GROUND CONTROL PI CIRCUIT

[Waikiki_Sweep]

Great Job! Not everyone can read assembly codes so freely!

QED is great machine but looks too high-tech for hobbyists.
Starting with some "light" version of QED probably would be interesting for many.
Without Snubber and Gating it looks much as regular PI detector but with advanced
sampling and ground balance.

[maikl]

QED is great machine but looks too high-tech for hobbyists.

It might be interesting... that do ground balance for Surf Pi... it would be interesting for hobbyists...

[6666]

Actually, as I figured out from source code, goes more like this: 60uS EF cancellation sample, followed by 150uS TX pulse. Then, without “hole shifting”sample, goes 10uS delay, then 10uS positive, then GB delay (integrator disconnected). This is variable and used to adjust GB point. Then, 150uS negative, again GB delay, same duration as before, then 140us positive. Optional “hole shifting” is 5uS negative before first sample.

Hi Tepco, thanks for explanation from the QED source code.

[6666]

Actually, as I figured out from source code, goes more like this: 60uS EF cancellation sample, followed by 150uS TX pulse. Then, without “hole shifting”sample, goes 10uS delay, then 10uS positive, then GB delay (integrator disconnected). This is variable and used to adjust GB point. Then, 150uS negative, again GB delay, same duration as before, then 140us positive. Optional “hole shifting” is 5uS negative before first sample.

I dont know how to write real code useing timers and such, just tried an amateur bit banging hack useing great cow graphical basic, dropped s2 5uS hole shifting pulse, but with variable GB width control.

[Taktyk]

You tried to compile and upload the code (orginal assembler code) to dsPIC?

[6666]

Hi Taktyk, I just put together some of my own code purely based on Tepco's description of the pulse train, I used great cow graphical basic and a recycled picaxe 18f14k22 running at 64 mHZ.
Here is a small video.
https://youtu.be/VEGTQ8emWUk

[Taktyk]

It is compatible with source code. It is only experimental platofrma? Do you change the sample efe (60uS) to eliminate efe?

Regards
TAKTYK

[6666]

Hi Taktyk
My code is purely experimental at the moment, the 60uS EF pulse is fixed.

[6666]

You tried to compile and upload the code (orginal assembler code) to dsPIC?

No I am not useing the original assembler code.

[Taktyk]

6666...

This is not consistent with QED - interpretation of the code is incorrect.

T.

[6666]

Hi Taktyk, thanks for the feedback, you say there is a problem with my code in the video, can you please tell me what is wrong with it so I can fix it
my code was based on the description by Tepco, Thanks.

60uS EF cancellation sample, followed by 150uS TX pulse. Then, without “hole shifting”sample, goes 10uS delay, then 10uS positive, then GB delay (integrator disconnected). This is variable and used to adjust GB point. Then, 150uS negative, again GB delay, same duration as before, then 140us positive

I have labeled this picture with pulse train , the camera is not square on to the cro screen.

[Taktyk]

Hi Taktyk, thanks for the feedback, you say there is a problem with my code in the video, can you please tell me what is wrong with it so I can fix it
my code was based on the description by Tepco, Thanks.

Check assembler code - in the main loop...

T.

[6666]

Hi Taktyk, thanks for the feedback
I dont have the QED asm code, I have been watching the work done by Green, he has made an interesting observation with his GB experiment,
he says that the EF pulse width should be equal to (the ground balance pulse width minus the target pulse width), and should track
is that what you are trying to tell me that is wrong with my EF pulse in my code ?

[Taktyk]

Hi, 6666, best to build a system to check


You only change the position of the first pulse (150uS) - the second pulse is immutable (140uS).

It is best to check yourself.

T.

[Teleno]

Please see attached extract from Inside the METAL DETECTOR - Published September 2012

What this boils down to is a 2-point fit of the signal to a 1/t function. Deviations are then interpreted as a "target".

The technique can be extended to more samples, fitting them together to 1/t and using the goodness of fit as indicative of ground/target.

http://math.stackexchange.com/questi...om-data-points A hyperbola takes the form y=k / x. This may be difficult to deal with. So instead, let's consider the reciprocals of our x values as J.M. suggested. For example, instead of looking at (2.5, 0.007713), we consider (1 / 2.5, 0.007713). Then since we have flipped all of our x values, we are looking to fit something of the form y=k / (1 / x) = kx. This can be accomplished by doing any standard linear regression technique.

Goodness of fit would be the mean squared error or other variants

Not that I've done it yet but I just leave the idea floating. Perhaps others already did it and can tell us how it pans out.