k-Wave

1. bastlwastl
   

   Member
   Joined: Apr '14
   Posts: 7

   Hi,

   i'm working on a simulation of our latest experiments and facing two major problems:

   1. I try the simulate the spherical sensor we used to match and compare the experimental results. For that, i use makeSphericalSection and provide the calculated number of grid point. Unfortunately, the resulting spherical section is tending to be larger then the provided grid points and thereby, i find it impossible to get the correct size for the sensor I'll like to simulate. Additionally, the help text is confusing. It says, that i have to provide the radius of curvature and the height in grid point, but from the resulting Plots, its seem these values are the radius of the corresponding disc of the sensor and the depth. Can you clarify this for me?

   2. At the moment, i simulate a 2000*430*430 grid, and although the memory consumption in the simulation itself is 'only' about 15gb, I'm facing Matlabs Out-Of-Memory Error during the pre-simulations (grid expansion, smoothing etc.) Since I'm already working on the best cluster the university can offer for this purpose with 64GB per working node, I'm looking for possibilities to minimize the memory consumption during this pre-simulation phase. Is there a way?

   Thanks for your time and help,

   Sebastian

   Posted 9 years ago #

2. Anthony
   

   Member
   Joined: Apr '13
   Posts: 96

   Hi Sebastian

   About your first point, you may find this topic helpful :
   http://www.k-wave.org/forum/topic/sensor-mask-of-3d-time-reversal-reconstruction-for-a-spherical-sensor#post-4393

   Best regards,
   Anthony

   Posted 9 years ago #

3. bastlwastl
   

   Member
   Joined: Apr '14
   Posts: 7

   Hi Anthony,
   i found this topic before and actually it's not that helpful for my specific problem. I considered modeling the sensor with Cartesian coordinates and converting it, since on top, my grid spacing is not uniform, i.e. dx!=dy. This is also not covered in the function makeSphericalSection.

   I wanted to point out, that the help text on this specific function is misleading, which is sad because it seems to be a fast-to-use function. Therefore is was looking for enlightenment but it seems that i have to write the function for the "Cartesian" approach.

   Anyway, thanks for your answer.
   Cheers,
   Sebastian

   Posted 9 years ago #

4. bastlwastl
   

   Member
   Joined: Apr '14
   Posts: 7

   Well, after all, it turned out the problem is not that hard. I'm posting my code for the inspiration of the reader. Here a function for creating the Cartesian coordinates of a spherical formed sensor (like in my case, an Olympus V311). Use cart2grid to convert this to an binary mask in your KWave simulation. You can shift the focal point of your sensor. Keep in mind, that the kgrid to which these coordinates are applied is centered around 0 in all directions, i.e. for 10(!) grid points from -5 to 4

   
%%Script creating spherical sensors via cartessian coordinates
function[final_cart_sphere] = createCartSphericalSection(focal_length, focal_shift, sensor_diameter, sample_points)
%focal_length is the radius of the original sphere in meter.
%focal_shift might be a needed final shift of the focal point in meter in all directions in this form: [x_shift, y_shift, z_shift]

   %sensor_diameter defines the maximal opening angle of the spherical section
   sensor_half = sensor_diameter/2;
   %sample_points defines, how many point in the orignial sphere are sampled, in my case 20000 is a good starting value;

   %create sphere in cartesian coordinates around zero
   cart_sphere = makeCartSphere(focal_length,sample_points, [0,0,0]);

   %convert to spherical coordinates for better modeling
   [phi, theta, r] = cart2sph(cart_sphere(1,:),cart_sphere(2,:), cart_sphere(3,:));

   %get opening angle of sensor
   alpha = asin(sensor_diameter/(2*focal_length));

   %select possible phi values
   possible_phi = find(abs(phi)<=alpha);
   %select possible values for theta
   possible_theta = find(abs(theta)<=alpha);
   %intersect
   possible_values = intersect(possible_phi, possible_theta);
   final_sphere = [];
   for counter=possible_values
   theta_max = sqrt(sensor_half^2-(phi(counter)*focal_length)^2)/focal_length;
   if abs(theta(counter))<=theta_max
   final_sphere = [final_sphere; [phi(counter),theta(counter),r(counter)]];
   end
   end

   [final_x, final_y, final_z] = sph2cart(final_sphere(:,1),final_sphere(:,2), final_sphere(:,3));

   %shift x coordinate by focus point on axial direction
   %now it should focus to the bragg peak

   final_x = final_x+focal_shift(1);
   final_y = final_y+focal_shift(2);
   final_z = final_z+focal_shift(3);

   %convert to structure for K-Wave function

   final_cart_sphere = [final_x,final_y,final_z].';
end


   Posted 9 years ago #

5. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,640

   Hi Sebastian,

   The makeSphericalSection function simply calls makeSphere, and then truncates the output. The radius is the radius of curvature, i.e., the radius used to create the corresponding sphere. The height is the total "thickness" of the transducer along the beam axis. I will take a look at the documentation and perhaps add a figure illustrating the different inputs. In any case, thanks for posting your alternate code!

   Regarding out of memory issues, we are aware this is a problem. As part of the upcoming release, we have added several new functions and examples to save HDF5 files for the C++ code in parts, so that only a small number of matrices are stored in memory at the same time. This seems to solve most of the issues.

   Brad.

   Posted 9 years ago #

RSS feed for this topic

Reply

You must log in to post.