Ultrafast Pulse Limitations

[Time]

I have a pulse I wish to measure thats very fast, relatively speaking.  Its not too fast for conventional high quality, high bandwidth probes but it is too much voltage.  The problem is that high voltage probes tend to have poor bandwidth due to the nature of their construction.

My question is: If you know your probes frequency response or bandwidth, is there a way to effectively estimate the real high frequency spectral content that gets attenuated in the measurement waveform in order to compensate and produce a more accurate representation of the actual waveform?  This would obviously be achieved analytically in post processing through some type of program written in MATLAB or something similar.

[jahonen]

You can use VNA to measure the complex frequency response of the probe. Both magnitude and phase must be considered here (thus the VNA), otherwise the phase of different frequency components are not corrected, which leads to pulse shape distortion, even if magnitude is correct.

In principle you could make same measurement using a scope, signal generator and splitter. Splitted signal would be your reference. But that is more tedious, as the scope is time domain instrument, whereas the measurement is primarily a frequency domain problem.

Regards,
Janne

[Time]

Ok, so I have a VNA and can take measurements of the probe to get the information I need.  I guess I really meant to ask how do I apply this information to an existing waveform that is in the time domain only?  Do I need a spectrum analyzer to give me the spectral data before I can make corrections?

[jahonen]

Ok, so I have a VNA and can take measurements of the probe to get the information I need.  I guess I really meant to ask how do I apply this information to an existing waveform that is in the time domain only?  Do I need a spectrum analyzer to give me the spectral data before I can make corrections?

The sampled data from the scope is enough. Just note the sampling frequency. You can make a digital filter with MATLAB etc. to realize the inverse frequency response (FFT-based frequency domain- or suitable FIR filter might be feasible approach). You might want to experiment a bit with this method before starting actual measurements, to get feeling how much frequency points do you need for the filter construction, so that correction is good.

Regards,
Janne

[Time]

One last question, is it better that I take measurements on a faster sampling scope with less analog bandwidth (~200 MHz) or is it better to use one of the multi-GHz scopes with a slower sampling?  The probe seems to be the analog bottle neck not the scope so intuition makes me lean towards more sampling.  Or maybe it doesn't matter.

Thanks for your help.  I knew Janne would be the one with the best and maybe only response.

[NiHaoMike]

How much voltage are you dealing with? A capacitive voltage divider might work. (Use metal plates in a dielectric like transformer oil or even air instead of actual capacitors.)

[jahonen]

One last question, is it better that I take measurements on a faster sampling scope with less analog bandwidth (~200 MHz) or is it better to use one of the multi-GHz scopes with a slower sampling?  The probe seems to be the analog bottle neck not the scope so intuition makes me lean towards more sampling.  Or maybe it doesn't matter.

Thanks for your help.  I knew Janne would be the one with the best and maybe only response.

Lower (but enough) bandwidth is optimum in sense that it produces less noise (more than enough analog bandwidth means just more noise). If Nyquist criteria is met, then you can always increase the "smoothness" of the curve by oversampling techniques (interp-function in MATLAB, etc.). But, OTOH, you can low-pass the signal if sample rate is higher and get rid of the noise that way.

One caveat came into mind, there might be some issues connecting the probe to the VNA since VNA is usually 50? stuff (and if the probe is made for usual 1 M? input impedance), but I think that can be worked around with some kind of matching circuitry (buffer etc).

Regards,
Janne

[Time]

10 kV and up.

I am familiar with capacitive dividers for probing high voltage but in much less temporally strenuous conditions.  How do you quantify a capacitive voltage divider probes bandwidth?

Yeah, I have already started to think about the match issue with the probe.  I didn't think it would be a setback though.  If it turns out something to be too involved, maybe phase and magnitude information is something that the company who produces the probe can provide?

[alm]

Tek has a probe for $$$ that has something like 50MHz bandwidth and is rated up to 20kV or so, derated with frequency. Not sure what your exact requirements are, I think that's the closest thing you'll find to a high voltage, high frequency probe.

Not sure how well DIY capacitors will perform at high frequencies (no idea what high is in this case, however). Ideally, you want the impedance and attenuation to be reasonably flat, depending on the output impedance. I would also be worried at the amount of current that fill flow at high voltage/high frequency. For example, a 10kV, 50MHz sine through 1pF will draw 3A (which is why even that Tek probe isn't rated for this). It is also likely to load the circuit.

When calibrating stuff with something like a VNA, keep in mind that components will behave differently at high voltages, many have a voltage coefficient, which is quite significant if you go from millivolts to kilovolts.

[Time]

I have a 75 MHz Tektronix HV probe but the rise and fall times I am dealing with are <2ns on an OK day.  It provides a good enough glimpse as to what the pulse looks like (its not super important that I have the waveform precisely) but I am really just curious to see how much pulse distortion the probe is introducing.  I am only concerned with single pulses at this point.  A company called Northstar makes some probes (http://www.highvoltageprobes.com/PDF/Vmonad.pdf) with higher voltages as well.  I am not sure if by max frequency they mean 3 dB.  I am not interested in getting a better probe though.

This seems feasible and I'd rather just learn how to do this or figure out why it can't be done, if thats the case.

Here is the probe manual: http://www.tequipment.net/pdf/tektronix/P6015A_Instructions.pdf

The specs start on pg. 40.

[drwho9437]

Are you switching a pockel cell or something? Basically you want perhaps 500 MHz and 10KV... What about a magnetic dipole antenna with a spectrum analyzer, and repeating the pulse a lot then doing IFT? Given even a reasonably nice analyzer you ought to be able to see down to -140 dBm, you have to do a very careful job of shielding to not get those amounts of power out into the room. You could use triggered sweeps if you can't have a high rep rate on the pulse. I will have to give this some thought...

[Time]

No, no pockles cell driver.  What its going to be used for is kind of irrelevant.  It can be switched up to a MHz if needed but that is it.

[drwho9437]

I was just curious. It may not be relevant to the problem solving but it may be interesting to for people to know... And it may be wise to interest people if you want suggestions. There is no reason a well matched antenna would not give you a lovely frequency domain spectra of your pulse, it would just be like Hertz's original work... with your voltages you could even spark it.

[Time]

I agree with you.  I want to do this with as little hardware as possible.

The pulse generator is meant for a lot of different things related to high energy physics and particle accelerators.  Pockles cells being one.  I am just playing with it and wanting to characterize it.  Right now it just operates into a high power 50 Ohm load.

[Time]

The reason I wanted to do this with as little hardware as possible is so I could characterize all the HV probes around the lab and create a program where you select the probe you measured with and a few other important parameters and apply the filtering alogrithm to your measured waveforms.