Someday when I get the chance I might look at the range-switch transients in my agc loop. The nonlinearity only costs 2 IN4148 diodes and a resistor.
BTW, once the oscillator has been in an oscillating state, the shock excitation you speak of is probably not necessary if the range switching is fast enough so that the oscillation does not die out too much during the switching interval. Make-before-break switching would mitigate this.
BTW2, settling time us usually a non-issue except on the lowest frequency range, as long as the agc time constants are switched to be appropriate for the chosen frequency range, as mine were.
Cheers,
Bob
For all that you want you might want to consider a digital solution.
It should be fairly easy to get THD down to below 0.001%
In the manual it says that's for the 60 Hz oscillator. The others are pretty quick. All 4 operate constantly. I'm not expecting it to be magical but it should be interesting. The packaging/graphics etc. do not connote the most premium of construction and I am not happy about RCA connectors etc. but I'm sure its a cost issue.
Victor has the same issue and doesn't offer low frequency oscillators because of the settling time issue. A SH/TH solution would address that issue but twice as complex as the basic oscillator.
When I get it I'll do a full test and expose on it. Its still a value for 4 oscillators in a box.
And another sidetrack- HP built for a short while a version of the 200 CD that had a single sweep ffrom 20Hz to 20 KHz, the HP 207 http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1957-01.pdf It used an interesting RC network to spread the tuning. I don't know if its applicable but it could be interesting.
Drive a tuned state variable filter with the digital source to reduce the harmonics from the -120 dB possible to almost -140 dB? Not really difficult and makes frequency accuracy a non-issue.
The component cost is trivial, I am mainly concerned that the system is not Linear Time Invariant anymore so all the theory is inapplicable.
Of course in practice it won't make much difference, it just bothers the ex-mathematics student in me.
On a more practical note, if the diodes and resistor were removed then would it not be possible to move the function of IC7 onto IC6 and remove an op-amp too? (Reference numbers as in the copy of "Audio" in your PDF, Part 1 p.39)
Yes. I meant it only for initial start-up.
Actually, when I looked at the schematic of your Distortion Analyser oscillator (same reference as above.) I noticed a connection from the power supply to the ALC capacitors. (on switch S2C, bottom corner).
If this is to precondition those capacitors to the correct potential then I don't understand, isn't the power off in position 1?
Yes I plan to do this too, seems an obviously sensible idea.
Best wishes
David
Last edited:
That's about 20 dB worse than I want.
Viktor's oscillators are much better than that and pretty simple.
A SVO with a sweep optimized ALC loop should not be much more complex.
Much simpler than a DDS module followed by a swept SV filter.
I am not completely averse to the "d" word.
A simple di*ital read out makes a lot of sense and there's a few ready made as panel meters.
Anyone know of a reasonable frequency readout module?
Best wishes
David
Hi David,
The diodes are conducting only during the transient interval when the error is initally large, so they do not compromise the analysis when the error is small. If you care about a complete analysis from initial startup when the transient error may be large, then these diodes will indeed make life more nonlinear. SPICE simulation may be very useful here to provide insight.
The primary purpose of S2C is to kill the oscillation when the output attenuator is in the off position. IIRC, the voltage divider formed creates a charge on the filter capacitors that is not overly far off from the nominal operating value.
Cheers,
Bob
OK, I had a half formed idea about that, thank you for the clarification.
One more question if I may, it concerns a difference between your oscillator and Bruce Hofer's Tek 505, so may interest others too.
The 505 has gain in the inverter, your gain is essentially 0 dB.
As far as I can see, the benefit of gain is that it allows the ALC feedback to be passive and eliminate an op-amp compared to your unit.
The variability of level around the loop probably means a minor trade-off - the optimal level can't be used everywhere but the "cost" is quite small.
Is there some other reason for inverter gain?
Best wishes
David
For what its worth you can reconfigure (open a connection) and the SVO is a filter. Feed your source into the agc node. The output is the second integrator which will connect to the AGC for leveling (not that its necessary). Bob could probably explain this better.
The current generation of high performance DAC's can get to -120 dB THD. The high Q of the SVF should get another 20+ dB?
The current generation of high performance DAC's can get to -120 dB THD. The high Q of the SVF should get another 20+ dB?
My superannuated mind may have confused Bob's presentation with somebody else's . . . but . . . I think he actually explained his oscillator's (theoretical) operation by starting with a state-variable filter (SVF), then closing the loop to make it oscillate.
With ideal amplifiers I'd expect a SVF to suppress harmonics by much more than 20 dB. But the high-Q that makes this possible means the filter bandwidth is quite narrow. This, and the practical tolerances of passive components, means that the filter must have some degree of tunability to get the filter's notch aligned with the digitally-generated signal's frequency - plus a control loop to detect misalignment and drive the tuning mechanism.
And, being an active filter, the SVF will add some of its own THD. The magnitude of this corruption may be small, but, as mentioned previously in this thread, when you get to the PPM distortion range EVERYTHING has a distortion effect.
This architecture (a digital generator followed by an analog clean-up filter) seems to have merit but I fear the details of practical implementation are as complex as a pure analog solution.
Dale
So add some damping to widen the bandwidth a bit to make the tuning easier.
It will bring the Q down and make the input signal amplitude more reasonable.
Without the damping the filter is likely to beat with offset input frequency.
When I experimented with this I had reduce the input signal to 3mV rms into the the SVF.
At frequencies above 10kHzand un damped SVF will oscillate spontaneously. There is a frequency between 10kHz and 15 kHz where the loop can be opened and the oscillator keeps on going at a stable level without control. Somewhere between the said frequencies the multiplier is applying damping only from there on up. The loop is no longer positive feedback it's negative.
Surely it won't be as complicated but more so?
A SV filter is, as Bob pointed out, essentially equivalent to an oscillator.
So you have all the complexity of a SV oscillator plus an additional DDSynthesizer on the front end plus extra care to ensure DDS hash doesn't leak thru.
Personally, I'd rather just add a counter and read-out.
To repeat my earlier question, anyone have any recommendations for a counter/read out module?
Presumably this is due to extra phase shift in the op-amps.
I recall that Glen Kleinschmidt had the reverse problem, he used "better" op-amps in an attempt to improve a functional oscillator and the reduced phase shift was sufficient to stop the oscillation.
Turned out the unit needed a little bit of extra phase shift.
Used to be on his web-site, not sure it's still there.
So what op-amps and oscillator type are your circa 10 kHz numbers from?
I plan to trim the feedback values at each of my switched frequencies so I can decouple the ALC loop more heavily and reduce distortion.
Seems I may have to make the trim bilateral, which I had not considered.
Best wishes
David
Last edited:
Hi David,
When I had designed my oscillator I had heard (and been inspired by) Bruce's SG505 AES paper, but was unaware of much of his specific circuitry, including his use of gain in the inverter. Saving an op amp was not of interest to me, and the noise/distortion of the op amp I used for the FET agc circuit is not much of a problem. I was more interested in providing a good way to get the desired feedback from drain to gate and driving the dc on the gate using the bridge approach I came up with. Properly implemented, both approaches are probably capable of delivering comparable performance.
A key issue in any FET agc design is how much signal is put across the JFET (setting agc distortion) versus added noise in the oscillator loop if the signal across the JFET is made small, all while maintaining adequate agc authority.
This is why it is important to keep track of the noise in various oscillator designs. FFT-based spectrum analysis of oscillator distortion can cover up oscillator noise due to the averaging inherent to the process. This is a case where a good old analog spectrum analyzer can reveal something better. It is too easy to get the oscillator distortion due to the agc element down while inadvertantly increasing oscillator noise.
Use of precision R-C tuning elements tends to reduce needed agc authority and allow smaller signal levels on the JFET agc element.
Cheers,
Bob
you mentioned the ability to sweep. this seems to me that this would make things very complex. that is why i suggested the digital solution.
Hi Dave,
I mentioned earlier we have to sort out what is theory from personal opinion.
Glen made it clear that this was his idea of the cause of the problem. He asserted that the quadrature relationship of the SVF is never quite exactly 90 degrees and that the added phase shift from lower GBP op amps cured this problem by adding enough phase shift to correct the difference.
Imbalance in phase or time constant of the integrator cause amplitude error, unequal amplitude, between the section outputs. It does not change the quadrature relationship.
If it did the the filter would not function at all. If the amplifiers are all the same the added phase shift is distributed equally through out the ring and will have little effect on amplitude and frequency.
There is a well documented phenomena call Q enhancement. Q enhancement is cause from finite bandwidth of the amplifier. The ring becomes unstable because of loop gain reduction. This causes the Q to increase eventually becoming infinite. From what I've been able to measure it's an exponential curve. The gain at resonance because so high it's outside my measurement range. The filter spontaneously oscillates at resonance. In closed loop as an oscillator the loop gain becomes so high it exceeds the authority of the multiplier and can't be controlled. The oscillator goes into hard clipping.
The right way to deal with the Q enhancement problem is to increase the bandwidth of the amplifiers so the Q remain flat over a wider bandwidth. Another way is to to severely limit the bandwidth using low GBP op amps. With this approach the gain fall off before the Q enhancement gets out of control. This seems like a backward approach to solving the problem. If greater bandwidth op amps are not available then the effect can be compensated for by adding a very small capacitor 3-7pF across the resistor that closes the ring from the LP output to the input of the HP sections. The capacitor add a phase lead to the ring compensating for the
Hi Dave,
I mentioned earlier we have to sort out what is theory from personal opinion.
Glen made it clear that this was his idea of the cause of the problem. He asserted that the quadrature relationship of the SVF is never quite exactly 90 degrees and that the added phase shift from lower GBP op amps cured this problem by adding enough phase shift to correct the difference.
Imbalance in phase or time constant of the integrator cause amplitude error, unequal amplitude, between the section outputs. It does not change the quadrature relationship.
If it did the the filter would not function at all. If the amplifiers are all the same the added phase shift is distributed equally through out the ring and will have little effect on amplitude and frequency.
There is a well documented phenomena call Q enhancement. Q enhancement is cause from finite bandwidth of the amplifier. The ring becomes unstable because of loop gain reduction. This causes the Q to increase eventually becoming infinite. From what I've been able to measure it's an exponential curve. The gain at resonance because so high it's outside my measurement range. The filter spontaneously oscillates at resonance. In closed loop as an oscillator the loop gain becomes so high it exceeds the authority of the multiplier and can't be controlled. The oscillator goes into hard clipping.
The right way to deal with the Q enhancement problem is to increase the bandwidth of the amplifiers so the Q remain flat over a wider bandwidth. Another way is to to severely limit the bandwidth using low GBP op amps. With this approach the gain fall off before the Q enhancement gets out of control. This seems like a backward approach to solving the problem. If greater bandwidth op amps are not available then the effect can be compensated for by adding a very small capacitor 3-7pF across the resistor that closes the ring. This compensate the lagging phase shift that causes the Q enhancement.
I know Bruce encountered the Q enhancement problem because the compensation technique was used in the Sys One oscillator and I believe the Tek 505.
Here is quote from Electronic Filter design Handbook, Arthur B. Williams and Fred J. Taylor.
"Another serious limitation occurs because of finite amplifier bandwidth. Thomas (see
Bibliography) has shown that, as the resonant frequency increases for a fixed design Q, the
actual Q remains constant over a broad band and then begins to increase, eventually becoming infinite (oscillatory). This effect is called Q enhancement.
The Q-enhancement effect can be minimized by having a high gain-bandwidth product.
If the amplifier requires external frequency compensation, the compensation can be made
lighter than the recommended values. The state-variable circuit is well suited for light compensation since the structure contains two integrators which have a stabilizing effect.
A solution suggested by Thomas is to introduce a leading phase component in the feedback
loop which compensates for the lagging phase caused by finite amplifier bandwidth.
Thomas is the engineer that realized the State Variable and Biquad filters. A more in depth study is available from his original works, if you can find it.
"I recall that Glen Kleinschmidt had the reverse problem, he used "better" op-amps in an attempt to improve a functional oscillator and the reduced phase shift was sufficient to stop the oscillation.
Turned out the unit needed a little bit of extra phase shift."
No, this is wrong.
I guess Glen just didn't know about the Q enhancement effect.
As far as the the feedback ratio is concerned for the ring, if you depart from unity and add gain to the input it causes an imbalance in the ring and the amplitude at the BP and LP output will be unbalanced. Try it in spice. Double the gain and see what happens.
By the way the Q enhancement effect doesn't show up in spice like it does with a real model. No help there.
Last edited:
Here is something for everyone to ponder.
The gain at resonance of an SVF, SVO is some 60dB or more depending on the amplifiers used.
I've measure this in an undamped SVF. It correspond to an SVF in spice.
As you all know the bandwidth of an op amp depends on the closed loop gain of the amplifier according to GBP. So what is the bandwidth of an undamped SVF if the gain at resonance is 60dB and GBP is 24MHz?
The gain at resonance of an SVF, SVO is some 60dB or more depending on the amplifiers used.
I've measure this in an undamped SVF. It correspond to an SVF in spice.
As you all know the bandwidth of an op amp depends on the closed loop gain of the amplifier according to GBP. So what is the bandwidth of an undamped SVF if the gain at resonance is 60dB and GBP is 24MHz?
I could have sworn I put in a tweak for Q enhancement on the 200kHz range of my THD analyzer, but I just went back and looked at the old schematic and could not find it. My foggy mind thought it remembered putting a small resistor in series with the tuning capacitors on that range. The design used 5534 ICs in the loop. If I did not use such a tweak, it may have worked OK because the agc in both the SVO and analyzer SVF had adequate control over the Q.
I worked in the same department as Lee Thomas and Jimmy Tow at Bell Labs in Holmdel. Both brilliant guys. Lee was a Widlar-like character with a wry sense of humor. He subsequently led the effort on the MAC32, the first 32-bit CMOS microprocessor chip. At about the same time, out of that same department came the first DSP chip.
Cheers,
Bob
Hi Bob,
Did everything good come out Bell Labs?
Q enhancement seems to get a bit worse with higher bandwidth op amps like the 1468.
Higher BW but maybe not high enough to combat the problem. I did find if I use a op amp with higher bandwidth than the 1468 in the high pass section I can go with out lead compensation. With the Q enhancement under control I find I can use a lot more decoupling between the multiplier and SVO. The control voltage stays in a tighter range and of course the distortion and noise is lower with the added decoupling.
I though you put resistors at the inputs of both integrators. I'll have to look again.
Yes you have 220 ohm resistors at the integrator inputs before the cap and tuning resistor.
Is that for a different reason?
Do you mean like this Widlar?
http://clifford.soup.io/post/419436041/digital-every-idiot-can-count-to-one
Last edited: