OK. That sounds like a good plan.
A super ultra low distortion gen of variable freq -- with or without the buffer/amp for driving low Z is still very helpful tool.
-Richard
Last edited:
One possibility is to gain step an amplifier with stepped compensation optimized for a particular gain. It's a dedicated amplifier and doesn't need to be compensated like a general purpose op amp.
I'm going to send it down as is. It will run equally as well with a +/- 10Vdc to +/- 15vdc supply. The on board shunts will run cooler with a lower supply voltage.
Output is 2.5Vrms. The shunts do 5Vdc and +/-5Vdc for the level controller and logic ic's.
The ADC in the controller takes 5Vdc and is configure for a +/-5Vpp input. Since the Multiplier is another ADC and Mdac configured as a four quadrant multiplier I was able to do a dual control loop. The inner loop acts directly with the multiplier ADC to form a pure proportional control. This loop stabilizes the oscillator but does not level it. The outer control loop is a PI controller which sets the level. The dual loop relieves the PI controller from having to stabilize the oscillator. It can operate in a narrower range and is optimized for it's purpose.
You will find common mode distortion here. I've searched without luck for an op-amp based Q multiplier completely free of the op-amp's distortions. The calibrated gains at the harmonics is very easy to tabulate and it does work remarkably well.
Scott do you know any voltage controlled gain circuits that operate with negative feedback.
Any attempt of mine to rap trans linear cells with feedback causes lose of the gain control.
The trans linear stuff is really quite high in distortion for what I need it for.
Does the use of positive feedback with a negative feedback amplifier negate the negative feedback? A Howland current pump would be an example.
Any attempt of mine to rap trans linear cells with feedback causes lose of the gain control.
The trans linear stuff is really quite high in distortion for what I need it for.
Does the use of positive feedback with a negative feedback amplifier negate the negative feedback? A Howland current pump would be an example.
I'm speculating that a buffer from the midpoint of the two resistors driving an isolated supply for the opamp may be a work-around. Or Bill Whitlock's "Engenius" trick might work.
The differential notch is not going to be an easy problem. I'm still surprised at how sharp and deep the notch is without Q enhancement. Conceivably a transformer across the bridge would work. Since the fundamental is attenuated so much the transformers residual distortions would not degrade the measurement. For a power amp the components could be scaled to much lower impedances.
The differential notch is not going to be an easy problem. I'm still surprised at how sharp and deep the notch is without Q enhancement. Conceivably a transformer across the bridge would work. Since the fundamental is attenuated so much the transformers residual distortions would not degrade the measurement. For a power amp the components could be scaled to much lower impedances.
The Dyna PAS 3 used positive feedback inside the negative feedback loop to increase the forward gain. The feedback still controlled the frequency response and gain. I'm sure there are consequences. . .
4-quadrant multiplying DAC have been used as gain control inside a feedback loop - but the basic deal is that the excess loop gain only causes more accurate tracking of the feedback divider - if that function is at all nonlinear, time varying or noisy you are stuck with the problems - just have a very accurate version of them at the amp output
our most accurate "multipliers" today are ADC/DAC - if the best don't meet your need then I doubt you can find better circuit techniques - otherwise the tech would be used in ADC/DACs
our most accurate "multipliers" today are ADC/DAC - if the best don't meet your need then I doubt you can find better circuit techniques - otherwise the tech would be used in ADC/DACs
The traditional solution is to not have a post amp. Make the output capable of driving a low impedance (50 Ohms?) with no degradation and use a passive attenuator. Use a variable reference voltage for the +/-1 dB range and a decade attenuator for the rest.
That's not as easy as it looks at first, and I don't recall having seen a prior art design without explicit output driver stage. If you use a 10 dB-per-step output attenuator, you'd need to vary the oscillator amplitude over a 10 dB range; this is not very practical as it causes inconvenient noise/distortion trade-offs in the oscillator circuitry (the oscillator is, generally speaking, noisier than a well-design broadband amplifier stage, so should be operated at a fixed, as-high-as-possible, level). Without further precautions it also alters the loop gain of the leveling loop significantly (which might provoke stability problems).
I'm running the oscillator at +16 dBu. The output driver stage consists of a first amplifier with fixed +4 dB gain. This is followed by a second stage with variable gain from -5.9 dB to +4 dB. A balanced output is provided by inverting the output of this second stage. At the end, a 50/100 Ohm (unbalanced/balanced) 10 dB-per-step output attenuator is provided for an overall balanced output voltage of below -80 dBu to +30 dBu.
With careful design, it is possible to achieve a THD+N (22 kHz BW) of better than -120 dB using this scheme (unless at low output voltages the thermal noise of the output attenuator takes over). One of the most daunting problems is the power coefficient effect of the resistors in the feedback networks and output attenuators. The for low noise required low resistor values cause high dissipation and thus distortion from power coefficient. Despite I'm using e.g. over 100 series/parallel resistors for the most critical stage in the 10 dB attenuator section, this appears to be the dominant distortion mechanism at 10 Hz.
Samuel
Last edited:
One consequence is that you must be quite sure that at overload or clipping, the pfb drops out before the nfb.
If it is the other way around you end up with a very high gain in the forward path and no feedback, which generally results in the output sticking to a rail.
Jan
The Krohn-Hite 4000 series and the Shibasoku oscillators have no output amp. The Shibasoku uses the reference voltage to adjust the output +/- 1 dB. The KH has a pot in series with the attenuator. I think this was used in some other KH low distortion oscillators as well. The current KH4403 does have an output buffer. All the above are discrete designs with higher rails and output drive.
Adding something like the LME49600 to the opamp in the oscillator may provide enough drive.
Adding something like the LME49600 to the opamp in the oscillator may provide enough drive.
I see how this can be made to work with a 600 Ohm output and modest max. output level--I was talking for a low-Z (50 Ohm or lower for unbalanced) and +30 dBuish (balanced) output, which for my needs is the way to go.
Samuel
I haven't seen a State Variable the doesn't use an output amp. They typically have to run at a lower output level to obtain a low level distortion. No more than +10dB for mine to get the harmonics below -140 b/t 1k and 10KHz. Victor's oscillator I think has a max of 2.67Vrms IIRC and would require a post amp for higher levels. If +10dB is all that's needed then no post amplifier is required.
So we are comparing apples to apples, B Cordel's uses a post amp. The gen in the Sys One
uses a post amp and IIRC so does the Tek SG. All SVO's. IIRC the KH uses a bridged T which like the HP339A operates at a high output level of about 20Vpp.The 339a has a buffer amp.
600 ohms is a very old standard. It seemed to gradually disappear with the onset of op amps and as discrete circuitry was replaced by IC's in the 70s and early 80s. It's still used in studio and broadcasting but not like it was.
I think 50 ohms would be more useful. 31.6Vrms is a lot and I don't know if I'm going there but I suppose it would be useful for bringing out distortions in passive components.
I pointed out to Dick Moore the 339a distortion will rise as the attenuator is loaded even though the buffer amp sees a constant 600 ohm load. I haven't confirmed the cause or isolated the cause to the attenuator network. A Best guess would be as Samuel has pointed out the power coefficient of the resistors. It is quite a significant rise in distortion.
This design has not been to simply copy designs or methodology of the past but more of a study on what it takes to get low distortion in an oscillator while applying SOTA technology. And of course to have a useful piece of test gear in the end. I had this in mind while doing the study on the lamp multiplier. There are no front panel controls. It operates entirely from PC UI control.
Samuel how are you going about determining R values for minimizing the effect of power coefficient of the resistors. Obviously physical size of the resistor must play a part. Would it be useful to mount resistors over a copper plane to spread the heat or is it this strictly an instantaneous power effect?
I tried adjusting the reference level in the 339a to study the effect of level vs distortion with a bridged T. The oscillator became unstable very quickly. It would require serious design changes to get it to work at different levels. The bridged T's seems to work better at a maximum level. The SVO's do better at lower levels. As Samuel pointed out it's difficult to change an oscillator's level over a wide range with out also changing the controller loop gain. You can get a way with small changes to reference level but a large change is at the cost of loop stability and settling time. For an SVO the loop gain relation is -1 / pie*Vp. If the gain is too high the control loop will oscillate at a low frequency and if it's too low then settling time is compromised. A variable frequency SVO requires a four quadrant multiplier. At high frequencies somewhere between 10kHz and 20KHz the phase of the multiplier must reverse 180 deg. A Bridged T, Wien Bridge or the like does not require a 4Q multiplier. A voltage controlled attenuator is all that's really needed.
The frequency response of the effect typically falls at 10 dB/decade for the thin film resistors I've tested. This appears to be a result of thermal diffusion, which is a complex process associated with various (actually distributed) time constants. I doubt that layout effects/heatsinking would give a significant benefit at audio frequencies--10 Hz is probably not much related to the case-ambient interaction (but I don't know for sure either as I haven't tried).
For MiniMELF resistors, I've derived a maximum permissible drive level (don't have the figure at hand) for a given distortion level at 10 Hz by testing several dozen specimen from different batches. This then (together with a maximum voltage limit to keep voltage coefficient effects negligible) leads to the required number of parallel/series resistors. If a voltage divider uses equal resistor values in its arms, there is a tendency to distortion cancellation (as long as they come from the same batch and have similar tempco) so fewer resistors might be used.
Larger resistors have lower distortion, but I prefer to use more small resistors as they tend to be more easily sourced, and then there's the effect of statistical averaging (both for distortion and resistance value).
Samuel
Drive is the easy part. I need it at a gain of about 2 to 3 with 0 distortion and noise.
Why do the emoticons never work for me?
The Krohn Hite 4000 series are state variable oscillators, dating from the '60s and '70s that had .001% distortion. They used 25V supplies and some pretty sophisticated discrete circuitry. The dominant distortion on them was from the AGC's sampling pulse.
Moving to opamps shifted everything- the typical voltage seems to be 3V RMS internally and then you add an output amp. That output amp needs to be exceptional as does the pot between it and the oscillator. FWIW I have measured the distortion of a number of pots using the CLT-1 (two of them) and the Alps and Nobel pots were way better than the Alpha pots (all 10K). So were the Bourns pots I tested: http://www.bourns.com/data/global/pdfs/Slimline.pdf
I would like a differential output amp with 50 Ohm drive for this stuff as well. Can it be done without compromising the really low distortion and noise of these oscillators?
Another possibility-This may or may not be useful- 10V drive into 600 Ohms- minimum loss matching pad 16-20 dB loss. The level is now below 1V but has a low source impedance.
Moving to opamps shifted everything- the typical voltage seems to be 3V RMS internally and then you add an output amp. That output amp needs to be exceptional as does the pot between it and the oscillator. FWIW I have measured the distortion of a number of pots using the CLT-1 (two of them) and the Alps and Nobel pots were way better than the Alpha pots (all 10K). So were the Bourns pots I tested: http://www.bourns.com/data/global/pdfs/Slimline.pdf
I would like a differential output amp with 50 Ohm drive for this stuff as well. Can it be done without compromising the really low distortion and noise of these oscillators?
Another possibility-This may or may not be useful- 10V drive into 600 Ohms- minimum loss matching pad 16-20 dB loss. The level is now below 1V but has a low source impedance.
What I have in mind in to make things as modular as possible. Direct access to the osc output. Buffered output with and with out attenuator, balanced and unbalance outputs at 50 ohms. Also outputs for quadrature access which would be useful for driving external control and syncing.
I think a composite approach with an op amp and discrete output might get us there or a parallel IC arrangement.
It might be useful to marry the control of an analyzer or variable notch to the quadrature signals such that it's locked to the oscillator. If the oscillator drifts a bit the controlled device changes with it. It might eliminate some of auto control required with independent equipment. I have to give this more thought.