PI active damping

[Teleno]

If allowed to oscillate at the beginning, the active damping may prove useful after all. Point is that even though flyback finishes with a slight overshoot, the damping becomes critical and the subsequent oscillations become effectively snuffed.

The small overshoot works just the same with a slightly overdamped resistor.

The circuit here also contains a snubber, and a gradient current compensation circuit.

Could you expand a bit on what gradient compensation is for? Thank you.

[Davor]

The small overshoot works just the same with a slightly overdamped resistor.
...
Could you expand a bit on what gradient compensation is for? Thank you.

I guess you thought underdamped? Anyway, a resistor of higher resistance than ideal. It does not snuff oscillations as fast as the active device, at least not in a configuration I presented. There is a little room for improvement here, as the most of the overshoot is a result of a MOSFET in active damping circuit working as an ideal diode for negative voltages. I'll see if it can be improved by a p-channel in series.

The gradient compensation circuit helps maintaining integrity of low level signals. MUR460 in series with IRF740 is a very nice device with low leakage current and low capacitance, but it is not ideal. After a flyback this diode cuts back current flow from MOSFET's capacitances, and hence they (capacitances) remain charged. Unfortunately, this charge is dispensed through leakage in diode and MOSFET, and the part observed from the coil, because it is not constant, is affecting reception of weak target signals. A constant current flow is compensated against with an EF sample, but the changing one, the gradient, not so much. Therefore, the compensation circuitry reduces a voltage at MOSFET drain to Vcoil+5V. Diode remains reverse polarised, hence present low capacitance to the coil, yet its leakage is constant, and therefore removed by EF.

Snubber reduces a flyback to a flat-top with lower voltage but longer duration and no avalanche noise. It will hopefully enable me to sample the famous X component there - just like VLFs do.

[Teleno]

Yes, I meant underdamped.

A few thoughts off the bat for brainstorming:

As I see the snubber voltage depends on both pulse frequency and pulse width. The voltage will be the same target or no target. An X compnent means the flat top lasts longer due to an increase in L. The transient slows down. The longer duration also increases the snubber voltage which in turn shortens the flat period, so you'll end up near the same place and measuring the difference becomes difficult. If you use an RX coil you can sample the voltage during the linear current ramp at the TX coil that corresponds to the flat top. An X component is detected as a higher induced voltage, a conducting target lowers the voltage.

Very interesting the explanation on current gradient. If your PI is of the "movement" type that looks at the difference inf sampled signals between periods, then the gradient will be subtracted out because it's the same from one period to the next, while the EF is not. A problem could arise if the leak is noisy or its magnitud is an order above the minimum signal you can detect causing it to be lost in common mode.

Just my two cents, no consistency with the true picture is guaranteed.

[Davor]

...A problem could arise if the leak is noisy or its magnitud is an order above the minimum signal you can detect causing it to be lost in common mode.

Both. You can speculate noise, but magnitude is well above the noise capability of even modest opamps, so yes, it is very good to get rid of it by the gradient elimination thing.

Snubber will not lose you any depth as it is integral of voltage in a flyback that counts. Within reason, of course.

The idea of detecting X component is to directly take a measure of bandpass filtered (motion) voltage at snubber capacitor, and in order for a whole contraption to work right, a coil must be charged with constant current. I already done some simulations, and it seem to work correctly. The only downside is that I can't as yet figure out the way to make it work in non-motion way, and most probably it will be overwhelmed by Tx noise so it will not be able to discriminate very small iron targets. But that's a start.

I already know how to "convince" my GB circuit to provide discrimination in a coloured metal range, so these both will get me covered. Gee, so much trouble just to make a PI work as it should.

[Qiaozhi]

[ATTACH]31274[/ATTACH]

Can you post the ".asc" file that you used to perform this simulation?
You will have to compress it as a zip file. Please also include any external models.

[Qiaozhi]

Please have a look at the attached image.
This shows two PI transmit circuits side-by-side. One is using the LTSpice inductor model with parameter passing, and the second is using an ideal inductor with externally defined parasitic components. The current waveform is the same for both examples, as measured via a zero-volt voltage source (which is a short-circuit). I have also plotted the current through the ideal inductor. Clearly the mysterious "kink" in the waveform is due to coil parasitics, and is not a bug in LTSpice.

Try plotting the current through C1, and you'll see the main contributor.

[Teleno]

Qiaohi, I already found out the lumped model and the discrete model coincide in LTspice.

My point was slightly different. The capacitance in the real coil is distributed, so there are two possibilities bugging me and I don't know which one is right:

1. The current in the distributed capacitor is circulating in the coil windings and contributes to the magnetic field.

2. The only current causing the magnetic field is the one corresponding to the idealized inductance.

In case 1, the induced voltage in the target would show the "elbow" caused by the capacitor current. In case 2 it will not.

Now, when couping an ideal target to an LTspice lumped inductor, the voltage induced in the target show no "elbow", so clearly LTspice thinks option (2) is the right one. My question: Is LTspice wrong or right?

[Davor]

LTspice will do exactly what you tell it to do. There are shortcomings in modelling for sure, but barking on a wrong tree will not lead us anywhere.
You may observe a distributed capacitance as a finite number of inductive coupled coils, say, each turn makes such coil. Now put a capacitor in parallel with each such coil, and what you get is a nice representation of a real coil. Because we are focused on the lowest of the resonances, higher order resonances are of little concern. Therefore, a coil in parallel with a coil that makes a tank in resonance that is equal to a measured one is precisely what we need. Increasing complexity of a model will not take us anywhere.

[Teleno]

Well I suppose nobody knows whether it's 1 or 2. Never mind.

[Davor]

It is 1 + all other capacitances, MOSFET being the largest of them all (~1nF++)

[Qiaozhi]

Qiaohi, I already found out the lumped model and the discrete model coincide in LTspice.

My point was slightly different. The capacitance in the real coil is distributed, so there are two possibilities bugging me and I don't know which one is right:

1. The current in the distributed capacitor is circulating in the coil windings and contributes to the magnetic field.

2. The only current causing the magnetic field is the one corresponding to the idealized inductance.

In case 1, the induced voltage in the target would show the "elbow" caused by the capacitor current. In case 2 it will not.

Now, when couping an ideal target to an LTspice lumped inductor, the voltage induced in the target show no "elbow", so clearly LTspice thinks option (2) is the right one. My question: Is LTspice wrong or right?

OK - I understand now what you're getting at.

The answer to your multiple choice question is #2.

Non-ideal inductors require the addition of extra components, such as a series resistance and parallel capacitance. If even more accuracy is required, you can also model the characteristics of the magnetic core using a second-order polynomial, and magnetic core losses with a parallel resistance. However, for PI metal detector purposes, a magnetic core is only present when a metal target gets close to the coil.

You can see what's causing the non-monotonic behaviour in the negative slope, by monitoring the current in the parallel capacitor. The "elbow" is due to resonance produced by the parallel combination of the inductor and its self-capacitance. If you then add I(L) to I(C) in the plot window, the result is the same as I(R). A lumped model is perfectly satisfactory for this application, and distributed parasitics are only required when you get into much higher frequencies where a transmission line model will be necessary.

So, to recap, the current measured in the series resistance is indicative of the behaviour of the real inductor, whereas the current in the ideal inductor is indicative of what's happening to the magnetic field. This is why the loosely-coupled target model does not display the same non-monotonicity in the current waveform. Basically, it's not a problem, and LTSpice is correct.

[Teleno]

Thanks for taking the trouble, Qiaozi. You understood my conundrum perfectly.

[Teleno]

The following is a PI experiment (breadboard).

Coil: 13uH, 15pF, 0.8Ohm
Pulse width: 15us
Vdd. 9V
Preamplifier: single-stage, transconductance, 3K3Ohm load, schottky transistor.
Damping: MOSFET with constant Vgs.
Oscilloscope: 2V/div, 5us/div.
Yellow trace: coil voltage.
Green trace: amplifier output.


I've found the trick to make active damping work is a trasconductance amplifier. Notice one single stage can amplify well into the volts range with very low noise.

[mickstv]

Teleno, just wondering why you use such a low coil inductance, the norm is usually around 300uh ?

[Teleno]

My goal is to detect small gold particles at a shallow depth (gold inside rocks or right on the bedrock). Kind of what the Falcon MD20 does but using PI instead.

The time constants are very short and the only way out is to use low inductance and high current. Active damping allows for earlier sampling. Since tau's are short pulses can also be short, which keeps consumption at bay.