lobo improved

[simonbaker]

Different designed preamp.
Lobo is designed somewhere at end of 80-ties. I think the desiner is not Jack Gifford because the preamp circuit of Lobo differs significant from other primitive preamp circuits used by Tesoro. The differences are:
1. The preamp is formed as three stage amplifier. Other Tesoro preamps use only a single stage.
2. The resistors of first stage are designed for low noise. The resistors in other Tesoro preamp circuits have much larger resistances, ie they generate more noise.
3. RX coil is connected to form second order band pass filter. In other Tesoro circuits, RX coil forms second order low pass filter. That means they can not suppress enough the audio frequency (330 - 550Hz) and
third harmonic of mains frequency (150 or 180Hz).
However:
1. Despite its three stages, the preamp operates with low gain. I see something is written for gain in the circuit diagram attached below.
2. Resistors and RX coil have low resistances, but this is useless because opamp generates spectral noise density more than 1kohm resistor. That means the designer can decrease the weight and price of RX coil without increasing input noise..
3. The tuned circuit (band pass filter) is damped with 6kohm resistor. That means increased noise because a) the damping resistor also generates noise and b) damping forms wide band input (noises are proportional to bandwidth)
Let we make analysis, revision and redesign of this circuit.

In other Tesoro circuits, RX coil forms second order low pass filter

I don't see how TGSL RX circuit is low pass filter -- isn't it a legitimate LRC tank circuit, with center freq around 16.5 kHz or something like that? The lobo circuit here seems to show a very over-damped RX coil tank with that 6.19K resistor in there, with main bandpass due to op amps.

-SB

[Davor]

It is in case of aperiodic operation. Just for a moment imagine there is no capacitor to make the coil resonant.

Hi Davor:

What do you mean by aperiodic input circuitry?

Would you also compare this frontend to TGSL/IGSL (your balanced version) and explain differences and advantages?

Regards,

-SB

First off, my IGSL frontend is not much of a solution as I realised recently. Center tap coil is what makes it work fine, but otherwise the frontend's CMMR is poor. It works well without shielding, but mostly because of the center tap.

So I reverted to learning the basics of the transducers interfacing and found that the line to follow is the microphone preamp art.

By aperiodic I mean lack of resonance near the frequency of interest in a coil and the immediate circuitry. Resonance works as an impedance transformer, and most efficiently at the resonant frequency of a tank. That very situation is not used because of steep phase change. When used at frequencies lower than resonance you have a modest phase shift, which is ~the same as if you use low input impedance and no resonance at all, e.g. aperiodic. While resonant operation is superb for op amps that suck at noise, low impedance is tolerant for wild variations in Rx coils' inductance, and hence much better for an amateur coil builder.

This particular Lobo solution is asymmetric and optimised for low noise. With op amp at hand, it is overly optimised, but that's good news for the Lobo owners - they can have a better rig by simply replacing the input opamp.

Asymmetric inputs do not sort out the common mode contributions (electrostatic) from the differential ones (magnetic induction), and apart from the coil shielding there is no way of suppressing the famous "wet grass" effect. Differential inputs are much better at that, they suppress the common mode component.

With nowadays op amps boasting with low noise, it is much simpler and phase-wise predictable to forget about the resonant tank and simply go directly to a low impedance frontend. That would allow you to use coils in wild ranges of inductances, and at the same time at a wide range of frequencies. In case of a balanced input, even without the center tap you'd be able to use unshielded coils. That's my current ideal.

[Qiaozhi]

By aperiodic I mean lack of resonance near the frequency of interest in a coil and the immediate circuitry.

The use of the word "aperiodic" is not strictly correct in this instance. Aperiodic means a non-periodic (not occurring at regular intervals) signal. It is a more appropriate description of the receive signal on a PI detector, which is damped to prevent oscillation.

For your case, it would be better to use the term "off-resonance".

[Davor]

True, because every coil has resonance somewhere unless it is damped really good. So for this Lobo thing you can say "way off-resonance"
However, what I have in mind is really an aperiodic frontend. Imagine a coil. Now imagine it picking up some magnetic flux at some arbitrary frequency, or better say any frequency. Now, extend your imagination to some Rx attached to it. It has some input impedance, and in case this impedance is lower than the critical damping - you have a real aperiodic frontend, the one that wouldn't even ring.

Now, what happens at resonance? From the coil's point of view in a resonance it sees the lowest possible impedance that is limited by coil's resistance, and that is the point where you have the best possible energy transfer - it is the voltage source after all. Because in resonance we have a sharp phase transition and we can't use it, I can attach a coil to a low impedance Rx instead and have frequency independent signal transfer, spoiled only by the coil's inductance. That kind of aperiodic.

In short - I can't use resonance, so why bother with resonance at all?

[golfnut]

Saying resonance isnt clear.

Parallel resonance the one we use - High Z case- (Not series resonance (Low Z case))- the impedances of the L and the C are the same and are HIGH. Opposite polarity or sign.

Giving a tank Z at resonance of L/CR (R is the resistance of the L)

You can broadband the tank with extra shunt or Parallel R makes it give less phase change with ground proximity - less touchy in use.

Yuo can make a tank resonate in a parallel way - without a cap - you just need lots of winding to winding capacitance and it will work

S

[simonbaker]

True, because every coil has resonance somewhere unless it is damped really good. So for this Lobo thing you can say "way off-resonance"
However, what I have in mind is really an aperiodic frontend. Imagine a coil. Now imagine it picking up some magnetic flux at some arbitrary frequency, or better say any frequency. Now, extend your imagination to some Rx attached to it. It has some input impedance, and in case this impedance is lower than the critical damping - you have a real aperiodic frontend, the one that wouldn't even ring.

Now, what happens at resonance? From the coil's point of view in a resonance it sees the lowest possible impedance that is limited by coil's resistance, and that is the point where you have the best possible energy transfer - it is the voltage source after all. Because in resonance we have a sharp phase transition and we can't use it, I can attach a coil to a low impedance Rx instead and have frequency independent signal transfer, spoiled only by the coil's inductance. That kind of aperiodic.

In short - I can't use resonance, so why bother with resonance at all?

Thanks for explaining your use of term "aperiodic", I get it. I agree with Qiaozhi that the word seems not what you mean; however, I would choose "over-damped" or "low Q" or "non-resonant" or "wide-band" coil circuit. "Aperiodic" to me describes waveforms that are not periodic. Just a language thing.

I would assume the purpose of RX resonance would be to increase S/N, especially regarding EMI; the synchronous detector probably can alias wide-band signals into its detection band, though maybe not as much as I think.

Also, RX resonance can greatly increase signal gain relative to resistance noise, etc. Of course the trade-off is a loss of phase stability relative to component tolerances, as you noted. So did Tesoro find a useful compromise with the simple RLC tank?

The TGSL/IGSL operates so far off resonance, I wonder if there really is an advantage in S/N over a completely "non-resonant/wide-band" coil circuit. You seem to think not, and I can believe you may be right. We should be able to do a noise analysis fairly simply with LTSpice. And it would be useful to include some broadband EMI noise into the coil as well.

Also bear in mind that the Lobo op amp bandpass filter will also make phase shifting for off center frequencies -- why don't you object to that?

I have always wanted to try a high-Q, "on-resonance" RX circuit design to optimize S/N, even though very difficult to achieve phase stability. Put the engineering effort into stabilizing the phase and reap the benefits of the superior S/N. S/N is really the what we are trying to achieve (until we hit the "ground noise" limit).

-SB

[Davor]

Right to the point.
There are a few points to clear before continuing to the specifics.

First off, a coil as a magnetic pickup is in reality a LPF device. It produces voltage. When observed without a parallel capacitor it is a wideband device with a high cutoff that depends upon a load resistance, or as we observe it - a preamp input impedance.
You can see that cutoff is higher as the load impedance is higher, but at cost of somewhat higher noise.

While you can spoil S/N with too high load impedance, you can screw it completely by too low impedance. In microphones world the rule of the thumb suggests using loads that are 10 times the coil resistance and up, but not too much because of the noise.
These cases are shown in pictures of the non-resonant circuit. One shows AC and the other shows noise. It is important to note that real effects of the noise must be normalised for the gain/loss because they will surely reflect at the preamp output. E.g. half the noise for a circuit that gives -6dB against the other circuit results in equal final result - you must pump gain and the noise for that 6db, and puff goes your noise advantage.

In case of resonance, you obtain some virtual gains because of the impedance transformation effects, and to fully grasp the mechanism follow the red line (1k) in non-resonant and the resonant cases. Without additional noise you get to the very same normalised noise performance this way or another.

The whole difference is in preamp noise. In case your opamp has 4nV/sqrt(Hz) it means that it is well matched with a 1 kohm equivalent resistance noise. Your system noise will not go below that, and you can optimise your frontend for that.

Every preamp can be seen as an infinite impedance voltage sensor with a shunt. Every coil can be seen as an auto-transformer with 1mH primary, and a secondary/tap at desired inductance. Point to note here is that with more inductance you gain more voltage, but phase and noise get worse. Goal is to reach maximum voltage at exactly the noise equivalent to the preamp input noise.

It goes like this...
Your opamp is noise limited to 1kohm equivalent, or 4nV/sqrt(Hz), and your working frequency is 10kHz. You want your phase to remain under 18° shift, and your input is configured as Lobo's with noninverting input shunted by some arbitrary but not too small resistor to keep self resonance and impulse phenomena at bay, say 10k (ten times 1k, rule of thumb thing).
18° shift is found at 1/3 frequency of -3dB point (12° is at 1/5), so it must be at 33kHz. At -3dB point the shunt resistance and coil reactance are the same, and a calculator gives 48mH. Upon checking the noise - in simulation of course, I find a tad below 4nV/sqrt(Hz).
So I get everything WITHOUT resonance. Largest possible input voltage at desired noise level. No resonance troubles.

Please don't get me wrong, but in case of VLF metal detectors resonance is so overrated.

Here go a few examples non-resonant and resonant, AC and noise

[simonbaker]

Right to the point.
There are a few points to clear before continuing to the specifics.

First off, a coil as a magnetic pickup is in reality a LPF device. It produces voltage. When observed without a parallel capacitor it is a wideband device with a high cutoff that depends upon a load resistance, or as we observe it - a preamp input impedance.
You can see that cutoff is higher as the load impedance is higher, but at cost of somewhat higher noise.

While you can spoil S/N with too high load impedance, you can screw it completely by too low impedance. In microphones world the rule of the thumb suggests using loads that are 10 times the coil resistance and up, but not too much because of the noise.
These cases are shown in pictures of the non-resonant circuit. One shows AC and the other shows noise. It is important to note that real effects of the noise must be normalised for the gain/loss because they will surely reflect at the preamp output. E.g. half the noise for a circuit that gives -6dB against the other circuit results in equal final result - you must pump gain and the noise for that 6db, and puff goes your noise advantage.

In case of resonance, you obtain some virtual gains because of the impedance transformation effects, and to fully grasp the mechanism follow the red line (1k) in non-resonant and the resonant cases. Without additional noise you get to the very same normalised noise performance this way or another.

The whole difference is in preamp noise. In case your opamp has 4nV/sqrt(Hz) it means that it is well matched with a 1 kohm equivalent resistance noise. Your system noise will not go below that, and you can optimise your frontend for that.

Every preamp can be seen as an infinite impedance voltage sensor with a shunt. Every coil can be seen as an auto-transformer with 1mH primary, and a secondary/tap at desired inductance. Point to note here is that with more inductance you gain more voltage, but phase and noise get worse. Goal is to reach maximum voltage at exactly the noise equivalent to the preamp input noise.

It goes like this...
Your opamp is noise limited to 1kohm equivalent, or 4nV/sqrt(Hz), and your working frequency is 10kHz. You want your phase to remain under 18° shift, and your input is configured as Lobo's with noninverting input shunted by some arbitrary but not too small resistor to keep self resonance and impulse phenomena at bay, say 10k (ten times 1k, rule of thumb thing).
18° shift is found at 1/3 frequency of -3dB point (12° is at 1/5), so it must be at 33kHz. At -3dB point the shunt resistance and coil reactance are the same, and a calculator gives 48mH. Upon checking the noise - in simulation of course, I find a tad below 4nV/sqrt(Hz).
So I get everything WITHOUT resonance. Largest possible input voltage at desired noise level. No resonance troubles.

Please don't get me wrong, but in case of VLF metal detectors resonance is so overrated.

Here go a few examples non-resonant and resonant, AC and noise

Thanks for the analysis, that's what I'm looking for. I'll need to study it for a while.

-SB

[Davor]

Just a hint, you can scale things from the known to the desired. Scaling is linear except for voltage/inductance where voltage rises with square root of inductance. Using Lobo's values you may reach to conclusion that, keeping the coil as is, the ideal op amp would have ~2.7nV/sqrt(Hz) of voltage noise. That would improve good old Lobo by ~10dB by mere op amp swap. Not bad at all.

[simonbaker]

Just a hint, you can scale things from the known to the desired. Scaling is linear except for voltage/inductance where voltage rises with square root of inductance. Using Lobo's values you may reach to conclusion that, keeping the coil as is, the ideal op amp would have ~2.7nV/sqrt(Hz) of voltage noise. That would improve good old Lobo by ~10dB by mere op amp swap. Not bad at all.

Thanks Davor - can you attach your LTSpice simulation ".asc" files for convenience?

-SB

[Davor]

It is not much of a simulation, but here you go...

[simonbaker]

It is not much of a simulation, but here you go...

Thanks.

Without doing any analysis yet, I'm thinking more about S/N and usefulness of "on-resonance" design. I may well agree with your conclusions, because my preliminary intuition is that although a high-Q resonant RX should indeed increase S/N regarding EMI, resistor noise, and op amp noise when looked over a wide spectrum, the fact may be that the Synchronous Detector forms such an extremely narrow filter, that the S/N within that narrow band simply is not affected much.

In other words, a resonant RX tank increases S/N by suppressing noise outside the RLC filter band. (That includes EMI noise and any resistor noise or op amp noise that is coupled to the RX tank.) However, the SD band is tiny and well inside the RX tank band. So there is no change in S/N in the SD detection band.

In order to conclude there is no advantage to a high-Q resonant RX tank, one assumes that all resistor noise has the opportunity to resonate with the RX tank -- my question is: do some resistors, such as the op amp feedback resistor, add noise that is not "coupled/amplified" by the RX tank. If those resistors add noise that is not coupled to the RX resonance, then we might achieve some gain over that noise. Also, some op amp noise probably is not coupled to the RX tank -- likewise, we should achieve some gain over that noise.

However, because of the availability of very low noise op amps these days, the improvements due to a high-Q resonant tank could very well be too small to overcome the huge hassle of dealing with the tempermental phase caused by a high-Q RX tank. In that case, I would agree, don't waste time playing with high-Q RX tank, and probably no point in having any capacitor at all. Well, perhaps there is a small purpose of having a capacitor simply to prevent a coil self-resonance from creating a large signal at some higher frequency which might so something weird (overdrive the inputs; alias the SD, etc.).

Those are my impressions without doing serious analysis yet.

-SB

[Davor]

You are right in most of the conclusions, and in most details, but there are some small bits that spoil everything.
First and most important is resonant coil being a very narrowband. It isn't. It is a low pass filter, with a ripple prior to the falling slope. It is sharp and pointy at the resonance, but you can't touch it, it is off limits, too steep and unpredictable phase change. So the only mechanism attributed to the metal detecting world is impedance transformation at off resonance spot, period. All other goodies can be achieved using far more predictable RC filters, just as is done for Lobo.

There are ways of approaching the theoretical noise limit of a coil itself, and this Lobo approach is darn close.

[simonbaker]

You are right in most of the conclusions, and in most details, but there are some small bits that spoil everything.
First and most important is resonant coil being a very narrowband. It isn't. It is a low pass filter, with a ripple prior to the falling slope. It is sharp and pointy at the resonance, but you can't touch it, it is off limits, too steep and unpredictable phase change. So the only mechanism attributed to the metal detecting world is impedance transformation at off resonance spot, period. All other goodies can be achieved using far more predictable RC filters, just as is done for Lobo.

There are ways of approaching the theoretical noise limit of a coil itself, and this Lobo approach is darn close.

No, I said Synchronous Detector is very narrow band -- I agree resonant coil is not narrow band -- that is my point -- even if you raise the Q of the resonant coil, you do not improve the S/N within the very narrow band of the SD. The EMI and even the coupled resistor noise are also boosted by the resonance within the SD band. Only if you consider a broader band than the coil RLC band do you achieve improved S/N metrics, I believe. I also think of resonant coil as band pass, not low pass -- why do you say low pass?

So yes, you can do the impedance transformation thing to at least optimize your resistors to the op amp you choose for noise, and perhaps choose the best coil inductance (there is a question - what do you think about RX inductance?). Now suppose we go with an "instrumentation amp" type of preamp -- does that remove the resistor optimizations because we have infinite impedance? Or is that just inherently more noisy to begin with?

As for operating at the resonant frequency of a high-Q coil tank -- I agree it is nasty, but if it had a vastly improved S/N, I would say it would be a challenge worth trying. However, I'm agreeing, based on my rough thinking, that it doesn't seem to help much because of the extreme narrow SD bandwidth, which seems to be the determining factor. But I'll continue to think about this S/N subject and am interested in the designs you are considering.

So I guess the reason you like the Lobo front end is because of the "over-damped" coil, even though it is a second order filter, (would you rather have a simple low-pass coil?). Do you like the high-impedance op amp input, or it is just a necessary evil to achieve the over-damped coil?

Regards,

-SB

[Davor]

Why a coil is a low pass filter, easy, it produces voltage from the applied flux, and does that in perfect order up to a corner frequency where series inductance becomes significant factor.

Instrumentation amplifier is an option, however, those are seldom optimised for low input impedance operation. If you happen to find some instrumentation amplifier boasting with low noise, it automatically means it is optimised for low input impedance as well. Point is that low voltage noise comes together with high current noise, and such solution is not optimal for high impedance operation. Otherwise it is a perfect choice for differential inputs.

For a perfect example, look at Lobo's frontend. All the gain setting resistors are in the inverting branch, while coil signal is introduced to the non-inverting input. The gain setting resistors are of low resistance value, hence optimised for very low noise. Such low resistor values are not seen in instrumentation amplifiers. In fact, those Lobo resistors are set for a much better op amp than you can find in a commercial Lobo.

Regarding bandwidth, remember that your system noise can get as close to the thermal noise floor only in a case your frontend does not introduce too much of its own noise. Out of that sum of natural noise and the introduced noise, your Rx will sort out the narrow bandwidth signal/noise, and ... this is important ... the noise floor there in the narrow band path will be ruined by the same number of dB as your frontend is responsible for.

So, in case your frontend introduces 20dB of noise on top of the thermal noise floor, your narrow band gain block will also have noise risen by the very 20dB on top of the narrow band noise floor.