PI limit to early sampling

[Qiaozhi]

OK. I hadn't seen it, there is so many topics on this forum !

I think this paper is very interesting because it gives formula useful to calculate the parameters of a PI system. This paper is just 60 years old !
Now, with modern electronics, we are more in the us range than in the ms range as in 1956 ...

Yes - the paper is still very relevant.
Perhaps you can answer the homework question in post #1, and derive equation 2?
As you can see from later posts, equation 3 appears to be correct; but I failed to figure out how F.B. Johnson arrived at equation 2.

[KingJL]

Originally Posted by Phiphi
Thank you all for your answers. It's clearer for me now.

What I understand is more or less what I was saying in my first post : between Tx off and, let's say 10 us after, it is impossible to distinguish no-target response to target response because the part due to coil is very strong... Is this analysis seems correct for you ?

Yes.

For arguments sake, lets assume you can isolate, capture and store the following:
1. capture and measure any voltage within the range of -5v to +5V with a granularity of 596nv (actually in practical terms we can double the range and still be at uv granularity).
2. the decay/response without ground or target for a given coil.
3 the ground response without target for the same coil.
4. the slow speed changes ( < 0.1Hz ) in the above due to temp changes, general slow changes to ground environment, etc.

Lets also assume that later, you can in (somewhat) real time manipulate a given target received by the same coil, to cancel out the effects of the above.

Would the above assumptions alter your response or your thoughts on the usefulness early sampling?

[KingJL]

For arguments sake, lets assume you can isolate, capture and store the following:
1. capture and measure any voltage within the range of -5v to +5V with a granularity of 596nv (actually in practical terms we can double the range and still be at uv granularity).
2. the decay/response without ground or target for a given coil.
3 the ground response without target for the same coil.
4. the slow speed changes ( < 0.1Hz ) in the above due to temp changes, general slow changes to ground environment, etc.

Lets also assume that later, you can in (somewhat) real time manipulate a given target received by the same coil, to cancel out the effects of the above.

Would the above assumptions alter your response or your thoughts on the usefulness early sampling?

The above was posted simply for discussion... not to dispute anyone's statements or understanding. In particular, I am interested in Carl's input.

[Phiphi]

Perhaps you can answer the homework question in post #1, and derive equation 2?
As you can see from later posts, equation 3 appears to be correct; but I failed to figure out how F.B. Johnson arrived at equation 2.

I'll try it but without any guarantee, I'm an electronics engineer not a physician and I'm not a magnetism specialist !

[Phiphi]

For arguments sake, lets assume you can isolate, capture and store the following:
1. capture and measure any voltage within the range of -5v to +5V with a granularity of 596nv (actually in practical terms we can double the range and still be at uv granularity).
2. the decay/response without ground or target for a given coil.
3 the ground response without target for the same coil.
4. the slow speed changes ( < 0.1Hz ) in the above due to temp changes, general slow changes to ground environment, etc.

Lets also assume that later, you can in (somewhat) real time manipulate a given target received by the same coil, to cancel out the effects of the above.

Would the above assumptions alter your response or your thoughts on the usefulness early sampling?

I agree that trying to cancel the different responses other than target response in post processing after sampling is not the right way to perform early sampling.

We have to re-think the process and perform in a different way. I have the idea to perform differential sampling to amplify only the difference between no-target and target response.
I have to perform some simulations to validate these ideas and I will come back in case of success.

Philippe

[Teleno]

I have the idea to perform differential sampling to amplify only the difference between no-target and target response.

Some analog PIs use the difference between a slow and a fast integrator. In digital signal processing this would be the difference between a short-term and a long-term average (mean). I would suggest using short-term and a long-term ranking filters (e.g. median filter) which allow smoothing of the signal (noise filtering) while preserving the transitions.

See for example: https://machinelearning1.wordpress.c...ilters-python/

[Qiaozhi]

Some analog PIs use the difference between a slow and a fast integrator. In digital signal processing this would be the difference between a short-term and a long-term average (mean). I would suggest using short-term and a long-term ranking filters (e.g. median filter) which allow smoothing of the signal (noise filtering) while preserving the transitions.

See for example: https://machinelearning1.wordpress.c...ilters-python/

The example runs exactly as described using Python, but fails with Python3 in Ubuntu 12.04.

Firstly you need to install python3-numpy and python3-scipy, as Ubuntu/Debian treats Python and Python3 (and their associated libraries) as different packages. This is because there are very few libraries that have a common executable for both Python versions.

Unfortunately, there is no python3-matplotlib available for Ubuntu 12.04, but it is apparently available for Ubuntu 13.04.
I then tried building matplotlib from source (downloadable from github), but the process fails.

I assume the same will happen on RPi, but haven't tried that. Wasted too much time on it already.

[Teleno]

Here's C code for Arduino: http://www.elcojacobs.com/eleminatin...tal-filtering/

This is a variation of the median filter called "mode filter". Advantages:

A median filter is much better in reducing the impact of noise than an averaging filter. It works like this: you take 100 measurements, sort them from small to large and take the middle one. A median filter is much less influenced by outliers, since they end up at the beginning or the end.
A mode filter is a combination of a median filter and an averaging filter: you sort the values and take the average of the ones in the middle

This C code implements a median filter without intensive sorting, just what is needed for embedding: http://embeddedgurus.com/stack-overf...ian-filtering/
Original article: http://m.eet.com/media/1173225/f-eckstro.pdf

[green]

Eric stated he has sampled at 1.5usec in another thread. How soon do we need to sample? Some thoughts. What is the lowest target TC we are looking for? The ground signal increases faster than the target signal if you sample sooner than the target TC, decreases S/N. The GEB hole is between the two samples, the sooner the first sample the lower the hole TC making it more likely the hole will eliminate a desired target. If there is no ground signal then the lower the sample delay time the better.

[Carl-NC]

For arguments sake, lets assume you can isolate, capture and store the following:
1. capture and measure any voltage within the range of -5v to +5V with a granularity of 596nv (actually in practical terms we can double the range and still be at uv granularity).
2. the decay/response without ground or target for a given coil.
3 the ground response without target for the same coil.
4. the slow speed changes ( < 0.1Hz ) in the above due to temp changes, general slow changes to ground environment, etc.

Lets also assume that later, you can in (somewhat) real time manipulate a given target received by the same coil, to cancel out the effects of the above.

Would the above assumptions alter your response or your thoughts on the usefulness early sampling?

It's pretty easy to accomplish first-order coil and ground cancelation with just a judicious coil design. For example, you can eliminate the coil decay by using an induction-balanced coil, and then you can easily sample down to 0us. I've done this. Furthermore, if you make the IB coil in a Bigfoot configuration (figure-8 receive coil) then you will get a first-order cancelation of the ground. In milder ground this might be sufficient, and you can take advantage of extremely short sample delays. Whether it holds for hotter ground, I don't know. The presence of hot/cold rocks tends to screw up all the good ideas that assume a homogeneous ground matrix.

[green]

It's pretty easy to accomplish first-order coil and ground cancelation with just a judicious coil design. For example, you can eliminate the coil decay by using an induction-balanced coil, and then you can easily sample down to 0us. I've done this. Furthermore, if you make the IB coil in a Bigfoot configuration (figure-8 receive coil) then you will get a first-order cancelation of the ground. In milder ground this might be sufficient, and you can take advantage of extremely short sample delays. Whether it holds for hotter ground, I don't know. The presence of hot/cold rocks tends to screw up all the good ideas that assume a homogeneous ground matrix.

Hi Carl

I keep reading I should be able to sample down to 0usec. I'm using a Bigfoot style coil, two 8 inch round coils Rx(647uH, 1.05MHz SRF). Oval Tx surrounding Rx(378uH, 895kHz SRF)SRF measured with coil leads connected to coil. Differential input preamp(gain about 375). Tx, 160usec, constant rate, 6250 amps/sec, 1 amp peak. The coil is very close to balance without trimming with a ferrite bead core. Including some scope pictures, coil volts and preamp out volts. Top, ferrite near one end to unbalance coil when charging. Middle, ferrite near center to balance preamp out when coil is on. Bottom, Ferrite opposite end. I can zero preamp out when coil is on and change the waveform when Tx is turned off but I can't get rid of the saturated signal. Any suggestions what or where my problem is? Coil, preamp or something else.

[Carl-NC]

Sampling down to 0 requires a very wideband null on the coil. I use lower inductances, I think the last time I did it, TX=100u and RX=250-300uH. Use a bit of ferrite for the X null, a bit of foil for the R null. I also run a low preamp gain, typically 100.

[green]

Hi Carl

I keep reading I should be able to sample down to 0usec. I'm using a Bigfoot style coil, two 8 inch round coils Rx(647uH, 1.05MHz SRF). Oval Tx surrounding Rx(378uH, 895kHz SRF)SRF measured with coil leads connected to coil. Differential input preamp(gain about 375). Tx, 160usec, constant rate, 6250 amps/sec, 1 amp peak. The coil is very close to balance without trimming with a ferrite bead core. Including some scope pictures, coil volts and preamp out volts. Top, ferrite near one end to unbalance coil when charging. Middle, ferrite near center to balance preamp out when coil is on. Bottom, Ferrite opposite end. I can zero preamp out when coil is on and change the waveform when Tx is turned off but I can't get rid of the saturated signal. Any suggestions what or where my problem is? Coil, preamp or something else.

Tried another coil I've been working on, similar to the one above. Comes out of saturation a little after 2usec. Some scope traces,(Ch#1 preamp out, Ch#2 probe laying near coil fet to show decay wave form). Top row, no target. Bottom row, target. First column, 99.9% pure 1 troy oz. copper coin about 5 inches above one of the figure 8 Rx coils. Second column, 1 qt. zip lock bag with California clay(6x6x1.5 inches) about 3 inches above the opposite figure 8 Rx coil. Get a large change in amplitude between 2 and 3usec, some to 4usec. Not much after 4usec. I'm thinking the large signal change is due to the Tx coil current decay especially since the copper coin and California ground are on opposite ends of the figure 8 Rx coil to get the same response. Maybe I'm thinking or doing something wrong, appreciate someone correcting me if I am.

[green]

Sampling down to 0 requires a very wideband null on the coil. I use lower inductances, I think the last time I did it, TX=100u and RX=250-300uH. Use a bit of ferrite for the X null, a bit of foil for the R null. I also run a low preamp gain, typically 100.

Thanks for the reply, I was posting another test at the same time. Hadn't tried the foil for a R null. The null on my coils seem narrow to me, how do I know if it's wideband and what is required to make it wideband? Sorry for the questions, I'm still trying to learn.

[Ferric Toes]

I found that the damping on the TX coil was critical at short delays. Try a fixed resistor in series with a reasonable wattage ceramic trimmer and adjust for the flattest RX response with quickest recovery. It looks in your last .png that there is some undershoot in the amplified RX and when you introduce a target, it is cancelling the undershoot before looking like a sensible signal. Try a damping preset trimmer and see if that gets rid of the undershoot.

Eric.