VLF MD with digital signal processing : Bee-Buzz 1

[Atul Asthana]

I became boring (to myself too) repeating this on forum many times; but author of FelezJoo PI has done a real little miracle with a minimalist approach.

............
I think Paul suggested the same thing only in a different way, by linking to the smart students project.

--------------------

Ivconic,

I am currently taking my time to browse through some patents, and it may take me weeks to identify the relevant ones, comprehend their details, and extract the knowledge applicable to this project.

However, those of you who have already explored these patents or worked on creating similar devices could share your insights, practical applications, and any associated challenges on this forum much more effectively and quickly.

For instance, you could delve into the student project you recommended, thoroughly analyze it to extract the relevant knowledge, and compare its methodology and outcomes with the goals of our project.

From this comparison, you could evaluate whether the project is feasible using the resources and constraints of the STM32 Blue Pill, and propose any necessary modifications.

Additionally, outlining the advantages of these suggested changes would help us understand how they enhance the project's performance or efficiency, providing a clear path for improvement.

The truth is, direct suggestions and improvements from a group of knowledgeable experts are far more valuable than one individual painstakingly analyzing the technology or patents, some of which I might struggle to fully understand or interpret. Collaborative discussion not only accelerates the learning process but also enhances the quality of solutions we can collectively achieve.

[moodz]

I would agree .. we had some discussions on this forum years ago about the KISS principle and you really need to make it simple.
A simple way to get a POC "proof of concept going" is these possible pathways / options.

1. Take a simple project off the web ( ie exists already ) that ppl can build and play without visiting the bank manager or the wife. Like the example I gave. Then make it better.
OR
2. Take some generic modules ( eval boards and cpu boards ) and connect them together with some glue bits ( power supplies , tx circuit )
OR
3. Just use the soundcard in an old laptop / computer with a simple RX / TX and write/download the DSP code and try things out.

Instead you are choosing the most difficult path of designing a custom detector from scratch using random chips and DSP blobs AND you already said it wont be very high performance ????? ( waste of time )
Building the custom unit is the step AFTER the POC is done !!

As an example I have gone down choice 2. and with an ADC7760 eval board and an XILINX spartan eval board and some glue bits ( about $150 ) I getting PI ( 1 meter detect ranges + ground cancel ) performance from a single frequency detector with no front end amplifier,
Best of all I have not committed my eval modules to a "soldered" design ... I can reuse them ( and bits of the code ) on something else. There are no amps or filters on the frontend and I did not try to use nice sounding DSP blocks I just kept what worked and threw out everything that didnt.
moodz.

[ivconic]

I became boring (to myself too) repeating this on forum many times; but author of FelezJoo PI has done a real little miracle with a minimalist approach.
That is enough proof that even with little a lot can be achieved.
Of course, just looking at the schematic doesn't say much in support of my claim.
But after a dozen DIYs and the final success of getting it right (there are a lot of strict conditions for success); FelezJoo PI becomes a real little wonder, very usable and competitive.
Of course, behind everything is a very smart approach to writing code. Everything is in the code.
That's why I said "The "trick" is not in the codec but in the DSP."
In the specific case, the author first relied on principes explained in patent No.5,506,506 (US5506506) by Candy.
In this case it is VLF I/B and not PI. (easier to do than PI imho)
I can't remember the exact patents, but I think Carl tried to include them all on this site...among them there are a couple of well written ones with detailed algorithms.
All it takes is to take some time and do a little research by reading some of the patents.
Some of the algorithms can be reduced to minimalism and successfully applied in your project.
I think Paul suggested the same thing only in a different way, by linking to the smart students project.

[Atul Asthana]

That's all hidden in code. However, based on patent US7579839 I assume Minelab is doing narrowband synchronous quadrature demodulation in software. Likewise, Noka has a patent (US11914095) on an asynchronous demodulation method.

great, I'll read up. but may not be able to directly use their patents in our design, due to patents.

[Atul Asthana]

I was considering whether we could explore using 2–3 different TX frequencies interleaved in adjecent time slots, frrquencies being on either side of the main TX frequency.

Our coils have a low Q and perform similarly good enough at 5800 Hz and 6200 Hz. With an op-amp filter bandwidth of ±400 Hz and precise control over the TX frequency, we can operate within these limits. Additionally, the signal processing in each TX slot is independent of the preceding or succeeding slots.

This allows us to transmit and process different frequencies in separate time slots: Slot 1 at 5800 Hz, Slot 2 at 6000 Hz, and Slot 3 at 6200 Hz. The received signals from these slots can be analyzed using the Goertzel algorithm to extract amplitude and phase data for each frequency.

To effectively process and display this multi-frequency data, we’ll need a new technique to consolidate these results into meaningful information for the user. Increasing the number of slots to 512 and reducing the sine wave cycles per slot from 16 to 8 may help accommodate the additional data.

I’m curious to know the advantages this multi-frequency approach might offer and what strategies could be employed to integrate and optimize these results for improved performance in the metal detector. Could this enhance target discrimination, depth estimation, or reduce ground noise interference?

please opine.

[Carl-NC]

I’m interested in understanding the sampling and signal processing strategies employed by direct sampling detectors.

Where can I find detailed information about these technologies?

It would be helpful to know if any manufacturers or research papers provide insights into their design or algorithms.

That's all hidden in code. However, based on patent US7579839 I assume Minelab is doing narrowband synchronous quadrature demodulation in software. Likewise, Noka has a patent (US11914095) on an asynchronous demodulation method.

[Atul Asthana]

Thanks ivconic for the schematics.

As per Carl's recommendations, if we stick to the original scheme, there isn't much hardware involved, as outlined in an earlier description.

Once the software design is finalized, we will focus on the appropriate hardware, which primarily consists of op-amps and filters. These components have already been calculated but will be revisited if the TX frequency changes.

Additionally, care will be taken to ensure the hardware supports efficient signal processing with minimal noise. Any changes in TX frequency will also necessitate adjustments to the filter bandwidth and response characteristics to maintain system accuracy.

[Carl-NC]

6kHz is fine. For the purpose of development, frequency doesn't matter and can easily be changed later.

[Atul Asthana]

The first direct sampling detector was the original X-Terra (30/50/70). They used a Wolfson CODEC WM-something, don't have my notes right now. The next was the White's Prizm 6T which was used a 24b ADC (not CODEC) to quadrature-sample the RX. Then everything ML did -- Go-Find, Equinox, Vanquish, XT-pro, etc -- were all direct sampling. As are the XP Deus models, the newer Nokta models (Legend, Simplex, Score, etc), and the Garrett Vortex.I know that Nokta uses a more mainstream TI CODEC, the PCM-something. Look at the PCM1808 as an example, it's only 96kHz but is the only part number I know off the top of my head. Most everyone is using a 192kHz CODEC. Generally, you want to sample at 4x so 192kHz supports up to 40kHz which you might recognize as the highest that any of the MF models run. There are also a few 384kHz CODECs.

I’m interested in understanding the sampling and signal processing strategies employed by direct sampling detectors.

Where can I find detailed information about these technologies?

It would be helpful to know if any manufacturers or research papers provide insights into their design or algorithms.

[Atul Asthana]

Personally, I think you're original plan of using the internal 12b ADC makes the best sense as it minimizes the number of hurdles you have to jump just to get initial results. Plus, it will give you good data on what minimum # of bits are needed.

.

I agree with Carl that lets keep things as simple as possible.

so, is 6 khz tx frequency the right choice or do we need to select some other frequency, and thr rrason thete of.

my next step is to analyse whether the signal processing methodology needs any improvement :
1. sampling strateg,
2. integration philosophy
3. amplitude and phase extraction,
4. ground and target response extraction
5. VDI calculation

And the administrstive functions management.

Please comment, describing logic behind your suggestion for change.

[ivconic]

Carl is absolutely right.
You're going to open a can of worms if you go down that road.
The "trick" is not in the codec but in the DSP.
In the case of X-Terra it is a TMS320VC5402 DSP.

[ivconic]

point to consider about using expensive opamps.
however, with 24 bit codecs, I'll probably need good opamps.

[ivconic]

Great conversation, I am learning

which low cost 24 bit codec would be suitable in this design?
and which direct sampling design should I study?

[Carl-NC]

Personally, I think you're original plan of using the internal 12b ADC makes the best sense as it minimizes the number of hurdles you have to jump just to get initial results. Plus, it will give you good data on what minimum # of bits are needed.

The first direct sampling detector was the original X-Terra (30/50/70). They used a Wolfson CODEC WM-something, don't have my notes right now. The next was the White's Prizm 6T which was used a 24b ADC (not CODEC) to quadrature-sample the RX. Then everything ML did -- Go-Find, Equinox, Vanquish, XT-pro, etc -- were all direct sampling. As are the XP Deus models, the newer Nokta models (Legend, Simplex, Score, etc), and the Garrett Vortex.I know that Nokta uses a more mainstream TI CODEC, the PCM-something. Look at the PCM1808 as an example, it's only 96kHz but is the only part number I know off the top of my head. Most everyone is using a 192kHz CODEC. Generally, you want to sample at 4x so 192kHz supports up to 40kHz which you might recognize as the highest that any of the MF models run. There are also a few 384kHz CODECs.

[Atul Asthana]

Your proposed detector is a single frequency @ 6 Khz ... Op Amp data sheets quote broadband noise and its usually worst in the first DC to 1 Khz ... far away from 6 Khz. No need to spend money on expensive opamps based on CE ( Catalogue Engineering )

The only way that low frequency noise in the ( cheaper ) opamps is going to get into your 6Khz signal is if there is non linearity or overload ( ie convolution ) occuring through bad design.
Do you reckon they did moonshots back in the 60s with these uber opamps ... no they didnt.
moodz

point to consider about using expensive opamps.
however, with 24 bit codecs, I'll probably need good opamps.