Now I decided on the first of many VFO rebuilds: a mixer VFO with insulated tank components. The idea was to separate the oscillator tank components from the heat source. So I built a PCB box lined with about 20mm of polystyrene insulation. There was no insulation lining on the front panel, just a thin PCB, because the tuning capacitor shafts are not long enough. The oscillator LC tank components are built on a small piece of PCB, which just sits in the middle of the polystyrene insulator, it is not fixed to the box itself. The three tubes of the oscillator, from the original chassis, were cut out onto an open-air chassis and soldered to the back of the insulated box, with thin wires leading to the VFO tank components. I also put the crystal in the polystyrene box too.

The result shows now a negative drift within 2.5kHz. Nice improvement! But still not satisfied. Nowhere near.

The chart below shows the same blue line of the oscillator drift but with the vertical scale expanded. There is a fast initial drift down, and then a slower upwards drift at a rate of approximately 2kHz/hour... still too much.

I concluded from this that the sharp initial downward drift was due to the heating of the valves section, the internal capacitances in the valves would all be changing as the heat built up; resistors, power supply components etc too. The subsequent slower drift would be caused by the high heat from the 3-tube oscillator chassis, which also heats up the insulated box. Eventually the heat would find its way through the polystyrene insulation and start to bring up the temperature of the VFO tank components gradually, resulting in the drift seen.

Now I decided I have to do some work on balancing the temperature coefficients of the components in the oscillator tank circuit. I read plenty about capacitors on the internet. Polystyrene capacitors have a slight negative temperature coefficient. This means that as temperature increases, their capacitance decreases. A decrease in capacitance in the VFO tank results in an increase in frequency. Therefore, ignoring the initial fast downward drift of the tubes heating up, what we see is an overall negative temperature coefficient. The remedy should be to replace things having a negative temperature coefficient, with other components having a positive temperature coefficient. Silver mica capacitors have a slight positive temperature coefficient!

The light blue line below, is the same as the on the chart above (original insulated version). At this stage ALL of the oscillator capacitors are polystyrene types. Next I tried replacing one of the oscillator capacitors with silver mica, this is the green line (not really much drift in the long term drift gradient). Next, replacing 2 capacitors (red line), but still lots of drift. By the time I replaced ALL the capacitors with the slightly positive temperature coefficient silver mica's, I got the purple line, which is showing a shallower gradient but still an overall upwards tendency, indicating still an overall negative temperature coefficient.

Next I read about the temperature coefficients of the popular micrometals powdered iron toroids. The type -6 material has a temperature coefficient of +35ppm, the type -2 material is +95ppm. Positive temperature coefficients, just as I need! See below - recall that the purple line is the one from the chart above (after having replaced the polystyrene capacitors with silver mica). The light blue line at the top replaces the air-cored inductor (plastic former) with 31 turns on a T50-6 core. That actually makes the drift WORSE, a faster upwards drift, than the air-cored inductor (purple line). From this we may conclude that the drift characteristic of the air-cored inductor is more than +35ppm - such that replacing it with a T50-6 makes the overall temperature coefficient MORE negative.

When I tried 31 turns on a T37-2, with +95ppm temperature coefficient, I get the perfect pink line! All temperature coefficients nicely cancelled out, this pink line shows the oscillator stable inside a narrow 50Hz band for more than an hour! One may conclude that the air-cored inductor has a temperature coefficient somewhere between +35 and +95ppm (T50-6 and T37-2 respectively). The tuning capacitors and trimmer capacitors presumably have negative temperature coefficients, and the combination of all the positive temperature coefficient silver mica capacitors, and the +95ppm inductor core, result in perfect cancellation. Very nice. However, there's still that fast downwards drift of 3kHz in the first half an hour, before the nice stable state is reached. It would be nice to tackle that too.