Moodz' Awesome Gold Pulse Induction Version 3 - MAGPI V3 Project

[Teleno]

doh ... typo on the damping mosfet ... Here is the FINAL FINAL schematic.

R6 (10K) across U2 is gone, was that intentional?

[Teleno]

Been studying and simulating the preamp and control loop of this circuit and want to share some observations.

R12 solves a ringing problem but is hast a downside. During the Tx time, current flows from VP through Q1, D3, the body diode of Q2 and R12 to GND. It amounts to a few hundreds of mA, very inefficient. It can be remedied with an NMOS type in series with Q2, its gate driven by the Tx pulse, closing when Q1 is open and vice-versa.

When switch U4A is open (low pass filter disconnected from feedback), U2 is configured in open loop, acting as a comparator with reference in the voltage held in C3. Since this happens during the target period I wonder how it can ampliffy anything. Perhaps there's some fact I'm missing, but it puzzles me.

It has been speculated that the early signal could be an X component. In my opinion it's just an artifact. As the coil current approaches zero, Q2 transitions from saturation to linear regime and negative resistance occurs at the sharp elbow caused by active damping. The time constant of the target affects this transition giving rise to unusual behaviour, such as short tau signals appearing under the baseline and long taus above it (from simulations), but later in the period they return to normal.

[moodz]

Been studying and simulating the preamp and control loop of this circuit and want to share some observations.

R12 solves a ringing problem but is hast a downside. During the Tx time, current flows from VP through Q1, D3, the body diode of Q2 and R12 to GND. It amounts to a few hundreds of mA, very inefficient. It can be remedied with an NMOS type in series with Q2, its gate driven by the Tx pulse, closing when Q1 is open and vice-versa.

......

I dont know what simulations you did but the transmit voltage is 5 volts and R12 is 330 ohms ... so the theoretical maximum current is 5/330 = 15.5 ma .. not a "few hundreds of mA".
In the actual circuit the current is less than 10 ma ( 5 volts - diode drop voltages / 330 = 9 ma approx. )

[Teleno]

I dont know what simulations you did but the transmit voltage is 5 volts and R12 is 330 ohms ... so the theoretical maximum current is 5/330 = 15.5 ma .. not a "few hundreds of mA".
In the actual circuit the current is less than 10 ma ( 5 volts - diode drop voltages / 330 = 9 ma approx. )

My bad, I was simulating with a higher voltage and a smaller resistor.

Did you remove R6 from any of your prototypes and still work?

Modulating the pulse width at U4B can be used for coarse/fine adjustment of the operating point.

Even if it's a motion detector, with a faster loop (about 10ms response time) the voltage at C1 reacts to a target as a step change. This could be used as a static target indicator, but its amplitude will be about 10 times smaller than the unamplified signal ( d(Vzp)/d(Vgs) ~ 11 in my simulation).

Instead of R12 I place a 50 ohm resistor between Q1 source and U2 (-) with R6 in place. The gain becomes less dependent on Rdson variations and the coil is still very overdamped for a soft-landing after the fast kick down. It's interesting to see that the currents from the target last longer and peak about 2us - 4us later, and the peaking depends on the time constant, so a fixed sampling delay benefits some targets in the detriment of others.

If only I had the spare time and peace of mind to build this...

[Teleno]

I think the gate capacitor C1 = 1uF is seeing current spikes from Q1 during Tx turn on and flyback, coupled through the drain-gate capacitance. It will accumulate noise from one period to the next destabilizing the bias point.

A solution would be to isolate this capacitor with a JFET voltage follower whose output controls a larger cap C1b (via a small 0.1 ohm resistor to avoid ringing) This second cap would drive the gate and absorb the spikes. The follower restores the charge in C1b back to the stable reference in C1, all within one period, so that the errors are corrected from one sample to the next and do not accumulate.

[Teleno]

A fast servo loop will reflect the target in the gate control signal. The control pulses are passed to a second leaky integrator with high gain to produce the amplified target signal. It will be quasi-static if the leak factor is small. A depletion type damping mosfet will settle to a Vgs not far from zero, which is good for integration (more balanced + and - pulses) .

Red: no target.
Blue: 1us target
Green 10us target

The bump on the left is the servo transient before reaching the steady state.

[Teleno]

When the coil is optimally balanced, its inductance L is virtually grounded through a low R, with a time constant

In the absence of a target this is a slowly decaying exponential (about 12us with L=300uH and R=25)

In the presence of a target the signal is a subtraction of two exponentials: the coil's and target's This is a family of functions that rise from zero to a maximum as in a sine, and then decay.

The time to the maximum depends on the relative amplitudes and taus:

[moodz]

... the original purpose of the current damping servo loop was to improve the damping speed of a monopolar PI over a conventionally damped PI and it does this ( there is at least one active patent still covering this today ... not mine )

However the damping loop can become messy and the servo action interferes with the target and ground balance requirements.
There is a solution though ... MAGPI 4 - the next generation .. uses alot less parts than version 3 because it drops the requirement for a servo loop and utilises bipolar pulsing so we have twice the signal to noise of a monopolar PI.
We dont even need the ringing clean up resistance as the damping is almost perfect in MAGPI 4. ( this is the basis of my patent TEM monocoil ).
There is a clue in the orginal patent that is behind MAGPI3 ..... the patent specifies a switch not a diode as the damping current element because whereas a diode can be a switch .. not all switches are diodes.
The dropout voltage across the diode is what causes problems in MAGPI 3 and the requirement for a ringing cleanup resistor ( residual damping ).
Below is the transmit current in the MAGPI 4 monocoil with the RX voltage ( from the same coil ) and then the RX voltage x 100 ( 40 dB ) post preamp. No TX tilt ( approx 40 uA ) No resistor at all. No ringing. One microsecond flybacks.

[Teleno]

... the original purpose of the current damping servo loop was to improve the damping speed of a monopolar PI over a conventionally damped PI and it does this ( there is at least one active patent still covering this today ... not mine )

Interesting. Care to share the patent number?

However the damping loop can become messy and the servo action interferes with the target and ground balance requirements.

I think ground balance is better left to DSP. The loop keeping the coil in a "superdamped" state is already challenging enough.

The dropout voltage across the diode is what causes problems in MAGPI 3 and the requirement for a ringing cleanup resistor ( residual damping ).

These problems can be avoided as in the schematic above.

At the end of damping the coil "wants" a low impedance (inverting configuration with input resistance of 25 ohms) and the gain of 10x is low enough to prevent saturating the LM6171 during the transition of the diode from ON to OFF. The op amp remains in total control of virtual ground never letting lose of the coil. The output is enough for DAC. The higher gain for the sevo loop is added in a second stage.

Below is the transmit current in the MAGPI 4 monocoil with the RX voltage ( from the same coil ) and then the RX voltage x 100 ( 40 dB ) post preamp. No TX tilt ( approx 40 uA ) No resistor at all. No ringing. One microsecond flybacks.

Looks good. Are you using a TIA?

[moodz]

Its hidden away in patent US9829598B2. .... note the expiry date of this patent
No I am not using a TIA just a straight 40 db low noise amp .. the bipolar TEM relies on reciprocal energy conservation .. or "ping pong" using a high inertial factor ( in this case a big inductor ).
The big inductor is a clamp for DC current ( so the ground cant tilt the pulses ) but a high impedance for target signals ( which are not DC ) so we can simultaneously recieve and transmit from one coil with no TX "off" time.
On the bench MAGPI4 is 4 times simpler and 4 times more sensitive than MAGPI3 .

The bipolar monocoil TEM patents will surpass the minelab patents that are trying to lock up bipolar PI technology till nearly 2040.
My reason for patenting is not to make money its to make sure the idea does not get hijacked.

[Teleno]

Its hidden away in patent US9829598B2. .... note the expiry date of this patent

They basically repackaged your patent, to put it mildly.
After al the obscurong legalese what remains in clear engineering terma is your idea:

The Active Element: A Controllable T/R Switch
Instead of relying solely on a static resistor, this invention uses the transmit/receive (T/R) switch itself as an active, controllable damping element. Rather than treating the T/R switch as a simple on/off gate (open during high voltage, closed during receive), it's operated as a variable-impedance device — it can behave as:
a constant-current sink,
a resistor,
a parallel combination of both,
or some other nonlinear current-voltage relationship
— and its exact behavior is tuned in real time.

The feedback loop:

Samples the signal coming out of the T/R switch right at the end of the back-emf decay,
Compares it against a target (ideally zero residual current/voltage),
Feeds back to adjust either the switch's drive (current magnitude, turn-on timing) or the duration/voltage of the preceding transmit pulse,
This self-corrects cycle-to-cycle, so it automatically compensates for coil inductance changes (e.g., from magnetic soil) without needing to know exactly when the back-emf actually ends.

[Teleno]

Its hidden away in patent US9829598B2. .... note the expiry date of this patent

The patent lists alternatives/combinations of the damping element that do not improve on the constant current sink, which is optimal, and therefore they defeat the stated technical effect.

Only a current sink gives non-exponential discharge. It forces the back-emf voltage at any given time to whatever level sustains that current, and the coil current ramps down linearly: dI/dt roughly constant. A linear ramp reaches zero at a calculable finite time. That's the physical reason a current sink gives a faster and more deterministic zero than than any other alternative or combination. It's a different decay law rather than an engineering preference.

Looking at what the patent lists (resistor-only, current source only, resistor+current source in parallel, various nonlinear admittances), the resistor-only case is exponential and slower, the parallel R+current sink combination isn't a different mechanism as it's still using a current sink element to do the fast part of the work, with the resistor handling a secondary/residual and exponential delayed role (post-decay damping). So the "alternatives" aren't such and can't achieve the speed of a current sink.

Contrary to the claims, the residual (which is mentioned as a prior art unsolved problem in Moody's) cannot be eliminated and the document admits this itself. It states outright tat a jitter of ~10 ps alone produces several μA of residual current and several mV of residual voltage, which after preamp gain becomes hundreds of mV. Any servo loop nulls the average or expected error over many cycles, but cannot achieve a nonzero residual on any given cycle. So a claim phrased as "achieving zero residual" should really be read as "minimizing residual" and that's true of both yours and this patent. The "improvement" is not clear.

This patent's broader language reduces to the same core physics of a current sink draining the coil's stored energy at roughly constant current, governed by a feedback loop that minimizes (not eliminates) residual error. Then the technical substance underlying both patents is the same, the "different" claim language in the later filing maps to functionally the same physical mechanism, just with the feedback loop's control variable redescribed: adjusting the Tx on-tine instead of the sink. Yet the current sink needs to be previously adjusted to a near optimal value, and the patent doesn't disclose this.

The document doesn't disclose or that K (the magnitude of the current sink) needs to be prest close to optimal for the Tx-on-time-control embodiment to converge in a well-behaved (low-sensitivity, fast-settling) regime. It shows the sensitivity problem near optimal K only in the context of the other control variable (switch timing), and seems to implicitly assume the reader will carry that caution over to the Tx-side case without stating it.

This is a real insufficiency of disclosure. If the only way to make "adjust Tx on-time, leave K fixed" work robustly is to first set K near its optimal value by some separate, undisclosed means (manual calibration, a startup auto-tune sequence, characterizing C in production) then the claimed single-feedback-loop architecture is incomplete as written. It's presenting one knob as sufficient while quietly depending on the other knob already being in the right neighborhood, with no disclosed method for getting it there. That's a legitimate "enablement" question (does the patent teach someone how to actually build the thing, or does it skip an essential step?) which is exactly the kind of issue a patent attorney would want flagged when assessing whether the later claims hold up against the earlier disclosure.

[moodz]

...they are persistant .. Nokta is battling them here :

https://ipsearch.ipaustralia.gov.au/patents/2023204592

this patent is trying to lock up bipolar constant current sensing till 2044.