Let's made a PC-base metal detector with usb interface !!!

[Tinkerer]

Noise at output

Attached is a noisy output signal. This is the noise level on the signal without the presence of a target.
The total time shown is about 5.35 seconds.
The noise level about 150mV.

With this noise level, the threshold would have to be set at about 160mV, that means that any target signal below 160mV would be totally ignored.

What would be the best way to filter this signal?

Tinkerer

[TheWizard]

Low-Pass?
First you need to know SampleRate, Fstop and Fpass.
Best way is FIR filter....
If you have not CPU speed or DSP core, try some averages filters
or Analog Filter after integrators(if is PI MD)

[Aziz]

Such accuracy is totally useless in the real world. There are so many other factors, which affect the phase-shift returned from buried objects, that the ability to discriminate to even 1 degree accuracy is unrealistic.

That's the opposite: I can start discriminating on more depth. The phase error on synthetic generated signals with sound card is accurate and stable enough. It can be synchronized with the lock-in amp much easier.

In contratry to an external LC osc., the phase error is much bigger. On my LC oscillators (Colpitts and CMOS inverter osc.), I couldn't use the phase information. But for all metal mode, the magnitude of the lock-in amp can be used well and it delivers more depth sensitivity.

I made some other experiments on my detector yesterday. To make the search coil as easy as possible, it can also be driven without capacitors on both sides (RX and TX). There is less sensitivity loss but the big advantage is changing the operating frequency and also the possibility of making a spectrum analysis (multi-frequency analysis).

My sound card has an internal noise of ~200 nV rms. This noise level is increased with external components (coil, amplifier, cable, etc.). I measured a noise of 1-3 µV with external components. Now I will start some experiments with as less external components as possible. The amplifier will be implemented in analog manner: using Faraday's law by increasing the number of turns for RX coil winding while keeping the resistance as low as possible (high resistance causes more noise).

More noise tests will be presented soon. So I can say you exact, how many nano Volts can be detected really (some 100 nV range).

Regards,
Aziz

[Aziz]

Attached is a noisy output signal. This is the noise level on the signal without the presence of a target.
The total time shown is about 5.35 seconds.
The noise level about 150mV.

With this noise level, the threshold would have to be set at about 160mV, that means that any target signal below 160mV would be totally ignored.

What would be the best way to filter this signal?

Tinkerer

Very simple answer: Using the lock-in amplifier.

It does a very narrow band filter and ignoring the noise on other frequencies. This is the reason why I am saying, that nV to µV signals can be detected.

Aziz

[amtech2005]

Let's make a PC-based metal detector with usb interface !!!

GOOD SUBJECT !!! I FOUND THESE HAND WRITTEN FILES LAST NIGHT , IT WAS POSTED ON GEOTECH ABOUT 5 MONTHS AGO ??? I HAD TO COMPRESS THEM TO ZIP FILE TO MAKE THEM FIT . I DO NOT AGREE WITH WHAT WAS SAID ON FILE P1, I BELIEVE THAT NOISE HAS FREQUENCY CHARACTERISTICS AND HAS HARMONICS ,AND IS SOMEWHAT POLARIZED. !!!!!!!!! IF YOU HAVE A METAL DETECTOR , IT SENDS A FREQUENCY AND A DC VOLTAGE WITH HIGH CURRENT TO THE SEARCH-COIL AND IF THE DETECTORS MAXIMUM FREQUENCY IS 10 KHZ. CAN YOU PUT A DC-10 KHZ BANDPASS FILTER CIRCUIT AT THE SEARCH COIL ??? OR JUST A 10 KHZ FILTER RIGHT ON THE TX COIL ?? WHAT I AM SAYING IS CAN GET RID OF ALL NOISE WITH ADDING FILTER CIRCUITS AT THE SEARCH-COIL BEFORE IT HEADS UP THE COAX CABLE ??? .............................EUGENE

[TheWizard]

Aziz,

your soudcard noise maybe is ~1549nV for bandwidth 96kHz.....
and this is without pre amplification....

[Aziz]

Aziz,

your soudcard noise maybe is ~1549nV for bandwidth 96kHz.....
and this is without pre amplification....

I am not focusing on the full bandwidth. No, I am just measuring the noise at the lock-in frequency (max. 24 kHz) at the full internal line-in amplification. This is what counts.
There is lots of more noise between 0 .. 1000 Hz. I can see the cross-talk of the DC-DC converter in the PC. On this frequency + multiple harmonics, the noise is higher and should not be used for MD. Anyway, on the frequency spectrum there are lots of frequencies for operation.

It is interesting to see also the cross-talk signal from the sound card output to short circuited line-input. So I can measure the cross-talk level and the noise. Also the DC-offset and noise of the internal sound card A/D converter.

So there is more accuracy when the sound card is quite good. There are some very good sound cards available today. At least as a PCMCIA, PCI-express cards or USB sticks. For a stand-alone project, look for some convenient codec chips. In the last years, they made big steps towards low noise and accuracy.


BTW TheWizard,

look to your lock-in output (R and P output). It doesn't looks good. Please check your code. Your low-pass filter has a big cut-off frequency. Increase your time constant. I have more stable results.

Regards,
Aziz

[TheWizard]

Last my pictures - RE and IM is raw data from demodulator without any filters...

now is filtered....

[TheWizard]

some experiments.....

[Tinkerer]

Noise in output signal

Attached is a noisy output signal. This is the noise level on the signal without the presence of a target.
The total time shown is about 5.35 seconds.
The noise level about 150mV.

With this noise level, the threshold would have to be set at about 160mV, that means that any target signal below 160mV would be totally ignored.

What would be the best way to filter this signal?

Tinkerer

Sorry about the confusion. This is basically the DC output after amplification x 56,000.
The signal, without a target, from the RX coil was pre-amplified with a gain of 560, sampled, the samples averaged then run through the difference amplifier and then amplified x100.

Now I would like to filter and amplify x 100 again, then feed the VCO.

Tinkerer

[Qiaozhi]

That's the opposite: I can start discriminating on more depth. The phase error on synthetic generated signals with sound card is accurate and stable enough. It can be synchronized with the lock-in amp much easier.

In contratry to an external LC osc., the phase error is much bigger. On my LC oscillators (Colpitts and CMOS inverter osc.), I couldn't use the phase information. But for all metal mode, the magnitude of the lock-in amp can be used well and it delivers more depth sensitivity.

I made some other experiments on my detector yesterday. To make the search coil as easy as possible, it can also be driven without capacitors on both sides (RX and TX). There is less sensitivity loss but the big advantage is changing the operating frequency and also the possibility of making a spectrum analysis (multi-frequency analysis).

My sound card has an internal noise of ~200 nV rms. This noise level is increased with external components (coil, amplifier, cable, etc.). I measured a noise of 1-3 µV with external components. Now I will start some experiments with as less external components as possible. The amplifier will be implemented in analog manner: using Faraday's law by increasing the number of turns for RX coil winding while keeping the resistance as low as possible (high resistance causes more noise).

More noise tests will be presented soon. So I can say you exact, how many nano Volts can be detected really (some 100 nV range).

Regards,
Aziz

When you eventually leave the lab and go into the real world, you will discover that target distance, composition and orientation will dramatically change the phase-shift at the receiver. Not to mention ground mineralization. Relying on an accurate phase-shift measurement by using perfect homogenous, or even simulated targets, will not work with real buried objects. But I guess you'll need to figure this out for yourself.

[Aziz]

Interesting lock-in amplifier experiment:
Measurement on line-input (no external amplification) with a line-output driven TX coil (83 turns, ca. 20 cm diameter).
A very small coil with one winding is placed on the center of the TX coil.
The signal is measured on 24 kHz (32 µV). The coil is then made 1 mm larger. The signal increases by 1 µV to 33 µV rms. Within the range of noise roughly ± 0.2 µV. The measured voltage is the mean value of some time constant (ca. 200 ms). The bigger the time constant, the better the accuracy.

This is showing the possibility of using lock-in amplifier.
Aziz

[Aziz]

When you eventually leave the lab and go into the real world, you will discover that target distance, composition and orientation will dramatically change the phase-shift at the receiver. Not to mention ground mineralization. Relying on an accurate phase-shift measurement by using perfect homogenous, or even simulated targets, will not work with real buried objects. But I guess you'll need to figure this out for yourself.

Yes of course. The field test will show me some surprising drawbacks. I have to wait until I have a laptop. This could be real problem. Anyway, I like problems.
Regards,
Aziz

[Aziz]

Hi guys,

as bbsailor said, one could take different time constants for different measurement modes:
- fast/middle swing mode
- slow motion mode for pin-pointing

With the slow motion mode, the sensitivity and the stability increases much. Also the discrimination accuracy. This mode will have a bigger time constant of the lock-in amplifier (in range of 100 .. 2000 ms or more).
Keep tuned.
Aziz

[Aziz]

Hi,

power consumption of the laptop MD is getting to be an very important issue. To maximize the battery life, there must be taken some steps:
a) switching unused hardware off:
- CD/DVD drive
- harddisk (after some period)
- WLAN
- unused ports (rs232/printer ports, etc., modem, network adapter)
- graphics display output (only multi-tone sound output for detection)

b) Using Windows XP as an operating system with minimum required software packages (switching all unused software services off)

c) Reducing physical memory (2GB->1 GB or less)

d) Reducing processor speed (we have enough computation power)

e) Programming in a sophisticated and optimized manner.

The laptop MD software has to take all this stuff into consideration. So we can achieve approx. 50 - 100% more battery life (up to 3-6 hours). For the ergonomic issue, it doesn't makes sense, to look allways to the display of the laptop. The laptop should be mounted somehow backpack. For a steady use of MD, the MD software can be started automatically during boot-up process. On the line-in/mic-in, a defined signal can be coded for some switches (pin-point/modes/etc.), which can be detected by the software.

Aziz