DSP in commercial prospecting and treasure hunting metal detectors ... and GPZ 7000 speculation

[scrungy_doolittle]

Cortex M4 is very attractive for even high-end metal detectors, lots of parts are around 100uA/MHz core power.

Yeah. The GG group is using a STM32F411. The nucleo board is cheap. 10.33

[Carl-NC]

Which part?

Cortex M4 is very attractive for even high-end metal detectors, lots of parts are around 100uA/MHz core power.

[scrungy_doolittle]

No, not really. Until recently, heavy DSP was too expensive, both in terms of price and power. ARM has changed that, and now for $3 and 10mA I can get what used to cost $30 and burn 100mA.

Which part?

[scrungy_doolittle]

Hi Wirelessguy,

I've looked into this just a little but found that the processor speed is the limitation. Many attempts have been undertaken and a uC can be very useful in providing programmable timing or high level processing. Even an Arduino can do some nice things. For adequate sampling microsecond pulse lengths however you need not only very fast ADC but also process this information in real time. I'm working with the LPC1768 and although this is a powerful microcontroller it will have no chance in adequately real time processing the fast signals.
In my opinion the only way to go is utilizing uC in combination with FPGA's. Technology is around for a while but not really matured in the metal detector hobby space. We probably need people like yourself to progress the hobby community in this space. I can program a uC but FPGA programming is something I have to look into. When searching on google for FPGA oscilloscope you will find some interesting leads.

Cheers mate, Jacob

I think you are quite wrong. The STM32F405/7 for example, has a 2.2 megasample pipelined triple ADC at 12 bits. The pic24FJ128GC010 has a 10 Megasample ADC. Either of these are adequate. Using DMA to directly transfer the data, and running at 168 mhz, there is more than enough processing to do multiple filters and analysis.

[Davor]

Thanks for encouragement. I might try to go back there a little, at least to give my son a proper introduction to the art.
It is like blonds - we all like them until they disappoint.

[Carl-NC]

I lost my liking of FPGAs and DSPs for a different reason - as soon as you become comfortable using one it disappears from the market.

ARM is changing this as well. Compatible core code across hundreds of devices, and written in C, makes porting much easier. All the vendors are now putting tons of effort into common frameworks, so even porting peripheral code is becoming trivial.

Edit: Even PIC is getting their act together; take a look at their new Harmony framework, quite nice and well-done.

[Davor]

I lost my liking of FPGAs and DSPs for a different reason - as soon as you become comfortable using one it disappears from the market. In a case of equipment manufacturer it is more or less OK if the supplier gives you heads up on the vanishing act, so you may stock enough of them knowing how many pieces of equipment you sold. After all, you are a reputable manufacturer that is capable of keeping your users' equipment running past the average smartphone lifetime. But in hands of an amateur it is a disaster. Altera did the vanishing act (late 90's) without notifying the small OEMs and it was a disaster. In my case that was a last straw.

That and power hogging.

Now that low power is a thing I might change my mind, but still analogue is the love of my life. With a little imagination you can do most of the functions in analogue (not all).

In a way VLF devices are lock-in amplifiers, as they do PSD and play with separate phases. Using PSDs in quadrature may drastically reduce AD requirements for your micro, and the complexity of DSP operations becomes so simple that all the magic can be performed in a simple microcontroller.

By a stretch of imagination the PI can be treated similarly, so the way to go is a middle path. Or instead you can use some monster AD/DSP and do it directly to the same end.

[Carl-NC]

The only reason no one is truly using the power DSP gives you is because most of the manufacturers don't have a clue where to start or what to do OR how to do it properly.

No, not really. Until recently, heavy DSP was too expensive, both in terms of price and power. ARM has changed that, and now for $3 and 10mA I can get what used to cost $30 and burn 100mA.

[Sean_Goddard]

The only reason no one is truly using the power DSP gives you is because most of the manufacturers don't have a clue where to start or what to do OR how to do it properly.

[Carl-NC]

Carl,

Do you see any potential to use the principals of "lock-in amplifiers" to extract very weak metal detector target signals in the presence of noise? http://www.thinksrs.com/downloads/PD.../AboutLIAs.pdf Chips are now available that can do the lock-in amplifier function.

I believe that Eric Foster's 3KHz TX rate metal detectors integrate about 1500 samples to boost the sensitivity on the RX side rather than pumping more power in on the TX side. This assumes that the sweep speed is controlled and that any potential target is in the TX field long enough to be sampled enough to increase the RX sensitivity in the sync demodulator circuit.

I believe this technology and the new DSP lock-in amplifier chips have potential when adapted for metal detecting.

Thanks

Joseph J. Rogowski

I don't see where a LIA would help any. We're not dealing with a weak signal in the presence of noise, we're dealing with a weak distortion of a really large signal. That is, in a VLF the residual RX signal is around 60dB lower than the TX signal due to an imperfect null. This is typically a few 10's of mV. So we have plenty of signal amplitude, and we even have a reference signal that is perfectly sync'd to it... no LIA needed. If we could achieve an absolutely perfect null, then maybe a LIA could be useful. And, in PI where there is only transient signals, a LIA isn't useful.

Most detectors use target integration to boost up the RX response. As a rough estimate, I use 100ms as the target sweep response width, so at 3kHz that comes to 300 integrated samples. A faster TX frequency usually helps. Eric's GoldQuest runs at 10kHz but has a lower turn-on current so the end results are about the same.

- Carl

[bbsailor]

Golfnut is right, pretty much everyone is doing DSP at some level. Most designs still use analog demods because direct sampling at the speeds needed for really good performance is still a little pricey. But Minelab X-Terra and White's Prizm6T/CoinGT are examples of direct sampling + DSP. Garrett was the first I know of to use a dedicated DSP chip (Analog Devices ADSP) in the GTi series, what, 15 years ago? Now most companies are rapidly moving to ARM cores and doing a whole lot more digital processing than ever before.

Carl,

Do you see any potential to use the principals of "lock-in amplifiers" to extract very weak metal detector target signals in the presence of noise? http://www.thinksrs.com/downloads/PD.../AboutLIAs.pdf Chips are now available that can do the lock-in amplifier function.

I believe that Eric Foster's 3KHz TX rate metal detectors integrate about 1500 samples to boost the sensitivity on the RX side rather than pumping more power in on the TX side. This assumes that the sweep speed is controlled and that any potential target is in the TX field long enough to be sampled enough to increase the RX sensitivity in the sync demodulator circuit.

I believe this technology and the new DSP lock-in amplifier chips have potential when adapted for metal detecting.

Thanks

Joseph J. Rogowski

[wirelessguy]

Thanks all for the feedback and clarifications.

[Carl-NC]

Golfnut is right, pretty much everyone is doing DSP at some level. Most designs still use analog demods because direct sampling at the speeds needed for really good performance is still a little pricey. But Minelab X-Terra and White's Prizm6T/CoinGT are examples of direct sampling + DSP. Garrett was the first I know of to use a dedicated DSP chip (Analog Devices ADSP) in the GTi series, what, 15 years ago? Now most companies are rapidly moving to ARM cores and doing a whole lot more digital processing than ever before.

[golfnut]

most modern machines do use uProcessors with either external or internal atod's - even entry level machines like the garrett 250 have a processor backend. !


A few one liners to help you grab some more info...


With your radio interests you will know that the noise performance is linked to the bandwidth of the Rx channel so as a consequence the IF BW of these machines is pretty narrow, 10's of Hz - to keep the noise down and S/N up - but often done in analog filters with high in band gain, some with AtoD backend others with comparators with adjustable dc ref for sens.

The trade off is a narrow range of sweep speeds the user must stick to - if narrow IF and low noise. Wide IF and you can sweep at any speed, but a bit deaf as S/N not as good.

The IF freq is normally centred on 8Hz ( linked to coil sweep speed )

The in band signals at this freq are ground, Iron, rusty large iron, etc etc...


The Rx channels are zero IF receivers with image at dc. The mixer gating is linked to the Tx at known phase offsets. A DC block cap after the mixer stops the dc offset.

They use one CH to remove the GND signal - it is done by straddling the zero crossings with the mixer gate time to average it out. Remaining signals which grt thru this single phase notch are Iron etc

The other Channel or channels ACCEPT signal phases from wanted alloys.


The sample rate required is faiirly low as say 10Hz signals at 10x oversampling is only 100 times a sec and so a microcontroller would do a reasonable job - thats what some use.


The main in band blocker is GND signal and its phase and amplitude jump about, so one fixed setting isnt great.

Identifying large iron is a chore as it occupies two quadrants and can play Iron or non ferrous tones..

[wirelessguy]

Jacob,

In my opinion the only way to go is utilizing uC in combination with FPGA's.

How sloppy of me. In addition to the two processors above, I forgot to add that the FCC test report includes the presence of a Microsemi M2S010-FGG484 which is a sophisticated chip containing a Cortex M3 with its associated memory and peripherals AND SRAM FPGA. This part is designed exactly for what you propose.

Technology is around for a while but not really matured in the metal detector hobby space.

I'm an ex-ham. (Let my license lapse after I got bored with it.) Seems to me the MD hobby space seems to never have taken off like decades of ham radio DIYers. Nonetheless, my puzzlement is over the professional production industry.

wirelessguy