So, what you guys are saying is that the tweeter is distorting as it nears the crossover frequency and a gentler slope might help it? This is again a design trade-off, accounting for poor performance or less than optimal choice of crossover frequency. Ultimately, it will have to be -6db at the crossover for LR.
It all depends on the drivers, the design choices and goals.
It all depends on the drivers, the design choices and goals.
To run the impulse generated by rephase you can either use a convolution plugin in your favorite mediaplayer (such as convolver for foobar for example), or use SoX to process your audio file off line.I have found some but not for 48db/oct slopes, and rePhase can only generate but not re-process existing files.
If one were to use filters of infinite order, then you would literally end up with the tweeter needing to output at full amplitude right up until the xover frequency.
Well, no, since with infinite order filter the group delay would also be infinite so the signal would never reach up to the tweeter. There simply would be no sound, never
- Elias
Still need to see specific examples for this to be anything more than yet another discussion on the merits of various crossovers. If there's nothing clear from an on-axis gated response measurement, then you need to get into polar measurements and/or phase. I don't see much point in a bunch of people guessing what it could be without knowing what the drivers are and seeing measurements and the speakers. Maybe you're just doing a better job of using lower order filters than you are with the higher orders.Frequency response analysis did not reveal any problems.
Whether or not you're concerned about phase tilting, absolute phase coherence at crossover, and so on, it's still important to look at the phase carefully, or at least see what it's doing by taking a lot of response measurements. To oversimplify a bit: if your driver phase is well-aligned at/around the crossover point on the middle axis of the speaker, then you just need to consider the null angles that result from whatever your driver spacing is, and whether there's any problem with that. If your phase alignment is anything else, then the entire vertical polar response should probably be evaluated.Yes I understand this well. However quarter wavelength at 2.5 kHz is just ~3.5 cm and this is obviously almost never possible to achieve between driver centers.Also when the listener is moving in the room phase will be shifting constantly and different depending on frequency; combined with offset drivers this gives phase corrections across a wide range of positions mostly meaningless IMO.
Without these it is all highly speculative, imho
Eelco
Hi,
Not really. 48dB/octave slopes have obvious repercussions.
not least of which is phase wrap and attendant group delay.
See Zaphs discussion here :
Zaph|Audio - ZD5 - Scan Speak 15W8530K00 and Vifa XT25
rgds, sreten.
So, what you guys are saying is that the tweeter is distorting as it nears the crossover frequency and a gentler slope might help it? This is again a design trade-off, accounting for poor performance or less than optimal choice of crossover frequency. Ultimately, it will have to be -6db at the crossover for LR.
It all depends on the drivers, the design choices and goals.
Er, yes, that's what we said. I took it for granted that people would appreciate the fact that everything is a compromise & that it is ultimately a question of selecting the one that best suits a given set of requirements.
Lukas,Hi,
For a while I have been trying miniDSP with various configuration of active filters on a couple of my DIY systems.
What I have noticed is subjective listening fatigue when 48db/oct filters are used vs lower order ones like 24db/oct.
I have some possible explanations for this:
a) Very abrupt change in polar pattern causing sharp changes in power response.
b) Significant phase shift so harmonics of instruments are shifted too much
c) Psycho acoustics
d) Something specific to my systems or electronics
Am I lone there or someone else has similar experience?
Regards,
Lukas.
Lower order crossovers have less ringing (faster transient response) than high order crossovers using the Mini DSP.
That ringing may be what causes your listening fatigue when 48dB per octave filters are used.
Some digital crossovers use linear phase filters which require more computer power than the MiniDSP has, and avoid the ringing present in the MiniDSP (or passive crossovers) using high order filters.
For most home applications with drivers of adequate response for a few octaves out of the passband, 12 dB per octave provides adequate protection.
6 dB may be adequate for low level listening with wide range drivers.
24 dB per octave may be needed to protect a tweeter used close to it's acoustical rolloff.
48 dB is just overkill for anything but drivers that have serious problems outside a very narrow pass band.
But you can't polish a turd, speakers that need a 48 dB per octave crossover will generally have more problems than the ringing it causes.
Art
Last edited:
Well, no, since with infinite order filter the group delay would also be infinite so the signal would never reach up to the tweeter. There simply would be no sound, never
- Elias
Can't you get very high order fir filters that also have linear phase? Or would these simply have a group delay that is constant vs frequency, but the group delay is very long
I don't expect anyone to accept the findings in this paper, but in blind tests they find that people can hear the effects of high order filters. As a result of their findings they recommend a conservative limit of filter orders less than 600...
http://www.acoustics.hut.fi/~mak/PUB/JAES_V57_6_PG413.pdf
Personally, I am completely undecided on the issue. Sometimes I think I prefer shallow filters, and sometimes steeper, with my own attempts at phase/amplitude flattening. But my speakers are currently a bit of a Frankenstein's monster, something which I am hoping to change in the very near future.
http://www.acoustics.hut.fi/~mak/PUB/JAES_V57_6_PG413.pdf
For digital crossover design we can draw the conclusion
that L–R filter orders up to about 8 should be safe in
most cases, whereas for FIR filters it is possible to go up
to about 600 without problems. Of course the safety limits
depend on many factors. For example, in the case of
coaxial loudspeakers the delay between driver elements
does not vary much with the listening angle, so higher
filter orders are possible. In general, going to the safety
limit of a filter order may not be well recommended.
Personally, I am completely undecided on the issue. Sometimes I think I prefer shallow filters, and sometimes steeper, with my own attempts at phase/amplitude flattening. But my speakers are currently a bit of a Frankenstein's monster, something which I am hoping to change in the very near future.
@Sreten,
Of course 48 db filters produce more phase wrap en GD. Whether this is audible is disputable. Furthermore, the off axis response due to the possibility of lower X-O frequencies might be better.
Simply stating the phase wrap and GD are the cause of listening fatigue, without even seeing the SPL plots, is a bit too short imho.
Regards,
Eelco
Of course 48 db filters produce more phase wrap en GD. Whether this is audible is disputable. Furthermore, the off axis response due to the possibility of lower X-O frequencies might be better.
Simply stating the phase wrap and GD are the cause of listening fatigue, without even seeing the SPL plots, is a bit too short imho.
Regards,
Eelco
Exactly.
Take, say, LR4 and LR6. Since LR4 begins its rolloff at at a higher frequency, amplitude demands on the tweeter are reduced compared to LR6 which forces it to work harder to a lower frequency. Which is the better tradeoff? YMMV on that score, especially given the host of other factors at work.
Here's the above mentioned example: it's talking LR2 & LR4 but you get the idea: Zaph|Audio
Edit: Finally found it. See the XO section & the remarks on tweeter power handling: Zaph|Audio - Bargain Aluminum MTM
Yes, passive crossover integrating natural roll off tweeter to get acoustic LR4. Slopes cross -6dB and -3dB occurs at lower frequency for higher order filter. Long before excursion of tweeter due to signal at -3dB point of LR4 limits use, excursion from signals in attenuation band below -6dB point produces high levels of IMD throughout most of tweeter pass band.
To effectively use active filter the crossover region of drivers must be sufficiently flat in response. Cross higher, or EQ drivers to flatter conditions in addition to crossover filter. Yes, potentially boost the natural roll off of tweeter to flat, a 6-12dB/oct process combined with 48dB/oct crossover to nicely suppress tweeter at resonance. Excess phase for path length distance to listener of the effective radiating surfaces is easily accomplished with digital active crossover, but all pass is used with analog active crossover. Flat responses through crossover region guarantee highly similar phase slopes/group delay when excess phase is conditioned. Poles of all pass filter are inserted to get GD match in crossover region.
Above is from thread: Active v passive #803
Peerless 2" full range is run with to differing conditions: Linkwitz-Riley 24dB/oct high pass filter and with 1024 point FIR high pass filter. Each condition is also brick wall filtered with 4096 point FIR 1kHz low pass filter. The upper trace shows a sea of output above 1kHZ for LR4 high pass filter consisting of harmonic distortion and IMD. FIR results show defined much lower simple harmonic peaks against the true noise floor.
Linkwitz Riley filter conditions sound is fatiguing, with FIR conditions sound is clearer and not fatiguing.
Steep filters cause overshoot. Overshoot causes ringing. The combination of ringing and music fighting over control of the cone or dome causes long term listener fatigue. Switching to a 4th order bessel filter has a profound effect on reducing listener fatigue compared to a 4th order LR. The 3 db peak can be finessed out with dsp (actually, dsp is anything but finessful) if you want. Even with the peak, the ease in which the music is presented is profound compared to other types of filters..
Last edited:
Steep filters cause overshoot. Overshoot causes ringing. The combination of ringing and music fighting over control of the cone or dome causes long term listener fatigue. Switching to a 4th order bessel filter has a profound effect on reducing listener fatigue compared to a 4th order LR. The 3 db peak can be finessed out with dsp if you want. Even with the peak, the ease in which the music is presented is profound compared to other types of filters..
If driver is used within suitable bandwidth, it slews with all frequencies within that bandwidth. Whole point of crossovers.
Remlab - I concur. Bessel filters really are the cat's meow in crossover design. The most vexing "problem" though is that the cones themselves exhibit their own efficiency-of-conversion that is frequency based. For this reason, I always like to drive each cone with no crossover whatsoever (at low volume) by itself with a pink-noise source and measure over many seconds at 1 or 2 meters with a 'reference microphone' and a good digitizer the response. With about 10 seconds averaging, and a little Fourier work (which takes noticeable time, having 480,000 samples, at 48 kHz sampling) ... shows a noisy, but very useful transfer characteristic of the cone ... in the enclosure, with damping and all.
If I'm using a 2-way speaker, then I just do it for both. Sometimes I'll additionally move the reference microphone around to side-lobe positions, since it can be better to design the crossover for a wider range "average" than straight head-on.
Armed with the curves, it becomes easy using a PERL algorithm to choose 4 poles for a combination filter that best achieves linear response. IIR simulation then shows group-delay for impulses ... which using Bessel filter designs is usually unsurprisingly already minimally impacted.
Implemented in ACTIVE form for bi-amplification, and she's ready to go. I have yet to listen to a resulting design that isn't just picture perfect. The method works, since it starts first "with the real" (the cone responses, in their enclosures, with damping, etc) and adds to it the most minimal energy-guiding through crossover filter design. The results are great.
Incidentally - the same technique is perfectly workable for 3-way and 4-way speaker designs. Parallel Bessel filters (for a 4-way: low pass, low-band pass, high-band pass, high-pass) working off the same input signal (no cascading!) although there are more poles and sections, also gives the best group-delay and impulse response.
GoatGuy
If I'm using a 2-way speaker, then I just do it for both. Sometimes I'll additionally move the reference microphone around to side-lobe positions, since it can be better to design the crossover for a wider range "average" than straight head-on.
Armed with the curves, it becomes easy using a PERL algorithm to choose 4 poles for a combination filter that best achieves linear response. IIR simulation then shows group-delay for impulses ... which using Bessel filter designs is usually unsurprisingly already minimally impacted.
Implemented in ACTIVE form for bi-amplification, and she's ready to go. I have yet to listen to a resulting design that isn't just picture perfect. The method works, since it starts first "with the real" (the cone responses, in their enclosures, with damping, etc) and adds to it the most minimal energy-guiding through crossover filter design. The results are great.
Incidentally - the same technique is perfectly workable for 3-way and 4-way speaker designs. Parallel Bessel filters (for a 4-way: low pass, low-band pass, high-band pass, high-pass) working off the same input signal (no cascading!) although there are more poles and sections, also gives the best group-delay and impulse response.
GoatGuy
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
- Home
- Loudspeakers
- Multi-Way
- 48db/oct crossovers causing listening fatigue?