Preamp

[Monolith]

That is what I wanted to have clamping for, so we can avoid saturation of the front end and have a nice low noise amplifier. Most Pi's I see, the RX signal bounces a lot at the sample point. What I would like to try is to eliminate as much of this as we can so we can use high amplification and filtering to get a smooth threshold but fast response. I have some ideas but am still researching some ideas.

If we can characterise the system noise, we can work out how best to reduce it. ��

Finding the source of the noise is obviously the first step. What exactly causes the noise.
A lot has been spoken about thermal noise. But, is this really the problem? If the RX signal bounces at the sample point, it seems this is not thermal noise.
It is a well known fact that pushing the sample very early makes the detector unstable. Why? This is one question we should find the answer to.

When we take a wide sample, say 20us wide, we integrate/average, the whole signal, including the high frequency noise during that time.
If we take a very short first sample, let's say 1us wide, we get a much higher amplitude of very short target TC response, but, any high frequency noise, makes the signal amplitude jump up and down.
Now, if we integrate/average the output, we reduce that noise again, but eliminating the noise before amplification is always better.

So, what is that input noise. Where does it come from? Would white noise/thermal noise bounce he signal? What is the frequency of the "bouncing"? Can we correlate this frequency to anything emanating from the detector circuit?
Is the noise totally random?

Let's start with this.

[daverave]

i tried a small 10 pf capacitor across the fed back resistor of the first amp...made the threshold more stable but i had to change the sampling position to get any signal...so i just now removed the 10 pf capacitor and put it back to normal...the noise from the first amp seems to be the major problem of instability of the threshold...hope you find some way around this problem.

[Sean_Goddard]

That is what I wanted to have clamping for, so we can avoid saturation of the front end and have a nice low noise amplifier. Most Pi's I see, the RX signal bounces a lot at the sample point. What I would like to try is to eliminate as much of this as we can so we can use high amplification and filtering to get a smooth threshold but fast response. I have some ideas but am still researching some ideas.

If we can characterise the system noise, we can work out how best to reduce it. ��

[6666]

Here is the link to the thread, discussing the initial circuit.

Thanks.

[Monolith]

Do you still have the circuit ?

Here is the link to the thread, discussing the initial circuit.

http://www.geotech1.com/forums/showthread.php?11986-Is-a-resonant-PI-possible&p=45228#post45228

Basically it is a soft clipping circuit.

When we use a hard clipping circuit, the reverse recovery of the diodes messes up the wave form.
Using specific very soft recovery Schottky diodes would help in this respect.

Using the transistor diode, again, choosing a specific type of transistor, with soft recovery is about the same. However, a further development of the circuit, I have not found it yet, uses biasing on the transistor base, applying feedback from the opamp output.
Using feedback, when the output goes near the rail, the transistor starts conducting strongly, so the higher voltage is not amplified, and then clipped softly so that the opamp does never saturate.

With PI, we are only interested in the part of the signal that is near 0V, therefore we can amplify the part of the signal that we are interested in, very much and discard the unimportant part.

[Teleno]

Actually ... no. In case Rx coil is balanced and you don't get high voltage spikes, so that you don't have non-linearities at front end, the Rx coil's response is a cyclostationary one, on top of which is piggybacking a target signal. If you subtract the cyclostationary component (the one that is non changing from one cycle to another), you'll see a target signal in perfect exponential form, perfectly linear in lin/log scale, starting much sooner than you expect.

Mono sucks at this department due to the preamp non-linearity at high voltages.

A target invariably unbalances the coil causing a spike ranging up to a few volts. This spke decays at the tau of the LRC around the coil. The target signal induces a voltage in the opposite direction.

[Old cart]

As I remember, the spiral wrap was to give a distance from the coil bundle to the shield. The coils themselves were bundle wound, not spiral wound.

Spider web or basket wind coils have indeed lower capacitance by themselves, but most of that advantage is lost when shielding.

I thought that the basket weave coil was self shielding so that no external shield is required.

[Davor]

In the presence of a target, the Rx signal is the difference of two exponentials: the Rx coil own decay minus the decay of the target. A low damping R means the Rx coil decays slowly and sampling will need to be done later, reducing S/N.

Actually ... no. In case Rx coil is balanced and you don't get high voltage spikes, so that you don't have non-linearities at front end, the Rx coil's response is a cyclostationary one, on top of which is piggybacking a target signal. If you subtract the cyclostationary component (the one that is non changing from one cycle to another), you'll see a target signal in perfect exponential form, perfectly linear in lin/log scale, starting much sooner than you expect.

Mono sucks at this department due to the preamp non-linearity at high voltages.

[6666]

Somewhere on a broken hard drive. Will have to redo it from memory if I find the time. Very simple circuit.

Thanks.

[Monolith]

That paper is outdated. It recommends a Teflon spiral wrap as the fastest option. Teflon has a dielectric constant of 2.

Instead, air has a dielectric constant of 1. The lowest capacitance coils are woven from enameled wire as spiderwebs such the examples on this page.

As I remember, the spiral wrap was to give a distance from the coil bundle to the shield. The coils themselves were bundle wound, not spiral wound.

Spider web or basket wind coils have indeed lower capacitance by themselves, but most of that advantage is lost when shielding.

[Monolith]

Do you still have the circuit ?

Somewhere on a broken hard drive. Will have to redo it from memory if I find the time. Very simple circuit.

[Teleno]

Good information in the linked articles below.

http://www.geotech1.com/pages/metdet...s/FastCoil.pdf

That paper is outdated. It recommends a Teflon spiral wrap as the fastest option. Teflon has a dielectric constant of 2.

Instead, air has a dielectric constant of 1. The lowest capacitance coils are woven from enameled wire as spiderwebs such the examples on this page.

[6666]

Years ago I posted a pseudo log pre-amp, with a transistor in the feedback loop

Do you still have the circuit ?

[Monolith]

Ivconic, excellent summary of the coil information. I have always thought a DD coil would be nice design for a PI but I don't see many designs using it so I dismissed it. I have had good results on my modified (very old) Garrett XL 500. It has a small 6" coil with low inductance 128uH and I have been able to reduce the sampling delay to about 14 uS after optimum damping the coil. This increased the depth on a US nickel and mans gold ring from 20 to 30 cm-50%. To summarize it would seem that the following are necessary for best performance, at least on low conductivity targets:

1. Critical coil damping
2a. Early sampling
2b. Well nulled coil or other means to minimize TX residual to prevent early amp stages from saturating.
3. Low coil inductance
4. Low coil capacitance
5. Short TX pulses ( just long enough for target TC)

Of of these I have implemented and proven 1,2a,3 and 4.

Good information in the linked articles below.

http://www.geotech1.com/pages/metdet/projects/fastcoils/FastCoil.pdf

http://www.geotech1.com/cgi-bin/pages/common/index.pl?page=metdet&file=/projects/coplanar/index.dat

[Monolith]

That is an impressive circuit. Looks like a compensated voltage divider. The question I have is this. In a PI do we have to carefully balance out the TX signal or is it just fine to simply reduce the TX residual signal to the point where the first stage does not saturate?
In a more advanced detector I can see the value in a processor based auto calibrate feature that would compensate for nulls and maybe even adjust for different coil LCR.
In a simple design though maybe just getting the TX residual down to less than say 8 volts peak to peak divide by first stage gain ( 10 in this example) or 0.8 volts p-p would keep the first stage out of saturation and go along way towards allowing earlier sampling.

You are right, saturation is one of the critical problems. If you can not avoid saturation, use a fast opamp, something like AD8055, that comes out of saturation fast. However, since it's noise figure is not all that good, use a low noise opamp for the first stage and keep that stage unsaturated.
Years ago I posted a pseudo log pre-amp, with a transistor in the feedback loop. It did not saturate. I tested it with a gain of up to 3000, saturation free. Amazing sensitivity.