Re: Five transistor version of the low distortion sine-wave oscillator

On 15/04/2025 11:12 pm, Edward Rawde wrote:

"Bill Sloman" <bill.sloman@ieee.org> wrote in message news:vtl0bc$364bt$1@dont-email.me...

On 15/04/2025 1:56 am, Edward Rawde wrote:

"Bill Sloman" <bill.sloman@ieee.org> wrote in message news:vtihob$sfdm$1@dont-email.me...

On 12/04/2025 6:27 pm, Bill Sloman wrote:

Edward Rawde posted an eight-transistor low distortion sine wave oscillator circuit recently, and John May pointed out that you
could leave out half the transistors.
>
I couldn't immediately see exactly how either of the circuits worked, though I could get the simulations to run under LTSpice
and
see roughly what was going on.
>
I've now dug a bit deeper. Here is a five transistor version of John May's four transistor version.

>
Out of curiousity,

>
Is that allowed Bill? I thought that making component changes to see if the circuit works better was design by evolution?

>
I didn't make the change to see whether it worked better - I did it to see if I'd correctly understood what it was doing. The fact
that it made it work better was incidental.
>

I upped the currents through Q1A and Q1B by about an order of magnitude (R2o down to 27k, R17 down to 22kk and R28 down to 68k)
and the worst case harmonic became the second at 2kHz, 155dB below the the fundamental. The fourth was close behind at at about
157dB down.
>
Essentially, their incremental resistance has dropped by an order of magnitude, and the ripple on the gain-control signal
produces less voltage excursion.

>

I think the only way forward with this circuit would be to build and

test it.
>
Agreed.
>

I'd do a first prototype with everything through hole except LT1679

and NSS40301MDR2G.
>
Why?

 Changing almost anything in this circuit in LTSPice changes the residual harmonic levels.
Assuming the same is true in reality I'd want to be able to change components easily.

I never had much trouble changing surface mount parts.

I'd also put four more resistors in series with each 68k (maybe

reduce them to 56k) for the four diodes so I can make the current pulses in the four diodes exactly equal.
>
Why? I can see an argument for removing all the 68k resistors so the current being fed through R11 is as high as possible, with
the smallest possible ripple. There is a risk that the diode current will turn off fast enough to drive them into snap-recovery,
but it is remote.
>
Increasing the 68k resisitors reduces the effect of the tolerance on the forward voltage drop through each diode, but choosing
diodes with a closer tolerance on the forward voltage drop would be a better way to go. The 1N914 doesn't seem to have one at all.
>
The Infineon-BAS3007ASERIES diodes at least specify 350mV typical and 400mV max at 100mA. I think NExperia had something better
back when it was Philips, but that's a long time ago.

 Changing to schottky diodes changes the distortion but not always down.
So I'd want to be able to make changes easily on a real prototype.

There are lots of different Schottky diodes. If I remember right they don't do step-recovery, which might help. The first thing I'd go for would be a tight tolerance on the forward voltage drop.
You might find that in a quad diode pack. I used to browse Farnell catalogue for that sort of stuff. Searching their data-base doesn't work as well.
The best I could do was the BAS40-05 common cathode dual from Nexperia and Infineon. That offers 250mV min, 310mV typical and 380mV max at 1mA, and looser tolerances at higher currents. Within the part the two diodes are matched to better than 20mV.

And add a capacitor (100n min) to ground where the resistors join.

>
Adding more phase delay along the feedback path and make the settling time even longer.

 One of us doesn't care if he has to wait 5 minutes for the purest sinewave.

What I actually want is critically damped - dead-beat - settling. Having a long period while the amplitude is ringing down is evidence that the circuit designer doesn't know what they are doing. This isn't the first time I've called your attention to this.

The other seems to put higher priority on the circuit settling in a few seconds.
I think we'll just have to differ there.

And use the remaining LT4167 (two quad packs) as an output buffer so

that whatever is connected to the output doesn't disturb the

operation of D10.

>
What D10?

 The one that's D14 in your circuit.

If John May didn't bother it's probably not worth doing.
More to the point, it's easier fit complex layouts around dual op amps rather than quad packages, and you'd be better off using the LT1678.
https://www.analog.com/media/en/technical-documentation/data-sheets/16789fs.pdf
A very low distortion sine wave source does call for careful layout, and   quad packs would be an invitation to disaster.
--
Bill Sloman, Sydney