Actually I was looking at one of your competitors products touted as a 'sigma-delta' RMS to DC converter, which is a current product and a lot cheaper. But the input level is quite restricted, and I don't have the idea I really grok how it works. Which bothers me
Jan
I thought the AD536 is still active, be it quite expensive ?
http://eu.mouser.com/ProductDetail/Analog-Devices/AD536AJHZ/?qs=sGAEpiMZZMv9Q1JI0Mo/tX1d6Ox%2b3J6Z
Patrick
http://eu.mouser.com/ProductDetail/Analog-Devices/AD536AJHZ/?qs=sGAEpiMZZMv9Q1JI0Mo/tX1d6Ox%2b3J6Z
Patrick
Yes, Demian proposed to me the option of an RMS converter quite recently, in >this< post.
But it seemed unnecessarily expensive for little to no benefit.
If you are in no hurry then a simple rectifier works well, as Viktors and other have demonstrated.
If you want to settle the amplitude fast then a track hold/sample hold is close to optimal, as recommended here by Davada.
If you want to control the amplitude in real time then a "sum of squares" should work.
Why waste a rare, heirloom AD RMS converter😉
Best wishes
David
Last edited:
Well, I don't know, I see pages and pages of lamenting ripple and settling issues for all these oscillators, and wondered whether an RMS to DC converter might be a worthwhile alternative.
But yes, Patrick, it is expensive, relatively, I get them from Mouser or Digikey, can't remember, for € 15 a pop. That's why I was looking at an alternative for the autoranger kit.
Jan
Last edited:
Actually that is not what you want. What you want is a control voltage, smooth and accurate and fast settling, that represents the oscillator output voltage.
Jan
I am interested in ripple and time to settle too, but an RMS converter seems to me to be "sub optimal", quite apart from the expense.
The first step of an RMS converter is, naturally, to square the input*.
Since a sine wave squared is another sine wave at twice the frequency we do reduce the ripple after we take the mean, but not that dramatically.
Whereas a sum of sine^2 + cos^2 reduces the ripple to zero.
Of course there is the need to match the sine and cos but even minor mismatch should be dramatically less ripple and allow faster response.
On an intuitive level, a sine oscillator is equivalent to a pendulum, two state variables, 2nd order Differential Equation and all that.
To set the amplitude of a pendulum we don't take the square of the velocity or position and take the mean over many cycles to smooth it!
We track and then sample at the peak, one cycle at most.
If we really want we can check it's position and velocity and determine the predicted amplitude essentially instantly.
So to use an RMS detector makes no sense to me, especially if it's an irreplaceable, personally blessed one from Scott.😉
Best wishes
David
* To be pedantic, the first step may be rectification to allow a one quadrant multiplier, no difference.
Yes, I want a smooth control, and to achieve that I want to take the two quadrature outputs and process them.
The "state variable" perspective seems to me to reveal more of the essence of the problem, that there are two variables to control, not to focus exclusively on the output, which seems initially the obvious approach.
The state variable view also makes it clear that the two variables are all we need.
Best wishes
David
Cross posted with Jan.
Perhaps I should have made it clearer that my comments were in the context of a state variable oscillator where sin and cos outputs are explicitly available.
Last edited:
The AD536/AD636 and disco'd THAT2252 are fun parts but in the context of basic oscillator stabilization I don't "get" RMS either.
For complex waveforms RMS provides a more accurate representation of work.
An ultra pure oscillator, by definition, produces a simple waveform.
The AD636 is in production. The 100 piece price in 14P cerdip is $22.62. http://www.analog.com/en/products/linear-products/rms-to-dc-converters/ad636.html#product-overview
Unless RMS detection provides a ripple advantage I don't see how it can outperform Vicktors' super-simple half-wave averaging detector.
Arguments against the $22 part:
1) RMS power measurement. For oscillator stabilization an accurate representation of power is not needed.
2) Dynamic range. The dynamic range requirements of the detector are small. The greatest detector dynamic range is required during lock; after lock the DR requirements are tiny.
3) Linearity and scaling. The AGC loop, being a feedback topology compressor, self-corrects large linearity errors in detector and VCA scaling.
The AD636 might be useful in a sum-of-squares detector but it would require two.
I've spent a lot of time replicating the THAT2252 for dynamics processors using transistor arrays. It's quite easy to do. THAT2252 RMS Detector Replacement Using A THAT300 Array.
Someone might consider using two "RMS" detectors for a state variable oscillator's sin and cos outputs and summing their unfiltered log-domain detectors similar to stereo "True Power Summing." This would provide "four phase" rectification. The output in the 3 mV/dB log domain would represent |sin|² + |cos|².
For those interested an example of the AD536 RMS detector used in the SSL 4000/6000 82E10 channel dynamics processor can be found here: http://www.ka-electronics.com/images/SSL/ssl_82E10.pdf
For complex waveforms RMS provides a more accurate representation of work.
An ultra pure oscillator, by definition, produces a simple waveform.
The AD636 is in production. The 100 piece price in 14P cerdip is $22.62. http://www.analog.com/en/products/linear-products/rms-to-dc-converters/ad636.html#product-overview
Unless RMS detection provides a ripple advantage I don't see how it can outperform Vicktors' super-simple half-wave averaging detector.
Arguments against the $22 part:
1) RMS power measurement. For oscillator stabilization an accurate representation of power is not needed.
2) Dynamic range. The dynamic range requirements of the detector are small. The greatest detector dynamic range is required during lock; after lock the DR requirements are tiny.
3) Linearity and scaling. The AGC loop, being a feedback topology compressor, self-corrects large linearity errors in detector and VCA scaling.
The AD636 might be useful in a sum-of-squares detector but it would require two.
I've spent a lot of time replicating the THAT2252 for dynamics processors using transistor arrays. It's quite easy to do. THAT2252 RMS Detector Replacement Using A THAT300 Array.
Someone might consider using two "RMS" detectors for a state variable oscillator's sin and cos outputs and summing their unfiltered log-domain detectors similar to stereo "True Power Summing." This would provide "four phase" rectification. The output in the 3 mV/dB log domain would represent |sin|² + |cos|².
For those interested an example of the AD536 RMS detector used in the SSL 4000/6000 82E10 channel dynamics processor can be found here: http://www.ka-electronics.com/images/SSL/ssl_82E10.pdf
Last edited:
Jan.
I tried using that sigma delta rms detector from LT but to get the ripple down the output filter tc was so long it wasn't any better than anything else. Plus the narrow output range, it just wasn't worth it in the end.
I tried using that sigma delta rms detector from LT but to get the ripple down the output filter tc was so long it wasn't any better than anything else. Plus the narrow output range, it just wasn't worth it in the end.
Since the signal in a sine wave oscillator is essentially a perfect sine wave catching the peak is all that's really required. For Jan's autoranger a peak detector may be more relevant since the RMS detector will ignore the crest factor and may settle with a low duty cycle over-range signal.
At those prices a sample & hold + track and hold seems reasonable despite its complexity. This guy http://www.ti.com/lit/ds/symlink/opa615.pdf for around $7 seems to be well suited to the peak hold requirement, or this http://www.ti.com/lit/ds/symlink/lf398-n.pdf for less than $2 could do the task.
Triggering on a cosine's zero crossing should be simple. These solutions do get really fast settling for a system. The Boonton 1120 gets 10 Hz to 140 KHz with no changes in the AGC loop.
The concern for noise/glitches getting into the signal are valid but addressable.
At those prices a sample & hold + track and hold seems reasonable despite its complexity. This guy http://www.ti.com/lit/ds/symlink/opa615.pdf for around $7 seems to be well suited to the peak hold requirement, or this http://www.ti.com/lit/ds/symlink/lf398-n.pdf for less than $2 could do the task.
Triggering on a cosine's zero crossing should be simple. These solutions do get really fast settling for a system. The Boonton 1120 gets 10 Hz to 140 KHz with no changes in the AGC loop.
The concern for noise/glitches getting into the signal are valid but addressable.
This oscillator used the LTC1968 RMS detector.
"Low-Distortion Sine Wave Oscillator with Precise RMS Amplitude Stability," Cheng-Wei Pei, Linear Technology Magazine, March 2005. http://www.linear.com/docs/7461
"Low-Distortion Sine Wave Oscillator with Precise RMS Amplitude Stability," Cheng-Wei Pei, Linear Technology Magazine, March 2005. http://www.linear.com/docs/7461
The LM13700 datasheet (figure 43) shows a simple Sample and hold made from half an LM13700: http://www.ti.com/lit/ds/symlink/lm13700.pdf
Just about every circuit block needed to make a tuneable SV oscillator is in that datasheet.
Just about every circuit block needed to make a tuneable SV oscillator is in that datasheet.
As Demian points out the fact is peak, average, and rms for a pure sine wave are only an easily computed scale factor apart. 0.003% is -90dB, I don't remember who discovered that a standard 120V Christmas tree bulb works very well as a level control. I got -110dB with nothing else.
A light bulb with a really massive filament to get a longer time constant would work better. Especially one with a vacuum instead of argon.
GR used thermistors in some of their battery powered oscillators. They may be worth a second look 40 years later.
GR used thermistors in some of their battery powered oscillators. They may be worth a second look 40 years later.
Someone has already looked into Thermistors in detail: Amplitude stability and distortion in thermistor-controlled oscillators - IEEE Xplore Document Probably not linear enough.
Yeah. I said the same thing back here: http://www.diyaudio.com/forums/equi...n-audio-range-oscillator-669.html#post5087544
This one has possibility.
https://www.google.ca/search?q=opa355&ie=utf-8&oe=utf-8&gws_rd=cr&ei=kIIkWfquH8TPjwS6tYb4Aw
No it was published in EDN a few years ago, the Mazda lamp or other small bulb only does -90dB or so.