Oscillator Distortion

Yellow Springs Instruments were the pioneers in glass-encapsulated thermistors. There are plenty of other manufacturers who make them today. I've used Betatherm parts.

What Cursitor Doom seems to be missing is that you need a particular value of resistance to stabilise a given Wein Bridge circuit. You can buy thermistors which have a wide range of resistances at room temperature, and he going to need one that has less resistance than the other resistors in the bridge, but more than enough that the reduction in resistance as it gets warm drops the gain around the loop to less than one at just above the desired working amplitude.

It would be simple enough to work out if you knew what you were doing.

My guess is type R23 is 2kohm at room temp (the R53/RA53 beloved by hobbyists in the 1960s/70s was 5k). eBay probably has some close enough (1.5k to 3.3k at room temp?) replacements.

This link might help you:

piglet

Thanks for the info, Erich; much useful info in that datasheet. Nothing showing up on Ebay at the moment, but I'm making enquiries of vintage parts sellers which hopefully might bear some fruit. Failing that I'll just have to build a new oscillator stage from scratch using a spare vacuum thermistor from my parts bin.

I saw a seller with Littelfuse GL202F9J which might do electrically if maybe a bit slow thermally? That part is also available from Mouser.

piglet

rlkey=jozht29aj1u6ocxnmw8okedrq&st=ohzxz0mw&raw=1>

According to the datasheet, this device boasts "a fast thermal response time" but doesn't quantify that. It could do the trick with a bit of bias tweaking.

That part is intended for temp measurement rather than self heated regulation so will be slower than the well isolated original R23 part - I expect it would work OK but you'd notice a longer amplitude bounce settling time after each change of frequency.

piglet

Got it, thanks.

Not entirely unrelated to this subject. Image a Wien bridge oscillator, stabilized with a lightbulb. This could be described with differential equations, including the heating and cooling of the tungsten wire.

I suspect a relation between the cooling time constant and the periods of an oscillator, and the distortion. A rule could be if you need N periods to have the distance to the the stable wave halved, you have circa 1/N distortion.

Is this known territory?

Groetjes Albert

You people who explain everything they see in terms of higher mathematics normally leave me baffled. But I believe I've deduced what you're trying to say and it's not an issue in this case. Thanks for the suggestion anyway.

The whole idea of making a sine oscillator is positive feedback with an AC gain  =1  while DC biased with negative feedback and not boosting to a pulse with high gain. So something must limit the gain smoothly like a hot bulb with lower impedance driven by a lower voltage.

Below the inverting gain can be  R1/(R2 + (R(Q1)) =2 maximum unless Q2 is slowly turned off. and quickly turned on ;) otherwise known as fast attack , slow decay.

Whatever is boosting the gain of your circuit or NOT cutting the gain with high R must be fixed.

Cheers

Tony Stewart, near Toronto

EE since 1975

Learning how to retire since age 54.

I forgot about image if not shown with plain text

There's no requirement that the amplitude regulation be continuous. It can also be done with a comparator that fires occasionally when the amplitude gets too large.

One approach is to have the oscillator's tail current controlled by an integrator, with a resistor causing the current to increase slowly, and a comparator causing it to kick it down by a fixed amount whenever the amplitude threshold is crossed.

Adjusting the phase of the signal at the comparator so that the kick arrives at the peak of the collector current waveform reduces the resulting small phase jitter.

Cheers

Phil Hobbs

Like any PLL or ILL or AGC , you only have samples to correction error and never "continuous" correction unless integrated to a DC level. It is a discontinuous error correction with a filter to hold between inputs.

There are many ways to correct error by limiting the gain > 1 yet ensuring it can reach 1 to oscillate. S/H, non-linear soft limiters, filters, peak detectors/comparators. It all depends on your specs for lock-in time and THD/IMD specs of the sine wave.

Any oscillator with a nonlinear or bilinear gain control element that has to respond during a cycle has to deal with the distortion caused by that element. OTAs, JFET variable resistors, PIN diode attenuators, Vactrols, light bulbs, and so on, all have that problem. Tail current sources can avoid it, because you can make them as stiff as you like by cascoding, and filter the control voltage as well as you like. (I often use two- or three-pole capacitance multipliers on the supply rails of discrete circuitry, which is a similar idea.)

Once the gain is close to 1.000, the amplitude growth or decay happens over many cycles, so the update rate can be correspondingly low--just enough to compensate for temperature and power supply variations.

Something like a baby-scale loop can help with that, to allow the amplitude to stabilize rapidly before slowing down the time constant on the control voltage.

If the oscillator needs to be tunable, you get into slightly different tradeoffs, but you still don't have to deal with nonlinear gain control elements.

Cheers

Phil Hobbs

The FET regulator is a filtered servo with fast cutoff and very slow build up over thousands of cycles. The same is done with other non-linear limiters using exponentially orders of magnitude higher resistance when regulating so the decay rate is slow and low distortion.

Phil I th>

You’re apparently missing the distinction between a control element that has to respond within a cycle, such as a JFET variable resistor used as a feedback element, and one that just sits at a very slowly varying operating point, such as a cascoded BJT tail current source with a big emitter resistor and lots of bypassing on the base.

The first kind gets run through its (inevitably somewhat nonlinear) I-V curve on every half cycle, regardless of the bandwidth of the control loop. This contributes an amount of distortion that isn’t improved by narrowing the loop BW. The second kind’s distortion can be reduced to any desired degree by careful design.

The active element’s nonlinearity is of the first kind, of course, but that’s just amplifier design.

Cheers

Phil Hobbs

Light bulbs and thermistors can have a controlling DC superimposed on a miniscule signal current, so that the distortion caused by the latter is negligible.

Another alternative is an indirectly-heated thermistor with a very small signal current in a large thermistor which is primarily heated by a separate resistive element. It would be slow to respond, but at 1 Kc/s and -90 dB distortion, a long response time is essential to avoid distortion from the amplitude-settling transient.

Depending on omega*tau_th, sure. The HP 200 exhibits increasing second-order distortion at lower frequencies.

Down at -90 dBc, depending on the signal level you might have to worry about deviations from Ohm’s law in an oxide thermistor. (Metals are pretty linear, but the carrier density in an oxide is going to be much much lower.)

Eventually it’s bound to be a tradeoff between distortion and noise.

Cheers

Phil Hobbs

Use the opamp dual-integrator sort of oscillator with a loop gain of

In the old HP Wein bridge oscillators, the light bulb had a huge influence on gain.