Can you measure the servo output on power up and see if it's working? It would be good to know if the servo is working, but the amp isn't responding to it, or if the servo is just too slow.
It would be interesting to see where the offset is originating from and see if it can be handled without the servo in circuit as well.
It would be interesting to see where the offset is originating from and see if it can be handled without the servo in circuit as well.
The DC servo is positive feedback.
Anything coming from the servo output is added to the input signal and comes out of the output node. That error signal just goes around the positive feedback loop and does not get the benefit of the circuits overall NFB to reduce the errors.
What one needs is a pure DC correction signal from the servo to add to the -IN node. If there is no interference/contamination/noise/distortion on that pure DC then there is no increase in the circuits distortion.
The +ve version of the DC servo is the better version to use for obtaining lowest distortion. It has an input filter that attenuates much of the audio signal. That gives the servo opamp an easier job since it now has a lower level signal to handle. In addition the majority of the remaining signal is at very low frequency. The opamp has enormous NFB capability at VLF so the distortion added by the opamp is inherently very low.
The integrator around the opamp tries to generate a DC output and it automatically attenuates/filters the incoming signal. Thus the +ve servo topology has two filters removing audio signal from the correction voltage.
There is a further enhancement, that has not been used, that can help attenuate spurious non DC voltages at the output of the servo opamp.
The 22k resistor that feeds the correction signal back to the -IN node is equal to the normal feedback resistor. It very effectively bypasses the normal feedback route. The amplifier cannot differentiate between the signals coming from either of these two 22k resistors.
If you double the servo resistor to 44k, the amplifier does not just add the two signals. It halves the correction signal relative to the full NFB signal. A bigger servo resistor reduces the correction signal and that reduces the interference/contamination/noise/distortion that goes around the +ve feedback loop.
We often see a 100k in the servo output for a 10k in the NFB loop. This is chosen to divide the added non DC correction by a factor of 10.
Try this and see what happens. There is a disadvantage to using this 10:1 resistor ratio. The range of input offset that can be corrected is reduced by that same factor of 10. i.e. if the servo has a maximum output of 10Vdc and uses 1:1 resistor ration it can correct an input offset of upto ±10V. The 10:1 resistor ratio reduces this to a correction range of ±1V
There is a further enhancement that can be made.
Split the 100k servo resistor into two 51k in series. At the junction attach a small capacitor feeding into audio ground. This filters the correction signal reducing the non DC part without reducing the DC part. You may find that just doing this, i.e. splitting the 22k into two 11k in series, does enough to allow the extra filter to remove much of the added distortion you measured.
But the filter adds another pole to the servo +ve feedback. It can make for instability. Simulation for many different component values should enable you to find a "safe" set of values that does not interfere with the stability margins.
Last edited:
I asked for confirmation of that 4V value shown on the sch. That would tell us a lot about what is happening.
That 4V value should be around 100mV to 400mV, i.e. the same as appears across the Rin setting resistor.
Yes Jeff,AndrewT,i have measure the servo out.It is 3v using R28=22K.
2.5v using R28=10K.
Using 22k the amplifier start from 3 v d.c out and go to zero after about 10''
Using 10k the amplifier start from 2v d,c out and go to zero after about 5''.
Last edited:
That tells us that there is something wrong with the amplifier or with the servo.
Remove the servo opamp, or lift it's 22k.
compare the voltages at the two INPUT nodes i.e. R1 & R2 node and R24 & R25 node.
These voltages should be very close.
Last edited:
Pete's prototyping schematic shows 4v on offset out,see post #1523.
Last edited:
That high voltage is what alerted me to there possibly being a problem.
The current through the two bases of the input pair should be the same.
Those currents pass through their respective resistors, R2 & R24. (my calculation indicates ~6 to 7µA)
The voltage drop across both these resistors should be the same.
That means the voltages at the +IN and -IN should be the same.
But clearly with 2V to 4V appearing on the -IN something is not fitting with that model.
disconnect the servo and look for where the problem is.
The few reports we are seeing about ultra slow offset correction may well be relating to the same fault.
Last edited:
Well, when there is 4V at the integrator's output, voltage at -IN is much lower, because of the divider. However, if there is more than 100mV at integrator's output, it always makes me worrying - something in the circuit is pretty much dis-balanced.
The closet to zero the offset is set with no integrator, the faster the whole thing will settle with the servo in place.
The closet to zero the offset is set with no integrator, the faster the whole thing will settle with the servo in place.
Andrew, this is something I can't agree with )
In overall, servo is a slow NFB. If the servo is inverting - we connect it to +IN, if the servo is non-inverting - we connect it to -IN (same as the global feedback loop).
Schematic in post #1528 is a little different(other resistors values)and shows 2.5v on offset controller out.
Attachments
Last edited:
If the 6 to 7µA base current is about right, then R2 should have ~150mV across it.
That same voltage (±10%) should appear at R25 & R24 node.
I agree with the rest, with addition that in some cases, some other (better?) places of offset control may be found - for example, in the design with symmetric CCSs feeding the diamond, great thing to control is the CCSs' currents balance (servo moves it up or down, zeroing-out DC at the output).
I would say - something dis-balances the front-end. Example - different base currents in the input LTP.