Originally posted by VelvetEars
Yeah, that paper's surprisingly relevant. The kind of data output I receive is more like this
http://www.rjmcompany.com/GPR-ground-probing-radar.jpg
and I'm trying to turn the smudge into a point basically.

But I like your approach, definitely give that paper a read-through.

iamatiger: what do you mean by 'reconstructing the beam as if they didn't exist'?

I meat,, if you can assume all the intervening scatterers are static, and you have enough measurements (at some diverse frequencies ideally) then you may be able to solve for their positions.

If you can time the data, you may be able to assume that the first lot of energy to come back is solely from the scatterers you want to remove. Then just use that in the solution for where the scatterers are, and use the solution to reprocess the data coming after that time.

Originally posted by sonhouse
http://www.eecs.umich.edu/~hero/Preprints/icip06_marble.pdf

I found this paper on 'see through walls' radar. Is this close to what you are doing?

Beware somewhat of algorithms proven entrirely in simulation. It is most important to a) work out the details of the waveforms you want to use, and then b) get lots of real measurements and work out how you can best grind them offline.

When you work out what is going on the may want to change the waveforms a bit, so there is some recustion implied by the above. also, be careful at all times of point solutions which are not applicable to the range of cases you will encounter in real operation.

Originally posted by iamatiger
Beware somewhat of algorithms proven entrirely in simulation. It is most important to a) work out the details of the waveforms you want to use, and then b) get lots of real measurements and work out how you can best grind them offline.

When you work out what is going on the may want to change the waveforms a bit, so there is some recustion implied by th ...[text shortened]... t solutions which are not applicable to the range of cases you will encounter in real operation.

Thanks for all your help and suggestions.

You're right about being wary of solutions that only work on simulated data, having played around with a couple of inversion algorithms that give lovely results in simulated data, the second you try it with actual data everything falls apart and you don't get any meaningful results.

The problem I'm currently facing is trying to come up with an expression that'll reverse the direction of the scatterers so they'll converge to a specific point (it's true location). Kirchkoff migration seems to be a bit too crude and the more complicated methods end up getting bogged down in complicated Fourier transforms.

Sorry if I'm being dense, but by changing the waveform are you referring to changing the signal we're sending out?