Did some eperiments on injection locking today. There's a hidden snag--the further you pull away from the natural oscillation frequency, and the higher the injection level is, the larger the authority of the multiplier needs to be. In my design I can hardly pull 0.1% without saturating the leveling loop. So it might not be the panacea it first looked like...
Samuel
Samuel
I didn't want to contradict Scott, but I have only used injection to lock a frequency never to pull it. As I mentioned I found the push-pull to not be symmetric when playing with it.
To me the benefit is to eliminate phase noise (Jitter).
However when using a wein bridge as a filter following an oscillator I aimed for 2% tracking accuracy or better.
I don't know how much this is useful to you as your design is quite different.
ES
In my experience the 4-phase peak detector is the 2nd best choice, the 1st one being T&H/S&H chain - the latter works way better, but is way more complex too.
L.
Are there any designs you know about which you can refer us to which uses this method?
Thx-RNMarsh
Rick this was all explored hundreds of posts ago. Fluke used it in one of there level standards.
OK.
next up -->
In another forum there was debate on the merits of voltage feedback and current -mode feedback. I pointed out the transfer chacteristcis were different.... I just read an article on error correction for rf circuits and there is the idea of predistortion (not new) but with the same characteristics needed that we have with VFB and CFB. They are complimentary and could thus be used/trimmed in a two stage osc/amp--> Like this:
[Might be useful for audio amp design idea as well as osc.]
Thx-RNMarsh
Last edited:
Update on lamp based multiplier SVO
I found I can lock the SVO without the integrator. The lamps in the multiplier provide long enough TC to run directly from the output of the MDAC. This chops two op amp stages lowering the noise and speeding up the settling time. The amplitude instability I mentioned a few post ago is gone or at least reduce to the uVrms. The distortion dropped further as a result and at a lower noise floor. This is a ADC/Mdac sample and infinite hold system which means no droop between samples of the peak. No ripple induced from droop.
The SVO settles from power up in under 5 sec at 20Hz and less than 1 sec at 1kHz and above. The settling time is proportional with frequency.
I found I can lock the SVO without the integrator. The lamps in the multiplier provide long enough TC to run directly from the output of the MDAC. This chops two op amp stages lowering the noise and speeding up the settling time. The amplitude instability I mentioned a few post ago is gone or at least reduce to the uVrms. The distortion dropped further as a result and at a lower noise floor. This is a ADC/Mdac sample and infinite hold system which means no droop between samples of the peak. No ripple induced from droop.
The SVO settles from power up in under 5 sec at 20Hz and less than 1 sec at 1kHz and above. The settling time is proportional with frequency.
These are relatively high levels of distortion from what we are dealing with but it's certainly worth a try. Distortion in RF has a different set of terminology. We need to understand what is meant by second and third order intercept distortion. We don't use these terms in audio. Some of these distortions are caused by multiple carriers in an RF systems and is unrelated to audio distortion sources.
RF has nothing to do with it... you missed the connection I was making ----> When i saw those curves, I thought of these curves:
This is the connection I imagined for two stage compound osc using complimentary circuit types. for pre-distortion correction/cancellation:
Thx-RNMarsh
This is the connection I imagined for two stage compound osc using complimentary circuit types. for pre-distortion correction/cancellation:
Thx-RNMarsh
Last edited:
Well.... dont keep us in suspense... show us the new results.
Thx-RNmarsh
It's hard to show settling time. I can do a screen capture and of ARTA and put it on Youtube.
The disto improvement is small but every bit counts. The noise floor at this point is more important.
I put some screen shots up later. I'm home for lunch.
I have to build a power supply to get rid of all the low frequency junk. It's not coming from the SVO.
I was running on the 339a supply but maxed the supply current and had to switch to a bench supply.
With the integrator the SVO started from clipping until settled. Without the integrator the SVO starts from 0V and grows from there. The over shoot is minimal.
It's much more graceful this way.
Last edited:
In a way this is what degeneration does. If we can use a none linear element where our de-gen resistor goes it might work.
I (somewhat) remember that part of the thread. Was a four-quadrant multiplier discussed? With two 90-degree phases for input, the output would be sin^2 + cos ^2, or the continuous amplitude (okay, actually something proportional to the square of the instantaneous amplitude), rather than just a value that's sampled at the peaks.
That need not be much of a problem, the oscillator should be set very close anyway. You could also trim the center frequency which also seems to work.
Ed - even if it's +-.001% you are still pulling it, the lock absorbs component TC's etc by forcing a fixed output frequency. If the oscillator had no frequency drift you wouldn't need to lock it. Jitter is not really an issue here just keeping the output centered on one f bin is needed.
The point is the lightbulb oscillator does have two loops a thermal loop (very slow) and the compressive non-lineartiy. His point is that if you defeat the later by making it too good the amplitude "squeegs". Jim Williams thought his work applied directly to the LTC oscillator which does have level detector, etc., the S&H and other sophisticated AGC aides might change the game (I don't know).
Scott,
Sometimes I wonder if you go out of your way to misunderstand things. Yes any injection requires "movement" of the oscillator. Pulling or pushing is used to mean to move it away from the oscillator's natural frequency. Locking it is to keep it from straying. Naturally the oscillator will wander a bit, but the center frequency is what you try to match to the injection frequency.
Now does that clear things up for you?
ES
Sometimes I wonder if you go out of your way to misunderstand things. Yes any injection requires "movement" of the oscillator. Pulling or pushing is used to mean to move it away from the oscillator's natural frequency. Locking it is to keep it from straying. Naturally the oscillator will wander a bit, but the center frequency is what you try to match to the injection frequency.
Now does that clear things up for you?
ES
Nothing was ever unclear, why do you always think people don't understand something if they have anything different to say? An oscillator drifts, when you lock it you push or pull the frequency back. I was mearly exploring the range that works. In fact .1% is 1000ppm what is the base temperature stability? At 50ppm/degree the simple injection lock could easily need to push or pull that much.
If you are building a precision oscillator a temperature drift inside the case of 5C after warm up is too much!
Now I have refrained from commenting on light bulb or thermistor based oscillators, but field use quickly shows their short comings, vibration and temperature.
It is also to good effect to make sure that the bridged parts are the same type to match tempco and other subtle parameters. (Of course somewhere someone may have missed this obvious concept.)
Now this is all common sense.
Then some of us were doing this not just to keep a stable FFT bin but to allow looking at 1/F offset noise. You know critical band issues.
ES