Delay and Phase with 3-Way DSP Crossover

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi!


My speakers are 3-way, with a BMS4590 co-axial horn and 15" bass-reflex, as seen in the picture below.
The pre-amp / digital crossover is a Sitronik Lucius 6K-V-2U driving a stereo THEL / DIY 400W Class AB for the bass drivers and two DIY Pass F5 for the horns. Crossover frequency is at 304Hz B/M and 6000Hz M/T, both 24dB/oct Butterworth.
full



While I'm fairly happy with the sound, I feel there might be some improvement to be made, particularly regarding the "livelyness".


I've set the delay of the drivers using ARTA in 2-channel mode and basically delaying the mid and tweeter so that the impulse from all three drivers arrive at the same time at the microphone, and phase by adjusting the polarity so that the first impulse of all 3 drivers is "positive" (polarity of the mid is reversed).


However, today a colleague suggested adjusting the delay as follows.


1. reverse the polarity of the mid driver so that it's "wrong"
2. play a sine wave at exactly the B/M crossover frequency and obersve the level at the microphone
3.adjust the time delay of the mid range until the level drops to a minimum, indicating cancellation
4. reverse polarity again
5. repeat with next driver "pair"


Has anyone tried this method?
Any other suggestions?


Cheers


Es
 
That method is ok. but don't you have a modern and cheap measurement mic and program? All speakers need equalization of responses too, to get acousic responses that match!

You should also try 12dB/oct xo, then transition is smoother, ie. speaker sound more coherent through xo range. The challenge of this kind of speakers is that the bass radiates omnispherically and the horn shoots sound only forwards like a cannon.
 
That method is ok. but don't you have a modern and cheap measurement mic and program? All speakers need equalization of responses too, to get acousic responses that match!

You should also try 12dB/oct xo, then transition is smoother, ie. speaker sound more coherent through xo range. The challenge of this kind of speakers is that the bass radiates omnispherically and the horn shoots sound only forwards like a cannon.
Hi!

Which method are you referring to? The cancellation method or "regular" delay method in ARTA?
I have a microphone, frontend and ARTA.


12dB I will look into! What about the crossover frequency B/M? I can't get the horn to go lower than 300Hz, but might it be better to move up the range?
Also, would a Linkwitz-Riley filter be better than a Butterworth?




Cheers


Es
 
diyAudio Moderator
Joined 2008
Paid Member
1. reverse the polarity of the mid driver so that it's "wrong"
2. play a sine wave at exactly the B/M crossover frequency and obersve the level at the microphone
3.adjust the time delay of the mid range until the level drops to a minimum, indicating cancellation
4. reverse polarity again
5. repeat with next driver "pair"
This can still leave the delay incorrect. Is it possible for you to take measurements of each driver while keeping the same timing reference between them?
 
All in-room tests and measurements below 500hz are plaqued with room effects. One should be able to go outdoors or to a large hall with the speaker on ladders.

I use REW and UMIK-1, so I don't have reliable delay timing for different sweeps. I set my dsp-speakers xo using the method TheNuge asked about (swapping polarity). It is not an easy method, because to get good summation/cancellation, acoustic responses must be symmetrical at least one octave up and down from xo! But when that is done, speakers sound just wonderful!

With this method, don't just look at spl but also step response changes! It happens rather easily that the delay is one full cycle off.
 
...
12dB I will look into! What about the crossover frequency B/M? I can't get the horn to go lower than 300Hz, but might it be better to move up the range?
Also, would a Linkwitz-Riley filter be better than a Butterworth?


Cheers


Es

If no other constraints for crossover frequency you should aim where directivity of the woofer matches directivity of the horn. Otherwise you'll get a step in horizontal polar response which is not optimal, smooth transition is better. I would guess it is somewhere above 600Hz where the 15" starts to narrow its response.

If you dare, take full set of measurements, start VituixCAD or some other crossover CAD capable of showing polar responses and you'll find the delays, crossover frequency, filter steepness and all in no time ;) Make a crossover, listen few weeks if its good or needs adjustment, rinse and repeat. There are helpfull guides how to measure here, at the bottom Software

Have fun!:)
 
I can select Bessel or Butterworth in 12, 18 and 24 dB or Linkwitz-Riley in 12 or 24 dB - I haven't tried them all ;)
The reverse phase nulling at crossover frequency is what I use with great success. Of course it is best done outdoors on a quiet day (unless you have access to an anechoic chamber) with the mic set at the normal listening distance from the speaker.
The Bessel crossover has the nice property that the phase change throughout the crossover frequency region is gradual and linear. This makes for effective time alignment in the crossover region. Butterworth and L-R crossover have a more rapid phase change through the transition region. The drawback with Bessel filters is that the initial roll-off rates is not as rapid as other alignments, although the ultimate roll off rates are the same for all alignments of the same order. This means it is good to choose drivers which are well behaved outside of their desired frequency band as the Bessel crossover will roll off only gradually initially. I don't think horns and large paper cone woofers qualify as "well behaved outside of their frequency band" though. If time alignment is your goal, Bessel is the best choice IME.
You probably want to go with the higher order (4th order, 24 dB/Octave) to avoid out of band issues with your choice of drivers.
 
diyAudio Moderator
Joined 2008
Paid Member
a more rapid phase change
This is the problem with the reverse null process.

Here are some examples of two drivers crossing at 500Hz, these are phase plots. All three examples will give a good null at 500Hz, but only one is correct. One is at +1ms, and one is at -1ms compared to the first one.
 

Attachments

  • Screenshot from 2020-01-17 12-27-06.png
    Screenshot from 2020-01-17 12-27-06.png
    47.4 KB · Views: 198
  • Screenshot from 2020-01-17 12-28-04.png
    Screenshot from 2020-01-17 12-28-04.png
    47.1 KB · Views: 208
  • Screenshot from 2020-01-17 12-28-51.png
    Screenshot from 2020-01-17 12-28-51.png
    48.4 KB · Views: 210
This is the problem with the reverse null process.
...
One is at +1ms, and one is at -1ms compared to the first one.
1ms delay is 34.4 mm driver alignment offset. This is a huge error margin. My miniDSP allows increments of 1mm or 0.005ms. The phase alignment between my mid-range and tweeter is typically to within 1mm tolerance at 2.5 kHz. It makes a dramatic difference to the magnitude of the out-of-phase null. The difference between the best mid/tweeter null and ±1mm can be 10 dB or more in the differential output. I don't see the problem you do at this level of discrimination. If anything, with 1mm increments, the phase null is too sensitive. Unless I am being super fussy, ±5mm is close enough for mid/treble phase nulls. For bass/mid ±5cm (±0.145 ms) at 200Hz is good enough. This is still nearly an order of magnitude sharper than the tolerance of your example.
When super matching the magnitudes and fine tuning the delay, the way the differential output drops like a stone close to the null, is truly dramatic. Of course it is very sensitive to small movements around the acoustic environment, but it is empowering to know how close is possible. Naturally it is audible in the final overall speaker alignment. No passive crossover comes close in my experience.
 
My speakers are 3-way, with a BMS4590 co-axial horn and 15" bass-reflex, as seen in the picture below.
The pre-amp / digital crossover is a Sitronik Lucius 6K-V-2U driving a stereo THEL / DIY 400W Class AB for the bass drivers and two DIY Pass F5 for the horns. Crossover frequency is at 304Hz B/M and 6000Hz M/T, both 24dB/oct Butterworth.

While I'm fairly happy with the sound, I feel there might be some improvement to be made, particularly regarding the "livelyness".

I've set the delay of the drivers using ARTA in 2-channel mode and basically delaying the mid and tweeter so that the impulse from all three drivers arrive at the same time at the microphone, and phase by adjusting the polarity so that the first impulse of all 3 drivers is "positive" (polarity of the mid is reversed).

I'm not quite sure from your post whether you adjusted the delays and polarities of the driver with the crossover included or not. It is the crossover filters themselves that impart delay, and that delay is often much greater than the delay you get from physical offsets in the loudspeaker. This is why many loudspeaker designers use the reverse-polarity null in combination with an even order crossover topology. It allows them to check the phase alignment (but not time delay alignment) in the crossover transition band.

When I design a crossover I use a model of the loudspeaker that includes all the sources of phase, both driver offset and the crossover filters. When you construct such a model correctly, you can model the effect of the crossover filters on the phase response while you are designing the crossover. Trying to fudge things afterwards always seemed a bit dubious of an approach to me.

You should know that the ear is not sensitive except when there are very large time delay differences between drivers on the order of a millisecond. You would get that if you had 0.3 meters offset between the acoustic center of the drivers. But when high order filters are used, the group delay can easily have a peak that gets close to or exceeds this limit, and then you might have some audible effects. If you stay below this 1msec difference in group delay threshold you will not hear anything and it will not influence anything having to do with "liveliness" in the sound. At the same time you need good phase alignment, but the phase can be well-aligned at the same time that there are group delay difference between drivers.

I'd like to see a good set of measurements of your loudspeaker, especially done off-axis, because it is more likely that there are issues with the frequency response, and that could be the source of the perception that the sound is "dull". My guess is that the culprit is the horn itself - it looks to me like it will have quite a high directivity index at higher frequencies, e.g. in the "treble". This means that the sound is largely beaming straight at the listener, and nothing goes in any other direction. What you hear is both the direct sound AND the "room sound" which is all the off-axis acoustic radiation from the loudspeaker bouncing off the room surfaces (floor, walls, etc.) and then reaching the listening position. The "room sound" can be as high as -6dB with respect to the direct sound. The midrange and woofer will radiate quite a lot to the side because they will act as a monopole, but not the horn. So the power response of the loudspeaker will not be uniform - in the frequency band of the horn the total sound power output will be much less and this is probably (in my opinion) the source of the "dull" sounding response. This effect will be more pronounced in smaller rooms than in larger ones. It's very difficult to "fix" this problem because it has been built into the loudspeaker when you chose to use the horn of this type. No fancy FirstWatt amplifier, coaxial compression driver, or any crossover tweak, can change it. Only better design choices.
 
Last edited:
1ms delay is 34.4 mm driver alignment offset. This is a huge error margin. My miniDSP allows increments of 1mm or 0.005ms. The phase alignment between my mid-range and tweeter is typically to within 1mm tolerance at 2.5 kHz. It makes a dramatic difference to the magnitude of the out-of-phase null. The difference between the best mid/tweeter null and ±1mm can be 10 dB or more in the differential output. I don't see the problem you do at this level of discrimination. If anything, with 1mm increments, the phase null is too sensitive. Unless I am being super fussy, ±5mm is close enough for mid/treble phase nulls. For bass/mid ±5cm (±0.145 ms) at 200Hz is good enough. This is still nearly an order of magnitude sharper than the tolerance of your example.
When super matching the magnitudes and fine tuning the delay, the way the differential output drops like a stone close to the null, is truly dramatic. Of course it is very sensitive to small movements around the acoustic environment, but it is empowering to know how close is possible. Naturally it is audible in the final overall speaker alignment. No passive crossover comes close in my experience.

Hi Bon, I don't know what miniDSP unit you're using, but I guess you have to be using a 96kHz plugin, to get the small delay increment you mention.

One thing to be aware of is that the delay increment cannot be finer than 1 sample of processing...1000 msec / 96000 samples per second, or 0.01ms with a distance equivalence of 3.6mm.
So really, best achievable physical alignment at 96 kHz is within +/- 1.8mm, unless you just get plain lucky with physical alignment falling right on a sample.
But that's dang good imo :)

Just know to ignore any miniDSP delay readout increment that reads finer than .01ms at 96kHz, or .02ms at 48kHz.

(I routinely use a 48kHz sampling rate, with double that distance error, for xovers as high as 6.3kHz.
1 sample, or .02ms delay increment, definitely makes an easy to hear change in pink noise, but i've never been able to hear it with any music. )

I think AllenB did a great job of showing the problem with using a sine wave null test for distance alignment.
Yes, you can get very deep nulls that way ...but nulls that can pinpoint timing at multiple locations, 1 cycle apart.
I'd add some bandwidth limited pink around the chosen sine frequency for a null test. The sine wave becomes the periodic indicator...and the pink summation would show which period produces the least total SPL.

But really, just measuring each section independently and getting phase traces to superimpose, like AllenB's first graph, is definitely the way to go, imo/ime.
 
Hi Bon, I don't know what miniDSP unit you're using, but I guess you have to be using a 96kHz plugin, to get the small delay increment you mention.

One thing to be aware of is that the delay increment cannot be finer than 1 sample of processing...1000 msec / 96000 samples per second, or 0.01ms with a distance equivalence of 3.6mm.
So really, best achievable physical alignment at 96 kHz is within +/- 1.8mm, unless you just get plain lucky with physical alignment falling right on a sample.
But that's dang good imo :)
I use a modified C-DSP 8x12. It operates at 192 kHz/24 bit, so that is where my ±1mm time alignment comes from.
 
Yes, you can get very deep nulls that way ...but nulls that can pinpoint timing at multiple locations, 1 cycle apart.
I'd add some bandwidth limited pink around the chosen sine frequency for a null test. The sine wave becomes the periodic indicator...and the pink summation would show which period produces the least total SPL.
I only measure at the listening position. it should still be close within the typical horizontal window for vertically aligned drivers. I like your suggestion regarding added pink noise, especially when measuring in-room.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.