VNA Sweep Method for Loop Response

[David]

Hi all,

Apologies for perhaps a silly question but I am getting myself confused...

Lets say you have a Vector Network Analyser (VNA) and you want to measure the loop response (gain/phase to produce a bode plot) of a SMPS. When the VNA sweeps over your desired frequency range, how is this done? Does the VNA produce a continuous (chirp) sine wave or is it a stepped swept sine? My head tells me that you would need to dwell for a least a few clock cycles at each frequency before taking a measurement? If the source is continuous then is the sampling also continuous? There surely must be a settling (of the loop)/sample time to account for?

Cheers,

Dave

[Andrey_irk]

Hi all,

Apologies for perhaps a silly question but I am getting myself confused...

Lets say you have a Vector Network Analyser (VNA) and you want to measure the loop response (gain/phase to produce a bode plot) of a SMPS. When the VNA sweeps over your desired frequency range, how is this done? Does the VNA produce a continuous (chirp) sine wave or is it a stepped swept sine? My head tells me that you would need to dwell for a least a few clock cycles at each frequency before taking a measurement? If the source is continuous then is the sampling also continuous? There surely must be a settling (of the loop)/sample time to account for?

Cheers,

Dave

All VNAs which I used could change their sweeping method. And user could set a delay too (sometimes you need it to measure an electrically long DUT, for example).
How are you going to use a VNA to measure loop response of the SMPS?

[cat87]

If a SMPS loop response is what you want to analyze, then the first thing you need is to figure out how to tap into the actual response loop. Usually this is done via a transformer, designed in such a way that it can go  down in frequency to a couple of Hz. If you don't already have one on hand, maybe this might help:

https://www.eevblog.com/forum/projects/diy-injection-transformer-for-power-supply-control-loop-response-measurements/

Also, this might be of help:
https://www.eevblog.com/forum/projects/dynamic-load-bode-plot-using-hp-35665a-dsa/msg309192/#msg309192

[rfbroadband]

first let's clarify "VNA". Many RF people refer to a VNA being per default a 50 Ohm RF unit measuring s-parameters.

There are low frequency "VNA" with low source and high load impedance (often called T/R ports) that work at lower frequencies. The E5061B is one example which is a unit that offer both 50Ohm s-parameters to 3GHz as well low frequency operation to just a few Hz. It also has a dedicated low frequency port with a low/high impedance interface.

What you are referring to is often called a bode plot analyzer, which is a low freq. non 50 Ohm VNA. This is what you would use for the loop measurements. Low freq. units and "lower cost" (a few k$) are available compared to insanely expensive units like the E5061B.

The next question is where to break the loop and how do you inject the AC signal without completely breaking the feedback loop that makes the SMPS work in the first place. For that you need injection transformers, as was pointed out in the previous post.

Then you need to understand where to "break" the loop  (or more accurately where to inject the AC signal). Next you measure the reference signal that you feed into the feedback loop and the output signal coming out of the regulator, calculate the ratio and plot mag and phase for your bode plot. You need to be very careful as to how to choose the amplitude of the AC you have to inject for the loop gain measurement. Some analyzers offer you to specify a power slope vs freq. for optimum measurements. The signal needs to be large enough so your equipment can pick up the signal out of the noise floor, but not too large to disturb the normal operation of the feedback loop of the regulator.
The analyzer allows you to specify start, stop frequency and step size and takes care of keeping the measurement receiver and source aligned for accurate measurements. You can also select the measurement bandwidth of your setup and trade off measurement speed vs accuracy.

These measurements can be extremely useful to tuning the frequency response of a regulator for solid stability. The issue is that you need a low frequency VNA. If you don't have one you are probably better by doing load step measurements and tuning the transient response of your response in the time domain and "draw" conclusion on stability this way without measuring the frequency response directly.

[Jay_Diddy_B]

Hi all,

Apologies for perhaps a silly question but I am getting myself confused...

Lets say you have a Vector Network Analyser (VNA) and you want to measure the loop response (gain/phase to produce a bode plot) of a SMPS. When the VNA sweeps over your desired frequency range, how is this done? Does the VNA produce a continuous (chirp) sine wave or is it a stepped swept sine? My head tells me that you would need to dwell for a least a few clock cycles at each frequency before taking a measurement? If the source is continuous then is the sampling also continuous? There surely must be a settling (of the loop)/sample time to account for?

Cheers,

Dave

Dave,

The answer is it depends on the VNA. I use an HP3577A VNA and it has a phase continuous sweep. That is there are no step changes in the frequency. It is a sine wave that continuously increase in frequency during the sweep.

The inputs on the HP3577A have tracking filters. I normally use the 100Hz setting. This is necessary so that the VNA rejects all other frequencies than the one you are measuring.

The sweep time can be set from the front panel. For a 100Hz to 200kHz log sweep I typically use a sweep time of 2-5 seconds. You can start with a fast sweep and increase the time until the Bode plot stops changing.

You also need to play with the injection amplitude. You want a signal that is big enough to avoid signal to noise ratio problems, but small enough so that you are sure you are measuring small signal behaviour. Again I start with a small signal and increase the signal to the noise disappears. The Bode plot should be fairly smooth, no discontinuities.

The threads recommended by CAT87 have some good details.

What kind of VNA do you have? What method are using for injecting the signals?


Regards,

Jay_Diddy_B

[David]

Thank you for all the replies. It's quite an interesting topic!

Onto my next question (this one may be dumber)...say I wanted to design a piece of kit to measure the gain/phase of a loop from say 1Hz to 10MHz. How could this be realised? Many VNA architectures I've seen use product mixers and or log amps but these all appear to have minimum frequency limitations. Would it theoretically be possible to simultaneously sample the injected signal and fed back signal with an ADC and then measure the zero crossing to determine the phase and the amplitude the gain? I'm not suggesting building one, just curious to how it could be done?

[David Hess]

Would it theoretically be possible to simultaneously sample the injected signal and fed back signal with an ADC and then measure the zero crossing to determine the phase and the amplitude the gain?

When I do a manual bode plot analysis with an oscilloscope and function generator, this is essentially how I go about it.  Usually I have some idea of the network configuration so it is sufficient to measure the gain at specific points of phase like 45 degrees, 90 degrees, etc.

Any oscilloscope with a waveform generator should be able to automate this process and display a real time bode plot but none do.  Some older oscilloscopes and high end oscilloscopes can generate a real time bode plot from the impulse response.  I always get the feeling that there is an untapped market here but at best it is currently used for market segmentation.

[Someone]

Any oscilloscope with a waveform generator should be able to automate this process and display a real time bode plot but none do.  Some older oscilloscopes and high end oscilloscopes can generate a real time bode plot from the impulse response.  I always get the feeling that there is an untapped market here but at best it is currently used for market segmentation.

There is a swept bode plot on some of the Keysight scopes, but they aren't updating the older products with these new features which is a real shame (since they're on the same operating system and have the same underlying structure):
https://www.eevblog.com/forum/testgear/frequency-response-with-a-tek-md3000/msg1047553/#msg1047553

[David Hess]

Any oscilloscope with a waveform generator should be able to automate this process and display a real time bode plot but none do.  Some older oscilloscopes and high end oscilloscopes can generate a real time bode plot from the impulse response.  I always get the feeling that there is an untapped market here but at best it is currently used for market segmentation.

There is a swept bode plot on some of the Keysight scopes, but they aren't updating the older products with these new features which is a real shame (since they're on the same operating system and have the same underlying structure):
https://www.eevblog.com/forum/testgear/frequency-response-with-a-tek-md3000/msg1047553/#msg1047553

I saw that discussion and almost mentioned it.  Even on the 3000T, it is a $1500 option.

Some DSOs (many LeCroy, older Tektronix TDS and 11K oscilloscopes) can calculate it from the impulse response *if* they return phase information from their FFT and this may be a more convenient test.  Syscomp Design makes some inexpensive USB oscilloscopes which support network analysis but I do not know how well they work.

I think EDN had an article showing all three ways to do it with a DSO but I was not able to find a link.

Another way to do it is with a sweep generator and two channel oscilloscope which can make gated or windowed phase and amplitude measurements.  Some analog delayed dual timebase oscilloscopes can do this with their built in timer/counter or an external timer/counter.

[Someone]

Any oscilloscope with a waveform generator should be able to automate this process and display a real time bode plot but none do.  Some older oscilloscopes and high end oscilloscopes can generate a real time bode plot from the impulse response.  I always get the feeling that there is an untapped market here but at best it is currently used for market segmentation.

There is a swept bode plot on some of the Keysight scopes, but they aren't updating the older products with these new features which is a real shame (since they're on the same operating system and have the same underlying structure):
https://www.eevblog.com/forum/testgear/frequency-response-with-a-tek-md3000/msg1047553/#msg1047553

I saw that discussion and almost mentioned it.  Even on the 3000T, it is a $1500 option.

Some DSOs (many LeCroy, older Tektronix TDS and 11K oscilloscopes) can calculate it from the impulse response *if* they return phase information from their FFT and this may be a more convenient test.  Syscomp Design makes some inexpensive USB oscilloscopes which support network analysis but I do not know how well they work.

I think EDN had an article showing all three ways to do it with a DSO but I was not able to find a link.

Another way to do it is with a sweep generator and two channel oscilloscope which can make gated or windowed phase and amplitude measurements.  Some analog delayed dual timebase oscilloscopes can do this with their built in timer/counter or an external timer/counter.

It would be an interesting project to write some desktop software to extract the response from an impulse perturbation. Averaging many many impulses should get a relatively clean measurement without impacting the DUT and the integrated waveform generators triggers accurately enough to pull out the signal from a lot of noise.

[David Hess]

It would be an interesting project to write some desktop software to extract the response from an impulse perturbation. Averaging many many impulses should get a relatively clean measurement without impacting the DUT and the integrated waveform generators triggers accurately enough to pull out the signal from a lot of noise.

Those DSOs which I mentioned can do this by themselves.  The requirements are being able to differentiate the input and then perform an FFT.  The trigger point becomes the phase reference so an external signal generator can be used.  If you have an impulse source instead of a step, then differentiation is not required.

This EDN article covers the basics of the three methods but leaves out phase.

This old Tektronix application note describes how to get the phase information from a step or impulse test signal.

I am sure the Syscomp Design instruments that I mentioned use a swept sine measurement.

[Someone]

It would be an interesting project to write some desktop software to extract the response from an impulse perturbation. Averaging many many impulses should get a relatively clean measurement without impacting the DUT and the integrated waveform generators triggers accurately enough to pull out the signal from a lot of noise.

Those DSOs which I mentioned can do this by themselves.  The requirements are being able to differentiate the input and then perform an FFT.  The trigger point becomes the phase reference so an external signal generator can be used.  If you have an impulse source instead of a step, then differentiation is not required.

This EDN article covers the basics of the three methods but leaves out phase.

This old Tektronix application note describes how to get the phase information from a step or impulse test signal.

I am sure the Syscomp Design instruments that I mentioned use a swept sine measurement.

By adapting the front end amplifier gain the swept sine measurements get a good dynamic range, and importantly uses 2 channels to compensate for the limited drive of the waveform generator. To do a proper ratio of the two channels FFTs could be done and subtracted on a scope (the chaining of maths functions on higher end scopes for instance) but its fairly fiddly and it could be done with these cheaper scopes and desktop software, the key trade off would be how accurate it is and how quickly it can update. By averaging in the scope you can transfer the noise reduced captures less often to the computer and still keep up an interactive plot.