The problem with my code is that it is all in FORTRAN. FORTRAN does not play well with modern languages and it took me quite awhile to work out how to do this. We never did work out the issues of using it in Python. I am not opposed to releasing my code, but I don't think that it would do anyone much good. I am not a professional coder and the code is now decades old.
Better would be for me to do exactly what I have done - show the math and how it needs to be implemented. I'll answer any questions.
Does your code assume that the directivity at DC is 0dB, e.g. that the source is a monopole? Would that cause problems if one were to try and use it to take and analyze acoustic data on a dipole?
The answer to that is complicated.
The inversion of field data to modal data (it is basically the inverse problem of sound radiation from an arbitrary source) tends to become unstable at LFs because the modal coefficients go to infinity as f->0. The higher the mode the faster this happens. These days with Singular Value Decomposition (SVD) one could likely handle this problem without too much difficulty, but when I started I did not use SVD. I had to find a way past this problem at LFs. So I assumed a monopole and got things to work well blending in the monopole mode with the calculations for higher frequencies. Hence, initially I could only do monopoles.
Later I tried to add in a dipole capability, but I was never happy with the results as it still seemed to go unstable at VLFs.
I'm sure that SVD could solve this issue, but I didn't go down that road very far before I stopped developing the technique.
The inversion of field data to modal data (it is basically the inverse problem of sound radiation from an arbitrary source) tends to become unstable at LFs because the modal coefficients go to infinity as f->0. The higher the mode the faster this happens. These days with Singular Value Decomposition (SVD) one could likely handle this problem without too much difficulty, but when I started I did not use SVD. I had to find a way past this problem at LFs. So I assumed a monopole and got things to work well blending in the monopole mode with the calculations for higher frequencies. Hence, initially I could only do monopoles.
Later I tried to add in a dipole capability, but I was never happy with the results as it still seemed to go unstable at VLFs.
I'm sure that SVD could solve this issue, but I didn't go down that road very far before I stopped developing the technique.
That's a brilliant paper. Certainly lays out the math in a very elegant manner. They are clearly academics.
What I did was identical if you consider only one plane. It gats a whole lot simpler - as I have posted before. Is the 3D versus 2D difference significant? It can be a problem in a poor design, but certainly adding in the 3D is no where near as important as having the 2D.
Another brilliant paper investigating an important question. We know that there can be errors from the spherical data sphere being too small, but the question is what magnitude?.
It's insane, even WITH a membership you often still have to pay for papers.Let me know which papers you want to see, but are behind that wall.
P.S.: there is more money in the scientific publishing business than in music industry
Don't really see how that is in the same spirit of the core values of science to be perfectly honest.
@aslepekis @gedlee
I have been trying to simulate the method in VituixCAD, but it seems one needs quite the vertical distance to make somewhat useful? That being said, it seems to be quite useful nevertheless, up to a certain frequency.
Which can easily be stitched again.
One other thing I was thinking about, does anyone here ever tried to put and helmholz absorber right on the floor?
(And the ceiling)
In theory this should also prevent any reflections @ tuning frequency of this HH absorber.
At 1 meter height, this probably could be even just simple pin hole absorber instead (just damping material probably won't quite do it).
I have been trying to simulate the method in VituixCAD, but it seems one needs quite the vertical distance to make somewhat useful? That being said, it seems to be quite useful nevertheless, up to a certain frequency.
Which can easily be stitched again.
One other thing I was thinking about, does anyone here ever tried to put and helmholz absorber right on the floor?
(And the ceiling)
In theory this should also prevent any reflections @ tuning frequency of this HH absorber.
At 1 meter height, this probably could be even just simple pin hole absorber instead (just damping material probably won't quite do it).
Helmholtz resonators are very ineffective because 1) they are narrow band 2) they only affect sound ray very close to the mouth of the device. They could affect a single tone at a specific angle, but that's about it. A huge array of random ones would work - that basically what is hanging on an anechoic chamber wall.
I know, but getting rid of the first floor reflection (I am not considering the ceiling atm), is already a big win in my book.
That won't give us an-echoic response, but it will provide a bigger time window = more frequency resolution.
(or less frequency smoothing, since I find freq resolution quite a confusing term to use).
I have tried a additional board on the floor with a steep(er) angle (facing the mic), but that didn't do much.
Apparently the wavelengths are to big to be effected by it I suppose?
Another idea that came across my mind (just brainstorming btw), was to use an active source (on the floor) to do something similar.
I haven't tried simulating this in VituixCAD, but I can say from empirical experience that performance improves with greater physical separation between reflective surfaces and source/receiver. However, I have seen that even a 48" elevation gets pretty good results down to 250Hz or so. Additionally, if you do two "layers" of data points spaced vertically so that the delay of reflection seen by one is half of the delay seen by the other, and the effect of the ground reflection is then greatly reduced when all the data points are summed together.
Additionally, since the best direct to reflect ratio is going to be best when the microphone is closest to the speaker, using incremental height steps for the data points should work better than incremental distance steps.
I haven't, but I am going to try a wire mesh support of open cell foam at some point. I've got a bunch of the stuff, but since it's a resistive absorber I'm thinking that its effectiveness would be improved by getting it off the floor.
It'd be an interesting experiment!
I'm probably reading this completely wrong, but the existence of MIMO room correction schemes would suggest that this could work... assuming the active cancelation wasn't corrupting the desired measurement.
I've used very large blocks of open-cell foam to try and absorb the floor reflection. It was only minimally effect at the frequencies where this reflection is an issue. It worked fairly well at HFs, but there it didn't matter.
Yeah, I'm kind of going into this assuming less effect where it matters. But I'm still interested to see what happens... that and I cut the wire a couple days ago.
Was that directly on the floor?
Because the absorption of materials improves greatly when leaving a bit of space (cavity).
This effect is even stronger with perforated board sandwiched between the layers of foam.
Which brings me back to another idea.
We are basically only interested in the direct path between the speaker and the microphone.
So instead of putting everything in a "dead room" we only have to make sure that this path is clear.
Meaning a mini an-echoic (tube like) "chamber" just between the speaker and the mic will already do quite a bit.
Also a (hyper)cardioid mic will help here, but I have never seen any were the freq resp is good enough for measurements.
Technically we can create one with multiple mics I guess?
Last edited: