Re: CSD
Hmm. That's weird. I looked at your site and found no CSD data anywhere. FR for drivers GR CSD??. How important is CSD?
BTW, what happened to all the measurements (FR,CSD,etc) for your speaker kits/systems? Those seemed to have done the disappearing act. Not important enough? Shame.
Me too
.
cheers,
AJ
Danny said:
Hmm. That's weird. I looked at your site and found no CSD data anywhere. FR for drivers GR CSD??. How important is CSD?
BTW, what happened to all the measurements (FR,CSD,etc) for your speaker kits/systems? Those seemed to have done the disappearing act. Not important enough? Shame.
Me too
.cheers,
AJ
I have a new server and on line ordering system. It won't allow pages that aren't product pages unless built into the drop down menu. We are working on a solution to get all the measurements back up.
That particular driver uses Dan's motor structure and yes he designed the motor structure for that one. T/S parameter targets and other materials were of my choosing. It is manufactured by Peerless of India. They do a great job with it.
They were quoting a background prior to founding the business in 1995.
Nice poke, and very typical of most of these forum debates. I am still waiting for that visit.
Re: on and on and on and on...
No Danny, you're wrong. Changing the moving mass must change the FR. A driver is a mass controlled device above Fs. It's also a bandpass. That bandpass, whatever it is, will without question exhibit a change in FR if the mass alone is changed. The Fs will change, thus the highpass will change. Also, the sensitivity and any influences on resonances will change. Remember, resonances are composed of a mass and a compliance. Change one, you change the resonance. You should recognize that if you really understand the principles of drivers. Your position on this is contrary to the physics.
Whether FR or CSD, the time window is paramount.
What point would that have been? You and I chose 4ms for entirely different reasons, apparently. Mine was constrained by the measurement conditions.
Again, as owdi put it, you're confusing "scale" with "window". Your argument was against the 12msec "window", not the 12msec "scale". A window is a window, it has nothing to do with the scale chosen for display.
So why then use a useless graph in the discussion?
You do like circumlocution, don't you? If I'm going to discuss a topic such as the CSD, with which I've experience, why would I not indicate how I might view it? That's immaterial, anyway, and just another straw-man argument. That argument is as useless as the CSD.
Hmmm, yet you posted a CSD of a measly 12msec that in fact was full of reflections by your own account. A 60ms impulse is mostly useless beyond the first reflection except for in-room resonse, so what's the point? The 12ms CSD didn't really illustrate much other than the fact that a poorly chosen window shows up just as badly in the CSD as it does in the FR and that the scale was also chosen poorly.
Yes, the thing to do. However, when I pointed out that I'd prefer a window of 12msec, you seemed rather surprised.
Oh, absolutely. One has to debate the issues as stated. If two drivers have THE SAME FR, there is no getting around the fact that they'll have an identical impulse response, CSD and anything else derived from that FR. Period.
Why do you read things into statements that aren't there? A "crossover" does not necessary have to only act within some narrow band around Fc. If that's what you consider a crossover, then I can understand why you'd stick to drivers with benign responses. I tend to do that to some degree, as I'm not a hard cone enthusiast, but in my lexicon a crossover encompasses the full bandwidth of the driver. That includes resonance control outside of the Fc region.
Oh, yes, it's possible in some cases to absolutely eliminate a FR abnormality. Done passively it's tough and some can't be done without DSP. But Minimizing to the point that a peaky driver is the equivalent in terms of FR, CSD, whatever, to some benign driver is entirely possible. It may be a costly crossover, but in the end the FR can be nearly the twin. Of course this is where the art of design comes in. You have your preference, others have theirs. Neither one is an example of some absolute that you seem to think yours is.
So what? That only says that the designer made a choice, one which displeases you. That may have been due to cost constraints, preference for "minimalist" crossovers, whatever. That in no way supports your contention. You're not debating the physics, just your preferences here.
It does not do so only if the designer chooses for it not to do so. Nothing prevents it from being done, one way or another. The above is just another baseless statement meant more to confuse or misdirect rather than elucidate.
BTW, I usually START with elimination (yes, I did say elimination) of the worst FR offenders, even if well "out-of-band" as it were. They may not be totally eliminated, but they can easily be reduced to be as benign as the small, untreated nonlinearities in "benign" drivers. I doubt that I'm the only one who takes that approach.
Dave
No Danny, you're wrong. Changing the moving mass must change the FR. A driver is a mass controlled device above Fs. It's also a bandpass. That bandpass, whatever it is, will without question exhibit a change in FR if the mass alone is changed. The Fs will change, thus the highpass will change. Also, the sensitivity and any influences on resonances will change. Remember, resonances are composed of a mass and a compliance. Change one, you change the resonance. You should recognize that if you really understand the principles of drivers. Your position on this is contrary to the physics.
Whether FR or CSD, the time window is paramount.
What point would that have been? You and I chose 4ms for entirely different reasons, apparently. Mine was constrained by the measurement conditions.
Again, as owdi put it, you're confusing "scale" with "window". Your argument was against the 12msec "window", not the 12msec "scale". A window is a window, it has nothing to do with the scale chosen for display.
So why then use a useless graph in the discussion?
You do like circumlocution, don't you? If I'm going to discuss a topic such as the CSD, with which I've experience, why would I not indicate how I might view it? That's immaterial, anyway, and just another straw-man argument. That argument is as useless as the CSD.
Hmmm, yet you posted a CSD of a measly 12msec that in fact was full of reflections by your own account. A 60ms impulse is mostly useless beyond the first reflection except for in-room resonse, so what's the point? The 12ms CSD didn't really illustrate much other than the fact that a poorly chosen window shows up just as badly in the CSD as it does in the FR and that the scale was also chosen poorly.
Yes, the thing to do. However, when I pointed out that I'd prefer a window of 12msec, you seemed rather surprised.
Oh, absolutely. One has to debate the issues as stated. If two drivers have THE SAME FR, there is no getting around the fact that they'll have an identical impulse response, CSD and anything else derived from that FR. Period.
Why do you read things into statements that aren't there? A "crossover" does not necessary have to only act within some narrow band around Fc. If that's what you consider a crossover, then I can understand why you'd stick to drivers with benign responses. I tend to do that to some degree, as I'm not a hard cone enthusiast, but in my lexicon a crossover encompasses the full bandwidth of the driver. That includes resonance control outside of the Fc region.
Oh, yes, it's possible in some cases to absolutely eliminate a FR abnormality. Done passively it's tough and some can't be done without DSP. But Minimizing to the point that a peaky driver is the equivalent in terms of FR, CSD, whatever, to some benign driver is entirely possible. It may be a costly crossover, but in the end the FR can be nearly the twin. Of course this is where the art of design comes in. You have your preference, others have theirs. Neither one is an example of some absolute that you seem to think yours is.
So what? That only says that the designer made a choice, one which displeases you. That may have been due to cost constraints, preference for "minimalist" crossovers, whatever. That in no way supports your contention. You're not debating the physics, just your preferences here.
It does not do so only if the designer chooses for it not to do so. Nothing prevents it from being done, one way or another. The above is just another baseless statement meant more to confuse or misdirect rather than elucidate.
BTW, I usually START with elimination (yes, I did say elimination) of the worst FR offenders, even if well "out-of-band" as it were. They may not be totally eliminated, but they can easily be reduced to be as benign as the small, untreated nonlinearities in "benign" drivers. I doubt that I'm the only one who takes that approach.
Dave
I didn't mean to ONLY change the moving mass.
Elementary stuff Dave. You're preaching that stuff to the wrong person. Yep, all that stuff has to change too if keeping the frequency response the same.
I have seen similar drivers play the exact same range at the same SPL levels (through a given pass band) yet have quite a difference in moving mass and CSD.
I don't know about the software you use, but in the latest Clio system the time segment selected for view sets the scale. You can't take a 3ms time window and put it on a 12ms scale. It doesn't work that way.
It wasn't useless, it showed exactly what I meant for it to. Going way out there to a 12ms time window on a CSD crunched everything that one would want to see into the first few lines and made it difficult to read and or get meaningful info from.
You just don't understand do you? I didn't pull that graph up from something I used. I didn't choose that 12ms window for any valuable informational purposes. I pulled up a saved impulse response (60ms measurement) of an M130 woofer and gated out 12ms of it just for you. It was so crunched up into the first few segments that you couldn't get anything from it, but right away you thought is was giving you more information and you started analyzing it not even knowing how it was taken.
Well if you want to get picky about it you can't find two identical drivers that have the same frequency response, not exactly the same.
But there is plenty of drivers out there that will have basically the same response through a certain range that will still have a different spectral decay.
More elementary comments... I think you have insulted my intelligence enough now, and I am beginning to think you have been a waste of time. Dave, I know crossovers. That is why many companies come to me to re-design their crossovers for their production models or to completely design their product lines for them. It's what I do and I do it well.
Sure it is possible in some cases. I have design several infinite slope crossovers that were pretty effective. But you know that is not what happens in most cases. I see it all the time with metal cone drivers. They never get all the break up out of it. Even you admit to it yourself in the last paragraph of your last response (which was extremely funny, BTW).
Next you'll tell me that it will sound the same.
Well you can "absolutely" think what you'd like. You can adopt the worst ringing metal cone driver you can find and put all the components in the world on it to fix and correct it into some useable fashion. I'd rather work with a well designed driver (what you call benign) that is free of such problems and does not require it to be fixed before use. Decide for yourself, but it looks like a better solution to me.
I am still laughing about this one. Dave you can take comfort in knowing that you are just like many others that take that same approach. Just keep telling yourself that if you put enough parts on it to fix all the problems and it will sound great.
I think by now (apparently way before) anyone WIth ability to reason seems to know who lost this debate. Everything that has to be said has already been said. Then why continue? I'm unsubscribing to this thread. Why spend more time on this? Let the loser be alone here. No more pointless debate please...
You need to write more precisely
Then you should say that. Your statement was flat out change the mass and the FR remains the same. That sort of inaccurate statement will mislead the inexperienced. Your words without elaboration were:
What do expect people to read from this, that you didn't really MEAN what you said?
Then say it, otherwise it is inaccurate.
They don't have an identical FR and that's the point. If they did, the CSD would be identical. How do you think that the CSD is calculated? It's a mathematical relationship that is programmed to be extracted from the impulse response. That impulse response can either be measured or generated from the FR. If the FR is the same, the impulse will be the same and the CSD will be the same. Period. If you can't recognize that or don't want to accept the physics of it, then you're sadly mistaken. Don't mislead others into thinking likewise.
I use LAUD, you should know that if you read the conditions page at my site. You've been there.
This has been around since at least 1996. I bought it in 1998. I can set the scale for whatever window is used. You could use Praxis and do the same. I've never been a fan of Clio. It does seem to have fewer options for maximum usage. If it can't change the scale of a measurement display, it's rather poor in my book. Don't use that to buttress a point when it just makes it clear that the software has limitations.
All it did was demonstrate a limitation of Clio.
You indicated that you take 60ms impulse measurements. Well, you actually don't specify, you get what Clio gives you, as is typical (unless it is quite different). Then you choose how much you want to use, unless Clio is an exception. Right away I assumed that you were providing something useful such as a longer window that was a good reflection free one, not one with bad data in it. The discussion had been on THE WINDOW, not the scale. The scale of the display in Clio apparently makes a long reflection-free window (the longer the better) somewhat unusable in some ways, due to its display limitations. That's a problem of Clio, not of the window length.
How is it being picky when I'm being accurate?
"Basically the same" is not "the same". Either the debate uses accurate descriptions or it is pointless.
It was not I that insulted your intelligence. I merely participated in a debate and pointed out errors. You felt it necessary to come over and jump in. Your first post started this way:
I think that was both the wrong attitude, counter-productive, condescending and in the end it has been you who have been misleading and inaccurate and that is sad given your "resume".
I don't doubt that one bit. I do doubt some of your understanding or at minimum your thoroughness when debating. You say one thing, then at a later time said that you meant something else. That doesn't cut it if one is not to be misleading and cares to be accurate and thorough.
Then why don't you acknowledge this up front rather than try to make some insupportable absolute statements? There'd be no disagreement with that, not on my part anyway.
That's immaterial to the discussion on the concepts that was the topic of the debate. What's possible and what people do are totally separate issues. That comment is just more obfuscation.
So why would acknowledging the limits of reality be funny? What has that got to do with debating an issue of the physics?
Ah, yes, back to the distractions, condescension and baseless nonsense. Good debating tactic, yes indeed. I'm sure that this impressed you when you wrote it.
And I take no issue with that whatsoever. That's largely my position as well, that will surprise you I suspect. As I said in an earlier post, I'm not generally fond of hard cone drivers, but that whizzed right by you. You see, we were debating the physics, not my preference and not your preference. You decided that you had to jump because some neophyte of yours informed you and said that it was funny. Well, what's transpired is more sad commentary on you and just makes me question your understanding even more, at least your desire and/or ability to accurately and thoroughly debate and that is unfortunate.
It's too bad that debates with you devolve into you making ignorant comments such as this. Your defense mechanism always seems to kick in. Some of this may impress those neophytes, but not those with a thorough understanding of the principles involved.
Any more "debate" is pointless when you have such an attitude.
Dave
Danny said:
Then you should say that. Your statement was flat out change the mass and the FR remains the same. That sort of inaccurate statement will mislead the inexperienced. Your words without elaboration were:
What do expect people to read from this, that you didn't really MEAN what you said?
Then say it, otherwise it is inaccurate.
They don't have an identical FR and that's the point. If they did, the CSD would be identical. How do you think that the CSD is calculated? It's a mathematical relationship that is programmed to be extracted from the impulse response. That impulse response can either be measured or generated from the FR. If the FR is the same, the impulse will be the same and the CSD will be the same. Period. If you can't recognize that or don't want to accept the physics of it, then you're sadly mistaken. Don't mislead others into thinking likewise.
I use LAUD, you should know that if you read the conditions page at my site. You've been there.
This has been around since at least 1996. I bought it in 1998. I can set the scale for whatever window is used. You could use Praxis and do the same. I've never been a fan of Clio. It does seem to have fewer options for maximum usage. If it can't change the scale of a measurement display, it's rather poor in my book. Don't use that to buttress a point when it just makes it clear that the software has limitations.
All it did was demonstrate a limitation of Clio.
You indicated that you take 60ms impulse measurements. Well, you actually don't specify, you get what Clio gives you, as is typical (unless it is quite different). Then you choose how much you want to use, unless Clio is an exception. Right away I assumed that you were providing something useful such as a longer window that was a good reflection free one, not one with bad data in it. The discussion had been on THE WINDOW, not the scale. The scale of the display in Clio apparently makes a long reflection-free window (the longer the better) somewhat unusable in some ways, due to its display limitations. That's a problem of Clio, not of the window length.
How is it being picky when I'm being accurate?
"Basically the same" is not "the same". Either the debate uses accurate descriptions or it is pointless.
It was not I that insulted your intelligence. I merely participated in a debate and pointed out errors. You felt it necessary to come over and jump in. Your first post started this way:
I think that was both the wrong attitude, counter-productive, condescending and in the end it has been you who have been misleading and inaccurate and that is sad given your "resume".
I don't doubt that one bit. I do doubt some of your understanding or at minimum your thoroughness when debating. You say one thing, then at a later time said that you meant something else. That doesn't cut it if one is not to be misleading and cares to be accurate and thorough.
Then why don't you acknowledge this up front rather than try to make some insupportable absolute statements? There'd be no disagreement with that, not on my part anyway.
That's immaterial to the discussion on the concepts that was the topic of the debate. What's possible and what people do are totally separate issues. That comment is just more obfuscation.
So why would acknowledging the limits of reality be funny? What has that got to do with debating an issue of the physics?
Next you'll tell me that it will sound the same.
Ah, yes, back to the distractions, condescension and baseless nonsense. Good debating tactic, yes indeed. I'm sure that this impressed you when you wrote it.
And I take no issue with that whatsoever. That's largely my position as well, that will surprise you I suspect. As I said in an earlier post, I'm not generally fond of hard cone drivers, but that whizzed right by you. You see, we were debating the physics, not my preference and not your preference. You decided that you had to jump because some neophyte of yours informed you and said that it was funny. Well, what's transpired is more sad commentary on you and just makes me question your understanding even more, at least your desire and/or ability to accurately and thoroughly debate and that is unfortunate.
It's too bad that debates with you devolve into you making ignorant comments such as this. Your defense mechanism always seems to kick in. Some of this may impress those neophytes, but not those with a thorough understanding of the principles involved.
Any more "debate" is pointless when you have such an attitude.
Dave
He's right. You have shown nothing to support your claim. I am still waiting to some someone predict what the CSD will look like just from looking at a frequency response chart.
All I have seen so far is one misread of one CSD that was made without even knowing how it was taken.
You and I know that well don't we? I still don't see anything to substantiate the uselessness of a CSD. All I have seen is semantics played with word choice.
The frequency response is the first line in the CSD. You can see FR in the CSD. However you can't see the rest of the CSD from the first line on the chart or from frequency response.
I have seen zero, zip, notta, to support any claims otherwise.
I think we are done here.
Hmmmm.............
Given the choice of a highly accurate CSD or high resolution frequency
response plot both on a single axis and no further data, which one
would you rather work with ?
Given the way they are derived from the same windowed data
why not simply show the windowed impulse response ?
Whilst strictly true I cannot say I agree with this .....
If one measurement I agree with this ......
Graphically this is true, but this does not mean a high resolution
frequency response does not tell you the same thing as a CSD,
it certainly does not present the information in the same way,
the frequency response in the CSD is far lower resolution.
What Zaph is implying can easily be reversed :
Show me a man who only needs to look at the impulse file .........
/sreten.
Given the choice of a highly accurate CSD or high resolution frequency
response plot both on a single axis and no further data, which one
would you rather work with ?
Given the way they are derived from the same windowed data
why not simply show the windowed impulse response ?
Whilst strictly true I cannot say I agree with this .....
If one measurement I agree with this ......
Graphically this is true, but this does not mean a high resolution
frequency response does not tell you the same thing as a CSD,
it certainly does not present the information in the same way,
the frequency response in the CSD is far lower resolution.
What Zaph is implying can easily be reversed :
Show me a man who only needs to look at the impulse file .........
/sreten.Listen, the whole point of this debate is Zaph saying that CSD contains nothing worthwhile. I find that ridiculous, evergy storage is of course a real phenomena. Then to try and prove that point the others in this thread have jumped in questioning the accuracy of CSD data, like distortion sweeps taken in somebodys basement are any more golden. We're not talking about submitting the CSD measurement back to the manufacturer so they'll change their documentation, we're just looking for trends in energy storage, and for that even less than perfect data will show what you want ...
Burst decay I think is a better (more accurate) method of measuring energy storage. What I do is a CSD first, and then investigate closer in the problem areas with a burst decay ... Linkwitz does both as well.
Nobody is arguing the inherent inaccuracy of a CSD, but it doesn't matter, it is repeatable and comparable data. Unless you can show that energy stoarage in a driver isn't real, people are going to need to measure it. And even a flawed measurement is better than sticking your head in the sand and ignoring the problem.
Burst decay I think is a better (more accurate) method of measuring energy storage. What I do is a CSD first, and then investigate closer in the problem areas with a burst decay ... Linkwitz does both as well.
Nobody is arguing the inherent inaccuracy of a CSD, but it doesn't matter, it is repeatable and comparable data. Unless you can show that energy stoarage in a driver isn't real, people are going to need to measure it. And even a flawed measurement is better than sticking your head in the sand and ignoring the problem.
Okay I have an idea. Enough word semantics. Let's see some real data, and let's see who can put up or shut up. This one is for all of those guys that think they can read a CSD in the frequency response.
Here is the premiss. We are looking at woofers for a two way speaker. These are all 6" woofers.
Here are the measured responses of three different woofers, made by three different manufacturers. They are all measured in the same box so they all fit the same hole.
The box is no larger than the woofers themselves in regard to width (on all four sides) but the box is very deep. It was a test box that was handy. It is fine for comparing these three woofers.
So that a huge baffle step loss wouldn't wipe out the bottom end I took near field measurements. This isn't all I'd look at if picking between drivers and I'd be using far field measurements, but these will work fine to illustrate the point.
Output levels were matched at one spot and the output levels were all very close and none needed any large amounts of additional power to make it level with the other two. All were very close in SPL.
Here are the three together:
Now here they are individually:
Don't ask what the woofers are. They are not any of mine and all belong of have belonged to other companies that I have done design work for. So I won't be telling what they are.
Okay answer these two simple questions.
1) In each graph what area has the greatest amount of stored energy?
2) Of the three graphs which woofer as the greatest amount of stored energy and where?
Okay were are the predictions?
Here is mine: The guys that pushed the hardest to claim that they can tell you what will be in the CSD by looking at the frequency response won't respond.
Here is the premiss. We are looking at woofers for a two way speaker. These are all 6" woofers.
Here are the measured responses of three different woofers, made by three different manufacturers. They are all measured in the same box so they all fit the same hole.
The box is no larger than the woofers themselves in regard to width (on all four sides) but the box is very deep. It was a test box that was handy. It is fine for comparing these three woofers.
So that a huge baffle step loss wouldn't wipe out the bottom end I took near field measurements. This isn't all I'd look at if picking between drivers and I'd be using far field measurements, but these will work fine to illustrate the point.
Output levels were matched at one spot and the output levels were all very close and none needed any large amounts of additional power to make it level with the other two. All were very close in SPL.
Here are the three together:
Now here they are individually:
Don't ask what the woofers are. They are not any of mine and all belong of have belonged to other companies that I have done design work for. So I won't be telling what they are.
Okay answer these two simple questions.
1) In each graph what area has the greatest amount of stored energy?
2) Of the three graphs which woofer as the greatest amount of stored energy and where?
Okay were are the predictions?
Here is mine: The guys that pushed the hardest to claim that they can tell you what will be in the CSD by looking at the frequency response won't respond.
Hmmm........
No way could you ever use an impulse file equivalent to 1/3 octave
frequency responses to generate a particularly meaningful CSD.
Quite the reverse in fact. Derive an impulse response from the 1/3 octave
data and then use this to generate some relatively good looking CSD's.
/sreten.
No way could you ever use an impulse file equivalent to 1/3 octave
frequency responses to generate a particularly meaningful CSD.
Quite the reverse in fact. Derive an impulse response from the 1/3 octave
data and then use this to generate some relatively good looking CSD's.
/sreten.
I'll answer, but Danny, you're not allowed to use a CSD to prove your point. That's the whole idea.
Let's make a couple points clear.
First, 1/3 octave smoothing is a joke, plain and simple.
Second, a nearfield measurement, while helpful below 1k, has to be interpreted very carefully above 1k due to the interference nulls that will occur that aren't really there.
Having said that, here's how I would interpret your rather poor quality measurements.
THE FOLLOWING IS BASED ON 1/3 OCT smoothing, which renders it very suspect.
-----
The red and blue curves have the best energy storage below 1k. However, the red would be a challenge to correct for and probably comes out below the blue because the peak and dip in the 1-2k region would be extraordinarily hard to equalize and hence, there would be quite a bit of energy storage here. Depending on the CSD windowing, this could look quite bad and the CSD would look artifactually bad below 1k. The dip of the blue in the 1-2k region, if real is also challenging to correct.
The green will have energy storage issues below 1k, but in fact, this would likely be the easiest of all to correct and so, in a good design would probably have the least overall energy storage.
Note also, the ones with the "worst" energy storage the way you are interpreting them, may not be the ones that affect sound quality the most. Perception of timbre change is worse with broad, low Q resonances. Of course, looking at a high Q peak, on first glance is visually unapplealing. But that's not the way our ears look at it.
-----
Now, it does come back to the smoothing. The CSD is taken from an unsmoothed impulse. The FR is intentionally smoothed. So you're giving us crippled FR info and expecting us to reconstuct the CSD at a higher resolution. Not very fair.
Based on the graphs you've shown us I stand by my analysis.
Let's make a couple points clear.
First, 1/3 octave smoothing is a joke, plain and simple.
Second, a nearfield measurement, while helpful below 1k, has to be interpreted very carefully above 1k due to the interference nulls that will occur that aren't really there.
Having said that, here's how I would interpret your rather poor quality measurements.
THE FOLLOWING IS BASED ON 1/3 OCT smoothing, which renders it very suspect.
-----
The red and blue curves have the best energy storage below 1k. However, the red would be a challenge to correct for and probably comes out below the blue because the peak and dip in the 1-2k region would be extraordinarily hard to equalize and hence, there would be quite a bit of energy storage here. Depending on the CSD windowing, this could look quite bad and the CSD would look artifactually bad below 1k. The dip of the blue in the 1-2k region, if real is also challenging to correct.
The green will have energy storage issues below 1k, but in fact, this would likely be the easiest of all to correct and so, in a good design would probably have the least overall energy storage.
Note also, the ones with the "worst" energy storage the way you are interpreting them, may not be the ones that affect sound quality the most. Perception of timbre change is worse with broad, low Q resonances. Of course, looking at a high Q peak, on first glance is visually unapplealing. But that's not the way our ears look at it.
-----
Now, it does come back to the smoothing. The CSD is taken from an unsmoothed impulse. The FR is intentionally smoothed. So you're giving us crippled FR info and expecting us to reconstuct the CSD at a higher resolution. Not very fair.
Based on the graphs you've shown us I stand by my analysis.
ucla88 said:
Sorry I'm a bit behind the thread here but I don't see why that has to be the case.
I agree with respect to the piston regime, as in that case it is easy to see that the response can be inverted (an easy test for minimum phase).
But it is less obvious with (for example) cone edge and breakup effects where there are intrinsic time delays and zeros resulting from that.
Or am I missing an assumption about the measurement that allows the situation to be simplified and the non-minimum phase behaviour to be ignored?
Ken
Post a proper set of measurements
The others are correct. The first thing noticeable was the 1/3 octave smoothing. No good designer would use them as the basis for choosing a driver, especially given the proximity.
Second, the proximity is definitely an issue. Please post a meaningful set of measurements, something 0.5m or more, but include the step, no smoothing whatsoever and properly windowed, square, no Blackman-Harris or any other. I never work with any smoothed responses for the mid and below, only a properly selected window. That's needed to have an idea as to what will be required to do a proper design as well as highlight the resonances that show up clearly in a clean FR. Since you're not posting large baffle quasi-anechoic data, we need a meaningful starting point. These measurements do not qualify based on your admissions as to their measurement conditions.
Also, please post the FR and impedance ascii files, I want to run do an initial crossover for one or more of them. I'll decide which one after I see a valid, useful set of measurements. But the step has to be there to be valid, since that is integral to a proper design. Of course, it is understood from the start, even if you provide a 0.5m measurement, that unless you can get a longer reflection-free measurement, all data below 1K cannot be considered to be accurately representative. But provide them anyway.
I want to avoid a repeat of the posting a bad measurement window that was posted for the 12ms CSD.
Dave
The others are correct. The first thing noticeable was the 1/3 octave smoothing. No good designer would use them as the basis for choosing a driver, especially given the proximity.
Second, the proximity is definitely an issue. Please post a meaningful set of measurements, something 0.5m or more, but include the step, no smoothing whatsoever and properly windowed, square, no Blackman-Harris or any other. I never work with any smoothed responses for the mid and below, only a properly selected window. That's needed to have an idea as to what will be required to do a proper design as well as highlight the resonances that show up clearly in a clean FR. Since you're not posting large baffle quasi-anechoic data, we need a meaningful starting point. These measurements do not qualify based on your admissions as to their measurement conditions.
Also, please post the FR and impedance ascii files, I want to run do an initial crossover for one or more of them. I'll decide which one after I see a valid, useful set of measurements. But the step has to be there to be valid, since that is integral to a proper design. Of course, it is understood from the start, even if you provide a 0.5m measurement, that unless you can get a longer reflection-free measurement, all data below 1K cannot be considered to be accurately representative. But provide them anyway.
I want to avoid a repeat of the posting a bad measurement window that was posted for the 12ms CSD.
Dave
kstrain said:
Hey Ken,
I guess it sounds overly dogmatic. However, I am not aware of a single well documented example of a driver not following minimum phase behavior. The possiblity does exist that cone breakup may be nonminimum phase, or some other novel driver would exhibit nonminimum phase behavior. I've read some AES journals that have suggested in an oblique way this is possible, but as a practical matter no one's ever come up with a real world driver that clearly isn't minimum phase, or not that has been documented. So it's not likely to be very common.
ucla88 said:
Isnt this Danny's point though? You're supposed to guess, get it wrong, then he shows you a CSD proving that your guess work was in fact bang off.
From what I understand here, if you are given a highly accurate and detailed frequency response, then you should be able to predict exactly whats going to happen in the CSD.
Danny gave us several examples from Zaph's measurements where this appeared to not be the case. If what you are saying is true, about CSD predictability, then this points towards two possibilities.
1) Zaph's frequency responses are not detailed enough to predict the CSD.
2) Due to limitations in the measuring process Zaph's CSD plots are inaccurate in the frequency range Danny asked us to look at.
Now we cannot deny here, that in the examples Danny gave, what he is saying is true, but at the same time we cannot ignore what everyone else is saying either.
If Zaph's measurements are not detailed enough to predict one from the other, then it would make sense to look at the CSD - it would tell us more information then the frequency plot can tell us alone. This doesn't mean you can't predict one from the other, it just means the data isn't sufficient to do it.
Now if we are using the same impulse response to create both then you could say - if we can't see it properly in the frequency response, then the CSD isn't going to be any more accurate, as its all the same data - However in these cases it was said that above 1k the CSD should be reasonably accurate and below will still show trends, even if the results can sway from reality.
Yes perhaps you can predict the CSD from the frequency response IF its detailed enough, but in these cases that seems to not be whats happening. Perhaps Danny isn't trying to argue that you cannot predict the CSD, more that 90% of the time having a CSD will give you more information then the frequency response simply because the FR is shoddy.
However on that assumption I DO find it very odd that Danny has posted heavily smoothed responses for his little 'test' to the pros. From anything I've gathered here, it's that usual FRs are not good enough to entirely predict the CSD - you need a completely unsmoothed, accurate, and detailed FR, without any anomalies from the measurement process, unless known about, for that.
Danny if you wish people to take you seriously on this please post a completely unsmoothed FR that is taken from distance. Whether or not this is required in your eyes, doesn't matter, it will simply help your argument.
I must say, I was surprised to see such smoothed FRs. I wont pretend to understand the mathematics behind turning an impulse response into FRs and CSDs, but it is clear you need a really good FR to do what these guys are saying. Danny don't give us a shoddy FR to work with.
One thing I would like an answer to though is Danny's original post of the comparison between the L18 and P18 from SEAS.
It was said that above 1khz the CSD will be reasonable.
Danny asked us to look at the FR of the L18 and P18 and predict the CSD below 4khz. This means the 1-4khz band should be okay for accuracy. You would expect both to be clean, yet the P18 is noticeably cleaner.
Are you saying that the small depression at 1khz and the bump at 1.5khz + the general loss of output before the resonance peak hits, is whats causing the energy storage? If so, isn't this only predictable if you are aware of the trend in stiff cones? Or does the presence of the high Q breakup associated with metal cones automatically describe this effect?
Originally posted by ucla88
I'll answer, but Danny, you're not allowed to use a CSD to prove your point. That's the whole idea.
Isnt this Danny's point though? You're supposed to guess, get it wrong, then he shows you a CSD proving that your guess work was in fact bang off
-----
Ha Ha. But this is my point. The CSD is not the gold standard. The impulse response is. Or, at least, its the most pure of the measurements. And since the FR is just the FT of the impulse, it is mathematically equivalent. The only error is a single windowing function.
The CSD on the other hand is a series of windows, many of which are in the worse possible place with respect to doing an FT. It is the CSD that is prone to error. If I look at a resolved FR and it does not agree with the CSD, it is more likely that the CSD is in error...
As far as the other points. Here's what you have to remember. Energy storage always shows up in the FR curve. And if you correct for the FR curve exactly to flat, you have equalized the energy storage away. I know this is contrary to what Danny and some others have said and implied, but it is the case. We had a very long and thoughtful, if not somewhat acrimonious debate on this on the MAD board a while back with John K, SL, Feyz, dlr, myself and a bunch of other folks. I was under the erroneous impression that energy storage was some mythical quality than couldn't be corrected for. Even SL was a believer in this. On SL's website, in the past he had a comment that linear and nonlinear distortion were inherent properties of the driver and couldn't be corrected. Sometime after our long discussion, this is now what he has on his website.
"It should be apparent by now why stored energy and non-linear distortion are such important driver parameters. Stored energy is difficult to equalize for and non-linear distortion cannot be corrected."
We've gone from an independent, fixed property of the driver to "difficult to equalize."
The way to minimize "energy storage" the way we're thinking about it is to correct as best as you can for FR irregularities. Some are really not fixable, except with very complex active eq, and even then, perhaps not. Choosing drivers that are flat in the passband using a highly resolved FR is a more accurate way of picking a driver with low energy storage issues than doing a 3ms CSD.
I'll answer, but Danny, you're not allowed to use a CSD to prove your point. That's the whole idea.
Isnt this Danny's point though? You're supposed to guess, get it wrong, then he shows you a CSD proving that your guess work was in fact bang off
-----
Ha Ha. But this is my point. The CSD is not the gold standard. The impulse response is. Or, at least, its the most pure of the measurements. And since the FR is just the FT of the impulse, it is mathematically equivalent. The only error is a single windowing function.
The CSD on the other hand is a series of windows, many of which are in the worse possible place with respect to doing an FT. It is the CSD that is prone to error. If I look at a resolved FR and it does not agree with the CSD, it is more likely that the CSD is in error...
As far as the other points. Here's what you have to remember. Energy storage always shows up in the FR curve. And if you correct for the FR curve exactly to flat, you have equalized the energy storage away. I know this is contrary to what Danny and some others have said and implied, but it is the case. We had a very long and thoughtful, if not somewhat acrimonious debate on this on the MAD board a while back with John K, SL, Feyz, dlr, myself and a bunch of other folks. I was under the erroneous impression that energy storage was some mythical quality than couldn't be corrected for. Even SL was a believer in this. On SL's website, in the past he had a comment that linear and nonlinear distortion were inherent properties of the driver and couldn't be corrected. Sometime after our long discussion, this is now what he has on his website.
"It should be apparent by now why stored energy and non-linear distortion are such important driver parameters. Stored energy is difficult to equalize for and non-linear distortion cannot be corrected."
We've gone from an independent, fixed property of the driver to "difficult to equalize."
The way to minimize "energy storage" the way we're thinking about it is to correct as best as you can for FR irregularities. Some are really not fixable, except with very complex active eq, and even then, perhaps not. Choosing drivers that are flat in the passband using a highly resolved FR is a more accurate way of picking a driver with low energy storage issues than doing a 3ms CSD.
You are correct, in that at first it would appear that ringing in a cone would still be present even if you EQ it out, however at second glance this would not so easily be the case.
After all we are dealing with energy here, it can of course not be destroyed nor created, merely changed.
So perhaps we have a driver with a bell like resonance at 1000hz with a peak of 20dB. The reason I am assuming the CSD would show a huge ridge in decay here is simply because the driver is way more efficient at that one specific frequency, so of course its decay is going to be longer then the rest, its decaying from 20dB more output, rather then the 'zero' reference, where the driver is flat. The energy having to decay from 90dB is obviously going to decay faster then at 110dB.
EQ that 20dB peak out and the energy put into the cone at that frequency is significantly less and hence it now decays at the same rate as the rest of the cone.
And as energy can obviously not be created nor destroyed, the cone cant magically create more energy, it CAN be more efficient at resonance however, but if you reduce the energy given to the cone at that frequency, we are now dealing with like amounts of sound energy being produced by the cone and hence it falls at the same rate.
I take it you're saying something like that?
After all we are dealing with energy here, it can of course not be destroyed nor created, merely changed.
So perhaps we have a driver with a bell like resonance at 1000hz with a peak of 20dB. The reason I am assuming the CSD would show a huge ridge in decay here is simply because the driver is way more efficient at that one specific frequency, so of course its decay is going to be longer then the rest, its decaying from 20dB more output, rather then the 'zero' reference, where the driver is flat. The energy having to decay from 90dB is obviously going to decay faster then at 110dB.
EQ that 20dB peak out and the energy put into the cone at that frequency is significantly less and hence it now decays at the same rate as the rest of the cone.
And as energy can obviously not be created nor destroyed, the cone cant magically create more energy, it CAN be more efficient at resonance however, but if you reduce the energy given to the cone at that frequency, we are now dealing with like amounts of sound energy being produced by the cone and hence it falls at the same rate.
I take it you're saying something like that?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.