chipKIT uno32 Arduino MD experiments

[hdphilip]

well here's a little progress on my Qed experiment

i finally got the smps power supply to sync with the tx pulses, i think a have a bit more work to do on it but,
so far so good. i did change the in and out on the IDF604, to make it a little more freindy with programming.

I hooked the qed power supply board up to my other metal detector experiment board, i even connected my minelab coil up to it, just to get a feel of things to come.

yup, the power supply turns off during sampling, or where ever i wish.........


Philip

[hdphilip]

power supply notes,
very tough, through all of my testing didn't burn up any components, which is a new record for me, even while programming, their were many times I'd look at my power supply in current limited condition, parts were hot, but let them cool off, and back at it.
I'm going to also build a switching power supply based on the LT1372 or LT1377, and give it a go.
why?
I think this chip can do a better job at controlling a power supply than me.
features.......
Shutdown and Synchronization
The dual function S/S pin provides easy shutdown and
synchronization. It is logic level compatible and can be
pulled high, tied to VIN or left floating for normal operation.
A logic low on the S/S pin activates shutdown, reducing
the part’s supply current to 12mA. Typical synchronization
range is from 1.05 to 1.8 times the part’s natural switching
frequency, but is only guaranteed between 600kHz and
800kHz (LT1372) or 1.2MHz and 1.6MHz (LT1377). At
start-up, the synchronization signal should not be applied
until the feedback pin is above the frequency shift voltage
of 0.7V. If the NFB pin is used, synchronization should not
be applied until the NFB pin is more negative than – 1.4V.
A 12ms resetable shutdown delay network guarantees the
part will not go into shutdown while receiving a synchronization
signal.
so with this chip i'll be able to use simple shutdown, at first, then down the road
i'll be able to change the software to do full synchronization, I'll wire it to an "OC" pin
on the pic32.
question or comments welcome!
Philip

[oroplata]

Watching this thread with interest. Looking forward to contributing fairly shortly - just have another Arduino-based project I have to concentrate on first. Can't let myself get side-tracked!

Keep up the great work!

[hdphilip]

Great!

having people instrested makes a difference,



Philip

[hdphilip]

here's something about the QED power supply synchronization if been thinking about, it sez in bug's patent "cycle time is 4,8,16 times mains freq,

so, I been playing with my timer spreadsheet, seems i can't do it with the way the chipkit is setup, here in north america our mains is 60 cycle, which didn't go into 80 megs very well,
but it seems i might be able to change a muiltiplier in the clock osc, if it works, i'll run the chipkit at 72 mhz, which would get me to 960Hz cycle time, which is a good place to start experimenting with.


here's his's patent, (already been posted on the internet)qed patend markup 1.pdf i was marking up, just trying to get some ballpark ideas on timining. i used my cad program, all it had was mm so i call it ms

Philip

[hdphilip]

here's something about the QED power supply synchronization if been thinking about, it sez in bug's patent "cycle time is 4,8,16 times mains freq,

so, I been playing with my timer spreadsheet, seems i can't do it with the way the chipkit is setup, here in north america our mains is 60 cycle, which didn't go into 80 megs very well,
but it seems i might be able to change a muiltiplier in the clock osc, if it works, i'll run the chipkit at 72 mhz, which would get me to 960Hz cycle time, which is a good place to start experimenting with.


here's his's patent, (already been posted on the internet)qed patend markup 1.pdf
i was marked it up, just trying to get some ballpark ideas on timining. i used my cad program, all it had was mm so i call it ms

that's it for now, the cat just "chased me" out of the room

Philip

[fixstuff]

timing and other questions

Hi,
I've been monitoring this and similar threads on the QED. Really interesting stuff. I took your timing numbers and converted them to msec and usec if anyone is interested. See attached. I have a couple questions though.

A/D questions:

1) During the small sample time, is 15-18 samples a reasonable number for the A/D input? I'm guessing at 100ksps for the 12 bit A/D, thats 1uSec per sample. With ~20uSec to work with, 15-18 samples.
2) During the large sample time, is ~180 samples a reasonable number for the A/D input?
3) Can all of the small sample averages simply be averaged, or is a rolling average better (say 3 sample averages)?

Switched Mode power supply questions:

1) What frequency would this be at? 100Khz?
2) Can this run independently (open-loop), or does the uC need to monitor the feedback and adjust the PWM in order to maintain a certain voltage?
3) Does the cycle for the metal detection sequencing need to be synchronized to the decay period of the SMPS PWM signal to avoid harmonics, or can it be run independently?4)

Coil questions:

1) What is the transmit frequency of the coil?
2) Is the coil compatible with other commercial coils?

Sorry if these are dumb questions. I'm still trying to wrap my brain around this. Is there any chance of getting real timing numbers from the actual source code? I realize the source is locked up in a vault somewhere, but timing numbers would really help - especially if the entire timing diagram in the patent app is not to scale (which means all these numbers are bogus)

Thanks

[hdphilip]

hello fixstuff,

excellent questions,

Hi,
I've been monitoring this and similar threads on the QED. Really interesting stuff. I took your timing numbers and converted them to msec and usec if anyone is interested. See attached. I have a couple questions though.

A/D questions:

1) During the small sample time, is 15-18 samples a reasonable number for the A/D input? I'm guessing at 100ksps for the 12 bit A/D, thats 1uSec per sample. With ~20uSec to work with, 15-18 samples.
2) During the large sample time, is ~180 samples a reasonable number for the A/D input?
3) Can all of the small sample averages simply be averaged, or is a rolling average better (say 3 sample averages)?

here's something to think about,

let's not use a A/D converter in the traditional sense,
when i read about the QED, from the schematics and patents,

here's what how i understand it,

through my experiments usuing my 16 bit adc, it worked ok, but still, i was only reading 1 target sample for 1 conversion.
problem is, the target sample is very very weak, so, what about usuing a 24 bit ADC ? same problem, 1 target sample for 1 conversion. and they still
have a "ENOB" similar to mine
http://en.wikipedia.org/wiki/Effective_number_of_bits

even when i used running averages, i tried up to a 100, it did help alot, and chudster software help even more, but running averages just dilute the signal peaks,

when it comes to usuing 24 bit adc's moodz is on top of it, alot more than me.

How about bugwhisker's QED idea?


let say we TX 5 times in a row, and we save the target sample in C2, so the target sample voltage gets added.
now we need to read the saved voltage, but how?

simple, at a certain point in time, we start a counter in the microprocessor, as the voltage drops, due to the discharge, the counter counts up, now, at a certain point, the comparator stops,
we read the counter, that number in the the timer may be better than a 32 bit adc.....................
then you could close T3, drain the cap, time to start over.

what happens next?

you could flip the switch U4, and subtract out the ground balance from the circuit,

does it work?

according to Minelab, it works very well,



anyways that how i see it, and i plan on trying it out as a experiment.

Switched Mode power supply questions:

1) What frequency would this be at? 100Khz?

I'm thinking it would run at a harmonic of the TX pulse period.

2) Can this run independently (open-loop), or does the uC need to monitor the feedback and adjust the PWM in order to maintain a certain voltage?

right now thats what i'm doing, although i'm am going to try to sync a LT1372,

3) Does the cycle for the metal detection sequencing need to be synchronized to the decay period of the SMPS PWM signal to avoid harmonics, or can it be run independently?4)

why not just turn the power supply off ? i bet thats what bug's does....

Coil questions:

1) What is the transmit frequency of the coil?
2) Is the coil compatible with other commercial coils

the transmit period 1/960hz, i'll use my Minelab coil,

Sorry if these are dumb questions. I'm still trying to wrap my brain around this. Is there any chance of getting real timing numbers from the actual source code? I realize the source is locked up in a vault somewhere, but timing numbers would really help - especially if the entire timing diagram in the patent app is not to scale (which means all these numbers are bogus)

great questions,
the more you think about it, the less need for the source code.

who knows if those number are to scale, seems like a good starting point.

Philip

[Davor]

even when i used running averages, i tried up to a 100, it did help alot, and chudster software help even more, but running averages just dilute the signal peaks

You can as well ignore peaks on the basis of irrelevance. Your target response is centred at about 10Hz sweep-related target response, so a hundred samples at 1kHz repetition is just about right. Running average is a perhaps the best compromise between time and frequency response. And is dead simple too. You could experiment with some windowing functions, but somehow I don't expect much improvement.

[fixstuff]

hello fixstuff,

...
let say we TX 5 times in a row, and we save the target sample in C2, so the target sample voltage gets added.
now we need to read the saved voltage, but how?

simple, at a certain point in time, we start a counter in the microprocessor, as the voltage drops, due to the discharge, the counter counts up, now, at a certain point, the comparator stops,
we read the counter, that number in the the timer may be better than a 32 bit adc.....................
then you could close T3, drain the cap, time to start over.

what happens next?

you could flip the switch U4, and subtract out the ground balance from the circuit,

does it work?

according to Minelab, it works very well,
...
Philip

Wow, that completely cleared that up for me. I think that is pretty interesting, since a lot a microcontrollers use this principle as a simple way to provide an a/d without the a/d. I really hadn't even thought about that when applied to this application.

That gives me some further questions on the tx pulse and the snubber pulse. How do they work together?

I guess first off, I would turn on the switched mode power supply, until it charged up to say ~7V or so.

After it charged to that, I would turn it off.

Then, I would turn on the tx pulse for X time (say 43.633uSec) then turn it off, and then turn on the snubber for Y time? (say 7.125uSec).

Is the snubber used to dampen the tx? The BEMF pulse on the waveform looks to have a flat top. Is this from the snubber on time? With no snubber, I'm guessing it would be much higher if true.

Assuming that's how the snubber works, after a small delay of 21.962 uSec, U4 would only be turned on to switch side A during the sample period of 20.38 usec. This would ultimately charge C2 a certain amount. U4 is then turned off. At this point the mcu counter can be used to count down a number, call it sample 1. After the counter stops, T3 is enabled, resetting C2 back to 0.

After this the process repeats. After five loops (we now have 5 samples), a longer tx pulse is used (69.237 uSec) with U4 switched to B after the large delay of 52.796 uSec. U4 is turned on for 167.115 uSec (large sample time). After this we have samples 1-5 representing the signal, and sample 6 representing the GB?

If its another way, the first 5 samples are simply used to charge the same instance, with each adding to the previous. After a total of 5, they as a sum correlate to the large single GB one. Is it this way instead? From your post I'm assuming this one. How does the 5 cumulative charges work? C2 charges over time, so I get that part, being C2's charge gets bigger and bigger. What does this do, having 5 small charges, as opposed to one big charge? Does it ignore some side affect of a single long charge? Is that what the two differences between method A and B are? Or am I way off in left field.

Anyways, during the processing time I turn the switched mode power supply back on.

As a side note, would anyone be interested in a flow chart of this sequence? Its not code, but a flow chart is only one step away from converting to useful code..

[hdphilip]

hello davor,

You can as well ignore peaks on the basis of irrelevance. Your target response is centred at about 10Hz sweep-related target response, so a hundred samples at 1kHz repetition is just about right. Running average is a perhaps the best compromise between time and frequency response. And is dead simple too. You could experiment with some windowing functions, but somehow I don't expect much improvement.

I agree about the windowing functions as far as improvement.

With the Qed, windowing may be more important.

When i was experimenting with the Adc's, I did try the internal 10 bit adc, i wanted to compair it to the 16 bit adc, with all things being equal, the 16 bit worked much better.

Philip

[hdphilip]

I guess first off, I would turn on the switched mode power supply, until it charged up to say ~7V or so.

After it charged to that, I would turn it off.

Then, I would turn on the TX pulse for X time (say 43.633uSec) then turn it off, and then turn on the snubber for Y time? (say 7.125uSec)..

pretty much, although, if you notice on the QED, theirs a signal that feeds back to the micro, SNUB_MON, that may feed adc input, so, sometime after TX, you control the snubber, and read voltage and adjust accordingly, That peice of code i'll be doing trail and error looking at waveforms on the scope.

After this the process repeats. After five loops (we now have 5 samples), a longer TX pulse is used (69.237 uSec) with U4 switched to B after the large delay of 52.796 uSec. U4 is turned on for 167.115 uSec (large sample time). After this we have samples 1-5 representing the signal, and sample 6 representing the GB?

If its another way, the first 5 samples are simply used to charge the same instance, with each adding to the previous. After a total of 5, they as a sum correlate to the large single GB one. Is it this way instead? From your post I'm assuming this one. How does the 5 cumulative charges work? C2 charges over time, so I get that part, being C2's charge gets bigger and bigger. What does this do, having 5 small charges, as opposed to one big charge? Does it ignore some side affect of a single long charge? Is that what the two differences between method A and B are? Or am I way off in left field.

humm, I haven't looked into it that far, but i'm thinking your on the right track.

I'm think my main goal would be to see the performance of the circuitry without ground balance, and then start to experiment with the ground balance ideas without it effecting the over performance.

As a side note, would anyone be interested in a flow chart of this sequence? Its not code, but a flow chart is only one step away from converting to useful code..

a flow chart would help a lot, do you have a copy of the patents?


Philip

[fixstuff]

...
a flow chart would help a lot, do you have a copy of the patents?

Philip

I believe I do, but there really isn't much detail compared to the usual patents I've seen (it just amazes me the patents that are awarded with so little supporting info).

I have the high res schematics, the patent info, and that's about it. I pulled up all the data sheets on all the ICs used, so I understand the circuitry pretty well now. Now I'm just trying to correlate those with the block diagrams, and the sequencing of the circuits.

I mainly just want to figure it out on my own. I long ago gave up asking about source code most of the time, since it's held onto so tightly most of the time. I think what your doing is probably about the only way to figure it out.

If I ever DID get the source code however, I would think the first step would be to break down the process into the flowchart so that anyone with another flavor of microprocessor could convert it over easily. I would like to show each piece individually, that way people could pick and choose which parts of it they use, and which they don't. Take that with a board for each function, and you could have building blocks. Maybe flowcharts would make it easier to avoid the wrath of the commercial companies.

On a side note, has anyone posted the schematics in eagle? I am about to start doing it myself, but if there was already some work on that regard, I could pick up from there.

[hdphilip]

fixstuff

check your email.

Philip

[greylourie]

........................................, I would think the first step would be to break down the process into the flowchart so that anyone with another flavor of microprocessor could convert it over easily. I would like to show each piece individually, that way people could pick and choose which parts of it they use, and which they don't. Take that with a board for each function, and you could have building blocks. Maybe flowcharts would make it easier to avoid the wrath of the commercial companies.

On a side note, has anyone posted the schematics in eagle? I am about to start doing it myself, but if there was already some work on that regard, I could pick up from there.

Hi fixstuff
Did you ever get to breakdown the process into a flowchart ? It would really help to understand how it works.