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

[Aziz]

Your offset looks like the sampling offset due to latency time of TX and RX.
So it can't be a DC offset.

[Aziz]

The first DSP Lock-In-Amp test - sine 12.5 kHz

Offset and Magnitude in samples
Angle in Deg

Rx-Tx direct connected, calculation - 300 frames per sec

Hi TheWizard,

nice, really very nice. So you can sweep the coil at 1 m/s and you will have every 3.3 mm a measurement. This is very good!
You have probably small window/frame size. How much? And how much is your sampling rate and bit depth (12)?

Your offset, is this the DC offset of the A/D converter?
Now I am looking forward to your progress.

Regards,
Aziz

[TheWizard]

The first DSP Lock-In-Amp test - sine 12.5 kHz

Offset and Magnitude in samples
Angle in Deg

Rx-Tx direct connected, calculation - 300 frames per sec

[Max]

Hi,

what about a speaking MD? I could imagine that a nice girl voice could tell me when I detect some gold:
"Stop Darling! You have found, whatever you seeked for!"

Even this is possible! A laptop with second sound-card (cheap, 16 Bit, USB or so) must be used also for the beeping. Even this runs without any problems. I have tested this many years ago, running the application with two sound cards: one for data processing and another for signal audio output.

Aziz

Now I need more whiskey....

[Qiaozhi]

Hi,

what about a speaking MD? I could imagine that a nice girl voice could tell me when I detect some gold:
"Stop Darling! You have found, whatever you seeked for!"

Even this is possible! A laptop with second sound-card (cheap, 16 Bit, USB or so) must be used also for the beeping. Even this runs without any problems. I have tested this many years ago, running the application with two sound cards: one for data processing and another for signal audio output.

Aziz

This is like Treasure Talk from Garrett:
Garrett's exclusive voice function found only on the GTI 2500 that audibly announces various settings, adjustments and target information continuously or on demand.


I have yet to find someone who actually likes this feature.

[Aziz]

Hi,

what about a speaking MD? I could imagine that a nice girl voice could tell me when I detect some gold:
"Stop Darling! You have found, whatever you seeked for!"

Even this is possible! A laptop with second sound-card (cheap, 16 Bit, USB or so) must be used also for the beeping. Even this runs without any problems. I have tested this many years ago, running the application with two sound cards: one for data processing and another for signal audio output.

Aziz

[Aziz]

Hi TheWizard,

it is a nice DLL, but costs a bit (shareware). Anyway, there are many sample source codes for this DLL library.

I made lots of noise experiments with lock-in amp. The result is incredible. Even when the signal is totally buried in more noise, the lock-in amp is quite stable. So I am going to develop my own digital lock-in laptop MD.
With this technique, I will be able to measure very low signals with ultra-balanced search coils. This would bring more depth sensitivity.

Aziz

[TheWizard]

BASS.DLL(http://www.un4seen.com/) is "free" audio library for MP3, MP2, MP1, OGG....
I use before for Stream Media Web Radio Player (XP widget)

[Aziz]

Hi,

I found some interesting measurements using sound-cards. There are some source codes and executables, I think in Delphi + DLL or so. If someone wants to try some coding or testing sound-cards, just check the site out.

Using the DLL any programming language can be used!

Web ist only in german.

http://www.michaelgaedtke.de/Key_Messen.htm

Aziz

[Aziz]

Now I'm playing around with the lock-in amp for a while. Did some additional noise on output channel (inductively -> input channel) and will also do some on input channel directly soon.
First tests showing a quite stable behaviour even if I take very much noise in the data (noise > signal).
Then, I will test a ultra-balanced search coil with µV range of signal rest.

Aziz

[Aziz]

You have RX frame buffer - If you know who sample index is TX cross-zero
n is TX cross
it is I part RX[n]
Q part is RX[n+"90deg"]
Just use and plot this two samples and look dynamic range of phase...

in my first picture phase error is ~ 1 deg for 1/4 input signal

The problem arises, when your index-offset for +90° has some fractional part: e.g. 5.4 (you will have then 5 for index offset).
To avoid this, one could use multiple of 4 samples per wave. This is just a limitation to frequency of interest selection. What happens, when your coil is not exactly matched to a particular frequency? You will lose some sensitivity. Anyway, a VLF/TR/Off-resonance digital-MD up to 24 kHz on 96 kHz SR should be possible. Even with 44/48 kHz and 16 Bit. It would be better, to have a 192 kHz SR, but my sound-card has only 96 kHz input. Output is of course 192 kHz but I am using 96 kHz for both.

Better accuracy is given on lower frequencies: e.g. 5-15 kHz.

So, who want's to develop a laptop MD now?
Need more proof?

Anyway, then implement this on a DSP/µC. It's almost same.

[TheWizard]

You have RX frame buffer - If you know who sample index is TX cross-zero
n is TX cross
it is I part RX[n]
Q part is RX[n+"90deg"]
Just use and plot this two samples and look dynamic range of phase...

in my first picture phase error is ~ 1 deg for 1/4 input signal

[Aziz]

OK,
but (if sync is good) you need only 2 samples at 0 and 90 deg (RxBuff[i], RxBuff[Q]) (I&Q) for ArcTan and Sqtr - why is FFT...

Less work at the moment, because I had allready implemented the FFT-Filter. For quick implementation it is ok. But for speed enhancements, LP-FIR-Filter/Averaging + decimation should be used. But on the Laptop/PC there is much more computation power. I can acquire 94 Frames/s of signals + signal generation for output + Ref-I +Ref-Q, can process 94 Frames/s doing much more computation on approx. 40(!) channels, and displaying only up to 35 Frames per second (lame Win32-GDI). My frame size is 1024 samples at 96 kHz SR (24 Bit). There must be a good thread-implementation of the signal acquisition/transmission, signal processing and displaying the results. So I have up to 8 threads running for the application at the same time. Of course, of different priority levels. If there is not enough power, I discard some frames for displaying (less important). But all frames are processed just in real-time. This is, why I like the Laptop MD. There is no other DSP/µC that can do this! Of course, I am using multi-core CPU's automatically!

By the way:
Phase offset for Ref-Q of +90 degree should not realised as an index-offset to Ref-I! Otherwise, you will lose dynamic range of phase!
Ref-I(t) and Ref-Q(t) should be generated independently by a generator, having Ref-Q-phase +90 degree of Ref-I-phase.
Also, you should use pure sine-reference signal (no TTL wave forms). Then you will be able to get more accuracy.

Aziz

[TheWizard]

OK,
but (if sync is good) you need only 2 samples at 0 and 90 deg (RxBuff[I], RxBuff[Q]) (I&Q) for ArcTan and Sqtr - why is FFT...

[Aziz]

Here is the main c-code of Lock-in amp for all others:

//------------Lock-in-------------
if (pMetDet->Config.bLockin)
{
fsample fi,fq, *pInp,
*pRefI,*pRefQ,
*pI, *pQ,
*pIOut, *pQOut,
*pMag, *pPh, *pF, *pIm, *pRe;
double f;
int i;

// Wave-Frame channels für Lock-in bereitstellen
//
pInp = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_INPUTL);
pRefI = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_REF_I);
pRefQ = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_REF_Q);
pI = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_I);
pQ = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_Q);
pIOut = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_IOUT);
pQOut = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_QOUT);
pMag = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_MAGNITUDE);
pPh = Wave_GetChannelPtr(pMetDet->pWaveSig, METDETCH_LOCKIN_PHASE);

// Buffer for FFT-Filter
pRe = Wave_GetChannelPtr(pTmpWave, 0);
pIm = Wave_GetChannelPtr(pTmpWave, 1);

// LP-Filter spectrum
pF = Wave_GetChannelPtr(pTmpWave, 2);

// I und Q Mixer
//
for (i=0; i < nFrameSize; i++)
{
f = pInp[i]; // signal(t)
pI[i] = (fsample)(f*pRefI[i]); // X
pQ[i] = (fsample)(f*pRefQ[i]); // Y

} // for i

// LP-filter I und Q
//
memcpy(pIOut, pI, nFrameSize*sizeof(fsample));
memcpy(pQOut, pQ, nFrameSize*sizeof(fsample));
FFTFilter(nFrameSize, pIOut, pF, pRe, pIm, FALSE, &Result.fLockinDCI);
FFTFilter(nFrameSize, pQOut, pF, pRe, pIm, FALSE, &Result.fLockinDCQ);

// R,P of DC Level of I und Q
//
Result.fLockinDCMagnitude = sqrt(Result.fLockinDCI*Result.fLockinDCI + Result.fLockinDCQ*Result.fLockinDCQ);
Result.fLockinDCPhase = atan2(Result.fLockinDCQ, Result.fLockinDCI)*NACH_GRAD;

// Amplitude and Phase
//
Result.fLockinAvgMagnitude = Result.fLockinAvgPhase = 0.0;
for (i=0; i < nFrameSize; i++)
{
fi = (fsample)(pIOut[i]); // Y: sin
fq = (fsample)(pQOut[i]); // X: sin+90 = cos

// Magnitude = sqrt(I*I + Q*Q)
f = sqrt(fi*fi + fq*fq);
pMag[i] = (fsample)f;
Result.fLockinAvgMagnitude += f;
// Phase = atan(Q/I), atan2(Q,I)
f = atan2(fq, fi)*NACH_GRAD;
pPh[i] = (fsample)f;
Result.fLockinAvgPhase += f;

} // for i

f = 1.0 / nFrameSize;
Result.fLockinAvgMagnitude *= f;
Result.fLockinAvgPhase *= f;

Result.dwFlags |= METDETRESULT_LOCKIN;
}
//------------Lock-in-------------

Damn! Format is gone to the hell!