Is the circuit sensitive to "C" and "R" tolerancies? Can I use a pot in series with one "R" to set the needed frequency, say +-10%? What for "Ref" is needed?

It is a Phase Shift oscilator. The three RC filters set the oscillator frequency.

REF is half the single supply voltage.

 

Is the circuit sensitive to "C" and "R" tolerancies?

Of course.
Their values are what determines the oscillator frequency.

Can I use a pot in series with one "R" to set the needed frequency, say +-10%?

Yes, but that changes the loop gain, so you would also need a pot in series with R2 to adjust the gain to keep it operating properly without clipping.

What for "Ref" is needed?

Ref is to bias the circuit at 1/2 the supply voltage so the circuit can operate with the output not clipped when operating from a single supply voltage.
It's not needed if you have dual supply voltages for the op amp.

 

I think @Audioguru again uses the Lancaster circuit followed by a switched-capacitor filter so that the filter tracks the frequency being generated.

I've done something similar in two different projects, both very successful. One was for a speech and hearing research project that needed bolt-down frequency stability. A similar but different counter-plus-resistors waveform generator followed by an 8-pole switched filter, with both clocks derived from the same crystal oscillator.

For this thread, I vote for the phase shift oscillator. Reasonable parts count, easily tunable over a small range, built-in amplitude stability. Personally, I like the version that uses opamp voltage followers between the R-C stages, with 3 (standard) or 4 (Bubba) stages. 1 LM324 does everything.

https://en.wikipedia.org/wiki/Phase-shift_oscillator

This section from a previous link covers the Bubba Oscillator: https://sound-au.com/articles/sinewave.htm#s561

Much lower loop gain, lower distortion, better amplitude stability, especially with tuning. The tuning range available by varying only one resistor is much smaller than with the 3-stage version, but the circuit's sensitivity to component tolerances also is lower.

ak

 

This is the schematic of the oscillator section of the HP 209A oscillator


This oscillator goes down to 4 hz. You could try to duplicate it or see if you can buy a second hand HP 209A. I bought one several years ago for £10.00
This is a link to the manual for the 209A http://bama.edebris.com/download/hp/209a/HP-209A-Manual-sn-818-Wide Sheet Schematics.pdf
Another posibility is the DDS Function Signal Generator 1Hz-500KHz if you don't mind a synthesised sine wave.

Les.

 

Of course.
Their values are what determines the oscillator frequency.
Yes, but that changes the loop gain, so you would also need a pot in series with R2 to adjust the gain to keep it operating properly without clipping.
Ref is to bias the circuit at 1/2 the supply voltage so the circuit can operate with the output not clipped when operating from a single supply voltage.
It's not needed if you have dual supply voltages for the op amp.

Thank you for the answer.
I' like to ask two more questions.

If real capacitors are used in place of C1, C2, C3 , say, 220nF with tolerance +-10 %, so C1=200nF, C2=210nF, C3= 230nf, and real resistors used in place of R6, R3, R5 , say 82 kOhm +-5 %, and those are not equal too, will the THD remains below 1% ?

Is it posssible to use another technique to stabilize the amplitude instead of diodes, as their parameters change due to temperature changes?

 

Is it posssible to use another technique to stabilize the amplitude instead of diodes, as their parameters change due to temperature changes?

As I have mentioned in post#15 - when using two classical opamp-based integrators in a closed loop (one inverting and the 2nd one non-inverting) you need no diodes (or other additional nonlinear parts) at all for a THD which fulfills your requirements.
In this case, it is best to use two different time constants for the integrators.

The oscillation frequency is wo=1/SQRT(T1*T2) with T1=R1C1 and T2=R2C2

In this case, the integrator with the smallest time constant T1 reaches the amplitude limit (power rail) with negligible clipping only - and the other one (larger time constant T2) with the smaller gain at w=wo delivers a filtered signal with low THD.

 

Is it possible to use another technique to stabilize the amplitude instead of diodes, as their parameters change due to temperature changes?

There is very little current through the diodes, and the self-heating from internal power dissipation is minuscule. When using a diode as a temperature sensor, the change in Vf is only around 2 mV per degree C. Unless this circuit is outdoors, I don't think that is an issue.

ak

 

Below is a quadrature oscillator that allows easier tweaking of the frequency to compensate for component tolerances:
Shown is the frequency change for the Fadj pot wiper settings of 0% 50% and 100%:

 

you need no diodes (or other additional nonlinear parts) at all for a THD which fulfills your requirements.
In this case, it is best to use two different time constants for the integrators.

So the integrator time-constant determines the output amplitude, thus the amplitude will deviate as determined by the time-constant tolerance (which likely is tolerable).

 

As I have mentioned in post#15 - when using two classical opamp-based integrators in a closed loop (one inverting and the 2nd one non-inverting) you need no diodes (or other additional nonlinear parts) at all for a THD which fulfills your requirements.
In this case, it is best to use two different time constants for the integrators.

Good circuit. THD=0.16%.
Voltage of power supply can be changed from ±12 V to ±6 V. It still works fine with ± 4.3 V OUT amplitude.


ADDED:
One more circuit:

 

If you read Don Lancaster's original paper it tells you that the circuit eliminates the first n harmonics, where n is the number of stages. The first harmonics is the clock freequency. Don Lancaster adds a Sallen & Key filter to remove the higher harmonics, which is great for a fixed frequency sinewave but not so good if you want it variable.
I think @Audioguru again uses the Lancaster circuit followed by a switched-capacitor filter so that the filter tracks the frequency being generated.
This is a good general paper on the topic of sinewave generation.
If the load is a solenoid or something else very inductive, then the steps won't matter.The inductor will be its own filter.

I was commenting on the circuit that was shown., not what could possibly be attached to it . I am not a master mind reader!!

 

So the integrator time-constant determines the output amplitude, thus the amplitude will deviate as determined by the time-constant tolerance (which likely is tolerable).

Yes - in this respect, the GIC-based resonator as mentioned in post#15 and #30 is preferrable.

 

I just came across an article about a solid state version of the classic "twin T" RC oscillator, That version used diode clamping to control the amplitude and a 741 opamp. The frequency for that model went down to 0.01Hz, still a good sine wave, not in need of serious filtering.

 

The Twin Tee circuit I saw today used an op-amp, and the feedback signal into the inverting input was clamped with a diode. so the amplitude stayed constant.