New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
WFS secondary source selection #8
Comments
IMO the cited "Only secondary sources on the surface A that can directly ’see’ the point source should contribute to the integral in order to obtain a more accurate result" is an oversimplification that should be avoided in future. I think it's correct to implement (2.114) and (2.115) from [Schultz 2016] for arbitrary SSDs, i.e. the black dots in above example, since this is what you get from math. However, I did not check this in detail->TBD. Please cite [Schultz 2016] as (Berlin changed to Rostock + release date) |
(I ran into the same problem, discussed it with Frank and he pointed me here. Some thoughts:) Let Consider the two secondary source selection strategies based on: I think the following holds: If the condition is not met, it seems that either (Sorry if that is either obvious or obviously wrong.) |
Good to see that someone else is also wondering about the selection. If we have non-convex surfaces, solution But I guess for a practical setup your are right that one should test what gives a good result. clear all;
conf = SFS_config;
X = [-6 6];
Y = [-6 1.5];
Z = 0;
xs = [0, 2, 0];
src = 'ps';
conf.xref = [0, -1, 0];
f = 700;
% Create convex + concave shaped secondary sources
conf.secondary_sources.number = 20; % number of one circle
conf.secondary_sources.size = 2;
x0_circle = secondary_source_positions(conf);
if isodd(conf.secondary_sources.number)
error('The number of secondary sources on the circle has to be even.');
end
% Split circle in two parts (the upper part has 2 sources more than the lower
% one)
x0_upper = x0_circle(1:conf.secondary_sources.number/2+1,:);
x0_lower = x0_circle(conf.secondary_sources.number/2+2:end,:);
% Reverse direction of lower secondary sources
x0_lower(:,4:6) = -1 * x0_lower(:,4:6);
% Put all together
x0(:,1:3) = [
x0_lower(:,1:3) - repmat([conf.secondary_sources.size 0 0], ...
[size(x0_lower,1) 1])
x0_upper(end:-1:1,1:3)
x0_lower(:,1:3) + repmat([conf.secondary_sources.size 0 0], ...
[size(x0_lower,1) 1])
];
x0(:,4:7) = [
x0_lower(:,4:7)
x0_upper(end:-1:1,4:7)
x0_lower(:,4:7)
];
x0 = secondary_source_selection(x0,xs,src);
conf.secondary_sources.geometry = 'custom';
conf.secondary_sources.x0 = x0;
conf.plot.usenormalisation = false;
[P,~,~,~,x0] = sound_field_mono_wfs(X,Y,Z,xs,src,f,conf);
plot_sound_field(10*P,X,Y,Z,x0,conf);
[P,~,~,~,x0] = sound_field_mono_wfs(X,-4,Z,xs,src,f,conf);
conf.plot.usedb = true;
plot_sound_field(10*P,X,-4,Z,x0,conf); clear all;
conf = SFS_config;
X = [-6 6];
Y = [-6 1.5];
Z = 0;
xs = [0, 2, 0];
src = 'ps';
conf.xref = [0, -1, 0];
f = 700;
% Create convex + concave shaped secondary sources
conf.secondary_sources.number = 20; % number of one circle
conf.secondary_sources.size = 2;
x0_circle = secondary_source_positions(conf);
% Split circle in two parts (the upper part has 2 sources more than the lower
% one)
x0_upper = x0_circle(1:conf.secondary_sources.number/2+1,:);
x0_lower = x0_circle(conf.secondary_sources.number/2+2:end,:);
% Reverse direction of lower secondary sources
x0_lower(:,4:6) = -1 * x0_lower(:,4:6);
% Put all together
x0(:,1:3) = [
x0_lower(:,1:3) - repmat([conf.secondary_sources.size 0 0], ...
[size(x0_lower,1) 1])
x0_upper(end:-1:1,1:3)
x0_lower(:,1:3) + repmat([conf.secondary_sources.size 0 0], ...
[size(x0_lower,1) 1])
];
x0(:,4:7) = [
x0_lower(:,4:7)
x0_upper(end:-1:1,4:7)
x0_lower(:,4:7)
];
x0 = secondary_source_selection(x0,xs,src);
% Remove secondary sources that are not illuminated by source
% see: https://github.com/sfstoolbox/sfs-documentation/issues/8
x0 = [x0(1:5,:); x0(9:15,:); x0(19:23,:)];
conf.secondary_sources.geometry = 'custom';
conf.secondary_sources.x0 = x0;
conf.plot.usenormalisation = false;
[P,~,~,~,x0] = sound_field_mono_wfs(X,Y,Z,xs,src,f,conf);
plot_sound_field(10*P,X,Y,Z,x0,conf);
[P,~,~,~,x0] = sound_field_mono_wfs(X,-4,Z,xs,src,f,conf);
conf.plot.usedb = true;
plot_sound_field(10*P,X,-4,Z,x0,conf); The results are a little bit interfered by the fact that I'm using only 20 secondary source, but I did the selection of the secondary sources for the second case graphical and was hence not able to use a higher number. Do you have an idea how we could come up with a formula for the second case (which corresponds to |
Assuming that
Then, for a non-convex volume V neither of the two approaches seems to work.
It seems that visibility of secondary sources from the listener is crucial for non-convex V to avoid back-propagation. (In addition to one of the other two selection criteria.) Suppose we want to avoid back-propagation for all listener positions in some region R contained in V. This seems to be possible iff the convex hull of R is contained in V. In which case it is possible for all points in the convex hull. Caveat: All this may be horribly wrong. |
Mh, another good point. For me it was obvious that the listening area can only start for values of y < -1, but you are right we have no mathematical definition of what the listening area is besides V, which obviously is not correct in this case. |
For the case above (finite, open, non-convex array) I'm not not even sure about the definition of V. I think with a bit of work, one could come up with a mathematical formulation of a "desirable" selection strategy for non-convex SSDs. On the other hand, I suspect that it could not be implemented without major changes. (An example: For closed arrays: Does Maybe we should stick with the current behaviour:
|
OK, for the implementation I fully agree and would also propose to simply stay with the current solution. For the documentation, should we then do the following?
For the last point I'm also not sure, if we have to state that "visibility" is the correct criterion in this case (as for example stated in the Lax and Feshbach paper) or if we only say "it might be better" to switch the criterion? |
For the non-convex case, personally I would be very reluctant to give any recommendation, at least for the moment. Some more nitpicking: To my understanding, the argument in the Lax and Feshbach paper does not really apply for us. They consider far-field radiation of a finite-extent source:
The second point is in contrast to our case: I think a useful criterion for the non-convex case depends on
|
The source selection for WFS is described as "Only secondary sources on the surface A that can directly ’see’ the point source should contribute to the integral in order to obtain a more accurate
result" (compare Herrin 2003 or page 27 in Schultz 2016).
I'm not sure if the current implementation of the source selection is correct as it does not consider shadow effects of other secondary sources. Here is an example of a difficult case.
The filled black circles show the loudspeakers that were selected by the secondary source selection for a point source, the empty circles show the non-active loudspeakers. But as I understand the visible elements, only the loudspeakers highlighted by the red area should be selected.
References
Herrin, F. Martinus, T. W. Wu, and A. F. Seybert, “A New Look at the High Frequency Boundary Element and Rayleigh Integral Approximations,” SAE Technical Paper, doi:10.4271/2003-01-1451 (2003).
Schultz, F., Sound Field Synthesis for Line Source Array Applications in Large-Scale Sound Reinforcement, Ph.D. dissertation, TU Berlin, Berlin, Germany, 2016 (submitted).
The text was updated successfully, but these errors were encountered: