Analysing and changing a resonant circuit

[F._Schubert]

I'm fighting with a setup that is prone to unwanted oszillation. The reality is a bit more complex, but I think I can simplify the circuit to a simple resonant circuit. I have an electronic load which seems to have an induction coil in the output (L), a current measurement resistor (R) and my DUT which can come in a wide range of capacitance (C).

At first some general questions about the analysis with LV. I am using 'Real FFT PtbyPt':

* Is it the right choice or should I use 'Complex FFT PtbyPt' or some other?

* I read that the FFT needs a potence of 2 samples, currently I did use sample length of 1000, should I use 1024 instead?

* How to obtain/calculate the frequencies?

More important to me is to change the resonance frequency to a higher frequency. My feedback circuity which amplifies the resonance to oscillation can be switched from fast to slow. For some DUTs the resonance frequency is high enough so I won't see it in slow mode (which is good enough for the application), but I can see it in fast mode.

I won't be able to change C and L in the circuity, and are limited for the values of R. But I can make some changes like putting a Resistor or a Capacitor in the output of my electronic load. In the past others were able to eliminated similar issues by doing this with trial-and-error to adjust the value for R or C. These tricks show some effect, but not as desired. So anyone with an idea on this?

Felix 

[johnsold]

Felix,

Getting rid of unwanted oscillations is often quite a challenge!  It seems that amplifiers want to oscillate and oscillators don't.

The Complex FFT PtbyPt VI is for use when the input data X is complex.  Unless you are measuring both real and imaginary values (sometimes called quadrature or I and Q), stick with the Real FFT VI.

FFT algorithms are more efficient with integer power of 2 sample lengths but recent improvements make the performance differences small.  Note that the help specifies that the default length is 100 which is not 2^n.

The standard FFT VI help  indicates that the frequency interval df = fs/N where fs is the sampling frequency or 1/sampling period and N is the sample length.  So the frequency of element k of the output array is k*df.

Adding capacitance in parallel will decrease the resonant frequency.  Capacitance in series will increase it, but if you need a DC path to the load you cannot use that.  Resistors will damp the oscillation but will not change the frequency.

An oscillation requires feedback.  Can you identify the feedback path?  Perhaps modification of the path can be doone in such a way as to keep the system stable.

Lynn 

[Henrik_Volkers]

It seems that amplifiers want to oscillate and oscillators don't.    😄 😄  Good one Johnsold

@Felix

 How I understand your explanation: You are testing a power supply and want to optimize the feedback. Unfortunately your (electronic) ballast include an L which lead together with your FB circuit to oscillation.

First thought: In one of the app notes Jim Williams from linear.com has written, there is a appendix about feedback circuitry....

Is it possible to add a constant load in parallel to your load?

Have you tried a filter ? Sound bad in the first view to add a ring core with a bifilar winding , but perhaps your electronic ballast don't like spikes out of your supply 😉  

Great task for a good old fast analog scope 🙂

Message Edited by Henrik Volkers on 04-29-2010 03:58 PM

[F._Schubert]

Thanks Lynn,

a lot of your answers assure me, that my own thinking about the FFT was correct.

The feedback is the sense input of the load and coming from the UUT. Everything fine without sensing, but my experiments have shown that I need it.

From the theory of an OpAmp, the (huge) capacitance of the UUT on the input sets the frequency of the noise gain to a low frequency. If I remember correctly, the noise gain is amplifying the input voltage noise and leads to instability if the intersection with the OpAmps gain exceeds 40dB. There are different compensation techniques, which basically reduce the amplifiers bandwidth. But the electronic load is off-the-shelf, so I prefer not to modify it.

What I've noticed is, that using the fast mode is reducing the frequency of the oscillation. Any idea what could do such a change? If I could make it even slower the same way, the oscillations might be pushed higher the same time!

The setup will need a DC path, but I currently reduced the complexity for the analysis, so I might try a capacitor in series.

I also increased the sampling rate, so I see more of the spectrum. There is at least one more resonant frequency. I see both affected in amplitude and to a lesser extend when I change the resistor; much more effect than predicted.

Felix 

[johnsold]

Felix,

Try a low pass filter in the sense path.  Make the series DC path low impedance (an inductance) so the normal feedback at DC still occurs.  Set the cutoff frequency above the UUT/load system bandwidth if possible.  If the required feedback bandwidth includes the frequency of oscillation, then look into all-pass filters to change the phase shift but not the amplitude at the frequencies of interest.

 it is difficult to be more specific without seeing the details of the configuration.

Lynn 

[F._Schubert]

Hendrik,

You got the problem a bit wrong. Those DUT's are power supplys of their own kind and can come from some crazy scientists. They commonly have a huge capacitance (which might vary by magnitudes). This leads to effects like oscillation when they are connected to a load. This is a known issue, and we have solved it for different kind of loads on different setups (and many others have given up this task). Because this is a problem in my industry, I need to be very careful about which kind of information of the system I post on an public forum (tons of NDAs and our own IP). But nevertheless the discussion here did give me a good lead.

So to continue the technical debate:

Is it possible to add a constant load in parallel to your load? 

For current testing purposeses, there is no DC current. I even get the oscillation when there is no current at all (so not even AC). In the final application, there will be an AC and a DC current generated. Ideally the load will handle both. I am able to control the DC and AC generation independently. Any ideas that I should look for some issues on this?

Have you tried a filter ?

Please explain your idea in more detail. I even get the oscillation when the DUT is 'off', so only producing noise (no spikes!). But I'm trying to impose some filter behaviour tomorrow, so your answer might hint me in a good direction. You can write it down in germanif it's more convinient (I had to look up 'bifilar').

Great task for a good old fast analog scope

I've got a NI low-cost DAQ card attached. At the moment I get 10kHz (5 kHz BW) without any changes to the real setup (including updating the graph at 250 Hz  ). I'm always amazed about the power you have with these devices and a good skill of LV (it's 5 min to implement an FFT or the like). And even if I get aliasing, I still can tell that I have a signal above nyquist frequency.

Felix

[johnsold]

Felix,

So you get the oscillation on the electronic load without the DUT being "on" although the DUT is connected?  Is that correct?  Can you describe the frequency or range of frequencies and the types of waveforms you see when it is oscillating?  How does the oscillation very with load current?

Lynn

[F._Schubert]

Lynn,

how would I make a LP filter in the sense?

Between sense+ and GND I have my DUT. The resistance of the DUTs I have for testing at the moment are 1.4 kOhm and 6 Ohm. There are devices in the range of 10 kOhm possible, but the next developement step will require a ohmic resistance (well, it isn't a resistor but just the linear approximation at the working point) of less than 1 Ohm. And once I am there, they will ask for configurations with IV characteristic on the range of mOhms (say the limit is the cabeling resistance). I also did place 400 uF (not small) between bothand it didn't help.

Between sense+ and Force+ (that's the measurement resistor) -> my task for tomorrow. The 400 mF mentioned above did remove the oscillation, but sadly also my signal. I'll have a bunch of different capacitors for testing tomorrow.

All-pass filters: I've ignored these things until now. But might be the step if I fail tomorrow. Can you recommend some further reading on them or write a short introduction yourself. I haven't looked at the phase of my FFT yet. Can I get valid information from the Real FFT or do I need a complex FFT using the current output at the resistor  (I also can apply my measurement AC signal and check any phase difference with a lockin amplifier).

Felix

[Ben]

You guys are way beyonf me but how about a series LC circuit in parallel with the load where LC and are chosen to create a notch filter at the resonant freq to effectively shunt the feed back to ground. It will look like an open to everything except the resonant freq.

Now I step bak to learn more.

Ben

[F._Schubert]

Lynn,

yes, I get the oscillation even if the DUT is 'off'. It is the voltage input noise + the capacitance of the DUT that get amplified and lead to the oscillation. I've seen frequnecies at 2kHz and above (I can't quantify above 5 kHz for now). Mostly they look like sinus with huge distortion (in the FFT I get them as peaks with a high level in the points next to the peak). I've also seen a (german) 'Schwebung' (dict says 'beats') when you have an interference of two sinoidal signals. Even more strangely, I have one configuration where the voltage signal is sinoidal, but the current signal seems a bit like saw-tooth with an severe offset (the offset must be current coming out of the sense input!).

I havn't done any tests with load variation. I can do it, but the basic configuration of the DUT is V(DUT)=0V. (yes, this is a bit nasty. Normally nobody cares about 0V conditions, and it is common for loads to behave like this if you want to measure at 0V, but for me it is standard test condition)

Felix