Facebook
Google
GitHub
Linkedin
Hello again,
I can see now why you did not see any compensation. That's because the 10M resistor limits the amount of compensation for any input offset. So this is perfectly understandable now.
I moved to a more theoretical circuit just to test the theory first. What i saw was only 240uv DC offset compensation, not nearly enough, for a 2mv input offset. Clearly that's not enough. Lowering the 10M hysteresis resistor to 1M brought that up to 1.2mv, which started to make a significant effect on the output duty cycle. With 2mv input offset the duty cycle is not very close to 50 percent duty cycle (as it would be if the circuit were always perfect with a perfect sine wave input) but with the 1M resistor the duty cycle beomces closer to 50 percent although still not exact. I suppose with more improvements this could be made much better. In fact, i have no trouble believing this now as i remembered something from the distance past while working with this circuit.
Back in the 1980's sometime we had a problem with our AC inverters/converters. This typically took in some AC voltage like say 120vac 60Hz and output maybe 240vac 50Hz but the various types could do almost any input output voltage and frequency because they were all mostly frequency converters used by other industries and various militaries.
The problem was really a pain because some of the clients would complain about the noise the units made, and i could not blame them because the technology we used back then relied on massive output transformers which were driven by transistor H bridges. The transistors could be 100 amp devices, so there were high peak currents.
The noise came from the varying DC offset in the primary of the transformer. It caused the flux to ramp up and down and up and down, repeatedly over several seconds, which caused the noise to modulate and that made it even harder to listen to in some place like a computer room where they were often used.
The solution was to integrate the output voltage of the bridge, which is the cause of the variable flux because the voltage is not always perfectly symmetrical for the positive and negative peaks. The bridge output is integrated (really a low pass filter) and then that is used to vary the output offset in a direction that corrects the offset before it gets too large. Problem solved, no more modulating audio noise.
So i see this same thing here. When the output is not 50 percent, the filter can act as an integrator/averaging network and thus correct the output duty cycle. For this circuit to work better though i would want to change some of the values, and make the integrator have a longer time constant too perhaps. It would depend partly on what that voice analysis chip does in response to these changes.
I could post some snap shots, but all you see is the duty cycle getting closer to 50 percent and the output sine getting closer to a zero output offset. It's not perfect though even with the 1M resistor instead of the 10M but it does get significantly better.
To see the effect more clearly, i changed the 10M to 1M first.
Then, add a DC offset voltage maybe 2mv, then short out the 11k resistor to ground (node that now connects to the input offset voltage source so the input offset is still in effect for the op amp but now no feedback) and run the sim noting the waveshapes.
Then, unshort the 11k and see the difference in waveshapes.
