Low Distortion Oscillator output puzzle

[Restoration73]

ISTR Twin-T gives lower distortion than Wien bridge e.g.Radford LDO(transistor version)

http://www.stancurtis.com/radford.htm

[Electronpusher0]

Originally Posted by Dekatron

I did so yesterday but forgot to mention it here!

/Martin

Thanks Martin, I did not realise that Bob was still contactable, please post any replies. I am still unsure what to do on my pcb but am tending towards just using a link instead of capacitors.

Peter

[Craig Sawyers]

I emailed Bob Cordell in November last year (2020) and received a very helpful and insightful reply two days later. That was regarding a vinyl preamp that he published in Linear Audio in 2012.

Craig

[Dekatron]

I wouldn't use a link as that presents a pure DC path to the output of the op-amp via just low ohms voltage divider, and on one setting directly through 620 Ohm (R38), which I don't think is intended here.

/Martin

[Dekatron]

I got this answer from Bob!

"Hi Martin,

There is no right way to obtain a truly non-polar electrolytic function The most common way is to employ a so-called non-polar electrolytic capacitor, which has two capacitors in series and yields half the capacitance of either one. In some applications, diodes of opposing polarity will be put across an electrolytic capacitor, especially in cases where the capacitor may be exposed to high-than-rated voltages (often under fault conditions), as in the feedback shunt decoupling capacitor in a power amplifier. In the case of my THD analyzer, I used parallel 100 uF capacitors connected in reverse polarity to achieve symmetry, presumably for lower distortion, while getting twice the capacitance of each one. This was at the output of the oscillator. In this application, the amount of DC across the capacitors is negligible.

Best regards,
Bob"

to my question:

"Dear Sir,

There is currently a discussion here: https://www.vintage-radio.net/forum/...01#post1397001 on the THD Analyzer that I understand that you designed in the 70's and had published, I found your link on your webpage here: http://www.cordellaudio.com/papers/b...analyzer.shtml

That discussion also got me interested in why the two 100uF capacitors are connected in reverse to each other and how that affect their working conditions. I have seen a similar design by Philips but there they used silicon diodes either one in series with each capacitor to protect it from reverse voltage or a pair of silicon diodes in each direction across the capacitors to limit the voltage to also somewhat protect the capacitors. But I've never seen your design with the just in parallel like that without silicon diodes, so if you could explain your design on this it would be very welcome.

Best regards,

Martin Forsberg, Sweden"

If anyone has any follow up question to Bob I could include that in my answer to him.

/Martin

[Craig Sawyers]

In the HP4470B transistor noise test set, the base/gate bias is controlled by a DC feedback loop to achieve a set collector/drain current. In addition an AC sine wave pilot signal is added as a reference signal against which the amplified transistor noise is measured.

That AC signal is coupled to the device under test via a capacitor. The value of the capacitor is set by the impedance into which it is working, and a lower operating frequency of 10Hz and that is clearly different for bipolar and FET transistors.

What that leads to is 6000uF for bipolar and 25uF for FETs.

HP used two 2,900uf 3V capacitors back to back in parallel (like the Cordell oscillator being discussed. They need to be done that way because it would have needed two 12,000uF capacitors in series, which would not have fitted on the circuit board, certainly not in the late 60's when this was designed. The reason that some sort of bipolar operation is needed is to cope with bias of both NPN and PNP. Only 3V is needed because the base voltage is low. Two diodes in series, back to back in parallel limits the voltage across the capacitors to 1.4V either way. That means of course that either one or the other capacitor is biassed the wrong way round by up to 1.4V if there is a fault condition (dead transistor under test, plugged in the wrong way round, controls set incorrectly etc), but in normal operation one capacitor is biassed the wrong way round up to ~0.7V.

The 25uF for FET's was made by two 50uF 25V capacitors back to back in series. Higher voltage is needed because of the bias voltage requirements of FETs, in particular MOSFETs. And the much smaller value and physical size meant that true series back to back was possible.

Craig

[Electronpusher0]

Originally Posted by Craig Sawyers

The two capacitors in that still well-thought-of design, make a total output capacitance of 200uF (2 x 100uf in parallel). To make the same value with two in series would need 400uF, or the nearest available value.

Alternative you could buy a Nichicon 220uF 16V non-polar electrolytic from https://www.hificollective.co.uk/cat...8_322_363.html for 66p and just use one part instead of two.

Or Farnell 2326022, 220u 10V non-polar for £1.13 each.

Craig

I am sure this has been discussed before but I cannot find it, what is the difference between a non-polarised electrolytic as above and two electrolytics in series back to back. (apart from the need to double the value for each series cap)
Do non polarised electrolytics require a minimum voltage as polarised ones do?

I am now tending towards buying a non polarised one as suggested.

Peter

[dave cox]

So, for the Bob Cordell osc, 2 * 470u in series would be less prone to reverse bias damage but you might get more 2nd (and even order) harmonic ?

dc

[Craig Sawyers]

Originally Posted by dave cox

So, for the Bob Cordell osc, 2 * 470u in series would be less prone to reverse bias damage but you might get more 2nd (and even order) harmonic ?

dc

Read Bateman's measurements of electrolytic capacitors. He measures back to back series capacitors, bipolars and back to back bipolars. Link is somewhere earlier in this thread.

Craig

[daviddeakin]

Aluminium electrolytics normally need about 10% of rated voltage as a minimum DC across them in order to maintain their polarisation.

Electrolytic coupling caps do not need a polarising voltage. Just use a regular 470uF cap. Since you're building your own thing you can even measure the opamp output offset and orientate the cap accordingly, not that it actually makes any performance difference, since the offset should be mV. Heck, it's 2021, use 1000uF! And if you're still panicking, just add a switch in parallel with the cap so you can bypass it and check for any change in your results.

[daviddeakin]

Originally Posted by dave cox

So, for the Bob Cordell osc, 2 * 470u in series would be less prone to reverse bias damage but you might get more 2nd (and even order) harmonic ?

It's not a simple as that, you can alter the results by biasing the midpoint. But it is also irrelevant if you use a big enough cap.

[Electronpusher0]

I can see that I will have to do little "suck it and see" when I build the sig gen, measure the actual offset and try various capacitor configurations.
And it started as such a simple question, but that's the joy of the forum.

Peter

[Craig Sawyers]

Originally Posted by daviddeakin

It's not a simple as that, you can alter the results by biasing the midpoint. But it is also irrelevant if you use a big enough cap.

The rising distortion of that little 1u electrolytic is that with 10V input into 1k ohm is entirely to be expected. Douglas Self also measures the same effect.

The rule of thumb is: decide the lowest frequency and the load resistance. Calculate the capacitance needed to achieve that. Then multiply by the capacitance by 10. That prevents the rising distortion at the lowest frequency.

So, suppose your load is 1k, and you want to get down to 10Hz at -3dB. That gives a capacitor of 16uF. Multiply by 10 and then pick the next highest value of 220uF. Which is about what Bob Cordell has with two 100uF in parallel.

Craig

[Craig Sawyers]

Oh and if you apply a bias voltage, the distortion goes up. Lowest distortion is always with zero bias. Remember that the AC voltage across the capacitor is the square root of sod all.

As an example, take the 220uF capacitor above into 1k, and take a frequency of (say) 1kHz, and shove 10V rms into it (so 28V p-p). The impedance of that capacitor is 1.38 ohms. So the voltage across the capacitor is 28 x +/-1.38/1001.38 V = +/-39mV.

Now Cordell's oscillator produces 5V rms and the attenuator sets an output resistance of 600 ohms, presumably to be loaded by 600 ohms to be in calibration. So the oscillator raw output has to be 10V (to give 5V into 600 ohms). His lowest frequency is 20Hz. So in the worst case there will be +/-1.65V AC across the capacitors at 20Hz.

Self measures the distortion of an unbiassed 220uF electrolytic into 1k fed from 10V rms, and measures 5ppm distortion at 20Hz (0.0005%). (Small Signal Audio Design, latest edition).

So I do not think there is a problem with Cordell's design.

You might get down lower by using a non-polar, or two non-polars back to back, but just Cordell's design looks like it does just fine.

Craig

[Craig Sawyers]

I've just been looking at the Tektronix SG505 ultralow distortion oscillator (it is about 3ppm distortion). There are no capacitors in the signal path, but they take DC out by putting a servo around the output op-amp (an NE5534).

Craig

[Radio Wrangler]

Originally Posted by Craig Sawyers

I've just been looking at the Tektronix SG505 ultralow distortion oscillator (it is about 3ppm distortion). There are no capacitors in the signal path, but they take DC out by putting a servo around the output op-amp (an NE5534).

Post 6, Plan B

It's a common technique where simple coupling capacitors won't do the job

David

[Craig Sawyers]

Sorry - missed the subtlety of post 6 David ;-)

The SG505 uses a jFET to buffer the servo feedback signal, which I have not seen before. The other way is to use active servo feedback using a second op-amp. Because you can make the resistors high value (usually) around 1Meg, the capacitors are in the region of 220nF.

Craig

Craig

[Radio Wrangler]

It's a good technique where you want to use an opamp (or discrete) with great AF or RF performance, but which is lousy at DC (Like the NE5534 for example)

However, it doesn't provide DC block for DC from whatever is downstream... it will fight it to the death.

David

[Craig Sawyers]

Indeed. I've seen servos run out of range and put the luckless opamp hammering back and forth between the rails. You definitely want a low-offset input.

Craig

I ask the question, are the anti parallel capacitors are doing anything at all?