Bad Servos?
It may be worth noting that in a feedback loop of a power amplifier which also has a servo, there are actually two signals being summed at the feedback node one of which is used to correct the DC offset but it also carries audio signals at a level shown by the graph above as well. The actual feedback resistor in a high quality amplifier would be of very high quality such as a Vishay bulk foil or similar as its primary job is to feed back a replica of the amps actual output to be compared to the input signal normally using a differential pair so as to correct any non linearity's that the amplifier may have. A standard resistor would not sound as accurate as a Vishay, as most would agree, yet it is only one component in the signal chain and its type does make significant subjective difference.
There is also the signal being fed back from the servo, which as it would probably be constructed from standard parts and an opamp chosen for low offset as a matter of course rather than is sound quality ability, then the signal appearing at its output would not have the kind of audio transparency that a single Vishay would. In fact if the servo circuit and its PSU were removed from the amp, its integrating capacitor replaced with a resistor to provide it with a fixed gain, it would probably be an amplifier of poor quality sonically compared to the audio transparency of a single high quality resistor and certainly something that you would not use to pass high quality signals through in say the output of a high end preamp.
Why then is it used to pass a division of the amplifiers output back into the feedback node as it does carry audio signals as well as a DC correction voltage. A sonically 'perfect' servo should have filtering in the audio band so that its output at 20Hz is lower than the noise floor of the amplifier so that any 20 Hz and above signals which have been corrupted by the servos poor quality (sonically) circuits are not being fed back into the feedback node. The audio feedback that the amplifier receives should only be from the actual feedback resistor alone, and should not be summed with low quality audio which has passed through the servo.
Regards
David
It may be worth noting that in a feedback loop of a power amplifier which also has a servo, there are actually two signals being summed at the feedback node one of which is used to correct the DC offset but it also carries audio signals at a level shown by the graph above as well. The actual feedback resistor in a high quality amplifier would be of very high quality such as a Vishay bulk foil or similar as its primary job is to feed back a replica of the amps actual output to be compared to the input signal normally using a differential pair so as to correct any non linearity's that the amplifier may have. A standard resistor would not sound as accurate as a Vishay, as most would agree, yet it is only one component in the signal chain and its type does make significant subjective difference.
There is also the signal being fed back from the servo, which as it would probably be constructed from standard parts and an opamp chosen for low offset as a matter of course rather than is sound quality ability, then the signal appearing at its output would not have the kind of audio transparency that a single Vishay would. In fact if the servo circuit and its PSU were removed from the amp, its integrating capacitor replaced with a resistor to provide it with a fixed gain, it would probably be an amplifier of poor quality sonically compared to the audio transparency of a single high quality resistor and certainly something that you would not use to pass high quality signals through in say the output of a high end preamp.
Why then is it used to pass a division of the amplifiers output back into the feedback node as it does carry audio signals as well as a DC correction voltage. A sonically 'perfect' servo should have filtering in the audio band so that its output at 20Hz is lower than the noise floor of the amplifier so that any 20 Hz and above signals which have been corrupted by the servos poor quality (sonically) circuits are not being fed back into the feedback node. The audio feedback that the amplifier receives should only be from the actual feedback resistor alone, and should not be summed with low quality audio which has passed through the servo.
Regards
David
I agree with most of the above. But, if the servo lf xover point is too high (like 10 Hz or so) you actually have feedback at that frequency in addition to the wanted feedback. Depending on the situation you can have net pos feedback meaning that at lower freqs your amps freq response may start to rise. In fact, that always happens, otherwise you cannot get at *infinite* feedback loop gain at DC. So, a servo should have its turnover point preferably below 1 Hz, to avoid it also reacting to musical assymetries as alluded to by JC above, and getting them amplified.
On the other end of the spectrum, if you have a servo amp that is so-so at hf, you can find a distorted version of the hi freqs in music at its output, and you are feeding that back also.
Bottom line: the servo should have wide band, low distortion amp and high-pass roll off not much higher than 1 Hz or so.
It is often the lack of proper implementation that gives servo's the bad name, but there is nothing inherently wrong with the concept.
Just my 2 eurocents worth.
Jan Didden
On the other end of the spectrum, if you have a servo amp that is so-so at hf, you can find a distorted version of the hi freqs in music at its output, and you are feeding that back also.
Bottom line: the servo should have wide band, low distortion amp and high-pass roll off not much higher than 1 Hz or so.
It is often the lack of proper implementation that gives servo's the bad name, but there is nothing inherently wrong with the concept.
Just my 2 eurocents worth.
Jan Didden
IF the servo is fed with lower voltage (most likely) than the main amp you must calculate one important thing to avoid distortion due to clipping.
At 20 Hz (as an example) and full power (100 watts @ 8 ohms, just an example), the servo must not deliver more than 11-13 volts peak if the servo is fed with +- 15V. If you know this, you will have the maximum speed of the servo. In my Gainclone case, with 15 volt (servo) and 35 Volts (LM3886), the speed will be approx. 7 Hz = Gain 1 for the integrator alone. The resulting speed will be lower, around 0.3 Hz = max speed!
Please note also that the integrator signal is attenuated at least 10 times when it's fed into the main amp. Resulting that the signal from the servo at 20-100 Hz is very small! (check this with SPICE if you don't believe me. Meaning, as small problem but it never hurts to use a good or a very good opamp, in my case an AD8620!
I have added simulation files for my Gainclone to play with. The simulations are general so they will fit most situations.
At 20 Hz (as an example) and full power (100 watts @ 8 ohms, just an example), the servo must not deliver more than 11-13 volts peak if the servo is fed with +- 15V. If you know this, you will have the maximum speed of the servo. In my Gainclone case, with 15 volt (servo) and 35 Volts (LM3886), the speed will be approx. 7 Hz = Gain 1 for the integrator alone. The resulting speed will be lower, around 0.3 Hz = max speed!
Please note also that the integrator signal is attenuated at least 10 times when it's fed into the main amp. Resulting that the signal from the servo at 20-100 Hz is very small! (check this with SPICE if you don't believe me. Meaning, as small problem but it never hurts to use a good or a very good opamp, in my case an AD8620!
I have added simulation files for my Gainclone to play with. The simulations are general so they will fit most situations.
Because you want to minimize the garbage at its output- remember, it's supposed to function as an integrator, which means that it rolls off higher frequencies, but only as long as it's actually functioning as an integrator. If it can't respond around its own loop fast enough to cancel HF at its input, that will be reflacted as unwanted signal at the output. You can band-aid it like I do, which allows lower performance opamps to be used.
In reality, something like a LF411/412 is more than good enough, but suspenders-and-belt types like me filter the output anyway.
peranders said:
Per-anders,
Assuming you use an inverting integrator, it will not clip assuming the Fc point is low enough. The servo will drive the inverting input to ground (the non-inv already is there) and the output level should never need to be more than a volt or so, depending on the attenuation in the summer network.
If you use a non-inverting integrator, presumably the input common mode could be several volts, but this can be adequately taken care of with an input RC low-pass filter ahead of the integrator.
So, I really don't see any problem with clipping in the servo, unless it is grossly misdesigned.
Jan Didden
I know this will generate a whole new "range" of feedback, but another way of doing a DC servo is to use a microcontroller and to measure the output offset (or anywhere else you want to measure it).
Now before everyone flies off talking about digital noise, etc. here is something to consider when implementing this:
- You do not need to make continual corrections to a servo loop. For most analog systems, you likely need one "gross" correction, i.e. at startup, that takes care of component mismatches, etc. This will get you pretty close in most circumstances, and may, though not likely, will be all you need. A micro can measure and correct this using a DAC without any difficulty.
- The remainder of the offset will be due to thermal differences while the unit is running. That can typically be taken care of by taking infrequent measurements and making small corrections.
- You can turn most microcontrollers off so the noise is virtually non-existant.
- The nice thing about a micro is that you can make a really fast DC fix on startup, then slowly (and imperceptably) correct for temperature related effects while running
... works for me
Now before everyone flies off talking about digital noise, etc. here is something to consider when implementing this:
- You do not need to make continual corrections to a servo loop. For most analog systems, you likely need one "gross" correction, i.e. at startup, that takes care of component mismatches, etc. This will get you pretty close in most circumstances, and may, though not likely, will be all you need. A micro can measure and correct this using a DAC without any difficulty.
- The remainder of the offset will be due to thermal differences while the unit is running. That can typically be taken care of by taking infrequent measurements and making small corrections.
- You can turn most microcontrollers off so the noise is virtually non-existant.
- The nice thing about a micro is that you can make a really fast DC fix on startup, then slowly (and imperceptably) correct for temperature related effects while running
... works for me
That's what I thought of a long time ago already. But I never tried it in practice as I have no use/time for microcontrollers in my designs.
I tried to convince one commercial manufacturer to pick this up. Makes sense, as commercial gear is bound to have remote control these days, and hence has a microcontroller on board that is doing next to nothing.
The controller only ever has to wake up when the continuous-time
measured DC-offset at the output exceeds a preset value.
I tried to convince one commercial manufacturer to pick this up. Makes sense, as commercial gear is bound to have remote control these days, and hence has a microcontroller on board that is doing next to nothing.
The controller only ever has to wake up when the continuous-time
measured DC-offset at the output exceeds a preset value.
This would certainly work. I think the most difficult part would be to actually measure the offset at the output in the presence of large AC. What one could do is built a zero-crossing detector, and measure the output DC when the zero-crossing detector indicates the AC is zero. But it will be hard to make sure you measure DC offset and not residual AC.
Jan Didden
Jan Didden
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.