Hello,
Very nice and generous proposal Jackinnj !
For this time, the opto-resistor used is relatively easy to find, i have seen that it is sale by Newark/Farnell for some euros.
I'm a little disappointed about the August EDN article.
i awaited it for some weeks and finaly this new oscillator is very good but not as that described in AN67 app note.
I don't think to build it, or when i will have nothing other to do.
At the moment, i work on my PSU project, sorry EUVL !
Regards.
Frex
Very nice and generous proposal Jackinnj !
For this time, the opto-resistor used is relatively easy to find, i have seen that it is sale by Newark/Farnell for some euros.
I'm a little disappointed about the August EDN article.
i awaited it for some weeks and finaly this new oscillator is very good but not as that described in AN67 app note.
I don't think to build it, or when i will have nothing other to do.
At the moment, i work on my PSU project, sorry EUVL !
Regards.
Frex
The LDR is actually cheaper at RS than Farnell.
But it is very easy to get and the price difference is not really relevant, compared to the rest.
So I am still looking for a since oscillator at below -140dB.
The one from Audio Precision System One is also not much better than -110dB.
Maybe Scott would kindly drop us some hints ??
Patrick
But it is very easy to get and the price difference is not really relevant, compared to the rest.
So I am still looking for a since oscillator at below -140dB.
The one from Audio Precision System One is also not much better than -110dB.
Maybe Scott would kindly drop us some hints ??
Patrick
I find it interesting that this is a fixed frequency design, though it might work well across a wide band.
But for me, when it comes to distortion products, my HP 239A actually does about as well or better -- but perhaps has more broadband noise. Using a spectrum analyzer, the 239 has a largest individual component of around -120dB, which is decent. So I guess I'll pass on building this one, and maybe try a 239 circuit using the optoisolator instead of the JFET for AGC.
But for me, when it comes to distortion products, my HP 239A actually does about as well or better -- but perhaps has more broadband noise. Using a spectrum analyzer, the 239 has a largest individual component of around -120dB, which is decent. So I guess I'll pass on building this one, and maybe try a 239 circuit using the optoisolator instead of the JFET for AGC.
Keeping the (THD + N) performance while making it tunable across a wide range may be difficult. There is a passive lopass (130 ohms/0.47 uF) right on the output terminal, with a corner frequency just slightly above (2.6 KHz) the 2 KHz fundamental frequency. My guess is, it was put there especially to control wideband noise. Then, the ALC loop is used to counteract the effects of having the fundamental frequency on the filter skirt.
To keep this architecture in a tunable oscillator you would have to switch the output capacitor as you changed tuning ranges. You could probably maintain the (THD + N) performance while tuning across an octave (2:1 range), or maybe a half decade (3.2:1), but I doubt that it could be used with decade (10:1) tuning ranges.
I find it curious that the optoisolator-based ALC is a step or two "backwards" in technology. The AN67 oscillator used Linear Technology's LT1228 operational transconductance amplifier (OTA) for the ALC control element, and JFET's seem to have been the preferred control element for designs from the 1980's and 1990's.
By my mental calculations, Williams kept the signal voltage to only about 5 mV pk-pk across the optocoupler. And there isn't a very wide control range - hence the need for the 100 ohm pot to set the optocoupler's control current to roughly 10 mA (and its signal resistance around 80 ohms). This operating point should probably be maintained if you transplant the ALC circuit into a different design.
Dale
Last edited:
Well, having a tunable low-pass filter switch Tau with frequency and range is extra switching, but is not hard to do. I think the magic, if there is any, is exactly what you point out Dale, that the signal voltage on the control element is extremely low, and this seems to hold regardless of the actual control element -- photocoupler or JFET.
I wonder what the actual performance of this circuit really is -- I'm betting it's a lot better than indicated in the article....
I wonder what the actual performance of this circuit really is -- I'm betting it's a lot better than indicated in the article....
Here is what I don't get: Why?
Article says Linear needed it to test stuff. Cool. But the AP has a lower floor, so they used it to test JW's design. Whaaaat? Why not just use the AP equipment? What's the point of a fixed oscillator that's not as good as the AP variable one? Linear can't afford an AP tester?
Article says Linear needed it to test stuff. Cool. But the AP has a lower floor, so they used it to test JW's design. Whaaaat? Why not just use the AP equipment? What's the point of a fixed oscillator that's not as good as the AP variable one? Linear can't afford an AP tester?
I want to mention something I haven't seen much if anything about in these threads using LED/CdS Photocell (LDR) devices. The light output of LEDs drops over time. Of course all components age, but LEDs appear to change their characteristics more drastically than most others. I recall a figure of about 50 percent reduction in light output at full current over six months. This oscillator may not be in operation over that time (unless someone leaves it powered up on the bench continuously), but I think it's something to be aware of. The adjustment of the 5V value across the 470 ohm resistor in series with the LED would have to be increased as the LED ages to maintain the original spec. I'm on a seismometer list where the idea of a LED and phototransistor are used with a vane in between to detect the position of a seismometer arm, and a small tungsten-filament light bulb is suggested instead of an LED as a long-term stable light source.
It might also be "important" in the threads on LED/LDR attentuators used for volume control, especially as these LEDs are run for long periods of time listening, but that solution is simply to change the knob setting.
It might also be "important" in the threads on LED/LDR attentuators used for volume control, especially as these LEDs are run for long periods of time listening, but that solution is simply to change the knob setting.
Of course a better method for a seismometer is to have a photo receptor monitor the led and keep the output constant.
The led is barely on for this circuit and it has it's level constantly adjusted by the output level. So it is in a feedback circuit.
Samuel Groner covers a lot here, Low Ripple/Fast Settling AGC For Oscillator.
He is worried a lot about fast settling, if you just want low distortion Jim's circuit built with composite super op-amps or discrete ones with better than -140dB THD and VERY carefully chosen passives (ask Ed
) should be able to do it. The key is to make the controlling non-linear element have as little as possible to do. BTW I don't think this is really the article promised, since the output amp could never do -160dB (or any that I could imagine).
I would give this a shot. Take a very good sound card and sum the channels with one attenuated by 1000. Program your fundamental into one channel and subtractive harmonics in the other until you get nulls at the harmonics. Verifying with a very good twin-tee (AP has a white paper on that).
Tedious yes, but if you just want to test something at a fundamental and pick out the first few harmonics this should work. Remember there is no magic trick to make an ordinary IC op-amp suddenly have -160dB THD.
Last edited:
I will mention a few of the interesting design items as I see it.
The resistors are a low distortion type I have not used. In the past JW has used vishays and other "Magic" parts.
He is quite specific about the capacitor types and I think the actual ones used are very important.
The output buffer has feedback to roll it off around 2K as does the passive filter on the output.
The Photocell is running around 80 ohms in use. This means any distortion contribution will be reduced by 50 db by the fixed resistors. It also means the gain trim is extremely tight.
Interestingly I think the noise floor might be lowered a bit by using a better filter on the 5 Volt offset reference, but I am not familiar with the particular part.
I would have to run distortion measurements on the Photocell to see what the actual limits of the circuit could be. That would be my guess as to the most critical part.
It seems to be chosen not just for response linearity but also low noise. I suspect a VCA across a resistor might yield better results.
The resistors are a low distortion type I have not used. In the past JW has used vishays and other "Magic" parts.
He is quite specific about the capacitor types and I think the actual ones used are very important.
The output buffer has feedback to roll it off around 2K as does the passive filter on the output.
The Photocell is running around 80 ohms in use. This means any distortion contribution will be reduced by 50 db by the fixed resistors. It also means the gain trim is extremely tight.
Interestingly I think the noise floor might be lowered a bit by using a better filter on the 5 Volt offset reference, but I am not familiar with the particular part.
I would have to run distortion measurements on the Photocell to see what the actual limits of the circuit could be. That would be my guess as to the most critical part.
It seems to be chosen not just for response linearity but also low noise. I suspect a VCA across a resistor might yield better results.
Last edited:
Reading between-the-lines in the schematic "Notes", I see the message "Attention to detail is important here.".
When I saw that I wondered if the Jung/Didden/Salas work on voltage regulators might point to a lower-noise design for the voltage reference.
If the moderators don't rap my knuckles for transplanting my comments from the thread, "R2R Attenuator for Volume Control without Transition Contacts in the Signal Path":
The atch image is Fig 7 (THD+Noise vs. signal level for a NSL-32SR2 configured as a 3 db shunt attenuator with a 2 KOhm source resistance) from the Silonex page at < Silonex Inc.: Products: Audiohm Optocouplers: Audio Characteristics >.
From the description on the Silonex web site, it sounds like the 32SR2's photocell was connected as the shunt element in a simple voltage divider (L-pad) with a 2 kilohm series element. Then the LED current was adjusted until the voltage divider produced a 3-dB loss. (I think this means the photocell resistance was around 4.8 kilohms.)
Quickly grabbing two data points, I see 0.010% (-80 dB) THD with -6 dBu input level; and 0.10% (-60 dB) THD with +4.5 dBu input. That's a 20 dB increase in THD for a 10.5 dB increase in input level. Thinking makes my brain hurt, so I will approximate it as a 2:1 ratio. If I recall correctly, that points to a predominantly cubic (third-order) nonlinearity in the photoresistor's V-I curve at this particular illumination level.
If I re-learned MATLAB I could probably determine the coefficients for a polynomial V-I curve, but maybe somebody who already knows MATLAB will do it.
There is more information about using the Silonex optocouplers as audio attenuators on the Silonex page called "Audio level control with resistive optocouplers". Scroll down to the discussion on the "Shunt Attenuator".
They say that a 3-dB shunt attenuator is a worst-case situation, distortion-wise. I don't know if they REALLY mean "a 3-dB attenuator will give the worst THD, regardless of the series resistance in the attenuator"; or if they mean "the optocoupler has its worst distortion when when the photocell is near the upper end of its usable resistance range - which is where it's operated in this example of a 3-dB attenuator". I'd like to see curves for THD versus attenuation for various values of series resistance - e.g., 2K ohms, 600 ohms, 200 ohms, etc.
DaleFrom the description on the Silonex web site, it sounds like the 32SR2's photocell was connected as the shunt element in a simple voltage divider (L-pad) with a 2 kilohm series element. Then the LED current was adjusted until the voltage divider produced a 3-dB loss. (I think this means the photocell resistance was around 4.8 kilohms.)
Quickly grabbing two data points, I see 0.010% (-80 dB) THD with -6 dBu input level; and 0.10% (-60 dB) THD with +4.5 dBu input. That's a 20 dB increase in THD for a 10.5 dB increase in input level. Thinking makes my brain hurt, so I will approximate it as a 2:1 ratio. If I recall correctly, that points to a predominantly cubic (third-order) nonlinearity in the photoresistor's V-I curve at this particular illumination level.
If I re-learned MATLAB I could probably determine the coefficients for a polynomial V-I curve, but maybe somebody who already knows MATLAB will do it.
There is more information about using the Silonex optocouplers as audio attenuators on the Silonex page called "Audio level control with resistive optocouplers". Scroll down to the discussion on the "Shunt Attenuator".
They say that a 3-dB shunt attenuator is a worst-case situation, distortion-wise. I don't know if they REALLY mean "a 3-dB attenuator will give the worst THD, regardless of the series resistance in the attenuator"; or if they mean "the optocoupler has its worst distortion when when the photocell is near the upper end of its usable resistance range - which is where it's operated in this example of a 3-dB attenuator". I'd like to see curves for THD versus attenuation for various values of series resistance - e.g., 2K ohms, 600 ohms, 200 ohms, etc.
Attachments
Yes, from the perspective of distortion the important parameter is the operating point of the CdS photoresistive element in the optocoupler. As the LED ages, the integrator in the ALC control loop (A7, an LT1006) will need to slew to a greater voltage to establish the steady-state operating point. This will affect settling time from a cold-start, but I don't think it'll change the distortion performance after the output level is in regulation.
Dale
Scott,
The signal passes through the gain control element in the Wien bridge version used here. It is followed by an inverting amplifier and then another with a 2K LPF to the output filter.
It is not just the precision of the unity gain that adjusts the distortion it is also affected by the parts choice. That is why the only common 1% MF resistors are used in the level detector path.
The gain adjustment range is .3%! That is why the resistors inside the loop must be .1%.
The signal passes through the gain control element in the Wien bridge version used here. It is followed by an inverting amplifier and then another with a 2K LPF to the output filter.
It is not just the precision of the unity gain that adjusts the distortion it is also affected by the parts choice. That is why the only common 1% MF resistors are used in the level detector path.
The gain adjustment range is .3%! That is why the resistors inside the loop must be .1%.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.