Superb mechanical construction of the inductor is a major factor. See Czech Tesla report cited below for specific advice.
If you can track down this 1941 [!!!] original and/or the many derivatives based on it using both "valves" and 'solid state" circuits you will probably find circuits with performance about as good as you can get.
"E. O. Seiler, "A Low-C Electron-Coupled Oscillator," QST, Nov 1941."
A circuit based on Seiler's work was published in QST with , I think, a title like "A synthetic rock" or "The synthetic rock. (Rock == crystal).
One reference noted that the oscillator output was designed to be plugged into a crystal socket and when used at eg 12 MHz to provide the reference fquency for a 2 m,etre band (144-148 Mhz) transmitter, produced a signal which was zero beat stable for hours ata time when compared with a crystal controlled test unit. Zero beat stable means drift is within a few Hz at most as otherwise you hear "rumbling" as the two signals heterodyne.
I don't speak or read Croatian and gargoyle was having a bad day but I believe the circuits below are based on Seiler's ideas. See pages 147-148 here. Translation to English (unless you read Croatian) looks worthwhile. Many other related circuits there too. These may come from US "hams' W8PK & W3EB.
Note that these two circuits do NOT look overly special. I think a significant part of the "trick" is in the relative (lack of) loading of the tuned circuit.
This mentions the Synthetic Rock in passing, with approval.
Valuable albeit old !!!
Resources for Understanding Oscillators {2003} - with references to MUCH older material.
Look for Seiler references.
The following are from the above reference -
E. O. Seiler, "A Low-C Electron-Coupled Oscillator," QST, Nov 1941.
- Describes the circuit that D. Stockton, ARRL Handbook, 1995+, p. 14.14, mistakenly calls "the original Colpitts circuit ... now often referred to as the parallel-tuned Colpitts..."
W. B. Bernard, "Let's Increase V.F.O. Stability," QST, Oct 1957.
- Suggests using grounded-cathode triode with separate buffer instead of the electron-coupled pentode circuit. Also suggests using the Seiler or Colpitts circuit with the largest available value of tuning variable capacitor instead of the Clapp circuit with a low value of tuning capacitor.
J. Vackar, "LC Oscillators and their frequency stability," Tesla Tech. Reports, Czechoslovakia, Dec 1949.
- He discusses mechanical design of tuning circuits, presents general analysis of oscillator circuits and their sensitivity to changes of internal capacities of valves (vacuum tubes). He reviews existing circuits, including Gouriet-Clapp, Seiler, and Lampkin, comparing their amplitude dependence on frequency and hence useful tuning range. He describes in detail the design process for circuit we commonly refer to as the Vackar oscillator, which has greater tuning range. He then goes on to describe a slight variation with still greater tuning range. This last circuit, which I have never seen elsewhere, he describes as a compromise between the first-mentioned Vackar circuit and that due to Seiler.
1949 original of the above report in Czech. Superb][(http://www.scribd.com/doc/62276627/Vackar-wholepaper)
Starts with a description of mechanical requirements for high stability.
J. K. Clapp, "Frequency Stable LC Oscillators," Proc. IRE, Aug 1954.
- Shows that Gouriet-Clapp, Seiler, and Vackar oscillators have equivalent frequency stability given equal resonator Q. They differ only in how much the amplitude of oscillation changes when they are tuned. The three circuits are useful over frequency ranges of 1.2, 1.8, and 2.5 to one, respectively.
The above suggests that the Vackar oscillator is potentially more useful.
See refs here Discrete Oscillator design- Rhea on page 261.
