I moved the bass localization posts to a new thread here:
http://www.diyaudio.com/forums/multi-way/200591-localizing-bass.html
http://www.diyaudio.com/forums/multi-way/200591-localizing-bass.html
I'd generally agree with this, and yet - someone like Nelson Pass who has spent the better part of his life "playing" with this facet in in his amplifiers, does find a certain character that is imparted to music reproduction.
Basically I believe that for most you can't reliably identify a difference under normal use (.. which is a bit different from not actually being able to hear the difference), but for some - they seemed to have *learned* to be able to reliably identify a difference.
If using FIR filters with sharp slopes and constant phase then 1.6khz would seem to be the most interesting point to put a crossover at:
In the absence of phase aberrations I suppose "accurate" could be reduced to "sensitive", and aberrations caused by sharp FIR filters (sharp level discontinuities and ringing off-axis if directivity is not perfectly matched and drivers not geometrically aligned) would be less noticeable at 1.6khz than at 800hz (or even 500Hz).
What would the optimal waveguide for a planar be like?
I was looking at this horn loaded planar:
Creative Sound - Product Details
And it piqued my interest.
I was looking at this horn loaded planar:
Creative Sound - Product Details
And it piqued my interest.
The ear is most sensitive to group delay at 2Khz ? Yes, I'll give you that. However does that mean the group delay of typical crossovers of 4th order or less exceed the threshold of group delay detectability even at this sensitive frequency ? Not that I've seen.
If you have a reference that says that the group delay of a 2nd or 4th order L-R exceeds the threshold of detectability at 2Khz that would help your argument. (Note: I still agree that 2Khz is probably the worst possible crossover frequency, but largely for reasons other than group delay)
The chalk board study, which I linked to recently in another thread actually said very specifically that the annoying frequency range was between 2Khz and 4Khz, not 2Khz and 5Khz, so unless there were two different studies you're mis-remembering the result.
Most who have listened to tone sweeps at moderate to high levels will not be surprised - there is always that part of the spectrum where we wince as the tone passes no matter what speakers are being tested. I seem to be someone who is particularly sensitive to sounds like scraped chalk boards and metal potato mashers scraping on metal pots. (Just thinking about that noise is making my skin prickle)
I think the takeaway from that study is that for speakers to avoid annoying you like a scraped chalk board, flatness of response through that critical region is absolutely crucial, and no response peaks can be allowed between 2-4Khz, even narrow band ones. Also important is that peaks don't form off axis despite a flat on-axis response, which means taking care of diffraction and directivity, as well as ensuring good phase tracking if there is a nearby crossover...
You keep speaking of "most sensitive" but are you referring to absolute sensitivity, which is indeed highest at 3Khz, or do you really mean "sensitivity to errors". EG it could be argued that the ear is more sensitive to variations in the response at some frequencies than others.
Last edited:
I think you mis-understand me.
What I was basically saying is that your hypothetical situation eliminated almost every important drawback of crossovers, so your question boiled down to "if I eliminate everything that makes a crossover bad, will it still be bad" ?
Mosts of the bad side effects of a crossover are not a result of the low pass and high pass transfer functions, (if done well) but due to non-coincident drivers. Make the drivers coaxial and the vast majority of differences that separate an actual point source full range driver and a multi-way system evaporate. In nearly all ways that matter the crossover ceases to be a problem.
Remember too that a dual cone full range driver is often thought of as a single driver with no crossovers, but the reality is that it is a 2 way coaxial driver with a mechanical crossover between the cones which just happens to share a common voice coil. This is readily demonstrated by looking at the excess phase and excess group delay curves which look very much like a 2nd order crossover. Does this matter ? If the response is smooth and the cones are concentric to form a point source, I say no. Same thing if it was a discrete coaxial driver with an electrical crossover.
Frequency response was only matched "fairly closely?". Not close enough then to draw any conclusions, as seemingly small frequency response variations can be significantly audible. Have you re-done the test with frequency response matched exactly ?
Simon, remember that speakers in which a horn is crossed to a cone midwoofer usually have some extra group-delay due to the fact that the acoustic centers are usually not aligned. In my speaker the acoustic center of the tweeter is about 11.3 cm behind that of the midwoofer. The combined group-delay of the crossover and the off-set tweeter is about 0.6 ms just above the 1khz crossover point. According to Blauert and Laws the threshold of audibility is 2 ms.
Last edited:
I should, and have used the term "acute" because that is more what I mean. Errors in this frequency range are very audible, more so than lower in frequency. Since a crossover is always an error, it should not be in this range.
When I refer to group delay I am not specifically refering to crossovers. For eample diffraction has group delay as do HOMs in horns and waveguides. These are the things that I am more concerned with because our sensitivity (re. acuteness) rises with rising SPL and they will become audible at higher SPL. This tends to be interpreted as "distortion", which of course it is, but its linear.
That is at a lower level SPL, it gets shorter with SPL level. Thats the point that I have been trying to make.
Maybe I have more confidence in my abilities in crossover design but I have never thought of excluding crossover points from any part of the audible range. The tools are available, and many people now have the skills, to design a crossover that is imperceptable in terms of the axial combined response. By that I mean that you can easily achieve crossover smoothness to a level where the usual driver and cabinet related response issues cause much greater error. On axis the crossover would be undetectable.
Now if we move to off axis curves we can achieve equally good response over a range of lateral off axis positions (assuming a vertical stack of units), again over a pretty wide angular range.
Yes, moving vertically will reveal the crossover point, as would measuring power response where a power dip will generally occur at each crossover. Still, I don't think a little restriction in listening position is out of the question and I'm sure you will agree since you design products with position restrictions as well (e.g. no listening on the horn's axis).
If this still bothers you then coaxial units are available (watch out for unintended response issues). I've also had good luck designing symmetrical arrays with fairly invisible crossover points in the vertical direction.
What remains as a clue to crossover point might be phase issues, but I'd agree with the consensus that typical crossover slopes give errors below thresholds of audibility. In my experience crossovers cause a fairly gradual phase transition from driver to driver.
Are we creating a problem where one doesn't exist?
David S.
Coaxial is not a solution because the waveguides are too small to be effective. It is the unavoidable lobing at the crossover that is the problem and without coax, which, as I said, is more of a problem than a solution, you will always have an uneven power response at the crossover. It is unavoidable.
The polars of the B&C 8CXN51 look pretty impressive:
B&C SPEAKERS
The bigger ones (12"/15") don't behave too bad around XO. The waveguide looks pretty big.
Last edited:
If anything that helps my argument. Often when people obsess over group delay of crossovers in multi-way designs they forget the fact that non aligned acoustic centres can generate more group delay than the actual filter transfer functions.
In the case of a coaxial design with closely aligned acoustic centre the only excess group delay left will be a very small amount from the crossover transfer function itself.
0.6ms is a lot compared to what you might find on a more closely time aligned system.
Out of interest I opened up a couple of measurements of my speakers to look at the excess group delay. The full range driver measured by itself (which is dual cone) has a maximum excess group delay peak of 0.115ms at 5.6Khz, which is where the mechanical crossover between cones occurs.
Shift in excess group delay between midrange and high treble (eg on either side of the mechanical crossover) is very minimal, with the midrange being delayed by 0.004ms relative to treble, suggesting that the effective acoustic centres of the two cones are aligned to within 1.7mm, which is pretty amazing.
(This is not always the case with dual cone drivers, on another model I've measured (Fostex FE207E) the treble is 0.023ms delayed relative to the midrange, which is an error of 7.9mm in the acoustic centres)
Audible group delay ? I doubt it.
Then I looked at the speaker as a whole - same full range driver as midbass crossed to the tweeter at 4Khz, 18dB octave. The tweeter is currently around 25mm further ahead than being acoustically aligned.
Maximum excess group delay occurs at 3.5Khz and is 0.171ms. Higher than the mechanical crossover of the driver, but not by much. Group delay of the midrange relative to treble is 0.08ms which corresponds to a distance error of about 27mm, which agrees very closely with the acoustic centre offset of the drivers.
Is a 0.171ms maximum excess group delay peak at 3.5Khz audible ? Is the 0.08ms offset between midrange and treble audible ? I've been thinking about changing the crossover to a 4th order L/R (acoustic) response and moving the tweeter back the 27mm to bring them into time alignement - this will eliminate the difference between midrange and treble, but probably increase the peak between them due to the higher order filter. Better or worse ?
If the 2ms figure is to be believed, (apparently not according to Earl) then the figures for any reasonable crossover where the drivers are closely time aligned will be vanishingly small by comparison, and only in systems with deep waveguides will there be a concern...
Last edited:
Those are the same two drivers as a Harper. Note the strong resonances of the woofer at 2-4 kHz - a real problem. Note that you don't see any sign of those in the polar plots. Why not? Those polars are just too rough to be of much use (12 dB steps!, I use 1 dB with major marks at 3 dB)
Coax favors smaller drivers because the small size of the wavguide is less of an issue. But the larger coax that I have data for - real high resolution data - show a sadly lacking horn performance at the crossover.
Last edited:
Moores work on GD is much more recent and much more enlightening.
Its not the group delay of the crossover that I am worried about. Its the GD of other things as I said. For the crossover its the inevitable dip in the power response.
Genelec seems to know how to do a coax with waveguide: Genelec 8260A - Sophisticated Tri-Amplified DSP Monitoring System
They show some impressive measurement data, although it would have been nice if they would have applied less smoothing.
They show some impressive measurement data, although it would have been nice if they would have applied less smoothing.
If group delay is so important, shouldn't all sources of group delay be taken into consideration ?
Your response made me smile because your comments indicate to me you agree with what I wrote.
You wrote 'reproduction is completely independent of 'art''. I wrote "reproduction is largely, an engineering challenge". I don't see much difference in those statements.
With regard to my 'art appreciation' statement, the 100's of different speaker designs that exist in the world today are testament to the varied subjective interests listeners have in their loudspeaker choices.
I have not heard any of your designs but am confident your efforts would indeed be satisfying when and if I should get that opportunity.