Beyond the Ariel

mige0 said:



There are good reasons to take CSD with a grain of salt.

First its a highly processed result in order to make obvious what's hidden (well, for the most of us ;) ) and not necessarily telling "the whole truth and nothing else than the truth".
And also not telling about limits where calculation is valid or not.

Second any graph has its limits where we have to decide between "truth" and what we can grasp at most. If you look at the trace below it should become clear what I mean ?we are lost in "true" details here:

ir.gif


Third, if you imagine the time span of sub 1msec decay - and translate this into how many cycles that my be for the frequencies of interest you easily see that it doesn't make sense to crank resolution (and conclusions) too far ?no?


;)

Michael
I have not seen such noisy data for a VERY long time(since I stopped using SpeakerWorkshop a few ears ago). I don't use any smoothing. I wonder..:confused: Windowing problem? Processing method?
 
xpert said:


...
Your misconception is quite common with Cumm. Spectr. Decay. You should leave that CSD reading behind.

so long
Some people find CSD usefull, and some don't. From the way you explained, it seems like you only use "group delay" as a "pass or fail" criteria like in many mass production lines. Some people can use CSD like evaluating food of very complicated tastes and flavors where the chef decides what he wants. So if the chef can find his own balance of ingredients, it's his talent. Why tell a chef to do it otherwise?
 
Disabled Account
Joined 2009
ScottG said:


I'm not entirely sure what you are requesting here. (..and I realize that English as a second language isn't exactly easy.)

I do think I understand what you are referring to with group delay (or at least partially). Presumably you mean that phase rotation from filtration (mechanical or electrical) is inducing time decay problems. ...


Hi,

"Resonance" is in German "Resonanz". It is the same physical concept meant. I do not think that we have a problem with respect to my mothers tounge, do we? You persumably referred to this:

http://members.aon.at/kinotechnik/diyaudio/diy_audio/Jantzen_AMT/30deg CSD_corr.gif

No resonance. There is no problem induced by high pass filtering (mechanically) here either. Why do You think Group Delay is a problem? We know about the threshold of perception that lies somewhere above 1ms @ 2kHz and is greater than 2ms below 1kHz and above 3kHz.

First - if perceptible - how much is it anoying to the listener? When the listener focusses on music instead off evaluating loudspeaker boxes! In a living room.

Second would the extra group delay - that is inevitable with any linear filtering - diminish when the counterpart is induced. Sum up the lower pass band to that waveguide and You'll experience that Your bad resonance vanishes by miracle.

Third. Did You ever wonder why with all that ripples in amplitude response the so called "phase issues" are gone in parallel?

CSD is often misused as a catchy no brainer. May be it is so because CSD shows nearly no numbers. You have to take an impression by pure eye. It encourages You to judge by Your very own feels. No good with engineering. Really.

so long
 
You are mixing things at will - Mr. Xpert.

First you stated that what could be seen in my high pass traces isn't a resonance at all - which for sure is *not* correct - CSD *is* displaying exactly this resonance.

- now - after some thinking due to my (and others questioning) you state that its no resonance due to the fact that there is a balancing counterpart in the low pass filter.

You should be more precise when you come along with your "I'm the Xpert !" attitude.

As your first statemet is plain wrong your second statement *might* not be not completely wrong - depending on several circumstances - which have to be controlled tightly to do the trick - which in fact is a challenge by its own.

- and also a seperate discussion...
You are welcome to contribute to this anytime - in a more positive and thought through manner, *if* you have the capability...

Michael
 
Disabled Account
Joined 2009
mige0 said:
You are mixing things at will - Mr. Xpert.
...
- and also a seperate discussion...
You are welcome to contribute to this anytime - in a more positive and thought through manner, *if* you have the capability...

Michael

Michael,

You shouldn't take that "xpert" attitude to serious. At least You shouldn't mix up my assertions due to any anger about it. If I'm not welcome to correct misconceptions, example given related to CSD, who else?

On that topic of CSD You may read at S.Linkwitz' site, Zaphs, a lot more too. You are an engineer. 'Hope some numbers won't bother You to much. How does a resonance of specific parameters fr, Q etc show itself in CSD - tell me by numbers, no pictues, please. What is the benefit of CSD in respect to show things in a way humans can understand? I mean - understand thoroughly.

What is the difference between Group Delay originated sensations seen in a CSD and those from resonances? You differentiate it (against my argumentation) in Your posting above.

The other topics could be reconsidered later on.

have fun
 
xpert said:
If I'm not welcome to correct misconceptions, example given related to CSD, who else?


I think that the problem here is your assertion that you are correct and everyone else is wrong, which from what I can tell is not at all the case.

The CSD that Michael showed has a clear resonance right at the point of the HP filter. All filters have resonances. If there weren't any resonances then the spectrum would be flat. There has to be poles (resonances) or there can't be a non-flat response. (All zero filters don't occur in mechanical systems, but can of course be generated in electronics.)
 
Disabled Account
Joined 2009
gedlee said:


I think that the problem here is your assertion that you are correct and everyone else is wrong, which from what I can tell is not at all the case.

The CSD that Michael showed has a clear resonance right at the point of the HP filter. All filters have resonances. If there weren't any resonances then the spectrum would be flat. There has to be poles (resonances) or there can't be a non-flat response. (All zero filters don't occur in mechanical systems, but can of course be generated in electronics.)

I think that the problem here is your assertion that you are correct and everyone else is wrong, which from what I can tell is not at all the case - accidentially.

The CSD that Michael showed shows a clear group delay at the point where the lines are curved. Some filters don't have resonances, even if they can be seen as poles. Example given the first order filter, a single null pole. In spite of being realized by fashonable energy storage things it filters a lot unwanted signals right away. So, You are wrong again (remember Your mistake in the decay topic). An other filter is the interference effect off axis, that has no energy storage in it but filters linear, but with cutey ripples. Comb filter. No poles, but group delay. It is not You personal. But You see how thin the surface over the established knowledge is. As to say, stored misconceptions.

I'm not the expert, I fiddle around with Your self confident expertize. To bring light to the catchy graphics I asked Michael to check out the relation of goup delay and such fancy ondulation curves in the pretty CSD picture, but in this case numerically. If he dares, and I hope so, he will see, that CSD has to be thought over. Shouldn't we give him time to check it out for us?

so long

ps, found whilst ckecking for spelling:
technology development is intervention in conventionalized decisions. Doesn't at all have to be true, bud nevertheless sounds profpound. There is one trick with sentences like that: In spite of discussing the content, the meaning right away, all intelligence is invested to get the sentence linguistic. After that no time is left. The contents remains in Your mind without critics. Aha! May be it is so with CSD. Take a second look!
 
The first order filter does have a pole, but I accept your point that only 2nd order and above can be called "resonances". There are no mechanical first order filters that I know of, but in any case a driver on a waveguide is well above first order and does have a resonance (at least one) where Michaels data shows it.

I said that only electronics can have all zeros, and a comb filter is an electronic filter, albeit a single reflection will act as a comb filter, but again this is not a mechanical system per-se in that it does not have mass, etc.

You continuously grab at ever more obscure details when your arguments are refuted and use those to bolster your position instead of providing concrete proof of your claims.

The data that Michael showed has a resonance, and, of course, it also has group delay - they go together. Your claim that it does not is just as false as your claim that the systems always decay at the driven frequency.
 
Disabled Account
Joined 2009
gedlee said:

The first order filter does have a pole, but I accept your point that only 2nd order and above can be called "resonances". There are no mechanical first order filters that I know of, but in any case a driver on a waveguide is well above first order and does have a resonance (at least one) where Michaels data shows it.

You continuously grab at ever more obscure details when your arguments are refuted and use those to bolster your position instead of providing concrete proof of your claims.

Crazy - I've proved You wrong with those "obscured" details. Plain wrong. What is unacceptable with that? I don't bolster my position, but prove it right against You.

By now You tell me some "obscure" details about mechanical systems that - as such - can NOT be seen from Michaels data. And exactly that is the point. CSD is totally useless to understand the system without tons of prejudices and sloppy estimations on its meaning. It's some kind of mantra.

Further detail required? Within a cone we find some interferences that occure due to its shape alone. NO resonances. But only an effect of limited speed of sound. These interferences will show up as "resonances" in the CSD, accompanied by ondulations on the sp.curve. NO resonance but interference. Two kinds. If You are a believer You would start to check the cone with laser or something even more weired, 'cause their "must be a resonance". How does CSD differentiate that? It can't because CSD is in its substance nothing - nothing - more that Group Delay and amplitude vs. frequency. But it is less, made unreadable to the educated analyst.

BTW that was Your non-order filter that You see as a mechanical impossibility. Should we discuss the effect of mechanical mass as a first order filter?

so long
 
xpert said:


...
Further detail required? Within a cone we find some interferences that occure due to its shape alone. NO resonances. But only an effect of limited speed of sound. These interferences will show up as "resonances" in the CSD, accompanied by ondulations on the sp.curve. NO resonance but interference. Two kinds. If You are a believer You would start to check the cone with laser or something even more weired, 'cause their "must be a resonance". How does CSD differentiate that? It can't because CSD is in its substance nothing - nothing - more that Group Delay and amplitude vs. frequency. But it is less, made unreadable to the educated analyst.

...
This is not true according to my measurements of various drivers. There are actually two types of interferences: 1. acoustic; 2. mechanical. For acoustic interference due to cone shape, it does not show up as resonances in the CSD. For mechanical interferences, it will show up as resonances in the CSD, and such that if you alter or breakup the resonance, you will see the energy shift in distribution to other spectrum regions.

I have already posted a series of measurements on JX92S in the "EnABL process" thread. It would be interesting to see some of your measurements as examples.
 
Hi,

soongsc said:

So what difference will it make when you stop the input at 0 of a sine wave vs peak of a sine wave?

I'm not quite sure what you mean.

Both signals - switching a sine of at maximum or at zero transit - have different spectra, so the result is different.

Concerning resonances in CSD:
you have really to be careful. There are two ways of generating a CSD. The most common one is, where the only starting point of the FFT is shifted to the end of the impulse response. The other one (not so common) is where _both_ - starting and ending point are shifted towards the end.
This is dangerous with reflections because such a reflection may look like a resonance. You have to distinguish between three cases:

1.) the first form of the CSD is used. As the reflection (e. g. a second peak a few milliseconds later) is in the FFT window as long as the starting point is not behind it, it adds to every spectrum until then. This may look like a resonance, because it seems that the decay is slow (which is not)

2.) the second form, when the starting point is before the main impulse and the ending point behind the (2nd, 3rd,...). In this case the second form behaves similar case 1.)

3.) the second form, when the starting point is before the main impulse and the ending point before the first reflection. Then the reflection will show up as a rising edge in the CSD, so you're able to distinguish between resonances and reflections.

You may think that the second form is better than the first. You have to remind, that reflections are often within only a few milliseconds, so a FFT window to get case 3.) will be very short, which degrades the effective frequency resolution.

The solution is to use a burst decay, either with a shaped tone burst or an optimized wavelet (like Morlet). From a burst decay you can clearly distinguish between a resonance and a reflection (above a frequency where the originate burst is down to a much lower level than the reflection). Also, the burst decay does not lie (so much) as the conventional CSD.

Bye


Baseballbat
 
Well, I don't think anyone measuring driver performance whould put the window end anywhere else than before first reflection.

I don't even know any software that would move both the beginning of the window AND the end of the window when generating a CSD. Do you know which software does this?
 
"I don't even know any software that would move both the beginning of the window AND the end of the window when generating a CSD. Do you know which software does this?"

CLIO for one

You can sample anywhere you want. The origin is always "0" for the starting point and the end is however long the sample is

Rob:)
 
Soongsc - Agreed

Any decent software and/or user is going to exclude the reflections for all windows. That's clearly what was done in Michaels data. What appears as a resonance IS a resonance. I should know I took the data and I know the system that this data is from. There is a resonance at about 1 kHz, plain and simple.
 
Hi,

soongsc said:
Well, I don't think anyone measuring driver performance whould put the window end anywhere else than before first reflection.

fine, and then you got a real low frequency resolution. Typical measurement setup, the floor reflection is around 5 ms. That gives you a frequency resolution of 200 Hz, nothing more, and the lowest frequency which is _reliable_ is 400 Hz. That may be enough for tweeters, but how can you evaluate a bass driver?

I don't even know any software that would move both the beginning of the window AND the end of the window when generating a CSD. Do you know which software does this?

As I said, this form is not so common (at least in my experience). I saw ARTA doing that once, but I don't know if it is possible.

Bye


Baseballbat
 
xpert said:



Completely wrong.



1st - When I stated that I was having a problem with your own statement is was specifically regarding this:

"With this You emphazise that You have concerns. But we don't know where they come from. If I don't bring those effects to my perception, what is the reason? Bad system, non critical listening?"

I'm finding it difficult to understand *what* you specifically are talking about. Is this an attempt to request this?:

2nd - What I was referring to (and have all along), is Michael's post #5398 here:

http://www.diyaudio.com/forums/showthread.php?s=&threadid=100392&perpage=25&pagenumber=216



Also, I don't think that this was terribly difficult to follow either. My post #5486 here:

http://www.diyaudio.com/forums/showthread.php?s=&threadid=100392&perpage=25&pagenumber=220

PRECEEDED Michael's post #5488 (same thread page). Additionally, Michael's post actually included as a reference my post from #5486.

My latter reply - post #5496 (same thread page). Specifically referenced *at the beginning* my post from #5486 - i.e. "My recent analysis". Moreover that portion of Michael's post #5488 that I did reply to was referenced - with his other plots/graphs and statements specifically left out.

I guess basically what I'm saying here is that if you are going to drop an "expert" *opinion* into the discussion, that you should probably read that discussion a little more carefully. ;)
 
I don't intend to get involved in this discussion because it, once again, is arguing over things well understood 3 decades ago. However, one simple thing I will point out is that you should not confuse the length of a window with the FFt length. The window length will set the lowest frequency for which the FR may be considered reasonably accurate. The FFt length, which is completely unrelated, sets the frequency resolution. Windowing (or 1/2 windowing as used in most audio software) is used to smoothly bring the impulse to zero at the end of the window. After that the impulse can be zero padded to what Levey length is desired to set the FFt length and the frequency resolution.
 
Robh3606 said:
"I don't even know any software that would move both the beginning of the window AND the end of the window when generating a CSD. Do you know which software does this?"

CLIO for one

You can sample anywhere you want. The origin is always "0" for the starting point and the end is however long the sample is

Rob:)
But does it move the end in the process of calculating the CSD? What I've seen is "no". As a matter of fact, CLIO could never show the resolution required from one of our suppliers. A few years ago, an agent here revealed that there was somthing wrong in the sample rate of data, details of which I cannot remember now. But if we talk about CSD from a CLIO system a few years back, I agree the data was useless. CLIO did have some other nice features though.