Is there any script or code around to perform a f-k filtering or velocity filtering ?
Ok, probably it's based on the fft2 command. But may be there is already an existing filter routine allowing to insert velocity intervals for filtering ?
hans
It's probably both a lot faster and simpler if you hack
this up yourself. It's only a matter of manipulating
the filter weights in fk domain, based on the simple
v -> k mapping I can't remember off the top of my head.
Rune
There is no such thing as a '2D filter' in matlab. Only
matrices. You are looking for something as simple as
y = real(ifft2(fft2(x).*H));
where x is your time-space data, H is your fk filter,
and y is the filtered data.
The 'hard' part is to fill in the numbers in H. You can
do that in a couple of minutes, once you have some sort
of idea what you want the filter to do.
Coming up with that idea is hard, though, as it requires
you to actually contemplate what you are up to, rather than
fiddling with ready-made software routines.
Rune
I have an idea about the "dip" in the signals I want to filter out in my data if we look at them in the non transformed data. So it's a linear relation between f and k or x and y, depending how You call the 2 dimensions.
How can I create a filter from my linear equations in the f-k domain ?
You can't. You fill in the numbers in the matrix (or N-D array, if
you have more than one spatial dimension) H I showed you a couple
of posts back. There is nothing more to it. You just have to know
what numbers to fill in. And they have nothing to do with anything
else than the job you want the filter to do.
Rune
Assuming I have a data matrix with Y X (Y=time;X=way) and I have a noise signal with Y/X = 1 contained in it, that I want to filter out.
Is there an easy way to tell, how the apllied (filter) matrix would like ? Or do You have a suggestion, where to look for ?
Thank You
hans
No. The basic principle is that any spectrum coefficient
you want gone should be multiplied with a small number,
roughly 0, while the spectrum coefficients you want to
keep should be mutliplied by a large number, around 1.
This is straight-forward idea is severely complicated by
the fact that any modification you do to the fk-domain
filter matrix has effect all over tx domain - be too hard
in one spot in fk domain will mess up everything in tx
domain. Which is why you need to know what you are doing.
Talk with your boss or supervisor for help on where to
find material suitable for your level.
Rune
Thank You for the discussion.
Hans
I think it should be possible too to generate a tool like this in Matlab .... but of cause it will be some work.
So it was much more comfortable to change data format.
hans