In a warmed-up oscillator, I found a drift of 0.2Hz during a five minute period:
Low-distortion Audio-range Oscillator
Regards,
Braca
Low-distortion Audio-range Oscillator
Regards,
Braca
drift
After a 1 hour warmup, I still get a drift of some 30ppm/minute (drift notch filter minus drift signal generator).
Cheers,
E.
Hi Pavel,
After a 1 hour warmup, I still get a drift of some 30ppm/minute (drift notch filter minus drift signal generator).
Cheers,
E.
Last edited:
What about locking the 48kHz oscillator of the Soundcard to the external oscillator? Just thinking out loud...
I derived an expression for attenuation offered by a single-section BPF at the 2nd and 3rd harmonic and am attaching results below.
I can also suggest a concept for a relatively (IMHO) simple tracking BPF if there's interest.
Tunable BPF ?
I apologize for teasing an idea and then being slow with follow-up.
I’ve attached a design recipe for the classic inverting active band-pass filter. These equations are widely available but those below are from “Operational Amplifiers, Design and Applications”, Graeme, Toby, and Huelsman ISBN 07-064917-0
IMHO and assuming I haven’t made errors in analysis, this simple circuit is astonishingly well suited to this application: it’s non-inverting so that it’s not susceptible to common mode distortion; it’s offers surprisingly easy tuning of center frequency without perturbing gain; and it provides a conveniently available 90 degree reference signal useful for phase detection and thus automatic frequency tracking.
Please note that the equations indicate center frequency can be tuned by adjusting R2, but R2 does not appear in the equation for gain. Wow!
The node voltage at R2 provides a signal that provides 90 degrees shift relative to the filter output. I learned this from studying a Khron-Hite schematic. It turns out the R2 node has a 2nd order low-pass function from the input. (Surprised me.) Center-frequency gain from this node to the output is 2*Q squared.
The final schematic shows a few possible implementation details. The component values were calculated using the recipe in the attachment. I’ve assumed the circuit is intended to filter a source that already has reasonably low distortion, so I’ve taken nominal gain =1, Q=10 for about 24dB suppression of 2nd harmonic, and 1kHz center frequency. I’ve assumed 22nF for readily available low distortion caps. Impedances are relatively high, so the FET opamps that Victor favors are probably good candidates. A non-inverting buffer at the R2 node might be perfectly fine but, to preclude any common-mode load impedance nonlinearities on the R2 node, I’ve assumed an inverting amp to provide the quadrature output. I’ve depicted the LDR (or FET) and fixed resistors; their nominal combined resistance is indicated.
Disclaimers and caveats--- this is a paper design, no bench testing or simulation.
I apologize for teasing an idea and then being slow with follow-up.
I’ve attached a design recipe for the classic inverting active band-pass filter. These equations are widely available but those below are from “Operational Amplifiers, Design and Applications”, Graeme, Toby, and Huelsman ISBN 07-064917-0
IMHO and assuming I haven’t made errors in analysis, this simple circuit is astonishingly well suited to this application: it’s non-inverting so that it’s not susceptible to common mode distortion; it’s offers surprisingly easy tuning of center frequency without perturbing gain; and it provides a conveniently available 90 degree reference signal useful for phase detection and thus automatic frequency tracking.
Please note that the equations indicate center frequency can be tuned by adjusting R2, but R2 does not appear in the equation for gain. Wow!
The node voltage at R2 provides a signal that provides 90 degrees shift relative to the filter output. I learned this from studying a Khron-Hite schematic. It turns out the R2 node has a 2nd order low-pass function from the input. (Surprised me.) Center-frequency gain from this node to the output is 2*Q squared.
The final schematic shows a few possible implementation details. The component values were calculated using the recipe in the attachment. I’ve assumed the circuit is intended to filter a source that already has reasonably low distortion, so I’ve taken nominal gain =1, Q=10 for about 24dB suppression of 2nd harmonic, and 1kHz center frequency. I’ve assumed 22nF for readily available low distortion caps. Impedances are relatively high, so the FET opamps that Victor favors are probably good candidates. A non-inverting buffer at the R2 node might be perfectly fine but, to preclude any common-mode load impedance nonlinearities on the R2 node, I’ve assumed an inverting amp to provide the quadrature output. I’ve depicted the LDR (or FET) and fixed resistors; their nominal combined resistance is indicated.
Disclaimers and caveats--- this is a paper design, no bench testing or simulation.
True, as I believe gain at center frequency is quite high. Maybe that's an Achilles heel. 
I believe gain at center frequency is 2*Q^2 With Q=10, that's G=200 or 46dB. But the gain at 2nd harmonic would be lowered by about 24dB, so closed loop gain at 2H would be about 22dB. More favorable feedback, but still worrisome. Maybe a composite amp would offer higher open-loop gain at 2H. Maybe the "Bandpass 2" just showing in the BPF3.pdf would help.
Thanks KSTR
I believe gain at center frequency is 2*Q^2 With Q=10, that's G=200 or 46dB. But the gain at 2nd harmonic would be lowered by about 24dB, so closed loop gain at 2H would be about 22dB. More favorable feedback, but still worrisome. Maybe a composite amp would offer higher open-loop gain at 2H. Maybe the "Bandpass 2" just showing in the BPF3.pdf would help.
Thanks KSTR
Jan, Why can't you tie it to a 10MHz signal like the time nuts folks do?
That is supposed to be stable. Then, can't you synchronize the FFT and the
filter, essentially locking their time signatures together? Much like the
principle of a tuning fork, that is two hearts beating as one. When finished,
they can lay spent in each others wires. How beautiful.
BSST, (BPF4) okay so what amps would you suggest? 1468 since I have them.
Don't know what to use, start there.
I've got a victors oscillator on the bench and working.
I see the in, the out, wondering where to place the 90 degree node?
And what about resistor values feeding the 90 degree opamp?
Or, are you saying 364 ohms for that combination, in parallel drawn, but one goes to the a trimmer,
the other to 90 degrees feedback with opamp?
Even thought I don't have working FFT right now, I could track the 2nd or 3rd harmonic
the Shibasoku's meter. Probably not. Never mind. Victors is already too low for me to look
at harmonics....I'm already pushing the measurements with out using a notch.
I'll have to shift that B&K over to plug that in and get that last -10dB outta the way.
LKA, thanks for the tip. lowering the output voltage for power amp.
Cheers,
That is supposed to be stable. Then, can't you synchronize the FFT and the
filter, essentially locking their time signatures together? Much like the
principle of a tuning fork, that is two hearts beating as one. When finished,
they can lay spent in each others wires. How beautiful.
BSST, (BPF4) okay so what amps would you suggest? 1468 since I have them.
Don't know what to use, start there.
I've got a victors oscillator on the bench and working.
I see the in, the out, wondering where to place the 90 degree node?
And what about resistor values feeding the 90 degree opamp?
Or, are you saying 364 ohms for that combination, in parallel drawn, but one goes to the a trimmer,
the other to 90 degrees feedback with opamp?
Even thought I don't have working FFT right now, I could track the 2nd or 3rd harmonic
the Shibasoku's meter. Probably not. Never mind. Victors is already too low for me to look
at harmonics....I'm already pushing the measurements with out using a notch.
I'll have to shift that B&K over to plug that in and get that last -10dB outta the way.
LKA, thanks for the tip. lowering the output voltage for power amp.
Cheers,
Last edited:
Yes, actually that is what I am checking out at the moment. It probably can use the same auto-track circuit I developed for the tracking notch. What has to be seen is the noise and distortion performance at the band-pass frequency, what I have read so far it doesn't seem promising, but maybe I do a prototype to be sure.
Jan
Jan
Not sure I get your drift. I don't need to sync the AP digital oscillator to anything, it is very stable and derives from the same clock as the FFT, but it isn't clean enough. The Viktor oscillator IS clean enough but slightly drifts over long FFT acquisitions.
So the choice seems:
- use the AP digital oscillator, cleaned up with a bandpass, or
- use the Viktor, with some sort of injection locking it to a freq standard (or to the AP digital oscillator).
Jan
Hi Sync,
Jan's post is spot on I think.
The sine wave source needs to be synchronized to the FFT analyzer clock to resolve the "test-at-bin-center" challenge. Synchronization could be via phase-locking an analog oscillator or by using using an analyzer that inherently provides digital syncing like Jan has. Or perhaps (as Damien suggests) a sound card that synthesizes the test signal from the ADC clock. A tracking BPF then cleans the imperfect source; Jan's tracking notch then comes to bear in suppressing the DUT fundamental to extend measurement dynamic range.
OPA1656 opamps seem to be good candidates as Victor demonstrates that they are capable. Others may be OK as well.
You've interpreted correctly re the unspecified resistor values. 10k resistors could drive the 90 degree buffer stage. Jan mentioned in an earlier post that he achieved good distortion when tuning range was +/- 0.3%. This implies cell resistance range should be about +/- 0.6% of the nominal 364 ohms and the fixed series resistor provides the rest.
The 90 degree output and the filter output provide the inputs to a phase detector very similar to that used in Bob Cordell's analyzer. It's followed by an integrator that drives the LED optocoupler.
Best,
Steve
Last edited:
Hi Jan,
When I was studying Bainter filters for my book, I seem to recall having successfully converted a Bainter notch into a nice bandpass in simulation. I'll try and look back and see how I did it. I'm guessing it was a simple arrangement with a Bainter notch in the feedback path of an op amp gain section, where a minimum amount of flat feedback prevented the gain from heading toward infinity at the notch center frequency.
Cheers,
Bob
Yes indeed Bob, my feeble attempts seemed to show gain going through the roof. Would be interested in your suggestion.
Jan
Let me restate the issue which may help
The current concept it a bandpass (or low pass) filter on the output of a digital source to address the drift issue and to ensure the source is centered on a band.
This really doesn't need tuning since the frequency is very stable. It does need to be really low distortion to improve on the already excellent performance of the best DAC chips.
The current concept it a bandpass (or low pass) filter on the output of a digital source to address the drift issue and to ensure the source is centered on a band.
This really doesn't need tuning since the frequency is very stable. It does need to be really low distortion to improve on the already excellent performance of the best DAC chips.