more loop gain can reduce distortions that are measured by the diff input
I collected a few ideas: http://www.diyaudio.com/forums/solid-state/45794-high-loop-gain-composite-op-amp-circuits.html
in the course of making my own high loop gain composite amps I did have to learn a little bit about designing conditionally stable circuits, actually used Popov, Circle Criteria
Noise power is a minimum when the current noise and the potential noise are equal.
For the AD797 typical numbers that means about 450 ohms, maybe a little low.
But the 797 can comfortably handle the traditional 600 ohms, and the noise is only a little worse.
600 ohm * 2pA /rt Hz = 1.2 nV /rt Hz
That 1.2 nV /rt Hz + 0.9 nV /rt Hz with RMS summation = 1.5 nV /rt Hz, a conveniently round number.
The LT1468 has typical 5 nV /rt Hz potential noise, just a trace over 10 dB more, before we even consider the current noise.
The exact calculation of noise in an oscillator is complicated but I think the relative values are correct.
A pity Scot Wurcer hasn't been in the thread for a while.
Hi Bob
So why not AD797 as per back-of-envelope calculations above?
Best wishes
David
Last edited:
Hi Richard
Did you eliminate the possibility of low level parasitic oscillation?
Cyril Bateman's famous capacitor distortion articles have a discussion of distortion problems with the AD797, but they were eventually fixed.
I suspect hi frequency parasitic oscillation because the symptoms exactly match Bob's description of the resultant distortion.
And it is known that the hi Gain Bandwidth Product of the AD797 make it susceptible to this issue.
Best wishes
David
The situation regarding optimum voltage/current noise tradeoff is quite complicated. First of all you have to consider the maximum capacitance you can afford. You might find that 1 uF is about the largest conveniently realizable capacitance if consistently excellent distortion performance is required (as far as I know, the APx555 still uses 300 nF for the lowest range). This results in a 10 nF capacitance for oscillator frequencies at the upper end of the audio frequency range, which is above 1 kOhm for most of the audio frequency range. So a BJT opamp optimized for low voltage noise in the 1 nV/rtHz region may not be the best option.
Another consideration is the operating level of the oscillator. It might be easier to increase the level than to reduce noise.
Samuel
Indeed handling large-signal conditions (in particular clipping, current limiting, slew-rate limiting and power up/down) is challenging for amplifiers with loop gain functions of order 3 and higher. I have not yet got to the point of using formal methods to proof global stability (but hope so in the future); with a bit of intuition and experience, suitable clamping circuits can be found nonetheless.
Samuel
The noise current will create significant noise voltage across the fairly high impedance of the R-C tuning elements associated with the integrators.
This will be worse at the low-frequency end of a given frequency range. Calculate the net value of impedance seen by the op amp looking back into the inverting input, and I think you will see what I mean.
One might argue that this impedance should equal the noise impedance of the op amp at the center of the frequency band as a compromise. Who knows, this could push one to consider the use of a low-noise JFET op amp.
BTW, I would think BJT op amp input current noise would be a significant problem for classic Wein bridges that use variable air capacitors.
Cheers,
Bob
Bruce Hofer's main point in using dual op amps in place of a single op amp is to isolate the sensitive input stage from garbage on the power rails from class AB output stage operation, and from thermal feedback from the output stage. This is accomplished by using 2 separate op amps powered from separate supplies (perhaps separate only in that the supply for the input op amp is just additionally filtered.
The use of a discrete external JFET differential pair in front of a single op amp would seem to serve the same purpose, at least in principle. And without as much compensation difficulty. Might cost a bit more, but I'm sure Bruce was not particularly sensitive to cost in this particular function.
Cheers,
Bob
Not sure I can help here, but I did build the two op-amp (all inverting) oscillator with two 797's and the SSM VCA (my idea) and got -130dB no oscillations no start up issues, etc.
I also built one as an experiment with a power resistor buffered off of the output glued to a precision PTAT resistor as gain control in an insulating box. Even at 1kHz it took 10 min to stabilize but the Q was so high the noise sidebands disappeared noticably from the fundamental. I was not able to measure lower distortion reliably though.
EDIT - To clarify the PTAT resistor was chosen because it had vanishing voltage coefficient of non-linearity and being 1K it was simply one leg of the Wein circuit, I truly tried to make an oscillator with no residual non-linearity to prove Barney Oliver's thesis. The settling time was orders of magnitude worse than with the VCA tweeked to get -130dB.
Last edited:
Yes but I still haven't found one that can match the distortion performance of bipolar input op amps. Maybe a composite design can do this.
Last edited by a moderator:
You're probably right, but bear in mind that the integrators operate the op amp in inverting mode, so the usual higher common mode distortion of JFET op amps is not an issue.
Cheers,
Bob
When thinking about the opamp for the ULD oscillator, then an discrete or composite opamp maybe is the real alternative for the best results. In this case, there is the one specific - inverted amp configuration is preferred because of the potentially lower distortions. Then when we build this amp, maybe we not need the differential cascade at it's input. When FETs are used, then the input cascade in the simplest case may be built as that:
This maybe can give us some preferences...
This maybe can give us some preferences...
Doesn't matter. The whole point is to determine the optimal impedance level and then select the R and C so that is what the op-amp sees.
Subject, as Samuel points outs, to availability.
A 10 uF polypropylene is affordable for the bottom frequency.
That puts the impedance just where we want, around 600 to 1k ohm.
A very small penalty in temperature stability is hardly likely to matter at 20 Hz.
Are there other problems?
I plan a simple switch rather than a frequency selector times a multiplier.
That means I can optimize impedance at each frequency individually.
So a BJT with noise around 1 nV /rtHz does look the best option in this case.
I assume that the level has already been selected to optimize the noise/distortion trade-off.
As I wrote above, my point is not to have hi impedance in the R C elements.
I haven't done this formally yet.
My first intuition is that the R needs to be around 1 k ohm, in quadrature with the C so the op-amp sees ~700 ohms.
Need to check in LTSpice.
I plan switched frequencies with a sweep potentiometer that has a "calibrated" position, just like an oscilloscope time-base.
Unfortunately that makes it less convenient to have the optimal value in the center.
The distortion hits a much sharper "wall" as the resistance decreases than the noise as the resistance increases.
So it would be better to increase the resistance and sweep down in frequency.
But I would prefer to sweep up in frequency.
Haven't found a smart solution to these issues yet.
I have considered a "defeat" switch that switches out/in the sweep pot plus some parallel resistance across the usual fixed resistance.
Maybe results in a funny control law?
Best wishes to all
David
Last edited:
Hi Scott
Thanks for physically verified data, -130 dB was about my expectation and is all I need.
In relation to the discussion above, did you consider noise optimization or was it mainly to test the leveler response and distortion?
Do you remember approximately what impedances you used around the op-amps?
Best wishes
David
I was interested in the distortion mainly, the values should be the same as in the Bateman article.
Now as I try to remember, the settling time might have been longish when tweeked to -130dB. BTW everyone the simulation is trivial, in LTSPICE take the Wein bridge oscillator and use an ideal VCVS with nothing but a third order poly transfer function (V = A*Vin - B*Vin**3) where B is made smaller and smaller. The amplitude and settling of the oscillation is a direct function of B (you need to kick start the transient analysis). You can get it to take hours to settle for VERY low THD.
Last edited:
Measuring oscillator noise
I'm glad we are finally having more discussion about oscillator noise in this thread, rather than just obtainable THD. We all know that there are different tradeoffs in oscillator design that can affect both oscillator noise and THD, sometimes one at the expense of the other.
Simulation of an oscillator just on the verge of oscillation can give us some good insight, but let's discuss some techniques for measuring and quantifying the noise of an oscillator that has been built.
Consider the following scenario. Two guys have built low-distortion oscillators at 1kHz using different circuits or component types or values. Both oscillators produce -120dB THD. But one design is noisier than the other. How best do we measure and quantify the noise on the bench with normally available test equipment?
Because of the bandpass filtering nature of oscillator designs, we normally expect the noise to be most dominant in the frequency region immediately surrounding the fundamental. This will generally consist of both voltage noise and phase noise. Agreed?
An analog spectrum analyzer should show this noise as broadening of the skirts of the fundamental (agreed?). But how best is that presented as data?
An FFT-based analysis, as available with a PC-based system, would seem not to give a satisfactory indication of noise due to the averaging used (agreed?).
What are our options? How do we articulate the noise performance of these oscillators?
Cheers,
Bob
I'm glad we are finally having more discussion about oscillator noise in this thread, rather than just obtainable THD. We all know that there are different tradeoffs in oscillator design that can affect both oscillator noise and THD, sometimes one at the expense of the other.
Simulation of an oscillator just on the verge of oscillation can give us some good insight, but let's discuss some techniques for measuring and quantifying the noise of an oscillator that has been built.
Consider the following scenario. Two guys have built low-distortion oscillators at 1kHz using different circuits or component types or values. Both oscillators produce -120dB THD. But one design is noisier than the other. How best do we measure and quantify the noise on the bench with normally available test equipment?
Because of the bandpass filtering nature of oscillator designs, we normally expect the noise to be most dominant in the frequency region immediately surrounding the fundamental. This will generally consist of both voltage noise and phase noise. Agreed?
An analog spectrum analyzer should show this noise as broadening of the skirts of the fundamental (agreed?). But how best is that presented as data?
An FFT-based analysis, as available with a PC-based system, would seem not to give a satisfactory indication of noise due to the averaging used (agreed?).
What are our options? How do we articulate the noise performance of these oscillators?
Cheers,
Bob
Your best bet is to use the same language as you would for an LO in a critical comms application.
And what is that and how can it be measured with ordinary equipment?
My recollection is that noise of oscillators in critical comm apps is specified as something like phase noise of some number of picoseconds, rms.
Cheers,
Bob