Yea, that's a very flat FR for a design like that. Mr. Kangaroo looks impressed. I'd think that with a crossover that high issues like phase become more important.
Its just a big full ranger in a quarter wave box. But apparently there's "nano particles" in the cone, so that fixes everything. Still, writing a review like that will basically mean no one will send him more speakers to review.
Sounds a bit like the principle behind fostex sigma line.
Just looking at csd's its apparent that some drivers have different decay patterns than other drivers even with the same FR. But of course the decay behavior of a certain driver is related to changes in the FR and cone break up of that driver.
Linkwitz gave accutons not so great scores in his tests due to energy issues even though they're supposed to avoid any kind of break up
Midrange distortion test
I don't know exactly what's going on but I do notice a splashy sound on a lot of drivers. It would be interesting to equalizer some drivers perfectly flat (hard to do) then see if they sound the same with different decay and break up profiles.
I have to say the 1-8 khz region is critical in the annoying fatigue department, but it does depend somewhat on the music you listen to and whether there are upper harmonics in that area.
As a matter of fact, you can have ringing without a peak in the frequency response, and sometimes even a notch. Notches at the lower frequencies like 1KHz or so can also be caused by surround ringing. This is one aspect that Klippel has brought to my attention after looking at a scan of a driver that I sent them for scanning. I later looked at CSD, and there was a noticeable ringing hidden below, not noticeable in the SPL, some 20db or more lower which varied from driver to driver. It was not so audible to me as I reduced it further down, never the less, it did improve.
Hello Soongsc,
Ringing without any dip in the SPL response (ruler flat that is) or small peak impedance? Please show me an example of this. BTW:I am well aware of the infamous surround bounce. Fairly innocent, but still visible by a mild dip in SPL and often in Z too.
Furthermore, CSD 's are difficult to compare. Even minor adjustment in SPL levels or the time window, or graphics make huge differences for the same DUT. They should be taken with a grain of salt, although many here seem to think otherwise.
Ringing without any dip in the SPL response (ruler flat that is) or small peak impedance? Please show me an example of this. BTW:I am well aware of the infamous surround bounce. Fairly innocent, but still visible by a mild dip in SPL and often in Z too.
Furthermore, CSD 's are difficult to compare. Even minor adjustment in SPL levels or the time window, or graphics make huge differences for the same DUT. They should be taken with a grain of salt, although many here seem to think otherwise.
Phase is always important in a crossover no matter what order. If the drivers don't phase track through the overlap region reasonably well you'll get sub par results on axis and odd off axis issues as well.
I had the challenge in this design that the acoustic centre offset between the drivers is massive due to the considerable depth of the 8" full ranger, and despite the short waveguide in the tweeter alleviating the difference a little.
Some may have noticed that I flush mounted the tweeter but left the FR driver mounted on the front of the panel - that wasn't a mistake or oversight, it was done on purpose to get the acoustic centre offset down to slightly more manageable levels.
(Also the shape of the driver basket is a complete PITA to flush mount, they were originally designed to be mounted recessed from behind the panel - as was common in the 70's when it was designed)
Measurements with flush vs proud showed that the diffraction effects of doing this were very minimal because at the higher frequencies where diffraction would occur at that raised edge the driver is already quite directional thus the bump at the edge of the driver is not "illuminated" and doesn't cause much if any diffraction.
Likewise the ribbon tweeter is quite directional in the vertical plane so there is very little energy travelling down the face of the tweeter and front panel to diffract off the top of the FR driver frame - if it was a dome tweeter with much wider vertical dispersion I wouldn't have got away with that frame bump being just below the tweeter without issues.
Despite all this the offset was enough that I still needed two cascaded first order all pass filters on the tweeter to bring the two drivers into time and phase alignment in the overlap region and allow me to use a symmetrical 4th order L/R acoustic crossover.
Was it worth it ? Absolutely. I'm very impressed by how smooth and neutral the crossover region is on these. On axis it is seamless and there are no hot spots off axis in any direction. Good phase tracking right through the crossover region really does make a difference.
Yeah, it was pretty harsh but fair review.
Not everything he said was strictly accurate and there was a lot of the flowery language that subjective reviewers seem to like to use, but yes, I don't think he'll be getting another review unit from them.
But what did they expect when they produced a speaker with a full range driver in a narrow baffle running fully open with no low pass filter, no baffle step correction and a ribbon tweeter with a single cap as the crossover ? That just isn't going to end well.
I've tried the approach of no filter on the FR (except some baffle step correction) and a high pass filter on the tweeter and it just doesn't work well at all.
The comb filtering in the treble (as shown in the measurements in that review) is horrendous, the on and off axis responses differ enormously, and any coherence and precision in imaging is lost. I get why people try to do it, but it's not for me.
Here's a simple questioned that has me very perplexed. My problem is a three way system with 8 ohm woofer and midrange, however my high frequency transducer is 3.8 ohms. In my crossover design I corrected for the impredance difference. Assuming all 3 transducers exhibit similar 1 watt/ 1 meter SPL, should I increase the impedance of the high frequency transducer and how. Any suggestions other than a different high frequency unit is fair game, thanks.
Last edited:
Well, the radial ribs the Sigma range use certainly mean that the driver is not radially symmetric and that may help somewhat but beyond that there is little similarity with what I'm doing.
The pattern I have around the outside of the cone as well as damping and rearranging radial modes in different sectors also controls and stabilizes concentric/circular modes to prevent them from chaotically "free running" around the edge of the cone.
Instead they are locked in place in relation to the position of the damping strips (since the strips form nodes) and also locked into a fixed phase relationship with the bending modes.
Once they are stable they can be corrected with damping or EQ, and this stability eliminates the "random" cone breakup "noise" that is common with cones that are breaking up severely.
Not sure what you're trying to say, but again if we assume minimum phase then a given frequency response will have a given CSD delay as well. Whether you can see subtle changes in frequency response that correspond to a different CSD decay tail is another matter.
The information is there in the FR but hard to see, the CSD makes it easier to interpret.
Not sure what you mean by a "splashy sound" so I can't really comment. It's not a description I've heard in relation to speakers before. MP3 files yes.
If you do it with a FIR convolver then you might be amazed at the results - at least at one point in space.
If you're just talking about taking something like a 31 band graphic equalizer and pushing the bands up and down until it is as flat as you can get it, forget it.
To EQ a discrete resonance you need to use a PEQ - as the centre frequency, Q and gain all need to match the resonance exactly for it to be truly eliminated. You can't just use something like a graphic equaliser and push the nearest band up or down because that band will have the wrong centre frequency and Q.
Now you have just added more resonances and not removed the ones that were originally there, making the time domain response even worse!
The only two approaches that would truly eliminate resonances is to either use a FIR convolver filter that is based on a measurement of the driver, which can theoretically be almost perfect, or if there are only a few discrete resonances (for example I corrected two resonances in the 2-5Khz range on my drivers) then you can individually address them with multiple PEQ's, or in the passive crossover domain, RLC notches.
But graphic equalisers are not usable in this task.
I agree, although I'd say it doesn't really start to get annoying and fatiguing until 2Khz, so I'd say about 2Khz to 8Khz is where you'll get most listening fatigue from.
That's actually the frequency range I spent the most time and components optimising on my speakers, including notches at 2Khz and 4Khz on the woofer, a 1dB 1/4oct notch at 7.5Khz on the tweeter for a small cavity resonance, and a lot of massaging of the high pass and low pass coefficients to get the response almost ruler flat from 2Khz to 8Khz whilst keeping good phase tracking.
And it really paid off. There is no trace of harshness, listening fatigue, edginess or anything like that which you would normally associate with upper midrange cone breakup or treble resonances on anything I've listened to. They sound utterly clean.
Even the 1dB peak at 7.5Khz from the tweeter is noticeable as a slight edginess on some specific recordings when the notch is removed again, even though it can't be heard on most recordings and there is no subjective change in tonal balance. The only change is the elimination of a slight edginess.
Attention to detail really gets you the last mile with speakers I think. I think many designs are just too willing to accept the warts that the drivers have, leaving them uncorrected, when you can get a lot better result by addressing some of them.
Last edited:
You certainly have ringing from a notch - a notch rings exactly the same as a peak, but in anti-phase. That's how the notch can cancel the ringing as well as the frequency response error, because it rings exactly the same in opposite phase so the ringing sums to zero.
As for ringing without a frequency response deviation - not possible unless the response is not minimum phase.
An example of when this could happen is a so called "buried" resonance in a multiway speaker.
Say you have a two way with 3Khz crossover, and your woofer has a sharp uncorrected resonance at 8Khz. Lets say the low pass filter is 25dB down at 8Khz.
At 8Khz the summed response of both drivers is almost completely dominated by the tweeter. The bump in summed response due to the 8Khz resonance may not be visible on a FR graph but will show up as a long tail in a CSD after the tweeter has decayed.
But the reason this can happen is because the summed response of the two way speaker is not minimum phase - you have a multi-path situation from two different drivers.
Yes surround dips are notch resonances so they will ring. They are indeed less audible than peaks because at the frequency they ring the amplitude is suppressed.
It's well known that subjectively peaks are more readily noticeable than notches.
My Coral drivers have quite large surround dips at 1Khz - they use a fabric roll surround of some kind, original factory Coral drivers had the fabric doped with some damping agent (I don't know what was used) but the damping agent used went hard with age (over decades) until the surround could barely move, so an original factory pair I have can't produce any useful bass because the surround is so stiff.
The pair I am using in these speakers are hand assembled units from new old stock parts, and whoever assembled them did not add any doping agent to the surround at all. 15 years after assembly the surrounds are still in perfect condition and bass response is excellent because of this, but they have a nearly 6dB notch at 1Khz due to it which is not there in the driver with the stiff surround.
I could try to add doping agent but the risk of permanently damaging the response of the drivers (which I cannot replace) is too great, so instead I have applied some correction in the network to reduce the dip to 3dB, and this sounds fine to me. (In fact I didn't really even notice with the original 6dB dip)
If I had several pairs of drivers and they were readily available I would experiment with doping the surrounds. But not on my only good vintage pair.
Last edited:
Yes, CSD's are very hard to compare when different people are making the measurements. In particular the floor adjustment (how low to go) is an arbitrary choice that can make the same driver look good or bad.
I tend to use 25dB as the floor on the basis that once a resonance has decayed that far it should not be audible. Many published measurements use floors of only 20 or 15dB and this doesn't tell you anything useful IMO.
CSD's are very useful when making before and after comparisons of changes on the same drivers (such as damping tweaks) if you keep the same configuration for the measurements though.
So you went with the wide dispersion ribbon instead of something like a horn tweeter. Can you notice a difference in horizontal dispersion around crossover?
This hasn't quite "sunk in". It sounds like what you and Geddes and others are saying is that if I sit down and draw a random FR you can turn that into a CSD with no more information required. That means any two identical FR's also have identical CSD's. It also means that the only way to a good CSD is a flat FR and if I flatten the FR I will also make the CSD perfect?
I've been using equalizer APO which is a PC based minimum phase equalizer to control and equalize and even crossover the drivers, (everything is done before the amplifiers). It sounds like it does the job of getting rid of the annoying resonances, of which the aurasound has its share.
http://www.diyaudio.com/forums/pc-based/301618-equalizer-apo-rew-rephase-wow-10.html#post5377707
Last edited:
When I used to cross them over at 4Khz 3rd order, yes I could notice a dip in the off axis response and a bit of a hole in the power response in the room.
But now that I've brought it down to 3Khz and 4th order, no I don't find any subjective problem with the off axis response or power response.
A well designed 8" whizzer cone driver like this has significantly better dispersion in the 2-6Khz region than a conventional 8" woofer / midbass single cone driver would have. That's one reason I'm using the dual cone driver in the first place.
I've proven this to my satisfaction years ago by de-wizzering one of my spare Coral drivers, (one with the stiff surrounds that wasn't usable for bass so I wasn't going to use it anyway) taking before and after measurements both on axis and off axis, and the improvement in dispersion with the whizzer in place is substantial.
For example at 4Khz removing the whizzer increased the on axis response by 3dB but reduced the 30 degree off axis response by nearly 4dB - the driver started beaming badly at this frequency with the removal of the whizzer as the ratio between on axis and off axis worsened by 7dB from 2Khz to at least 4Khz, which is a lot...
Practically speaking with the whizzer it has the dispersion of a 4" full range driver from 2-6Khz, so right through the crossover region, because of this there isn't the large dispersion discontinuity you might expect from 8" driver to ribbon.
I'm sure there will still be an off axis dip there when I get around to measuring it properly, but it won't be any worse (probably better) than say a 6" midrange.
I would also add that the tweeter does have a small waveguide so there is some control of the horizontal dispersion as well, from about 2Khz up - its an Aurum Cantus G2:
Aurum cantus
So the tweeter is more directional than a typical dome even in the horizontal plane. This will help match it to the other driver a bit better.
There is an implicit assumption here that the signal being measured is minimum phase though, which means the phase can be derived from the amplitude response.
When you're talking about driver cone breakup resonances then except for unusual driver types like whizzer cones that are effectively a multiway system, the resonances will be minimum phase in which case there is a direct correspondence between amplitude response and impulse response, and between impulse response and CSD, yes.
But the level of signal that corresponds to a resonance tail that you can see at the bottom of a CSD may be so small that it's not even a "line thickness" on a graph.
The information is there in the impulse response that generated the frequency response line, but that line doesn't have enough visual fidelity to resolve that information. So anything you "draw" with a pen is not necessarily going to have the level of detail necessary if you tried to convert it back to an impulse.
A CSD can take that detail that is present in the impulse and display it in a way that you can clearly see what you're looking for, where it would be an imperceptible wiggle in the frequency response graph among other wiggles.
However not everything is minimum phase, most multi-way speakers sum to a non minimum phase result, in that case the amplitude response alone is not enough, you'd need the measured phase as well.
Last edited:
So I can take my cheap aurasound, and using a very precise equalizer like equalizer apo, match the FR of a RAAL ribbon and the CSD will be identical. So then except for dispersion (which the aura does OK), and non-linear distortion it will sound the same. It would seem that it wouldn't have the same quickness because of the heavier mass, but maybe this isn't in the CSD.
Last edited:
I need to look through old data, but you may be right about the impedance. However, normally it would be had to distinguish whether the spider or surround is influencing the impedance. I the impedance wriggle was caused by something else. But in the CSD, I did do some comparison before and after some surround tweaking. Lots of CSD differences I am finding seem to relate with Windows as I am finding out. This is quite annoying because XP seems to deliver lower noise floor. I am still investigating this as I evaluate shifting to different computers.
Currently just looking at CSD of sound card loop backs I see such significant differences, which suggest why I hear differences as well. Bear in mind that differences are heard first before I started to look into the situation. Even with the same cards and software on different computers, there are significant differences. So my question would be, why such difference caused by the different Windows version or the computer?
Last edited:
It is hard to quote longer responses in a phone, so I am going to skip the procedure.
Once the cone is ringing, I would not consider it minimum phase. As a matter of fact, it cannot be minimum phase if there is a breakup mode (normally called first bending mode in general engineering terms) because clearly the inner and outer parts of of the cone are not in phase with each other. Under some studies, it is possible to assume minimum phase, but in the process of perfecting a device, one cannot assume such, but rather best look at the actual physical dynamics which is where the Klippel scanner is very useful. This cross reference of measurements allow better understanding of the results from different data processing methods of commonly used measurement methods.
Once the cone is ringing, I would not consider it minimum phase. As a matter of fact, it cannot be minimum phase if there is a breakup mode (normally called first bending mode in general engineering terms) because clearly the inner and outer parts of of the cone are not in phase with each other. Under some studies, it is possible to assume minimum phase, but in the process of perfecting a device, one cannot assume such, but rather best look at the actual physical dynamics which is where the Klippel scanner is very useful. This cross reference of measurements allow better understanding of the results from different data processing methods of commonly used measurement methods.
Also what if a driver rolls off and has no ringing, then does making it flat create a ring?
Simon is not making the point strong enough. You could draw any line so long as a few simple rules were met, like singular value (at any freq there can be only one amplitude) and there can't be any jumps in the curve - it must be continuous. But for the impulse to be unique I also need to know the phase. I could assume that the system is minimum phase, which is almost always the case because non-minimum phase requires some kind of multi-path for the energy to take through the system - like a wizzer cone. Hence, if I assume the system is minimum phase then I can derive the phase from the magnitude and I have a complete frequency domain picture and all time results are then know - impulse response, CSD, anything in time.
But I could also allow you to just draw any phase response (with the same rules as above) and that too would completely determine all time responses. The time-frequency relationship is quite strong and is never violated (as long as the system is linear.)