I was casually reading about the Joseph Audio Pulsar out of curiosity based on someone's strong endorsement on ASR. In the review they mentioned the crossover was around 100dB/octave or more. The idea has occurred to me, since with DSP, you can do that as easily as any other filter... but I was under the impression that a smooth handoff was preferred to avoid jumps in directivity. I have seen few posts talking about more than 48dB/oct before.
But if directivity is a good match at the crossover frequency... any other reasons you wouldn't want a sharp transition like that? According to the reviewer, they did a stellar job with the xover so that you couldn't tell where it was by listening.
Assuming you nail the FR match and directivity... (a big assumption, certainly) am I missing anything? Seems like an interesting way to stop worrying so much about center-to-center distance, at least...
But if directivity is a good match at the crossover frequency... any other reasons you wouldn't want a sharp transition like that? According to the reviewer, they did a stellar job with the xover so that you couldn't tell where it was by listening.
Assuming you nail the FR match and directivity... (a big assumption, certainly) am I missing anything? Seems like an interesting way to stop worrying so much about center-to-center distance, at least...
I asked myself the same question a little while ago and since I can easily do the test with my digital Xover (DEQX HDP-4) I did give it a try.
My system is 3 way active/tri-amplified and the drivers are SB Acoustics Satori MW16TX-4/TW29BN and SEAS L26ROY with a passive radiator: I would qualify it as a high-resolution system. The Xover selected frequencies are 150 Hz and 1,6 kHz, with a respectively 84 and 120 dB/octave linear phase filter. The DEQX provides 4 profiles for which you can store and configure parameters differently, so I set two profiles with exactly with the same parameters, except for the filter's attenuation. The "low" profile was set to a highish 48 dB/octave because it is the lowest value allowed by the DEQX for the mid/tweeter linear phase filter. I wished I could have done the test with a 12 or 24 dB filter and possibly the results would have been more determinant, but... hey, we can't have it all !! I can listen to each profile just by hitting a button on the remote control from my listening chair, which is perfect for an A/B comparaison. My observations:
My system is 3 way active/tri-amplified and the drivers are SB Acoustics Satori MW16TX-4/TW29BN and SEAS L26ROY with a passive radiator: I would qualify it as a high-resolution system. The Xover selected frequencies are 150 Hz and 1,6 kHz, with a respectively 84 and 120 dB/octave linear phase filter. The DEQX provides 4 profiles for which you can store and configure parameters differently, so I set two profiles with exactly with the same parameters, except for the filter's attenuation. The "low" profile was set to a highish 48 dB/octave because it is the lowest value allowed by the DEQX for the mid/tweeter linear phase filter. I wished I could have done the test with a 12 or 24 dB filter and possibly the results would have been more determinant, but... hey, we can't have it all !! I can listen to each profile just by hitting a button on the remote control from my listening chair, which is perfect for an A/B comparaison. My observations:
- going that low for a tweeter (1,6 kHz) is normally kind of a risky busines: however, with a 48 dB/octave filter, it is already more than adequate and for sure, going with 120 dB/octave, that ensures you that your tweeter is totally protected and will not be modulated at all by lower frequencies that may introduce distorsion or unwanted excursion;
- since each individual driver is muted very rapidly outside the desired pass band, in theory, there is less room for interference to take place between the drivers (comb filtering) in the shared frequency band. Does this impact negatively on the directivity? I can't measure it, but according to my old ears, there was no obvious change, and anyway, my serious listening is always done in the sweet spot in the main axis of the speakers;
- measuring the frequency response at my listening position with a calibrated microphone did not reveal any "observable" difference on the curve; the measurement is rather dominated by the room response;
- listening test: purely SUBJECTIVE, so, please do not take this as the "truth"! Not night and day, you have to listen carefully and preferably use a high quality recording. Bass/medium: a bit better defined or cleaner: this correlate with theory that says there should be less additions/cancellations of the emitted waves in the shared frequency band; the vinyl and tube aficionados might find that they lose some warmth: personally, I prefer precision to warmth. Mid/highs: same thing, a bit better defined or cleaner, can be played very loud without challenging the mids and as a bonus, a little more precise location of the instruments in the soundstage.
@Sined thanks for the response, this is much more substantive than I was hoping for!
I think since the room response at the MLP was about the same, we can assume the off-axis / directivity was also about the same. Very interesting. All else held equal, steeper crossovers should produce an overall "cleaner" output, so your subjective confirmation there is pretty encouraging. When I finally get around to building my 3-ways I'll definitely try this out.
I think since the room response at the MLP was about the same, we can assume the off-axis / directivity was also about the same. Very interesting. All else held equal, steeper crossovers should produce an overall "cleaner" output, so your subjective confirmation there is pretty encouraging. When I finally get around to building my 3-ways I'll definitely try this out.
Maybe, I am not sure if the Hypex filter designer has allpass filters or not 😀
But yeah in all seriousness the phase distortion is a pretty interesting question.
I've not really found it necessary to go too steep in practice, but crossover requirements can get pretty tight sometimes. That is to say that sometimes you need to go deep into what really makes a difference in order to avoid having to go steep.
I've also seen them suggested to reduce the overlap region for matters like lobing, but clearly this is a compromise. If a person has the opportunity it would be appropriate to manage the lobing.
I've also seen them suggested to reduce the overlap region for matters like lobing, but clearly this is a compromise. If a person has the opportunity it would be appropriate to manage the lobing.
My pleasure to share this little experiment!
And yes, this is one of the advantages of digital xovers for DIYers: you can try the silliest things without having to store a lot of parts and do a lot of soldering each time you change your mind!
Wish you as much fun as I have playing with such a versatile tool!
The high slopes are really just for marketing. Throwing around terms like "infinite slope crossover" is really going too far. The crossover has one or more notches in it. As the response falls into the notch, the instantaneous rate becomes very high, as high as thousands of dB per octave. But that is only happening for a very small interval and then the response comes back up again. So you could just as well claim the crossover has infinitely high GAIN (where it is coming up out of the notch). Anyway, the last I checked thousands does not equal "infinite". But it makes for good marketing. This is just one example why engineers hate marketing people.
Anyway, I am working on an article that I hope to publish next year on crossover with notches. Something new and (to me at least) interesting. So I know a little bit about the subject...
Away from the notches the slopes are "high" e.g. over 24 dB or more usually, it depends on the exact implementation. Some of these are actually really good as crossovers, so I am not knocking crossovers with notches in general, just the way it is advertised.
One issue that can happen with very steep crossovers is that the group delay develops a peak around the knee region. Once this is as large as about 2msec it becomes an audible effect that generates ringing in the time domain. But again, this depends on the exact crossover filter that is being used, so you can only make some general statements about it.
Also, up to now I have been only talking about analog/active/IIR type crossovers. You can use FIR to break the relationship between phase (and therefore group delay) and the rate of change of the amplitude response (e.g. via HBT) and so you can create very steep slopes and have linear phase (flat group delay) given enough taps and some intelligent kernel design approaches.
Anyway, I am working on an article that I hope to publish next year on crossover with notches. Something new and (to me at least) interesting. So I know a little bit about the subject...
Away from the notches the slopes are "high" e.g. over 24 dB or more usually, it depends on the exact implementation. Some of these are actually really good as crossovers, so I am not knocking crossovers with notches in general, just the way it is advertised.
One issue that can happen with very steep crossovers is that the group delay develops a peak around the knee region. Once this is as large as about 2msec it becomes an audible effect that generates ringing in the time domain. But again, this depends on the exact crossover filter that is being used, so you can only make some general statements about it.
Also, up to now I have been only talking about analog/active/IIR type crossovers. You can use FIR to break the relationship between phase (and therefore group delay) and the rate of change of the amplitude response (e.g. via HBT) and so you can create very steep slopes and have linear phase (flat group delay) given enough taps and some intelligent kernel design approaches.
Yes, unless high order, very steep crossovers are linear phase, you can bet it's mainly a marketing claim.
(like the Joseph Audio for example)
Imo, for IIR or passive, anything beyond 24 dB/oct is pragmatically a non-starter.
But if linear phase, then 100 dB/oct can work extraordinarily well in the right applications.
Such can be especially useful for higher multi-way counts, (some 3-ways and almost certainly with 4-5 ways etc.
Not really useful for two-ways, where summation ranges are most likely needed to be wide.
(like the Joseph Audio for example)
Imo, for IIR or passive, anything beyond 24 dB/oct is pragmatically a non-starter.
But if linear phase, then 100 dB/oct can work extraordinarily well in the right applications.
Such can be especially useful for higher multi-way counts, (some 3-ways and almost certainly with 4-5 ways etc.
Not really useful for two-ways, where summation ranges are most likely needed to be wide.
A long time ago I was into crossing domes very low with very low order crossovers.
Anyway I found that going more steep could compensate for the low output I was getting from them being crossed low, before they topped out... but only to a point. Going very steep was unable to help further.
Anyway I found that going more steep could compensate for the low output I was getting from them being crossed low, before they topped out... but only to a point. Going very steep was unable to help further.
IME it's more about the width and the shape of the knee in the transition region (down to -20...-30dB) than the final filter slope that counts.
For example, a digital Neville-Thiele XO (as implemented in the Acourate software package) has infinite slopes but the transition region can be made as wide or narrow as you want.
Here an example of a NT with about 0.8 octaves -20dB transition width, compared with a 8th-order Linkwitz-Riley with about the same transition width and shape:
Personally, I've never used anything steeper than 6th order in analog or IIR realisations, and typically only 4th order maximum for lower frequency XO points below 1kHz as phase wrap / group delay issues become relevant.
With linear-phase XO's you can go steeper and sharper knee but too kinky results in severe pre-ringing off-axis that may easily become audible. Further, too abrupt a handover from one driver to the next makes individual driver's perceived location more apparent in near field listening unless it's a coaxial or D'Appolito.
For example, a digital Neville-Thiele XO (as implemented in the Acourate software package) has infinite slopes but the transition region can be made as wide or narrow as you want.
Here an example of a NT with about 0.8 octaves -20dB transition width, compared with a 8th-order Linkwitz-Riley with about the same transition width and shape:
Personally, I've never used anything steeper than 6th order in analog or IIR realisations, and typically only 4th order maximum for lower frequency XO points below 1kHz as phase wrap / group delay issues become relevant.
With linear-phase XO's you can go steeper and sharper knee but too kinky results in severe pre-ringing off-axis that may easily become audible. Further, too abrupt a handover from one driver to the next makes individual driver's perceived location more apparent in near field listening unless it's a coaxial or D'Appolito.
https://xx3stksm.hatenablog.com/entry/2020/06/09/000548
Japanese, but would be very beneficial to this discussion!
Japanese, but would be very beneficial to this discussion!
Exactly imo. That's my experience too.
Yep. Although I haven't heard any evidence of off-axis pre-ring, I do not use brickwall linear-phase xovers simply for theoretical caution sake.
Unless they can specify how sharp the knee is, or what their transition bandwidth is.
Linear-phase 96 dB/oct Linkwitz-Riley have consistently given me excellent audible results on synergy builds, as well as on more conventional builds that don't share a synergy's inherent coaxial design. And they keep the transition region sufficiently narrow enough.
So, I've kinda quit giving further consideration to other xover types and orders, .....and to potential pre-ring.
Power handling for the HF driver can be an issue, since the point of maximum excursion for these is > the crossover frequency, so the higher the order, the lower you're forcing the HF unit to run ~flat. You also need to watch the group delay, & some think the phase rotation in the transition band is audible if you're not using FIR type active filters. YMMV as always.
Last edited:
Yeah, IME with compression horns, 1st order rules, though unfortunately I learned this too late to protect my (as it turned out) 'rare as Hen's teeth' Altec ultralite 1" CD diaphragms. 
Think that even with two identical drivers and linear phase there is a problem with phase difference of axis.
The LR principle is targeting both FR and constant difference in phase response.
Not shure if there will be constant difference in phase response even with linear phase off axis
If it is not we know LR handles this well up to 24 and struggles a bit more with more than 24 dB slopes
But the whole thing can be easily simulated by summing two filtered signals and introducing a time delay to simulate a path way difference.
Headphones should reveal any artifacts
The LR principle is targeting both FR and constant difference in phase response.
Not shure if there will be constant difference in phase response even with linear phase off axis
If it is not we know LR handles this well up to 24 and struggles a bit more with more than 24 dB slopes
But the whole thing can be easily simulated by summing two filtered signals and introducing a time delay to simulate a path way difference.
Headphones should reveal any artifacts
Acoustically you profit from a 48db/per octave filter for subsonic filtering for example for bass reflex systems. Helping them not being used beyond the reflex resonance.
Acoustically for crossover networks I would really try out which filter sounds best. As with dsp assistance and measurement equipment you can try them out all.
Usually 18db filters are traditionally used by the BBC for their high end loudspeakers. With dsp filters I would use Linkwitz type filters (24db). All depends of implementation of the drivers and the boxes if they do profit from this steep filtering or not.
However I always avoid any filtering in the midrange (by using fullrange drivers) - only parametric EQs for flattening the response curve but never to make cross over two drivers in the middle tones where the ear is sensitive to phase faults caused by traditional passive crossovers.
100 db filtering is in my eyes only a selling argument but acoustically it is not necessary.
Acoustically for crossover networks I would really try out which filter sounds best. As with dsp assistance and measurement equipment you can try them out all.
Usually 18db filters are traditionally used by the BBC for their high end loudspeakers. With dsp filters I would use Linkwitz type filters (24db). All depends of implementation of the drivers and the boxes if they do profit from this steep filtering or not.
However I always avoid any filtering in the midrange (by using fullrange drivers) - only parametric EQs for flattening the response curve but never to make cross over two drivers in the middle tones where the ear is sensitive to phase faults caused by traditional passive crossovers.
100 db filtering is in my eyes only a selling argument but acoustically it is not necessary.
Steep xover slopes make the stereo image appear to jump around between drivers and cause weird localization errors that sound very unnatural. I've used very steep filters with dsp and came to the conclusion not to use steeper than 4th order filters only when absolutely necessary. Notching a drivers pass band response is also problematic with very steep higher Q digital filters. It just doesn't sound right acoustically and creates alot of odd sounding artifacts, sort of like a selective on/off switching of specific frequencies that sort of disappear then come back. Not my cup of tea.
I would think that in a passive crossover such as in the speaker in the first post that they are using elliptical filters to get the really fast slopes. As such the slope can be very steep but the ultimate rejection is not unless the network gets very big and lossy.
My SW (Audio Weaver) will let do up to 8th order LR etc. with biquads and the associated computations needed for anything beyond that. Never found the need to go beyond 4th order in the present system and that's where it sits. It provides a reasonably flat response and is pretty detailed.
Can't do IIR; not enough taps unless I get heavily into decimation for the LF/MF ranges. I am too lazy for that.
My SW (Audio Weaver) will let do up to 8th order LR etc. with biquads and the associated computations needed for anything beyond that. Never found the need to go beyond 4th order in the present system and that's where it sits. It provides a reasonably flat response and is pretty detailed.
Can't do IIR; not enough taps unless I get heavily into decimation for the LF/MF ranges. I am too lazy for that.