U-LULU (Updated LULU) for 48 meters, 20 Watt Class-E AM Transmitter

[Antennae]

Charlie, thats a really professional write up.  Thanks for sharing it. 
I've always daydreamed of a transmitter system where there's multiple exciters for the different bands and one big amplifier that is broadband. 

[redhat]

These parts have low breakdown voltages which probably forces you into using a CMCD architecture, as opposed to a Class E architecture, in order to get the RF output power you are claiming, because of the peak drain voltage excursion in Class E. Nothing wrong with CMCD (I'm actually doing some myself) but it does bring a slightly higher complexity, with the need for a non-overlapping clock generator/due diligence to make sure you aren't cross conducting and some sort of output balun, be it a traditional wound one or an LC balun, because at these power levels the typical output impedance is significantly less than 50 Ohms.  On the other hand, if you are intent on using a low BV transistor in Class E, a cascode architecture should work.

Actually, the voltage difference is not all that great.  Class E is around 4x Vcc, CMCD is around 3.14x in my experience.  There are plenty of good devices in the 600V class now that will work nicely in either architecture.  Also, dead time control is nice but not necessary for push-pull operation.  I've had PA's running at 80 meters or so with no dead time control circuitry.  I even built a 2 mosfet 500W MW PA that is running in commercial service with just a 7486 setup as a phase splitter...no drive loss protection though 

For me one of the appeals of CMCD was the elimination of most of the cumbersome ferrite in the output section.  I even started using hardline (0.141") coax as a combination tank resonator and output transformer.  No ferrite required!  Typical PA impedance with respect to Vmod is on the order of 10 ohms or so, assuming the PA is looking into 50 ohms.

+-RH

[Stretchyman]

Thanks RH. I need to adjust my layout to accommodate a higher Q shunt tank inductor as the current one is the only thing getting warm!

Ok I looked at the peak drain V and noticed it wasn't as high as class E but still, as you say I can see the need for the higher voltage parts so will stick with them.

Single SOIC driver easily drives 3 and would drive 5 devices.

I've taken the little beast to my local university where they'll give it their critique. Will be fun for them I think and make a change from LMBA & Doherty at 40+GHz! There's some learned folk there, one quite well know in amplifier design circles. Hopefully they can help me improve it further.

Oh, just to add, I've never seen the need for any duty cycle adjustment and rather crudely just use a dual inverter at the input, inverting once to drive one driver, then inverting again to drive the other. Using a sine wave at the input adjusting it's level from 3V to 5V give a 40-47% adjustment.

The CMCD design I was given used duty cycle adjustment and I don't see any measurable difference between that more complex design and mine.

Str.

[Charlie_Dont_Surf]

Ok I looked at the peak drain V and noticed it wasn't as high as class E

That's what I see in simulation and in reality too. CMCD peak drain voltage excursion is less than Class E. In my simplistic back-of-the napkin look, CMCD excursion theoretically shouldn't be a lot more than Vdd but in reality I'm seeing something in the range of 2 to 2.5xVdd, depending upon the tank circuit Q, loading and all that jazz.

I've taken the little beast to my local university where they'll give it their critique. Will be fun for them I think and make a change from LMBA & Doherty at 40+GHz! There's some learned folk there, one quite well know in amplifier design circles. Hopefully they can help me improve it further.

I going to make an educated guess that you're referring to Dr. Steve Cripps at Cardiff U. I took a couple short courses from him a very long time ago. He's a world-renown expert in PA design.

https://www.cardiff.ac.uk/people/view/364356-cripps-steve

Oh, just to add, I've never seen the need for any duty cycle adjustment and rather crudely just use a dual inverter at the input, inverting once to drive one driver, then inverting again to drive the other. Using a sine wave at the input adjusting it's level from 3V to 5V give a 40-47% adjustment.

Because CMOS/TTL propagation delay for an inverter is ~3-40 nanoseconds, depending upon the family chosen. A 7 MHz signal has a period of ~140 nanoseconds and a half cycle is ~70 ns. With what you have described, you have probably at least 10 nsec propagation delay on one phase (including the FET driver) and so an additional ~5-15 nsec on the other phase, aside from the 180 degree delay. Ten nsec out of 70 nsec is a significant portion of the waveform and apparently enough for you and the settling time of your CMCD circuit. So in this case, leaving it to chance (which totally doesn't surprise me, coming from you) works, but then you probably haven't looked at the rise and fall times on the gates and drains thoroughly over all conditions, multiple DUTs anyway. I like to have options with the ability to adjust for safety margins and maximize the possible conduction angle, hence a non-overlapping clock generator with some adjustability.

So while your crude circuit isn't adjustable, it's still a dual-phase clock generator at least for some range of frequencies but it definitely doesn't lock out/prevent overlapping.

A non-overlapping clock generator isn' t big deal. At it's simplest level, it's a pair of NOR or NAND gates and an inverter in front.




Taken from: http://individual.utoronto.ca/schreier/lectures/3-6.pdf

Edit: Ha. Just realized I know the professor at U Toronto that made these slides. So that's two people I (vaguely) know referred to in one post.

[Charlie_Dont_Surf]

Charlie, thats a really professional write up.  Thanks for sharing it. 

Thank you. I'm glad that you got something out of it.

[Charlie_Dont_Surf]

Also, dead time control is nice but not necessary for push-pull operation.  I've had PA's running at 80 meters or so with no dead time control circuitry.  I even built a 2 mosfet 500W MW PA that is running in commercial service with just a 7486 setup as a phase splitter...no drive loss protection though 

Let's put it this way: maybe you are not explicitly controlling it and providing adjustability to it, but one way or another you are probably not cross-conducting much or at all.  How that happens - or rather doesn't happen - only you can say. Otherwise, if something - anything - was not there making it not cross-conduct, either by dumb luck or intentionally prevention, you might have somewhat different results.

[redhat]

:shrug: It doesn't seem to matter much until you get over 4 MHz, then things get hairy, and efficiency tends to nose-dive.  My main transmitter has overlap control and drive loss protection, FWIW.

+-RH

[Charlie_Dont_Surf]

:shrug: It doesn't seem to matter much until you get over 4 MHz, then things get hairy, and efficiency tends to nose-dive. 

+-RH

 "Quite curious, Captain."

The behavior above 4 MHz is what I would expect almost everywhere there was actual significant overlap.

[Stretchyman]

I'm fine upto 15 MHz so far but then devices with a Qg of 2.2nC aren't exactly difficult to drive. Also I'm comparatively QRP in comparison!

Str.

[redhat]

My 'daily driver' PA has a drive delay and protection circuit onboard, and seems fine up to 10MHz or so.  I haven't investigated any higher as I feel there is little merit in doing so, and have no interest in anything above 30 meters.  The simple drive circuits consisting of just an xor configured as a phase splitter start having efficiency drop off after 4 Mhz.  This makes sense as the drive pulse distortion begins to be a larger fraction of on time with increased frequency   Granted this was a PA I built several years ago now and there are other devices and drivers to try.

+-RH

[Charlie_Dont_Surf]

For me one of the appeals of CMCD was the elimination of most of the cumbersome ferrite in the output section.  I even started using hardline (0.141") coax as a combination tank resonator and output transformer.  No ferrite required! 

I have been unable to find a reference to allow me to design these sorts of things. Even output transformers like TL1 & TL2 & TL3 below, which do use ferrites. Obviously I can copy something somebody already did but I'm looking for some sort of reference on how to design them. Short of having that, I can make some guesses and assumptions as starting points (L and C per foot of the coax, permittivity of the ferrite, etc.) but I'm thinking that there is more to it than that.

I dunno, maybe people just start winding and see where they end up and adjust as necessary?

[Stretchyman]

I used that design as a starting point but ended up just using a single balun on the o/p.

As I'm learning ADS I'll simulate the thing at some point but as it's already working at 93% efficiency I don't see too much to be gained.

I'm interested in any other design that leads to simplification and an increase in efficiency so please post up.

I'm happy to design PCBs and fit the fiddly bits if anyone is interested.

Str.

[redhat]

I used W1VD's design as the basis of my work.  All I did was remove the 1:4 output transformer and directly couple the tank to the balun.  This, with a change to modern SiC devices, makes it possible to get kilowatt level carrier power from just two devices.  In my current design, the balun has been replaced with a hardline transformer wound into the tank coil, the outer being the primary as well as the tank inductor, the inner being the output winding.  The RFC's don't seem to be real critical, just aim for 10x reactance of the nominal line output impedance.  Doing these modifications raises the PA modulator impedance to around 10 ohms or so, easing current carrying requirements at high power.

+-RH

[europirate]

My first build attempt at the U-LULU, with a board made by JLCPCB using CDS's gerber files.



Learning how to solder all over again with a hot-air rework pencil, great for the chip but how do I stop the tiny caps flying away 

On a test-jig I'm getting about 15w carrier, still more work to getting the output coil right and finding a suitable modulator.  Any ideas on something still available from China in 2023?

[redhat]

The TPA3116 amplifier boards and their clones seem to work well.  Tie the positive outputs together, and add a solder blob to tie the wipers of the pot together.  This will ensure that outputs are as close to the same as possible, so one channel doesn't fight the other.  Seems to be good to 15W or so.

https://www.ebay.com/itm/284160314693?mkcid=16&mkevt=1&mkrid=711-127632-2357-0&ssspo=hksnm2-7qz6&sssrc=2047675&ssuid=XS3HxYCuQ6C&widget_ver=artemis&media=COPY

+-RH