Low-distortion Audio-range Oscillator

[davada]

Originally Posted by RNMarsh

1. I took a relook at the LME49600... that buffer is and must be used within an opamp feedback loop. I measured it a year or more ago and the thd and it was too high for audio. It has great dc characteristics but ac is poor unless it is within a feedback loop to get low overall thd.

2. Has the EDN 11.10.94 issue for an oscillator design which has THD below 1 ppm been tried by someone we know? authored by Jeff Smith at Analog devices.

3. I only report changes that lower thd. However, with this old gear, some parts get replaced to keep it reliable. One is the 1000mfd ripple filter caps at the power supplies. [they had no affect on thd].

4. I use the LT1468 everywhere.

-RNMarsh

Sorry Rick I quoted the wrong part number. It's the LME49710HA metal can.
I did try a 1468 with the LME49600 and got exactly the same results so I didn't see the point in using the LME buffer and went back to the LME49710HA.

I had the opposite effect of a reduction of 3rd H when changing the level vernier pot to the Bourns cermet from what you found.

If we can't get consistency with these changes then they're are not worth doing.

[RNMarsh]

It is pretty consistent. However, one can trim for lowest H2 or lowest H3 or null on the thd number. I null on THD. But if you null on 1KHz you get higher thd at 10KHz.... but if you had nulled on 10Khz, the distortion is higher at 1Khz ..... Much of this is all about the tolerances of the R and C being switched.... they only used 1% parts... or 2% total deviation when switching decades or range etc. One percent is OK for the frequency but for nulling at -110db and below the trim changes enough. Right now, I dont want to select 0.1% caps and buy 0.1% R's. If I did, the freq changes will stay at lowest trimmed thd value everywhere.

Anyway, the pot change is like the other trimmers and electro cap upgrades... drops distortion some more. -Thx RNMarsh

[davada]

Originally Posted by RNMarsh

It is pretty consistent. However, one can trim for lowest H2 or lowest H3 or null on the thd number. I null on THD. But if you null on 1KHz you get higher thd at 10KHz.... but if you had nulled on 10Khz, the distortion is higher at 1Khz ..... Much of this is all about the tolerances of the R and C being switched.... they only used 1% parts... or 2% total deviation when switching decades or range etc. One percent is OK for the frequency but for nulling at -110db and below the trim changes enough. Right now, I dont want to select 0.1% caps and buy 0.1% R's. If I did, the freq changes will stay at lowest trimmed thd value everywhere.

Anyway, the pot change is like the other trimmers and electro cap upgrades... drops distortion some more. -Thx RNMarsh

Hi Rick,

The loop gain changes with frequency so the gate voltage and Jfet R has to be changed to compensate this. This is why the distortion is different at 10KHz if the oscillator is tweaked for 1KHz. If you tweak for 10KHz then it's wrong for for 1KHz. Aside from other fundamental reasons like lower open loop gain at higher frequency. No doubt the components contribute to this but not this much. All this was discussed in earlier posts.

[richiem]

RE 339A "level cal rheostat" -- sorry, I had just been looking at the 239A prints, not the 339A.

I do question the effects of F2 and especially the four protection diodes CR10-13.

[dirkwright]

Originally Posted by davada

Sorry Rick I quoted the wrong part number. It's the LME49710HA metal can.
I did try a 1468 with the LME49600 and got exactly the same results so I didn't see the point in using the LME buffer and went back to the LME49710HA.

I had the opposite effect of a reduction of 3rd H when changing the level vernier pot to the Bourns cermet from what you found.

If we can't get consistency with these changes then they're are not worth doing.

One use of the LME49600 (and BUF634) inside of the feedback loop with an otherwise ordinary opamp is to enable a designer to drive a lower impedance feedback network to reduce total noise. So, not much benefit will be evident by using the buffer in an existing circuit unless the original design was driving an exceptionally low load (ie, it was a bad design).

[RNMarsh]

Originally Posted by davada

Hi Rick,

The loop gain changes with frequency so the gate voltage and Jfet R has to be changed to compensate this. This is why the distortion is different at 10KHz if the oscillator is tweaked for 1KHz. If you tweak for 10KHz then it's wrong for for 1KHz. Aside from other fundamental reasons like lower open loop gain at higher frequency. No doubt the components contribute to this but not this much. All this was discussed in earlier posts.

I expect some relaxing of the thd spec at higher freq but this is a 100KHz instrument... dont expect to see much at 1/10th that. Phasiness perhaps affecting maintaining deepest null across the board... decades and tenths of the dials. Actually, what I was refering to behaves more like loose tolerance parts. I dont want anyone to get the idea these are big differences -- I can stay under .001% from 100-10KHz. THD+N.

Meanwhile ... today my 339A has crapped out!!!! Puts out a continuous 256Hz rather than 1Khz. Now what?!? Into trouble-shooting mode for awhile. When i get it up and running again, I can try removing some of the protection diodes etc, suggested to take a look-see affects with the ShibaSoku.

Thx-RNMarsh

[davada]

Originally Posted by richiem

RE 339A "level cal rheostat" -- sorry, I had just been looking at the 239A prints, not the 339A.

I do question the effects of F2 and especially the four protection diodes CR10-13.

Actually I seem to remember having lifted the protection diodes.

[davada]

Originally Posted by RNMarsh

I expect some relaxing of the thd spec at higher freq but this is a 100KHz instrument... dont expect to see much at 1/10th that. Phasiness perhaps affecting maintaining deepest null across the board... decades and tenths of the dials. Actually, what I was refering to behaves more like loose tolerance parts. I dont want anyone to get the idea these are big differences -- I can stay under .001% from 100-10KHz. THD+N.

Meanwhile ... today my 339A has crapped out!!!! Puts out a continuous 256Hz rather than 1Khz. Now what?!? Into trouble-shooting mode for awhile. When i get it up and running again, I can try removing some of the protection diodes etc, suggested to take a look-see affects with the ShibaSoku.

Thx-RNMarsh

I'm wondering if we could drop the distortion in the notch filter more by using a multiplier with current output in place of the phase controlling LDR. AD has one that has 50nV/rootHz voltage noise. The 633 I think would be too noisy at 0.8uV/rootHz.

[RNMarsh]

I have not studied such subjects in depth. You and others seem to be more familiar than I. So, if you think so, I will see what and where i can add to the conversion. - Thx RNMarsh

[RNMarsh]

Originally Posted by RNMarsh

I have not studied such subjects in depth. You and others seem to be more familiar than I. So, if you think so, I will see what and where i can add to the conversion. - Thx RNMarsh

BTW -- since one can trim for lowest H2 or for lowest H3.... maybe tune for lowest H2 and add a notch for the H3. reason being H3 is further away from fundemental and notch will have least affect on it.

I can measure distortion directly, without calcs and tricks -- -110dB Full Scale..... and easily -20 below that. Direct measurement to below -130. Useable to about -135dB. Then there is the FFT and notch filters to apply. I'll be able to test most oscillators. -RNMarsh