New PI circuit idea?

[Billy]

Some time ago I saw an old patent that described a seach head for PI and a new approach to sampling. I had it for years in my mind, so recently I deside it was time to try it.
We have the usual transmiter that gives the decay curve to sample. The idea is that instead to wait some usec and then sample, why not sample a fast timer with a good comparator at fixed voltage levels say 5volt, 3volt and 1volt. This will give us three timings with say reference the stop of tx pulse. Subtracting each from the other will give us coordinates for discrimination. The decay curve does not change shape with distance.
I believe it's a good idea and worth a try.
I'll use a microcontroller for all the timing and comparator work (propably the AT90S2313).
Any opinions?

[jackdetect]

Re: New PI circuit idea?

What about ground balance?
Number of patent?
Salute,
Jackdetect.

[Billy]

Re: New PI circuit idea?

Unfortunately I can't remember now the patent, I'll have to dig huge piles of papers. The most important thing the patent had was that the inventor suggested that in a graph of a decay curve, the geometric area enclosed between the "flyback" and decaying of voltage was the same for vacuum and for say ground. That means the peak is lower and the decaying is larger than vacuum which is sharp. If I remember correct, the patent was for some industrial automation.
As you understand my first and primary concern is not ground balance but sensitivity. If it has enough then a lot of things are possible. The interesting one is that it may do some discrimination because you can calculate the shape of the curve.
For sensitivity, I'll start with 10MHz clock. That will give me 100ns plus or minus 50ns relative timing of voltage crossing each sample point.

[jackdetect]

Re: New PI circuit idea?

I can not belive in inventor claim about same graph for the ground and vacum ( maybe he mean about ground in industrial purpose).
Industrial automation ( known conditions ) and wildness ( always unknown conditions ), are very different things.
I was see many detectors with exellent sensitivity on the air, but almost useless on the field.
Without ground compensation, I think is not possibly to utilize other good features of any metal detector.

Salute,
Jackdetect.

[sevzirfo]

Re: New PI circuit idea?

Hi Billy!

Very interestingly...
It is not enough speed of the processor.
It is possible to use the external clock generator 100 mhz and fast counter, and result to transfer in the processor.

AT90s2313-10pi Stably works on 20 mhz

[Billy]

Re: New PI circuit idea?

Thank you,
I didn't worked it above 10 MHz. I'll try first to use its internal comparator linked to capture Timer/Counter1 and see from the resolution from there. If it captures reliably then I'll try to increase clock for better resolution. When I have somthing at hand (hardware-software) I'll post. In the meantime I'll work it.
I'm drawing some pictures to visualise the concept to be better understood.

[Billy]

Re: New PI circuit idea?




I hope from the image to be better understood the whole concept. Beware that nothing is scaled but rather enlarged to point things.
Not shown is the clock that I count but you can imagine that I have a steady pulse train at 10 MHz frequency.
I am interested on the three points t1, t2, t3. t2-t1 will give our x-axis coordinate (lets name it Tx) and t3-t2 will give y-axis (lets name it Ty).
Tx and Ty will be independend from object-to-coil distance (at least I hope). Tx and Ty from air count stored into memory will give us a good distance from coil approximation.
Ground conditions will not interfere with measurement.
I'm waiting for more opinions.

[sevzirfo]

resolution

AVR AT90s2313-10

10 mHz
5 microsecs = 50 cucles
8 microsecs = 80 cucles
12 microsecs = 120 cucles
50 microsecs = 500 cucles

20 mHz
5 microsecs = 100 cucles
8 microsecs = 160 cucles
12 microsecs = 240 cucles
50 microsecs = 1000 cucles

I think, that the resolution is small even for 20 mhz.
It is possible to try to combine two or three results for reception of good resolution.
The decision of a task can be the external fast counter and external clock generator, but it will complicate the detector.

[Billy]

Re: resolution

why? It is suggested that sampling at 10usec - 13usec is sufficient enough for small nuggets. Of course bigger resolution is better but with the added cost of a more complicated schematic. For now we have only the switching supply, the TX mosfet and a front receiver for signal conditioning. This means that it will be a small device.
Now, I need to explain that you don't have to take three counts at the same pulse. You can sample once at each pulse. And if we have for TX an H-bridge, we can play with positive and negative pulses for field neutralization (land mine detection). I'm thinking for reference voltage for the comparator to use the uc and one of the comparator pins to charge a capacitor for a precise amount of time. Then the voltage will be used as reference.

[Cossaro Fernando]

Re: New PI circuit idea?

>why not sample a fast timer with a good >comparator at fixed voltage levels say 5volt,

the idea is ok (not new)
I think it work with a lot of false signals.
The main problem is the noise that false each sample. For example, few mVolts ripple at the power coming back from the TX coil will false the Voltage comparator.

The solution is the medium-sample so you perform some kind of “integration” computing the average of (n)samples and your circuit (or the software) became more complex to make the same task.

The second problem is the ADC resolution with intrinsic non-linearity that impose some undetection conditions so you must amplify the signal to use the full ADC range (and this grow the noise too!)

PS: my english is too bad!

[Billy]

Re: New PI circuit idea?

I'm sorry but you probably misunderstood. I don't propose to use an ADC nor to sample the curve when it is some uvolts but to sample a quick timer (imagine it as pulse train) which is tripped by the falling voltage crossing the 3 volts. All of that can be done on the same chip provided that the 10Mhz clock is adequate.

[Cossaro Fernando]

Re: New PI circuit idea?

The ADC include more comparators that perform a sample each one.

Your idea is based on one comparator but the problem is the same. The edge is more
noise-affected (compared to an integrator) so you keep a lot of errors that an integrator will aritmetical eliminate.

Did you consider this aspect?

[Billy]

Re: New PI circuit idea?

Yes, but again, I don't say to use an ADC. I don't even care what happens when you are close to ground. Of course there will propably be noise problem but 1st this can be filtered out with a digital filter and 2nd the amplification needed is less so less noise. If you read the concept (and of course if you look at the decay with an oscilloscope) you 'll see that the sample is not fixed in time - it moves. But even this doesn't matter because my interest is the difference between 3 succesive time samples. For instance in DvDt, Dt is what I count. Dv is allready known in contrary to usuall PI, in which at the same t in time you try to find Dv voltage.
The main point still remains: do you believe that the slope (sampling it at say 5v) contains enough information that will distinguish a target reliably? I believe yes. Of course I don't believe that it will give more sensitivity (at least for now) but it will offer something we all want - discrimination.

[Billy]

Re: resolution - thank you sevzirfo

Thanks to you, I just saw that the new tiny2313 goes to 20 MHz. I ordered a few to try. Eric Foster, commented as 'workable' and 'worth to try' so I'm counting to you too for remarks.

[Cossaro Fernando]

Re: New PI circuit idea?

OK, the difference at the decay give some kind of discrimination, this is a fact.

The method to compute the voltage "delta" between the T1 and Tn must consider the error.
If you try to read the voltage or the time when the decay cross the reference voltage, the error condition don't change.

So the the time delta between V1 and Vn must be the result of an average.

Consider that a classical PI integrator perform an average over 500-1000 samples to give an acceptable result (some signal) and some errors still occour. Did you calculate 100 time-delta (in a FIFO buffer!) per second, at neccessary precision, with a PIC?

The idea is ok but the problem (IMHO)is still the average computation.