I am open minded on the subject of servo vs. caps. In my collection of amps at home, I have both types, and have xperimented with both on the same circuit (Version A with cap, Version B with a DC Servo), and in all honesty I cannor swear either sounds better. Of course, there are so many variables and possibilities, it's hard to even liste them all, let alone try them all out.
The only variable I know will make a difference in sound is when you bypass a large cap with a smaller vaue film high quality cap, like say the usual 220 uF electrolytic with a 470 nF (just as an example).
Using a DC servo does give me peace of mind, though, as Richard pointed out these circuits are usually good for the life of the device, and high quality electrolytics are not cheap and will need raplcement after say 10 years or so.
On the performance front, I agree with Andrew's choice in his e-180 amp project, in which he used a 1,000 uF cap in the NFB. It's large enough not to act above say 1 Hz or so, and so it should be outside the audible range. Then there are some potentially dangerous designs which do not use a servo or a cap, relying on very hard core matching of devices to drop the DC offset level down to insignificant, BUT avoidance of input coupling opens the door too wide for my comfort for some DC producing source to do nasty things, since any input buffers have also been avoided for the sake of a simple signal path.
The only variable I know will make a difference in sound is when you bypass a large cap with a smaller vaue film high quality cap, like say the usual 220 uF electrolytic with a 470 nF (just as an example).
Using a DC servo does give me peace of mind, though, as Richard pointed out these circuits are usually good for the life of the device, and high quality electrolytics are not cheap and will need raplcement after say 10 years or so.
On the performance front, I agree with Andrew's choice in his e-180 amp project, in which he used a 1,000 uF cap in the NFB. It's large enough not to act above say 1 Hz or so, and so it should be outside the audible range. Then there are some potentially dangerous designs which do not use a servo or a cap, relying on very hard core matching of devices to drop the DC offset level down to insignificant, BUT avoidance of input coupling opens the door too wide for my comfort for some DC producing source to do nasty things, since any input buffers have also been avoided for the sake of a simple signal path.
Brad,
This work you did for H/K was in the 90-ies, right? I can't swear on it, but I think you didn't do it in vain. My HK 680 integrated amp uses FET switching, specifically using 2SK364 BL, for switching of all line level inputs, i.e. everything but the phono, which has a separate board all of its own.
And I guarantee with zero switching noise, no FET turn-on spikes, nothing but background silence. If so, you did a hell of job there. The model was manufactured until late 1998, and was their top of the line model of integrated amps.
This work you did for H/K was in the 90-ies, right? I can't swear on it, but I think you didn't do it in vain. My HK 680 integrated amp uses FET switching, specifically using 2SK364 BL, for switching of all line level inputs, i.e. everything but the phono, which has a separate board all of its own.
And I guarantee with zero switching noise, no FET turn-on spikes, nothing but background silence. If so, you did a hell of job there. The model was manufactured until late 1998, and was their top of the line model of integrated amps.
You are theoretically right. Reason why i use two OPAs in serial for the purpose. Let-us take a cheap TL072, with an open loop gain of 200V/mV. it makes a 4 exp11 gain. Large enough ?
We can argue too that the caps are not perfect, with inductive and resistive components. Using 48dB/oct, 1MO of resistance values and Jfet inputs impedances, plus the output attenuation what did we have to fear in the audio band ?
Nothing i can measure or hear, on my side.while i can feel differences with various lytic caps qualities.
If you consider, with Current feedback amplifiers, the low impedance of the feedback path, even using very large values of caps, instead of a servo, you will measure a real impact on distortion and phase in the low part of the audio band, that are cured easily by Servos. As they are my preferred amp technology, you can understand my position.
0.003% of distortion at 100Hz with 2200µF of the VSSA to 0.0006% of distortion with diamond and servo, in my example. And a factor >10 in the phases FC.
Last edited:
I have my microprocessor control board with script-able delays and
full DC/overcurrent protect - no worries about ever seeing that
transient.
I have just observed what is actually happening on the amp itself.
Richard's solution of always having the servo running (separate PS)
would work , but .... another PS is not too DIY friendly.
Another issue with that , what if the amp was not powered and
the servo failed ?
Having the servo tied to the CCS power has a few advantages.
(start-up timing/simplicity).
Edit - as the rails "fold" , the timing of the CCS- servo voltages
scale together .... no large transient on shutdown either.
- Servo >200khz - yes, it breaks down and does nothing (usually sits
at 0V) simulated offset rises. Never rectified a 200k sine ?
Never saw anything "break through" ??
Old 70's servo's = ne5532 + just a simple integrator.
Most newer ones = TL072 + filter + integrator.
Have not seen many OPAxxx servo's ?
OS
Last edited:
Reason I asked is that a few years ago, when I was looking into this, I could easily measure differences in power amp distortion between different servo opamps. I don't think I could hear a difference, but then again I am not a golden ear (I think).
Anyway as you say, the effect does exist, and maybe some care in selecting the servo opamp and cap is is order, but I don't really know the magnitude of the issue.
Maybe something for Bob Cordell to address in his 2nd edition ;-)
Anybody any experiences pro or on?
Jan
Interesting. Do-you had kept some datas left ?
What was the frequencies where you measured differences ?
No data left I'm afraid, but the differences were in the higher frequencies, with some opamps in the servo the upturn of THD with frequency occurred earlier than with some others. I really should redo that, I still have that amp.
Jan
Hi Jan,
I think Esperado is correct in his comments on servos. However, as I pointed out in my book, the dc servo IS effectively in the signal path, so we must not be complacent about its design quality. I believe it should be designed with audiophile quality in mind. I personally would never use a TL072 in a DC servo in a high-end amplifier. I think a reasonably audio-grade op amp should be used. I usually use something like an OPA604.
Indeed, I actually prefer a servo that uses 2 op amps - one an inverting servo and one that inverts. This means that only one capacitor needs to be used, as opposed to the necessary 2 capacitors in the non-inverting servo many like to use. A dual op amp is not significantly more expansive than a single. I also beieve that the integrator capacitor should be of good quality - I always use a 1uF polypropylene.
Servo noise injected into the signal path is not a problem as long as it is kept in mind in the design. Attention to detail matters in the high end, and this includes the servo. I discuss this in my book.
Your point about the limited open-loop gain of the op amp letting some stuff through at high frequencies is interesting, but not likely a problem. One only needs to do the numbers or a simulation to see this. A decent FET op amp will have unity gain bandwidth as an integrator up to the 10MHz range. One also must recognize that the amount of signal current being injected into the integrator input is very small, as the integrator resistor is often 1M. I sometimes put a little bit of passive LPF after the integrator to further reduce noise, and this will also kill any HF feedthrough that might get through the integrator. Any such added LPF must be placed at a frequency where it does not create any ringing or instability in the servo. Putting in some such added filtering is easy in the two-op-amp servo. Of course, if desired to avoid two net poles, one can add a zero to the integrator with a small resistor that is placed in series with the integrating capacitor to cancel the subsequent passive pole, yielding a first-order servo, but with the higher-frequency part of the roll-off done passively. This is all probably guilding the lilly.
One must also make sure the servo never clips and has the right amount of authority. High-amplitude LF signals at the output of the amplifier can generate significant signal level at the output of the servo integrator. The servo should be designed not to clip with a full-power 10Hz signal at the output of the amplifier, AT MINIMUM.
Cheers,
Bob
That's why you place the pole at least a decade higher than the integrator. For the servo in my modified Adcom 555, the servo frequency is about 1.6Hz, the RC pole is at about 50 Hz. That pretty much puts the opamp type out of the equation and it can be chosen for DC properties rather than worrying about slew rate or anything like that.
Well, could-we think to add a little inductance in serial with the output resistance of the servo ?
Why, apart his own offset ?
In my example CFA, the output level after the fist OPA is ~450mV at 10Hz for near full power at output. The output of the second OPA is attenuated by a 10K/22.5hms when it reach the emitter. So, noise, distortion etc are divided by 500.
John and Steven, I should be more cautious with my short sentences.
That sentence should be followed by : “ as measured at the coil gap”.
The target magnetic flux there, as well as the dimentions of the magnetic gap are key points and have to be determined at the design stage (there should be a stable point at the magnetic load line diagram).
The pole piece composition then has to satisfy these two design criteria.
Control of leakage fields would be an exercise in geometry (physical forming of the magnetic path).
Electric Circuit Quantity / Magnetic Circuit Quantity
Current, i / Magnetic flux, φ
Voltage, v / "MMF", N.i
Conductance, G / Permeance, L
L= µA/l
where, µ is permeability, A is area and l is length.
http://product.tdk.com/en/products/magnet/pdf/e371_circuit.pdf
George
A DC servo is in the signal path. A TL072 should never be in the signal path of a modern high-end audio amplifier.
Would you put a mylar capacitor in the signal path just because you cannot see its distortion on a THD analyzer?
Cheers,
Bob
George,
Now your throwing science into it! Thanks George. If only we could do things to their theoretical best, the problem is we have other considerations that change things. You compromise one thing for something else. In my case I am more interested in a longer linear field in a gap than getting the highest flux density possible.
Now your throwing science into it!
Thanks George. If only we could do things to their theoretical best, the problem is we have other considerations that change things. You compromise one thing for something else. In my case I am more interested in a longer linear field in a gap than getting the highest flux density possible.862s are amazingly good although they usually require cascoding to address breakdown voltage and dissipation limits in a typical circuit. The pinchoff voltage is such that operation at Idss is near the zero drain current tempco operating point, hence alleviating d.c. drift problems at the outset. This is true of the higher-Idss SK170-like parts as well, or the half-sized SK117.
One drawback for diy is the SOT-23 package. It's also wise to include some lumped inductance in the gate lead to push the self-oscillation frequencies down a bit, lest things scream away at several hundred MHz.
The remarkable thing about the 862 is that the transconductance is achieved with about half of the input capacitance of an SK170. So with two in parallel the resulting composite is ~3dB lower in voltage noise and with about the same capacitance as the SK170. This may not be optimal for certain applications (see Wurcer's discussions in Linear Audio about condenser mic preamps).
I stacked about ten of them with a piece of wire soldered to the gate leads (which is also the best thermal path to the chips). It worked fine without oscillations in a tight layout. Perhaps a little impractical for automated assembly.
I'm cheered that the part may be around for a while, as it is used in volume in China for AM radio front ends. They are also cheap.
Brad
Especially if the stabilized main channel is more than just another IC, there is almost always a place to inject the correction signal from the servo, obviating the need for a differential servo. If for some reason you absolutely need a noninverting integrator, in addition to the sensible suggestion of a following inverter, one can resort to the Deboo integrator topology, with positive and negative feedback---but the resistors have to be well-matched.
I don't know the history of differential servos with two capacitors, but I suspect it's something that was less than well-thought-out, and then started being copied without a lot of examination.
Brad
- Status
- Not open for further replies.