2007-01-22
The "Wee Willy" RF Power Amp
I first saw the Dick Pattinson VE7GC Wee Willy on the "Popcorn" site. It is quite an interesting little rig, but some things about the design concerned me, so until now I've largely ignored it. In particular, the final power amplifier stage is essentially class-C which seemed unlikely to work well for a DSB signal. The Solder Smoke guys were talking about it in an episode I heard on the way home from work tonight (21 I believe), so I decided to build and test it out. The circuit is so simple, I figured I could whack it together before dinner - which indeed I did, took less than 2 hours to build and customise the device.
I did not have VN10KM devices, but they are /fairly/ similar to 2N7000 devices, so I built it with 2N7000 devices instead. The VN10KM has slightly worse properties in most ways, except its dissipation capability, which is somewhat better, I suspect due to the metal tab most versions have embedded in the TO-92 case.
If you do the math on the output network at a 6 V supply you'd expect gain compression around 500-600 mW out. I *really* doubted the claims of 1 W out at 6 V. At 12 V it is definitely possible to get 1.5 W or more out, but the efficiency at either voltage will be pretty bad due to the modest on-resistance of the devices.
The driver stage is clearly designed for high-Z in and out, at 50 Ohms its power gain is only about 12 dB. My completed amplifier gave 25 dB more or less flat to 18 MHz (without the LPF network). Gain drops to about 23 dB at 1 MHz because of the transformers, and the gain is still usable at 30 MHz (about 20 dB). The 2N7000 has quite low capacitances, it is practical to build a HF-flat amplifier that puts out about 2 Watt with them paired up in push-pull.
Built as specified in the Wee Willy there is a noticeable "threshold" effect from the unbiased output stage. I assume either the noise or carrier leakage from earlier circuits make this a minor problem in the actual Wee Willy. Either that, or the VN10KMs have much lower threshold voltages. To avoid this problem with my amplifier I added another pot to bias the output stage as well.
Best efficiency is achieved with about 10 mA on the driver and 25 mA on the final pair. Efficiency isn't that good however. The best I achieved was 43% at 12 V, delivering about 1.7 W into 47 Ohms. At 6 volts I could get about 680 mW out at a woeful 20% efficiency!
Envelope linearity seems fairly good, but I am yet to do IMD tests on it. The waveform without an LPF looks absolutely terrible! The LPF is *not* optional, without it you will be heard quite well on 40 metres.
Without a load the circuit is fairly well behaved. It will not pull much current from the supply and appears to be able to tolerate the abuse continuously. Stability seems good with the input floating as well, however is it possible to cause self-oscillation by touching the input and output at the same time - not very advisable! Removing your fingers will stop it. When the input is terminated in a low impedance this goes away.
I didn't try shorting the output, but I did try running a torch globe from it. It wasn't a good match to 50 Ohms, so it largely refused to load up the circuit properly. The amplifier pulled normal amounts of current, but produced little output (a dim glow - and poor voltage across the load). The output devices got a bit warmer than usual, but survived the test (about 2 minutes). This is probably how it will behave into low-Z loads. I didn't fiddle much with reactive loads, and lacking a ready-to-go tuner for 80 metres I suspended testing at this point (I might have a fiddle with an L-network later). I did cook some 1/4 Watt 47 Ohm resistors though - the 20 Watt dummy load got warm to the touch, so there is real power there.
The output devices get quite warm, no doubt because of the poor drain efficiency. Being superglued top-down to the PCB does cool them somewhat. It doesn't get that hot that it will melt the wax I used to hold the toroids in place. The bias current drifts up after cooking the amplifier with 1.6 W out for a few minutes. As the devices cool it comes back down again, this seems to be non-damaging, the 2N7000 is a pretty robust device. Note that this is CW power, not what the amplifier would actually have to contend with in a phone rig like the Wee Willy.
I intend to rebuild the amp at some point with the specified VN10KM devices. Jaycar carries them, but at about $4.00 each, I am hesitant to invest the money when I know the 2N7000 (much cheaper) works just fine. It would be interesting, however, to compare the performance with the devices specified.
Update
I built a LPF for the unit and started playing with it in a practical DSB transmitter using a few other bits and pieces I had laying around.
The signal looks *much* better with this filter in place.
The insertion loss is less than 0.5 dB in-band. The efficiency and power output were not significantly affected by its addition. The stability seems unchanged as well. I tried shorting the output too, it reflects the transformer primary reactance to the collector in a similar manner to open-circuiting the output, so the circuit is poorly loaded and pulls much less current. I still haven't experimented with pathologically reactive loads.
Update 2007-04-09
I've since built the entire Wee Willy (which will be the subject of another article). This time I used VN10KM FETs for the final stage. I can now confirm that the 2N7000 performs identically in all ways, the only difference being the easier heat sinking of the VN10KM devices with the drain-connected tabs which you can solder a heat sink to.
I've also had the 2N7000 and the VN10KM operating at 6 metres delivering over half a watt per device. The VN10KM has internal protection zener diodes, which makes it more robust in theory, but I am still yet to kill either device by load mismatches or severe overdrive. The 2N7000 has less capacitance, which may mean it will operate better into the VHF region, but this is yet to be confirmed.
4 comments.