My experience measuring suggests that second is usually the bigger culprit. When its not you are usually way down in all the harmonics. At that stage cancelling may be too difficult to maintain.
When the distortion get to the level where the passive components can be significant (-150 or less) I suspect it will be a losing battle to go further.
When the distortion get to the level where the passive components can be significant (-150 or less) I suspect it will be a losing battle to go further.
I am very interested to undesrtand this ' distortion ' stabilization effect
Do you have a ref where I can read Olivers paper ( allways challenging)
Thanks JPV
In essence its never statically stable and will always need some form of steering to maintain stable operation.
I have had good success when ultra low distortion is not required using back to back Zeners to stabilize the output. With two low pass stages past the Zeners the distortion is quite low for such a crude solution.
HP Journal, April/June-1960, v.11 n.8-10 The Effect of u-Circuit Non-Linearity on the Amplitude Stability of RC Oscillators, by Bernard M. Oliver
L.
Gee Scott.
The math is very nicely presented and at least to me quite approachable. Must have been an absolute bear to get it that far. (I do have issues with some of the assumptions such as flat frequency response of the amplifier. But I certainly don't want to get into them.)
But the interesting detail is the 1949 copyright claim on a 1960 paper!
ES
The math is very nicely presented and at least to me quite approachable. Must have been an absolute bear to get it that far. (I do have issues with some of the assumptions such as flat frequency response of the amplifier. But I certainly don't want to get into them.)
But the interesting detail is the 1949 copyright claim on a 1960 paper!
ES
The kibbitzing is not productive.... more ideas and actual circuits that can be effective in reducing osc thd+n is needed. Specifically, cancellation techniques... new and old.
Thx-RNMarsh
Thx-RNMarsh
Current copyright 1949-1998 obviously added after the fact to each page by HP corporate. I don't know what you saw.
Jim Williams thought this one of the finest pencil and paper technical articles he ever read, that's why Mr. Brisbois' oscillator surprised him (I assume HP had access to the Christmas lamps).
Ed, from now on I will expect to see the same mathematical rigor out of you. Since an op-amp works obviously flat frequency response does not matter and certainly does not affect significantly the important results. Of course you knew that?
Last edited:
My take is that the seconds are there as an artifact reflecting the usual seconds dominant op-amp distortion spectra. In an ideal world a post facto thirds cancellation could make the simplest nearly perfect oscillator.
2h and 3h is all you see as dominant... the other harmonics are always much lower for various reasons. All osc have a 2h null feature inside. but it can often tune to the 3h instead. I have done this before with osc. It leaves you with one or the other remaining dominant harmonic to cancel. cancelling can then be done with a passive notch filter or, here, i hope to find a circuit to do it instead. So far i have one from Samuel to try.
Thx-RNMarsh
Last edited:
good advice... thank you.
idea.... in a combo unit like the Hp339a, we already have a deep notch filter on the fundamental... could there be a way to feedback the residual harmonics to cancel them in the osc ??
Thx-RNMarsh
A notch filter introduces phase shift to the harmonics, It's the phase relationship that makes cancellation difficult.
Difficult is not Impossible. How many degrees shifted?
post phase shift it as needed and lock onto fundamantal etc.
-RNM
post phase shift it as needed and lock onto fundamantal etc.
-RNM
If you can filter that sharply why not just add a filter to the output of the oscillator. 2 pole, would give 12 dB reduction of the second and the higher harmonics would essential vanish.
An active filter would leave you a the mercy of the amplifiers however there may be a topology that has the passive elements after the active one, self-filtering in effect. Tie the AGC to the output to maintain the overall absolute level accuracy.
if you used something like the Maxim programmable resistors getting four stages to track would be pretty easy.
It took me the longest time of looking at the notes on the Shibasoku 725 to realize that it has three cascaded, tuned notch filters and no servo for the notch. Thus no analog multiplier/LDR/etc. to add distortion. Unfortunately the oscillator needs an AGC to stabilize (as we have learned) but low pass filters on the output don't need any active interactions and should be even lower distortion.
An active filter would leave you a the mercy of the amplifiers however there may be a topology that has the passive elements after the active one, self-filtering in effect. Tie the AGC to the output to maintain the overall absolute level accuracy.
if you used something like the Maxim programmable resistors getting four stages to track would be pretty easy.
It took me the longest time of looking at the notes on the Shibasoku 725 to realize that it has three cascaded, tuned notch filters and no servo for the notch. Thus no analog multiplier/LDR/etc. to add distortion. Unfortunately the oscillator needs an AGC to stabilize (as we have learned) but low pass filters on the output don't need any active interactions and should be even lower distortion.
can you breadbpoard the Lp filter with Maxim resistors to see how well it does the job on the output of an osc?
-RNM
Notch phase -
beyong 2H the phase flattens out at a constant which mean the 2H and higher can be useful in harmonic subtraction if appropriately massaged/interfaced for the purpose.
Thx-RNMarsh
beyong 2H the phase flattens out at a constant which mean the 2H and higher can be useful in harmonic subtraction if appropriately massaged/interfaced for the purpose.
Thx-RNMarsh
In all my copious spare time? I'm still working to pay my bills . . .
I'll be happy to get the parts for a volunteer. Even possibly a PCB (two state variable filters, one with an AGC to become an oscillator but I think Davada already tried the trick, in a sense.
Another thought if you are committed to the distortion cancelling you could program a DSP (http://www.analog.com/static/imported-files/data_sheets/ADAU1701.pdf) to detect the oscillator frequency and generate the target cancellation tones. Since the tones would be added at less than -100 dB the artifacts would be minimal at probably -180 db or more.
Another idea for the DSP handy- use that chip as the AGC. It could detect the output level (and frequency) and create a signal with the appropriate level and phase to adjust the output for both amplitude and frequency. At 24 bits it should be clean enough and you would not need the full digital output except possibly to start the oscillator or stabilize it after a transient. The normal SVO only needs a small amount of steering to maintain level so the added signal would be tiny. If you actually generated the target sine wave with it into the SVO its a form of injection lock that could work well across a narrow band. Retune the SVO for the next band (with the Maxim digital resistor).
The Sigma Studio software is pretty powerful but this would need some special expertise. Perhaps Scott could get one of the ADI guys working with the family to look at the concept?
Just to prevent some misunderstandings--unless I'm myself missing something (I can't claim to be sufficiently strong in theoretical circuit analysis to keep up with Oliver's writing), Oliver's proof does not apply to the linear leveling circuits (level detector, error integrator and multiplier) we're usually concerned with while thinking about ultra-low distortion oscillators. It just applies to simple oscillators which rely on compressive nonlinearity in an amplifier stage.
A classic case where you could convert the SFV oscillator to have an elliptical response and notch out the third without an additional low-pass stage.
Samuel