Curious about ZAPH's designs. Seas L18 and P18RNXP

[ucla88]

I know I won't change Danny's mind, but a few comments to everyone reading this thread are in order.

Again, windowing an impulse is a very unnatural event. There is significant error introduced by the windowing process. Due to the way the impulse is chopped up, subtle changes are just as likely due to the way the impulse is windowed as anything else.

For example. None of the CSD's offer much relevant information below 1k. Most of what you see is heavily distorted by the windowing. Making audibility judgements based on measurement like this are questionable.

Most of the CSD's behavior above 1k can be more or less predicted by the FR. Now Danny has made some subtle points about resonances that "don't show up in the FR curve." Well, if they don't show up in the FR curve, and the FR and impulse are one and the same, perhaps we should wonder if the effect is artifactual.

I have spent a lot of time looking at CSDs, SL style tonebursts, and as detailed FR curves as I can on my site. I think you'll find a careful analysis will show the majority of what you need to know about stored energy will be found in a detailed FR curve. You're better off going outside and making a 12 ms FR measurement than post processing your 3ms in room measurement and thinking you can pick out the stored energy in any detail.

Also, be aware that what "looks bad" on an FR curve or CSD is not exactly what sounds audibly offensive. Google O'Toole's whitepapers at harman and read about resonance audibility-how they measure it and study it's audibility.

Anyway, forget about what Danny and I have to say and look at the links/research it yourself.

[dlr]

Originally posted by Danny
[B]

Or so you think (in this case).

The content of the debate was indicative.

In a CSD it is not that there is a point at which everything below point A is not accurate and everything above point B is. It is a graduated curve. Obviously accuracy gets less and less as frequency decreases. Depending on the measurement set up and gated time window accuracy below a 1kHz range becomes less accurate.

Yes, but you were trying to use just such measurements and that measurement range in part to argue your position. This correctly says that your previous argument in this frequency range was based on questionable measurement accuracy and precision. This is what I would call misleading.

Below 200Hz (or even 300Hz) it cannot be observed to have any real value as accuracy below that point is not possible.

If YOU think otherwise, then that's a clear demonstration of misunderstanding the limits of the measurement.

This is a strawman argument. I made no such reference. You're arguing for the sake of arguing with this.

Clearly you and I know that both are derived from the impulse response measurement.

Yes! But it's more than that. The impulse resposne can be reproduced from the frequency response. Just do an inverse FFT.

That does not diminish the value of looking to a spectral decay to see stored energy problems. To dismiss a CSD as having no value is absurd.

The CSD is only of value to those who do not see the information in the FR. As John said, the CSD is nothing more than the FR in another form. I have no need of the CSD as I can see what I need to see in the FR.

[QUOTE]Maybe you would like to rub your crystal ball and tell us what the spectral decay will look like from looking at a frequency response curve? I could post a few for you if you'd like to decipher them for us.

I don't need to. It's EXACTLY as John said, ALL of the information is in the frequency response. If I had a desire (not need) to see the CSD I could extract it from the frequency response. Even if the measurement is a swept sine (such as LMS that does not produce an impulse, it has to be determined from the frequency response), I can convert it to an impulse, then to the CSD. Again, it's all in the frequency response. There's nothing in the CSD that is not also in the frequency response. I just no longer learn anything from the CSD, it's in the FR.

This is where you are absolutely wrong. I can take a frequency response and do an iFFT to retrieve the impulse response. From that impulse response I can then determine the CSD. That is where your argument fails.

[QOUTE]Furthermore to suggest that knowing the information shown on a CSD is of no value is even more ridiculous.

I essentially know what's in the CSD by seeing the FR. I've seen enough of them to know that there is little if anything to be gleaned from the CSD.

Dave

[Danny]

Again, windowing an impulse is a very unnatural event.

Unnutural event? Now that is funny.

There is significant error introduced by the windowing process. Due to the way the impulse is chopped up, subtle changes are just as likely due to the way the impulse is windowed as anything else.

I have measured drivers in and out of my anechoic chamber and can get identical results in this regard. It is necessary to gate or limit a time window to get meaningful information when looking at a spectral decay. Why look at a 60ms time window if the impulse response only last for 2 to 3ms?

The only way you can alter or chop up the impulse response where as to alter the data is to not catch all of it. If the impulse is 2ms long then chopping it at 1.8ms would alter it. Looking at a 4ms time window will not.

Most of the CSD's behavior above 1k can be more or less predicted by the FR.

Not true. And why try to predict it when you can look at it? That makes no sense.

Just look at the examples I posted from Zaph's measurements. There are differences in the amount of stored energy from one driver to the next where both drivers show a flat response with no breakup at all.

Now Danny has made some subtle points about resonances that "don't show up in the FR curve."

Ah, now I have made a point and it is subtle.

So if a made a subtle point about resonances that you do not see in the frequency response measurement then how do you know they are there only by looking at the frequency response? Are you kidding? Good grief!

Well, if they don't show up in the FR curve, and the FR and impulse are one and the same, perhaps we should wonder if the effect is artifactual.

Oh gee! The frequency response curve only displays maximum amplitude per frequency range. It does not display a time element.

You're better off going outside and making a 12 ms FR measurement than post processing your 3ms in room measurement and thinking you can pick out the stored energy in any detail.

Are you sure you understand this? The entire impulse response without reflections from anything might only be 2 to 3ms long with a 1 watt input. How much more dead air space do you need beyond the initial impulse response?

Dave, you are just playing semantics and missing the point.

The CSD is only of value to those who do not see the information in the FR.

Not true. See next response.

As John said, the CSD is nothing more than the FR in another form.

Yea, in a form that includes a time element that the frequency response does not include.

I have no need of the CSD as I can see what I need to see in the FR.

Good luck to you Dave. I am sure you don't even need to listen to it either. You'll know what it sounds like.

I essentially know what's in the CSD by seeing the FR. I've seen enough of them to know that there is little if anything to be gleaned from the CSD.

And from looking at these two frequency responses how do you know that from 3kHz and down one has a lot of stored energy and the other doesn't?





I know you have seen a lot of them and you probably won't need to see these, but for the sake of everyone else here are the CSD's again.

[terry j]

Well, I at least, am very inexperienced in this whole area of 'reading FR graphs', and I must say that I've learnt a lot about them in this thread, and have learnt a lot about 'reading' csd graphs as well.

I most certainly do NOT want to get into (what seems to be developing as) the unfortunate debate that is going on, but now after reading this thread (and the links within it) I am grateful for what I have learned and would like to say that at my very limited level of experience of looking at FR's that I can see a LOT of worth in seeing the results displayed in csd format.

So, they have value at least for us dummies in the field.

I must be honest though and say that the series of graphs displayed by Danny of csd's vs fr's seemed pretty convincing to me that we can, at times, glean 'extra' information form viewing the data in a different way.

Perhaps it's a little like architectural drawings, the info a builder gets from viewing the data in 'plan' can be more valuable to him than the information he gets from 'elevation', and at times vice-versa.

To complete the imperfect analogy, both plan and elevation (and of course end elevation, isometric view etc etc) are all aspects of the same measurement if you will, and all are used simply because over time they have been found to be useful!

Nonetheless, this has been a very educational thread for me, thanks. I hope it does not deteriorate because I want my education to continue.

[dlr]

It's as I said. Below 1K with an FR that is limited to 300Hz, there is no way to say with any assurance exactly what the true response is below 1K. The CSD's shown are not those of the FR's shown, since the FR is shown down to 10, but the CSDs are limited to 300Hz. The FR is probably a splice, so it's a good indicator, but not reliable for any absolute analysis. The FR above the splice point and I suspect that it is, given the low end extension, is still only as precise as the sampling length will allow. This is likely still only precise (or nearly so, not necessarily absolutely) for integer multiples of the low limit, possibly 300Hz.

Likewise, the CSD is limited in that it's resolution is 300Hz. Anything between the FR points of 300, 600, 900, etc. are representative approximations. Any resonances that might actually exist between any of those points will be missed. Those particular CSDs are not fully reliable below about 1KHz because of this. Change the window type and you'll see a change in the FR and the "resonances". Some might actually "disappear". I would have thought that you'd know all of this.

I'll let Mark respond to the points that you make to his, with one exception. If you think that an impulse taken outdoors and in such a way as to lengthen a clean impulse without reflections is still only good for 2-3 msec, you do have a bit more to learn about impulse testing. I'm surprised at your comment. Remember, it's the length of the impulse used (the windowing) that determines both the low end extension and the FR resolution from that impulse. You did know this, didn't you?

Dave

[ucla88]

Originally posted by dlr
.

I'll let Mark respond to the points that you make to his, with one exception. If you think that an impulse taken outdoors and in such a way as to lengthen a clean impulse without reflections is still only good for 2-3 msec, you do have a bit more to learn about impulse testing. I'm surprised at your comment. Remember, it's the length of the impulse used (the windowing) that determines both the low end extension and the FR resolution from that impulse. You did know this, didn't you?

Dave

What? I'm going to respond.?

I doubt I will change Danny's mind. dlr's points are well taken. What good is 2-3ms at looking at the midrange. The period of a 500Hz signal is 2ms. How much decay will there be in one or 1.5 cycles...

Anyway, as I mentioned before, those of you that want a better understanding, check out the references.

[noodle_snacks]

I would say there is no extra information in a CSD plot, however for someone inexperianced it might make things easier to interpret. If this is the case you need to be careful because as Danny has clearly evidenced with his examples, the information can easily be mis-interpreted.

[Danny]

You did know this, didn't you?

Certainly I do. It is my job. I have had to take many a far field measurement including those made in my anechoic chamber. The further you can be away without reflections the lower you can accurately measure. I was referring to the use of such measurements for a CSD. Seeing as how the range below 300Hz is of little to no value and the accuracy will still be limited even as you approach that. And then why try to get a clean 12ms time window to measure a CSD when the software won't show it anyway. These are ranges more dominated by room gain and room responses than anything else.

Also while accuracy is limited below 1kHz in a spectral decay is does give some approximation as to any problems. As one comes up from 400 to 600 to 800 and 1000Hz the SCD gets closer and closer to good reliable figures.

Danny has clearly evidenced with his examples, the information can easily be mis-interpreted.

Funny! What's even funnier are those that can interpret it having not even seen it. They can just look at the frequency response and will know all about the CSD.

[PB2]

Originally posted by Zaph
Well, for starters, the P18 didn't exist at the time I designed the L18 system. It came out about 1.5 years after that. But, even if I had to choose today, I'd probably still pick the L18 over the P18 by a teeny margin based on a few characteristics.

You're making a very common mistake about cumulative spectrum decays. But don't feel bad, it's a mistake that even industry professionals make. The CSD is the most over rated and useless form of measurement there is. There is absolutely nothing in there that can't be seen in a response curve. They both are generated from the same impulse and are simply different ways of looking at the same linear distortion information.

One might ask why I provide them on my web site. I often ask myself the same thing. The sad fact is that I think I've been brainwashed by people who think they need to see them. If I didn't provide them, there would be hoards of people saying "yeah, but he didn't provide any CSD plots". I'm guilty of posting CSD plots to avoid having to explain linear distortion to people who aren't likely to understand it. There are people who think a CSD contains info that a response curve doesn't because one has time slices and the other doesn't. Consider this question: If both the P18 and L18 were filtered with a proper crossover to shape the response curves into perfect LR4 rolloffs, and a CSD was then taken of the result, what would the differences in the CSD's of each be? The answer is that they'd be exactly the same. This is hardest thing to grasp about CSD plots.

A response curve and a CSD are just different ways of looking at linear distortion. linear distortion can and should be fixed in the crossover when possible. Most proper crossovers for metal cone woofers do have the the breakup filtered out, and thus ringing in the CSD is not a problem.

snip ....

Let me ask you Zaph, since you mention issues with some of your measurements below 800 Hz for example, would you say that your distortion measurements are reliable? Especially for the L18?

Several here are making valid points viewing these drivers as linear time invariant systems, however when one asks to compare the L18 and the P18 we have to look at the distortion performance also. Zaph you make the valid point that if notch filters are used to match the frequency response then the linear characteristics are the same. Let me point out that I think it is rare when all the resonance modes are notched, and so the FR is often more of an approximation to the target.

Let's move on to the non-linear comparision, interesting that the L18 has peaks in F3 between 2 and 3kHZ, in F5 between 1 and 2 kHZ, and more peaks between 6 and 10 kHz. Here we enter the world of non-linear analysis and where notching cannot possibly fix these distortion issues. One should also keep in mind that while these are harmonic distortion tests, they are indicative of a non-linearity, and that non-linearity will also produce intermodulation distortion.

It is important, from my view to step back and ask why a driver has various peaks and dips in the amplitude response, since if it is a perfect piston with a simple small chamber, as in a tweeter, it should follow a simple 2nd order filter function. Cone breakup, pole vent and rear chamber resonances, and reflections are most of the reasons for deviations in the FR, and often especially with cone breakup peaks are also seen in the distortion plots. The CSD tends to magnify FR issues, makes them easy to spot and also find points where there might be distortion issues. Seems the trend is that distortion is usually low, or at least well behaved when the CSD is clean. The clean CSD of the D26NC55, T25CF001, and the K1 is a quick identifier of some of the top performers. I think the CSD is helpful from this perspective.

Some might prefer the P18 from the perspective of THD performance.

Pete B.

[Dave Bullet]

Consider this question: If both the P18 and L18 were filtered with a proper crossover to shape the response curves into perfect LR4 rolloffs, and a CSD was then taken of the result, what would the differences in the CSD's of each be? The answer is that they'd be exactly the same. This is hardest thing to grasp about CSD plots.

This is the point I struggle to get my head around.... I presume this is true when looking at the CSD above the xo point (assuming we are talking about a lowpass here). However, below the xo point even if the FR curves are identical due to the implementation of identical 4th order acoustic slopes, it doesn't change the stored energy behaviour below the XO point? Unless we are saying the xo itself will act as an energy "release" (damper - excuse my struggle to find the right terms).

In other words, I need to really bend my head to understand how 2 different drivers with identical dB output at a specific frequency can end up with identical CSD information...

David.