DDS (Direct digital synthesis) based feedback control with phase lead (option1) or just uC (option2) / Tesla Coils / Forums

Forums

4hv.org :: Forums :: Tesla Coils

« Previous topic | Next topic »

DDS (Direct digital synthesis) based feedback control with phase lead (option1) or just uC (option2)

Move Thread

LAN_403

axelro

Sun Sept 04 2011, 08:48PM

Registered Member #3640 Joined: Sat Jan 22 2011, 12:16PM
Location: Germany close to Heidelberg
Posts: 39

Hi,
there's a wild idea going around in my head since a couple of days, so I thought I'd validate with you:
Â
Using a DDS chip (like Analog's AD9833) as part of the feedback loop of a tesla coil. In a way, this is nothing else then a digital PLL, but with some control advantages. The idea is:
Â
1. Use a mikro controller =uC via it's interrupt pin to measure the frequency of the primary, e.g. by counting up a 32bit register with max speed from one interrupt to another. The signal would come from a CT circuit no different then typically applied.
2. Calculate the DDS frequency tuning word from this and program it to a DDS chip to adjust it's output frequency.
3. At the same time (time of the interrupt) calculate a phase shift word and program it to the DDS chip phase register to shift the phase to match the output of the DDS chip with the zero current timing, taking the delays of the driver/DDS/uC in consideration.
Â
So far, this should take care of taking the input from the CT circuit and provide an updated frequency with the correct phase shift to compensate the delay of the driver circuit. If the uC had a clock frequency of 20Mhz (like an Atmel 8bit uC), one could do this with a 50ns resolution. As the processing time of the uC is deterministic, it just adds to the overall delay and can be compensated with the phase shift word.
Â
4. The next challenge could be to switch the current carrying IGBT off early by a predefined value. This could be achieved by the preprogramming of an appropriate waveform, in case the DDS chip supported this. Actually, this is somewhat of a weak spot, as this seems unreasonably complex. Another approach could be to phase shift appropriately in above step 3 to achieve the early switchoff.

Ok, this is far from being mature, however it seems, this hasn't been done yet, so it's interesting.

Call for discussion, please.

Edit: The uC idea is explained a little further down the thread.

Dr. ISOTOP

Sun Sept 04 2011, 10:53PM

Registered Member #2919 Joined: Fri Jun 11 2010, 06:30PM
Location: Cambridge, MA
Posts: 652

The problem with all these software-based controllers is that they're highly susceptible to noise. Even the hardiest of micros is not going to enjoy running in the EMI-filled environment of a Tesla coil. A bit of noise, suddenly the output of your controller latches high, and poof goes the bridge =(
Besides, the Hammertone Predikter + flip-flop based driver already achieves near-perfect ZCS...no need to improve what's already perfect.
The advantage of a software-only controller is flexibility; it lets you do cool tricks with the driver.

tsar

Sun Sept 04 2011, 11:23PM

Registered Member #4084 Joined: Sun Sept 04 2011, 10:51PM
Location: Germany, near Gelsenkirchen (FC Schalke04)
Posts: 4

Hello axelro and all other members of the 4hv community.

I would like to introduce myself quickly to you: My name is Dennis, from Germany. Currently, Iam studying "Powersemiconductor" at the University of Bochum. i am also a member of the german CTC-labs community. I have build several teslacoils in my life, most of it were soild state coils. Since 2 years i am working on my masterpiece, a "intelligent" version of a DRSSTC.
With "intelligent" i mean: Ich control of the teslacoil is done by a Xilinx Spartan 3 FPGA which is connected to every part of the system (the IGBT driver, the current and feedback transformers, a big SMPS with active PFC for powering the coil, and so on). The point why i am posting here is that i will do exactly what you mentioned. I am doing it with the FPGA by doing some heavy math "shit". I dont just measure the time for one or two cycles of the sinewave, i do a curve sketching to find the zero-crossing BEVOR it comes. I have tested it alread, it works fine, but it is to complex for a simple teslacoil. I will go back to the basic "timemesure"way. When i know the exact point of zero-crossing you can compensate all static delays of your driverboards.
I think this type of controling(delay compensation) will increase the power of teslacoils in a mesureable way.
I hope your idea with the DDS chip will work, it is much simpler then my way, so if it works you can build it into smaller coils, too.
Please post if you got any results of your "research".

tsar

axelro

Mon Sept 05 2011, 05:36PM

Registered Member #3640 Joined: Sat Jan 22 2011, 12:16PM
Location: Germany close to Heidelberg
Posts: 39

Hallo, thanks for your replies.

@bwang,
yes, if I ever realised this I needed solid HF capsulation. I once had a AVR uC drive an early SSTC - it was not very stable. I know Finn's Predictor and Steve's UD2.0, but actually so far only tried a bread board version of their logic, which - not astonishing - wasn't very stable. It worked a bit though. The charm of the DDS solution is, that noone seemed to have tried it yet.

@tsar,
no need to introduce - we actually exchanged already, on ctc's forum and probably at mikrocontroller.net as well. Your FGPA approach is certainly high end. Please report when you have first results.

All, while I am still further interested in replies/discussion of the DDS approach, I have been thinking last night, if a pure, simple uC approach (shielded) wouldn't probably suffice as well.

Quick estimate of worst case error/switching current:

uC clock 20Mhz, one command per clock ==> 50ns resolution
200Khz resonance frequency assumed ==> 5us period. At 90Â° phase angle, 1.25us have passed.
At max 50ns miss of the zero switching point I will have: 50ns/2/1250ns times 90Â°= 1.3Â° phase angle to the zero switching point
At 1000A max primary current, I will have 1000A times sin(1.3Â°)= 22,6A to be switched off or on.
i took 50ns/2, because I assume it's a window over the zero switching time point

Not too bad, considering the whole logic simplifies to catching an interupt (falling & rising edge), count an integer up and compare to some values to trigger the gate drivers. The upper and lower IGBT's could nicely been provided with their individual switch signal. Initially, the uC could switch the coil for a single pulse, just to measure the start frequency (and calculate the trigger values). From there, with each feedback rising/falling edge, numbers could be adjusted. With that, of course easily a phase lead could be introduced as wished for.

What am I missing - seems simple enough?

tsar

Tue Sept 06 2011, 03:16AM

Registered Member #4084 Joined: Sun Sept 04 2011, 10:51PM
Location: Germany, near Gelsenkirchen (FC Schalke04)
Posts: 4

The shielding of the ÂµC is very important. All supplys for the controller have to be filtered, because the wires will act like an antenna. If you can, transport the signals to the ÂµC via fiber.
For better performance you could use a CPLD for the counting part of your AVR. You can run the CPLD at a highe frequency (arround 50MHz) to increase the resolution.

I will come up with first results maybe in january 2012. My problem is to find capacitors for the main tankcircuit.

axelro

Tue Sept 06 2011, 05:23PM

Registered Member #3640 Joined: Sat Jan 22 2011, 12:16PM
Location: Germany close to Heidelberg
Posts: 39

Tsar, OK, I'll try at the next rainy weekend and report. For now it's just a feasability study, but it should actually work.

As for your caps, try ebay. There's enough stuff to get started, the perfect caps can always be bought later. If you zero switch perfectly, their quality is a little less important.

tsar

Wed Sept 07 2011, 01:08AM

Registered Member #4084 Joined: Sun Sept 04 2011, 10:51PM
Location: Germany, near Gelsenkirchen (FC Schalke04)
Posts: 4

I have an old MMC lying around from an SGTC with 100nF. I think i use that one until i blow it up.

I have discovered a new problem with my current transformers: Just for fun i have used some rogowski coils for measuring and my selfmade integrator used for the coils has conversionratio that is depending on the frequency of the current

that is my first problem i have got solve, otherwise my currentlimiter will not work....
These coils are tricky to handel, but when the integrater works...... hell yeah!

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob