Concept Design

[Willy Bayot]

I saw it and it is splendid post, touches some of my questions.
Before I saw that post again, now when you pointed me on it; I was going to propose direct sampling on only first 30uS of the decay.
Because we will looking for small gold nuggets, that's the main role of our detector here.
Anything above 30uS is far out of our interest here. Or am I wrong?
Even those 30uS are negotiable, maybe we could "chirp" even few uS less than that.
Which left us not so much data to process further, don't you agree?
So the question repeats; in which case we will have more data to process; Deus case with complete sine reproduced or here with only 30uS decay reproduced?
If you don't agree on my proposed 30uS; ok, lets put in math all 85uS from that post.

Ground control needs to process the late slots of the decay, thus, they also need to be recorded.

[Willy Bayot]

So I have set the sampling to 2.5 MSPS continuous ( as if you use the sync input it resets the internal FIR and glitches the data till the device settles ) and the FPGA bins 512 samples in a dual port ram oversampling by 4096 ( 12 bits for 6 bits of oversampling gain) in each bin

Is it not necessary to synchronize the slot captures with the pulse period so that each slot corresponds to a given offset in the exponential decay?
In that case, an oversampling of 4096 applied on a pulse period of 200µsec would only give a net DSP result every 820msec. This is making the system too slow in target response.

Am I wrong?

[moodz]

Is it not necessary to synchronize the slot captures with the pulse period so that each slot corresponds to a given offset in the exponential decay?
In that case, an oversampling of 4096 applied on a pulse period of 200µsec would only give a net DSP result every 820msec. This is making the system too slow in target response.

Am I wrong?

You can't have the sample times moving around in a pi type detector. Yeh I was wrong missed the sample count factor making the target response too slow....

[Willy Bayot]

You can't have the sample times moving around in a pi type detector. Yeh I was wrong missed the sample count factor making the target response too slow....

The best we can do with a pulse period of 200µsec is an oversampling of 2^(2*3) = 64 to get a gain of 3 bits and giving a target response of around 13msec (13mm @ 1m/s)

That can also be complemeted by a short delay IIR filtering applied on each slot.

[Carl-NC]

Anything above 30uS is far out of our interest here. Or am I wrong?

There are fast eddy targets, slow eddy targets, viscous magnetic ground, and iron. Generally for any eddy target you want to sample as early as possible and if all you want is raw detection then that is good enough. If you want to look at the decay and generate a target ID then you need more samples. Viscous ground can be nulled using an early sample and late sample (the classic method) or by using multiple TX pulse widths. Iron decay requires multiple samples to distinguish its BH-curve decay from eddy decay. So the simple answer is, no, we don't want just the first 30us.

So the question repeats; in which case we will have more data to process; Deus case with complete sine reproduced or here with only 30uS decay reproduced?

With VLF you want enough data points to resolve a phase angle so this requires 4 samples per TX period. For a 10kHz detector this means sampling every 25us. This is probably why current direct-sampling MF models are limited to 40kHz; they all appear to use audio codecs and I haven't seen a codec that samples faster than 192kHz. Anyway, the amount of data depends on the max TX frequency. It is similar for PI. You can probably do everything you need to do with only 4 samples per decay period. In a square-wave PI that means 8 samples per complete period, and we have only talked about 20kHz max. That comes to 160kSps, exactly the same as for a 40kHz VLF. In either VLF or PI, oversampling more points per period can buy you more SNR so you do it if you can.

[moodz]

The best we can do with a pulse period of 200µsec is an oversampling of 2^(2*3) = 64 to get a gain of 3 bits and giving a target response of around 13msec (13mm @ 1m/s)

That can also be complemeted by a short delay IIR filtering applied on each slot.

The faster oversampling can be used for searching / swinging and more oversampling for slower swing fine sensitivity. I have 255 for 4 bits set at the moment but obviously can adjust.

[moodz]

There are fast eddy targets, slow eddy targets, viscous magnetic ground, and iron. Generally for any eddy target you want to sample as early as possible and if all you want is raw detection then that is good enough. If you want to look at the decay and generate a target ID then you need more samples. Viscous ground can be nulled using an early sample and late sample (the classic method) or by using multiple TX pulse widths. Iron decay requires multiple samples to distinguish its BH-curve decay from eddy decay. So the simple answer is, no, we don't want just the first 30us.



With VLF you want enough data points to resolve a phase angle so this requires 4 samples per TX period. For a 10kHz detector this means sampling every 25us. This is probably why current direct-sampling MF models are limited to 40kHz; they all appear to use audio codecs and I haven't seen a codec that samples faster than 192kHz. Anyway, the amount of data depends on the max TX frequency. It is similar for PI. You can probably do everything you need to do with only 4 samples per decay period. In a square-wave PI that means 8 samples per complete period, and we have only talked about 20kHz max. That comes to 160kSps, exactly the same as for a 40kHz VLF. In either VLF or PI, oversampling more points per period can buy you more SNR so you do it if you can.

At least for research purpose the ADC must be fast with sufficient resolution and sample at any and as many desired timings.

[moodz]

I saw it and it is splendid post, touches some of my questions.
Before I saw that post again, now when you pointed me on it; I was going to propose direct sampling on only first 30uS of the decay.
Because we will looking for small gold nuggets, that's the main role of our detector here.
Anything above 30uS is far out of our interest here. Or am I wrong?
Even those 30uS are negotiable, maybe we could "chirp" even few uS less than that.
Which left us not so much data to process further, don't you agree?
So the question repeats; in which case we will have more data to process; Deus case with complete sine reproduced or here with only 30uS decay reproduced?
If you don't agree on my proposed 30uS; ok, lets put in math all 85uS from that post.

Those curves have been posted many times in many places in different formats and people look at them and say how "easy" it will be to build a detector to exploit them.
Well those curves are the "el Dorado" of pulse induction .. everyone knows ( most designers ) know what they look like but they dont know how to achieve the goal.
This is not helped by the now requirement of PI detectors to look for very small or very deep targets which provide very little target info for discrim because they are very weak targets.
For PI depth in mineralisation is no 1.
Descrim ? whats that ?

moodz

[Willy Bayot]

Those curves have been posted many times in many places in different formats and people look at them and say how "easy" it will be to build a detector to exploit them.
Well those curves are the "el Dorado" of pulse induction .. everyone knows ( most designers ) know what they look like but they dont know how to achieve the goal.
This is not helped by the now requirement of PI detectors to look for very small or very deep targets which provide very little target info for discrim because they are very weak targets.
For PI depth in mineralisation is no 1.
Descrim ? whats that ?

moodz

This is why we need the components of best quality and good PCB layout in all the modules to get the best SNR from the AFE before the DSP.
The last parts of the decays are the most difficult to process as their SNR is minimal there.
SOME designers DO know how to exploit these curves.

[Willy Bayot]

There are fast eddy targets, slow eddy targets, viscous magnetic ground, and iron. Generally for any eddy target you want to sample as early as possible and if all you want is raw detection then that is good enough. If you want to look at the decay and generate a target ID then you need more samples. Viscous ground can be nulled using an early sample and late sample (the classic method) or by using multiple TX pulse widths. Iron decay requires multiple samples to distinguish its BH-curve decay from eddy decay. So the simple answer is, no, we don't want just the first 30us.



With VLF you want enough data points to resolve a phase angle so this requires 4 samples per TX period. For a 10kHz detector this means sampling every 25us. This is probably why current direct-sampling MF models are limited to 40kHz; they all appear to use audio codecs and I haven't seen a codec that samples faster than 192kHz. Anyway, the amount of data depends on the max TX frequency. It is similar for PI. You can probably do everything you need to do with only 4 samples per decay period. In a square-wave PI that means 8 samples per complete period, and we have only talked about 20kHz max. That comes to 160kSps, exactly the same as for a 40kHz VLF. In either VLF or PI, oversampling more points per period can buy you more SNR so you do it if you can.

With Direct Sampling, we can define as many demodulators as we want, we are not limited by the number of analog channels. We can assemble consecutive slots into any number of virtual sampling windows and apply filtering on each of them separately.

[ivconic]

Based on the criteria set by Tony (and Carl) ; what is the total uS number to be sampled?

[Willy Bayot]

The faster oversampling can be used for searching / swinging and more oversampling for slower swing fine sensitivity. I have 255 for 4 bits set at the moment but obviously can adjust.

True, the oversampling at 255 would still give a net results every 50msec in pinpointing mode.

If we decide on an LTC2380-24 ADC and a standard CPU , we could get slots 1µsec wide. With a pulse period of 200µsec, we get 200 of them.

• Continuous Capture of the raw data under SPI with DMA support into a circular DMA buffer.
• Synchronizing to the start of each pulse, save the current DMA buffer index and read the FIFO from there to fill (increment) the 200 slots one after one and then back to synch as many times as necessary to make the required oversampling.
• Assembling consecutive slots into virtual windows
• LP Filtering
• DSP and reports

[Willy Bayot]

Based on the criteria set by Tony (and Carl) ; what is the total uS number to be sampled?

Depends on the pulse period size.
The pulse period size depends on the average TC of expected targets. Long TC targets (large nuggets) require longer pulse periods while short TC targets (small nuggets) can accomodate shorter pulse periods (more PPS) giving the option to increase the digital integration without increasing the response time.
The number of µsec to be sampled IS the length of the pulse period (or the half-period if using the bipolar CC type XMIT)

Note that it does not cost more to capture the whole decay period than to only capture the first N µsec.

[ivconic]

Depends on the pulse period size.
The pulse period size depends on the average TC of expected targets. Long TC targets (large nuggets) require longer pulse periods while short TC targets (small nuggets) can accomodate shorter pulse periods (more PPS) giving the option to increase the digital integration without increasing the response time.
The number of µsec to be sampled IS the length of the pulse period (or the half-period if using the bipolar CC type XMIT)

Note that it does not cost more to capture the whole decay period than to only capture the first N µsec.

I see.
So, the maximal sample number will determine the choice of ADC.
Carl would probably suggests even more than that! Much more samples as possible. Ok.
I circulate with questions about one and the same thing; what would be the minimum ADC specs that would get the job done here.
Carl said 18bit, maybe 16bit. Moodz and you said 24bit is the real deal.
But 4-8 channel, 24bit, with high(est possible) sampling rate... either there is none, they don't exist, or I didn't find them, or if they do exist;
they are probably very expensive and complicated to implement.
So...
In order to relax the criteria and in that way reach an existing ADC that would finish the job in full; first we need to clarify
if we really need high repetition rate TX?
25kHz according to Tony.
I understand why he demanded it.
But is it really mandatory?
We know of some detectors that successfully locate the smallest gold nuggets and yet do not have such a high pps.
How much in reality do we lose if pps is lower? What will we miss?
Another aspect; doesn't the influence of soil increase at higher pps?
Aren't the problems with mineralization drastically complicated, with the increase in pps?
What is the optimal measure? Which choice is the best? With the least complications and tradeoffs.

[Willy Bayot]

I see.
So, the maximal sample number will determine the choice of ADC.
Carl would probably suggests even more than that! Much more samples as possible. Ok.
I circulate with questions about one and the same thing; what would be the minimum ADC specs that would get the job done here.
Carl said 18bit, maybe 16bit. Moodz and you said 24bit is the real deal.
But 4-8 channel, 24bit, with high(est possible) sampling rate... either there is none, they don't exist, or I didn't find them, or if they do exist;
they are probably very expensive and complicated to implement.
So...
In order to relax the criteria and in that way reach an existing ADC that would finish the job in full; first we need to clarify
if we really need high repetition rate TX?
25kHz according to Tony.
I understand why he demanded it.
But is it really mandatory?
We know of some detectors that successfully locate the smallest gold nuggets and yet do not have such a high pps.
How much in reality do we lose if pps is lower? What will we miss?
Another aspect; doesn't the influence of soil increase at higher pps?
Aren't the problems with mineralization drastically complicated, with the increase in pps?
What is the optimal measure? Which choice is the best? With the least complications and tradeoffs.

With Direct Sampling, we just need a single channel ADC.
The slot width depends on the total conversion rate of the ADC (tCYC)
The LTC2389-24 can sample at 2MSPS but the SPI clock of standard CPU's do not allow that speed. I have calculated the maximum rate using both PIC32MZ and STM32. I get a rate of a bit more than 1MSPS, thus, 1µsec per slot if the capture is made with DMA support, otherwise, if it is made through polling or interrupt, it is not possible to get this rate.
I would say that 1µsec wide slots are reasonable enough to allow a good DSP.

The repetition rate (PPS) has nothing to do with the selection of an ADC.
High PPS give the opportunity to get a faster response time with higher oversampling (integration)
25Khz is just the size of the period (or half-period for CC) = 40µsec. This high PPS is only valid for small nuggets.

Mineralisation does not influence a PI as much as a VLF. but it does and it should be taken care of.
The normal ground mineralisation is less of a problem than HOTROCKs which are detected as targets and should be masked.