DEEPER PI DETECTION DEPTH

[simonbaker]

It seem only appropriate to continue here where we left some time ago. Now this is big. I present you a LTspice simulation of a logarithmic weighted PI Rx with dead time of less than 2 microseconds. Furthermore, its output is logarithmically compressed so there will be much fewer bits required for ADC, where required.

There are some additional repercussions of this approach that enable some very interesting solutions to discrimination (say, you may play with C3 and observe consequences), but finally it is THE weighted solution I was mumbling incoherently for the past few months. Also, it is an inverting solution and thus the signal decays opposite to the "Surf" solution.

On the schematic you'll find two switches that are used to divert signal to a common node, and prevent two Rx-es to interfere with each other, and to draw output in nice colours. The opamp with a bunch of diodes is a log detector: nothing fancy, but works as a horse. I'm aware this is going to perform a bit differently with a real world opamp, but here you go - a place to start experimenting. This can sample very early.

Patent trolls - keep off!!!

Nice creativity, good contribution!

-SB

[Aziz]

It seem only appropriate to continue here where we left some time ago. Now this is big. I present you a LTspice simulation of a logarithmic weighted PI Rx with dead time of less than 2 microseconds. Furthermore, its output is logarithmically compressed so there will be much fewer bits required for ADC, where required.

There are some additional repercussions of this approach that enable some very interesting solutions to discrimination (say, you may play with C3 and observe consequences), but finally it is THE weighted solution I was mumbling incoherently for the past few months. Also, it is an inverting solution and thus the signal decays opposite to the "Surf" solution.

On the schematic you'll find two switches that are used to divert signal to a common node, and prevent two Rx-es to interfere with each other, and to draw output in nice colours. The opamp with a bunch of diodes is a log detector: nothing fancy, but works as a horse. I'm aware this is going to perform a bit differently with a real world opamp, but here you go - a place to start experimenting. This can sample very early.

Patent trolls - keep off!!!

Hi Davor,

this kind of log-amp is horrible noisy and unstable (temp drift). Nevertheless, it could still be sufficient for our purpose. Someone should try this log-amp and report the results here.

Cheers,
Aziz

[Davor]

I know, but with all other effects in play the temp drift is the least of my worries. Besides, dedicated multi-decade log amplifiers are incredibly expensive, and same as this one must hit the noise floor ... eventually.
Using BJT in a feedback loop is not famous for stability so simple diodes were a logical choice. Noise is also not high on my worries list because it seem as if the most interesting part of the curve is waaaaay above 1mV of a PI decay.
I think the most of the minor problems will be tackled with the tricks already in our toolbox, and we don't need to call them different names. EF can stand for a slowly decaying value past the log slope knee (the ski-like bend) e.g. low signal + offset + noise. That should keep it a bit more stable.
Another possibility may be provided by mere window comparators, and a set of voltage values, hence time differences. That could do as an ADC in a way - a simple counter.

It can work as well with bipolar pulses with only minor changes, and i wonder what it will be like with TEM pulses.

[Aziz]

Hi all,

let's apply the Occam's Razor to the log-amps (see more http://en.wikipedia.org/wiki/Occam's_razor ).

"Occam's razor (also written as Ockham's razor, Latin lex parsimoniae) is the law of parsimony, economy or succinctness. It is a principle urging one to select among competing hypotheses that which makes the fewest assumptions and thereby offers the simplest explanation of the effect."
(Quote from Wikipedia)

Question:
Why it (the log-amp) hasn't been used in the top performing detectors yet?

Hypotheses (not complete):
1. Engineers are bloody dumb and do not know that log-amps exist (I bet a lot of them are really dumb).
2. Engineers do know the log-amp but can not migirate the log-amp into the MD application.
3. Engineers do not believe, that it has any potential ("gut feeling").
4. Engineers do not like new ideas ("Never touch a running system!!!"). Why bother with speculative designs, if there is a working solution already?
5. Engineers don't know deep math except the basic calculus (*,/,+,-).
6. Engineers do not really exist. So everybody who is working in the art is a bloody amateur.
7. No one has tried it before.
8. There is a reasonable ground not to use it.
9. There isn't a reasonable ground not to use it.
10. The weather was fine and the girls were very sexy. Engineers didn't have time to look at it.

So, why the f... h... hasn't the log-amps been used yet?

Aziz

[Tinkerer]

Years ago, I used this pre-amp. It was giving good results. However, because of latent temperature drift etc. I abandoned the design.

The Opamp does not saturate, because of the soft clipping by the transistor feedback. The negative TX pulse that we do not need, is clipped.

The response is somewhat logarithmic. The gain is enhanced for the low voltage here we need it.
The higher voltage has less gain and still higher, the signal is soft clipped so that the opamp does not saturate.

Possibly the latent problems could be solved and then it would make a good alternative.

Tinkerer

[moodz]

Log amps = Bog amps

Myth Buster
Myth 1 : log amps are expensive ..

You can buy the AD8307 92 db logamp DC to 500 Mhz on ebay for $2 each. They are excellent ... at least they are excellent when used for right application.

Myth 2 : log amps are useful for sensitive metal detectors ..

Log amps are used for things like Radars where your reciever is very sensitive and you want to measure the difference between a very weak signal and very strong signal ( like 10000 times as strong ) Problem is in metal detectors you want to measure difference between 1.0000 and 1.0001 signal .... log amps are wrong amp for this application LOL.

comment .... even in some radio applications where they used to use log amps they now use very high resolution ADC ....since you can have over 100 db of dynamic range ( ie smallest to biggest signal ) ... you dont need a log amp.

moodz

[Altra]

I had saved this post from Eric Foster. It came from the the PI forum on Findmall.
Maybe helpful comming from someone who has actually tried a logamp?

Quote.
Hi Randy and All,
I would be very interested in the log amp you mentioned. Do you have the part number? This is a good way to go, either for DSP or normal analogue processing. Although linear amps are invariably used, the signal strength varies so dramatically with object range, as a result of the sixth power law, that they rapidly run into saturation. The other benefit of a log amplifier is that it is possible to enhance the differences in the decay curve between ferrous and non-ferrous metals. A non-ferrous object has a decay that is exponential in shape for later times i.e. times longer than one time constant. Run this through a log amp and you get a linear decay. The decay from a ferrous object when run through a log amp is anything but linear. If you take a long sample window of the signal and differentiate it, then non-ferrous objects come out as a constant reading, depending on the slope of the linear decay, which depends on the time constant. Hence an indicator could be calibrated for different coins etc. Ferrous objects could be distinguished by the fact that the reading changes during the sample window as the slope changes. Tomorrow I will try and post some log plots of different objects which were taken in 1968 using a wide band log amplifier (yes it was an IC!) which clearly shows the differences which one would be hard pushed to notice on a linear display.
So why isnt this system currently used? The main bugbear is noise. As the signal decays into the noise, the gain of the log amp goes up and with no signal the output is flapping about all over the place. Cant have a log of zero or negative values. The gain of the amplifier would have to be curtailed or its function otherwise inhibited when no signal is detected. The other problem, which has likely been overcome, is temperature stability. The amp I tried was good as long as the ambient temperature didnt change.
Ive been looking for a good modern log amp for some time to re-evaluate this approach. Perhaps this is it?
Eric.
End Quote

[Davor]

I'd go along with Earl's observations, except that I'm not too much into the noise story. Yet.

Problem is in metal detectors you want to measure difference between 1.0000 and 1.0001 signal ....

...which you can only benefit from the log amp. Log amp detectors are famous for AM detection of low modulation indexes.

I still think the whole story about a nasty noise is far too much exaggerated because I can decide NOT to sample deep in noise but half a decade above. It is the slope that matters, and simulation shows nice splitting of curves waaaay above noise. Besides, even a traditional PI Rx does not go that deep into noise. There is a tendency to sample a bit earlier.

I think the perfect way of harnessing data from such a setup is by means of window comparators. Level reference instead of time, and time result instead of level. Two such windows could yield discrimination. Voltage reference for the windows can be set by diode forward voltage drop, hence a temperature compensation. Easy.

Log amps are a different pack of animals altogether, and they need different care and feeding. But they are not saturating. And they are darn fast.

[Davor]

Years ago, I used this pre-amp. It was giving good results. However, because of latent temperature drift etc. I abandoned the design.

Your design is noninverting one and both inputs are exposed to full signal swing and rails limitation. The solution I propose is an inverting one, and opamp via its feedback fights the full wrath of monocoil Pi voltage. OK, there is one diode to help it - just in case. But the whole point is that you can have a full PI decay exposed to analysis and decomposition. It would be funny seeing it on scope.

(I'm still lab-less )

My personal view of PI so far is that there is far too much power in Tx, while the only thing that really matters is di/dt or voltage. Even in that department there is too much juice flowing.

[Davor]

Hypotheses (not complete):
...
Aziz

11. Engineers are in deep doo-doo if their chief does not understand what the heck they are talking about. (which happens a lot)
12. Only the competition is entitled to radical design changes. (no competition - no changes)

[Aziz]

Hi Davor,

why do you insist on the log-amps? There are better solutions available.
1. VCA (voltage controlled amplifier)
2. PGA (programable gain amplifier)
3. Fixed gain amplifiers (more cascaded stages of them)

The latter one offers you the best performance (noise & stability). If you need different gains, you could use more amplifier stages (2 or more). The VCA is critical and does not offer the benefit of the fixed gain amplifier. Note, that fixed gain amplifiers offer a stable loop-back regulation.

The problem with the MD signals is, that the ultra tiny response signal is "riding" on the high flyback voltage. Best to detect the response signal is to integrate over a window time and look at the change of it. The tiny integrated signal change can be detected easier at the end.

Why bother with high flyback voltage? You can use the IB coils.

If log-amps would offer more benefit, it would have been used long before.
(Occam's Razor fact)

Cheers,
Aziz

[moodz]

I'd go along with Earl's observations, except that I'm not too much into the noise story. Yet....which you can only benefit from the log amp. Log amp detectors are famous for AM detection of low modulation indexes.

I still think the whole story about a nasty noise is far too much exaggerated because I can decide NOT to sample deep in noise but half a decade above. It is the slope that matters, and simulation shows nice splitting of curves waaaay above noise. Besides, even a traditional PI Rx does not go that deep into noise. There is a tendency to sample a bit earlier.

I think the perfect way of harnessing data from such a setup is by means of window comparators. Level reference instead of time, and time result instead of level. Two such windows could yield discrimination. Voltage reference for the windows can be set by diode forward voltage drop, hence a temperature compensation. Easy.

Log amps are a different pack of animals altogether, and they need different care and feeding. But they are not saturating. And they are darn fast.

Are you sure Davor ? .... Log of 1.0000 = 0 and log of 1.0001 is 0.00004342 ... In other words by logging the value you made the difference even smaller. ... A log amp is a compression amp. .. Not good for seeing small differences ....

Regards moodz

[Midas]

....
If log-amps would offer more benefit, it would have been used long before.
(Occam's Razor fact)
...

Lol. Not that I'm pro or con logamp (honestly I'm well out of my depth). But if you start applying Occam's razor to the effect of "The simplest reason that no one has ever done it, is because it doesn't work" Then you've pretty much put a halt to ALL innovation.

"Common man everyone knows lightning is the only way to get fire. If it was as simple as rubbing two sticks together don't you think someone would have tried it by now?"

Midas

[Aziz]

Lol. Not that I'm pro or con logamp (honestly I'm well out of my depth). But if you start applying Occam's razor to the effect of "The simplest reason that no one has ever done it, is because it doesn't work" Then you've pretty much put a halt to ALL innovation.

"Common man everyone knows lightning is the only way to get fire. If it was as simple as rubbing two sticks together don't you think someone would have tried it by now?"

Midas

*LOL*

I was only showing the simple principle of questioning things of course. Why, why, why..?

However Davor principle tried to get max. target response energy. Nothing wrong with it. There is a lot of response energy in the early decay time. But the reactive response (ground signals) are huge at this stage too, which makes ground balance difficult.

BTW, the induction balanced coil types reduce the unwanted flyback decay voltage and the ratio of the signal/flyback voltage increases heavily.

Anyway.

The main core issues are:
- low noise designs
- avoiding pre-amp saturation/overloading
- synchronous demodulator
- filtering the output of the demod

Aziz

[Davor]

There are several good points to it, and I'll try to elaborate a bit. My gut feeling says that PI demodulation can only benefit from applying some appropriate weighting function to a Rx signal that will flatten its response over time. My first thought was a ramp or a sinus. There are some good points to these. Ramp is feasible via VCA, and its family, while sinus is feasible via LC tank. Trouble with linear ramp is that it will not help much beyond, say, one decade of signal span, and it is not flattening the signal a lot. Non-linear ramp is readily available as a log-amp, and here you go. The sinus weighting was already patented and as far as I know did not fare too well, partly because the time constant of a tank was IMHO far too short. I'll get back to that solution eventually.

Common feature of all log amps is flat top and bottom where the log curve hits its limitations. On the bottom side there is a (mostly) linear part that is usually buried in noise .. but not necessarily so. With the design I posted previously I can decide how deep the curve follows the log law. By addition of a single resistor in the existing design in a feedback path I can assure a fixed gain for low signals, and detection as usual. 2.2Meg fixes gain to 100 which is same as the "Surf" counterpart below. I traced the response without the resistor for comparison (dotted line). There is something else apparent there: the response is not symmetrical against "Surf" response, and "Surf" response (for no apparent reason) quenches much faster.

I could speculate about the "Surf" result falling short, but it seem as if the 1k in series with limiting diodes in effect damps a coil a bit too much at higher signal levels, so there's not much energy left for it to do its thing. One may guess its role in noise, but I don't expect much difference with larger one because it goes to a hi-impedance non-inverting input.

Anyway, here is a log-lin amp with max gain of 100.