Only two reasons you'd need time alignment
-Your active crossover filters are linear phase and you're targeting a linear phase system response
-The physical alignment of the drivers is so out of whack that you can't get minimum phase filters to track acceptably. Often this is the case in car audio. Almost never the case in a conventional multi-way home audio speaker.
Of course, even if you're using minimum phase active filters there is no reason not to use delays to make your job of designing the crossover easier. This allows you to first shape each driver into a target amplitude response/roll-off, THEN apply delays to get them phase aligned. Without delays you have to find a compromise between amplitude response and phase alignment as per passive crossover design*.
*Although, passive all-pass filters do exist which are almost like a delay
-Your active crossover filters are linear phase and you're targeting a linear phase system response
-The physical alignment of the drivers is so out of whack that you can't get minimum phase filters to track acceptably. Often this is the case in car audio. Almost never the case in a conventional multi-way home audio speaker.
Of course, even if you're using minimum phase active filters there is no reason not to use delays to make your job of designing the crossover easier. This allows you to first shape each driver into a target amplitude response/roll-off, THEN apply delays to get them phase aligned. Without delays you have to find a compromise between amplitude response and phase alignment as per passive crossover design*.
*Although, passive all-pass filters do exist which are almost like a delay
waveguide can set back tweeter more or less favourably, align with woofer or get even further back and then delay is needed for the woofer. no problem with dsp
I think a large part of my own confusion is that in building passives I've never cared about precisely implementing an LR2 or any other type of crossover slope. I've cared about the phase and amplitude matching, and driver excursion and dispersion. So whether I got 18 dB / octave or 16.5 dB/octave wasn't that relevant to me so long as I got deep and wide notches when a driver was inverted. Call me a pirate if you will. 🙂
However I got there by tweaking 0.1uF at a time, not something there's a real analog for in DSP land I know of. So knowing this is true and that setting accurate time delays makes it possible to use ideal slopes seems like the easy way to go.
However I got there by tweaking 0.1uF at a time, not something there's a real analog for in DSP land I know of. So knowing this is true and that setting accurate time delays makes it possible to use ideal slopes seems like the easy way to go.
Glad to have helped earlier 🙂
Use minimum phase shelving filters, both in band and out, for overall shape. And para EQs for bumps in otherwise smooth order rolloff.
A series of out-of-band shelving filters can do wonders to adjust drivers' acoustic rolloffs to meet desired order pre xover.
After a little practice you can bend a drivers acoustic order to about anything you want. I bend them to the closest fully complementary order that works for both sides.
Must strongly disagree. Desirability of time alignment has nothing to do with linear phase or minimum phase.
Time alignment is about in-phase summations over as wide a listening axis as possible.
It benefits any any all speakers with multiple driver sections. The better the time alignment...the better the speaker...from two-ways thru 5-ways ime.
If your crossover filters are linear phase, then you have no way of getting the speakers in phase by adjusting highpass/lowpass filters, because the filters are exactly that... linear phase. They don't alter the phase versus frequency. You can invert the polarity of one of the speaker drivers but that only gets you +/-180degrees adjustment. If your drivers turn out to be say 45degrees out of phase at the crossover frequency, then you are screwed without having the ability to delay one of the speakers to get them into alignment.
If you're using minimum phase filters then the filters have an inherent frequency versus phase relationship. By using asymmetric filters (e.g. 1.9kHz lowpass paired with a 2.1kHz highpass, and/or a 3rd order lowpass paired with a 2nd order highpass) you can introduce more phase shift into one driver than the other, and as a result get them aligned in phase. The difficulty is that while say 1.9kHz and 2.1kHz filters (as opposed to 2kHz and 2kHz) may achieve perfect phase tracking, the amplitude may not be what you wanted - there may be a slight peak or a dip in the on-axis system response at the crossover frequency, therefore it's always a balancing act between on-axis system response and phase tracking if you are only using minimum phase filters and nothing else. Speaker drivers are minimum phase devices, so using minimum phase filters is often exactly what you need to get them phase aligned.
Last edited:
Hi eriksquires!
Yes, you understand correctly the crossover with textbook slopes and time delay, and this is true for all alignments, not just the aforementioned L-R 24dB/octave.
You still need to do this even after time aligning. I use a crossover simulator in the ordinary way when investigating active filters, and transfer the data to other software when implementing. You may find different ways to do the same thing.
Well, that ignores the fact it takes preconditioning minimum phase work, to get the drivers to where linear phase xovers are usable.
That preconditioning works is the same thing you describe below.
Whether the xovers end up being linear phase or minimum phase remains immaterial, if the correct minimum phase preconditioning work was done.
And if acoustic complementary xovers are achieved from the summation of the minimum phase preconditioning and whatever electrical xover orders are used, the delays needed to synchronize times-of flights to acoustic centers will be the same, whether xovers are lin phase or min phase.
The real key, and principal objective in all xover design........imho......is to achieve fully complementary acoustic magnitude and phase throughout xover summation.
Min phase or lin phase xovers don't matter......both work fine, and both require same delays when done right.
It's easy to tell when "done right" is achieved.... delays end up being exactly the distances between drivers' acoustic centers, without any regard for asymmetrical electrical xover shifts. Because the electrical asymmetrical timing gets swallowed by achieving net acoustic symmetry (if done right 🙂
It depends on how you do it. Digitally, if you flatten the driver responses first as much as possible, then applying the desired crossover slopes, you only need to set the time delay and you have the textbook crossover. Or textbook-like crossover, depending on how the driver flattening came out and how precisely the timing has adjusted.
How is this not the same thing (for a single axis crossover)? Erik's current skills/knowledge remain valid in that respect.
In my opinion, it all depends on how big the time misalignment is. For example, in a sound reinforcement / cinema application, the horns are often very deep (sometimes >1m), digital delays can set things right quite easily. They are also the only choice (tape-loops are obsolete) when it comes to extremely long delays (>100ms), like in case of surround channel speakers, stadium sound systems etc. However, shallow speakers used for personal listening usually have much less acoustic misalignment and can be corrected using simpler methods like inverting the HF etc.
Digital delays are also programmable, and therefore highly suitable for experimentation purposes. However, a possible disadvantage is that the delay settings (specified in samples) needs to be recalculated every time the sampling frequency changes.
Digital delays are also programmable, and therefore highly suitable for experimentation purposes. However, a possible disadvantage is that the delay settings (specified in samples) needs to be recalculated every time the sampling frequency changes.
Huh? Linear phase means the phase changes linearly with frequency. You might recognize that phase increasing linearly with frequency is a pure delay.
Any DSP worth its salt will apply a global buffer/delay such that whatever collection of linear phase filters you have applied all achieve the same delay - in effect none of the linear phase filters impart any delay and therefore have 0deg phase shift at every frequency. Otherwise you'd be forever having to readjust delays every time you altered one of your linear phase filters.
I find that "threshold of audibility" is always of interest. Loudspeakers and even the recordings themselves are all compromises based on actual acoustic performance sound fields. Not asking whether it's audible, to me, indicates over reliance on measurements over whther you can hear it or not.
________________________________________________________
Years ago, I acquired a pair fully horn loaded loudspeakers from commercial cinema duty (Klipsch Jubilees)...which was designed with a requirement where no seat in the theater can be a bad seat acoustically, i.e., no "head in a vise" compromises. I started to look for a center loudspeaker to support a 5.2 array because I couldn't accommodate a third Jubilee in the center position. My first trials included various loudspeakers and one "horizontal MTM center loudspeaker"-like you showed, above. I immediately found out just how critical the center channel sound quality actually is. By far, the center channel is the most critical loudspeaker in a multichannel array.
What differences did I hear for a center between two Jubilees (TAD TD-4002 drivers, dialed in carefully using DSP)? Huge timbre mismatches and apparent source width problems (the apparent width of the center channel vs. the left and right front channels). After a few passive-crossover loudspeakers were tried, I used a Belle (also fully horn loaded, but much different size and performance than the Jubilees). This loudspeaker was elevated to be on nearer the centerline with the Jubilee HF horn centerlines. In short: it was terrible. As a last resort, I tri-amped using DSP crossover and found it still sounded bad but using passive crossover conditions (i.e., no time delays).
Then I dialed in the phase/time alignment of the Belle. Suddenly, it matched the Jubilees--in terms of timbre and also clarity. I later swapped out the tweeter (K-77-->Beyma CP25) which solved a high frequency harshness issue, then I swapped out the midrange horn/driver (K-600/K-55 --> K-510/P.Audio BM-D750, a.k.a, "K-69-A").
I used that setup for a couple of years. The timbre of this center channel matched the Jubilees. It wasn't quite seamless when walking across the entire 15.5 feet of room width just behind the listening positions: you could hear a small change in coverage when walking through the 1/3 and 2/3 positions laterally, but it was "good enough". I later found a better solution (i.e., the K-402-MEH).
__________________________________________________________
Another example that comes to mind is the dual-diaphragm compression driver, BMS 4592ND (dual ring radiators). With the time misalignment of the mid frequency and high frequency diaphragms at crossover (...the high frequency leads the lower frequency by one wavelength), the driver doesn't sound the same as the TAD TD-4002s--i.e., there is a timbre shift if the 0.145 milliseconds of time delay through the ~6 kHz isn't corrected through bi-amping using a DSP crossover.
Bottom line: the number of wavelengths of phase shift (down to 1/4 wavelength) through the crossover region is important: you can hear it in timbre shifts (at least) if you've got high performance loudspeakers. For rooms with lots of early reflections and poor phase fidelity of the loudspeakers, adding one more loudspeaker without time alignment doesn't seem to make a lot of difference: you can't hear it among all the other issues coming to your ears simultaneously.
This isn't really true in my experience. Using PEQs to change the slopes though the crossover interference band allows a continuous range of crossover values. In fact, you don't have to use "named" crossover filters at all, thus avoiding the 90+ degree phase growth through each crossover filter set.
Chris
Attachments
![gallery_26262_6_1571719[1].jpg](/community/data/attachments/960/960679-620041af5a93814f191b5522f8a50855.jpg?hash=YgBBr1qTgU)
Last edited:
No. Linear phase doesn't changes with frequency. It's linear, zero change in the phase response, independently to the frequency response. But why the phase doesn't changes with frequency? Because the DSP are delaying the specified frequencies with specific timing to assure us of this.
Last edited:
I think you're confusing "zero phase response" with linear phase response.
What term do you use when the group delay plot is zero (or constant), but not not the phase plot? I call that "linear phase", don't you?
What term do you use when the group delay plot is zero (or constant), but not not the phase plot? I call that "linear phase", don't you?
Last edited:
Interesting. I certainly don't use that definition. Minimum phase implies that there is no added "all pass" response to the loudspeaker's response, and I can use EQ, etc. , to correct the net response using DSP. But the acoustic center of the drivers/horns can certainly move around within their passbands, and it's still "minimum phase"--right?
I think it's probably better if we preface our definitions with "I call that...". Then arguments of "right vs. wrong" are avoided, and no ego clashes are necessary.
Chris
I think it's probably better if we preface our definitions with "I call that...". Then arguments of "right vs. wrong" are avoided, and no ego clashes are necessary.
Chris
Hi Chris, nice to see you here.
I'd like to offer a few definitions/observations to help the discussion.
First, a traditional definition of linear phase, is phase that is a straight like on a linear frequency scale. Only occurs on a linear, not log freq scale (like in virtually all common measurements).
Second, get rid of the constant fixed time delay in the measurement, and if true linear phase exists, phase will be at 0 degrees across the frequency spectrum, whether linear or log frequency scale.
Third, and this was a bit of an ahaa moment for me....linear phase is quite simply a minimum phase special case......
It is simply minimum phase where phase is flat across the spectrum, once fixed fixed delay is removed. That simple imho.
By across the spectrum, i mean the usable spectrum, 20-20kHz if we are lucky...something I've yet to achieve acoustically at my desired SPL levels.
I 1000% agree it all comes down to audibility.
I keep finding over and over, getting mag and phase right (the "only two truly root measurements") correlates well with audibility improvements.
But it takes a level of correctness before it can be heard, and then it can't be unheard.
By the way, the reason for the distinction that I made with the term "linear phase response": I haven't heard any differences if the group delay plot is flat (within certain plus or minus constraints vs. frequency), regardless of the tilt of the phase response (of course, within reason). In this domain of interest--electro-acoustic loudspeaker performance--I think typical definitions used in electrical circuit theory might actually clash with loudspeaker/room acoustics usage. YMMV.
Chris
Chris