HFU HF Underground
Technical Topics => The RF Workbench => Topic started by: redhat on January 13, 2023, 1546 UTC
-
From an edited version of last night's live stream, watch for two painful hours me build my version of the LULU. This is just the initial build without optimization. Schemo's and BOM to follow once I get everything dialed in.
https://youtu.be/Mme7L0eR954 (https://youtu.be/Mme7L0eR954)
+-RH
-
Don't bash it with a hammer like some folks on the old "Homebrew Pirate Group" were tempted to do when they couldn't get the right waveform. Lulu is a shy thing that likes to be treated just so.
-
Comments added!
Nice job btw, keep going! Took me a few attempts and that was at 2MHz before I progressed to 7.
Str.
-
I didn't watch the whole thing but I did go through probably the high points. I'm fairly sure that you know this already but for the other folks watching, the reason why you have ended up lower in frequency is the unavoidable presence of parasitic capacitance on your "hand carved PC board."
The online calculators to determine component values are all fine and good*, but you have to understand that the models can't incorporate PCB parasitics because the the parasitic values depend upon the PCB layout, the PCB materials, the choice of transistor, the packaging of the transistor, etc. They work OK as starting points but there will be refinement later. This is why I recommend the "driving a car in deep snow" method: https://www.hfunderground.com/board/index.php/topic,92791.msg297947.html#msg297947 (https://www.hfunderground.com/board/index.php/topic,92791.msg297947.html#msg297947) Make your first try, figure out how far off target you are and then make course corrections to get closer to the goal.
And, yes, as I heard you say in the video, Class E can be very picky about component values on the matching network. Whatever values you come up with may be suboptimal for someone else whose "hand-carved PCB" technique is a little different from yours. This is one place where a manufactured PCB may be helpful since it is likely to be more uniform from batch to batch, though that is not guaranteed either. (FR4 Er can be all over the map.)
Unless I missed it, you didn't show what the drain waveform at the frequency where the power output peaks (~5500-5600 KHz?). If the drain waveform looks kinda/sorta OK there, then I think your problem is simply a tuning issue and not anything else.
One more thing: if your PCB is dual-sided copper PCB material, you can try to cut away some of the bottom copper directly under the active components and the tuning/matching network. This may dramatically reduce the parasitic capacitance to ground. (On a quasi-related matter, on microstrip and grounded-coplanar waveguide, removing ground will raise the characteristic impedance of a transmission line for similar reasons.)
* Here's another one, also from Australia: https://people.physics.anu.edu.au/~dxt103/calculators/class-e.php (https://people.physics.anu.edu.au/~dxt103/calculators/class-e.php)
-
The online calculators to determine component values are all fine and good*, but you have to understand that the models can't incorporate PCB parasitics because the the parasitic values depend upon the PCB layout, the PCB materials, the choice of transistor, the packaging of the transistor, etc. They work OK as starting points but there will be refinement later. This is why I recommend the "driving a car in deep snow" method: https://www.hfunderground.com/board/index.php/topic,92791.msg297947.html#msg297947 (https://www.hfunderground.com/board/index.php/topic,92791.msg297947.html#msg297947) Make your first try, figure out how far off target you are and then make course corrections to get closer to the goal.
And, yes, as I heard you say in the video, Class E can be very picky about component values on the matching network. Whatever values you come up with may be suboptimal for someone else whose "hand-carved PCB" technique is a little different from yours. This is one place where a manufactured PCB may be helpful since it is likely to be more uniform from batch to batch, though that is not guaranteed either. (FR4 Er can be all over the map.)
Unless I missed it, you didn't show what the drain waveform at the frequency where the power output peaks (~5500-5600 KHz?). If the drain waveform looks kinda/sorta OK there, then I think your problem is simply a tuning issue
Yup, no argument there, and its one of the things I neglected to add into the dialog. As I mentioned in the video, its easy to get tunnel vision and forget simple but important things. This was more of a 'hey, lets try this and see if it works' kinda thing. Appreciate the comments!
+-RH
-
I saw it. Thanks for creating the video. 8)
Z
-
Yep, tuned with a scope. There's a pdf of class E on the net somewhere and shows a whole page of 'half sine' traces indicating what state the tuning is in. Max power and max eff never meet alas.
Btw parasitic capacitance from the PCB will be minimal but the devices C will have a strong influence.
You can use LTSpice (free) to simulate and get good accurate results.
Seeing as most electronics is designed this way (sim-build) then it would be good to show if you have time.
Good work BTW.
Str.
-
Yep, tuned with a scope. There's a pdf of class E on the net somewhere and shows a whole page of 'half sine' traces indicating what state the tuning is in.
From the person who is considered perhaps the inventor, or at least someone who spent a lot of time advocating for Class E, Nathan Sokol. Refer to page 16. I took a short course from him years ago and still have the book.
Sokol, N., Class-E RF Power Amplifiers, QEX January/February 2001. https://people.physics.anu.edu.au/~dxt103/160m/class_E_amplifier_design.pdf
Max power and max eff never meet alas.
That's often the case with RF power amplifiers though. OP1dB is , for example, peaked at a different tuning than best OIP3, than the tuning for the the best gain or S11/S22, etc.
Btw parasitic capacitance from the PCB will be minimal but the devices C will have a strong influence.
Here's my problem with that statement: if that were true, then the tuning of my builds would have been close to spot on after simulation with the transistor models straight from the transistor vendor and they never are. Then I change PCB layout with the same transistor and passives and then the tuning is a bit different yet again. So, no, that doesn't agree with my experience. In theory, proper Class-E design is supposed to "absorb" the transistor Cds and resonate against it. Either the models aren't accurate at 7 MHz (which would not surprise me at all) and/or something else is going on, like, oh, for example, unmodeled parasitic C on the passives and the PCB.
Speaking of transistor models - they should be "accurate" in a sense and they extracted parasitics and put them into their models but in many cases they are apparently not planning on you running their transistors at 4, 6, 10, etc. MHz, even though the published rise and fall times would seem to allow it. These are meant for switching power supplies and inverters and, though this is starting to change, a lot of the applications for these inexpensive transistors we (I) use are not going go any higher than 1 MHz, if that. As a result, the models from some manufacturers create "convergence errors" in a normal Class E and sometimes Current-Mode Class D circuit above maybe 1 MHz. A convergence error basically means that the computer can't converge on a solution to the math problem you are asking it to solve and it generates an error. (Transient simulations are essentially solutions to differential equations.) No simulation is possible in that case.
I figured this out about a year ago. A lot of very simple circuits (not even Class E circuits) generate convergence errors at my frequencies of interest. But then I can reduce the operating frequency (or slow the slew rate, or whatever) and then they start to work. Some Nexperia and TI transistors have been a pain in the backside this way. Infineon transistor models have been pretty good and I have yet to have serious issues with them. (They are my new favorite transistor company.) I have had good luck so far with Cree and Transphorm too, though Cree makes you create a place for their thermal model in order to configure their transistors.
-
Sokol needs to get in line. I've got a Class-E transmitter that was built by a non-engineer in 1992. It's called a "Grenade". I was on the air with it before he published that paper. He needs to meet Boomer T. Dog and apologize.
-
The differences are subtle but a grenade style PA is not class E. It was never optimized as such.
+-RH
-
I took one look at your bench and felt right at home lol.
-
I took one look at your bench and felt right at home lol.
Yes Me too. :)
-
My VP at work asked me to tidy up my part of the lab and I told him, "I either have a tidy lab or I get work done, not both."
-
"clean desk, empty mind."
+-RH
-
Was anyone else thinking rh was gonna drill his hand?
-
Great video...Thanks for sharing it, Redhat!
Molvania Calling,
MP