OK, I can do French and have help for Deutsch and, less easily, Italian, Japanese, Dutch, Spanish, Danish, Chinese and a few others (have to love a multicultural society).
So any dialect will do.
Sounds impressive, can it be used as a stand-alone unit or does it need the PC app?
Expected Time of Availability and cost?
Best wishes
David
It is controlled by a micro controller and PC app via USB. The uP is completely ground isolated.
The communication is 3-wire SPI.
I'm working on a stand alone controller for it so that there is an option to either use the USB or panel. External control is down the raod though. I've also entertained the idea of remote control as well. It high performance but not cheap as the parts are expensive.
The communication is 3-wire SPI.
I'm working on a stand alone controller for it so that there is an option to either use the USB or panel. External control is down the raod though. I've also entertained the idea of remote control as well. It high performance but not cheap as the parts are expensive.
Samuel Groner has a publication in German that gives a good treatment. Maybe check with him for the title.
Samuel,
I find the drop in the distortion of the APx555 from 1,000 hertz to 3,000 hertz and above quite interesting. Allowing for band switching at 2,000 hertz for the drop and the rise again above 20,000 hertz would seem to explain a bit. But my suspicion is that if there is almost no measurable distortion this high in frequency but not lower there is something else going on. I would consider that that the unit is calibrated in this band. As a result it reads as close to zero here as the calibration can get it and the error shows up in the other bands.
ES
Samuel has written some of the best information I found.
I will check if there's any untranslated work that I may have missed, thank you.
Best wishes
David
Yes, this is strange. I noted it during the measurements, but didn't had time to verify. There is a non-zero chance that I messed something up.
Samuel
Samuel,
If I think you made a mistake I will mention it.
What I see in your measurements is that the SG505 looks best on the 200 - 2,000 hertz range. The System One from 2,000 - 20,000 hertz, The SYS-2722 from 200 - 2,000 hertz and as mentioned the APx555 from 2,000 - 20,000.
Now as you showed both frequency and level measurements, there seems to be a family of spectra for a given frequency that is slightly modified by the level adjustment.
The Apx555 does show this family resemblance best at 5 volts out on 3,000 & 10,000 hertz. So I suspect this is the band where the system either is adjusted or tweaked during design and testing. I also suspect your measurements show where the other units were optimized.
Now it is possible to calibrated the internal A/D processes to compensate for the test oscillator. Of course this would give false readings on other sources. I do not suspect that here. My OPINION is that the APx555 really is that good on those frequencies. This would lead me to suspect that the parts that were tweaked to value for that band could also have additional band switching tweaks to them. That would allow for better results in the other bands at a significant increase in complexity and initial adjustment. (Although if the band switching is just done with capacitor changes then tighter matching of the capacitors may accomplish the same result.)
ES
I don't think we know enough about the APx555 to make too many guesses on how it works. For example is the source is digital with an analog cleanup or pure analog? The analyzer may be pure AD or it may have a tuned filter in front of the ADC. Further, if the number is derived from software analysis of the digitized waveform the bandwidth of the analysis matters. iIbelieve the posted figures on the APx555 use a 20 KHz upper limit (and may have an automatic BW switch when the fundamental increases). They also show best performance at 2V RMS.
I have also seen issues around synchronous measurements- using the same clock for a DAC and an ADC, that seem to hide some artifacts and get better results than the actual measured analog performance, as well as the inverse with an analog source, usually having to do with the "close in phase noise" of a pure analog oscillator.
I would think to get any real handle on either the sources or the analyzers you would need more than one of each in every measurement. That would be quite a circus. More useful would be the ability to show how each instrument affects the measurements and doesn't always show the absolute truth.
I have also seen issues around synchronous measurements- using the same clock for a DAC and an ADC, that seem to hide some artifacts and get better results than the actual measured analog performance, as well as the inverse with an analog source, usually having to do with the "close in phase noise" of a pure analog oscillator.
I would think to get any real handle on either the sources or the analyzers you would need more than one of each in every measurement. That would be quite a circus. More useful would be the ability to show how each instrument affects the measurements and doesn't always show the absolute truth.
Artifact are usual coming from false bit size use/mix & math rounding.
Coherent measurement (multiple of sample size / FFT size) allows you to get ride of the FFT smearing...
Here is nice reading on B&K Dyn-X (dual 24 ADC) how get into the 160dB dynamic range http://www.bksv.com/doc/articles/dynx_tech_article.pdf
Hp
The generator is a conventional RC oscillator. The analyzer is an evolution of the System One/System Two approach with two cascaded notch filters; the second filter is implemented in the digital domain (see Technote 124, AP High Performance Audio Analyzer & Audio Test Instruments : Downloads). This approach ensures that the ADC of the analyzer does not have any significant impact of the performance.
Samuel
I found more, of course, but not much that was focused on audio applications.
I came across this Rosens Oscillator but it has multiple errors.
So I searched to see if there was a correction but the first search result was this thread - turns out this is the oscillator that inspired the circuit of the OP in the first place.
So has anyone corrected the maths?
I have worked out some of it myself, which is educational but slow, would like to kick some ideas around.
Samuel, do you have any untranslated work that would be relevant?
I have a new idea for a leveller circuit if you're interested.
Best wishes
David
Why don't post your idea here? Unless it top secret.
Is this a state variable design or something else?
Oscillators are essentially band pass filters with positive feedback.
You can find a lot on the subject of band pass filters, mostly state variable, on the net.
Bobs write up on his oscillator analyzer is on of the best. Samuel's thesis is available on this thread. There is a link posted some 100s of entries back. There are web pages around but be careful filter out what is personal from what is actual theory.
The optimal control loop loop gain for best settling time is said to be 2 for a state variable oscillator. You can distribute this gain through out the loop. This includes gain at the difference amplifier, integrator, summing amplifier and multiplier.
Probably the most difficult is getting the time constant right in the integrator. It's different for every frequency but usually one can get away with one TC for each decade range and still have reasonably good settling time. If you don't care about settling time then the subject of TC is more relaxed.
The integrator adds gain to the loop. The more integral gain used the more you have to back off on the proportional gain. Without integral, proportional control runs at an offset.
Proportional on it's own is more suited to manual control rather than automatic control. The integrator accumulates the proportional errors from cycle to cycle. Once the oscillator has settled the integrator holds the total error which allows the proportional error to reach zero.
The only purpose of the integrator is to drive the error to zero. Best to use the minimum integral gain to get the job done.
Too much loop gain causes level swaging. if you see this then back off the controller gain.
It is more often caused by too much integral gain. Usually lengthening the integrator TC can clears this up or you just have to much proportional gain. If the gain is too high then the controller hunts which means overshoot or undershoot but never just the right amount of error.
Is this a state variable design or something else?
Oscillators are essentially band pass filters with positive feedback.
You can find a lot on the subject of band pass filters, mostly state variable, on the net.
Bobs write up on his oscillator analyzer is on of the best. Samuel's thesis is available on this thread. There is a link posted some 100s of entries back. There are web pages around but be careful filter out what is personal from what is actual theory.
The optimal control loop loop gain for best settling time is said to be 2 for a state variable oscillator. You can distribute this gain through out the loop. This includes gain at the difference amplifier, integrator, summing amplifier and multiplier.
Probably the most difficult is getting the time constant right in the integrator. It's different for every frequency but usually one can get away with one TC for each decade range and still have reasonably good settling time. If you don't care about settling time then the subject of TC is more relaxed.
The integrator adds gain to the loop. The more integral gain used the more you have to back off on the proportional gain. Without integral, proportional control runs at an offset.
Proportional on it's own is more suited to manual control rather than automatic control. The integrator accumulates the proportional errors from cycle to cycle. Once the oscillator has settled the integrator holds the total error which allows the proportional error to reach zero.
The only purpose of the integrator is to drive the error to zero. Best to use the minimum integral gain to get the job done.
Too much loop gain causes level swaging. if you see this then back off the controller gain.
It is more often caused by too much integral gain. Usually lengthening the integrator TC can clears this up or you just have to much proportional gain. If the gain is too high then the controller hunts which means overshoot or undershoot but never just the right amount of error.
Dave it would help if you defined what you expect from an oscillator.
Ultra low distortion
Fast settling.
Continuously tuned.
This sort of thing.
Any oscillator that used clamping to control the damping of the oscillator won't be low distortion.
Most of the phase shift oscillators I've seen out there use diode clamping to control the level. It works, but if you want really low distortion don't consider them.
Ultra low distortion
Fast settling.
Continuously tuned.
This sort of thing.
Any oscillator that used clamping to control the damping of the oscillator won't be low distortion.
Most of the phase shift oscillators I've seen out there use diode clamping to control the level. It works, but if you want really low distortion don't consider them.
Just an idea that should reduce leveler distortion and time to settle.
Not so radical as to require "top secret" but perhaps worth some work.
Just wanted to check if it had already been invented before I decided what to do. I'll PM you
The idea would work with State Variable and similar quadrature oscillators.
Yes I have both Bob and Samuel's papers.
Samuel's work is probably beyond what I require but very educational.
Yes, it was when I looked closely at the Rosens paper that I noticed the errors.
"said to be"? I would like to be able to derive this, as well as optimize time constants etc.
The "PID" control theory nomenclature is not the way I think about it, I visualize Bode plots and Bode Maximum Feedback.
So I need to think more about this, to have both perspectives. Thanks for the comments.
Thank you, if my quick check is correct then the specifications are not very impressive but I will have a closer look when the library re-opens after Easter.
Well, initially I just want to understand the trade-offs, because I can find almost no information on this.
For instance lower distortion oscillators usually settle more slowly.
But what's the mathematical limit to how fast an oscillator can settle for a specified distortion?
Or if we demand, say, 2x less distortion does this mean the oscillator takes 2x as much time to settle, or 4x times or ?
As an ex-mathematics student, I want some metric to see if one type really is better than another.
On a practical note, I want continuously tuned, ultra low distortion and fast time to settle.
Also simple
I do have an idea for continuously tuned with ultra low distortion spot frequencies at 1, 2, 5 multiples.
Probably in 4 decades, 10 Hz to 200Hz up to 10kHz to 200kHz.
The last continuous sweep mostly for frequency response of amplifiers and filters, distortion less critical.
Ric Lee would probably say "Phase shift oscillators are EVIL".
I'm less dramatic but no, I have NO plans to consider them.
They are actually an example of a circuit that is simply worse, more distortion for no benefit.
So, not all circuits are equal, what has the best measure of merit?
Best wishes
David
Last edited:
Just an idea that should reduce leveler distortion and time to settle.
Not so radical as to require "top secret" but perhaps worth some work.
Just wanted to check if it had already been invented before I decided what to do. I'll PM you
The idea would work with State Variable and similar quadrature oscillators.
Yes I have both Bob and Samuel's papers.
Samuel's work is probably beyond what I require but very educational.
Yes, it was when I looked closely at the Rosens paper that I noticed the errors.
"said to be"? I would like to be able to derive this, as well as optimize time constants etc.
The "PID" control theory nomenclature is not the way I think about it, I visualize Bode plots and Bode Maximum Feedback.
So I need to think more about this, to have both perspectives. Thanks for the comments.
Thank you, if my quick check is correct then the specifications are not very impressive but I will have a closer look when the library re-opens after Easter.
Well, initially I just want to understand the trade-offs, because I can find almost no information on this.
For instance lower distortion oscillators usually settle more slowly.
But what's the mathematical limit to how fast an oscillator can settle for a specified distortion?
Or if we demand, say, 2x less distortion does this mean the oscillator takes 2x as much time to settle, or 4x times or ?
As an ex-mathematics student, I want some metric to see if one type really is better than another.
On a practical note, I want continuously tuned, ultra low distortion and fast time to settle.
Also simple
I do have an idea for continuously tuned with ultra low distortion spot frequencies at 1, 2, 5 multiples.
Probably in 4 decades, 10 Hz to 200Hz up to 10kHz to 200kHz.
The last continuous sweep mostly for frequency response of amplifiers and filters, distortion less critical.
Ric Lee would probably say "Phase shift oscillators are EVIL".
I'm less dramatic but no, I have NO plans to consider them.
They are actually an example of a circuit that is simply worse, more distortion for no benefit.
So, not all circuits are equal, what has the best measure of merit?
Best wishes
David
Hi Dave,
It a misconception that settling time has anything to do with distortion.
My oscillator design has both fast settling and extremely low distortion and 16 bit tuning resolution in each range. It's also very low noise. But you won't get it all in simple or cheap.
You should try to get a grasp on PI control because just about every sine oscillator uses at least proportional control. This is with exception to thermal stabilizing elements such as lamps etc.
Victor's is an example of very good ultra low distortion oscillator but it's limited to fixed frequency. Long settling time yes and there is virtually no ripple at the Jfet gate. The long time constant is what filters the ripple out but it's not the only way.
Samuel's oscillator design is near completion. It has 1800 parts. This is SOTA.
Yes, I wrote "lower distortion oscillators usually settle..." precisely because I don't think the trade-off is fundamental.
Reasonably simple and cheap is one of my requirements.
So the trick is to work out just how excellent is possible within that constraint.
Yes, I understand feedback theory fairly well, just that I don't think in terms of PID.
So how would Bode's Maximum Feedback results apply?
My respects to both of them, I am particularly impressed by the sheer perseverance required to finish a project with almost 2k parts.
My limit is about the complexity of Bob Cordell's SV oscillator.
But I want sweep frequency option and whatever improvements I can make.
I am usually skeptical of people who fiddle cluelessly with op-amps but in this case AD797 should lower noise and potentially distortion.
Best wishes
David