Tuesday, May 12, 2009

Super Resolution

When people say they have 'super-resolved' their source-- say, by having a 0.75" beam for VLA 20 cm A config observations-- what do they mean? Did they just ask CLEAN to give them a 0.75" restoring beam? This makes me feel very nervous. What do you guys think? Is it a good idea in some cases?


Adrienne said...

There's something called "super uniform" weighting. I'm guessing this is what "super-resolved" might mean.

I ran across this yesterday randomly while looking up a good way to explain uniform vs. natural weighting to some people. In the process I discovered Dan Briggs' thesis (yes, of the robust parameter!) and his explanations/derivations of how different weighting schemes work. You can find it here in Chapter 3, but I'm not sure how much useful information there is. Apparently you can end up with a smaller beam at the expense of higher sidelobes.

It also made me a bit nervous when I first heard about it, but have never actually used it (and probably won't without a very convincing argument and a scientific reason). I guess I just don't understand how you can do better than the diffraction-limited resolution because that seems like a set physics principle.

Laura said...

Hmm...interesting. Do you know how you can actually tell IMAGR that you want super-uniform weighting? The robust parameter only gets you to uniform...

Adrienne said...

It looks like you can set "bmaj" and "bmin" to be whatever you want, but I don't think that changes the dirty beam at all. It would just change the size of point sources that have been restored in your image in a completely artificial way. (This could still be what they mean by "super resolved" though.)

You could also check the bit in the EXPLAIN file about "Data weighting options". There are weighting schemes available in IMAGR that are more complicated than ROBUST, but I've never used them.

It could also be hidden in the nebulous "IMAGRPRM".

Laura said...

So, I just talked to Eric Greisen, and he says that super-resolving through image weighting can be accomplished by increasing UVBOX-- to 1 or 2 or maybe even 3. This will give more weight to points on the outsides of the uv plane.

However, he thinks changing the size of your restoring beam is probably a more reliable way to super resolve in most cases. This can be accomplished with bmaj and bmin in IMAGR, or can be done after imaging using CCRES.

Rob Reid said...

"Superresolution" usually means deconvolving to resolve angles that are smaller than the dirty beam's FWHM. It's not quite the same as using superuniform weighting to shrink the dirty beam, but the goal is similar.

You can use a smaller restoring beam with CLEAN, and that works to some extent. If you go too far though you'll just get an image of the CLEAN components. Fundamentally the achievable resolution is SNR dependent, and the deconvolution method should take it into account, like smear fitting and Maximum Entropy do.