DC offset

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi All,

I am trying to design a simple, no feedback direct coupled (except for the input) Class A amplifier (please see the attached schematic). Some of you will like the design, some won't. It's not intended to be clever or original in any way, merely a tailored design to produce a sound I think I'll like. It uses a brute force approach to audio and very old principles - A HV 60mA single ended input stage which will easily drive those lateral output fets, in turn chosen because I can almost get by with no part matching. Anyway, I digress.

As you will see I have made a rather basic discreet DC servo. It works well for this purpose. The one problem I am having us that it is not totally benign and is acting as a feedback path, though not as much as you might think.
This may not surprise some of you, nor myself. I know it is not as nice as an opamp based circuit.

Anyway, what I would like to know is:

1) Given that I have no problem with an amplifier having up 100mV of offset, do I need a servo at all? Would a simple precision voltage reference bias for the front fet suffice? (I know, I should have 0mV, but I honestly don't care about the affect of 100mV of DC on a hundred watt speaker).

2) Any ideas on how to make the servo more benign? A larger filter cap does only so much. I have tried giving the servo a heap of "miller" style capacitance to try to numb its HF response. It's a nice theory, but the whole circuit ends up becoming a VLF oscillator. I have also tried filtering the input to the servo with a CRC filter and also the base drive of Q2 , both mods also turned the amp into a VLF oscillator.

Anyway, I look forward to your thoughts.

Greg.
 

Attachments

  • Image 021.jpg
    Image 021.jpg
    105 KB · Views: 639
I've used this kind of servo on my TGM3 and it worked like a charm - without it I was always wondering if the dc-offset had wandered off to a nasty value. Designed well it can be benign - I haven't looked at your design in detail but what is needed is to ensure the operating frequency of the dc-servo is really low. I don't think you have it quite right yet - it needs to be configured as an 'integrator' - I think you have a capacitor missing somewhere ?

The downside was that you get a significant start-up event whilst the servo settles down - in my case the speaker cone would move in and out a fair bit on power-up and then again a few seconds later when the servo came up.
 
I've used this kind of servo on my TGM3 and it worked like a charm - without it I was always wondering if the dc-offset had wandered off to a nasty value. Designed well it can be benign - I haven't looked at your design in detail but what is needed is to ensure the operating frequency of the dc-servo is really low. I don't think you have it quite right yet - it needs to be configured as an 'integrator' - I think you have a capacitor missing somewhere ?

The downside was that you get a significant start-up event whilst the servo settles down - in my case the speaker cone would move in and out a fair bit on power-up and then again a few seconds later when the servo came up.

Integrators have a tendency to bounce back and forth. This may not be a big problem in audio but in video it's bad. Instead of an integrator an OTA performs a similar task but without the oscillations. This chip works

Special Function Amplifier - Transconductance Amplifier - OPA860 - TI.com

 
Any ideas on how to make the servo more benign?
Hi Greg

I like John Curl's approach to servos, which is to firstly get the DC offset as low as possible without the servo, so that when the servo is connected it has very little work to do. Then the servo can be very loosely coupled to the main circuit, to minimize the feedback in the audible range.

In your circuit you have two gain stages in the feedback loop (Q3 and Q2) and neither of them have degeneration, so loop gain is high and the feedback signal is also unnecessarily distorted.

I'd recommend adding emitter resistors to both to reduce the gain and improve linearity. There's a bit of detail missing in your schematic near Q3. That 0.672V voltage source looks like you might be using a long tailed pair?

Another thing I'd be a little uncomfortable with is the tight coupling between Q2's collector and the input. Three things to note about that:
a) Almost the entire input signal voltage is on Q2's collector.
b) That puts a nonlinear load on the input signal. (OK, it's a small effect but if we can avoid it, why not?)
c) The servo's control range is far greater than needed. If it weren't for the zeners, it could set U1's Vgs anywhere from 0V to 25V.

So anyway, I'd try to decouple that. If I come up with any bright ideas, I'll post them later. (I already came up with a couple of dumb ideas - you know the ones where you get half way through working out the details before realizing it's a complete ****-up because you forgot something important)

Cheers - Godfrey

p.s. Congrats on the FETZILLA project!
(Which reminds me, I think I forgot to reply to a PM at the time - sorry about that:eek:)
 
Last edited:
insert
gootee
dc servo
and it reports 14 results.
No.4 is what I was recommending.
Thread / Thread Starter Last Post Replies Views Forum
You have 2 post(s) in this thread, last Yesterday Smile
6 Attachment(s) Go to first new post Light Dependant Resistor Current Control (Multi-page thread 1 2 3)
Chris Daly

Today 09:14 AM
by Chris Daly Go to last post
100 1,995 Analog Line Level

Adcom gfa 565 problem
Bablo

20th March 2012 12:34 AM
by gootee Go to last post
8 165 Solid State
You have 29 post(s) in this thread, last 7th February 2012
34 Attachment(s) paralleling film caps with electrolytic caps (Multi-page thread 1 2 3 ... Last Page)
jarthel

10th March 2012 08:13 AM
by HiFiNutNut Go to last post
381 41,145 Power Supplies

Use DC servo as a Rumble filter ?
gurevise

10th January 2012 07:50 PM
by Calvin Go to last post
9 293 Analogue Source
You have 20 post(s) in this thread, last 27th September 2007
18 Attachment(s) DC Servo question... (Multi-page thread 1 2 3)
Dxvideo

25th December 2011 04:12 PM
by gootee Go to last post
122 18,268 Chip Amps
You have 2 post(s) in this thread, last 16th October 2011
6 Attachment(s) Balanced line driver (DRV134) and noise
ettos

16th October 2011 10:20 PM
by AndrewT Go to last post
34 4,026 Analog Line Level
Wink
2 Attachment(s) Urgen question about de-coupling capacitor for PCM1704
Kenneth Zhu

30th June 2011 10:56 PM
by gootee Go to last post
30 1,339 Digital Source
You have 16 post(s) in this thread, last 10th February 2008
48 Attachment(s) Thoughts On LM1875 Gainclones (Multi-page thread 1 2 3 ... Last Page)
Zero One

19th February 2008 08:07 AM
by danielwritesbac Go to last post
196 22,448 Chip Amps
You have 1 post(s) in this thread, last 7th February 2008
2 Attachment(s) Op amp Dc offset
legarem

7th February 2008 04:48 PM
by gootee Go to last post
12 3,594 Solid State
You have 3 post(s) in this thread, last 26th June 2007
Check my PSU design and LTspice sim
gootee

29th June 2007 06:12 AM
by gootee Go to last post
23 3,753 Power Supplies

4 Attachment(s) Please explain a servo circuit
ppcblaster

18th June 2007 12:02 PM
by ppcblaster Go to last post
6 3,186 Chip Amps
You have 3 post(s) in this thread, last 17th June 2007
PA100 with DC SERVO?
YFW

17th June 2007 10:17 PM
by gootee Go to last post
35 4,682 Chip Amps

Capacitor for DC servo
borges

13th June 2007 03:29 PM
by gootee Go to last post
5 859 Analog Line Level

18 Attachment(s) Anyone have SPICE models for PSU Toroidal Xfmr ??
13DoW

2nd April 2007 04:01 PM
by kelticwizard Go to last post
49 5,454 Power Supplies
If you have not read Gootee's post on the DC servo thread then you should never post a question on DC servos.
If you have forgotten some of the content on Gootee's DC servo webpage then go back and read it again.
 
Last edited:
Hi, thanks all for your responses:

AndrewT: Found it. Gootee's circuit looks very good, using an opamp based low pass filter. However it is a little more complicated than I was hoping for given the simplicity of the amplifier itself! If all else fails I'll investigate further. Definitely looks like a good circuit though.

Stratus46: Looks like a very impressive chip, though I'm not sure how I would go about using it. Going to read the datasheet a little more. I don't suppose you have a schematic of an example circuit with it anywhere? I'd rather stay away from surface mount devices but would use it if necessary.

SRH: Thanks for the response, I finally found a legible schematic but it's going to take me a while to work my way through it and even longer to figure out how the servo works.

Godfrey, long time no speak! Thanks for the encouragement. Fetzilla was a great learning experience and I am much better off now than I was then, but still pretty ignorant! I have built quite a few amplifiers now though with literally 50 or so incarnations of fetzilla with different input, vas and output stages (bjts/fets/etc).

Anyway, thanks for the servo ideas. You're right, the transistors could use some degeneration. I did actually have some degeneration in there at one point but removed it in the name of simplicity without really thinking of the consequences. In fact if I could linearise the servo, the feedback aspect might not be such an issue. The real problem is the VLF resonance that the servo seems to impose on the amplifier. There's a definite 80hz or so element in all the simulated FFTs I do, though it almost falls into the noise floor. Looking back, I didn't make this issue clear in my first post. The feedback is also certainly a problem, but I guess if I degenerate the transistors and it might almost be a non issue.

Hmm, it might be possible to reduce the amplitude of the oscillation if, as you say, I tighten up the intrinsic operating point of the input fet and reduce the ability of the servo to control the input voltage... Need to fiddle a bit.

The reason I have it the way it is is because, as you would be aware, the servo can only pull the input voltage down (output voltage up), so the gate of the fet has to tend to want to be up rather than down. So as a result I set the bias string to make sure of this. But you are right, it could certainly be tightened up to give the servo less control and reduce the upwards swing at the input gate.

I was also a little nervous about having the input directly coupled to the servo and I look forward to hearing any decoupling ideas you might have. This is why I used a fairly high impedance bias string for the input and between the servo and fet gate - trying to get as many ohms between the servo and the input as possible!

I have also tried adding a decoupling cap to the top of R34, to try to prevent rapid changes in the current through the bias string. This basically turned the amplifier into a VLF rail to rail oscillator.


Oh, and the voltage reference I was going use for the servo was going to be a simple LED and pot, an arrangement which worked very well with fetzilla. Though a LTP would work equally well I guess. I have actually experimented with a LTP based servo, but it was prone to the oscillating problems too. I'm happy to hear of a better servo arrangement using a LTP if you come up with one.

I'm off to keep fiddling!
 
Last edited:
Thinking about this a bit more.....

PSRR is a bitch. The input signal goes directly to a voltage gain stage that's referenced to -75V, so any ripple on the -75V rail will appear at the output, multiplied by the full gain of the amp.

Switch-on thump could be a nightmare too, since the input coupling cap has to charge up to about 70V before the amp can begin to stabilize.

It would be a lot easier if you had an input stage referenced to ground, with it's output driving the voltage gain stage. Only problem is the design starts looking a bit normal then.:p

Anyway, heavy filtering of the -75V line can get the hum low enough. Maybe CRCRC filtering with big enough time constants that the voltage rises sloooowly at switch-on to avoid a nasty thump.
 
Hi Greg

Here's a slight reworking of your servo. With the values shown, frequency response rolls off below 1.5Hz due to the input cap, and again at 0.5Hz due to the servo. Looking at it the other way around, the feedback factor reaches unity at 0.5Hz, is -40dB at 10Hz, and -60dB at 30Hz. In theory.

The high supply voltage makes choosing Q1 a bit difficult. MPSA92 has lousy current gain. BC556 is probably OK, but close to it's max voltage. Maybe BC560C plus cascode would be better?

The trimpot should allow very accurate trimming of DC offset. If it fails open circuit, the two resistors across it get the voltage there close enough that the servo can still sort out the bias fairly accurately.

If you're not too fussed about accuracy, you could leave out the trimmer, R8 and R9, and just use one zener of about 7.5V.

One thing I'm not totally comfortable with is the 100uF electrolytic cap. Hopefully leakage and ESR won't be too much of a problem. The nice thing about opamp based servos is you can use plastic film caps.

Cheers - Godfrey

p.s. Sorry for any typos. Tired, need to sleep now.
 

Attachments

  • servo.GIF
    servo.GIF
    8.1 KB · Views: 253
On second thoughts, the servos shown above will misbehave badly because they can't cope with low supply voltages. It's probably a better idea to use something like the circuit below, which works OK over a wide range of supply voltages, down to about +-5V or so.

One possible improvement: Ideally the voltage across zener D2 should be matched to the Vgs of the MOSFET, so it might be better to replace the zener with a trimmable Vbe multiplier. If you do that, you probably don't want or need the other trimmer.

p.s. Note the klutzy transistor choices in the LTP. :D
 

Attachments

  • servo2.GIF
    servo2.GIF
    7.5 KB · Views: 245
Thanks for putting all the time in Godfrey!

I like what you've done, essentially turning the input low pass filter into an even lower pass filter for the servo by using a super high resistance. I toyed with that idea but I got nervous about having such a high impedance bias network. Not sure why that bothered me, guess I've just never seen it done! Completely makes sense though.

Very good!

My only question is the purpose of D2 in your first schematic. Is that to ensure Q2 is on at turn on?

Can I ask if you have tried the circuit in a turn on situation with an input signal applied? I came up with a few interesting servo ideas that worked well until I did a 0v startup with a 1kHz 1Vp-p input signal i.e. the effect of turning on the amp when the preamp is already attached with music playing. I got terrible rail to rail oscillations. I think your circuit should be fine though. I didn't understand what was going on before but I think you have addressed the problems that were causing it.

As for PSRR, yes it's a real problem and I was thinking of smoothing the HV rails using a ridiculous CRCRCRC filter. Need to think a little more about the actual values, but I was thinking of something like 100R and 1000uF, off the top of my head.

The other thing I wanted to think about, if I were to go the CRCxxx route is the effect of such a high impedance supply on the distortion of the input stage. Anyway, something to think about.

I was also thinking of things like a huge zener string, a basic shunt regulator, a feedback regulator and also a few other ideas. Always keen to hear your thoughts.

I like to think of this thing as a back to basics brute force amplifier which basically (hopefully) uses terrible inefficiency to perform well with only one voltage gain stage and no global feedback. It certainly sims surprisingly well considering the circuit (I know simulation results don't mean much, but its relative performance compared to some more complicated designs is not bad). So the simpler the power supply solution the better I think.

I have stuck with a singleton input because I love the way they sound. All that H2. I have also used a high rail voltage and current for the input stage to keep it well away from clipping and hopefully increase linearity. I'll probably drop the voltage a bit to suit the kind of capacitors I can get for a reasonable price.

I have elected to use lateral output fets, which from my fetzilla experience I know can make a pretty decent 20kHz square wave with only 5mA of drive current. 60mA in the input stage should do the job OK I think, though the input fet will need to be on the main sink. The lateral gets can probably get by with no part matching too.

Anyway, gotta run but thanks so much for taking the time to draw up a circuit and think about it!
 
My only question is the purpose of D2 in your first schematic.
Mostly to make the mental arithmetic easier. :eek: There may have been some other reasons I thought were good at the time, I forget.

D2, Q2, R5 and R6 form a "mirror with gain" i.e. Q2's collector current is directly proportional to Q1's collector current, without a huge DC offset. This means the voltage across R4 is directly proportional to the voltage across R1.

Somewhere along the line, I had a useful insight:
a) The voltage across R1 is 1V (due to LED biasing).
b) I want 35V across R4 to give the servo maximum range.
c) Therefore the DC voltage gain of the servo from base of Q1 to collector of Q2 is going to be 35, no matter what current gain is chosen for Q2 or what idling currents are chosen.

That gets rid of a lot of uncertainty and makes life much easier. Without the diode, everything affects everything else.

Can I ask if you have tried the circuit in a turn on situation with an input signal applied?...
Nope, I didn't try any start up/shut down stuff, but I'm surprised you got oscillation. What sort of frequency was it? I would expect the output to be stuck hard to the negative rail for a few moments while the input cap is charging.

I like to think of this thing as a back to basics brute force amplifier....
I was seeing it as a clash between minimalism and the quest for maximum open-loop linearity. :D One of the things I love about this forum is you get to see, and play with, all different kinds of design philosophy. The high rail voltage and heavy degeneration should indeed give good linearity.

I have stuck with a singleton input because I love the way they sound. All that H2....
I think a nice compromise would be to add an LTP input stage with heavy degeneration, so it has little (if any) gain and negligible distortion. That way, distortion will still be dominated by the MOSFET voltage gain stage, there's no need for a separate servo, and no worries about switch on/off thumps.

...it would seem there's no real point to making this circuit direct coupled anyway...hmmm. Might as well make it single rail and capacitor couple it.
I'm not really a fan of that approach, and prefer no coupling cap and no thumps when switching on or off.

If you do decide to go that route, I'd suggest putting a coupling cap between the voltage gain stage and the buffer as well, so you can still run the VGS at high voltage.
 
Thanks for the explanation Godfrey, I get it now.

Actually I just revisited my simulations and the oscillations were only there for some of the more elaborate versions I came up with.

Your idea of using a heavily degenerated LTP isn't a bad one, though I hate the thought of adding another gain stage, no matter how benign! I have no problem with a big turn on thump provided it's not damaging. Going to play around with a few simulations over the coming days and see what I come up with. Do you have any thoughts on the HV rail smoothing? I am also tempted to try an opamp servo though I'm still yet to come up with a circuit for this amplifier.
 
Do you have any thoughts on the HV rail smoothing?
What you mentioned earlier sounds OK. The output impedance of the RC(etc) network will be quite low and reasonably linear in the audio band, so I don't think it will cause distortion. OTOH, it might cause some cross-talk depending on the amp circuit, so having the last RC separate for each channel might be a good idea.

Something else to consider is a cap multiplier. They can give a very long time constant and good ripple rejection, without using huge caps. They also have very low output impedance but it's very nonlinear, so a bit of a mixed bag.

I'm thinking of something like shown below, but maybe that's overkill.

btw, It occurred to me you don't need a separate transformer for the high voltage rails. You can use a diode-capacitor voltage multiplier fed from the transformer that powers the output stage. Should save some cash that way.
 

Attachments

  • psu.GIF
    psu.GIF
    5.2 KB · Views: 193
Wow, you whip these things up in an instant!

Ah yes, a cap multiplier. Nice circuit, those things still intrigue me. I must build one at some point, in fact it could be a good choice for this amplifier. I wonder how a fet based version would work, in keeping with the overall theme of this design?

What I have been thinking though is this. Correct me if my reasoning is wrong here as we're on the edge of my understanding.

1) One big disadvantage of a single rail amplifier, normally, is that the output cap is outside the feedback loop, unlike the rail caps in a direct coupled amplifier. Feedback normally helps correct any power supply capacitor induced distortion. Since this amplifier has no global or interstage feedback we can pretty much ignore that one and assume that having a capacitor in the output is going to have the same sonic penalty (if any) as capacitors in the negative rail of a dual rail version. So this particular aspect is perhaps not as significant as one might expect...I think.

2) Running a single rail, with a HV input stage means an interstage blocking cap is needed. Not so good...but I do often think the supposed sonic penalty of capacitors is overstated. Plus a good friend tells me that it is alleviated somewhat by placing a high potential across the capacitor, as would be the case here.

3) By running a single rail, I could use two off the shelf transformers (one HV and one LV) with dual windings and have a dedicated winding for each channel and almost full dual mono construction. This lightens the requirements for preventing the crosstalk you mentioned above.

4). As we discussed months ago, a single rail can be set up in a way which keeps the current through the output stage more or less constant, reducing supply ripple and distortion.

5) A single rail with a coupling cap provides inherent DC protection and removes the need for any servos.

Soooo, as much as it is tempting to always go more complicated, the single rail version is looking pretty promising. The only problem is that turn on thump, which I need to go model I think.

Only having one HV supply per channel would also give me a little more motivation to try something more exotic in the power supply (cap multiplier).

It's funny. I really hated the idea of a cap coupled amplifier which is why I came up with the servo in the first place. Now it seems I have talked myself into it!
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.