Since KSTR pointed out worrisome noise gain in my earlier circuit, I’ve taken another run at the BPF paper design, using “Bandpass 2” as inspiration. ” As before, attachments are from “Operational Amplifiers, Design and Applications”, Graeme, Toby, and Huelsman ISBN 07-064917-0
The attached revised design uses the same inverting BPF filter but with values adjusted to set Q=1 and Ho =1; then positive feedback raises the resulting Q to 10 or higher. To emphasize similarities with Victor’s oscillator design, I adjusted the capacitor values to imply NPO caps, but revised to 20nF to allow lower impedances. It potentially uses the same NPO caps and opamps Victor uses. Additionally, positive feedback makes the filter operate just short of self-oscillation.
The filter stage topologies are different but performance is similar. At center frequency, my design has Q=1, G=1, noise gain =2; Victor’s has Q=0.5, G=1, noise gain =2. At harmonic frequencies, the noise gains differ slightly but both are low. So there is reason to hope the filter’s residual distortion might be comparable to Victor’s results.
A schematic side note suggests a pot that allows adjustment of feedback Q. Adjustable Q may be insightful and at the extreme setting, the filter should oscillate. This may be beneficial in initial trimming of center frequency.
I’ve included the transfer function and noted the multiplied Q and gain as primed variables. Component sensitivities in the attachments should be applicable. Closing comments in the attachments claim Q of 50 is practical. Gain will grow rapidly with Q' but input resistor R9 can be made larger to reduce gain.
Again, my usual caveats: this is a paper design, neither tested nor simulated.
The attached revised design uses the same inverting BPF filter but with values adjusted to set Q=1 and Ho =1; then positive feedback raises the resulting Q to 10 or higher. To emphasize similarities with Victor’s oscillator design, I adjusted the capacitor values to imply NPO caps, but revised to 20nF to allow lower impedances. It potentially uses the same NPO caps and opamps Victor uses. Additionally, positive feedback makes the filter operate just short of self-oscillation.
The filter stage topologies are different but performance is similar. At center frequency, my design has Q=1, G=1, noise gain =2; Victor’s has Q=0.5, G=1, noise gain =2. At harmonic frequencies, the noise gains differ slightly but both are low. So there is reason to hope the filter’s residual distortion might be comparable to Victor’s results.
A schematic side note suggests a pot that allows adjustment of feedback Q. Adjustable Q may be insightful and at the extreme setting, the filter should oscillate. This may be beneficial in initial trimming of center frequency.
I’ve included the transfer function and noted the multiplied Q and gain as primed variables. Component sensitivities in the attachments should be applicable. Closing comments in the attachments claim Q of 50 is practical. Gain will grow rapidly with Q' but input resistor R9 can be made larger to reduce gain.
Again, my usual caveats: this is a paper design, neither tested nor simulated.
I agree completely. The only incentive for the design I posted is the availability of the 90 degree output that would be supportive of adding autotune tracking.
A Victor oscillator and a like filter would have very similar drift behavior
A filter and an oscillator slightly drifting together would still not solve the issue of long synchronous FFTs unless you can derive the FFT clock from the oscillator (and then still I think it won't work).
The idea for the filter was to use the AP or soundcard digital oscillator which is locked to the FFT, but has relatively high distortion.
Basically trying to mate two unlikely parties: low distortion of an analog oscillator and frequency stability of a digital generator.
But the drift of the filter when used on the digital oscillator does not impact the FFT, that can still be locked to the oscillator. The drift of the filter would only impact the harmonic attenuation and that drift may be small enough to make it still work.
I am waiting delivery of another Viktor oscillator for this purpose, but mail transit times seem unusually long. (And mail from EU to oversees has stopped completely unless you pay € 50 'premium').
Jan
The idea for the filter was to use the AP or soundcard digital oscillator which is locked to the FFT, but has relatively high distortion.
Basically trying to mate two unlikely parties: low distortion of an analog oscillator and frequency stability of a digital generator.
But the drift of the filter when used on the digital oscillator does not impact the FFT, that can still be locked to the oscillator. The drift of the filter would only impact the harmonic attenuation and that drift may be small enough to make it still work.
I am waiting delivery of another Viktor oscillator for this purpose, but mail transit times seem unusually long. (And mail from EU to oversees has stopped completely unless you pay € 50 'premium').
Jan
Interesting idea, I will test that when I get the Viktor.
BTW Any news on the resistor distortion front?
Jan
I don't think that this is the ultimate solution. The oscillator still must limit,
be it free running or helped by the injection. If you get the usual output
voltage, you get the usual limiting.
With normal filters, you can stay away far from any limiting. And the
individual filter sections should be low-Q, since the Q inflates the
internal AC voltages inside the filter.
An oscillator is close to the resonance, artificially large signal included.
Far-off from the resonance, you get 6 dB/octave per pole, not more.
2 or 3 times the resonance/corner frequency is always far-off, so there
is not much of a reward to compensate for the large signal drawback.
Cheers, Gerhard
Maybe I need more coffee to formulate that.
be it free running or helped by the injection. If you get the usual output
voltage, you get the usual limiting.
With normal filters, you can stay away far from any limiting. And the
individual filter sections should be low-Q, since the Q inflates the
internal AC voltages inside the filter.
An oscillator is close to the resonance, artificially large signal included.
Far-off from the resonance, you get 6 dB/octave per pole, not more.
2 or 3 times the resonance/corner frequency is always far-off, so there
is not much of a reward to compensate for the large signal drawback.
Cheers, Gerhard
Maybe I need more coffee to formulate that.
I think this is what Billam und Bonello have analyzed in their respective papers on the distortion in active filters.
Got a brainwave, looks like it works.
See attached. I did a loop back from the AP dig. generator, via the 1kHz passive notch. That shows the generator distortion only, red trace, and you can see that the 3rd and 2nd are about -113dBr and -122dBr, respectively. Confirms reasonably to the specs.
Then I applied my idea to reduce the 3rd. This is the blue curve. 3rd is gone, or at least lower than -150dBr! I am not sure why the 2nd and the other high-order products are also reduced, most probably because they were the result of the analyzer A-D converter distortion excited by the high 3rd. The original 3rd was about -60dBr wrt to the notched fundamental; in the blue graph it is much lower.
Reducing the 3rd increased the autoranging gain and decreased the A-D dynamic range requirement.
Anyway, this was with dig generator at 48kHz clock, the analyzer A-D at 96kHz, FFT SR 32kHz (highest available), 4Msamples FFT length, 4 averages.
It all comes at a price: total analysis times about 20 mins per run.
Nevertheless, happy camper!
Jan
See attached. I did a loop back from the AP dig. generator, via the 1kHz passive notch. That shows the generator distortion only, red trace, and you can see that the 3rd and 2nd are about -113dBr and -122dBr, respectively. Confirms reasonably to the specs.
Then I applied my idea to reduce the 3rd. This is the blue curve. 3rd is gone, or at least lower than -150dBr! I am not sure why the 2nd and the other high-order products are also reduced, most probably because they were the result of the analyzer A-D converter distortion excited by the high 3rd. The original 3rd was about -60dBr wrt to the notched fundamental; in the blue graph it is much lower.
Reducing the 3rd increased the autoranging gain and decreased the A-D dynamic range requirement.
Anyway, this was with dig generator at 48kHz clock, the analyzer A-D at 96kHz, FFT SR 32kHz (highest available), 4Msamples FFT length, 4 averages.
It all comes at a price: total analysis times about 20 mins per run.
Nevertheless, happy camper!
Jan
Last edited:
I may be mistaken or misinterpreting, but for purposes of filtering, I don't think these issues pertain to a state variable filter in the same way. You will indeed get gain in proportion to Q with an SVF bandpass, but not objectionably high "internal" signal levels. Things are pretty much "out in the open" in an SVF. At resonance, for example, a reasonably high Q SVP BPF might have a gain of 10, just like an ordinary amplifier stage, but far lower gain away from resonance. If you really want just unity gain at resonance for concerns of overload, you just put an attenuator in front of it. If you really want just unity gain at resonance and overload is not a concern, you just put an attenuator after it. The very simple way that Q can be controlled in an SVF is one of its advantages.
The fact that it uses 3 op amps instead of one, as compared to a Sallen-Key, for example, does not imply that it is 3 times as complicated. Op amps are small, simple and cheap. Two duals or a quad do the whole thing. Tuning Rs and Cs are identical in value. About 3 more resistors than a Sallen-Key completes it.
Cheers,
Bob
Hi Jan,
Are you using DHL? The only guaranteed way to get them in about a weeks time.
My last order was shipped out on
Tue, Apr 14, 5:26 AM (12 days ago)
Shipping Method:Yunexpress Global Driect Economical Line
I was in no rush, we shall see when I get them. When I ship post I can avoid paying HST, where as DHL nails me for HST+ every time
Rick
Are you using DHL? The only guaranteed way to get them in about a weeks time.
My last order was shipped out on
Tue, Apr 14, 5:26 AM (12 days ago)
Shipping Method:Yunexpress Global Driect Economical Line
I was in no rush, we shall see when I get them. When I ship post I can avoid paying HST, where as DHL nails me for HST+ every time
Rick
Last edited:
Hi Jan,
This is a very good point worth looking into. I would observe, however, that the noise gain of the two integrators is just unity, and that op amp distortion degradation is usually in accordance with the noise gain.
That leaves the summer. It is not frequency dependent, and its noise gain is a little higher than 2. So when the SVF is operating at its fundamental with a gain of 10, the summer op amp is not really operating at a gain of ten insofar as its noise gain is concerned.
The gain of 10 at the fundamental is mainly due to the positive feedback of the global SVF loop reinforcing the gain. At frequencies away from the fundamental, that positive feedback begins to disappear. Also coming into the picture in a good way is that the BPF output comes after the first integrator, which attenuates harmonic distortion.
These questions could probably could be answered by a SPICE simulation using an op amp with some known distortion and furthermore looking at filter distortion with different BPF Q settings.
Cheers,
Bob
Has anyone mentioned going "the other direction" for FFT syncing? Have the ADC driven by a PLL driven by the 1kHz sine oscillator?
It seems this should be (relatively) easy. From the time-nuts list, they have 10MHz precision frequency standards running from the 1 pulse per second output of a GPs receiver.
It seems this should be (relatively) easy. From the time-nuts list, they have 10MHz precision frequency standards running from the 1 pulse per second output of a GPs receiver.