I would like to measure control loop response (loop gain and phase) of a DC to DC switching regulator. Various resources online suggest using a Vector Network Analyser together with an injection transformer. VNAs are very expensive and I am not sure if a full blown VNA is really necessary. Assuming that I already have a suitable injection transformer, could someone recommend alternative solutions under $1200?
The new Siglent 4ch low cost SDS1x04X-E scope series might do it too if you have/buy a suitable sig-gen (in a similar way to Keysight, scope has bode plot function built in). If this is a one-off then maybe you could do it manually, measuring with scope FFT + sig-gen at a range of frequencies. Bandwidth requirement won't be that large, a few MHz at the very most. The thread about Dave's Bode 100 teardown video also has info about a bunch of DIY and low-cost isolation transformers (short answer - wind one yourself or find a video isolation transformer).
If you don't need that fancy frequency / phase response plot, I'd suggest testing the DC/DC by applying a step load change.
Nothing more required than a homebrew step load generator and a scope. Look up Jim Willams' app notes regarding this technique.
If you don't need that fancy frequency / phase response plot, I'd suggest testing the DC/DC by applying a step load change.
Nothing more required than a homebrew step load generator and a scope. Look up Jim Willams' app notes regarding this technique.
There really isn't much to look up. All you need is an additional load resistor, a transistor to switch it and and out of circuit, and something to drive the transistor. You might get away with just a resistor and a switch, but switch bounce might spoil the results with that setup. You don't need anything finely calibrated. You only need something to abruptly increase or decrease the load somewhat, and a scope to capture the effect on the power rail.
If you don't need to measure much beyond the audio band and you don't mind figuring out probing, etc, you can get very nice VNA measurement capability from a USB sound card and some software. I find that ARTA (
http://www.artalabs.hr/) gives some really powerful analysis capability. It's especially nice because sometimes a swept sine input to a converter breaks things, but a PRNS does not.
John
capture the effect on the power rail.
Is it possible to estimated loop bandwidth and phase/gain margin from this?
If you don't need that fancy frequency / phase response plot, I'd suggest testing the DC/DC by applying a step load change.
Nothing more required than a homebrew step load generator and a scope. Look up Jim Willams' app notes regarding this technique.
There really isn't much to look up. All you need is an additional load resistor, a transistor to switch it and and out of circuit, and something to drive the transistor. You might get away with just a resistor and a switch, but switch bounce might spoil the results with that setup. You don't need anything finely calibrated. You only need something to abruptly increase or decrease the load somewhat, and a scope to capture the effect on the power rail.
Noob question, will something like this ->
Dynamic Electronic Load Project is enough ?
I mean, especially for "low cost".
capture the effect on the power rail.
Is it possible to estimated loop bandwidth and phase/gain margin from this?
Why do you want to know the frequency and phase response? So I can work out the impulse response. Why do you want to know the impulse response? So I can work out the frequency and phase response.
Its one of those "Is the Fourier transform half empty or half full" kinda things.
capture the effect on the power rail.
Is it possible to estimated loop bandwidth and phase/gain margin from this?
Why do you want to know the frequency and phase response? So I can work out the impulse response. Why do you want to know the impulse response? So I can work out the frequency and phase response.
Its one of those "Is the Fourier transform half empty or half full" kinda things.
I'm designing a PSU and I want to know if it is fast and stable and how different components affect its performance.
This can be done very cheaply:
1. Read these two app notes:
- http://www.ti.com/lit/an/slva381b/slva381b.pdf
- https://www.richtek.com/Design%20Support/Technical%20Document/AN038
2. Get a suitable load power resistor to do the load step test
- Note that if you immerse the resistor in water you can probably run 2-5x rated power for a short time
3. Hook up your DSO and run the test. The "switch" can be as simple as manually touching an alligator clip/wire to the PS output terminal.
Thanks for the links. As for switching, I just directly drive a mosfet from my signal generator (using square wave). By regulating upper and lower voltages I can simulate not just load ON / OFF scenario, but also, e.g., transient 0.1A => 1A (and back). Of course mosfet drifts a bit as it heats up and 50-ohm drive will not switch mosfet lightning fast, but it was good-enough to characterize PSUs I have. As for heating, I just keep duty cycle low, this helps a lot.
I think manually touching wires will create a lot of ringing and not repeatable. Perhaps, NE555+mosfet will be a good and cheap alternative.
MOSFET + siggen is definitely a more elegant solution. However, manual alligator switching is an empirically verified method.
20Support/Technical%20Document/AN038]https://www.richtek.com/Design%20Support/Technical%20Document/AN038[/url]
2. Get a suitable load power resistor to do the load step test
- Note that if you immerse the resistor in water you can probably run 2-5x rated power for a short time
You're only interested in the step response, so load time can be very short and with a low duty cycle. Unless the power level is such that there's a risk of the resistor rapidly combusting or exploding ( yes, this happens, guess how I know), then average dissipation in the resistor can be negligible
I think it was Bob Pease who phrased step-load testing as "just bang on the output and see what happens"
While it might be hard to get a clean step response using an alligator clip, it is not a bad test for robustness. I once saw a test setup in a lab that designed very reliable converters. It was a giant "computer grade" electrolytic cap with a big knife switch. Good for testing transient robustness of inputs and outputs
John
While it might be hard to get a clean step response using an alligator clip, it is not a bad test for robustness. I once saw a test setup in a lab that designed very reliable converters. It was a giant "computer grade" electrolytic cap with a big knife switch. Good for testing transient robustness of inputs and outputs
John
It can be good for robustness testing, as it will catch things like overvoltage spikes due to inductive ringing. This on the input of a small DC/DC with only ceramic input caps can kill a converter running close to its max input voltage.
Been there, Done that , added a zener
added a zener
Did you connect the zener between the output terminals or accross the input capacitor? (I think it should be on output).
added a zener
Did you connect the zener between the output terminals or accross the input capacitor? (I think it should be on output).
Input - I had a situation where a 32V max input converter IC would literally explode if you connected a 24V supply via direct contact contact ( as opposed to turning the PSU on). Scope showed input spike up to 40V from combination of cable inductance and ceramic input filter cap.
capture the effect on the power rail.
Is it possible to estimated loop bandwidth and phase/gain margin from this?
Yes, the load step response works and is also the easiest but you need an oscilloscope which can differentiate the response and then return magnitude *and* phase response from the FFT which the mentioned Keysight and Siglent oscilloscopes won't do. I do not know of any current production oscilloscopes that can do this for less than the cost of a new car; since it is just firemware, I suspect market segmentation is involved.
This article from EDN covers the three ways (swept, noise, impulse) that oscilloscopes can make this measurement:
https://www.edn.com/design/test-and-measurement/4441000/1/Measure-frequency-response-on-an-oscilloscopeThe load step response is not quite the same test but has the advantage of measuring large signal response which may be more relevant than the small signal response.
I'm designing a PSU and I want to know if it is fast and stable and how different components affect its performance.
The load response test above will do exactly what you want but like I said, no current DSOs even close to your price range will work.
Two other options for DSO based VNAs include Cleverscope which it designed exactly for this application but is just outside your price range and Syscomp Design which is well within your price range:
https://cleverscope.com/https://www.syscompdesign.com/products/
added a zener
Did you connect the zener between the output terminals or accross the input capacitor? (I think it should be on output).
Input - I had a situation where a 32V max input converter IC would literally explode if you connected a 24V supply via direct contact contact ( as opposed to turning the PSU on). Scope showed input spike up to 40V from combination of cable inductance and ceramic input filter cap.
Linear tech (as usual) has an informative app note about this:
http://www.analog.com/media/en/technical-documentation/application-notes/an88f.pdfHave definitely seen this as well, can also be solved with a series resistor on the input (before capacitors) if you can eat a little loss. Bonus points when someone goes through a stack of boards "trying to find a good one", managing to blow them all up in this manner (board was designed to go into something with a PSU directly upstream and no switch).
Yes, the load step response works and is also the easiest but you need an oscilloscope which can differentiate the response and then return magnitude *and* phase response from the FFT which the mentioned Keysight and Siglent oscilloscopes won't do.
I'll try to export raw data from the scope and do analysis in Python.
Any 4-channel oscilloscope should be able to do it.
Put a current shunt in series with the load and use two channels to look at the difference across it (i.e. the current reaching the load).
Use the other two channels to look at the input and output voltages.
It won't produce pretty graphs automatically but grabbing the data from the scope over USB/Ethernet is usually easy and you can make graphs using excel (or whatever).
Any 4-channel oscilloscope should be able to do it.
Put a current shunt in series with the load and use two channels to look at the difference across it (i.e. the current reaching the load).
That is not going to work because you'll be looking at milli-Volts on top of several Volts (or more). The resolution of an oscilloscope won't be enough. A better way is to put the current shunt at the ground point with one end and use 2 channels: one for voltage and one for current.
Any 4-channel oscilloscope should be able to do it.
Put a current shunt in series with the load and use two channels to look at the difference across it (i.e. the current reaching the load).
That is not going to work because you'll be looking at milli-Volts on top of several Volts (or more). The resolution of an oscilloscope won't be enough. A better way is to put the current shunt at the ground point with one end and use 2 channels: one for voltage and one for current.
Not if you use the math function.,
But yes, putting one end of the shunt on GND is easier and saves a channel.
Any 4-channel oscilloscope should be able to do it.
Put a current shunt in series with the load and use two channels to look at the difference across it (i.e. the current reaching the load).
That is not going to work because you'll be looking at milli-Volts on top of several Volts (or more). The resolution of an oscilloscope won't be enough. A better way is to put the current shunt at the ground point with one end and use 2 channels: one for voltage and one for current.
Connecting one end of the shunt to ground only works moderately well if you are careful to ensure the ground point for the scope is right on the shunt. The ground bounces around too much for millivolt level sensing to work unless the ground for the measurement is well placed.