Design Challenge

[Sean_Goddard]

I am struggling to get a circuit to work despite all the calculations showing it SHOULD, so....Here is your mission;

I have a signal with an dv/dt of 15mV/us for 56us then the signal levels off at 710mV final value.

What I am trying generate a logic "1" for the rise period, but as soon as the signal levels off, I want a logic "0". Now I know a differentiator & schmitt trigger should do the trick, but I simply cannot get the circuit to work.

I have spotted something in the Rx waveform of a PI which gives some VERY consistent results and using some magic circuitry processing, I get this sloping output. By processing the slope in the manner above, I want to gate another signal which will provide REAL discrimination (yup Dave Emery isn't the only one who has cracked this at last). The only thing is that I can't post any schematics of the PREVIOUS part of the circuit which generates the slope, and I do not intend to patent this idea as there will be only TWO units made (one for me and one for a friend).

The theory all works out according to the simulations and calculations and I have but this one stumbling block so any ideas people? NO, why don't you try it a different way is NOT the solution, this is the ONLY way (unless I use a micro which I don't want to do for noise reasons).

[Qiaozhi]

Have a look at the envelope follower circuit of the "Off-Resonance" probe in ITMD.
The circuit produces a logic one at the output of the comparator while the input signal is rising, and a logic zero shortly after it levels off.
You'll just need to adjust the time constants to suit your particular needs.

[Sean_Goddard]

Therein lies the problem. The slope and thus discrimination is, sadly, VERY coil dependent and it is VERY time consuming to wind and trim coils of different sizes (one is a 10mm diameter coil for ferreting out nuggets in rivers) as the machine is a 6KHz fast PI designed specifically for very small gold. The idea is to work the spoil heaps of a local mine on which I have been given free reign to keep WHATEVER I find, but working the surrounding areas requires a license and the cost is just stupid. The best part is I have access to a rock crusher to turn the rocks into gravel. I don't expect to get rich, but it will be fun and pay for the beers.

The mine is due to reopen, but the old extraction methods were not very efficient thus there are (from what I could see by eyes only) a LOT of VERY small nuggets (pin head sized) and flakes in the spoil heaps and it's all MINE (pardon the pun).

[Skippy]

A few more specifics may help us.
Does the signal always start off from zero volts before rising? Is the dv/dt critical, ie. if it rises at 10 mV/us or 50 mV/us , you don't want it flagging a Logic Hi ?
Is there any other 'false positive' condition we should know about?

If you generated a signal that FELL at 15 mV/us , and added it to your waveform, you would end up with a flat-ish output, which might be easier to distinguish from the unwanted signals?

[Teleno]

If you could at least post a timed waveform of the input and the desired output it would help.

[dbanner]

Micro is probably the best way for doing this task. Why prohibit it's use?
There exists a patent which has a circuit using a differentiator to do this sort of magic. Can't remember the title of it.

[green]

I am struggling to get a circuit to work despite all the calculations showing it SHOULD, so....Here is your mission;

I have a signal with an dv/dt of 15mV/us for 56us then the signal levels off at 710mV final value.

What I am trying generate a logic "1" for the rise period, but as soon as the signal levels off, I want a logic "0". Now I know a differentiator & schmitt trigger should do the trick, but I simply cannot get the circuit to work.

I have spotted something in the Rx waveform of a PI which gives some VERY consistent results and using some magic circuitry processing, I get this sloping output. By processing the slope in the manner above, I want to gate another signal which will provide REAL discrimination (yup Dave Emery isn't the only one who has cracked this at last). The only thing is that I can't post any schematics of the PREVIOUS part of the circuit which generates the slope, and I do not intend to patent this idea as there will be only TWO units made (one for me and one for a friend).

The theory all works out according to the simulations and calculations and I have but this one stumbling block so any ideas people? NO, why don't you try it a different way is NOT the solution, this is the ONLY way (unless I use a micro which I don't want to do for noise reasons).

Differentiator in spice. Set output=1.5v with 15mv/us in. Had to change time from 56us to 47.3us to get 710mv with 15mv/us in. Added sine input to show effect of noise in. Needed to add R2 and C1 to help filter noise. Causes a delay, how much delay is allowed? How much jitter is allowed? How much noise does your signal in have?

[Sean_Goddard]

Thanks Green!!

For the rest of you, here are the waveforms (truncated as there is "secret" info in some parts)





For Green: Jitter is NOT a problem as I can average this out. I'd rather NOT have any delay, but the less the better. If predictable (repeatable) I can compensate in later stages by using subtraction from the DC voltage in later stages. Generally, the lower all "errors" the better, but this world isn't perfect ...YET.

[green]

Thanks Green!!

For the rest of you, here are the waveforms (truncated as there is "secret" info in some parts)

[ATTACH]45060[/ATTACH]

[ATTACH]45061[/ATTACH]

For Green: Jitter is NOT a problem as I can average this out. I'd rather NOT have any delay, but the less the better. If predictable (repeatable) I can compensate in later stages by using subtraction from the DC voltage in later stages. Generally, the lower all "errors" the better, but this world isn't perfect ...YET.

Looking at Waveform 11. Ramp doesn't start at zero volts, maybe 25mv in zero time. 100 to 200mv in 7us(14000v/sec). 500 to 600mv in 10us(10000v/s). 600 to 710mv slope changes a lot. Are you looking for certain slope change or when it flattens out?

[Sean_Goddard]

When it flattens out. At that point, the magic with the other part of the circuit times (not mathematically, but spatially) another signal then, based on that time difference, these are the "X" and "Y" of the signal decay, a pot across the two provides virtually full range discrimination, from nail though to silver coins. The best part is that the sample time is <8us so this thing is insanely sensitive to the small gold flecks I'm looking for.

[Sean_Goddard]

Messing around in between call outs at work, I came up with the following;



It's NOT the most elegant solution, in fact it is cruder than Bernard Manning, but IT WORKS, it shouldn't but it does! OK, it has 6us delay and quite a bit of "follow-on" on the tail edge of the rise period. The other problem is the terrible jitter which, so far, I have not traced the source of (suggestions as to source / remedy please).

I'm going to play with some monostables as I think I have a cunning plan to use the RESET input to cut short the pulse and eliminate a lot of the tail edge jitter as I might try synchronising signals a la FPGA techniques to ensure all inputs are timed correctly. I'll have to think a bit more about how I can do this.

[Qiaozhi]

Now I'm a bit confused.

In your initial post you said: 'What I am trying generate a logic "1" for the rise period, but as soon as the signal levels off, I want a logic "0"., but then later on you've posted a diagram of your desired waveform which shows the generated output staying at logic "1" when the signal levels off.

Which is correct?

[green]

Initial slope approximate 15mv/us. Lot less before levels of. What slope is levels of? 1mv/us, 100uv/us, 10uv/us? What changes with target change, 710mv amplitude, time to level off or both?

[Sean_Goddard]

Ah I see where some have got the wrong end of the barge pole .

The BLUE signal is the pulse that generates the slope via the magic circuit, the GREEN (apt really) trace is the actual slope.

Logic "1" when the signal is rising then falling to logic "0" when the signal levels off. The logic signal is the derivative of the green analogue signal. What I need to do is use the signal to gate off the green signal or the discriminate part "folds back" on itself and I get a disc range from nail, though to pull tabs, then instead of rejecting everything up to silver coins, the disc somehow flip and accepts these "bad" targets and rejects the good ones which is, as you can imagine, not very useful.

Regarding the start level of the slop, yes it is around 25mV, this is due to the blanking circuit further up not reacting to the Tx pulse (+200V) no acting fast enough (plus a little shoot through) but, funnily enough, this gives a better nail knockout capability. I set the blanking to kill this and it becomes an all metal machine, with this small step it will reject nails as Disc zero.

As stated, I suspect my design is very similar to the TDi even though I have not seen one of those machine in the flesh or had chance to examine one.

One thing I WILL give away is that the disc method works ALSO uses information from the RISE portion of the Tx waveform. The wanted info is VERY subtle, but it's definitely there. It is how you process the decay signal in conjunction with this information which gives the discriminate signal. That is as much as I'm letting on . I believe the Pulse devil uses a VERY similar system, and that is one machine I HAVE played with (yes, it really does exist).

[Carl-NC]

One thing I WILL give away is that the disc method works ALSO uses information from the RISE portion of the Tx waveform.

I assumed so. You can't discriminate without a reactive component. I did a lot of work on this at White's, but the TDI doesn't use it.

The jitter is likely due to the ramp waveform itself, not the detection circuit.