Hello Eric
Here are the current and SPL at 1m of this test. As it was a 1st attempt, I haven't pushed it really... I remember now that I was a bit surprised looking at both!
Yes exactly
Good observation. No such sources close but you reinforce the assumption of an effect of power supply. My old USB soundcard has an external switching power supply (a model of replacement because the original one died). The laptop was operating on battery. There is also the power supply of the power amp (it is a very old one I built in the early 80's... yes! inspired from Hiraga amp but in AB class...).
Christian
Christian,
Im still considering why the calculated SPL is not increasing with frequency when using either simple averagng, or my attempt at coding the Rayleigh integral directly. Any ideas?
Ive been doubting our assumption that the plate velocity is just w (omega) times the displacement. We actually want the normal component of displacement, but idk if that would make any difference.
Ive asked the Elmer people how to output velocity, as my attempts have given errors.
Also found some Fortran code for a more robust method of calcuating SPL, related to boundary element method. Would take a few days, but may be worth converting. Also the Python bempp code can probably be used to do the same job.
Im still considering why the calculated SPL is not increasing with frequency when using either simple averagng, or my attempt at coding the Rayleigh integral directly. Any ideas?
Ive been doubting our assumption that the plate velocity is just w (omega) times the displacement. We actually want the normal component of displacement, but idk if that would make any difference.
Ive asked the Elmer people how to output velocity, as my attempts have given errors.
Also found some Fortran code for a more robust method of calcuating SPL, related to boundary element method. Would take a few days, but may be worth converting. Also the Python bempp code can probably be used to do the same job.
Hi Paul,
You are in advance on my own thinking... Currently I go slowly on the FDM for orthotropic free edge plate and I have also open a side topic with the impedance measurements. The impedance being I think something interesting for all, I hope to close the assembly of a not too bad set up this week. It leads me also to the idea to read agoin some papers about the electrical impedance of DML; papers that were aside for now but I should understand now and I hope to find information that will help in modeling.
The step in front of the Rayleigh integral is for me to get the speed in any point, any frequency based on the location of the driven force. Not fully clear in my mind...
Yes the normal speed is needed. But as I suppose displacement in the other directions very limited, it shouldn't make the difference.
Christian
Hi Paul,
I tried to find elements going down in the paper I collected. About the Rayleigh integral there are few doubts. Here is an extract of
Analysis of Flat Panel Speakers Luke Humphry. This one is based on the displacement.
In the same document, there is the calculation of the displacement as a sum of the modes depending on the driving point location similar to the Putra's thesis. This is probably something I should implement in the FDM script to see where it leads.
Where I am not able to follow those papers is when the topic comes to the radiation efficiency. The coincidence frequency is introduced while it is not in the Rayleigh integral... something missing.
Christian
Christian
Yes I can't really see much wrong with our assumptions. I also have been looking for example to prove our assumptions about it. This equation you found confirms it well. And omega squared is there as expected.
I think that the Rayleigh integral should in principle show coincidence. Because another related technique using the 2D Fourier transform called 'supersonic intensity' does show it. It involves taking the transform into k-space, filtering to keep only those wavenumbers (confined to a circle about the origin) which radiate, and then taking the inverse transform. It shows clearly the edge modes and corner modes below coincidence, and would of course show coincidence itself, when all modes are inside that circle. So the required info is there in the panel velocity.
I would like to try this also with scipy, but the 2D Fourier transform requires a regular rectangular grid, not triangular. I would need to make the exciter area a square, berhaps with just one or four elements.
The simple application of the Rayleigh integral (called the 'simple source' technique becomes 'unstable' for locations near the plate, but we only need the far field, so I suppose it should be OK out there.
We need to get to a point where we can replicate the results of various papers to validate the techniques. Or use multiple techniques and see how well they agree. It may be that these results are OK. Increases toward coincidence may just occur at higher frequencies (Ive only gone to 500Hz). The steep low frequency rolloff below 100Hz is probably mostly due the the front-back cancellation due to finite baffle. The increased cancellation between adjacent peaks and troughs is probably correctly accounted for, but it is counterbalanced by the omega squared term.
PS: I said some posts ago that the number of modes is constant with frequency, but you were right. It does increase with frequency. If you plot the node count less than f divided by f, versus f, then you get a straight line.
Yes I can't really see much wrong with our assumptions. I also have been looking for example to prove our assumptions about it. This equation you found confirms it well. And omega squared is there as expected.
I think that the Rayleigh integral should in principle show coincidence. Because another related technique using the 2D Fourier transform called 'supersonic intensity' does show it. It involves taking the transform into k-space, filtering to keep only those wavenumbers (confined to a circle about the origin) which radiate, and then taking the inverse transform. It shows clearly the edge modes and corner modes below coincidence, and would of course show coincidence itself, when all modes are inside that circle. So the required info is there in the panel velocity.
I would like to try this also with scipy, but the 2D Fourier transform requires a regular rectangular grid, not triangular. I would need to make the exciter area a square, berhaps with just one or four elements.
The simple application of the Rayleigh integral (called the 'simple source' technique becomes 'unstable' for locations near the plate, but we only need the far field, so I suppose it should be OK out there.
We need to get to a point where we can replicate the results of various papers to validate the techniques. Or use multiple techniques and see how well they agree. It may be that these results are OK. Increases toward coincidence may just occur at higher frequencies (Ive only gone to 500Hz). The steep low frequency rolloff below 100Hz is probably mostly due the the front-back cancellation due to finite baffle. The increased cancellation between adjacent peaks and troughs is probably correctly accounted for, but it is counterbalanced by the omega squared term.
PS: I said some posts ago that the number of modes is constant with frequency, but you were right. It does increase with frequency. If you plot the node count less than f divided by f, versus f, then you get a straight line.
Last edited:
The most interesting thing here is that
- mode cancellation due to acoustic short-circuit leaving isolated uncancelled areas which can radiate to far field and
- Modes having one or other directional wave numbers (kx or ky) inside the radiating circle giving zones of supersonic intensity via the Fourier integral.
Hi Paul,
One question to be sure to follow what you mean. Acoustic short-circuit : is it the short-circuit between the front and rear waves ? this one is present in all our observations (mainly in in-room measurements, less in outdoor one) and not in the model which suppose a full separation of the 2 waves (infinite baffle)... Or is it to describe what happen below the coincidence frequency when the capability of emitting waves from the panel is limited?
Here you are one or even more steps ahead from me! I have seen in paper the use of the k-wave space without for now understanding it. Seems I have to go back to that!
Christian
PS
Having a square mesh for simulation even if the exciter is simplified might be a good idea in order to open to possibility of check and Python calculation.
@pway
Hello Paul,
Going into the papers I collected, there is Optimization of orthotropic distributed-mode loudspeaker using attached masses and multi-exciters. Have a look to fig 3; there is no LF roll out shown... The method used is similar what you do.
Christian
Hello Paul,
Going into the papers I collected, there is Optimization of orthotropic distributed-mode loudspeaker using attached masses and multi-exciters. Have a look to fig 3; there is no LF roll out shown... The method used is similar what you do.
Christian
Acoustic short circuit explains both of those, but of course the Rayleigh integral does not consider the front to back part, as it assumes an infinite baffle. To account for front-back we would have to use either BEM, or an empirical correction factor which I think was given in one of the Thompson papers.
Acoustic short circuit is just where there are zones of opposite pressure (like a potential difference) but the frequency of oscillation is too slow, or the distance between them is too small, so that instead of producing sound waves, the air simply moves or sloshes about (like a current in response to a potential difference, hence the term). Of course to quantify that you need the math, but that is the intuition. Reality is also more complicated because of the way multiple sources add to give a resultant far field SPL. Our intuition is often a poor guide, and is a bit like the story of the three blind men and the elephant
Hi Guys - I flick visited the most recent pages of this Mind Muddling Mix of Maths and Masochism (just kidding
) and noticed this bit above. Does this mean you are ignoring a non-spine mounted exciter case (ie inertial magnet mass) which is used in many circumstances? You would need to add a magnet mass to the spider support which would have both a movement and delay effect on the panel And if I can ask a more global question - just exactly what do you hope is the practical end game of this research/analysis work.
AND ...no sign of chdsl and his patents here! (LOL)
Cheers
Eucy
Hi Christian
Ive not seen that paper.
It interests me because they are considering the mode spacing with a view to reduce the degeneracy and variation in mode spacing! I thought that was my idea!
Their measure of variation in mode spacing is interesting, I used standard deviation of mode spacing but they use mean^2/variance, which they say reduces the SD of the SPL, which would be a better objective function.
There does not seem to be any increase in response with frequency as you say, and they go to a higher frequency. I suspect its a deficiency of the simple source technique.
I find it a little strange that they are focussing on mode distribution when adding weights. On the one hand, adding weights at asymmetric locations should tend to remove degeneracy and even out mode spacing. But in my mind, weights are mainly about removing specific problematic resonances - the weight has to go to the location which is optimal for that, and you just need to accept whatever effect on the modal distribution that produces. The main technique for getting a smoother modal spacing in my view, is to use an asymmetric panel shape to begin with.
Ive not seen that paper.
It interests me because they are considering the mode spacing with a view to reduce the degeneracy and variation in mode spacing! I thought that was my idea!
Their measure of variation in mode spacing is interesting, I used standard deviation of mode spacing but they use mean^2/variance, which they say reduces the SD of the SPL, which would be a better objective function.
There does not seem to be any increase in response with frequency as you say, and they go to a higher frequency. I suspect its a deficiency of the simple source technique.
I find it a little strange that they are focussing on mode distribution when adding weights. On the one hand, adding weights at asymmetric locations should tend to remove degeneracy and even out mode spacing. But in my mind, weights are mainly about removing specific problematic resonances - the weight has to go to the location which is optimal for that, and you just need to accept whatever effect on the modal distribution that produces. The main technique for getting a smoother modal spacing in my view, is to use an asymmetric panel shape to begin with.
It's about developing a usable model of panel speaker response, so that different designs can be compared, so that any one design can be optimised, so that they can be designed from first principles rather than impossibly long and difficult trial and error.
On the other forum, what do you have, really, after so many years and so many thousands of posts?? Not a whole lot. You have folklore, anecdote, lots of pompousness and ego, rumour, myth, and failure to agree even upon the meaning of basic terms. You have some characters whose only purpose is to dominate discussion, or thwart rational discussion it if they cannot dominate. Now in my view, THAT is the real masochism. It's like flogging a dead horse.
So I would counter that the end game here is very much practical and focussed upon the real world. It's just that the real world is, in fact, complicated - you can either ignore it, or you can try to understand it. Can you say the same of the main forum?
"Examples ... show how difficult it often is for an experimenter to interpret his results without the aid of mathematics."
— Sir John William Strutt, Lord Rayleigh
That's more complicated to model, and since the spine-mounted case is better anyway, why would you bother?
Hmm.. very definite there...They're certainly different but do we have a conclusion based on results that a spine mount is superior.? People claim more bass response with a spine, but I haven't found that to be the case, at least on the cedar panel on which I tried both approaches. At certain low(ish) frequencies, isn't it possible/likely that the inertial magnet adds effective mass (and some damping) to the panel which actually may improve bass response?
Oddly, with my admittedly quite heavy exciters used with suspension but without a spine, I can't see or feel any movement in the casing at any frequency...no doubt there is some but it can't be detected without some laser work.
Anyway - it is certainly a more complex model ...but it would be interesting to pursue it for completion sake ( I know you're now going to invite me to try it
)OK good, and please accept that in no way am I being critical, I'm a structural engineer so this stuff is in my blood.....it's just that I've seen more than a few cases where people disappear down rabbit holes and lose sight of the overall picture in the minutiae. I also accept that I'm a big believer in intuitive design --it's probably intuition based on solid scientific training, but I like manually stuffing around with different concepts, some of which would be very difficult to effectively model. So it also becomes a question of how flexible and extensible any resulting model will be.
As an amusing aside, I saw this rabbit hole effect in play many years ago when a government bridge design unit decided to automate standard road bridge designs by writing their own software. I was invited to watch the program in action. Plug in parameters, press a few buttons and out pop plots of the design. This was back in the days of pen plotters and when the software generated a cross-section view of the reinforced concrete bridge deck, the model showed reinforcement in end view and shear stirrups (of which there are many along the length of the deck beams, and when it was plotted, the plotter just kept going over and over the same location drawing all of the stirrups, which in the model end view, were behind one another, until it tore holes in the paper (which didn't take very long). Having taken months in development, the software was never used.
I agree that the 'other' forum is a bit of a meandering mess which has probably and unfortunately exceeded its use-by date.
Keep pushing - I look forward to the results and I won't contaminate the thread with any more left-of-centre queries
PS - I installed Elmer - I haven't any spare time at the moment but later on....
Cheers
Eucy
Last edited:
Again, it's a difference of approach. It's perfectly possible that the mass and spring constant of some particular exciter may give a flatter response via accentuating just the right frequency in some particular design. But good luck matching the random exciter with the random design. Every exciter will have a different mass and spring constant, every design will be different, and the exciter+panel will form a coupled system.
In attempting to model a complex system to achieve some result - lets say a frequency response flat and smooth, and good impulse response, it makes sense to simplify the design so that the remaining parameters give a predictable result. Unless there is some good reason to suppose that a free exciter would contribute something unavailable to a simpler design, you lose nothing by eliminating it, and gain design simplicity. But I am confident it does not contribute anything extra, because if you wish to tune response with added mass or springs, you can do that, in a more controlled fashion, without having to contend with a mass (on a spring!) over which you have no control. So why would we waste time analysing a needlessly complex system, when we have not yet reached the main goal of being able to model the simpler system??
Besides, there are other (practical!) reasons to suppose that a spine-attached exciter is better. Unbalanced forces will give a shorter exciter life, and possibly add non-linearity or noise.
And yes, if you have a burning interest to know specifically about predicting the effect of free exciters, please do go ahead and model it!
Last edited:
Ouch!! - OK - consider me told
However it's not a random exciter or a random design - they're both known upon selection, and parameters can be modelled. Your model will have to contend with a range of different/random designs otherwise you're never going to determine which is the best. The exciter is just another element to be taken into account,
Forces are also not unbalanced on a well suspended exciter, and we have no evidence that the lack of a spine will affect exciter life, or add non-linearity and noise - you appear to be falling victim to the faults you ascribe to the 'other' forum .
Anyway, enough, just make sure the simple model is also applicable to real life designs
I'll butt out entirely
Good luck
Last edited:
Certainly a risk I think we are aware of. But this endeavor is not a job, it's a hobby. So if I didn't find exploring those rabbit holes intrinsically rewarding, I wouldn't be doing it. In any case, that's the groundwork you need to do to develop any good intuition, because there are a lot of phenomena involved that are unintuitive or even counter-intuitive.
The behavior of panel speakers are at the confluence of a lot of recent developments in theory and software which makes the goal attainable for the layman. And I believe that we have made significant progress already.
Then you should join us spelunking a few rabbit holes. That would increase the chances of reaching the goal. With a critical mass of people sharing insights and results, we are more likely to achieve the goal.