zero padding and time axis. (matlab)

[JoannesPaulus]

Apparently the only information you have about the sampling frequency (needed to recreate the correct timing) is that the signal capture is 2-3ns and it is 401 samples. This corresponds to a sampling rate of 200-133GHz. Do you think this is reasonable (I am not an expert in UWB but it seems a little high)?

 

[tjuby]

Well, sort of. All I know is that the signal has 401 sample points from 1Ghz to 12 Ghz. And that this is the bandwidth of the UWB pulse emitted.

I dont really know the time capture, It's just a reasonable guess, because this data is from an experiment that is comparable to other experiments of which I have read some peer review articles on.

When it comes to UWB, im not an expert of any sort. Im a newbie on DSP in general, so it's a bit difficult to be honest.

I got a tip though, from a professor, that to recreate the signal in the time-domain, I should first use zero-padding as a sort of interpolation technique. This should give better resolution in the time domain, which is pretty common stuff. He even suggested 100GHz as a reasonable value for Fs. If not even Fn (nyquist frequency). But, this is done with the zero-padding of course, and is not the actual sampling frequency. As far as I know, you could zero-pad for a sampling frequency of infinite hz, though you'd need an infinately fast computer. I might be wrong though.

The whole experiment is that a UWB antenna illuminates (maybe not the correct word) a cylinder containing a medium. Inside this medium somewhere, an object is located. By recording the backscatter from several angles, one should be able to picture this object.

One measure was made with the object placed in the center of the cylinder. Since I have the dimensions, I have calculated that the total time for the signal path from antenna -> object -> antenna, will be 0.49 ns. This response would be contained in the late time response of the signal. The early time response will contain backscatter from the cylinder "skin" and when entering the medium.

Now if the axis was 2ns, this would place the object somewhere in the area which would contain the late time response of the signal. As it is now, with the example I gave you, the object would seem to be located very very early in the early-time response. And that can't be correct.

I don't know if I need more information, or it's just a matter of understanding the relations of number of samples in freq-domain vs time-domain, Nyquist frequency, etc. (That is, (I)DTF in general)

My problem is that I think you should be able to decide the total time window from the information I have, then, increasing the resolution by zero padding will not affect the total time after IFFT, but would only result in more points, ie. better resolution. WHen I go by this, it seems to me the only correct approach is to use T = N/Fb, where Fb is the bandwidth. But maybe I could use the fact that I know the size of the frequency bins ? Which is 27.43MHz. This is where I blank out..

 

[JoannesPaulus]

I believe the only way you can figure out the sampling rate is through the physical size of your device: you know how long it takes your signal to hit the target and you know how many samples there are between the signal start and the backscattering.

Added after 46 seconds:

Where did your first post go?