Pi Detecting, ADC values

[Hesamavr]

Hi,

I want to build a PI detector. I have read the interesting part of the signal using an ADC which consists of 40 samples which if we plot it, it will represent the decay curve of the signal, which will change in presence of a metal.

Now what is the best action with these 40 samples? I was thinking of averaging, but averaging destroys small changes. averaging is something like moderating changes.

What is the best mathematical operation here?

[moodz]

Are you a hacker / designer etc ...

http://www.anadigm.com/_apps/AN231013-U308.pdf

There are many and varied opinions on your question ... LOL ... there are many posts in Geotech on this subject and even more opinions.

[Hesamavr]

Are you a hacker / designer etc ...

http://www.anadigm.com/_apps/AN231013-U308.pdf

There are many and varied opinions on your question ... LOL ... there are many posts in Geotech on this subject and even more opinions.

maybe there are some, because I could not find through google search, but they might be inside different topics. actually this is a fundamental question.

[waltr]

As Moodz said, read the forum here as all the theory and practice has been highly discussed.
There are even a few PI MD's here, and in the ITMD book, that uses an ADC. Look up and read them. One is Temos' Pulse MD.

Fundamental Yes, but not so in real world details.
Another good starting place is the Geotech Projects here:
http://www.geotech1.com/cgi-bin/page...e=projects.dat

Study how these simple detects work. Then transfer function to Digital.

Interesting chip Moodz and good app note.

[Hesamavr]

As Moodz said, read the forum here as all the theory and practice has been highly discussed.
There are even a few PI MD's here, and in the ITMD book, that uses an ADC. Look up and read them. One is Temos' Pulse MD.

Fundamental Yes, but not so in real world details.
Another good starting place is the Geotech Projects here:
http://www.geotech1.com/cgi-bin/page...e=projects.dat

Study how these simple detects work. Then transfer function to Digital.

Interesting chip Moodz and good app note.

I will have a look, but if you know the answer, it will not bother to just reply what can I do with these 40 samples

[Mdtoday]

As others have said, there are a few alternatives to sampling.
I am assuming you are aware of the the integration methods employed in PI designs.
I would suggest to start with a simple running average of 1 sample per cycle over 16 to 32 cycles , get a feel for the sort of offsets and noise you are dealing with first , then move onto your multiple sample per cycle.
You can use look up tables to compare previous groups of samples and apply all sorts of algorithms and filters.
I have a question for you, what were you thinking of doing with those 40 samples?
Did you have a thought as to why you wanted to digitise multiple samples per cycle?
It could be that you were actually on a thought path that you should continue to experiment with.
My advice is to try out what you were thinking but start simple.

cheers

Mdtoday

[Hesamavr]

As others have said, there are a few alternatives to sampling.
I am assuming you are aware of the the integration methods employed in PI designs.
I would suggest to start with a simple running average of 1 sample per cycle over 16 to 32 cycles , get a feel for the sort of offsets and noise you are dealing with first , then move onto your multiple sample per cycle.
You can use look up tables to compare previous groups of samples and apply all sorts of algorithms and filters.
I have a question for you, what were you thinking of doing with those 40 samples?
Did you have a thought as to why you wanted to digitise multiple samples per cycle?
It could be that you were actually on a thought path that you should continue to experiment with.
My advice is to try out what you were thinking but start simple.

cheers

Mdtoday

Thank you.

Actually the ADC is fast enough (one sample = 0.5uS or 2Msps) which I don't need to use several pulse to acquire whole length of the decay curve. Therefore this whole length in my sampling consists of 40 or 60 numbers. if I just simply add these numbers, then it means that I have calculated the integral of the pulse decay. if I use two pulses, I will have 80 or 120 samples and therefore more stable results. but the sensitivity is not as much as it should be. it detects a 2cm coin at 15cm with a 20cm spider shape coil. which should be at least at 30cm distance.

What is your suggestion now? You think integration using an opamp is a better solution?

[green]

Thank you.

Actually the ADC is fast enough (one sample = 0.5uS or 2Msps) which I don't need to use several pulse to acquire whole length of the decay curve. Therefore this whole length in my sampling consists of 40 or 60 numbers. if I just simply add these numbers, then it means that I have calculated the integral of the pulse decay. if I use two pulses, I will have 80 or 120 samples and therefore more stable results. but the sensitivity is not as much as it should be. it detects a 2cm coin at 15cm with a 20cm spider shape coil. which should be at least at 30cm distance.

What is your suggestion now? You think integration using an opamp is a better solution?

The way I look at it. Including a chart calculated with HyperPhysics. To increase detection distance from 15cm to 30cm with a 20cm coil, S/N ratio would have to increase by 30 times. My thought, maybe not correct. Best way to get there??

[waltr]

Thank you.

Actually the ADC is fast enough (one sample = 0.5uS or 2Msps) which I don't need to use several pulse to acquire whole length of the decay curve. Therefore this whole length in my sampling consists of 40 or 60 numbers. if I just simply add these numbers, then it means that I have calculated the integral of the pulse decay. if I use two pulses, I will have 80 or 120 samples and therefore more stable results. but the sensitivity is not as much as it should be. it detects a 2cm coin at 15cm with a 20cm spider shape coil. which should be at least at 30cm distance.

What is your suggestion now? You think integration using an opamp is a better solution?

I did not know you actual have the detector circuits and ADC/processor working which is the first Why I suggest reading a lot. You sounded like someone without any experience building MDs.

Very possible that you may need more gain in the front end and/or simply the digitized SNR is too low. This can cause 'low sensitivity' and depth loss. This is a tricky balancing act in detector circuit design.
Sample integration can be done with op-amp or in the digital domain. In digital the simplest is to do running average on samples which is basically what the op-amp integrator does. You may want more bits resolution so a simple method is just keep adding the newest ADC value and subtract out the oldest ADC value. The more samples added in the lower the noise and higher SNR.
You may want some analog LP filtering before the ADC to assist.

As for multiple ADC sampling per Tx cycle---- there are many theories as to what information can be derived from the shape of the waveform decay. I just do not know of anyone actually doing this with success.

[Carl-NC]

Sampling the decay produces less response than a sampling integrator. I would ignore that for now. Take the 40 samples and average each 32x. Save a static (no-target) version and subtract it from the dynamic signal. That eliminates the static decay and leaves only the target response. Play with that and see what happens. Esp try variations in preamp roll-off, maybe damping and even different coils. The divvy the 40 samples into maybe 4 zones with 10 added samples per zone. See what that offers. Lots to play with here, I did a lot of this in Labview years ago. Will it make a good swinging detector? So far it hasn't, but it's a good way to learn.

[bbsailor]

A lock in amplifier (LIA) will sample the same place in a signal cycle to reduce noise that is broad and random. By taking the sample in exactly the same place, such as in a PI metal detector, integrating many samples operates much like a LIA and improves the signal to noise ratio. See this tutorial. http://www.compadre.org/portal/items...l.cfm?ID=13360

Assume that a 12 inch diameter coil is swept about 3 feet in one second, then a target will stay within the 12 inch coil for about 333 milliseconds. Now assume that the PI pulse rate is 3K PPS then approximate 1000 samples could be integrated by my example above. The more samples that are integrated the better the signal to noise ratio. Eric Foster used this integration technique to improve the detection range on some of his PI metal detectors. This is just another alternative to increasing the power to increase the range.

Joseph J. Rogowski

[Mdtoday]

I meant to ask, what is the resolution of the ADC you are using?

cheers

Mdtoday

[Hesamavr]

I meant to ask, what is the resolution of the ADC you are using?

cheers

Mdtoday

12 bits. therefore the sensitivity would be around 1mV.

I think my MD should detect a 2cm coin at at least 30cm with a 20cm spider coil, but now it detects it at 15cm. Amplification is high enough.
The detection is triggered at the same time as I see some changes in the signal in the oscilloscope screen, but as you know oscilloscope just shows something when the magnitude is high enough to be visible in the volt/division range. it means detection must happen much sooner than any visible change in the oscilloscope screen.

[green]

12 bits. therefore the sensitivity would be around 1mV.

I think my MD should detect a 2cm coin at at least 30cm with a 20cm spider coil, but now it detects it at 15cm. Amplification is high enough.
The detection is triggered at the same time as I see some changes in the signal in the oscilloscope screen, but as you know oscilloscope just shows something when the magnitude is high enough to be visible in the volt/division range. it means detection must happen much sooner than any visible change in the oscilloscope screen.

How much amplifier gain do you have? I'm guessing you need at least 20 bits total, probably more?

[Hesamavr]

How much amplifier gain do you have? I'm guessing you need at least 20 bits total, probably more?

No, I don't think I would need more bits since it actually will go down to the uV range which accompanies with many noises either, Besides if you look at the similar designs, they have used 12 bits or even less 10 bits ADCs for the PI MD, such as Chance MD, Clone PI, ...

I have used a 1M resistor in the inverting mode with a 100R resistor in the input. Therefore it provides around 10K Gain.