After all that total failure with the sub-mini valve, I decided to rebuild my confidence a bit and try to build the 20MHz crystal oscillator next. Crystal oscillators should be easier to do than VFO's, right? Stick with what works - in this case the triode Pierce oscillator used in my 1-valve CW transmitter. For the triode I used one half of the dual-triode 12AT7. Perfect, it worked - testing with various crystals in the junk box found many were 3'rd overtone - but I found one on 20MHz fundamental, perfect for this.

Next for the other parts of the circuit - for which I relied heavily for inspiration on various circuits in my 3'rd edition RSGB handbook, published 1961. The VFO part of the oscillator is a Colpitts oscillator in the other half of the 12AT7. I used a 6BE6 pentagrid mixer, with a 26MHz tuned circuit in its anode circuit, driving an EF91 buffer, with another 26MHz tuned circuit.

These photos show the VFO construction in a box made from PCB material. Size was 90 x 75 x 70mm approximately. The heat dissipation inside that small box totaled 10.7W! Bear in mind the heater filaments alone consume 6.3V at 300mA each - which itself is nearly 6W. Consequently, that box got HOT, so hot, you couldn't put your hand on it and keep it here. RF output was by RCA (phono) socket and feedthroughs for the power connectors. The top and bottom lids fixed on with nuts soldered into each corner. Note: these photos were taken BEFORE I heard that toroids are not stable with temperature, so I removed the toroid in the LC tank circuit and replaced it with an air-cored 4uH inductor.

Below see the circuit diagram, a nice photo of the power supply (260V HT and 150V regulated from the VR150/30 valve on the left), and the HP1741A oscilloscope and Racal 9911 frequency counter showing the nice 26MHz output. Note that there are two variable capacitors for tuning: one is the main tuning capacitor, the other one is for bandspread (fine tuning) - it adjusts the frequency by approximately +/- 3kHz. The circuit diagram of the tuning capacitor arrangement isn't shown in the diagram below, but it's the same as in the battery valve circuit seen later.

After the construction, came the measurements. In the chart below, the first measurement was the blue line. Over 50kHz drift in 45 minutes, and no sign of slowing down! Wow! That is the WORST oscillator I have ever built. Next, on freinds' suggestion, I added a grid leak resistor, 100K, and a 50pF coupling capacitor. The theory was that without this grid leak resistor, the zero bias at the grid would result in high valve current which would cause additional heating. The result was the squegging seen in the oscilloscope photo to the right here. The squeg frequency was approximately 75kHz. This was easily remedied by reducing the coupling capacitor from 50pF to 25pF. No problem. The resulting drift is the red line... EVEN WORSE! Over 70kHz in an hour! Just when you think you built the worst oscillator in history, it gets worse, even worse.

The third experiment was to replace all the horrible cheap orange moulded mud ceramic capacitors (you know those) with high quality polystyrene capacitors. Ceramic capacitors are lossy and they generally have poor temperature coefficient characteristics. They are available in both positive and negative coefficient versions, and NP0 (close to zero temperature coefficient), but mine were just junkbox types and I believe that these have typically a negative temperature coefficient. The result was the green line, much better, but still 20kHz drift in the first hour. Progress, nonetheless.

Finally the purple line shows the result with a little less heating, by leaving the lid off for ventilation. Drift something like 7 or 8 kHz, about 10% of the previous 70+ kHz, but still we have a long long journey ahead.