the option to use an allpass instead is still there.
for more delay bandwidth, you use multiple sections.
Yep, i can certainly understand how one would like to avoid going "all in" with multi-channel FIR and amplification.
And I don't see any point in imitating analog or IIR filters with FIR either...
(if we are speaking of xover filters).
I do think it is most often "best practice" to use linear phase xovers in lieu of one "group delay EQ" applied to the input, especially as the number of muti-way sections increases. (when resources are available)
I know theoretically the two approaches should be the same, for on-axis at least. But afaict, theory presumes the analog or IIR xovers first achieved a rather perfect phase matchup between drivers through the critical summation region.
But like you say, how you go about is totally preference.....
I was soo disappointed to learn all-pass filters shared the same phase rotation pattern as xovers in general. (LF lagging HF)
The rotation was what I wanted to get rid of...not add more to....😱
Using all-pass reminds me of the three stooges drilling holes in their boat to let the water leaks out 😀
🙂,
I've found learning linear phase xovers (and FIR, of course to facilitate them) ...is sooooooooooooooooo much easier than IIR, it's like comparing a high school course to a grad school course.
And the added benefit I think, to first working with linear phase xovers, is that once linear phase is learned, it makes understanding IIR xovers much easier...
I think I can see the temptation..
In understanding acoustics, why not learn to speak its language. 🙂
I learned about resonance before I taught myself to make a useful crossover. When things don't line up in IIR I want to know why before I go altering the evidence by force.
In understanding acoustics, why not learn to speak its language. 🙂
I learned about resonance before I taught myself to make a useful crossover. When things don't line up in IIR I want to know why before I go altering the evidence by force.
I hardly ever understand your add-on point(s), as they seem to continually shape shift and bring in previously non-discussed associations....in a way that appears to sidestep keeping focus on what was being discussed.
You know, like saying "I learned about resonance before I taught myself to make a useful crossover. When things don't line up in IIR I want to know why before I go altering the evidence by force"
Oh well. maybe it's just me and the way I think....that can't appreciate your point...sorry if so...
Anyway, FIR is not about forcing alterations and ignoring evidence.
To the contrary, FIR allows one to look at the evidence in finer fashion, and to separate variables so they can be more appropriately handled than with IIR alone.
Basically, FIR allows one to separate and inspect what can fixed with IIR, from the unnecessary and complicated phase rotation of IIR xovers.
If you've never experienced the relative ease and precision of aligning phase for drivers using linear phase xovers, well imho ...you've truly missed a big section of fundamental xover education...
When you fix something minimum phase, the phase gets fixed with it. If it doesn't then it's not minimum phase. Not all non minimum phase issues should be fixed by flattening.
I meant minimum phase...and my bad in responding /asking.
I misread your quote "Not all non minimum phase issues should be fixed by flattening."... as saying not all minimum phase issues ....
My understanding is all minimum phase issues should be fixed....so when I misread you, i replied asking...
I totally agree with your correct quote....
I'd even strengthen it to say not any non minimum phase issues should be fixed by flattening
Further thoughts.....
The ideas that all minimum phase issues should be fixed by flattening,
and that not any non-minimum phase issues should be fixed by flattening,
are at the root of why I don't think using FIR on the input stage before xovers, works.
Xover regions are not minimum phase.
FIR that spans these regions just becomes a hit or miss set of corrections, ime.
The ideas that all minimum phase issues should be fixed by flattening,
and that not any non-minimum phase issues should be fixed by flattening,
are at the root of why I don't think using FIR on the input stage before xovers, works.
Xover regions are not minimum phase.
FIR that spans these regions just becomes a hit or miss set of corrections, ime.
My learnings
Thanks for the inputs and comments, I think I've been able to address my doubts. I'm walking away with the following very useful insights:
Thanks a lot for your responses and patience.
Thanks for the inputs and comments, I think I've been able to address my doubts. I'm walking away with the following very useful insights:
- People who do speaker measurement and design the way I do do not need allpass filters for phase alignment in their analog active xo (AAX). Allpass filters in AAX are probably of great use for those who use textbook electrical LPF and HPF, not optimised acoustic LPF and HPF tweaked using speaker simulation software. Since I use the latter approach, my xo (whether passive or active) already give me phase alignment at xo point. And to reach there, my AAX filter blocks have strange non-standard Q, almost never 0.5 or 0.707. Such filter parameters are only possible to arrive at using speaker sim software.
- Linkwitz uses textbook electrical active filter blocks in his xo. The gentleman from TI who designed the 2-way analog active xo also used textbook electrical slopes. Both these gentlemen use allpass filters later for phase alignment. Neither gentleman appears to use speaker simulation software and its optimiser to arrive at non-standard filter parameters for precise phase alignment.
- Analog active allpass filters, whether the CR type or RC type, have a delay-vs-frequency curve which looks like a low-pass filter, i.e. it's flat in the low frequency range, and slopes down in an S shape to less delay in upper frequency ranges. What we may need may be the opposite -- we may need to delay the signal going to the tweeter, so we may need high delay in the upper freq ranges. So, analog active allpass filters are a rather imperfect approximation to what we need in typical real speakers. (This was a big insight for me.)
- Thorough phase correction and alignment can probably only be done using FIR digital filters. Just a flat, absolute freq-invariant delay for an xo output, of the kind digital xo allows us to do easily, may not really straighten out phase alignment on a larger scale. I don't know anything about FIR digital filters, so that's an area for future study. But I now realise that a flat, absolute delay figure of XYZ msec is not the same phase "straightening" in a real-world speaker.
Thanks a lot for your responses and patience.
Linkwitz if I have understood, had an admirably methodical approach. I believe he saw not just a random response in front of him, but knew intimately about the many parts that came together to produce that result. He then picked them off one by one.
In my humble opinion this is an elegant approach on so many levels.
My understanding of his approach exactly matches what you've described so well.
Yes, it has great engg elegance, but I also wish he would have shared one specific kind of data which he has never shared. I wish he would have shown us (i) the raw driver SPL, and (ii) the final simulated SPL after applying his crossover.
I know that he worked most of the time (the last part of his life at least) with OB designs, and OB speaker modelling software isn't as mature as box speaker s/w, but I'm not really talking about that. I'm just saying that I know he measured his speakers very carefully -- I wish he had shared his raw measurement curves and his curves after applying (in simulation or in reality) his xo.
I'm a huge fan of SL, but I wish he'd shared this with us.
I appreciate his approach to design, a lot, but my experience of tweaking crossovers by ear at the end of the testing phase makes me feel that measurements alone take you close, but you have to step in, listen, tweak, and then measure (or at least simulate) the outcome of the tweak. I know SL tweaked his designs, and there are clear descriptions of how he tweaked the xo after his friend did listening sessions on the LX521, or how he found subjective differences in other models after tweaking their construction, etc. But he never shared his measurement/simulation data of these tweaks.
I'm not blaming him. I'm just saying that this aspect of his design process are not shared, and I wish we had been given a chance to learn from them.
Last edited:
I think this would allow you to easily reverse engineer and make your own crossover. Of course the information is all there still if you want to digest it.