "it really doesn't matter very much".

 

[Trifonov Audio]

Here you go—a graph of the BU18/Oct Butterworth filter, showing both in-phase and anti-phase responses.

View attachment 535280

The impulse responses from the drivers must be "assembled" (synchronized) in time into a "beautiful" TRIANGLE! This is done by introducing a delay to each adjacent upper one... and not only that! Can you control such a delay separately for each driver?
Can you set a different crossover with a specific Q-factor for each driver? Can you set the same polarity for all drivers?
These are the "small", but important (mandatory) conditions, if you have selected suitable drivers and defined optimal frequencies for the cross-slopes?......?

 

[LSPhil]

The impulse responses from the drivers must be "assembled" (synchronized) in time into a "beautiful" TRIANGLE! This is done by introducing a delay to each adjacent upper one... and not only that! Can you control such a delay separately for each driver?
Can you set a different crossover with a specific Q-factor for each driver? Can you set the same polarity for all drivers?
These are the "small", but important (mandatory) conditions, if you have selected suitable drivers and defined optimal frequencies for the cross-slopes?......?

I think you are overcomplicating a very simple measurement.

All I did was take two filters, each with an 18 dB/octave slope, sum them together, and then measure the result with my sound card and ARTA.

I am not building a multi‑way loudspeaker here. I am simply checking what the electrical sum of two filters looks like in a real circuit.

There are no driver impulse responses, no time alignment, no polarity reversals, and no separate delays needed for this test. The miniDSP outputs are already time‑coherent because they come from the same internal clock and the same firmware processing.

Your comments about triangles, driver delays, Q‑factor per driver, and polarity conditions would be relevant if I were designing a full loudspeaker system with physical drivers in a box. But that is not what I am doing here.

My measurement is valid for what it is: a simple check of the summed filter response. No medications are needed — just a clear understanding of the measurement setup.

 

[KL....]

I think you are overcomplicating a very simple measurement.

All I did was take two filters, each with an 18 dB/octave slope, sum them together, and then measure the result with my sound card and ARTA.

I am not building a multi‑way loudspeaker here. I am simply checking what the electrical sum of two filters looks like in a real circuit.

There are no driver impulse responses, no time alignment, no polarity reversals, and no separate delays needed for this test. The miniDSP outputs are already time‑coherent because they come from the same internal clock and the same firmware processing.

Your comments about triangles, driver delays, Q‑factor per driver, and polarity conditions would be relevant if I were designing a full loudspeaker system with physical drivers in a box. But that is not what I am doing here.

My measurement is valid for what it is: a simple check of the summed filter response. No medications are needed — just a clear understanding of the measurement setup.

Thank you, @LSPhil, why have you done this and what will it achieve other than it can be calculated?

 

[kimmosto]

@robh36062
First round of LR12 XO tests done with XO frequencies of 50, 71, 100, 141, 168, 200 and 282 Hz. Overall impression is that it's nicer and not so critical to XO frequency than LR24. Not a surprise due to smoother transition slope and shorter delay with equal XO freq. All XO options up to 200 Hz were audible, but at least I would not mind crossing at 200 Hz. It has been dynamic and "good enough" with speakers also in practice. Lower XO -> more audible.

Test signal. 10 Hz fundamental does not stress much ears so it's possible to try to catch some lame skin effects too

 

[LSPhil]

Thank you, @LSPhil, why have you done this and what will it achieve other than it can be calculated?

First of all, so that I don't rely on incorrect opinions from other people who try to explain things based on their own flawed logic. Secondly, for myself — to have a reliable reference when building my own loudspeakers. For many years I have been searching, and I have finally found the best designs for me. These include acoustic 3rd-order Butterworth (or even 5th-order), heavy aluminium woofer membranes, and a construction that allows me to listen to my favourite music in the kitchen — coming from the living room.

 

[Palladium]

It is probably obvious that I am a strong advocate of matched-delay subtractive crossovers. However, having listened to a variety of different crossovers -- Gaussian-derived MDS, Bessel-derived MDS, Butterworth and LR both passive and active -- my opinion has evolved to "it really doesn't matter very much".

It is technically trivial to do match delay on actives?

 

[gberchin]

All I did was take two filters, each with an 18 dB/octave slope, sum them together, and then measure the result with my sound card and ARTA.

Exactly that, and more, are summarized here.

 

[gberchin]

It is technically trivial to do match delay on actives?

Requires DSP, because delays are necessary. As for "trivial", depends upon your definition of the word. No more difficult than implementing a traditional crossover.

 

[KL....]

First of all, so that I don't rely on incorrect opinions from other people who try to explain things based on their own flawed logic. Secondly, for myself — to have a reliable reference when building my own loudspeakers. For many years I have been searching, and I have finally found the best designs for me. These include acoustic 3rd-order Butterworth (or even 5th-order), heavy aluminium woofer membranes, and a construction that allows me to listen to my favourite music in the kitchen — coming from the living room.

Thank you, @LSPhil, reasonable :=)

Re allows me to listen to my favourite music in the kitchen — coming from the living room

• It can be suggested that this is a Harmonic/Dispersion (Lingering, Step Response related) characteristic/behaviour and occurs with 1st order XOs although as @Trifonov Audio suggests/mentions is not easy as (everything) requires (Pre) Matching/Alignment/Analysis/Construction.... which is what you have suggested/worked out, for you, that is acoustic 3rd-order Butterworth (or even 5th-order), heavy aluminium woofer membranes, and a construction that allows.... .... well done :=)
• It also suggests that your Kitchen is acting as part of your Room Treatment in a Diffuseing manner/way, that is reduceing/controlling Modes/Nodes....

 

[Trifonov Audio]

I think you are overcomplicating a very simple measurement.

All I did was take two filters, each with an 18 dB/octave slope, sum them together, and then measure the result with my sound card and ARTA.

I am not building a multi‑way loudspeaker here. I am simply checking what the electrical sum of two filters looks like in a real circuit.

There are no driver impulse responses, no time alignment, no polarity reversals, and no separate delays needed for this test. The miniDSP outputs are already time‑coherent because they come from the same internal clock and the same firmware processing.

Your comments about triangles, driver delays, Q‑factor per driver, and polarity conditions would be relevant if I were designing a full loudspeaker system with physical drivers in a box. But that is not what I am doing here.

My measurement is valid for what it is: a simple check of the summed filter response. No medications are needed — just a clear understanding of the measurement setup.

So - there are no problems for designing filters with good Time Domain!
I apologize for my instructions, explanations and unsolicited "training".
If you have specific questions on the "most important" topic - I am at your disposal.....

 

[KL....]

So - there are no problems for designing filters with good Time Domain!
I apologize for my instructions, explanations and unsolicited "training".
If you have specific questions on the "most important" topic - I am at your disposal.....

Thank you, @Trifonov Audio, no need, (for me) very much appreciated, very very good/fabulous.... please (freely) continue :=)

 

[kimmosto]

Few decades ago local DIY influencer had some models with quite good timing, but without 1st order acoustical slopes.
This tiny image was scanned from a book so it looks too clean. XO is probably close to Le Cleach with 3rd order Butterworth slopes and inverted tweeter. Delay of tweeter is a bit longer than in Le Cleach's spec. to lift inverted peak of above 0 Pa.

I have listened that speaker many times for several hours at a time, but timing is not it's major strength or weakness.

Step of full range with equal XO concept would look something like this in theory.

 

[Thomas_A]

I had seen that a while ago. I've been designing my (passive) system crossovers exactly that way since my first 3-way about 25 year ago. I had no clue to the specifics related to distortion, though I made assumptions about it. I did it because that was the best way to approximate the ideal lowpass SPL response of a midrange or woofer in the pass-band and into the stop-band. I've only recently started doing distortion measurements with new microphones and software so I may investigate this. Good to know that what I did also resulted in much improved distortion results.

I'm using DSP for my main system now so that's unecessary or so I think. It would be interesting to see if a combination of passive peaking SPL correction along with DSP would make a difference in the distortion profile. My initial thought is that it would not. Past practice by many was often just using the passive filter to push any SPL deviation into the stop-band, often not very far down. I found it better to reduce them directly as shown at that link by Purifi Audio. Simply pushing into the stop-band still left "remants" of it. No doubt that often had to do with the increased cost of a passive crossover due to the additional components required. I had no limitation, so I used whatever was required.

The physics at work reducing this type of distortion occurs between amp and driver. DSP filter before the amp stqge does work.

 

[Waveform Fidelity]

Few decades ago local DIY influencer had some models with quite good timing, but without 1st order acoustical slopes.
This tiny image was scanned from a book so it looks too clean. XO is probably close to Le Cleach with 3rd order Butterworth slopes and inverted tweeter. Delay of tweeter is a bit longer than in Le Cleach's spec. to lift inverted peak of above 0 Pa.
View attachment 535335
I have listened that speaker many times for several hours at a time, but timing is not it's major strength or weakness.

Step of full range with equal XO concept would look something like this in theory.
View attachment 535336

Something wrong with ^^ the Y axis is in Pascals (sound pressure) - once the driver has reached the end of the step, there is no further motion, so the SPL decays.

 

[Trifonov Audio]

Thank you, @Trifonov Audio, no need, (for me) very much appreciated, very very good/fabulous.... please (freely) continue :=)

And thank you for your attention!

 

[thewas]

Just some reminders.

Step response is not a good way to visualise time or phase alignment in regard of audibility, group delay should be used.

The typical electrical 6 dB/octave crossovers are usually not sufficient, the total acoustic slope which counts for the filter characteristic is usually higher so it was a bit of a misunderstood myth.

What makes 6dB filters so special – Just Blogging

Is it possible to make a first-order crossover in a real-world situation? – Just Blogging

Also such a shallow slope has also significant disadvantages on the vertical radiation unless a coincident design is used, even the "time incident step" won't be such anymore at vertical angles and thus for example from reflections from the floor and the ceiling.