What would call the starting frequency, and ending frequency, for establishing that rolloff?
I was trying to figure that out. So I just took the point where it crosses 65 dB which looks like 4500 Hz, then went left an octave. Best I could do.
Thank Pano, but I dunno...that picks the steepest part of the rolloff after alot of shallow rolloff has already occurred..
I mean, looks to to me like rolloff has begun all the way down at 400Hz....???
I had been working on a post as to what I was seeing the rolloff, and stopped to see what's going on before posting...
I'll go ahead and post it anyway now...all about learning huh?
I mean, looks to to me like rolloff has begun all the way down at 400Hz....???
I had been working on a post as to what I was seeing the rolloff, and stopped to see what's going on before posting...
I'll go ahead and post it anyway now...all about learning huh?
I don't think Bessel orders can be compared to Butterworth or LR orders.
They don't follow a 6 dB per octave rolloff pattern increase as their order increases,
which is why their phase rotation doesn't increase 90 degrees per order, as well.
I mean look.... that '16th order Bessel squared filter' posted starts rolling off at < 500Hz, and is down <1 db at 1kHz, 5 dB at 2kHz, and only 22 dB at 4kHz,...4 octaves later. That's a long way from steep !
I'm no expert on bessels at all...bewildering array of options for them in prosound dsp's.
Anyway, below is a grab from the chart previously posted.
It has the same BW 1st order LP, and lin phase LR 16 LP as before.
And I added two bessel 16th LP's, designed to have matching xover points to the BW and LR, so that slopes can be compared. dsp outputs measured, not sims.
The red trace is bessel 16th to match the BW -3dB at xover.
The green trace is bessel 16th to match LR -6dB at xover.
Looks to me like a bessel 16th is more akin to first or second order.....when you look at the entire rolloff.
Attachments
Last edited:
Thanks for all that Allen,
So far, for every multi-way I've built, steep xovers have made for smoother sets of polars, than trying 4th order or less.
As you say, the varying horizontal and vertical coverage patterns vs frequency, of two drivers summing, is always a matter of compromise and full of considerations.
But I think the simple fact that two drivers summing come from different geometric points in space, tends to almost invariably🙂 favor a steep xover, once any reasonable pattern matching has been achieved.
Anyway, that's just my experience so far. But I do want to explore shallow xovers in depth again with the synergies I've built.
To see if the closer acoustic centers inherent in the design let shallow work as well as the steep (that i currently have in place).
With regard to traditional filters having the ability to slide up or down and keep phase relationships intact....
yes, can fully agree, if they are fully complementary xovers.
LR is the only fully complementary IIR xover that I know of...and not being an IIR enthusiast, I need to ask are there others?
In trying to use complementary IIR xovers to slide up and down, I've found the need to cascade LR's per Linkwitzlab once above a two-way.
(And the need to add an allpass filter for each way above 2-way.) Woofer crossover & offset
Virtually every regular dsp i'm aware of is set up to apply xovers as parallel signal paths, which works fine for linear phase.
The only ones that can handle cascading xover filters appear to be the open architecture types like Symetrix, Q-Sys, etc.
I think in practice, the idea of having easily sliding IIR xovers once above a 2-way, is really not so doable unless an advanced dsp is used. (one that's bound to have linear phase capability.)
You are forgetting the most important character of High order Bessel function, the linear phase shift. With 8th order the phase shift remains linear down to -60db. At first this means nothing, but with more attention, you can realize that it is a constant delay, hence the relative phase shift is 0°. The signal gets attenuated without any relative phase shift. This why if it is subtracting the same delayed signal, it wields a second order high pass without any phase shift in attenuation.
LR is just a special case of in-phase allpass crossover. They are derived by subtracting the output of two equal cascaded highpass or lowpass filters from the output of a single allpass of same order and poles like the aforementioned filter. Only if Butterworth parameters are used the final filter slopes are symmetrical. All other cases are asymmetrical but always sum flat and they are always in phase synchronity. But the LR case has the best lobing behaviour. But the other ones can still be interesting depending on the use case. If you use two 2nd order Bessel lowpass filters and one 2nd order allpass you end up with a crossover that is 4th order lowpass but only 2nd order highpass. The advantage is lower group-delay distortion than LR. One can of course use two Bessel 2nd order highpass filters and a Bessel allpass and end up with a 4th order highpass with second order lowpass crossover. The crossover is always at -6dB. Therefore different pole frequencies have to be used for the same crossover frequency for either of the asymmetrical cases. If it weren't that hot at the Moment there would be more details that could pop up but maybe later on …….
Regards
Charles
Last edited:
Mark, my point wasn't that shallow or steep is better. It's that two devices, which themselves are imperfect, can work together to create a more desirable system directivity, notwithstanding but including their non-coincidence based interaction. The result being entirely system dependent.
I don't identify with Linkwitz-Riley or any basic filter types. I use what works and complementary is no problem. My DSP has plenty of options, but passive has no such limits.Mark100 said:
No, FIR fails once you get off-axis (as will any electrical filter).CharlieLaub said:
But it does at least achieve it on axis where it has the most effect. Others don't achieve it on any axis at all exactly like you have stated.
Regards
Charles
I get all that Allen...and i applaud the folks who are willing to juggle all kinds of variables to find the best combination for their purposes.
Me, I'll continue stick to and recommend relatively simple straightforward techniques that use steep complementary linear phase xovers.
I believe if nothing else, the measurements I've posted from various builds speak for the techniques.
BTW, where did you pull the second quote from????
This one...you can create a filter with perfectly flat group delay everywhere
Yes, you've posted measurements, in another thread, which you "believe" prove your point when in actual fact they do the exact opposite. 😉 I applaud your self-belief, but I find your arguments totally random at times.
Thank Charles, i get some of what you say, but most is beyond.
I do like to run transfers of various IIR xover / filter implementations, and study papers like the Rane Notes..... but mostly just out of curiosity ...
Very cool, the knowledge you guys amassed re IIR and passive.
An understanding of a standalone phenomena is a very helpful thing. People search long and hard for such information. If you can identify the aspects of your horns that demonstrate the effect in isolation, that's a good thing. Then it can be considered in any design as one of the requisite sliding compromises.
On the other hand adding that it is invariably a fix, is hyperbole. Not saying that it isn't can also be misleading. Nutting out these inconsistencies and ambiguities is an aspect of peer review. Where does the buck stop?
Post#19 by CharlieLaub
Ok, fine. I can accept all that.
Other than i would ask when you quote and name one person in a post, that when you pull additional quotes from additional people, you also name the source along with those quotes.
I will try to carefully document any effect in isolation before over describing its effect as invariable. Or just simply state, that so far I keep getting these results with this effect.
That should prune the 'hyperbole' to which you object.
And to hold both of us to the same standard.....
A few of your quotes from this thread:
"OK, here's a starting point for LR96, uF and uH, the impedance stays above 8 ohms."
"I don't identify with Linkwitz-Riley or any basic filter types. I use what works and complementary is no problem. My DSP has plenty of options, but passive has no such limits."
When you say passive has no such limits vs DSP's plenty of options...
....and when much of the thread has referred to the unique ability of linear phase xovers to provide both steep xover and flat phase....
.... and when i asked what was the phase was that LR96 you were building and i received "Just what you'd expect it to be"
Well, i expect it to have completely butt ugly unusable phase rotation and group delay.
To the point i think it's pretense to even propose a passive LR16 as viable, other than for speculative debate purposes.
If i may borrow one of your quotes and use it as my own..
".....I see no handle, I can't tell whether you're speculating or claiming so all I can say is I disagree, prove it."
So how about showing both sides of a 96 dB/oct complementary passive xover, that sums to both flat magnitude and flat phase across the full spectrum.
One that has a xover summing range width in the same ballpark as LR 96's, not one that sums over octaves, and then finally rolls steeply down.
Allen, not trying to start anything...just saying, let's both be real, both be fair...