Looking at the department below roughly 1kHz we see from the plots that dampening and EQing will become a critical task to achieve tonal balance there.
One more simu that I find "half ways" related:
We see that the FR of such a closed box speaker *to some extent * follows the same pattern as a dipole overlay.
I attenuated the back wave to some degree by inserting a lowpass filter and set the delay accordingly to the box depth .
Sure - this simple model does not account for many things - but I found it interesting enough to throw it in
😉
Michael
One more simu that I find "half ways" related:
We see that the FR of such a closed box speaker *to some extent * follows the same pattern as a dipole overlay.
I attenuated the back wave to some degree by inserting a lowpass filter and set the delay accordingly to the box depth .
Sure - this simple model does not account for many things - but I found it interesting enough to throw it in
😉
Michael
You are making too much of this. What is happening at low frequency (the continued roll off) is more a result of the windowing of the impulse. It is an artifact of truncating the impulse with the window, and then zero padding so the FFt can be performed. The response below 1/T window is not meaningful. In this case, 1/3.4msec = 294Hz so the response below 300 Hz is not representative of the system response.
I can not post the IR of my simulation as there is no way to dump an impulse file from the code I use. The circuit was simple. Just a 3 band parametric EQ with center frequencies set to 1, 1.2 and 1.4 K Hz, Q = 10, gain = 15dB, as I recall.
An externally hosted image should be here but it was not working when we last tested it.
I can not post the IR of my simulation as there is no way to dump an impulse file from the code I use. The circuit was simple. Just a 3 band parametric EQ with center frequencies set to 1, 1.2 and 1.4 K Hz, Q = 10, gain = 15dB, as I recall.
Agree (beside the notch past 1kHz possibly).
It's been intended more as a starter towards comparison and discussing "real" resonance and what just looks like a "real" resonance. - No good start, I admit.
What a pity.
I try if I can create the IR file - not sure it works though.
Thanks - this page I found.
Was just curious if there possibly could be found more of a *concept* (in contrary to "just modding") in his postings, but seems he isn't active in this any more - at least the postings seem to be rather old and also restricted to only a few specific Tang Band speakers.
An interesting approach though - kind of selective stiffening of the membrane by embossing IMO.
Michael
Last edited:
Thanks @JohnK for his most insightful post. So a seemingly related wiggle can show up even in a perfect LTI system from beating between narrow spaced resonances. If those were consistent within some reasonable off-axis range and vs. large signal conditions this would be correctable to some extent with simple analog inverse filters, wouldn't it?
Last edited:
One more question, is there a chance for two distinct phenomens mapping into the LTI part of a system's IR, assuming it was generated by a logsweep+convolution process?
Assume the wiggle came indeed from a damped (bandpass-filtered, slightly ringing) and delayed mechanical/acoustical reflection (let's assume one a single reflection for simplicity) as mapped into the IR by the measurement method, wouldn't then the FFT of this IR look akin to the example John has posted, although the root cause is different IHMO?
- Klaus
Assume the wiggle came indeed from a damped (bandpass-filtered, slightly ringing) and delayed mechanical/acoustical reflection (let's assume one a single reflection for simplicity) as mapped into the IR by the measurement method, wouldn't then the FFT of this IR look akin to the example John has posted, although the root cause is different IHMO?
- Klaus
Can't directly dump from software either, so did a measurement with the DCX set to 1000 / 1250 / 1500Hz at Q=10 and A=6dB:
(not a perfect solution but should be sufficient for the purpose)
Below a wavelet analysis that shows that there is a distinct difference to what we see in the plots of the TD15m regarding the 2.3ms wiggle:
The main difference to focus at is that its a periodic behaviour (which could also be mimic'ed by looped reflections though) but even more important, there is *no* delay involved - meaning - for such a case EQing is a perfect solution for correcting - also meaning: there is *no* CMP distortion involved here.
The mechanic aquivalent here would be three mass spring systems coupled to the cone. The mass being "compact" and the spring being "ideal" - quite a difference to a "standing wave" mechanism.
Michael
(not a perfect solution but should be sufficient for the purpose)
Below a wavelet analysis that shows that there is a distinct difference to what we see in the plots of the TD15m regarding the 2.3ms wiggle:
The main difference to focus at is that its a periodic behaviour (which could also be mimic'ed by looped reflections though) but even more important, there is *no* delay involved - meaning - for such a case EQing is a perfect solution for correcting - also meaning: there is *no* CMP distortion involved here.
The mechanic aquivalent here would be three mass spring systems coupled to the cone. The mass being "compact" and the spring being "ideal" - quite a difference to a "standing wave" mechanism.
Michael
Last edited:
Haven't seen your posting, Klaus - but I guess I'd shade some light with my posting above.
😉
Michael
I don't understand, maybe John can comment.
The measurement I usually do is MLS.
The FR responses are not exactly any meaningful if it comes to CMP behaviour as - simply put - FR changes with time.
Hence a wavelet analysis or any other form of time-frequency plot is way more revealing.
But I guess you had something else in mind ?
Michael
I think he might have just found no further interest in communicating here. Basically, it's proper mass loading at the right points in the direction the wave travels. It's basic physics, just how one looks at data to find the right points is a skill. I wish higher education were free, but unfortunately, it's not.😉 Note that his samples as well as EnABL are mostly applied to softer cone material.
Last edited:
Michael, I was trying to say that "optically" similar looking IRs could stem from different mechanisms, either from closely spaced resonances or delayed copies of the impulse, bandfiltered and damped, reflected repeatedly. Therefore they may give the same similarities when viewed in the frequency domain, with windowing. Or is this even a duality I'm overlooking here?
- Klaus
- Klaus
The original procedure seems to be related to dimpling a diaphragm - no?
So, no mass loading necessarily involved, it also works with stiffening (lateral to the dimples) and softening (perpendicular to the dimples) at the right places as it seems.
http://www.madspeaker.com/Proof.pdf
With my CMP concept I try to bring in some more clarity (mostly for myself I admit).
In the frequency domain there might be not "that much visible/audible" difference - depending at which time span we look at respectively our ear brain system likes to judge/weigh its clues.
From a technical point of view there is a fundamental difference between a resonance (in the strict sense of an ideal mass spring system) and interference (or even energy transfer) effects of reflections either sort.
CMP - as kind of missing link in audio - accounts perfectly for that IMO.
On the other hand its not mere hairsplitting I set up here.
At the example at hand we easily have demonstrated that the "uncorrectable time smear" of dipoles - Lynn is concerned with - happens just as well with closed box speakers.
CMP distortion happening here as well as there...
Michael
Last edited:
First I want to correct an error in a previous post. I said that the simulation was for what was essentially a 3 band parametric EQ. That was incorrect. The circuit is shown here:
It is the sum of three individual resonances.
Second, Michael, you really need to stop talking about time dependent frequency response. A LTI system has one and only one frequency response. Frequency response is, by definition, a steady state characteristic. What you are looking at is transient response. When you look at a wavelet analysis, a CSD, etc, you are not looking at a time dependent frequency response. You are looking at how different frequency components decay in time for a system with a given frequency response.
Third,
Klaus, I don't think the effect could come from reflections. We are seeing distinct resonant peaks in the frequency response. For the peaks to come from reflections there would have to be multiple reflections which were all filtered through sharp, narrow, band pass filters. However, whether they are the sum of individual resonances or sort of a parametric eq effect, if the result is minimum phase then it could be corrected at least at one in space, and perhaps over some window where the response was constant.
Also the resonances are showing up at 2k and higher. That means they should be present in the response even if a window on the order of 1 msec or less was used. That is, using a shorter window would help reveal if this is a reflection or inherent in the driver's behavior.
Anyway, I think this discussion is getting way off base. This looks like a simple cone breakup problem and I would not use the driver without an LP filter at 600 Hz or so.
An externally hosted image should be here but it was not working when we last tested it.
It is the sum of three individual resonances.
Second, Michael, you really need to stop talking about time dependent frequency response. A LTI system has one and only one frequency response. Frequency response is, by definition, a steady state characteristic. What you are looking at is transient response. When you look at a wavelet analysis, a CSD, etc, you are not looking at a time dependent frequency response. You are looking at how different frequency components decay in time for a system with a given frequency response.
Third,
Klaus, I don't think the effect could come from reflections. We are seeing distinct resonant peaks in the frequency response. For the peaks to come from reflections there would have to be multiple reflections which were all filtered through sharp, narrow, band pass filters. However, whether they are the sum of individual resonances or sort of a parametric eq effect, if the result is minimum phase then it could be corrected at least at one in space, and perhaps over some window where the response was constant.
Also the resonances are showing up at 2k and higher. That means they should be present in the response even if a window on the order of 1 msec or less was used. That is, using a shorter window would help reveal if this is a reflection or inherent in the driver's behavior.
Anyway, I think this discussion is getting way off base. This looks like a simple cone breakup problem and I would not use the driver without an LP filter at 600 Hz or so.
Last edited:
Well, if you look at the related patents... the long dimple can also serve similar purposes if additional weight is not desireable. Dot like dimples can serve different purposes. So, the goal is to control how the wave travels starting from the lowest modes. A point he addressed to me in personal message was less than 10 words, and it shed light on this basic concept I missed.
Last edited:
Sure. It's nothing radical. I just look at the periods of the oscillations and the overall decay envelope.
In John's example the time between peaks for the highest frequency oscillations looked to be about 0.7 msec. Since 1/(0.7 msec) is 1.5 kHz I expected a significant peak in the frequency response (magnitude) at that frequency. If that were all that was going on, then the impulse would have looked like a 1.5 KHz sine wave that started with the impulse and then died exponentially. The more damped the resonance is, the wider the peak will appear in the frequency response plot and the fewer the number of oscillations you'll see in the impulse response. In John's example the 1.5 KHz ringing didn't die monotonically, there was a lower frequency undulation in the envelope of the decay. I didn't catch the fact that there were three resonances so when I tried to explain the behavior based on only two I misjudged the frequency of the second one.
Had I been less lazy I could have fiddled with a few guesses by plugging them into the online Wolfram Alpha site and done a bit better, although I'm not sure I would have thought to try three resonances. Here's some quick code you can plug in to generate something like what John posted, but without any damping and therefore without any decay in the amplitudes. I'm using cosines instead of sines just because I wanted the waveform to start at a maximum at time zero.
plot (cos 1400*t ) + (cos 1200*t) + (cos 1000*t) from 0 to 0.05
I hope this quick explanation makes at least a little sense.
Few
[Edit: I just read John's post of the actual resonance frequencies so I've used those instead of the values I eyeballed from the original plots.]
Last edited:
Thanks Few! That's pretty much what I thought you were doing and my experiments completely agree with what you say.
Dan
Dan
I will – John – I will - as soon we find agreement on :
„Who's ranch now belongs to whom?“ (not meant *that* seriously of course)
http://www.diyaudio.com/forums/multi-way/100392-beyond-ariel-136.html#post2249887
😉
and also and *especially* as soon as you come up with somthing regarding point 4.) :
„I hold my position until you show that my measurements presented (the sine bursts) are flawed.“
http://www.diyaudio.com/forums/multi-way/100392-beyond-ariel-136.html#post2251309
😉
You'll gonna keep that 10 words as a secret ??
Michael
Last edited:
Nothing wrong with you measurement, but as you put it, it is a burst which is a transient signal. It takes time for the system to respond to it, and take sime for the system to stop after the signal is removed. How long depends on the form of the impulse and test frequency.
It is not the measurements which are flawed, but your logic.