Allison KB1GMX has provided an update on her experimentation with flattening out the power curve on the µBITx.
Gain and power line up with a few possible solutions listed:
Mixer – output at full tilt about -10 dbm. Go higher and the spur gets bigger so lower output here is better.
Q90 – no mods at 80m about 16db at 10M about 10-11db and not more than 3mW power. Replace it with a really hot transistor. BFR106 has a FT of 5ghz and can do over 100mW. This should keep the stage gain at about 16-18db from 80 through 10M and deliver the 3mw! It is the easiest stage to make flat and high gain as this get more difficult at higher power out.
Pre-driver – takes that 1mw and boost it to about 40mw, We hope. What she saw was about 45mW at 80m and barely 8mW at 10m . Note the gain at 80m is about 16db and at 10m maybe 9db.
We need the gain again to be more stable and higher as well. Transistors tried that worked better included the TO18 2n2222A, 2n3866, and 2n5109. The 5109 was best even though she used only one. Second best was tie between a single 3866 and two 2n2222. MPSH10 was disappointing.
Allison plans to try an oddball 2n6661A, [at 14$ each most will choke] a VMOS fet has potential and requires many circuit changes[Bipolar bias to MOSFET]. They worked really well in another project at VHF.
Driver – takes that 40mw and make about .4W at 80 and sinks to barely .08W at 10m. Gain for that stage is supposed to be about 16db and likely is at 80m but at 10m measured for different 3904s from around 8 to 11db.
This is where the 2n3904 really fails. Its bandwidth at high currents actually does down. The 2n2222A has the reverse, the gain and bandwidth increased with increasing current. While not an ideal part it does work better. The ideal parts here would be two devices in push pull [not 4] and with higher FT. Allions hasn’t yet tried 5109 parts here but its a solid bet. She wishes 2n3553s, 2SC2166, 2SC799, 2sc1306 and others were still around.
She plans to try an oddball 2n6661A here too as its good for up to 6W.
Finals IRF510 at 80m approaches 15-20W with a gain of over 17db, at 10m its about 13db, but the drive is so low that yields maybe 2W.
If there is enough power at the gate it does the job we ask. RD16HHF does not offer more gain or bandwidth as MOSFETS do not have the equivalent of FT. These parts do offer a handy tab that can be grounded.
To go from -10 dbm[.0001W] to +40 dbm[10W] we need 50 db of gain over all. That includes losses to transformers and other circuit features. So we need more gain than that available. Maybe 6 to 10DB more gain would allow some latitude on the gain adjustment pot.
So with four stages and the last being limited to the IRF510s. We set a few rules.
- We never run a transistor wide open as one may exceed and another fall short.
For 2n3904s that happens and the lowest common denominator for them is around 10db. Use better transistors and lots of feedback. More likely to work for everyone than a bet on Monte Carlo.
- We only have four stages! The board is laid out that way. Reality sucks.
- IRF510 or RD16HHF you get about 13-16db of gain, period. More
is wishful thinking or running wide open and risking stability. Hint IRF510s blown cost about US$2, RD16HHFs blown cost US$10. Going to IRF520 and 530 are not better [for 10W output] as the internal capacitances are significantly higher and these parts don’t make the job easier.
- Based on total gain needed and what the finals can be reasonably expected to do: The three prior stages must deliver 45 to 47 db of gain. Its also a lot so attention to stability is everything.
- The result must be stable. Oscillation will kill the finals.
As it turns out, item four is the killer as its basically asking for 16db from all three stages. You need good parts for that. Also the interstage coupling must be up to the job while not introducing uncontrolled losses. This is a tall order. So far I have only partial answers.
Answers so far:
Dump the 3904s and use 2n22222(TO18), for those interested in 80-17M it works remarkably well and give a boost up high too. If you go to 2n5109s you may need one as pre-driver and 2 in the driver stage. They are big and the space is small. Use the SHORTEST leads possible.
Replace or parallel R941,R911, R96,R942 to get 11 ohms each (I paralleled 22ohms across them). Lower emitter resistance helps the gain and power out to finals.
Replace Q90 with BFR106. Note R81 has to be increased to between
2K and 2.7K for this part. (for those making suggestions I tried 2n2369 in
SMT, it was better but not great). Mouser has the BFR106 for a whopping 38 cents each.
Change C81 to 470 pf, This flattens the 80 and 40M runaway power and helps the higher bands.
There are many changes to transformers possible but for the moment the above are best bang for the buck. Also changing the transformers might be a handful for some.
Warning every time I tried the 2:4 turn transformer with any ferrite the finals were heating a lot and the stage efficiency was well under 40% [terrible with IRF or RD16]. The 2:3 was better at high power but below 50% efficent. This is still in the grinder…
Ashhar Farhan VU2ESE observes that ALC is just one way of controlling output power from a linear amplifier stage and that an easier approach is to do it in software.
JerryW0PWE notes that in the posts on boosting and/or leveling output power he saw pictures of a binocular core being used as the output transformer in place of the toroid that comes with the uBitx board. He asks “What size is that binocular core?”