To DC servo or not to DC servo,this is the question NEW INFO

to DC servo or not to DC servo, this is the question.

In the last few years I read tons of posts about people prefering or not DC servoed audio circuits. I remember at the old place there was a thread dedicated to this subject and there were so many contradictory afirmations that I thought this is something worth to be investigated. The opinions are divided between those who think that DC servo is a bandaid because it eliminates the coupling capacitor but has other problems (sound unnatural, constrained, has problems with low end etc, etc) and the ones who think that DC coupling is better because it eliminates the capacitors (low freq phase problems, dielectic absorbtion that smears the sound, etc)

One of the recent threads, where one member was expressing his not liking DC servos ( quote: Question is, what are remedies for that beside DC servo (which I tend not to like for nontechnical reasons) ") reminded me that this is an unsolved issue, and nobody did a serious controlled test, and draw the conclusions about this subject (or at least didn't make the results public).

So I decided to do myself this test. In fact it is a more extended test because I'm also interested not only about DC servo but also about the claims related to the audibility of various coupling capacitors.

Before anything I decided to do my tests with two different discrete opamps, with and without DC servo. The first opamp is jfet opamp based of Fred Forssell's paper, and the second is based on the JML Hybrid (which is a circuit similar to some used in some expensive comercial devices). The advantage with the Hybrid is that is has a DC servo circuit that can be activated or not depending on application. For the Jfet opamp I had to build a separate one, so I used the DC servo circuit from Fred : http://www.forsselltech.com/schematics/JE16%20Mike%20Preamp.PDF
The first setup was built with four Hybrid type opamps as pairs of circuits for stereo evaluation, one pair with DC servo and one pair without.

The only way to make a correct evaluation is to use a coupling capacitors for each setup, because the one without servo can't be DC coupled to the next device. To reduce as much as possible the sound alteration induced by the coupling caps, I used some very high quality polyprops (MITCAP) so if there is any difference between the servoed and non servoed circuits it should be below the damage induced by the capacitor.
Anyway, based on how radical were the opinions about the sound of servo versus non servo, I was expecting some serious differences.


The setup: computer playing various test files for calibration, and 24bit 96Khz mastertape quality records, minimalist, two microphone orchestral, choir and jazz music. The computer connected to Lavry DA10 converter, the analog outs distributed two the two pairs of discrete opamp based gain blocks (one with servo, the other without). The outs of each pair of the gain blocks into the coupling capacitors and then into the inputs of the Cranesong Avocet. The Avocet was used as an A/B test device because it has a very transparent audio path and the switching is relay controlled.
For the first test I made both pairs of gain circuits as 20dB noninverting circuits, all four opamps configured in class A (LED biasing). I opted for the 20dB gain and not unity gain to make the opamps "work harder". Also everything was calibrated with the RME Totalyser to less than 0.1dB, to avoid the chance to draw wrong conclusions because of any level related problems. So the resistors in the gain blocks were perfectly matched to obtain this precission.


The audio test. I played lots of materials to be sure that my conclusions are correct. I forgot to mention, that my choice for using pairs of circuits and not mono chanels, was that in mono you can make judgements related to timbral changes, but only in stereo with coerent materials you can hear if there is any change in space definition because of phase distortion or other issues.

After listening for an hour and switching from servo to non servo and back for at least 100 times, I can say that I couldn't detect any difference in sound between the two situations. So if there is any difference in sound between a servo and not servo circuit it should be under the distortions introduced by an expensive polyprop capacitor. I'm a bit shocked by the results considering how radical opinions were expressed related to this subject.


Later I did three more tests.

The second test was with the jfet opamps in the same circuit, and had the exact same conclusions. No difference in sound between servo and non servo.

The third test was to use four identical mic preamps with the same test setup. The line level signal was distributed in pairs, and attenuated 40dB before the Lundahl input transformers. everything was calibrated like in the first test. First pair (A) was non servo and the second pair (B) was with servo.
At this test I finally could hear a difference in the width of stereo image at high frequency. Interesting was that the servoed version sounded a bit wider than the non servo but without any timbral change. Before drawing any conclusion I swapped the gain blocks between the pairs, now pair a is servoed and pair B is non servo. The patern of sound was exactly the same but this time the non servo sounded a bit wider. I can say it without any doubt that the small width difference I heard was because of the input transformers, and there is no difference in sound between DC servo and non servo.

Later I redid the same mic preamp test, this time with the Jfet opamps and had the exact same conclusion. The B pair sounded a bit wider (but timbraly identical) regardless of servo or non servo.


So my final conclusion is that there is no audible differece between gain blocks with DC servo and no DC servo. Also I think that direct, capacitorless coupling (with DC servo circuit) has the cleanest audio because there are no artifacts induced by the DC servo circuit and there is the advantage of eliminating a coupling cap.
I sugest anyone who can do this test in controlled condition to do it to confirm my foundings.


chrissugar

P.S. The next test I intend to do, is to make a setup with a gain block with DC servo, and couple it's output to the next stage directly or with various type of capacitors (polyprop, polystiren, elco), to evaluate how much damage produce various capacitors compared to direct coupling. But this will be in another day, I hope in the very near future.

Finally a good experiment on this!
Good work, and thats a lot of hours bolting all the combos together.
You are probably right about the input xfmr being the difference on the high 3end, the servo should only affect low frequencies, right?

I always wanted to try a nickel core output with an extra winding for counter dc.
In other words, the servo action would be done inside the transformer, not another opamp.
This is a standard technique in the instrumentation racket.


But then whats the point of the servo if you can not direct couple?

It seems like the question, "What is better: fat free cookies, or sugar free sousages?"
I always though that the topology must reflect desired end results, and many topologies may be implemented for similar results, some of them are more optimal, some are less optimal, some are equal in cost and results with different technologies in hands.
Trying to eliminate a capacitor using servo equals to moving the capacitor to another place and adding more DC and low end feedback gain. Stabilizing working points of some precious amplifier that can't be stabilized another way is the different story.
However, usage of a servo to compensate errors in calculations or wide tolerance of resistors is another story, for example in hybrid thick film ICs where adjustment of resistor values is costly, but anyway proper technology implementation helps to reduce the need for adjustments...
IMHO, everything may be used if used properly, either sugar or fat.

And you have to know what is being plugged in, and what is being plugged out.
If you DIY a whole system, you can cut corners.
If you sell to the the world, you have to anticipate unorthodox applications and design accordingly.
So the DIY rack wins everytime, as far as S/N and Sonics.

I have never had a Jensen circuit on the bench, but I am willing to bet thst tempeture is a major concern.

If you put a 6.2 1/2 watt zener on the bench in series with enough resistance to current starve the backwards p/n, and put a fluke 4/1/2 dijii on series with the supply, you can get an idea of what is going on in a servo circuit.
The current kind of oooozzzes up and down, like a roller coaster stuck in a dip.

While this may be just a semantic distinction, servo's do not eliminate the blocking capacitor they instead trade a crap big one for smaller value good ones and a hopefully neutral opamp path, that does no harm in the process.

I used a servo in a no compromise phono preamp. While I concede I did it partially for the design exercise and marketing hook, I was pleased with the result. I'm not from the camp that all blocking caps must be bad. I mainly don't like them for low impedance applications or when you can't control immediately before or after. I am enthusiastic about Wayne's effort to eliminate them from phantom powered mic preamp front ends (low impedance). I've had similar thoughts for a while, but never reduced my scratchings to practice, so good job Wayne. Now, I Just need to do it betta (in my spare time). :wink:

JR

That 2 Hz looks like ringing.

I suppose tube amps may ring on infralow frequencies when 3 stages with interstage caps have a global feedback. Some people like to mix micked sound and output from speaker leads. But your picture shows too high amplitude for the single channel summed with others, most probably it is something related to CD...

80 Hz HPF is what doctor prescribed for guitars. I have 20, 40, 80 Hz switch on my tube mic preamps, 12 dB/Oct.

[quote author="mediatechnology"]

So my final conclusion is that there is no audible differece between gain blocks with DC servo and no DC servo. Also I think that direct, capacitorless coupling (with DC servo circuit) has the cleanest audio because there are no artifacts induced by the DC servo circuit and there is the advantage of eliminating a coupling cap.
I sugest anyone who can do this test in controlled condition to do it to confirm my foundings.

Click to expand...

Thank you for that. Very interesting. I tend to agree. I suppose like most things there is a "right" way to do it and a "wrong" way. I think this may be an example of the wrong way:



Check out that bodacious base-line shift in the right channel. 2 Hz? It's the guitar intro to Switchfoot's "We Are One Tonight."

Don't know we can hear it, but the asymmetry in the guitar drives this one nuts.[/quote]

I would expect a properly executed DC servo to look like a simple HPF.

OTOH guitar amps, preamps whatever, can and do routinely clip asymmetrically for some "nice" sounding distortion. An otherwise 0 DC content AC waveform if clipped asymmetrically will have a transient DC component that then works it's way through any following AC coupled stages drifting toward zero DC equilibrium again. If the LF perturbation seems to be following signal peaks perhaps that's an alternate explanation.

JR

Nice job Chris. I confess I think about things a lot more than I have the opportunity and time to listen to them, so I always like to hear some real results of careful tests, particularly when I trust the person and know they have at most a minimally skewed preconception.

A long time ago, when I was working on a system design to be used in conjunction with Floyd Toole's speaker mover facility at Harman, people at Madrigal were queried as to the likely success of my designs up to that point. It was interesting when it came out that they had basically abandoned servos in favor of plain old coupling caps---IIRC just Wima polypropylenes.

But then about that same time, popping the hood on an expensive preamp of theirs revealed they had also migrated to monolithic opamps, which led an associate to write them off as beneath consideration as serious high-enders from then on :roll:

Thank you all.
I'm happy that after many years of reading contradictory opinions about this, I have a real world answer to this question. :grin:
There are two more tests I'm very interested in, and hope I will have the time to make a setup to test them.
One is about the sound of various coupling caps compared to direct coupling, and the other about audibility of low frequency phase shift (at what freq and what angle it starts to be hearable). From my experiments I have the feeling that I hear diferences but untill I make an A/B test in controlled conditions I can't draw any conclusion.

chrissugar

Thanks for this contribution! I suggest that you could include a third test to your setup, i.e. a precisely trimmed DC coupled amplifier without servo. For this you may use a low drift trimming circuit as shown e.g. here: B_r1.pdf

Samuel

Yes Samuel, thank you, I had this in my mind.
I just have to make a circuit that has a very low DC offset from the beginning and add that trimming circuit. But that would probably eat more time because of fine tuning the circuit, so I leave it as a test to be done after the capacitor test.

chrissugar

[quote author="chrissugar"]Yes Samuel, thank you, I had this in my mind.
I just have to make a circuit that has a very low DC offset from the beginning and add that trimming circuit. [/quote]

Thanks for sharing your research. I have been thinking about this alot. I dont like cap-coupled outputs because the impedance of the load is unknown, in order to assure LF preformace you have to size for minimum load. this leads to the desire for 1000 uF that have all the problems that implies.

Im personally on the fence about servos. Ive witnessed bad sounding servos. wierd bouncy bass, etc. And im not sure about the real word reliability of direct coupled, servoed outputs.

I have been playing with DC-trimmed, direct coupled output stages. IMO the solution is to cap couple into the output stage, where you can setup favourable conditions for that cap to operate in. Im not saying this is totally drift-free, but it works.



(there sould be a 10k resistor in the top leg of those biasing diodes)

the problem is, when you DC couple into an unknown input that has low DC-Z ( like a transformer). If it is at even slightly different potential than your output, the current will start to flow like crazy. same story for a servo. This is what doesn't sit well with me. If you are certain that the next stage will be cap-coupled then this is fine, but then why try to get ultra-low offset at all? ( for my personal equipment I am perfectly happy with up to 1 volt of offset into a film cap-coupled inputs.) In a way a servo output stage makes alot of sense, but in the "real" world (a much stupider world than I live in) Im just too afraid it will blow up when joe sixpack plugs it into the "power amp in" jack on his guitar amp, or a mic amp input with phantom on, or whatever. I guess it worked for the SSL 9000 though, and there are plenty of other examples out there.

mike p

[quote author="mediatechnology"]OK John it makes sense. Could it also be a beat frequency from doubling?
[/quote]

Could be alot of things. sometimes detuned guitars do weird LF stuff, ever listen to the Melvins or Sunn-O)) ? the "b-wow" sound. there is a rougly 1/2 a Hertz modulation when loose strings are strummed hard and let to ring out. not saying that is what caused this, but anyway

[quote author="mikep"]
Im personally on the fence about servos. Ive witnessed bad sounding servos. wierd bouncy bass, etc. And im not sure about the real word reliability of direct coupled, servoed outputs.

mike p[/quote]

Most people don't understand feedback control theory very well, or bother to simulate, so there are doubtless some very bad servo designs out there.

I recall Dan Siefert telling me about Jim Bongiorno's hassles with servos. Dan "knew that he (Dan) knew not" but conjectured that Jim's dilemmas arose from wanting very high gains in the loop, and got some things working better just by figuring out how to reduce them. But these were the days when powerful simulators were generally unavailable, and you had to know and do the maths.

As John R has said, and PRR has remarked (loosely paraphrasing) they are a way to get a low-frequency cutoff based on a good film cap instead of a big not-so-good electrolytic. They are not "out of the signal path" just because the topology doesn't look the same as one with a coupling capacitor in series with something.

However, the devil is in the details---where do you inject the correction and what does it do to the rest of the circuit? How much noise comes from the servo amp that is not reduced by other circuit responses? How much noise is associated with the higher impedances? What happens in overload? What about component tolerances---for example, in the popular differential integrator approach, what does cap mismatch do?

I abandoned my usage of servos for a project i was working on for bipolar caps.

glad I did too. It actually ended up cheaper to use a single BP cap than add an opamp and a couple of poly caps.

I did a similar test to Chris's, between the same circuit with a servo and a bipolar coupling cap. I didn't hear much difference either. I'm sticking with the BP caps.

MORE INTERESTING RESULTS!

Although I'm busy in this period, I managed today to make two more tests.
I was afraid to make the servo DC coupled versus non servo DC coupled, because I was sure that it will be a big pain to make a gain circuit without DC servo to have low offset and not to drift in time.
I installed each discrete opamp I had, one at a time, into the gain circuit. Powered up and left them for a while to stabilise, and measured the DC offset without the DC servo. I found an opamp that had less than 0.5mV offset at the output and it was stable, without more than 0.2mV variations over 15 minutes. I thought I have the perfect candidate to make the most relevant test, the direct coupled with DC servo versus direct coupled without DC servo.
The setup was the same like in the previous test, everything calibrated to less than 0.1dB.

The result? I couldn't detect any hearable difference between the two situations. No timbral change. No change in details, space information, reverb tails.
So I can say without doubt that in my particular test setup I couldn't detect any difference in sound between the two situation.


I was so excited about this that I decided to take some more time and do some direct coupled versus capacitor coupled tests.
The test was not very extensive because my time was limited and probably will do this test again with different type of capacitors from different manufacturers, but I managed to test three situation:
1-expensive 47micro polyprop (biggest value I have)
2-medium price/quality 47micro polyprop
3-electrolyt 2200 micro Panasonic FC parallel with 100nano WIMA polyprop

1-the overall difference is very small, a very little bit the sound is smeared. The biggest difference is at low freq where the direct coupled has more energy and the sound goes lower (but I'm not sure that the reason for this is not the value of the capacitor, maybe not big enough for linear phase/freq response)
2- the same thing but a bit more smeared the sound.
3- the biggest surprise, the Panasonic FC bypased with the WIMA poly. This combo sounded a little bit more smeared than the expensive polyprop but without any timbral change, and the bass was not affected like with the first two situations. (which leads me to the idea that the bass difference at the first two tests was determined by the capacitor value)
In my opinion this is good news because we should not forget that the Panasonic was not polarised at all (nonpolarised elcos have higher distortions). Also it is a big surprise because anyone would expect that an electrolyt will sound significantly worse than a polyprop, which was not the case. Of course the polyprops sounded a bit closer to the direct coupled connection compared to the Panasonic, but we are talking about nuances.

This conclusion contradicts all the radical opinions about electrolyt capacitors. Too many times I read opinions like "ELNA is the only way and Panasonic is crap", or "Black gate is perfect, all the others are crap" or any other combination.
If the difference between a Panasonic FC and direct wire is this small, then even if there are many other capacitor better than the FC we cant talk about differnces like day and night.


For me this test is also relevant because it will surely affect my decissions in my future projects. For perfectly transparent sound (mastering,orchestral recording) the DC coupled with DC servo is the way to go, but for non critical application high quality polyprop or elco (Panasonic, ELNA, Nichicon) can produce very good results.


chrissugar

P.S. One more thing, the test was done only for one single capacitor against wire in the audio path, so I can imagine that in a sitution where more caps are in the signal chain they will make bigger differences.

Congratulations Christian!

You've discovered for yourself that the question, "What is better: fat free cookies, or sugar free sousages?" has no direct answer and depends on the situation.

next up: voltage biased electrolytic caps?