Jiri Vackar [correctly pronounced as "Vatzkaarz"] invented his VFO oscillator during late 1940s. It is probably the most stable VFO oscillator known.
Thanks George!
The Vackar oscillator configuration is rarely used because of known reason.
(NIH, the not-invented-here syndrome).

Vackar oscillator
The frequency tuning range is above 2.5, not observable in any other type of oscillator.
The coupling ratio is fixed; typical range is 1:4 up to 1:9. The frequency tuning is provided independently of coupling. Transistor's parametric variables are isolated from the resonator. The transistor input is not overloaded as Clapp or other circuits. The collector output is at low impedance providing low gain just to maintain the oscillation. The feedback division ratio is fixed. Even if the VFO is tuned, the impedance divider is fixed. The stability is close to XO - crystal oscillator. Jiri Vackar published his work in a book, providing theory and analysis of each type of his oscillator. What was the last model, V66? Who knows?

A while ago, I thought about fine tuning oscillator for PLL based VHF synthesizer. Few designs failed high expectations. I tried Colpitts, Clapp, Hartley, Pierce, and Seiler. Junk. The Butler is better than single active component oscillator, but it is not good enough. Commonly used oscillator configuration does not guarantee good performance. The signal clipping by diodes guarantees additional phase noise and thermal frequency drift. The best oscillators use two or three active devices. This is valid for the VCO, VCXO, TCXO, and OCXO. The second active device acts as an impedance converter, isolation amplifier, AGC circuit, series RF power dissipation device, and a phase shifter. Be aware the articles in QST very often copy old mistakes, and limiter diodes. U.L.Rohde from U of Washington wrote few articles about the poison of limiter diodes in oscillators. Kenwood.
Ordinary oscillator has poor tuning range, the output voltage swing is unstable, and the frequency stability is poor as well. The industry tries hard to make its sale pitch, to replace single oscillator with 50 ICs, digital dividers, approximation registers, thermostats, and other junk. Now what?

Many times I heard this and that is "not the right cup of coffee". Many times because of bias. Search around and make some math derivations and experiments. Exactly same story I found as a student, hearing "the Cuk's power inverter is somehow unstable". Politics was behind. Without that, there won't be any space program and sattelites.

I checked the Vackar. First measurements with frequency counter were quite positive. The Vackar VFO was running in freezer at -30 C. That is about -22 F. Not bad for Yukon Territory climate. Stable. Can it run more stable? The ferrite and iron-powder tuning slugs went out, down the pipes.
There is a strong public belief the iron powder tuning slug cores are good. It is not. In thermal stability, they are better than ferrites. They can take higher maximum magnetic flux than heavy ferrites. Micrometals cores use iron powder technology from the 50's, nobody around use it any more. The iron powder cores are lossy in terms of resulting Q. The grains are not properly bound together. You will have "brown fingers" from handling these cores. Generally, ferrite is NiZn or MnZn alloy.

Ferrites
Quite good are Ferroxcube-Philips ferrite cores, presently manufactured in Spain. The Spanish sampling service for US works fine. The Ferroxcube's sales reps in Poland asked $230 for delivery of $1.0 stuff. Sort of naive robbers. Wrong people and wrong management.
Iskra-Feriti from Slovenia manufactures good ferrite toroidal cores and two hole cores, excellent for transformers (1C material ui = 900). I recommend this company.
MMG Canada LTD. (TT electronics) have excellent heavy ferrite cores for HF SWR meters.
The experience with German Epcos is weird. Epcos is not able to deliver anything in time, delivering "we forgot" and the promises take 3-6 months to learn nothing will ever come. RIP, rest in peace Epcos. Same story is with another German company Neosid. Europe is full of morons without healthy patriotic approach.
Good and verified supply of balun cores and ferrite toroids is from www.jpmsupply.com Houston, Texas. They ship by surface mail. You can order and pay with visa card on the phone. Flexible and friendly. Once again, you get the feeling there are still normal people around, and you are the customer. Texas. God bless them. Another source of ferrites k5nwa, reliable source.

There are few important components
good caps of known properties, inductors, voltage regulator, and the transistor. It will run with dual gate MOSFET, JFET (low 1/f noise), as well. Low noise MAC01 voltage regulator (10V precious low noise reference) will do the job. Even LM78L06 with 50uV/Hz of noise is fine. The LM317 is good for car lead battery charger with 350uV of noise.
The best generic voltage regulator of all times is maybe the LM723 (MC1723CP) with 7-20uV of noise. Consider the input voltage range 40V. 24V are used for transmitter, the features, flexibility, output noise, thermal shutdown, thermal management for PA current runout, current shutdown, reliability, stability. Motorola has the same opinion about LM723. Ignore the +5.5V Absolute Max Input Voltage regulator technology.
For the oscillator, expect 80dB spur free spectrum.
The mechanical design has to be stable. Tuning cap needs reduction drive. The coupling with buffer is loose, and at low impedance. The fine-tuning varactor is coil tapped, or cap-divider tapped. The fine tuning with varactor (+- 1.0 kHz) will slightly change the frequency-temperature coefficient. Direct tuning with varactor will ruin temperature characteristics and phase noise of any oscillator.
Watch how many designs with varactor, limiter diode, and a 500 kHz tuning range ruined all advantages of the oscillator (Kenwood TS-850). Varactor behaves as a non-linear resistor and variable capacitor. All parameters change with temperature, frequency, DC voltage, and RF voltage. Direct varactor is for designs with Kvco = 150MHz/V @2.4GHz where the Q is low. Direct varactor tuning for low noise oscillator is not a good idea.
I used ceramic coil form with diamagnetic tuning slug for better temperature stability (brass, aluminum). The brass slug works as a single short turn, and tunes the frequency up. Teflon coil form is good, or Plexiglas will work as well. Keep the high Q, and shield the whole box. Stability of 2Hz at 7MHz was measured. Under 1ppm? Here I stopped.
"The VFO can create stable beat with crystal oscillator, and it will stay like that for hours".
The concern was why to use another PLL loop? The oscillator phase noise is lower than any frequency synthesizer use to have. The frequency reference multiplication adds noise, the dividers add noise, the phase detector adds noise, the buffers add noise, the active filter and regulators add noise. Outside the DC PLL regulation loop you won't get similar VCO phase noise as free running, but worse. National Semiconductor used this idea for Bluetooth with open loop. The micro controller used for synthesizer burns power, radiates spurs, and generates broadband spur spectrum.

Maybe DDS?
The DDS is power hungry. Recent DDS chips are from Analog Devices MS, using DAC. The output is "sine wave". It is improving.
The second source is Silicon Labs TX. The output is rectangular or sawtooth. Depends on the frequency. 600 to 900MHz oscillator use programmable UHF divider for lock, and the output is divided down. Check for jitter and discrete spurs. Jitter can be translated to phase noise. There are no similar chips from Europe.
The DDS phase noise and residual noise floor is not for EME receiver, or weak signal reception. The VFO is still better solution than the AD9850, AD9891 DDS - direct digital synthesizer chip.

The DDS chip has fine-tuning of fractions of Hz. With crystal tolerance of 25 ppm. Major performance limitations are digital clock radiation from the chip state machine, clock multiplication from the MCU controller (more spurs), DAC aliasing, discrete spurs, spurs from display controller, spurs from display backlight, discrete dynamic spurs while tuning, rich grass type of parasitic spectrum, finite level of residual flat background phase noise, and the output frequency is never a round number (5,000.000 kHz). The phase noise parameter is the worst performance issue. It takes 10-14 bits to create a single sine wave. The signal reconstruction filter is not there. If you want phase noise of -80dBc/Hz @10kHz and worse, use the DDS.
The beat frequency signals are misleading and can come from supporting hardware. Beat signal from MCU, from keyboard, from a color LCD/TFT display driver, backlight LED step-up switcher are disturbing. It is not possible to fully clean this spectrum. Sometimes the beat signals are hidden between the specific frequency steps. For HF reception of static noise, few recent DDS chips are promising. It improves every year.

Careful observation of the oscillator performance reveals known issue, the limited dynamic range and S/N ratio of HP (Agilent, Keysight) Spectrum Analyzers compared to Rohde&Schwarz.
But R&S has different technical challenges, including ego issues, feeling you talk with the God, stolen noise feedback oscillator challenge (DL1xx atv person). R&S should be sued for stealing and mischievous competition. They never got permission for commercial use.

The oscillator circuit works well with JFET (square law). Components with cubic transfer characteristic (dual gate, tunnel diode) have good inherent frequency stability and performance.

Links related (UK):



Vackar oscillator, Ge pnp



Vackar UHF oscillator. Something is missing.




Click to zoom

The genuine Vackar oscillator circuit by G3PDM. JFET

With C1/(C4+C6) and C3/C2 = 6. Use a high-quality variable capacitor with ball bearings, two-wheel transmission. Adjust feedback control C2, an air-dielectric trimmer, so the circuit just oscillates. Use a strong box from solid metal. C1, C3 and C6 are silver-mica or ceramic types glued to a solid support to reduce sensitivity to mechanical shock. The buffer amplifier is essential. Circuits using external gate-to-ground diode suffer from high phase noise and instability. The diode loads the circuit. The signal is rectified by the diode, and dynamically shifts the operating point of the JFET. Single-point grounding is fundamental. The inductor used a ceramic form. Use thick solid wires (#16 to #18 gauges) for mechanical stability. The coil is always mechanically fixed by paint. My choice was transparent nail polish with lacquer thinner. Mechanical 20:1 reduction gearing with anti-backlash may do the trick. Take out the Zener and replace it with voltage regulator. Clean all components and the box in ultrasound cleaner.
The bias circuits and feedback caps ratio influence the close-in phase noise. A wonderful sine wave output does not mean it is - a low phase noise oscillator with excellent stability. Calculate a five-element low pass filter with 1dB ripple. Place it on the output. Chebyshev is fine. Use the termination resistor. Think about the frequency plan and frequency dependency, and different types of pulling. The buffer is essential. You can divide down, but there is no need.

What is the TEMPCO compensation?
The ceramic capacitors are manufactured with different Temperature Coefficients of Capacitance (K_c). It means, by choosing the right combination of capacitors you get zero thermal frequency drift. The cap combination you have to find out by thermal measurement. Get the numbers, calculate the caps. Test it. Standard TEMPCO K_c values for capacitors are:
+135 (blue dot, Porcelit)
+33 (white dot, Stabilit L33P)
+0, -33 (N)
-47 (J, dark gray, Stabilit K47N, NPO, COG),
-125 (?)
-470 (U)
-750 (V) purple dot, Rutilit
-1500 (green base, dark gray dot, Negatite)
and multiple specialized values. Other ceramic dielectric types are temperature non-linear and not fit for oscillators. Try polyester caps, they are good and stable, low loss. The TFT plastic multilayer caps from AVX are good.

The resulting cap value is:
C' = Co *(1 + deltaTemp*K_c*1E-6).    [pF, pF, C or Kelvin, K_c temperature coefficient]
Co = capacitance at room temperature (24C)

Variable caps have high negative Kc. Coils have positive Kc. Capacitor manufacturers make caps with different Kc. There are about fifty manufacturers. AVX, MuRata, ATC, CDE, ... Expect lead time of few weeks. The TEMPCO compensation is done for a single frequency. By slight schematic diagram modification, it is possible to reach zero TEMPCO at two frequency points. Keep a set of caps with different Kc.


Good Luck!
va3diw


After more than ten years, this article still evokes questions to think about.
That was the purpose.
Initial discussion formed a concept for K3 Elecraft.



Resources
[1] Iskra-Feriti, shutdown? Ferrites for broadband transformers 1C material, and R-1F070506-02 two hole ferrites
[2] Ceramic Magnetics Inc., NJ
[3] Bourns.com, CA
[4] TSC International.com, IL
[5] Ferroxcube, a Yageo company
[6] Elnamagnetics, NY
[7] Fair-Rite, NY
[8] MuRata.com, GA
[9] Tesla capacitors
[10] CDE Cornell Dubilier mica capacitors
[11] MMG Canada LTD, TT electronics
[12] Neosid.de, weeks to months
[13] Epcos now TDK, 13 to 21+ weeks lead time

The DDS spectrum

Span 24MHz, 0 - 24MHz, 10dB grid

Measured AD9833 DDS:
sine wave, 10Bit DAC with 100pF/50-ohm load @9MHz. Spectrum Analyzer, span 24MHz, /10dB scale. The 48MHz span shows even richer spectrum. Theoretically infinite spectrum response. Next time, it will be better.

This DDS chip is not bad for audio and 137kHz. High order anti-aliasing filter can clean some spurs. But not the noise floor and phase noise. The chip was intended for medical ultrasound, sonar, ocean buoys, modulators, and automated testers.