No, the Q is not infinite (let alone plus/minus a little as in inf +1....)
The regulation in the oscillator affects only the amplitude of the loop gain
and adjusts it to 1.000 for a round trip.
Q is -d phase / d frequency of the loop gain and often surprisingly low
(as in "resonant like a wet sand bag").
All that is needed for a fixed frequency is that the phase delay of the
feedback loop goes through 0 at this frequency and that the round trip
of the signal makes no amplitude gain or loss for constant amplitude.
I have once written this up in
< http://www.hoffmann-hochfrequenz.de/downloads/downloads.html >
in a crystal oscillator article for DUBUS ham radio magazine.
The state variable filter/osc may be lucky in that it may not feature
a large Q, but just think at a parallel tuned circuit, where the currents
in L and C are much bigger (by Q) than the little net current that flows
through the combination.
I have to think about that.
Gerhard
Last edited:
This is conceptually very interesting. But probably far more complex than building an oscillator with very low distortion directly. The cancellation would need to be dynamic, as the distortion of an oscillator is unlikely to be sufficiently stable. To sample the distortion (in order to dynamically adjust the cancellation) we'd need a distortion analyzer, rated to/beyond the desired final oscillator performance, and tracking the oscillator frequency. Lots of efforts required!
Samuel
An update-
I have the Shibasoku 725 and the Shibasoku 590A connected together and just sitting running for a while now. I have observed the following- the distortion shifts with time or temp or ? from -120 to better than -145. for all the harmonics. Its all well below the meter but the -140 is only visible on the external spectrum analyzer.
If I connect the computer near the pair the distortion shoots up, the 2nd goes up a bunch (2-30 dB), remove computer and it drops. I'm not sure yet why. It all needs time to settle to the really low numbers.
The supplies in the oscillator are +/-30V so the 10V peak is less that 1/2 of the available rail voltages. Possibly the supplies and the level just needs to be much larger to get the residuals lower. The +/- 15 volt supplies in IC based solutions could limit the possible performance.
I have the Shibasoku 725 and the Shibasoku 590A connected together and just sitting running for a while now. I have observed the following- the distortion shifts with time or temp or ? from -120 to better than -145. for all the harmonics. Its all well below the meter but the -140 is only visible on the external spectrum analyzer.
If I connect the computer near the pair the distortion shoots up, the 2nd goes up a bunch (2-30 dB), remove computer and it drops. I'm not sure yet why. It all needs time to settle to the really low numbers.
The supplies in the oscillator are +/-30V so the 10V peak is less that 1/2 of the available rail voltages. Possibly the supplies and the level just needs to be much larger to get the residuals lower. The +/- 15 volt supplies in IC based solutions could limit the possible performance.
oscillator noise vs distortion
Here's a question whose discussion I may have missed in this thread, in which case I apologize. There is a tradeoff between agc-element-induced oscillator distortion and oscillator noise. It seems that a fair comparison of oscillators must also include some measurement of the noise.
When I long ago designed the SV oscillaor for my THD analyzer, I faced the issue of operating level of the JFET agc element. In my design, I was free to arbitrarily select the amplitude of the signal across the JFET. If I made that signal smaller, I would have to increase the injection gain of the agc result into the oscillator loop, in order to maintain a given amount of agc control range. See the oscillator schematic at CordellAudio.com - Home where the THD analyzer is described.
Given that I trim the signal gate feedback optimally to kill 2nd harmonic contributions from the JFET, the main remaining contributor is 3rd order. So, net, reducing the operating level of the JFET "always" reduces the oscillator distortion that is due to the JFET agc element nonlinearity, (even though the injection gain must be increased) but at the expense of increased oscillator noise.
So are we adequately evaluating our different oscillators by somehow documenting their noise as well as their distortion?
Cheers,
Bob
Here's a question whose discussion I may have missed in this thread, in which case I apologize. There is a tradeoff between agc-element-induced oscillator distortion and oscillator noise. It seems that a fair comparison of oscillators must also include some measurement of the noise.
When I long ago designed the SV oscillaor for my THD analyzer, I faced the issue of operating level of the JFET agc element. In my design, I was free to arbitrarily select the amplitude of the signal across the JFET. If I made that signal smaller, I would have to increase the injection gain of the agc result into the oscillator loop, in order to maintain a given amount of agc control range. See the oscillator schematic at CordellAudio.com - Home where the THD analyzer is described.
Given that I trim the signal gate feedback optimally to kill 2nd harmonic contributions from the JFET, the main remaining contributor is 3rd order. So, net, reducing the operating level of the JFET "always" reduces the oscillator distortion that is due to the JFET agc element nonlinearity, (even though the injection gain must be increased) but at the expense of increased oscillator noise.
So are we adequately evaluating our different oscillators by somehow documenting their noise as well as their distortion?
Cheers,
Bob
Some creative out of the box thinking will be needed to get a source which can cancel its own distortion. Drift in the existing designs of the null and tuning might be tracked or held at null better if the phase-locking circuitry was added/used.
The drift on the newer ShibaSohu is much lower than the old one as I dont see such levels of change as Demians old model (he owns both but I have his newer one). The ShibaSoku (new) looks like it has some digital artifacts in it... a clue to its generation, perhaps, and why it is so stable. Note: Also, the KH4402B drifts a little and heat seems to be the issue inside which affects the bipolar output devices characterists ....
See attached of 1KHz/1v spectrum --> [The thd is about the same as the modified KH4402B at 1/10 the cost]
Thx-RNMarsh
The drift on the newer ShibaSohu is much lower than the old one as I dont see such levels of change as Demians old model (he owns both but I have his newer one). The ShibaSoku (new) looks like it has some digital artifacts in it... a clue to its generation, perhaps, and why it is so stable. Note: Also, the KH4402B drifts a little and heat seems to be the issue inside which affects the bipolar output devices characterists ....
See attached of 1KHz/1v spectrum --> [The thd is about the same as the modified KH4402B at 1/10 the cost]
Thx-RNMarsh
Last edited:
yes. I include noise and distortion reduction in the process. I believe progress is made by getting away from the jFET to reg level. It improved THD byat least 6db across the board with same oscillator IC.
Thx-RNMarsh
Gerhard,
We may just be looking at this in different ways and may both be right. See this link for design of SV filters, especially SV bandpass filters, http://www.analog.com/static/imported-files/tutorials/MT-223.pdf. The Q of the bandpass filter is directly proportional to R6/R7, where R7/(R6+R7) defines the feedback ratio around the first integrator. For R6 very large, Q gets very large. In an oscillator, the agc element controls this factor. So if you take an SV oscillator arrangement and feed an input signal to it so as to use it as a BPF, you will get a BPF with extremely high Q for very large R6.
Cheers,
Bob
There may be an additional advantage to phase locked nulls and cancelling harmonis -- fast response time... not the long filter time constants to deal with. Of course, if it can be done, the application would benefit audio amplifier circuits as well..... possibly.
Anyway, getting another 10-20dB (to where the noise floor is at) using existing methods has reached a limit of sorts with the null/tuning drift issue. Solve that with existing designs and I will get interested again. Otherwise a fundamentaly new approach would produce better and stable results for hyper low distortion generation. It might be hard at first but the resolution limit will only be noise then.
Thx-RNMarsh
Anyway, getting another 10-20dB (to where the noise floor is at) using existing methods has reached a limit of sorts with the null/tuning drift issue. Solve that with existing designs and I will get interested again. Otherwise a fundamentaly new approach would produce better and stable results for hyper low distortion generation. It might be hard at first but the resolution limit will only be noise then.
Thx-RNMarsh
I think that could be done with a DDS scheme. Adding/subtracting harmonics
with a predetermined phase /amplitude to the fundamental would only
cost one or two adders for phase and a multiplier/adder for amplitude, that
should easily go into a Spartan 6 FPGA for quite a few harmonics.
The DAC could be a Maxim 16B/500MSPS; it works nicely in **** mixed signal
chip testers, and probably there is something better in the last years. Anyway,
it gives a lot of room for dithering and oversampling.
Gerhard
with a predetermined phase /amplitude to the fundamental would only
cost one or two adders for phase and a multiplier/adder for amplitude, that
should easily go into a Spartan 6 FPGA for quite a few harmonics.
The DAC could be a Maxim 16B/500MSPS; it works nicely in **** mixed signal
chip testers, and probably there is something better in the last years. Anyway,
it gives a lot of room for dithering and oversampling.
Gerhard
How do you know whether the computer is effecting the oscillator or analyzer or both?
We need a more scientific method here. How can we isolate this?
Is it a magnetic or electric field that's effecting? Or is it a static field?
If it's electric a Faraday shield will take care of it. If it magnetic some EMI shielding material will have an effect.
This is a very good point and something I was going to bring up. I'm facing the same decision even with out a jfet in the picture with the lamp multiplier. I'm not happy with the amount of noise in order to achieve the needed gain range. I have a greater restriction with the lamp R range compared to a Jfet. It's a same we must attenuate the signal so much just to hit it with a pile of gain to get the functionality of a 4Q multiplier. This gain is noise gain. We need to improve the method to gain a better SNR.
It may require greater complexity of the multiplier to satisfy both low distortion while maintaining low noise. I'm considering a dual element multiplier which operates by attenuation only on both the non inverting and inverting inputs of an op amp and leaving the amplifier gain fixed. Another possibility is a single element controlling the non inverting attenuation on a fixed gain amplifier, but this has limited range.
Last edited:
That sounds GREAT ! Offers all I could imagine in a third generation oscillator for the 21st century. What is the first step on this road to making a working prototype??
Break it down in developemnt sections to do then combining them would be a good way to start. A block diagram approach to building the sections.
Thx-RNMarsh
Last edited:
You suggested to me last year an elliptically tuned filter can theoretically completely null the 2nd H or maybe it was the 3rd H. Exotic filter tuning might be a better approach than distortion cancellation.
I know that removing the computer causes the distortion to drop. I have not dug into why, it most likely is some EMI affecting the analyzer or maybe the oscillator. it seems related to connecting the QA400 USB connection. It may even be the DC-DC converter in the interface board. Thats a chore for later.
At this level its very tedious to find. I'll explore further when I have time. My point is that these systems are very delicately balanced to get the performance.
However an active filter has distortion from the active elements and noise. A passive filter can do it. The one in my CLT-1 does a fine job. The inductors are the size of softballs made from some ferrite cup cores. Attached is the low pass filter that removes the remaining harmonics to get the 3rd harmonic below -160dB. It is capable of passing 2W undistorted. This is not tuneable however.
Attachments
"My point is that these systems are very delicately balanced to get the performance."
Indeed.
A lowPass with a sharp cutoff and null at 2H and 3H ?
None of these -- other than all passive filtering (which we can sim to see the filter response) --- offer anything but brute force approach. Thats why we need a more clever approach -- A cancellation circuit or the DDS is good as it is very stable with XtL osc. As with dithering to remove harmonics/artifacts.
More along these lines and less of the same as has been done or we'll get the same results as has been done. A new way -- not for low distortion -- for No harmonics above noise. It isnt going to happen by the means already tried.
-RNM
Last edited:
I think the idea is to tune the SVO filter to an elliptic response. Not an add-on filter.
Another very old technique for creating low distortion sine waves is the beat frequency oscillator. If the mixer is good and the two oscillators are good then the difference product should be free of anything but the target frequency. The harmonics of the two sources could all be way out of band. GR made them for years. Today it would be easy. The mixer is critical but available. Here is a primer: http://www.ietlabs.com/pdf/GR_Experimenters/1942/GenRad_Experimenter_July_1942.pdf