Rotation of Gridded Spherical Coordinates to the Same Grid

antennaist

I have a uniform grid of data in spherical coordinates. e.g. theta = 0, 1, 2, ... 180 and phi = 0, 1, 2, ... 359 which forms a 2D matrix. I wish to rotate these points around a cartesian axis (x, y, z-axis) by some angle alpha. To accomplish this I currently do the following:

1. Convert to cartesian coordinates
2. Multiply by rotation matrix
3. Convert back to spherical coordinates
4. Non-uniform interpolation over the original uniform grid

The problem with this is two-fold:

1. The non-uniform interpolation of step 4 e.g. using MATLAB's griddata function is slow. For instance a non-uniform interpolation of a 360 x 181 matrix on my machine takes about 1.8 seconds.
2. Most non-uniform interpolation functions such as MATLAB's griddata are not in spherical coordinates which means that convex hull typically does not contain theta = 0 and 180 or phi = 0 or 360. I currently get around this by seeding the non-uniform data with points outside of phi = 0,360 and theta = 0,180 which is both inaccurate and adds to the computation time.

My main problem is speed. It's simply too slow for what I need. It seems like this would be a common problem that must have some elegant solution that I have not heard of. My question is simply:

Is there a better way to do this?

mfb

This does not look uniform (as seen on the sphere) to me.
Do you need this special point density (denser at small and large theta), which depends on the choice of your coordinate system anyway?

antennaist

It is uniform in the sense that phi and theta angle increments are the same. I understand that this means they have non-uniform distance on a sphere.

If it helps, these are measurements of an antenna's radiation pattern. This uniform angle spacing is typical of many setups which measure the radiation pattern. As for your other question about whether it is necessary: I do not control the data I obtain. They are obtained by instruments for me. I could, of course, interpolate this to obtain your version of uniform but this would just require another interpolation and thus more time.

I do not care if you want to call this uniform of not, my problem still remains.