Very interesting. Not long ago I was trying to very accurately measure the relative acoustic centre depth of a tweeter and midbass (ideally within a few mm) and my first attempt was to do it by impulse response, with the mic vertically exactly half way between both drivers. (and approximately 1m away)
I measured each driver separately and accurately measured the path delay to the impulse. For the tweeter with a high pass filter the location of the beginning of the impulse is of course very precise. But for the midbass which is being limited by the low pass section it wasn't clear to me what I should be looking for.
Trying to line up the "peak" of the mid-bass drivers response didn't give a believable result, and I realised that the beginning of the rise is probably the correct (or closer to correct) place, as the low pass filter is causing a slow rise time. But deciding exactly where a gradual rise begins is not easy to do with sufficient accuracy.
I then tried measuring the driver without any filter, but the unfiltered response of the driver has a "double rise" in the impulse response, presumably because it's a dual cone full range driver, and the high frequency impulse from the whizzer cone is launched slightly ahead of the lower frequencies from the main cone - effectively it's a two way design with the high frequency driver slightly ahead of the lower frequency driver, thus a misaligned impulse response.
That also made me realise that even trying to choose one of the two peaks wasn't necessarily accurate - because the double peak implies the acoustic centre is different at different frequencies for that kind of driver.
What really matters is where the acoustic centre is across the crossover region - if it moves a bit elsewhere, it doesn't matter.
What I ended up doing was doing it by cancellation instead - I connected both drivers out of phase to each other directly to the tweeters crossover, so that the ribbon tweeter was still protected, but that both drivers have identical drive signals with no phase shift except for the deliberate 180 degree reversal.
I then used the real time broad band analyser mode in ARTA and simply moved the tweeter back and forth until I found the point of maximum cancellation at and immediately around the crossover frequency.
I was able to find a deep notch with an accuracy of about +/- 1mm using this method, and the notch was pretty close to occurring at the same physical position from an octave below and above the crossover - beyond that it started to diverge a bit, presumably due to phase shifts of the drivers.
Have I reached a valid conclusion from this approach, at least for getting the time alignment at the crossover frequency range right ?
Once I determined I had it in the right place I decided to go back to the impulse measurement approach, and found that aligning the initial steep rise of the tweeter with the beginning of the slow initial rise (not the peak) of the low pass filtered midbass driver was almost but not quite right, and aligning with the broad peak was definitely wrong - introducing about a 40mm error in the acoustic centre.
What I learnt from this exercise is that measuring acoustic centre very accurately is quite difficult, and that the acoustic centre can and probably does shift with frequency, although maybe not quite so much with a single cone driver.
That's really interesting. So what you're saying is that you really only noticed the amplitude variations and imbalance not so much ringing, until the Q was very high ?
What Q would you say a typical room mode has ? Much greater than 2 ?
I agree completely. Slow bass results when the upper bass range is depressed compared to the lower frequency range, in other words its a frequency balance problem.
Floor bounce cancellation can be one culprit, by putting a big fat notch in at a certain frequency in the upper bass - the low bass note whose second or third harmonic falls into this hole will lose some of its attack. And fixing floor bounce cancellation with an appropriate design (eg low woofer) can solve this particular cause.
It also happens if the low end bass response is jacked up a lot by the room resonances/gain. Typical rooms and speaker positions usually tend to conspire to put notches in the mid/upper bass, and boost the low bass - the end result is "slow bass".
The persistent audiophile myth that annoys me no end is the "small woofers have faster bass" one. In my opinion the explanation for this is simply that a small speaker will tend to have a much higher cutoff frequency thus negating some of the low bass room gain and giving an overall more balanced sounding bass.
If you're lucky, the roll off of the small speaker will complement the room gain and you'll get a reasonably balanced response to quite a low frequency. If you're lucky that is.
A much larger speaker flatter to a lower frequency will have its low frequency response boosted hugely by the room and can sound "slow" without something being done about the room effects, but move the speaker to minimize any mid/upper bass holes as much as possible and then use EQ to bring down the lower room modes occurring below 80Hz to give a more balanced response and it will no longer sound "slow".
Eg its just a frequency balance issue.
Last edited:
We would like to have a loudness measure?
My point is that it is not possible to determine the loudness from the impulse response alone, no matter how one windows it.
What is needed is to convolve the impulse response with a signal having duration longer than the loudness integration time. Even this tells only half truth, because real life signals have random durations and they may or may not excite the loudness detection in fully. Thus in any case a loudness model gives the loudness measure only for a particular signal type.
- Elias
My point is that it is not possible to determine the loudness from the impulse response alone, no matter how one windows it.
What is needed is to convolve the impulse response with a signal having duration longer than the loudness integration time. Even this tells only half truth, because real life signals have random durations and they may or may not excite the loudness detection in fully. Thus in any case a loudness model gives the loudness measure only for a particular signal type.
- Elias
Thanks for the post Sane--I'm going to have to read it again. The ear does need several looks to make sense of a sound FWIW. It's less total looks in the bass end, but the signals are much longer.
So I guess the real question for me would be "at what point does our integration time become shorter than our measurements duration?" Then, to what degree does that change when listening to more complex signals if at all?
I tried web researching last night and came up with a lot of interesting stuff, but nothing on point.
So I guess the real question for me would be "at what point does our integration time become shorter than our measurements duration?" Then, to what degree does that change when listening to more complex signals if at all?
I tried web researching last night and came up with a lot of interesting stuff, but nothing on point.
Last edited:
Time alignment of the top of the drivers will not, IMO, effectively time align the drivers as a system. Keep in mind that the rise in the response of a driver is the result of applying a lowpass to an impulse.
Consider a 2-way for simplicity, first with theoretically ideal drivers before one considers the bandpass nature. An ideal driver, i.e. one that is not a bandpass, that is also perfectly aligned to the other driver and that has a lowpass applied will have an impulse response that is the perfect representation of the lowpass transfer function. The same applies for a perfect tweeter with a highpass. For this theoretically ideal case, the perfect alignment with no delay due to no driver bandpass issues will not have the driver impulse response peaks aligned. The reason is simple, the impulse response is at this point purely a function of the frequencies passed by each driver, i.e. the equivalent of the electrical transfer functions of the 2-way crossover sections. The only point of alignment that matters in this case is the initial rise of the impulse responses. Time alignment in the ideal case has nothing to do with the peaks of the impulse responses.
Now add in the bandpass nature of the individual drivers, make them real drivers. What changes? Not the initial rise of the impulse response, it may be infinitesimally small, but it is there nevertheless. The delay added by the lowpass of the driver natural response delays as one sees in the group delay curves, but at the limit, the rise will be immediate at higher frequencies, even though highly attenuated due to the driver lowpass. For the woofer it will be even smaller, but in theory at some level it is immediate.
For our purposes the areas in the driver output that are far down in level largely don't matter due to our hearing, so we concentrate on the areas of most significance. In delay issues the stop-band area is still of some significance due to some influence on group delay, but let's assume a well behaved driver well into the stop-band.
Adding in the highpass influence of the drivers simply adds complexity. But as to time aligning at the limit, it's all in the initial rise of the impulse. The problem is partly a function of the type of crossover we use as well. One that is a minimum-phase crossover vs. a transient-perfect target may actually require slightly different physical alignments of the drivers. The reason I posit this is that there can be no ideal version of either type. The bandpass nature of drivers prevents that, so the best that we can do is find the closest response to some ideal. What is that response for each type?
For a minimum-phase system it's hard to argue that it is anything other than driver alignment with drivers/crossovers such that the phase response in the intended crossover area is as close to ideal for each leg as is possible. On-axis, of course, no point complicating it further.
For a transient-perfect (T-P) system target I would submit that the best alignment will be close to the point of initial impulse rise, but adjusted slightly due to the bandpass nature of all of the drivers. There can be no perfect combination since no driver can have a perfect response, they are bandpass devices. So then how would one determine the optimal response, lacking the ability to have an ideal? For one, intended usage has to come into play.
I would suggest that one should do what John Dunlavy did. I believe that he judged his (T-P) systems on a combination of on-axis frequency response coupled with a measure of the step response at 10 feet. Since no ideal system exists, the step response will have overshoot and ringing. There can be no perfect alignment IMO as well. There will be a narrow range of adjustment of any of the drivers that will still result in a reasonably good step response, just as there can be small variations allowed (or required) in the crossover to achieve an optimal response.
I would also suggest that it's reasonable at this point to say that in neither of the cases described above will the peaks of the impulse response of any two or more drivers be aligned. Alignment of peaks for woofer to mid or mid to tweeter may be close, but it should be obvious that alignment of peaks for woofer to tweeter will no doubt not occur, yet the on-axis frequency response and step response will show excellent results if done well.
If one intends to have some ideal (optimal) system, it would have to be measured and designed tightly to the targeted listening spot. Multi-way systems need more distance for this integration. Dunlavy always impressed me with his adherence to this. Most of us, of course, don't have our own anechoic chambers and don't use forklifts to measure our systems as I believe he did.
Personally, I pay no attention whatsoever to the peaks of impulse responses with regard to any alignment of any kind.
Dave
Last edited:
Hi Simon, my 2Cts :
if you checked the impulse afterwards, it' OK, but generally this method can drive to incorrect results as there is a lot of points where the cancellation happens.
Drivers in phase opp, sure, but starting together, not automatically...
if you checked the impulse afterwards, it' OK, but generally this method can drive to incorrect results as there is a lot of points where the cancellation happens.
Drivers in phase opp, sure, but starting together, not automatically...
Measuring acoustic center is nearly impossible, one can get close. Measuring relative acoustic offset, however, is rather easy. Yours is a good example of doing so with a crossover, though this does add some dependency on how close each leg of the crossover is to the ideal. Make some change to the crossover and there may be added variance.
If using software to model and design, there are a couple of methods to determine the relative acoustic offset that should work for most practical crossover implementations. The benefit is that there is no requirement to design/implement the crossover at this stage.
Dave
I think you missed one small but important point of my test set-up. When doing the 180 degree phase shift "move for the best null" measurement I had both drivers connected directly in parallel with exactly the same drive signal, apart from the polarity inversion.
The high pass filter was only there to protect the ribbon tweeter from over excursion from a broad band test signal at low frequencies where the comparison is irrelevant. I could have done the same with a line level (or software) high pass function, but using the existing tweeter crossover for this was convenient.
Because the two drivers are directly in parallel there is no relative phase shift introduced between the drivers by any crossover, so for the purposes of finding an amplitude null in the frequency range of interest the result should be identical to that if both drivers were driven unfiltered. (But without the damage to the ribbon...)
The one remaining source of error I can see is if there was significant phase shift introduced by the drivers themselves, for example due to a sloping down response on the mid-bass driver, but if both drivers are fairly flat across the range of interest the determined relative acoustic centres should be pretty close ?
Not sure what you're saying here ? Perhaps you can elaborate.
To be clear, I started with drivers connected in reverse phase to each other. There is only one location which will give a broad null across a wide range of frequencies. (Within the limitations of the response of the two drivers tracking adequately)
If it was a full wavelength out (at the crossover frequency of 4Khz) then the notch would only occur over a narrow range of frequencies because the phase error is different at different frequencies for a given offset.
Besides, a wavelength at 4Khz is 86mm - and it's physically clear that these alternate locations are incorrect, as they're either in front of the cone of the mid-bass driver or behind it's magnet, both nonsensical locations 🙂
Last edited:
Excellent, that's basically the conclusion I came to when comparing the results I got from the out of phase cancellation test back with the impulse response measurements - when driven through their respective high and low pass filters, it was the beginning of the slow impulse rise of the mid bass driver than nearly but didn't quite line up with the beginning of the impulse rise of the tweeter when the two had acoustic centres aligned.
Trying to decide exactly where that slow impulse rise begins is something that's hard to do though, so I think I got greater accuracy with the cancellation test, as I was able to home in on a notch consistently within +/- 1mm of the same location, while the best I could get purely by comparing impulse response time delays was around +/- 3mm, partly due to coming close to the sample time resolution in ARTA, even at 96Khz sample rate.
Very informative post by the way. 🙂
Polarity connection is immaterial in finding relative acoustic offset if it's done prior to designing the crossover and should provide for a more accurate result that is independent of the crossover itself. It is taken into account in the process and can be done with either polarity for the method I've used. Technically, acoustic offset is not related to polarity.
I do strongly disagree with the statement that the result should be identical for drivers driven with and without a crossover. A raw arrangement as you described may introduce significant error because the phase in the crossover region is partially dependent on the lowpass of the driver. A lowpass that is raw vs. the lowpass with a corrected stop-band may easily have significant difference in the phase at the intended crossover point. Highpasses are generally easy, so I'll limit my comments to the lowpass.
Adding a protective cap can also be easily taken into account in the process, though I've not been in a situation where this was needed.
Much depends on how well behaved the driver is raw in what becomes the stop-band and the intended crossover Fc. If you use a simple electrical crossover (or equivalent) that does not have any correction for stop-band error, you may be correct. In that case you've accepted a lowpass response not fully matching the targeted response and there will be some deviation from flat above or below Fc due to the uncorrected stop-band region. However, if there is significant correction in the stop-band area to achieve the desired target, there may also be significant alteration of the phase response in the area of the Fc. This will be a result of both the correction for peaks/dips in the raw response above Fc (plus how close to Fc they are) as well the ultimate change in the group delay due to the slope achieved in the stop-band. In this case there may easily be a different result when comparing the raw method vs. using an applied crossover.
The method I show that uses raw measurements, when complete, will be correct for the relative acoustic offset as long as the models of the drivers created during the process are not altered (there being no reason do so in any case). If various crossover designs are to be examined, then the measure of relative acoustic offset is more important.
If physically aligning the drivers is then desired, the offset change required will still depend upon the targeted design as I posited above, but the change required can be reasonably determined during the design process directly from the phase of the driver/crossover combinations shown in the design software. There is no need to measure anything other than the initial raw response of the drivers, save for re-measuring due capture any change in diffraction influence from driver re-positioning if that was done and/or testing for things such as step response. If the driver positions are fixed, nothing more is needed. On any specified axis, that is.
Dave
Last edited:
Sorry, it was not for you...I saw that you control the problem, I was writing my comment for the future readers of the forum that will take this as the perfect method. The subject has been deeply treated by Le Cleach, IIRC it's only correct for symmetrical even order slopes.
I think that "generally" it's much better to look directly at the step or applying FFT filters to the impulse (ok, in your case that was not ideal).
I think that "generally" it's much better to look directly at the step or applying FFT filters to the impulse (ok, in your case that was not ideal).
Good point, I usually include that caveat and forgot to do so. With that additional issue, it makes the modeling method that much more practical. It works for determining offset for all cases of any crossover type since it is dependent solely upon the raw response of the drivers as modeled prior to doing any crossover design.
Dave
Hey Dave
I have predicted this result on many ocasions. Since the room dominates the LF sound field it should not be surprising that the audiblity of details of the LF sources would be relatively small. I expect that even differences in source types, Qs, resonances etc. will all go away once the LFs of the room were properly setup and EQ'd. The obsession with bass alignment just isn't relavent.
"Fast bass"! I always loved that term as a classic example of misguided audiophilia. If its fast then it ain't bass.
Earl,
It really depends on room size. In an arena woofer placement needs to be reasonably close to the rest of the sound source. Same goes for large outdoor systems. Made that mistake once.
Of course the assumption here is in reasonable size rooms, where I think we are in agreement.
ES
The reason I reversed the polarity of one driver for the test should be obvious - it's far more accurate to find a deep null when they cancel than a broad 6dB increase when they sum. For a given phase error you'll get a lot more amplitude change around a notch, so it can be found with greater accuracy. So if you want the best accuracy it's not "immaterial".
You still seem to be misinterpreting what I said. I said that introducing one high pass filter driving both drivers together during the test does not change the relative phase between them, and thus from the point of view of testing for proper cancellation at acoustic alignment there is no difference in the result obtained than if they were both driven directly. Do you disagree on this point ?
As for phase error in the drivers raw response, I already acknowledged that as a possible source of error at the end of my post. The low frequency driver was tested in it's normal cabinet and has a natural roll off at about 45Hz, so there should be very little phase shift contributed at 4Khz by the 45Hz acoustic high pass almost 2 decades below.
That leaves phase shift contributed by the high frequency roll off of the driver, and any excess phase. Because it's a dual cone driver it does have significant excess phase in the low treble region around ~>8Khz where the two cones cross over, and the on axis response is relatively flat up to about 8Khz.
A single cap is not enough to protect a ribbon tweeter subject to either broad band noise or a sine sweep. In ARTA you can set an additional high pass filter up to a maximum of 2Khz in noise mode, but there still isn't a steep enough roll off to prevent ribbon excursions at bass frequencies, likewise in sine sweep mode you can't restrict the minimum sweep frequency. Using the tweeters own normal high pass filter for protection is an easy and obvious choice if it can be done without affecting the accuracy of the result.
You're racing about 3 steps ahead of me here, I was only trying to determine the relative acoustic centres of the drivers without any crossovers applied, to use as a starting point.
As for the stop band performance of the drivers, the crossover is 18db/oct at 4Khz, the ribbon tweeter measures dead flat down to it's 1.4Khz resonance, the full range driver whilst not nearly as flat narrow band (with small spatially dependant ripples in response) 1/3 octave averaged it's pretty close to flat on axis up to 8Khz, and actually rises a bit from 8-12Khz. So you have a total flat driver overlap from 1.4Khz to 8Khz available.
A small amount of correction might still be needed for the flattest response of course, but even without doing anything special the summed response from 2Khz up is already remarkably flat.
When is it not important though ? Whether you choose to acoustically align the drivers or deliberately chose not to and design around it, you still need to know what the offset is.
Personally I dislike designs where the equivalent introduced phase error at the crossover frequency is more than about 45 degrees. No amount of all pass delay or phase correction tweaks in the filter can compensate for the change in acoustic offset when going off axis.
Assuming you're using software that does all this stuff for you. Some of us like to do a lot of the design process manually 😉
Last edited:
To me it goes without saying that everything that we talk about here is small room Hi-Fi. The large venue problem is very different.
A lot of "arguments" happen arround here when this assumption is not held, but I would claim that it is important for the poster to point out when they have changed it. Large rooms, telephones, all different problems and comments on small room hi-fi do not apply.
Last edited:
if its active no, if its passive then yes, I do disagree, because the loads on the passive are different so the phase is different.
No you don't. I don't measure it and I don't care what it is. If the data is taken in-situ then all appropriate acoustic delays, diffractions, whatever, are present and the crossover design will take them into account. It is not necessary and I don't even think it useful, to measure all of these things seperately.
Think again carefully about it for a second. Everyone seems to be very quick to jump to the clever and complicated explanation whilst ignoring the simple and obvious.
It doesn't matter for this test whether the high pass filter is active or passive - sure, the absolute amplitude response measured will be somewhat different between the two due to extra loading on the filter. (I'm not that naive or inexperienced that I don't realise that)
However, it has no bearing on finding the acoustic offset which with reverse polarity provides the maximum depth of notch because the filter introduces no additional relative phase shift between the drivers. I don't know how I can be any more clear on this point, so I'm really not sure why people are having such a hard time seeing this.
It's absolutely no different for the test at hand to connect both drivers directly to the output of an amplifier in reverse phase, and applying the protective high pass filter before the amplifier - what we're looking for is the physical position where the cancellation is deepest across the crossover overlap region.
So you're happy to throw together two drivers with different and unknown acoustic centre depths, measure their amplitude and phase, and design whatever crossover correction is necessary to account for the unknown difference in acoustic centres ?
Well I guess that's one approach to speaker design, certainly not one John Dunlavy would have adhered to 😉
Surely you are aware that even a pure time delay applied to a driver whose acoustic centre is too far forward is only equivalent on axis ? If you only design for on axis response that's fine but your last dozen or so posts suggest that you don't, and that you don't much care for on axis response in fact.
I'm not saying that compensating for acoustic centre with the filter is necessarily bad (after all, the majority of speakers don't bother to physically align acoustic centres for very practical reasons) but it's certainly a different result, so to suggest it doesn't matter at all is a bit disingenuous.
Last edited:
I will not argue that a lowpass circuit does not have an effect... but the reality is that group delay increases dramatically at low frequencies even for a full range speaker and regardless of the application of a lowpass circuit in the signal chain.
How do we explain the difference in the delay of the impulse response when there is also a lowpass circuit applied to the midrange? In addition, how do we explain the difference in the delay of 2 different sized drivers playing within the same frequency range with the same crossover circuit applied or no crossover applied for that matter?
I really suspect that the primary reason that a delay is not added to the higher frequency drivers, in order to bring them into alignment with the lower frequency drivers (at the crossover frequency of course) is that this is practically impossible to implement in an analog circuit.
SL has some interesting insight into a cascaded crossover topologies that helps to correct for the phase anomalies of crossover circuits that readers may find to be helpful.
Woofer crossover & offset
I'd say it depends on how one does it. Done manually the crossover type will be important as pointed out. In your specific example, then yes, a reverse null of an even order system is reasonable, as long as the possible caveats are avoided. It's not possible at all for odd order of course.
The reason I started using the raw measurements was due to the software I had available back in 1996. This was the cheaper version of CALSOD that did not have the option to use direct measurements, so creating models was a necessity. I have found that using this method, inverted connections provide no benefit because the raw measurements usually provide a whole series of constructive/destructive interference in the summed response that is easily matched in the model and can in fact be done to the tenths of a mm if desired, though that level of detail is not necessary.
Ah, excuse me for that, I missed the part about both drivers with the same highpass. However, I think that I agree with Earl that it does matter if done passively, because of the difference in the interaction of the highpass (cap in this case) with the differing complex impedances of the drivers that affects more than just phase. In any case, it looks to me that there will be other errors that remain, related to the change that occurs when a full crossover is applied. Nulls of raw or nulls with the same highpass on both, neither one addresses the (more general) case of the lowpass change due to band-pass tailoring.
I'll point out also that my comments are more general in nature, not necessarily directed at your particular case. There can be exceptions depending on the specific case.
I think that I understood you, but I still disagree that the individual centers can be determined, neither via the impulse response nor via modeling. The relative offset is all that actually matters in any case.
Certainly your case has more flexibility, so it will lend itself to your method more easily than others may.
As is the case for Earl, I now seldom create models where the offset needs to be known, direct measurements include the correct phase relationship to the measurement point. I do that only if I'm looking into the off-axis response in the software since all software that I use requires a minimum-phase response for the model to do this that then dictates the need for the relative offset on the z-axis. The physical geometry is then used in the calculation of phase.
If in the end it works for you, that's what counts.
Dave
- Status
- Not open for further replies.