Re: Mic Preamp DC-Servo: The "backwards" way
Posted: Tue Feb 24, 2015 5:03 pm
Sorry for getting this thread back to live.... but it was really worrying me that reality worked better than simulation...
For reference I'll link in the image from my first post again:
My proposal was to inject the DC servo's (differential) correction voltages into the bias voltages (non-inverting inputs of the opamps) to "indirectly" set the relative collector current for the transistors. So in effect instead of connecting R7 and R9 to gound they get (differentially) moved up and down by the servo.
Then we went on to discuss how the effective frequency of the DC servo changes with gain. The two "traditional" ways of attaching the servo are:
A) sample DC offest across the gain set resistor and inject servo voltage into the bases -> the "gain" the servo sees does not change with gain changes, but we amplify any remaing DC offset by that amps gain
B) sample at the output and inject into the bases -> gain changes change the servo's frequency, so at low gains it takes ages for the DC offset to get servoed away
Now I finally saw the real difference in my approach: When altering the gain by changing Rg we also change the gain of the transistors (because Rg is the emitter resistor for the transistors). So in other words, the higher we set the gain, the higher the gain of the transistors gets (and the "remaining" gain is added by the opamp). The crucial thing is that the gain of the transistors does not stay constant. What follows is that any voltage change injected into the non-inverting inputs does not see the full gain of the complete composite amp.
To sum up, when sampling the output DC offset and injecting the servo voltages into the bias voltage nodes the servo's frequency will not change as much as it would when injecting into the bases. Advantages are that we still sample at the very output, so the amp will not further amplify any remaining offset left by the servo (so no need for super special low-offset opamps or manual trimming). And of course we don't have to mess with the bases, so the common-mode bootstrapping (not shown above) does not get compromised.
Olaf
For reference I'll link in the image from my first post again:
My proposal was to inject the DC servo's (differential) correction voltages into the bias voltages (non-inverting inputs of the opamps) to "indirectly" set the relative collector current for the transistors. So in effect instead of connecting R7 and R9 to gound they get (differentially) moved up and down by the servo.
Then we went on to discuss how the effective frequency of the DC servo changes with gain. The two "traditional" ways of attaching the servo are:
A) sample DC offest across the gain set resistor and inject servo voltage into the bases -> the "gain" the servo sees does not change with gain changes, but we amplify any remaing DC offset by that amps gain
B) sample at the output and inject into the bases -> gain changes change the servo's frequency, so at low gains it takes ages for the DC offset to get servoed away
Now I finally saw the real difference in my approach: When altering the gain by changing Rg we also change the gain of the transistors (because Rg is the emitter resistor for the transistors). So in other words, the higher we set the gain, the higher the gain of the transistors gets (and the "remaining" gain is added by the opamp). The crucial thing is that the gain of the transistors does not stay constant. What follows is that any voltage change injected into the non-inverting inputs does not see the full gain of the complete composite amp.
To sum up, when sampling the output DC offset and injecting the servo voltages into the bias voltage nodes the servo's frequency will not change as much as it would when injecting into the bases. Advantages are that we still sample at the very output, so the amp will not further amplify any remaining offset left by the servo (so no need for super special low-offset opamps or manual trimming). And of course we don't have to mess with the bases, so the common-mode bootstrapping (not shown above) does not get compromised.
Olaf