What is the best frequency response for the preamp in a PI ?

[moodz]

1.When designing a band-pass filter before demodulation, you need to find out what Spectrum the target signal has. When I mentioned the bandwidth, I meant the Spectrum of the target signal. If the target's constant time is 1u, then its Spectrum starts from 1 MHz and above.
Harmonics of the pump signal are the spectrum of the pump signal but not the target signal.

2.The filter you mentioned above as a band-pass filter BEFORE demodulation is definitely a high pass filter.
My problem is that after this filter, a clamping circuit to some level (for example, ground level) is needed.
Another problem is frequency-dependent phase shift in the filter.
And yes, the filters we are discussing here are frequency domain filters, but oddly enough, the way we process signals is time domain, we use integrators which are time domain tools.

3.Regarding the schematics you provided: this is my version. I think it’s less complicated, although with the same phase shift drawbacks

I dont think you understand what the problem we are solving is .... The target spectrum can only be in response to the TX signal .... A simple bandpass filter you propose can remove low frequency components but it cannot remove modulation components due to noise / earth field etc.

Below are two spectrums from a simple PI operating at 1 khz. One of the circuits is exposed to 60 Hertz mains hum .... you can see it in the primary spectrum spike at 60 Hertz but you can also see modulation sidebands on the components of the RX signal. The filter you propose will never remove the sideband modulation from the RX signal components.

The demodulating bandpass filter I use does solve this problem.

[Aziraphel]

1.When designing a band-pass filter before demodulation, you need to find out what Spectrum the target signal has. When I mentioned the bandwidth, I meant the Spectrum of the target signal. If the target's constant time is 1u, then its Spectrum starts from 1 MHz and above.
Harmonics of the pump signal are the spectrum of the pump signal but not the target signal.

2.The filter you mentioned above as a band-pass filter BEFORE demodulation is definitely a high pass filter.
My problem is that after this filter, a clamping circuit to some level (for example, ground level) is needed.
Another problem is frequency-dependent phase shift in the filter.
And yes, the filters we are discussing here are frequency domain filters, but oddly enough, the way we process signals is time domain, we use integrators which are time domain tools.

3.Regarding the schematics you provided: this is my version. I think it’s less complicated, although with the same phase shift drawbacks

[moodz]

This may explain what is happening in the MAGPI 3 circuit for example. The amplifier A amplifies the signal from the RX coil however the only place the 1Khz LPF is in the feedback path ... so the noise ( earth field and EMI ) that is modulated onto the wanted signals ( target signals ) is not amplified by the amplifier. however the wanted signal is.

The noise at the output never falls to zero ( or there would be no feedback signal to remove the noise LOL ) However for an amplifier gain of say 1000 .. the wanted signals will be amplified by 1000 but the gain applied to the unwanted noise will be aproximately 1..... so 1 millivolt of wanted signal will be 1 volt but 1 millivolt of EMI will still be 1 millivolt at the output of the amplifier.

[moodz]

Although at second glance something worries me, something is wrong here. 10 us of response from the target corresponds to 100 kHz, on the other hand the cutoff frequency of the low-pass filter should be less than 1 kHz.
Which in turn provides 1 kHz bandwidth of the source signal at best.
Do you see the discrepancy?
Such a design will only detect large objects. V.L.O.

Not entirely sure what you are referring here .. however If you have a PI that works at 1Khz repetition ... lets say we have a equal turn off an on time ( ie square wave or symmetrical wave ) then the fundamental tranmit frequency ( ie lowest frequency the target sees ) is 1 khz ..... because we are transmitting a square wave ... we will also have every odd harmonic at a decreasing intensity ...
1 Khz ( fundamental )
3 Khz (3rd Harmonic )
5 Khz (5th Harmonic )

etc etc

So the "bandwidth" extends from 1 Khz up to the 10 odd harmonic or so.

But this is before you demodulate ...

How fast can you sweep a target ???

As the coil sweeps past the target ... it effectively modulates or distorts the TX signal ... the RX picks up this "distortion" and demodulates to recover a "target signal".
Going back to the question above .. how fast can you sweep the coil ??
The target modulation that needs to be detect probably tops out at 10 Hertz or so .... nowhere near 1 Khz.

The filter mentioned above is a bandpass filter BEFORE demodulation so its not a low pass filter.

Low pass filters might be applied AFTER demodulation.

The target responses quoted in literature ( wrongly IMHO as they imply the target response is like a radar response which is completely different physics ) are TIME DOMAIN responses ...
The filters we are discussing here are FREQUENCY DOMAIN filters ... not the same thing.

( note: the capitalisation is not yelling at you its to really emphasise the different terminology that applies in two different domains time and frequency )

As an example I have a PI that only examines one frequency ( about 20 Khz ) ... it easily detects a .05 gram nugget ...so the assertion that only large targets may be detected is not quite right.

You could write a whole book about this and two thirds would be taken up just explaining the signals before moving onto a practical circuit. For instance I did not touch on the phase changes that occur.

moodz

[Aziraphel]

Although at second glance something worries me, something is wrong here. 10 us of response from the target corresponds to 100 kHz, on the other hand the cutoff frequency of the low-pass filter should be less than 1 kHz.
Which in turn provides 1 kHz bandwidth of the source signal at best.
Do you see the discrepancy?
Such a design will only detect large objects. V.L.O.

Of course, there is another way - to transfer multiplication up, and to cut down all lower frequencies. (let say Fo=100 KHz with 150 - 250 KHz cutoffs?) the resulting signal wouldn't be resemble its original, it would be shape shifting transformation, and I think it will be difficult to distinguish a large object from a medium one.

[moodz]

Yes I have seen that article it explains basically how to solve the problem.
For modelling I used the free trial of MATLAB. But ltspice is fine for modelling the demodulating preamp.

[Aziraphel]

As soon as you mentioned it, I immediately remembered this article:

https://www.analog.com/en/resources/analog-dialogue/articles/synchronous-detectors-facilitate-precision.html#:~:text=Synchronous%20detectors%20can%20extract%20small,as%20the%20frequency%20approaches%20zero.

A long time ago I wanted to model this approach in LTspice using the ADA2200, but the spice model didn't have an .asy file and I never tried.
You are now awakening my interest in this approach, which I put on the back burner many years ago.

[moodz]

... But I don't like this solution, partly because turning the input on and off itself causes transients.
I wonder if you have found a simple and elegant solution to this problem?
....

You have made the key point ... if you consider the transmit and recieve waveforms from a discrete sinewave ( fourier ) analysis ...every waveform in the system no matter what amplitude or period can be decomposed to an additive series of sinewaves. (note : the keyword here is additive ).

If a non linear event occurs ( eg diode clamping, blocking, amplifier overload, switching ) which causes distortion of the waveforms then due to the non linearity a convolution ( mulitplication ) will occur and you end up with frequency products you proabably dont want..... note : here the unwanted event is multiplication not addition as previously mentioned.

So imagine we have a PI fundamental tx frequency of 1 kHZ pulses ... then the Fo of this pulse train is a 1 Khz sine wave ... imagine also that we have a 50 or 60 hertz mains interference signal of 1 millivolt in the recieve coil.
If a non linear event occurs ( eg preamp overload, clamping diode distortion etc ) then this will cause the 50/60 hertz mains frequence to be modulated onto the 1 Khz Fo signal. ( multiplication in the time domain cause frequency addition / subtraction AKA "difference" in the frequency domain. )
So now our 1 Khz Fo will have 3 components .... 1000 - 50, 1000 and 1000 + 50. ie 950 hertz, 1000 hertz and 1050 hertz ( for the 50 hertz case ).

We could call this "intermodulation".

So when you take a sample from the output of the preamp at 1 khz rate ... you will in fact then demodulate the difference signals from 1 Khz ( 950 and 1050 hertz ) .... and your sampled signal will have 50 hertz interference on it.

Because it would be very difficult to construct a filter that can separate 1050 and 950 hertz from 1000 hertz signals ... some PIs use sampling rates that are multiples of the mains rate to cancel the effect .. however this is not optimal and does not cancel other noises that may occur at the rx coil.

This same scenario applies to earth field ( modulated by swing speed ) for example or any other interfering signal at the RX coil.

This effect is the same in mono or DD PIs.

A solution I found was to demodulate the preamp output signal by mixing - mulitplying ( ie demodulating ) it with Fo ( eg 1 Khz ) and then LP filter the demodulated ouput and subtracting this from the input of the preamp thus cancelling noise below Fo ( 1 Khz ) by negative feedback.

In summary it could be called a demodulating negative feedback band pass amplifier.

In this way any convolution products produced by non - linearity in the frontend of the RX ( clamping / amp overload etc ) are substantially blocked and prevent modulation of the desired frequency components ( 1 Khz upwards ) from being distorted by earth field or mains etc.

You can see this preamp in the MAGPI 3 circuit. Even though this circuit uses active damping the preamp in that circuit will work on any PI including regular resistor type damping.

The solution has been out there for some time... unfortunately lots of folk have missile lock on expensive low noise amplifiers as the solution to their problem.

moodz.

[Aziraphel]

The solution is to put a HP filter on the input and disable the input during transients. Similar circuit solutions existed in the early days of pi metal detectors. But I don't like this solution, partly because turning the input on and off itself causes transients.
I wonder if you have found a simple and elegant solution to this problem?
P.S. and in the picture, I think, high-frequency noise is depicted against the background of a voltage drop in the coil. I don’t know how this relates to low-frequency noise filtering.

[moodz]

I see what you're hinting at, you're hinting at a DD type receiving coil located like in an IB metal detector.
If this is so, then another completely non-trivial task appears here: to balance the coils in a wide frequency range.

Not at all ... it also applies to mono coil PIs ....

The use of a "DC to daylight" bandwidth preamp in PIs is just plain wrong ( mono or DD etc ).

Modulation theory tells us that the lowest frequency in a PI recieve signal is the fundamental Fo of the TX ....

eg if you are pulsing at 1000 hertz .. then the the lowest frequency of interest at the RX is 1000 hertz.

A preamp that extends down to DC is plain dumb ... there is no DC from the target. ( or anything you want for that matter below 1000 Hertz for the example above ). The simplistic sampling approach that PIs use is just wrong ..( anyone with a radio background can see this straight away ).

All this talk of low noise opamps is wasted as most opamps are noisest below 1000 Hz ... but they do make a difference if your preamp passband extends down to DC ( which it should not .... hence the need for high pass filters / preamps. ) ... so two wrongs kind of equal a right in this case.

[Aziraphel]

The main problem with your filter though is that you are filtering the transmit step response ... You don't want to do that.

I see what you're hinting at, you're hinting at a DD type receiving coil located like in an IB metal detector.
If this is so, then another completely non-trivial task appears here: to balance the coils in a wide frequency range.

[moodz]

The main problem with your filter though is that you are filtering the transmit step response ... You don't want to do that.

[moodz]

Ah yes I understand what you are getting at. The high pass filter is responding to a step change in DC at it's input .... The solution requires a high pass filter that only responds to target decays as the Eddy current in the target is zero before the step an d the decays to zero in a short time eg few microseconds. The high pass filter input capacitance must not be charged by the input step as the step is a tx signal not a Target signal.

[Teleno]

The average DC output of the waveform of your filter ..if it's working correctly should be zero. The output looks like it has a DC imbalance. I realize you have a single supply amp but shouldn't you have a vcc/2 reference to set the DC bias?

reference is biased at 1.57V.

The problem is, as I pointed in a previous message, is that the step response of a high-pass filter is a decaying exponential, and much worse than the coil's for a 500 Hz cut-off frequency.

[moodz]

The average DC output of the waveform of your filter ..if it's working correctly should be zero. The output looks like it has a DC imbalance. I realize you have a single supply amp but shouldn't you have a vcc/2 reference to set the DC bias?