PI limit to early sampling

[mickstv]

You'll have a slow response in your long TC integration and a fast response in yout short TC integration, they wont cancel each other so you'll have a response on the output. I've tried similar intergrator methods like your version before and it doesn't work.

[Phiphi]

No, I don't agree. If you look at the waveforms I have joined, you will see that, without target, the fast and slow responses are the same because the input is the same but when the target enters in the coil field there is a fast variation modifying the output of the fast integration but the output of the slow one doesn't move significantly so there is a variation of the difference.
It will be the same with a earth field varying slowly.

[Qiaozhi]

R9 and R27 are forming a voltage divider so the input to your first amplifier is 1/10 vcoil. The first amp gain is approx 10 ... so your amplifier gain is effectively lost by the input divider R27/R9.

I've already pointed out that problem in post #69, and here was his reply:

Yes, you are right, the damping resistors and the diode form a potential divider but not a linear one. When the voltage is high, the current in the diode is high and it's equivalent resistance is very low so there isn't a significant current in the 22 ohms resistor but at low voltages, the current in the diode is very low and all the current runs in the 22 ohms resistor and we have a potential divider 22/( 22 + 738 )
That's the same in the classical solution for high voltages but for low voltages, the equivalent resistance of the diode is very much higher than the 1k resistor and there isn't any attenuation.
However, I think my solution remains to be considered. It seems possible to optimize the ratio between the two damping resistors to decrease the decay time without attenuate too much. For example, with 675 ohms and 75 ohms, the attenuation is a 5 factor and with an amplifier with a 5 gain, the equivalent circuit has an early sampling at about 4.5 us to reach 100uV at amplifier output.

I don't want to put a damper on things (pun intended), or to be excessively blunt, but ...
Many designs that look good in the simulator fall flat on their what-sits once they get introduced to the real world. My suggestion to Phiphi is to start prototyping some of these "ideas" before going any further down a rat-hole.

[mickstv]

No, I don't agree. If you look at the waveforms I have joined, you will see that, without target, the fast and slow responses are the same because the input is the same but when the target enters in the coil field there is a fast variation modifying the output of the fast integration but the output of the slow one doesn't move significantly so there is a variation of the difference.
It will be the same with a earth field varying slowly.

Instead of working in sims, build it and see what happens in the real world.

[Phiphi]

It's what I intend to do but simulations are a fast way to try many different parameters more easily than on a breadboard !

And, being in France, I'll have "difficulties" to test it in Australia !

[mickstv]

You can test earth field and magnetic ground response by waving a magnet over the coil.

[moodz]

Instead of working in sims, build it and see what happens in the real world.

Sims have their place as do building real things ... :-)

To simulate a varying magnetic field ... put a sinusoidal voltage source in series with the RX coil with a peak to peak voltage of approx 1 - 10 millivolts. I usually use a frequency of 50 Hertz ... ( to solve the mains ingress problem at the same time also it is easier to see in the results ) however there is no reason why you cant use 1 to 10 hertz to simulate waving magnet. If the voltage source appears in your sampled / processed output then your earth / magnet / mains cancellation is not working.

One of the problems with building a real circuit and test field cancellation is that the ferrite of the magnet .... not just its field can also cause a problem. In this case the sim is easier to debug earth field than real world. :-)

[Phiphi]

OK, I am going to try that but the simulation risks to be very long as I have to simulate some periods of 50Hz signal (let's say 100ms min for 5 periods) with a high resolution (< 1us) to see details.
Next time with the results !

[6666]

has anyone tried Active Damping, and by that I mean using a FET to "slug" the first part of the decay using severe overdamping then in the region of interest, switching to a lower damping coefficient (resistor)?

I wonder if this would provide and benefits.

While this is not exactly what you are after, its an interesting thread

http://www.geotech1.com/forums/showt...UIT-FOR-A-COIL

[Phiphi]

Next time with the results !

I come back with new simulations. Here is the schematics with some modifications :



I have added a voltage source in series with the coil with 2mV offset to simulate a constant field and a 4 mV sinus at 20 Hz to simulate an other varying slowly.
At the outputs of U3 and U4, I have added a high pass filter to attenuate low frequencies. This has a differentiating effect so I have two thresholds created with resistors R17 to R20 used by two comparators U18 and U19. These comparators associated with the D flip-flop U20A create at the flip-flop Q output a pulse when the target is under the coil.
Here are some waveforms :
First, a zoom on the preamp output showing the effect of the voltage source in series with coil



Now the outputs of fast and slow samplers :



The same but after the high-pass filters :



And finally, at the bottom the output of U5 amplifier and the upper and low thresholds and on top the digital signal at "1" state during target detection :



So it's possible to detect a target even with a field varying slowly. OK, I know that I have to build it and prove it works really but simulations are really useful before doing that !
Let me some time to test it in the real life and I will come back to give you some news !

[green]

Hi Eric,

If you can post the circuit(s) you have in mind, then I'll knock up a simulation.
Perhaps. in the process, we can also get you indoctrinated into the world of SPICE.

Eric hasn't posted any circuits yet. Thought I would try the MPP integrator to understand it better with LT spice. Tried voltage controlled switches instead of fets. Wanted to do an AC analysis to plot frequency response when changing sample times or gain. Analysis is way to slow. Tried a 5Hz input with 10 and 50usec sample times, still slow but usable. Didn't get a 5times increase in signal with the 50usec sample that I would expect with a 1C integrator. Any ideas on how to improve the setup? Tried the AC analysis again, different but still doesn't look right. Doing something wrong.

[moodz]

Eric hasn't posted any circuits yet. Thought I would try the MPP integrator to understand it better with LT spice. Tried voltage controlled switches instead of fets. Wanted to do an AC analysis to plot frequency response when changing sample times or gain. Analysis is way to slow. Tried a 5Hz input with 10 and 50usec sample times, still slow but usable. Didn't get a 5times increase in signal with the 50usec sample that I would expect with a 1C integrator. Any ideas on how to improve the setup? Tried the AC analysis again, different but still doesn't look right. Doing something wrong.

Eric hasn't posted any circuits yet. Thought I would try the MPP integrator to understand it better with LT spice. Tried voltage controlled switches instead of fets. Wanted to do an AC analysis to plot frequency response when changing sample times or gain. Analysis is way to slow. Tried a 5Hz input with 10 and 50usec sample times, still slow but usable. Didn't get a 5times increase in signal with the 50usec sample that I would expect with a 1C integrator. Any ideas on how to improve the setup? Tried the AC analysis again, different but still doesn't look right. Doing something wrong.

I think you will find he has published some ccts here .... http://www.geotech1.com/forums/showt...640#post223640

Particularly post number #3 ... its so close to a a very significant improvement in PI performance .. its one step away


You may also refer to the PUBLISHED PRIOR ART ...

Sample and hold demodulator with feedback for reduced riffle
US 4617521 A

I was retired and had time to work on this ... I lodged a provisional patent ... however now that I am back in the workforce and no time so I am putting it out there ...

Eric's circuits are briliant ... a few tweaks and they will shine even more. RE: ferrite discrimination ;-)

moodz

[green]

I think you will find he has published some ccts here .... http://www.geotech1.com/forums/showt...640#post223640

Particularly post number #3 ... its so close to a a very significant improvement in PI performance .. its one step away


You may also refer to the PUBLISHED PRIOR ART ...

Sample and hold demodulator with feedback for reduced riffle
US 4617521 A

I was retired and had time to work on this ... I lodged a provisional patent ... however now that I am back in the workforce and no time so I am putting it out there ...

Eric's circuits are briliant ... a few tweaks and they will shine even more. RE: ferrite discrimination ;-)

moodz

Thanks for the reply. When I replied (Eric hasn't posted any circuits), it was about using spice for modeling circuits reply #76. I tried the MPP integrator because I've never understood how it works.

[Qiaozhi]

Eric hasn't posted any circuits yet. Thought I would try the MPP integrator to understand it better with LT spice. Tried voltage controlled switches instead of fets. Wanted to do an AC analysis to plot frequency response when changing sample times or gain. Analysis is way to slow. Tried a 5Hz input with 10 and 50usec sample times, still slow but usable. Didn't get a 5times increase in signal with the 50usec sample that I would expect with a 1C integrator. Any ideas on how to improve the setup? Tried the AC analysis again, different but still doesn't look right. Doing something wrong.

I contacted Eric, and he will provide the circuits when he has some spare time.

In your post you stated that you wanted to run an AC analysis on the MPP integrator. However, your waveforms are from a time domain analysis.
Have a look at the analysis command in the schematic. It says ".time" in both cases instead of ".ac".
An AC analysis is also known as a small-signal AC analysis. The simulator linearises the models around the current operating point, and therefore the results are only good for small signal changes. For a large-signal analysis you would need to use a different technique known as harmonic balance simulation, but this is not available in LTSpice. Also, even if you set up the analysis as AC, the switches (or even fets used as switches) are essentially time domain devices, so the results of an AC analysis would be nonsensical.

Your only option is to run the simulation in the time domain, and then perform an FFT on the results. However, I'm sure just how useful the results will be in that case.

[green]

I contacted Eric, and he will provide the circuits when he has some spare time.

In your post you stated that you wanted to run an AC analysis on the MPP integrator. However, your waveforms are from a time domain analysis.
Have a look at the analysis command in the schematic. It says ".time" in both cases instead of ".ac".
An AC analysis is also known as a small-signal AC analysis. The simulator linearises the models around the current operating point, and therefore the results are only good for small signal changes. For a large-signal analysis you would need to use a different technique known as harmonic balance simulation, but this is not available in LTSpice. Also, even if you set up the analysis as AC, the switches (or even fets used as switches) are essentially time domain devices, so the results of an AC analysis would be nonsensical.

Your only option is to run the simulation in the time domain, and then perform an FFT on the results. However, I'm sure just how useful the results will be in that case.

Tried a square wave input, easier to see response. (_2) 10 and 50usec sample time with R5(R22 MPP) 56k, (_3) 10 and 50usec sample time with R5(R22 MPP) 150k. Not what I was expecting, maybe something wrong with my simulation.