Questions on sampling

[Beenthereuk]

I'm looking to sample the amplified signal first at the 'knee' and later at a time that will allow me to determine a steady state value.
I've read that subtracting 2 such samples helps eliminate ground effects / balance the signal.
To understand this better, I've tried a simulation (attached) where a non-inverting amplifier is biased using a zener diode (maybe noisy but I dont really care at this stage) and a simple BJT switch stage is used to sample and hold the voltage at the knee. The circuit would be microcontroller driven so that both the sampled knee voltage and the steady state voltage would be acquired by ADC.




From the graph, VOUT and VSW would be ADCd by the microcontroller.

My questions are:
1. At what point is it best to sample to eliminate ground effects?
2. Would simply accumulating + averaging the samples (after subtracting them) be sufficient to make a sensitive detector?

[Davor]

There are some effects not accounted for. You need to separate ground effects from target tau and coil tau (L in parallel with damping resistor). Because coil tau is under a microsecond it vanishes fast, but not as fast so that your approach actually works. You may expect target response to overpower fast declining coil response at/below 0.1V for a common 300uH coil and few hundred volts flyback.
The ground response due to its 1/t nature is dominant at the beginning and the end. Because sampling the coil response is meaningless, you can't use samples immediately after flyback. Then comes a period where target response is dominant, and after that the ground is dominant. This is a reason why samples are delayed after flyback to 15us or so, followed by a target sample, and after that a ground sample. In case of unipolar Tx pulses you also have a third sample which is designed so that all the samples and their gains are in equilibrium, in order to remove any offsets, which is a reason you may consider this "EF" pulse to turn the Rx contraption into a chopper stabilised amplifier.

You may consider using a log amplifier prior to applying your approach, and it should work.

[Beenthereuk]

Hi Davor, OK, so using the signal graph above as a reference, what would you say the sample time for each of the three samples should be?

[Davor]

That would depend on a coil inductance, parasitic capacitances, damping, speed of ADC, and drive current, but using common values of 300uH, say 100pF, 1A before flyback ... 15us for a first sample minimum if you go for weak targets, down to 10us if you are after strong ones.

[Bill512]

There are some effects not accounted for. You need to separate ground effects from target tau and coil tau (L in parallel with damping resistor). Because coil tau is under a microsecond it vanishes fast, but not as fast so that your approach actually works. You may expect target response to overpower fast declining coil response at/below 0.1V for a common 300uH coil and few hundred volts flyback.
The ground response due to its 1/t nature is dominant at the beginning and the end. Because sampling the coil response is meaningless, you can't use samples immediately after flyback. Then comes a period where target response is dominant, and after that the ground is dominant. This is a reason why samples are delayed after flyback to 15us or so, followed by a target sample, and after that a ground sample. In case of unipolar Tx pulses you also have a third sample which is designed so that all the samples and their gains are in equilibrium, in order to remove any offsets, which is a reason you may consider this "EF" pulse to turn the Rx contraption into a chopper stabilised amplifier.
You may consider using a log amplifier prior to applying your approach, and it should work.

Hi Davor,
my opinion is that every "target", metal or ground, has in general a variation of 1/t response.
About the dominant response at the end:this is probably true for small or low conductivity targets but not necessary for the big or higher conductive targets.
For example,do you say that the ground has a longer response from a big ferrous target (iron box)?
Have you done some actual measurements on this?

[Davor]

That would mostly depend on a sort of ground you have. Not many terrains have viscous magnetic properties, and in PI those do have a very long response. You most probably have red clay or terra rossa that has lots of iron, but seldom has pronounced viscosity. If that is the case, you don't have to worry about it. All metallic targets do have 1/t response in a small portion of time. Because viscous ground response continues to have 1/t past the metal objects' time, it is the key for subtracting it and detecting on difficult grounds.

[Carl-NC]

Beenthere, your sim response looks slow. The raw flyback (red) is about 20us wide, should be closer to 5us which will end up settled at the preamp output in 10-15us.

Ground has a 1/t response, most metals have an exponential response, at least as far as we're concerned. A typical subtractive system would have a target sample at a delay of 15us with a width of 10us and a ground sample delay at, say, 30us with either a variable width or a separate variable gain path or a mixture of both. Depending on the strength of the ground and the tau of the metal, either one can be dominant at later delays, so the output of a subtractive GB circuit can be either positive or negative, depending on the GB setting and the tau of the target.

Yes, accumulating & averaging the samples increases SNR. That's what pretty much everyone does.

[Beenthereuk]

Thanks Carl, yes, the signal was quickly generated by a large (700 uH) coil model I happened to have on file - not ideal, but I understand the basics now.
I had been toying with the idea of compressed amplification so will take Davor's suggestion and look into a log amplifier too. I have to build some circuits and play for a while. Fortunately, the control side of all this will be trivial using a pic.

[Bill512]

Because viscous ground response continues to have 1/t past the metal objects' time, it is the key for subtracting it and detecting on difficult grounds.

Davor, consider the case of a big metal (iron) target .The response from this metal in time length, is well beyond, from any ground response.
I'm talking for a big metal target in a "normal" distance from the coil, because close to coil the things changes fast in favor of metals.

[Bill512]

Ground has a 1/t response, most metals have an exponential response, at least as far as we're concerned.

Carl, speaking strictly mathematical, the response curve from ground or metals, is not a "genuine" 1/t function of time.
From the other side, all of these curves are exponential.
Let's assume that we have plot a curve from some data.
Next, for modeling we consider the closest simplified mathematical function.
So we have loose our terms now,but for practical reasons.
This is a normal procedure ,but I can't see why the ground has a response more close to 1/t and most metals not.

[Davor]

The effect is called magnetic viscosity, and it does respond in 1/t fashion. Non-ferrous metals respond exponentially with a decay related to the conductivity and skin thickness, and the factor often mentioned is tau. For a coil in series with a resistor tau = L/R. Because 1/t can be synthesized with a sequence of exponential decay contributions, you may find every non-ferrous target being nicely aligned with 1/t some portion of time.
Ferrous metals also have exponential decay due to the conductivity, but because their presence influences the coil inductance by means of increased permeability, you have some additional effects that help in discrimination in VLF IB devices. In PI this sort of discrimination is not as straightforward.