Completely different thinking about the problem and solution if tasked:
1. make an oscillator/generator with low distortion/harmonics (how low?).
2. make a signal generator with no harmonics.
Dither is a good example of thinking of not reducing the level but removal of the artifacts or appear to do so.
-Richard
1. make an oscillator/generator with low distortion/harmonics (how low?).
2. make a signal generator with no harmonics.
Dither is a good example of thinking of not reducing the level but removal of the artifacts or appear to do so.
-Richard
Last edited:
yes, harmonics of the sources would be out of band, but intermodulation products would not.
For 100 and 101 KHz we would get:
101 - 100 = 1 KHz (wanted)
101 + 101 = 202 KHz (not bad by itself, but will be further processed in the mixer) IMD2
100 + 100 = 200 KHz (same) IMD2
202 - 200 = 2 KHz and that hurts. IMD3
Mixers are large scale nonlinearities by concept.
Last edited:
Correct--you can adjust the notch frequency to an arbitrary value. Don't have the references at hand, I think it was a WW article, and there's also an AP patent. It nulls only the distortion contribution from level detector and multiplier, passives and opamps are not significantly affected. However, as level detector and multiplier are usually the more challenging things, this can be pretty helpful nonetheless if either 2nd or 3rd is strongly dominant. If bothe are ~equal in magnitude, one can tune the notch at a frequency in-between--that's what's the AP patent about IIRC.
It isn't a valid proof by saying that because I have not found a solution, the solution cannot exist.
Samuel
BTW, I tried a thermistor multiplier years ago, probably somewhat analogous to a lamp multiplier, and I recall getting some distortion from it at low audio frequencies, like 20-50Hz, due to thermal modulation.
In regard to IC 4-quadrant multipliers, my recollection was that they were noisier than the JFET approach in the end. But I must admit I have not tried IC 4-quadrant mutipliers in a long time.
Cheers,
Bob
They are still noisy.
I had the lamp multiplier down to 40Hz with out problems. I'll try it again to confirm. I might have it up and running this weekend if Fedex delivers today.
Last edited:
The technique I have used is the driven oscillator. If you build any bridge type oscillator it requires gain of exactly one to oscillate. If the gain is higher you get distortion and if lower it does not oscillate. (Duh every one here already knows that!)
So I use one where the gain is set below 1 but as close as one can get it. This does not oscillate. It now has an added signal from a driving oscillator that you can either treat as pushing it into oscillation or just treat the driven oscillator as a filter.
For my low distortion version I used dbx VCA's and level detectors (again the VCA contribution is kept very small.) For loss of gain at higher frequencies I use relay switched trimmer resistors.
The technique to keep noise down is limited bandwidth along with low impedances.
I use a simpler version of this in my power line cleaner. An isolation transformer is placed in series with a small buck boost amplifier. (After AC line filters) the reference oscillator must be tied to the AC line frequency and phase, but I don't want the noise!
I have not used my bench version of this for years. So if I can find it I will post a measurement.
So I use one where the gain is set below 1 but as close as one can get it. This does not oscillate. It now has an added signal from a driving oscillator that you can either treat as pushing it into oscillation or just treat the driven oscillator as a filter.
For my low distortion version I used dbx VCA's and level detectors (again the VCA contribution is kept very small.) For loss of gain at higher frequencies I use relay switched trimmer resistors.
The technique to keep noise down is limited bandwidth along with low impedances.
I use a simpler version of this in my power line cleaner. An isolation transformer is placed in series with a small buck boost amplifier. (After AC line filters) the reference oscillator must be tied to the AC line frequency and phase, but I don't want the noise!
I have not used my bench version of this for years. So if I can find it I will post a measurement.
The Linear Tech ULD oscillator uses the LT1228 CFB amp avec gain control.
LT1228 - 100MHz Current Feedback Amplifier with DC Gain Control - Linear Technology
The SSM parts are similar, VCA's optimized for audio use very different from general purpose 4 quadrant multipliers. That LT part specs at .2% THD at 30mV rms. I asked Jim Williams why a distortionless oscillator has a distortion trim and he laughed. He said the trim turned down the control range to the least possible amount mentioning the Bernie Oliver paper.
Jeff Smith, my tech at the time, built it and we did measure -130dB which was good at the time.
Last edited:
The only reason why I didn't use the SSM parts was back when SSM was starting I ordered a bunch of parts to play with and they never came in! dbx back then was also hard to get, but I did get a bunch and still have a few.
I was building a Speech Transmission Index (STI) meter and VCA's with log detectors were required. I built one instrument but no manufacturer wanted to produce them as they had over 1000 parts and used then top of the line 1% metal film resistors and 2% polycarb capacitors.
using a simulator I have achieved over 200db with a SVF.
So Im sure a fairly deep notch is possible.
If this is how everyone read it, then I am at fault for not describing the message well enough. I am sure a solution exists. Just not so sure it exists using the known methods so far. I used dither as an example of not fine tuning the existing but coming in with a new approach to eliminating harmonics/artifacts.
Can you do similar with a new approach?
Thx-RNMarsh
Last edited:
I see. Well, I'm pretty confident that it is possible to design oscillators with better distortion performance than presented in this thread, without resorting to fundamentally new approaches. The first step to improve a given design is to understand the dominant distortion mechanism. We have four known fundamental distortion sources--passives, opamps, multiplier and level detector ripple. To find out which one's the culprit, it is best to measure each one in isolation. Just looking at the oscillator output (as you've done it so far) is not very helpful. This applies particularly because various distortion sources may cancel. If you swap in another opamp type and observe that the oscillator output has lower distortion, this may have two reasons. First, it could be that indeed the old opamp was the dominant distortion mechanism, and the new one is better. Or it could be that the new one actually has higher distortion, and that happend to cancel better with another distortion source. There's no way to decide which one's the case if you just look at the oscillator output. But if you hit the second one and believe it's the first, you'll get tremendously fooled and run in circles forever. So looking at each of the four known fundamental distortion sources in isolation is absolutely inevitable.
So we need to know how this can be done. The passives may be checked on a suitable bridge setup, or you can replace them with series-parallel combination which lowers their distortion contribution. The opamps can be measured at high noise gain (luckily a Swiss chap has already done much of this hard work for you). The multiplier can be isolated in the oscillator and tested by feeding it with a suitable sine input and DC control voltage. Level detector ripple distortion contribution can be estimated by operating the oscillator and taking an FFT of the voltage at the multiplier control voltage port. The various frequencies can be translated into resulting harmonic distortion levels by the simple amplitude modulation equation.
That's a lot of work, but before thinking how to make it better we need to know what's the issue...
Samuel
So we need to know how this can be done. The passives may be checked on a suitable bridge setup, or you can replace them with series-parallel combination which lowers their distortion contribution. The opamps can be measured at high noise gain (luckily a Swiss chap has already done much of this hard work for you). The multiplier can be isolated in the oscillator and tested by feeding it with a suitable sine input and DC control voltage. Level detector ripple distortion contribution can be estimated by operating the oscillator and taking an FFT of the voltage at the multiplier control voltage port. The various frequencies can be translated into resulting harmonic distortion levels by the simple amplitude modulation equation.
That's a lot of work, but before thinking how to make it better we need to know what's the issue...
Samuel
Pls keep in mind that what is writtten for DIY lay people to replicate and what has been done here and over the years are not the same. Far more gets done than gets put up here.
All those areas have been covered in one way or another... thru the manufacturers products. We know there are oscillators published with super low thd and thd+n. We own some of them and have tried some of them.
Looking for ideas here that are practical, low cost and not have to do a lot of build/fab ... thus the mod approach to products which Others have already gone over all the points you have outlined in thier commecial products. How to close the gap in them vs the super low thd in other postings and publications.
So, where do You think the weak link is in these existing products/designs? Or, is it just a lot of fine tuning of existing designs... making the thd lower and lower?
I wouldnt mind a new concept/approach to try if you have one. My goal has been for multiple freqs output at below -140 THD + N or better.
Thx-RNMarsh
All those areas have been covered in one way or another... thru the manufacturers products. We know there are oscillators published with super low thd and thd+n. We own some of them and have tried some of them.
Looking for ideas here that are practical, low cost and not have to do a lot of build/fab ... thus the mod approach to products which Others have already gone over all the points you have outlined in thier commecial products. How to close the gap in them vs the super low thd in other postings and publications.
So, where do You think the weak link is in these existing products/designs? Or, is it just a lot of fine tuning of existing designs... making the thd lower and lower?
I wouldnt mind a new concept/approach to try if you have one. My goal has been for multiple freqs output at below -140 THD + N or better.
Thx-RNMarsh
Last edited:
Rectification and ripple in the AGC -- I built Bob Cordell's SVO and it has a full-wave rectifier for the AGC signal. It also has a balance pot that lets you increase the level of one half wave while decreasing the other. I've tried watching the distortion products via my active T-T filter, the EMU 0204, and ARTA. Turning the balance one way and the other had almost no affect on the distortion levels. Making those changes significantly affects the ripple level and dominant frequency, and yet it didn't much matter -- all this with 2nd H and 3rd H levels below -120dBV.
Fluke used four-phase rectifiers (0, 90, 180, 270 deg -- FW rectification of the primary and quadrature signals) in several of their lab standards in the 5700 series and I assume they had a reason. But I haven't seen any difference between 1/2-wave and full-wave rectification, using the HP 339 and the SVO as DUT examples.
I just wonder exactly how important the rectification is given some adequate method of filtering ripple -- and when in the Cordell SVO the ripple frequency was cut in half and the ripple level increased by 6dB without any consequent penalty, well, I wonder.
I know these anecdotal experiences don't carry true analytical weight, as Sam or Demian or David, among others, will quickly point out. But it leads me to think the the AGC distortion is in fact due to the control element as I think Richard said.
Fluke used four-phase rectifiers (0, 90, 180, 270 deg -- FW rectification of the primary and quadrature signals) in several of their lab standards in the 5700 series and I assume they had a reason. But I haven't seen any difference between 1/2-wave and full-wave rectification, using the HP 339 and the SVO as DUT examples.
I just wonder exactly how important the rectification is given some adequate method of filtering ripple -- and when in the Cordell SVO the ripple frequency was cut in half and the ripple level increased by 6dB without any consequent penalty, well, I wonder.
I know these anecdotal experiences don't carry true analytical weight, as Sam or Demian or David, among others, will quickly point out. But it leads me to think the the AGC distortion is in fact due to the control element as I think Richard said.
The decision to use as much of existing hardware as possible and upgrade them, add-on to them and customize them has worked well so far..... without going back to day one in the design process. Many years of experience are in those existing topologies and designs.
We are already at the -140dB/ <.00005% THD+N level using this fast-track approach. Can we go lower - into the noise?
Dick Moore; yes, that is my conclusion as well. It just doesnt matter all that much. I looked at it and wondered but nothing I did seemed to matter there.
The jFET isnt the way to go -- nor LDR - anything that isnt linear can only be minimized.... better not to have them. If you only want to go down to .001% all is fine with them.... and there would be no need to do anything more than what is available now.
Thx-RNMarsh
We are already at the -140dB/ <.00005% THD+N level using this fast-track approach. Can we go lower - into the noise?
Dick Moore; yes, that is my conclusion as well. It just doesnt matter all that much. I looked at it and wondered but nothing I did seemed to matter there.
The jFET isnt the way to go -- nor LDR - anything that isnt linear can only be minimized.... better not to have them. If you only want to go down to .001% all is fine with them.... and there would be no need to do anything more than what is available now.
Thx-RNMarsh
Last edited:
Hi Dick,
It really depends on how much filtering the ripple undergoes before reaching the element.
In Bob's design I see a lot of filtering when the oscillator is in a settled state. The use of speed up diodes, as Bob put's it, is the key to success with this AGC design. Without the speed up diodes there's probably too much filtering for the oscillator to settle in any reasonable amount of time if at all. So I'm not surprised that forcing an imbalance of the rectification had little effect.
However, all this tells us is that this is the case for an increase of ripple with the AGC's tested. What I'm interested in is the effect of less ripple. Your experiment does not draw any conclusion to this end. It is possible that other factors are masking the effect.
Level detection and AGC-
Remember one aspect, especially on a production oriented instrument like the Fluke is settling time. Those are potentially part of an automated system so waiting for the oscillators to stabilize to the target amplitude accuracy is not good. I suspect 4 pase rectification can really accelerate the settling time but may have little impact on the actual ripple on the agc element. Most of those have sample and holds with long time constants so there will not be a lot of ripple, but you do often see feedthrough of the sample pulse or related on the distortion out.
I think Samuel is right, to make progress we need to see the distortions inside the system. Something like the Shibasoku is very useful because it makes seeing very low distortions easier and identifying the relationship of the distortion at that stage to the distortion at the output.
I'll need to look up the Swiss guy's work to see if there are clues in it to distortion sources in an oscillator. At -140dB and below passives may become a roadblock. Certainly AGC's etc. are a roadblock and possibly the big one.
The benefit of the state variable oscillator I read was that it had two low pass filters to reduce the distortion. A thought would be to cascade 4 identical low pass filters, the first two are the SVO and the AGC comes from the last one so the output is stable with frequency. This would give 24 dB attenuation of the second harmonic and more for higher harmonics. Its just a duplication of the existing tuning networks.
Remember one aspect, especially on a production oriented instrument like the Fluke is settling time. Those are potentially part of an automated system so waiting for the oscillators to stabilize to the target amplitude accuracy is not good. I suspect 4 pase rectification can really accelerate the settling time but may have little impact on the actual ripple on the agc element. Most of those have sample and holds with long time constants so there will not be a lot of ripple, but you do often see feedthrough of the sample pulse or related on the distortion out.
I think Samuel is right, to make progress we need to see the distortions inside the system. Something like the Shibasoku is very useful because it makes seeing very low distortions easier and identifying the relationship of the distortion at that stage to the distortion at the output.
I'll need to look up the Swiss guy's work to see if there are clues in it to distortion sources in an oscillator. At -140dB and below passives may become a roadblock. Certainly AGC's etc. are a roadblock and possibly the big one.
The benefit of the state variable oscillator I read was that it had two low pass filters to reduce the distortion. A thought would be to cascade 4 identical low pass filters, the first two are the SVO and the AGC comes from the last one so the output is stable with frequency. This would give 24 dB attenuation of the second harmonic and more for higher harmonics. Its just a duplication of the existing tuning networks.
The model KH4402B doesnt use the jFET but injects its control into the oscillator via a multiplier - type AD633 - . Thats where I am focused right now. The output of the multiplier is via an opamp with discrete output stage. changing that opamp dropped thd and noise a noticable amount. Now I'll be looking more at and around the 633 and its I/O. But more thinking about how to inject some opposing residual as well to do some cancelling.
Thx-RNMarsh
Thx-RNMarsh