Low-distortion Audio-range Oscillator

[Dave Zan]

AD797 LTSpice model?

Soon I will need to stop my hand-wavy discussion and actually start into Spice.
Anyone have any comments on the LTSpice model for the AD797?
May be I should ask this question in a different thread but there are a few experts here and perhaps Scott Wurcer would care to comment.

Best wishes
David

 

[dimitri]

Glen uploaded PCB design files for his oscillator here

 

[scott wurcer]

Soon I will need to stop my hand-wavy discussion and actually start into Spice.
Anyone have any comments on the LTSpice model for the AD797?
May be I should ask this question in a different thread but there are a few experts here and perhaps Scott Wurcer would care to comment.

Best wishes
David

Macro models are what they are, not too much IMHO. I wouldn't count on much and certainly not fine scale distortion.

 

[davada]

I "set the Wayback machine to 6th of May 2012" to check on what Glen actually wrote >Here<
It's not obviously incorrect but it is odd.
The substitution of a faster op-amp will reduce the Q enhancement at hi frequency and that seems to have happened, as predicted by theory.
The odd bit is why his oscillator needed Q enhancement to function.
The Q enhancement seems to have coincidentally cancelled out low Q at hi frequency.
I don't understand why his amp had low Q at that frequency, it shouldn't be the op amps.
Perhaps lossy capacitors?
It is also a mystery why this comment was removed.
Perhaps it was accidental when he reformatted the site but I wonder if he found some problem and removed it deliberately.
I have sent him an e-mail but no response so far.
It has sparked a few new ideas.
Seems it should be possible to have an oscillator on the borderline of oscillation just by correct trim of the inverter and integrators.
Then the leveler loop has only to maintain the level, can be very heavily decoupled for ultra low distortion and noise.
Simple to implement and tune.





Thanks to you both. I hadn't seen either of these before.
Looks like I am OK to at least 100 mV and that's all I plan to use.

Best wishes
David

Maybe I'm not understanding your comment but oscillators don't need Q enhancement to operate. It's just a phenomena of active filters that use feedback. Using wider BW amplifiers just pushes the beginning of the effect further out. Hopefully beyond the working bandwidth. It's an exponential curve. The gain rises with the Q and this in itself lowers the GB. So it's two fold.

I suppose you could exploit the Q enhancement effect to you advantage but I think you would have to go about this empirically.

I haven't been able to duplicate the effect in spice. I'm not sure the macro models support it. But this doesn't make sense to me.

 

[Samuel Groner]

Was your trim in the usual loop from the first integrator back to the inverter, or in the outer loop?

I'm not sure what you mean by "outer loop"; the trim is in parallel with the multiplier (does the same thing), so from bandpass out to inverter.

I just see it as a consequence of the Bode relations and clear from the Bode plot. But perhaps I miss some issues, any references?

EN195.pdf is a good summary (linked before?).

It occurred to me yesterday that the separation of the leveler loop dynamics from the oscillator loop is very similar to the same problem in switch-mode power supplies. There is a substantial literature on that, have you looked at any of it?

Modeling PLLs is also very similar.

Samuel

 

[Edmond Stuart]

AD797 model

Soon I will need to stop my hand-wavy discussion and actually start into Spice.
Anyone have any comments on the LTSpice model for the AD797?
May be I should ask this question in a different thread but there are a few experts here and perhaps Scott Wurcer would care to comment.

Best wishes
David

Hi David,
Have a look at this post for more info about the input stage and noise.
Below you see a simplified model at the device level. Probably not 100% correct, but better than a macro model.

Cheers, E.

 

[scott wurcer]

Hi David,
Have a look at this post for more info about the input stage and noise.
Below you see a simplified model at the device level. Probably not 100% correct, but better than a macro model.

Cheers, E.

Yes that will do much better. There is Ib comp but I would not worry about it.

 

[Dave Zan]

Macro models are what they are, not too much IMHO. I wouldn't count on much and certainly not fine scale distortion.

No - at this point I don't expect to model distortion with any accuracy, just understand the overall behavior.

Hi David,
Have a look at this post

Thanks Edmond, it is to your usual standard of professionalism.
More than I need now but it inspires me to try for the next level of accuracy.

Best wishes
David

 

[Dave Zan]

...oscillators don't need Q enhancement to operate. It's just a phenomena of active filters that use feedback.

Sorry I wasn't clear when I tried to be brief.
"Q enhancement" is just a manifestation of the effect of phase shift on active filters.
I meant that he had used the same phase shift to keep his oscillator within the bounds that the leveler loop could control, and so operate correctly.

I suppose you could exploit the Q enhancement effect

Yes, that's pretty much what he's done.
Glen has sent some helpful email comments since my initial speculation and I think we are on the way to some clarification.

I think you would have to go about this empirically.

I think I can do this from the Bode plots, which would please me.

I haven't been able to duplicate the effect in spice. I'm not sure the macro models support it. But this doesn't make sense to me.

Or to me, it should model perfectly well even with macro models.
I will try this in a while, have a few tasks to finish first.

Best wishes
David

 

[Dave Zan]

I'm not sure what you mean by "outer loop"; the trim is in parallel with the multiplier

OK. I meant that it should be possible to trim for oscillation just with adjustments to the inverter and integrators in the main (or "outer") loop.
This is as opposed to the usual, additional loop from the bandpass out to inverter input.
Not sure if there's any benefit yet, but I haven't seen it done so I wanted to learn if it was a new trick

EN195.pdf

Don't know if it's been linked before but it's new to me, looks excellent, thanks.

Best wishes
David

 

[1audio]

Rexotech ULD4

I got the latest Rexotech ULD 4 sample and made a quick measurement of the distortion. Because I can't figure out how to do an offset in ARTA the plot will need to be scaled- 0dB is actually -70 dB. So the -60 line is -130 dB. The distortion shown is really at the limits of the 725. At this level the potential of phase shifts causing cancellations between the source and the internal distortions is significant.

In any case it does deliver very good performance for a very reasonable price. I'll do more measurements as soon as I get some time.

 

[davada]

Manipulate the voltage setting in the calibration.

 

[forr]

Because I can't figure out how to do an offset in ARTA the plot will need to be scaled- 0dB is actually -70 dB. So the -60 line is -130 dB.

This is a setup I used with input levels at around 1 V :

 

[1audio]

My problem is that 0dB would be 3,000V. The spectrum is the output of the Shibasoku 725. I'm not sure ARTA can handle that much offset. I have another way that works to get the info. I'll make some more measurements soon a better way. I'll make sure they are properly calibrated.

 

[davada]

My problem is that 0dB would be 3,000V. The spectrum is the output of the Shibasoku 725. I'm not sure ARTA can handle that much offset. I have another way that works to get the info. I'll make some more measurements soon a better way. I'll make sure they are properly calibrated.

Demian I'm not clear about the reference.

The output from the 725 monitor is 1Vrms FS.
Do you mean you want to change the scale of ARTA by +10dB?
Where is the 3V coming from?

You can calibrate ARTA at any level below the sound card FS.
For example, input -20dBV and in ARTA cal type in 100mV. After accepting the level (the accept button), ARTA will display -20dBV.
If you want to trick ARTA input -20dB and type in 1000mV. Then ARTA will display -40dB with a -20dBV input.

 

[1audio]

I know my description is confusiong. What I want to do is scale the output of the 725 to match the input. If the 725 is boosting the input by 70 dB I want to scale the display so the indicated levels are scaled correctly and a -130 dB harmonic is displayed at -130 dB.

I'll try your method and see if I can make it work. Essentially enter 1V and set the cal to .1 mV. In Praxis there is a simple offset field independent of calibration.

 

[davada]

I know my description is confusiong. What I want to do is scale the output of the 725 to match the input. If the 725 is boosting the input by 70 dB I want to scale the display so the indicated levels are scaled correctly and a -130 dB harmonic is displayed at -130 dB.

I'll try your method and see if I can make it work. Essentially enter 1V and set the cal to .1 mV. In Praxis there is a simple offset field independent of calibration.

No don't bother ARTA won't do it.

We'll have find another piece of software.
ARTA functions well but no control over the graphics.

I'd like to find something that shows phase of the harmonics relative to the fundamental. We might able to use that to sort out the harmonic cancellation.

 

[1audio]

I'm not even sure how you would describe the phase of a harmonic with respect to its fundamental. I understand what you want, just not how to get there.

Praxis will display in a useful way. Unfortunately Bill Waslo had to discontinue it because its not happy with Win 7 and later and Borland (the development system) is history. Something current would be nice.

 

[davada]

I'm not even sure how you would describe the phase of a harmonic with respect to its fundamental. I understand what you want, just not how to get there.

Praxis will display in a useful way. Unfortunately Bill Waslo had to discontinue it because its not happy with Win 7 and later and Borland (the development system) is history. Something current would be nice.

When did Borland disappear? I used Delphi years ago. Did MS drive them into the ground.

 

[1audio]

Here is the ULD4 and Victors 1 KHz compared. Victors is better BUT the ULD4 comes ready to use.

These are scaled correctly. First is ULD4 , second is Victor's. Also shows how clean things are with battery power. I just wish 9V batteries were not so expensive, especially when I forget to turn it off.