Using FFT and neural network on a PI detector

[Atul Asthana]

Thanks thats a lot of useful information.
Are there any patents filed on these?

I'd have to refresh my maths to understand these papers

We should try using these papers to try out

Two methods for estimating the distribution of time constants for target classification with pulse induction.

http://www.geosensors.com/global/Hol...SAGEEP2004.pdf

https://zonge.com.au/wp-content/uplo...7/perc2000.pdf

[Carl-NC]

Although it's technically VLF and not PI, the Tarsacci metal detector probably does something similar. The TX voltage is a square wave so the RX signal consists of exponential responses, and they are demodulated in the time domain and curve-fitted to determine what's going on. There is a patent for it, US10969512. I see he also has a new patent application, US20220268961, but I haven't read it yet.

[Nozimo]

I guess, interpretation of 'use' in using the students, is due to the cultural difference.

Yes, definitely! I work in academia and have witnessed too many times "big professors" using students as cheap workforce with little to no credit for their work, hence my comment on "using students".

Students, will be able to take up such projects with great zeal, learning a lot in the process. And many students must already be working on [...] The md companies could learn from NASA's approach.

Definitely, this is probably one of the best approach. In fact, the more I read about PI and the more I'm convinced it's the same thing as TDEM or TEM, just with a different goal and scale. There must be research teams working on optimizing TDEM who developed signal processing techniques that can be borrowed by PI MD.

Funny enough, one of the papers shared by Teleno is from Zonge. We rented a TEM system from them once, it was terrible! Their customer service was even worse...

[Atul Asthana]

This happens in India too, where profs exploit the students, and we simply call it exploitation.

Since you are in academia, it may be a good idea for you to 'use' students around you to do innovative projects. May be, even a FFT+NN metal detector !!!!

Yes, definitely! I work in academia and have witnessed too many times "big professors" using students as cheap workforce with little to no credit for their work, hence my comment on "using students".



Definitely, this is probably one of the best approach. In fact, the more I read about PI and the more I'm convinced it's the same thing as TDEM or TEM, just with a different goal and scale. There must be research teams working on optimizing TDEM who developed signal processing techniques that can be borrowed by PI MD.

Funny enough, one of the papers shared by Teleno is from Zonge. We rented a TEM system from them once, it was terrible! Their customer service was even worse...

[Atul Asthana]

Thanks for the patents Carl.

I guess, the difference between the PI an VLF technologies/methodologies is narrowing.
Immaterial of the fact that the pulse is single or repeated, the processing could be either in time domain or in frequency domain.

I'd go a step further, and sample during the transmission, the transmit pulse shape, and collapsing field waveform, itself carries a lot of information. This can be better utilised with an FFT. THis waveform is too complex to be easily interpreted in terms of sums-of-different-exponential-decays or any kind of curve fitting algorithm.
My feeling is that a Neural Network approach will do a better job in analysing FFT output.

However using just first 3-4 frequencies in present day VLF-multi-frequency circuitry/software, should be able to exceed in performance than present day detectors.

[Atul Asthana]

I have a faint feeling Carl, that you are some how connected to some MD company.

Although it's technically VLF and not PI, the Tarsacci metal detector probably does something similar. The TX voltage is a square wave so the RX signal consists of exponential responses, and they are demodulated in the time domain and curve-fitted to determine what's going on. There is a patent for it, US10969512. I see he also has a new patent application, US20220268961, but I haven't read it yet.

[Carl-NC]

Faintly connected with First Texas Products (Fisher/Teknetics/Bounty Hunter).

[Atul Asthana]

I have requested one of the embedded engineers in my network to write code for ESP32, running at 240 MHz. From what Ive read, the fft shouldnt take more than a milli second.

The ESP32 will send a pulse of width controlled by a pot, and take 64 samples starting from the start of pulse, at 5 micro seconds interval for 64 samples, do fft of these 64 samples, and send the packet over blue tooth.

I plan to ask an android developer to just do the basic processing : average out the response over 4 pulses as target, and arunning average over 256 pulses as ambient.
Show the result on the tablet/mobile, without any target characterisation.

Let us see, how much time these developers eill take, if and when they are free from thair livelihood jobs, and whats the result.

I dont expect quick results, since they will probably start only after Deepawali (https://en.m.wikipedia.org/wiki/Diwali), our festival, celebrating win over evil mindsets.