DEEPER PI DETECTION DEPTH

[Tinkerer]

Pretty much every real component has resistance (except for those damn superconductors...) and having no capacity to enter the ESR of the parasitic capacitance makes the inductor model useless in any switched inductor application, and probably introduces significant inaccuracies in EVERY application. Sure we can work around and make a discreet model but we shouldn't have to.

Yes its probably not a large ESR but there's an infinite gulf between very small and zero. You can tell how unrealistic it is because its basically impossible to switch the mosfet in this model without getting massive drain-gate shoot through.

The Drain-Gate shoot through with real mosfets, when using high A coil currents, is exactly the problem that I have. To me, used to real circuits, the simulations look quite good.
Of course, I still have a lot to learn about simulations, but most of the time when there are differences, I can find the cause in my own mistakes.

Tinkerer

[Aziz]

The Dark Pulse Theory

Hi guys,

the Dark Pulse Theory™ ( nice name - isn't it? ) states, that there is no free lunch for you! I'm so sorry for this bad news. But it is very overdue now.

The EM pulse has a spectral energy, which is "seen" by the inductively coupled target and is responding to this pulse accordingly. The TX -> Target -> RX system has a frequency response as well (well, we can see it as a multi-stage filter response).

The nature offers you the maximum possible available information by physics law. All you can do is making either worse or loosing/throwing away that amount of available information.

You can easily see the proof of this statement by making a simple AC analysis (frequency response analysis) for different time constant (TC) targets. I have made this for four different TC targets for you. For your convenience, I'll put the spice file and you can play with it.

Now looking forward to the interesting discussion.

Cheers,
The Dark Aziz

[Midas]

Since LTSpice was developed primarily to analyze switching power supplies, it would be surprising that the inductor model would be so inadequate. Maybe this case should be posted on the LTSpice group to see if this is a known problem or if they can shed light on our simulation. Is it possible that: huge spikes exist in real life but are absorbed in other ways; or the spikes we see are not the fault of the inductor model but the other components around it (like ideal wires, mosfet model, etc.)?

-SB

Indeed, I wouldn't have expected it to be so fundamentally flawed either. Thinking about it a bit more the neglected resistance isn't so small either, since the parasitic capacitance is coupling to the mosfet drain through the coil it has go through at least some of the coils resistance. Also there should be resistance between the coil and its parasitic capacitance, otherwise the coil could potentially oscillate forever, which of course isn't possible.

With those thoughts in mind I made the following discrete model. Its hard to get your head around how much resistance there should be between all the connect points for a 15 ohm coil but its a start. It still has some weird current spike during switch on but the switch off is a lot better. Also removing the damping resistor doesn't result in the kind of oscillations I would expect so its obviously still not perfect.

Edit: Yeah actually.. its electrically identical to one of Simons earlier post. Lol I think we can burn this extra wheel.. never mind its been fun.
The question over what proportion of the total resistance is taken by each component remains though. It may not be 50/50.

[simonbaker]

Pretty much every real component has resistance (except for those damn superconductors...) and having no capacity to enter the ESR of the parasitic capacitance makes the inductor model useless in any switched inductor application, and probably introduces significant inaccuracies in EVERY application. Sure we can work around and make a discreet model but we shouldn't have to.

Yes its probably not a large ESR but there's an infinite gulf between very small and zero. You can tell how unrealistic it is because its basically impossible to switch the mosfet in this model without getting massive drain-gate shoot through.

Since LTSpice was developed primarily to analyze switching power supplies, it would be surprising that the inductor model would be so inadequate. Maybe this case should be posted on the LTSpice group to see if this is a known problem or if they can shed light on our simulation. Is it possible that: huge spikes exist in real life but are absorbed in other ways; or the spikes we see are not the fault of the inductor model but the other components around it (like ideal wires, mosfet model, etc.)?

-SB

[Midas]

Ok, my slowness, I didn't get that, thanks. Yes, it seems we may have a circuit where we need a more sophisticated model for an inductor than the LTSpice model. Of course you can keep going, dividing the coil into more and more elements, have to stop somewhere...

On the other hand, does a real coil's parasitic capacitance have an ESR? What physically would cause that? Wouldn't the electric field just jump across the windings with almost no resistance? Or are we talking about a tiny-tiny but significant amount of resistance?

Who knows, maybe those 60A glitches in your sim exist in real life, but our measuring equipment can't pick them up, and they too transient to matter?

-SB

Pretty much every real component has resistance (except for those damn superconductors...) and having no capacity to enter the ESR of the parasitic capacitance makes the inductor model useless in any switched inductor application, and probably introduces significant inaccuracies in EVERY application. Sure we can work around and make a discreet model but we shouldn't have to.

Yes its probably not a large ESR but there's an infinite gulf between very small and zero. You can tell how unrealistic it is because its basically impossible to switch the mosfet in this model without getting massive drain-gate shoot through.

[simonbaker]

It would be a good start if we could get a simulation of the target response and get this response into a RX coil and from the RX coil into the preamp.

Not to forget that the coil arrangement should be induction balanced.

Then we could really compare the real circuit with the simulated circuit and fine tune them both.

Let's do it!!!!!

Tinkerer

Does anyone have a front end simulation of the PI you are working on?

If not, basically just need to slap in the RX coil, add some parasitic components, add the preamp circuitry, to your existing sim. Delete the unused target coupling coefficients (or delete the unused targets completely).

Then start tweaking the coupling coefficients of the TX coil to RX coil and target to RX coil.

Of course good target models are desirable, but start with what we have.

-SB

[Tinkerer]

Let's get a full simulation including the output of the preamp that agrees with the Carl's real tests showing 15% more output for the flat top charging pulse. I'd like to see what accounts for that extra 15%.

Up for it, Tinkerer?

-SB

It would be a good start if we could get a simulation of the target response and get this response into a RX coil and from the RX coil into the preamp.

Not to forget that the coil arrangement should be induction balanced.

Then we could really compare the real circuit with the simulated circuit and fine tune them both.

Let's do it!!!!!

Tinkerer

[Aziz]

How do you capitalize on that wide-band information? If you don't use it wisely, you are better off using narrow band techniques because you are picking up a lot more noise with wide-band, it would seem.

-SB

Hi SB,

that's a good question. Any ideas?

I for one would take all possible spectral energy of the response. And not all spectral bands do have the same noise level.

BTW, the VLF principle is taking an huge advantage of the narrow band (high gain at resonance and high suppression on the other frequency sides). And it's so efficient.

Despite of the fact, that a PI detector performs a wide band principle, it throws away a lot of the spectral response however. All the super high dI/dt (100 kA/s .. 1 million A/s) is mostly wasted.

Aziz

[simonbaker]

I don't think Midas is disputing that the LTSpice inductor model is behaving the same as an identical discrete model. His comment is referring to the fact that the parallel capacitor in the LTSpice inductor model has zero ESR. In other words there's no way to define an equivalent series resistance for the parallel cap. Hence the glitches. If you build the same network with discrete parts, you can then include the missing ESR. The best option would be to use a parameterized subcircuit for an air coil in place of the inductor model.

Ok, my slowness, I didn't get that, thanks. Yes, it seems we may have a circuit where we need a more sophisticated model for an inductor than the LTSpice model. Of course you can keep going, dividing the coil into more and more elements, have to stop somewhere...

On the other hand, does a real coil's parasitic capacitance have an ESR? What physically would cause that? Wouldn't the electric field just jump across the windings with almost no resistance? Or are we talking about a tiny-tiny but significant amount of resistance?

Who knows, maybe those 60A glitches in your sim exist in real life, but our measuring equipment can't pick them up, and they too transient to matter?

-SB

[Qiaozhi]

How do you come to that conclusion? The parasitic cap parameters are there to be specified by you if you want to. It's your choice.

You just have to know when your are comparing apples to oranges. If one coil has the parasitic parameters specified, and the other coil does not and you instead include them externally, you can no longer just compare the "coil" currents and expect them to be equal, because one coil includes the parasitic components, and the other doesn't, right?

If instead you compare the two "network" currents, they match up nicely, showing LTSpice is working as expected.

Is there any mystery? Isn't that the way LTSpice should work?

-SB

I don't think Midas is disputing that the LTSpice inductor model is behaving the same as an identical discrete model. His comment is referring to the fact that the parallel capacitor in the LTSpice inductor model has zero ESR. In other words there's no way to define an equivalent series resistance for the parallel cap. Hence the glitches. If you build the same network with discrete parts, you can then include the missing ESR. The best option would be to use a parameterized subcircuit for an air coil in place of the inductor model.

[simonbaker]

Morale of the story?

We need to design a simulation target model that can reproduce all the variations that real targets could present.

Big job.

Suggestions of how to design this universal simulation target model?

Tinkerer

That would be so valuable; you on the money!

We first need some kind of data about real targets.

I would propose using our best, widest-band PI machine to actually bang away on a whole bunch of real targets.

Capture the responses with a good digital storage scope. Also capture the stimulating pulse.

Then we have to divine a general model for a target, and probably a linear dynamic system is our best start. We'll pick some arbitrary order like 5, and then try to estimate the system parameters from the response and stimulating signal using fancy shmancy optimal estimation theory.

Ooboy, that is a big job. Ok, let's just start by capturing some real data. Then we could at least tinker with target models heuristically and have something to match against.

-SB

[simonbaker]

Well, it's coming into the region of the VLF detector principles. Except, we have a wide band frequency response.

Aziz

How do you capitalize on that wide-band information? If you don't use it wisely, you are better off using narrow band techniques because you are picking up a lot more noise with wide-band, it would seem.

-SB

[simonbaker]

I though we had already determined the answer to this particular question.
Carl did some comparisons on a real circuit, and posted the results in #137 ->
http://www.geotech1.com/forums/showp...&postcount=137

Let's get a full simulation including the output of the preamp that agrees with the Carl's real tests showing 15% more output for the flat top charging pulse. I'd like to see what accounts for that extra 15%.

Up for it, Tinkerer?

-SB

[simonbaker]

Good work Qiaozhi. You've nailed down the main issue, zero ESR parasitic capacitance in the inductance model. Pretty sloppy work on LT's part.

How do you come to that conclusion? The parasitic cap parameters are there to be specified by you if you want to. It's your choice.

You just have to know when your are comparing apples to oranges. If one coil has the parasitic parameters specified, and the other coil does not and you instead include them externally, you can no longer just compare the "coil" currents and expect them to be equal, because one coil includes the parasitic components, and the other doesn't, right?

If instead you compare the two "network" currents, they match up nicely, showing LTSpice is working as expected.

Is there any mystery? Isn't that the way LTSpice should work?

-SB

[Carl-NC]

ok, the glitch problem isn't a real problem. Let's leave it and focus to the interesting dark matter.

No, not a problem, but I like to know!

The flat-top (t-on > TC) versus (almost) "linear" current ramp (t-on < TC).
What is better and why?

This is another "big coil vs small coil" comparison. Each has advantages. The linear ramp has more useful "on" time info. Flat-top has an advantage in magnetically viscous soil. Either one can be done in short TX times for high PPS rates. We're time domain here, we can do whatever we want. So why not do both?