AMX TX

[Carl-NC]

Its unfortunate that the term CC has been taken to mean Constant Current when a better term would be ContinuousCurrent.
The bipolar TX always has current flowing in it the notable thing being that the polarity swaps from period to period. (crossing zero point )

Both of the following waveforms can be considered "constant current" because the pulses have constant current amplitudes:



Maybe the former should be CCC-PI: continuous constant current.

I have done alot of work in unipolar systems where this principle is applied and it explains why with huge sample pulses spanning the whole RX period the detector can still resolve sub 0.1 gram nuggets. ... because the integration of the sample pulses and ref pulses is done across the time constant of the receive coils which are widened because they are 'shorted'.

When you see the results there is a real AHA moment.

When the time constant is 'widened' by 'shorting' the coil there are no sudden target transitions occurring because they are smeared across the L/R response of the RX coil.

I don't understand this. Even with a current-mode RX coil the turn-on response should be instantaneous, not an integrated response. In the past I have intentionally slowed down the RX preamp which can drag super-fast targets out in time, allowing you to see something that otherwise may be too fast. But the no-free-lunch scenario applies: stretching out the target response also weakens its amplitude.

[moodz]

... here is the complete post because the board logs you out while typing ... doh.

BUT

• Signal Offset at ADC input : + 2.4V to –2.4V
  • If more receive gain than 50x, Saturation
• XMIT Coil Current Ramp : 33mA over 100µsec
• RCV Voltage Ramp : 380mV

THUS,

• Needs for Automatic Compensation of energy losses.
• This keeps the XMIT coil current CONSTANT
• Dramatically reduces the signal OFFSET at the ADC level

Its unfortunate that the term CC has been taken to mean Constant Current when a better term would be ContinuousCurrent.
The bipolar TX always has current flowing in it the notable thing being that the polarity swaps from period to period. (crossing zero point )

The main thing is that the "FLAT" di/dt is much less than the "TRANSITION" di/dt.

The other thing to note is that if you use some means of regenerative active damping / fast damping ( where the coil is NOT shorted ) and then apply a TIA style ( zero input impedance AKA shorted coil ) preamp to the RX function ...then we all ( should ) know that the time constant of the RX coil will be L/R.
If we have a really good system and our RX coil is 300 uH and 1 ohm loop impedance then the time constant will be 300 microseconds !!
Ok if we dont have such a good RX system and the loop resistance of the RX 'shorted coil' is 10 ohms the time constant is still 30 microseconds.

I have done alot of work in unipolar systems where this principle is applied and it explains why with huge sample pulses spanning the whole RX period the detector can still resolve sub 0.1 gram nuggets. ... because the integration of the sample pulses and ref pulses is done across the time constant of the receive coils which are widened because they are 'shorted'.

When you see the results there is a real AHA moment.

When the time constant is 'widened' by 'shorting' the coil there are no sudden target transitions occurring because they are smeared across the L/R response of the RX coil.

This means there are no sub microsecond events occurring at the RX ... everything ... even very small targets need to be integrated across 10s of microseconds to get a good signal to noise. ( and because of the wide target integration the noise is much lower).

So you may need a fast ADC if you want to take lots of samples for integration ... but the speed wont be for catching target response features and if you look in certain top shelf detectors they are using multichannel sample/integrators with slow but accurate ADCs on each channel.

just sayin ....

moodz

[moodz]

BUT

• Signal Offset at ADC input : + 2.4V to –2.4V
  • If more receive gain than 50x, Saturation
• XMIT Coil Current Ramp : 33mA over 100µsec
• RCV Voltage Ramp : 380mV

THUS,

• Needs for Automatic Compensation of energy losses.
• This keeps the XMIT coil current CONSTANT
• Dramatically reduces the signal OFFSET at the ADC level

Its unfortunate that the term CC has been taken to mean Constant Current when a better term would be ContinuousCurrent.
The bipolar TX always has current flowing in it the notable thing being that the polarity swaps from period to period. (crossing zero point )

The other thing to note is that if you use some means of regenerative of active damping and then apply at TIA style ( zero input impedance AKA shorted coil ) style preamp to the RX function then we all ( should ) know that the time constant of the RX coil will be L/R.
If we have a really good system and our RX coil is 300 uH and 1 ohm then the time constant will be 300 microseconds !!
Ok if we dont have such a good RX system and the loop resistance of the RX 'shorted coil' is 10 ohms the time constant is still 30 microseconds.

I have done alot of work in unipolar systems where this principle is applied and it explains why with huge sample pulses spanning the whole RX period the detector can still resolve sub 0.1 gram nuggets. ... because the integration of the sample pulses and ref pulses is done across the time constant of the receive coils which are widened because they are 'shorted'. When you see the results there is a real AHA moment.

When the time constant is 'widened' by 'shorting' the coil there are no sudden target transitions occurring because they are smeared across the L/R response of the RX coil.

[Willy Bayot]

[ATTACH]n409296[/ATTACH]

The TX current can decrease or increase depending on the TX voltage setting.
When the TX voltage setting is just right, the TX current is nearly flat.
The losses are resistive including RDS of the Mosfets, resistance of the coil and cable.

In the simulation we added a RX circuit.
You will notice that if you change the K L1 L2 0.3, and/or R19, R10,R3, The absorption of the RX coil changes and correspondingly, the slope of the TX current changes.

Now, if you approach the coil to a target, the target absorbs some energy from the TX coil field. This is especially noted with large targets and ground. Again, this changes the TX current slope.

Changes in TX current slope produce an offset in the RX as seen in the added RX preamp. In this RX preamp, no offset adjustments are made. [ATTACH]n409297[/ATTACH]

This is the description of initial conditions and results of the second simulation in POST #45 (typical but powerful CC-based PI system).

• XMIT Coil dia 8”, 50 turns resistance 1 ohm, Inductance 850µH
• RCV Coil dia 4” resistance 2 ohm, Inductance 300µH
• MOSFET VDS : 800V RDSON : 360mOhm (not good)
• XMIT battery voltage : 2.45V
• Pulse period 200µsec = 5Kpps, two half-periods of 100µsec = 10K decays / sec
• Differential Receive chain : gain 50x to ADC input

Results

• Coil Current : +1Amp to –1Amp = delta 2 Amp.
• 2 Amp x 50 turns = 100 Amp.turn
• Flyback Voltage : 900V
• Resonant Frequency : 160KHz, Flyback Width : 3µsec
• Battery Power Consumption : 4.7W
• Pulse Delay : 1.2µsec

BUT

• Signal Offset at ADC input : + 2.4V to –2.4V
  • If more receive gain than 50x, Saturation
• XMIT Coil Current Ramp : 33mA over 100µsec
• RCV Voltage Ramp : 380mV

THUS,

• Needs for Automatic Compensation of energy losses.
• This keeps the XMIT coil current CONSTANT
• Dramatically reduces the signal OFFSET at the ADC level

--> This woiuld allow for a much higher amplification gain without saturation.

[Teleno]

A problem with this circuit is that the RX damping and input resistors get reflected back to the TX coil because of mutual coupling. This happens whether you increase k or present ferrite ground. A circuit that avoids this is a current-mode RX amplifier, like this:


Both sides of the RX coil are at virtual ground so that there is no reverse mutual coupling. RD can be anything you want and it doesn't affect the transmitter. During normal operation it appears RD has no effect, although it actually does because of the finite GBW of the amps. Since there are no flyback spikes on the opamp inputs, the diodes are only needed if the TX circuit starts running before the opamps are powered up.

The gain of the circuit is RG/RL so coils are a little more critical, and the RGs may need to be PTC resistors. A potential drawback with this circuit is that the output signal is no longer the derivative of the incident RX magnetic field. Instead, it is the same waveform. This means ground is now a square wave instead of exponential spikes. But his might actually be easier to deal with.

The advantages as I see it:

In an ideal shorted coil (zero resistance) the current is the integral of the EMF divided by L

This implies that the induced current has a "memory" and there's no need to sample early. The tau of the target becomes irrelevant by integration.
In a real shorted coil the memory decays at a rate L/RL, which is much slower than L/Rdamp so we can sample later without much loss of gain.
- This approach relaxes dramatically the timing constraints.

The disadvantages:

- The information on the target's tau is lost.
- Any error in the induction balance is also integrated and can saturate the preamp.

The approach is suitable for a monocoil too, as in the following schematic:



The depletion MOSFET transitions sharply from high resistance to low Rdson. By dimensioning R6 and Rdamp the switching instant can be made close to the zero crossing of the coil's current. Together with a constant energy drive scheme the set up should be stable. The Rdson of the depletion MOSFET should be as low as possible in order to short the coil to virtual ground. This is similar to Moodz's trick but not exactly.

[Teleno]

True, an error in the null shows up as a square wave output, same as a ground signal. It might not be difficult to compensate.

In my simulation of the circuit the error decays with time constant L/RL. An RC circuit with the same time constant can be charged/discharged synchronously and its decay subtracted from the signal.

[Willy Bayot]

A problem with this circuit is that the RX damping and input resistors get reflected back to the TX coil because of mutual coupling. This happens whether you increase k or present ferrite ground. A circuit that avoids this is a current-mode RX amplifier, like this:


Both sides of the RX coil are at virtual ground so that there is no reverse mutual coupling. RD can be anything you want and it doesn't affect the transmitter. During normal operation it appears RD has no effect, although it actually does because of the finite GBW of the amps. Since there are no flyback spikes on the opamp inputs, the diodes are only needed if the TX circuit starts running before the opamps are powered up.

The gain of the circuit is RG/RL so coils are a little more critical, and the RGs may need to be PTC resistors. A potential drawback with this circuit is that the output signal is no longer the derivative of the incident RX magnetic field. Instead, it is the same waveform. This means ground is now a square wave instead of exponential spikes. But his might actually be easier to deal with.

Carl,
I think this interesting part of the discussion should be moved to the AMX RX topic.

[detectormods]

Also double the noise figure but they may have negligible or little consequence as a current amplifier.

[Carl-NC]

True, an error in the null shows up as a square wave output, same as a ground signal. It might not be difficult to compensate.

[Teleno]

A problem with this circuit is that the RX damping and input resistors get reflected back to the TX coil because of mutual coupling. This happens whether you increase k or present ferrite ground. A circuit that avoids this is a current-mode RX amplifier, like this:


Both sides of the RX coil are at virtual ground so that there is no reverse mutual coupling. RD can be anything you want and it doesn't affect the transmitter. During normal operation it appears RD has no effect, although it actually does because of the finite GBW of the amps. Since there are no flyback spikes on the opamp inputs, the diodes are only needed if the TX circuit starts running before the opamps are powered up.

The gain of the circuit is RG/RL so coils are a little more critical, and the RGs may need to be PTC resistors. A potential drawback with this circuit is that the output signal is no longer the derivative of the incident RX magnetic field. Instead, it is the same waveform. This means ground is now a square wave instead of exponential spikes. But his might actually be easier to deal with.

The problem is that a slight imblalance between the Tx and Rx coils will shift the baseline of the signal, kind of a variably floating signal you need to chase.

https://www.geotech1.com/forums/forum/technology/tech-forum/13221-pi-negative-capacitor-instead-of-critical-damping?p=204212#post204212

[Carl-NC]

In the simulation we added a RX circuit.
You will notice that if you change the K L1 L2 0.3, and/or R19, R10,R3, The absorption of the RX coil changes and correspondingly, the slope of the TX current changes.
[ATTACH]n409297[/ATTACH]

A problem with this circuit is that the RX damping and input resistors get reflected back to the TX coil because of mutual coupling. This happens whether you increase k or present ferrite ground. A circuit that avoids this is a current-mode RX amplifier, like this:


Both sides of the RX coil are at virtual ground so that there is no reverse mutual coupling. RD can be anything you want and it doesn't affect the transmitter. During normal operation it appears RD has no effect, although it actually does because of the finite GBW of the amps. Since there are no flyback spikes on the opamp inputs, the diodes are only needed if the TX circuit starts running before the opamps are powered up.

The gain of the circuit is RG/RL so coils are a little more critical, and the RGs may need to be PTC resistors. A potential drawback with this circuit is that the output signal is no longer the derivative of the incident RX magnetic field. Instead, it is the same waveform. This means ground is now a square wave instead of exponential spikes. But his might actually be easier to deal with.

[Carl-NC]

Such high gains are unnecessary and counterproductive when using an ADC of 16 bit and up. Unnecessary because the noise level of the signal is already higher than the resolution of the ADC. Counterproductive because amplification reduces the dynamic range, the bandwidth, adds noise, offset and delays.

I agree, the whole reason for using a 24b ADC is to get good sensitivity without massive gain. An analysis of dynamic range and noise will tell you how much gain to run and how many bits are really useful. The noise analysis is easy but dynamic range requires knowledge of the worst-case ground response which I don't know. Probably I should build a transmitter and do some tests.

[Willy Bayot]

This is the exact same problem encountered in a square wave drive VLF detector. A square wave TX voltage produces a triangle wave TX field, and ground then induces a square wave voltage on the RX coil. In other words, big offsets. It's really the same problem you encounter in sinusoidal VLF designs where the X-channel ground signal limits the dynamic range. In those designs, you simply limit the X-channel gain. In the square wave design, I was once looking at methods to dynamically buck out the ground component. I should go dig up those notes again.

If we are now convinced that a compensation is required, then let us ask to the electronic specialists (which I am not) in this project to find how to best make the required measurement of errors and the way to apply the corrections to compensate for the energy losses.

Note that this function is not required in the raditional PI technology since the whole energy accumulated in the coil is spent in th edamping resistor at each pulse period.

[Teleno]

.

That would mean that a ARDUINO NANO could do the job??
Thus, we have found the right platform for this project??

If the future experimentations in this project prove that you are right, all is well for me.
The main condition to success is first to keep the SNR high enough before data capture.

In the past, we have failed to get a high enough sensitivity using direct sampling with an internal 12-bit ADC (DMA-based at 0.7µsec per slot on a STM32) and software oversampling/decimation.
Maybe we have made some mistakes in our design or in our implementation!!!

I didn't use a development board. I installed an Atmega328P on the detector's PCB so I could carefully layout the analogue ground. Development boards mix the snalog and digital grounds.

It worked for me. You can see some code and the de-noising process in this thread https://www.geotech1.com/forums/foru...noise-question

The schematic in this thread https://www.geotech1.com/forums/foru...260#post313260

[Willy Bayot]

Such high gains are unnecessary and counterproductive when using an ADC of 16 bit and up. Unnecessary because the noise level of the signal is already higher than the resolution of the ADC. Counterproductive because amplification reduces the dynamic range, the bandwidth, adds noise, offset and delays.

In my experience even the lower spec 10 bit ADC of an Atmega328P proved excellent sensitivity (0.3g gold nugget at 5cm, spiral monocoil) with a preamp gain of 17 using the internal 1.1V reference voltage plus software oversampling/decimation (64 samples to add 3 bits) and low-pass filtering.

.

That would mean that a ARDUINO NANO could do the job??
Thus, we have found the right platform for this project??

If the future experimentations in this project prove that you are right, all is well for me.
The main condition to success is first to keep the SNR high enough before data capture.

In the past, we have failed to get a high enough sensitivity using direct sampling with an internal 12-bit ADC (DMA-based at 0.7µsec per slot on a STM32) and software oversampling/decimation.
Maybe we have made some mistakes in our design or in our implementation!!!