Well that is a nice opamp.. but also very expensive.
It is important that all the components are avaible as leaded devices, because it must be easy for people to build.
I think we don't agree about what a DC servo is..
No audio signal must run through the servo, only the DC must go back.. The audio signal have to be totally removed before the servo result is injected back into the amp..
I disagree.. If i but a servo on my circuit with same filter as the one you suggest, the THD will by very high.. Implementation of a servo like that will be difficult, because the servo input to output gain in the circuit most be extremly low.
If 20Hz is below -120dB, I would say that the audio signal is below the noise floor, and totally gone.
Spice plot is uploaded to my website.. scroll to the bottom DIYhifi.dk - Forside
1mHz is +50dB and 20hz is -103dB a bit worse than I rembered..
I think the response is nice, but maybe the +50dB is to much, and that is what makes the swing..
That is a plot of the amplitude of the DC servo open-loop gain? Phase matters too. Can you simulate the whole amplifier? Or do a Bode plot?
50dB at 1mHz is not a problem, as this sets the DC offset error. The problem will be somewhere between 100mHz and 10Hz, where the servo has some gain but also lots of phase shift. If you plot the frequency response of the whole amplifier you will see a big peak in this region. Then you can adjust the DC servo to reduce the size of the peak.
50dB at 1mHz is not a problem, as this sets the DC offset error. The problem will be somewhere between 100mHz and 10Hz, where the servo has some gain but also lots of phase shift. If you plot the frequency response of the whole amplifier you will see a big peak in this region. Then you can adjust the DC servo to reduce the size of the peak.
Servo phase added: DIYhifi.dk - Forside
I can not simulate the hole amp, because there are too many components.. but as i remember the servo input to output gain in the transistor stage is 0.8 to 1.0 depending on level.
I have made real life measerments, and the frequency response is totally flat from 0,05 hz to 5mhz..
What is a Bode plot?
OK, it looks like the integrator dominates so by the time the phase has changed sign (to positive feedback) the amplitude is sufficiently low. I am surprised you see a flat response, as I would expect a peak somewhere below 1Hz. The damped ringing of switch-on suggests a peak somewhere. You could estimate the frequency by measuring the frequency of the ringing. Anyway, it doesn't look as bad as I feared.
Google "Bode plot".
Google "Bode plot".
I think that globally, Lars didn't make a bad job:
The Bode plot is pretty clean, and the analysis with a step excitation shows a mildly oscillatory response. Not exactly optimum, but certainly not alarming either.
The red trace is for a 500n cap in position C28; clearly too small.
Cyan is the original 680n: some overshoot followed by a small undershoot, but acceptable
With 900n (magenta), there is still a hint of overshoot; quasi-optimal
1µ2 (gray) is already too much.
In conclusion, I'd say 1µ is optimal.
Note: I estimated the gain of the various stages rather grossly, and for an accurate analysis, an exact calculation would be necessary.
The Bode plot is pretty clean, and the analysis with a step excitation shows a mildly oscillatory response. Not exactly optimum, but certainly not alarming either.
The red trace is for a 500n cap in position C28; clearly too small.
Cyan is the original 680n: some overshoot followed by a small undershoot, but acceptable
With 900n (magenta), there is still a hint of overshoot; quasi-optimal
1µ2 (gray) is already too much.
In conclusion, I'd say 1µ is optimal.
Note: I estimated the gain of the various stages rather grossly, and for an accurate analysis, an exact calculation would be necessary.
Attachments
That is some real nice work you did there.. Thanks
Really nice simulation results.. And it explains the behavior of my servo. I am not able to do such fine results myself, so I am very exicited about this.. If there are any thing you need to now about the transistor stage, just ask...
You have rather few actually so it's quite possible. If you draw the schematic you can use generic parts. You don't have to get the exact spice model since it's a LF simulation. If you haven't used LTSpice yet, this is a golden opportunity to start.
Last edited:
Gootee has a great thread and website showing LTspice simulations of DC servos. (the thread was not started by Gootee)
A must read before you commit.
Peranders also has good info on his website, and many comments in the threads that should be listened to.
The slow oscillation from 1V offset to 0.1mV offset should be eliminated.
The time to correct from start up should be shortened considerably.
Can you reduce or remove the start-up offset of 1V?
Rx & Cx.
15k4 is fixed. Make 220p fixed as well.
Add a 4way dip switch. Two poles add in 15k4 or 6k8 in parallel to 15k4. Two poles add in 100p and 150p in parallel to 220p.
A little label inside the lid explaining the switch settings. Better still, the two 4way dip switches on the bottom of the PCB accessed through a little hole/s in the floor of the chassis.
A must read before you commit.
Peranders also has good info on his website, and many comments in the threads that should be listened to.
The slow oscillation from 1V offset to 0.1mV offset should be eliminated.
The time to correct from start up should be shortened considerably.
Can you reduce or remove the start-up offset of 1V?
Rx & Cx.
15k4 is fixed. Make 220p fixed as well.
Add a 4way dip switch. Two poles add in 15k4 or 6k8 in parallel to 15k4. Two poles add in 100p and 150p in parallel to 220p.
A little label inside the lid explaining the switch settings. Better still, the two 4way dip switches on the bottom of the PCB accessed through a little hole/s in the floor of the chassis.
Last edited:
Okay.. Do you have a link?
Well, first of all I find listening tests and real life THD measerements very important, a servo with to low filter values brings in a lot of THD, and bad sound.. I have tried a servo like Peranders suggest, but it can never work if the mission is best possible sound.
The change of C28 as Elvee suggest will make that much better.
And the opamp protection diodes DF96 suggest, eliminates the startup charge of the servo filter..
The offset of the transistor stage without servo is still below 100mV.
Bandwidth limit as fixed? Why? This will make bad sound when the stages is used for preamps..
No switches in signal path..
did I miss read your post?
The table shows three gain settings with the three resistor values and the three capacitor values. It appears that the way you presented the information that 220pF and 15k4 are a compensated pair that work together to give a stable output when set to that gain.
It further appears that adding a 15k4 in parallel increases the gain but when doing this, the 220p must be replaced by a 330p.
Have I read that right or wrong?
As for a the two fixed values for Cx & Rx they are soldered in and will dominate the the sound coming from that tiny part of the circuit.
To switch in a parallel resistor and switch in a parallel capacitor and suggest that these parallel pairs will degrade the sound audibly is beyond my understanding of sound quality issues.
At Page 1 of the schematic, it is mentioned that bandwidth limit is optional for DACs. But I think Table 1 must be edited, for better understanding.
Adding 15k4 will decrease the gain..
Well.. I am trying to build something that can match the best high-end gear.. It may not be possibel to hear one switch, but the idea of unnecessary swithces in the audio path is bad in general.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.