Questions about phase response

[Wilk]

I have a few phase based questions

Phase response: What exactly are you guys looking for in this chart when designing crossovers? Most people on other forums truly to seem to ignore it. You guys are more detailed in your designs so maybe I can get a decent answer here.

Also what about the number of phase changes? How many, where, why? For instance I have been fiddling with Jon's Modula NeoD CC crossover. I have gotten decent frequncy responses with 1 phase change on one design, and 2 on another one. Why is that? Which one is preffered?

Phase Impedance: I understand that the phase will tend to follow the impedance, and you want to minimize the swings near low impedance points to make it an easier load on the amp. Ids there any quidelines on the size of the swings, or the phase should stay between here, and here?

[Jed]

https://www.htguide.com/forum/showpost.php?p=353375&postcount=150

And check out the rest of the thread because phase was discussed a bit IIRC.

[JonMarsh]

Jon

I have a few phase based questions

Phase response: What exactly are you guys looking for in this chart when designing crossovers? Most people on other forums truly to seem to ignore it. You guys are more detailed in your designs so maybe I can get a decent answer here.

Also what about the number of phase changes? How many, where, why? For instance I have been fiddling with Jon's Modula NeoD CC crossover. I have gotten decent frequncy responses with 1 phase change on one design, and 2 on another one. Why is that? Which one is preffered?

Phase Impedance: I understand that the phase will tend to follow the impedance, and you want to minimize the swings near low impedance points to make it an easier load on the amp. Ids there any quidelines on the size of the swings, or the phase should stay between here, and here?

In the Modula NeoD CC there should be one complete phase rotation from high to low, with the midrange driver wired out of phase. Properly executed, a Duelund three way is a single rotation all pass network summing to unity with the drivers in phase at all frequencies (well, over the range that the drivers can operate, and at least down to -20 dB). I.E., you have to select drivers carefully for this configuration.

To repeat some of what was posted in the previous Modula NeoD thread,
================================================

While the greatest interest in developing the Duelund style crossover holds for the larger systems, a brief examination shows it holds potential for this center channel HT design also.

The most interesting aspect is that if properly executed the drivers are all in relative phase with each other at any frequency, with a net total phase rotation of 360 degrees from the low bass to the treble, similar to the phase rotation of a two way LR-4 crossover, but with lower peak group delay.

For a first look, a damping coefficient of 2.828 was used, with a nominal center frequency for the midrange of 1500 Hz.





This proposed transfer function should support playback levels over 100 dB; notch filters will be required to suppress the midwoofer breakup as well as the midrange breakup, but as the nominal transfer function level will be below - 24 dB, impact on amplitude and phase for the overall transfer function should be acceptable.


Here's a link to a page with a walk through on the Duelund crossover concepts written by Steen Duelund.

On Building a Duelund Speaker



This is the example posted there for amplitude and phase calculated in MathCAD, normalized to 1 kHz (the same aleph coefficient as I'm using in the Modula NeoD CC design, but not the same target frequency for the center of bandpass).




==============================================

At the center of the bandpass element (midrange), the woofer and tweeter contribute enough output to raise the midrange level by 2 dB. Which means both must have some output capability to 1500 Hz.

==============================================

Here's a link to a PDF you can download for "The Art of Building a loudspeaker to the end".

duelund-filter.pdf

There are easier ways to do this, of course, once you have some insight into the Duelund concept. The equations for filter components don't seem to take into account BSC or other equalization; what I do is use MathCAD to generate my target function, use similar filters (Bessel constant delay) for HP and LP as starting points. Works pretty well.

==============================================

If the original driver measurements don't have accurate phase data, then a crossover simulation will be quite flawed (GIGO).

JonW had some questions about this and what to expect to see.

This plot is representative of what one might see with a wideband midrange, and all drivers in principle should look somewhat like this, with variations in behavior at the upper breakup/roll-off point. On the bottom end, the transfer function is normally some form of 2nd order high pass, with Q described by Qtc in the cabinet.



In the area of relatively flat frequency response, you should have a pretty constant phase value, with slight drifts up or down depending on how the driver SPL is trending. Errors usually come up from incorrect transit delay phase compensation making the measurements. Some software, like Fuzzmeasure 2.0, can be set to do phase compensation manually, or to convert a measurement to a minimum phase version. Then you'll still have to line up driver acoustical offsets - something fairly straightforward in LspCAD, but I don't know how Sound Easy handles that.

Hope this helps a bit.

~Jon

[Wilk]

I have been trying to work on the RS180-4 version of Jon's center channel design on the side. Instead of just waiting around for him I figured this would be a great chance to try and learn something. I can design it on my own, and see how close I got to his design. (and then build the design he came up with :lol: :lol: :lol: )

I measured all the drivers in a PE sealed box using Speaker Workshop. Imported the FRD, and ZMA files into PCD, and started playing around.

I kept very close to the crossover points, and slopes Jon has explained in the build post, and used his posted crossover as a starting point. I got to a point where it started to look like something that might work so I figured I would post it here, and ask for some constructive critisim, and ask a few questions.

I am not looking for anyone to redo this sim, just some thoughts on what looks good, and what looks bad, and a few pointers.

Does anyone got any decent ideas on how to raise the impedance in the 800 to 1500 Hz range? it gets a little low right in that area.

The entire freq plot has a slow 1-2 db rise across it. Is there a way to cut this down a little other than using additional padding. The padding just lowers the exisiting impedance dip?

Jon uses a 300 nf cap in his design that is in parallel to the first stage of the lowpass section of the filter for the RS52 (C8 I think). Is there a way to model this in PCD, or do you just make a 15 KHz notch filter with one of the parallel sections that PCD gives you?

[JonMarsh]

PCD is the FRD tool for crossover design? There's a lot of configurations one might come up with that it can't model; if you measured in Speaker Workshop, might be better to stay in that for modeling; that's what I used for the first pass of the crossover before I could get ThomasW's Praxis dongle, and then find my own. It would help to see your network to evaluate what you're doing. And yes, you're GOING to have to pad it down, because a series setup with two 4 ohm woofers wired in series will be about 3 dB less sensitive, all things otherwise being the same. That impedance dip in the midrange is a bit "odd".




This was my impedance plot- it's not "right" either, but it's a result of the "fudging" you have to do with the drivers not mounted with their acoustic centers all in the same plane.




The midrange and tweeter networks should be nearly identical to the ones I posted in the final design, but with different Lpad values.



Note how you've got some peaking in the tweeter response over the summed response of mid and tweeter at about 5 kHz- that's a likely sign of crossover phase misalignment or measurement data that isn't quite right. It's symptomatic of destructive interference- which shouldn't happen with the drivers in matching relative phase at each frequency.

In my experience, it may be best to first get the treble and midwoofer acoustic transfer function fig correctly, and make sure they're adding properly, and then go in and finish the midrange fitting.


Remember this description from one of the earlier posts, if you haven't generated an acoustic target with MathCAD or Excel and imported it as a target-

==============================================


This was undertaken as a mental diversion from my normal responsibiities- the basic results and method posted below for those interested in dabbling with this without a major time and intellectual commitment.


This does not include having to use driver data with baffle step fall off and the other usual irregularities- those are dependent on the drivers chosen and the cabinet design. This is merely to describe the basic filter impelmentation and targets - much like the difference between a text book LR4 two way, and what you have to do in the real world to pull off the complete design.

The basic target functions are derived from the LspCAD Flat Delay functions- the LP target is the Flat Delay 4th order with a Fs of 1000 Hz, which hits the target -18 dB level at ~1.6 kHz, and the HP Flat Delay target with Fp set to ~2200 Hz, again hitting the -18 dB crossover target at ~1.6 kHz.

The midrange target is somewhat more difficult, as it must be generated from the combination of HP and LP 2nd order Flat delay targets; the LP set to 1400 Hz, the HP to 1600 Hz, and the target level adjusted for the peak at ~ -3 dB.

Components can be calculated from the Duelund equations, or use an LR4 and LR2 networks as the starting point.

The basic network result is shown next:



As these are idealized drivers, with simple impedance, there are no zobel networks or equalization used in this demonstration. The network configuration for the bandpass is my preferred configuration for 2nd order bandpass. I do not recommned cascading two conventional filters.

After the LP and HP networks have been optimized to the Flat delay targets, and the bandpass likewise for the synthesized target, response is close, but still with a dB or so ripple- this is because the flat delay target is not exactly equivalent to the chosen Duelund (aleph = 2.828), but very close. The final step is to optimize ONLY the bandpass components, and setup the target as flat response between 200 Hz and 10 kHz, to bring the midrange passband into compliance.


The result will look like this-




The relatively low Q crossover transistions have higher impedance than most "conventional" filters in the transistion region, and the network impedance even with a 4 ohm target woofer was at or above four ohms at all frequencies. How that will fare in a network with BSC and driver equalization is yet to be seen, but my Force sense is optimistic...

Note that through out the overlap range the drivers are in phase (midrange is wired backwards), and reversing the midrange produces the expected nulls.

Note the phase characteristic- All Pass- with one 360 degree phase rotation between 20 Hz and 20 kHz.



===============================================

Having a bit of roughness in the curve of your simulated phase rotation is probably to be expected- but take care of the amplitude fit, and the phase should take care of itself, for the most part.




Notice how the midrange and tweeter have a large area where their contributions sum substantially, with the nominal "crossover" points being each 6 dB down in level? This wasn't the final version of the network; in it the midrange combining is better, due to a better phase fit. Because we don't have the drivers mounted with aligned acoustic centers and offset faceplates, we have to modify the midrange repsonse target a bit- it's best once midwoofer and tweeter are dialed in, to lock them, and go for optimizing the flatness and phase in the transition region.

And that RS180 will need a CE notch filter similar to the NatlieP to clean up the HF cone resonance. :W

~Jon