Low-distortion Audio-range Oscillator

[scott wurcer]

Originally Posted by Samuel Groner

The pudding is about making a low distortion oscillator with reasonable settling time (whatever this may be for a particular application--just saying that at 10 Hz, 1000 cycles are awfully long). Also see my rather long post back in this thread about the THAT1281-based multiplier.

Samuel

Yes, in fact the C7 lamp oscillator might be very low THD for a lamp based one but at 4kHz it takes 10's of thousands of cycles to stabilize.

[RNMarsh]

Originally Posted by Samuel Groner

I think I've commented on this design before--I would not say that "this oscillator has -145 dB distortion", I'd say that "the authors believe that this oscillator has -145 dB distortion". Their measurement setup is very questionable (they have rheostat-connected trimmers in the direct signal path of the notch filter, which is a substantial invitation for distortion at these levels), and they have not cross-validated their result with a different measurement setup. There is simply no establish method to reliably measure distortion below the -130 dB level, so any claims at and below this level must be looked at with highest scepticism.



It doesn't prove anything--because the distortion contribution from the multiplier element is not just a function of its inherent distortion performance, but equally as much of its decoupling. If you can stand very low multiplier authority (i.e. a lot of decoupling, and resulting long settling time), its quite easy to make a low distortion oscillator even with a mediocre multiplier (and level detector). The pudding is about making a low distortion oscillator with reasonable settling time (whatever this may be for a particular application--just saying that at 10 Hz, 1000 cycles are awfully long). Also see my rather long post back in this thread about the THAT1281-based multiplier.

Samuel

I agree with all of this... and yes, the circuit was seen this spring but forgotten about as we were try to just get the THD down on the 339A's circuitry which is cheap and plentyful for DIY'ers. Not to reinvent the wheel. But now all the easy stuff has been done. The result is well below -100db over a wide range of freq.

In that context, and the direction we have been going, it is time to re-evaluate the direction by using the example given.

As for the time constant --- me, personally, I dont care how long it takes to settle down. Ten minutes is OK. 100 seconds is no problem.... We dont need it for (and the 339A wasnt made for) fast assembly line testing. Nice if it wasnt that long, though. I'm going to do that freq thd test once in a blue moon. However, the sine wave thd and individual harmonics is a more important priority to be as low as possible - is what is needed to be able to even measure what we can design as amplifiers. For a budget price, no less.

-130 ... -145 ... -160 and more even are seen on FFT plots from SIM's. That's why test equipment is needed to verify. Reality checks. If the distortion is below what can be measured with accuracy and precision and repeatabliity, then thats just about the right amount..... Thx-RNMarsh

[davada]

Originally Posted by Samuel Groner

I think I've commented on this design before--I would not say that "this oscillator has -145 dB distortion", I'd say that "the authors believe that this oscillator has -145 dB distortion". Their measurement setup is very questionable (they have rheostat-connected trimmers in the direct signal path of the notch filter, which is a substantial invitation for distortion at these levels), and they have not cross-validated their result with a different measurement setup. There is simply no establish method to reliably measure distortion below the -130 dB level, so any claims at and below this level must be looked at with highest scepticism.



It doesn't prove anything--because the distortion contribution from the multiplier element is not just a function of its inherent distortion performance, but equally as much of its decoupling. If you can stand very low multiplier authority (i.e. a lot of decoupling, and resulting long settling time), its quite easy to make a low distortion oscillator even with a mediocre multiplier (and level detector). The pudding is about making a low distortion oscillator with reasonable settling time (whatever this may be for a particular application--just saying that at 10 Hz, 1000 cycles are awfully long). Also see my rather long post back in this thread about the THAT1281-based multiplier.

Samuel

Is this -130dB THD or just -130dB for any harmonic?

Something that raises my eyebrow is the possibility of harmonic cancellation with in the signal chain most notably from inconsistent measurement. The 339a does this for the same input level seen by the input amplifier but with different attenuation between the analyzer input and oscillator output. I can understand the noise changing but the distortion??? The only real change is the input amplifier is seeing a different source resistance. The analyzer input resistance is constant and the output resistance of the oscillator is constant as well.

[Samuel Groner]

Is this -130 dB THD or just -130 dB for any harmonic?

Just a rough order of magnitude--usually only 2nd and 3rd are significant, so THD is anyway not much higher than the individual harmonics.

Something that raises my eyebrow is the possibility of harmonic cancellation with in the signal chain.

That's exactly the point--the distortion contribution of the notch filter does (with a 50:50 chance) either add or subtract to/from the oscillator, depending on the phase relation of the harmonics (if the phase relation is exactly +90 or -90, the magnitude will remain unaltered--but this case is extremely unlikely). So we don't even know if the notch filter increases or decreases the observed distortion!

Samuel

[dimitri]

So we don't even know if the notch filter increases or decreases the observed distortion!

I think that insertion of pi(T) attenuator between UUT and notch filter will help

[davada]

Originally Posted by dimitri

I think that insertion of pi(T) attenuator between UUT and notch filter will help

Hi Dimitri,

Why would that help?

[dimitri]

Hi David,

As I understand the distortion originated in the notch filter and(or) in the load resistor is the main concern. Assume that the filter is fully passive and we are using the same best components (R and C) as in oscillator feedback network(s). With 6-10dB attenuation distortion which may be generated in the notch filter will be reduced considerably. If there will be small difference between 6 and 10dB measurements one can say that this is oscillator residual.

One can ask - what about distortion is attenuator resistors? I would replace one resistor with parallel resistors in series

[davada]

Originally Posted by dimitri

Hi David,

As I understand the distortion originated in the notch filter and(or) in the load resistor is the main concern. Assume that the filter is fully passive and we are using the same best components (R and C) as in oscillator feedback network(s). With 6-10dB attenuation distortion which may be generated in the notch filter will be reduced considerably. If there will be small difference between 6 and 10dB measurements one can say that this is oscillator residual.

Well yes we can see if the notch filter is poking it nose in with distortion by testing this way
but there is also the spectrum analyzer or FFT. There is usually a buffer before an ADC and then there is the ADC residual as well. It is difficult to know exactly which is the contributing factor.

[dimitri]

You are correct. AD797 stages before notch filter don't allow me to measure reliable <-125dB with AP2. After rejection of fundamental amplifier before an ADC and ADC itself aren't very important, at least there is always the option to use different amplifier/ADC and compare results.

[davada]

Originally Posted by dimitri

You are correct. AD797 stages before notch filter don't allow me to measure reliable <-125dB with AP2. After rejection of fundamental amplifier before an ADC and ADC itself aren't very important, at least there is always the option to use different amplifier/ADC and compare results.

This true for the buffer but we are assuming the ADC behaves the same way. Different animal.