Some time ago Gingertube suggested to incorporate a bias servo in a design I was working on. I had not heard of a bias servo for tube amps before (I'm coming from the bone head tech of old guitar amps), so it grabbed my attention. The latest topics I could find about bias servo's date from some time back. I briefly looked at various designs which should work with class AB amps, like the ones from Gingertube/Menno van der Veen, Morgan Jones, John Broskie, Norman Koren to name a few. I'm also aware of the solution offered by Guido Tent. Overall the subject of bias servos is still way over my head.
The basic question I still have: is a servo worth the effort over more conventional bias schemes in a e.g. class AB 2xEL34 amp? I'm not aiming for a zero maintenance amp and am comfortable with adjusting bias every now and then. What are the biggest gains I can get?
In ltSpice I could see that when the amp has some form of feedback around the bias servo (so gnf or Schade) it tends to oscillate at a low frequency, unless I make the time constants in the servo very long. Is that the way to stabilise it?
Got a lot more questions about how a servo should work, but this should do for now...
The basic question I still have: is a servo worth the effort over more conventional bias schemes in a e.g. class AB 2xEL34 amp? I'm not aiming for a zero maintenance amp and am comfortable with adjusting bias every now and then. What are the biggest gains I can get?
In ltSpice I could see that when the amp has some form of feedback around the bias servo (so gnf or Schade) it tends to oscillate at a low frequency, unless I make the time constants in the servo very long. Is that the way to stabilise it?
Got a lot more questions about how a servo should work, but this should do for now...
ALL servos work the same basic way: a sense resistor in series with the cathodes of the output tube(s) is a current-to-voltage converter; tapping its rapidly varying (and low level, due it being to a low value resistor) signal, and running that into something that'll time-average it to a near-DC voltage.
That near-DC voltage in turn is one input to a "servo amplifier", which in turn inverts the slightly positive averaged-out DC voltage, and turns it into an appropriately engineered negative "bias" voltage. It self-adjusts after that.
This is done to not "resistively" develop a bias voltage (using cap bypassed cathode resistors or CCS's), but to have an active long-time-duration negative bias, well below ground.
Let's see what others have to say.
GoatGuy
That near-DC voltage in turn is one input to a "servo amplifier", which in turn inverts the slightly positive averaged-out DC voltage, and turns it into an appropriately engineered negative "bias" voltage. It self-adjusts after that.
This is done to not "resistively" develop a bias voltage (using cap bypassed cathode resistors or CCS's), but to have an active long-time-duration negative bias, well below ground.
Let's see what others have to say.
GoatGuy
I think the desirability of a bias servosystem depends on amplifier bias sensitivity. If performance and/or tube life suffers disproportionately from small bias changes, and other approaches to bias stabilization are judged undesirable, then the servo can be useful. A nice bonus is that it's usually pretty easy to include a bias fault indicator.
Bias with a cathode resistor has two problems. You lose operating voltage for the powersection (can be up to 50V and more).And since the power is proportional to the square of the voltage it can make much difference. It is only good for classA and more or les classAB not classB.Because the bias voltage is derived from the mean anode current.That current goes up with output power, so does the bias but inverse.ClassB get into classC monentarely, sounds bad. On the other hand fix bias doesn't "listen" to the tube and do nothing if the tube changes by ageing or overheating.You have to use lower grid resistors, bigger coupling capacitors and more drive from the previous stage.
Mona
Three anwers at the same time
Mona
Three anwers at the same time
Thanks for all the swift answers. Here's the circuit Gingertube drew.
AFAIK supposed to de class AB as well. R46/47 are the sense resistors. Rock solid bias: even if simmed with a pair made out of an KT66 and EL34 the current draw, or better said dropped voltage over the sense resistors is perfectly equal. However, it only sims when C26/28 are as big as they are. With lower values and corresponding time constants the servos oscillate at around 2.5 Hz. As I see it, the slow performance of the servo is not a problem at all when up and running. Luckily tubes don't age that quickly. But how does it perform at startup of the amp? My guess is that it would take about a minute for the bias to settle.
What bothers me most is that the bias still shifts with a big transient, but that could also be the rest of the design. Have a look at it, maybe that's where I go wrong.
AFAIK supposed to de class AB as well. R46/47 are the sense resistors. Rock solid bias: even if simmed with a pair made out of an KT66 and EL34 the current draw, or better said dropped voltage over the sense resistors is perfectly equal. However, it only sims when C26/28 are as big as they are. With lower values and corresponding time constants the servos oscillate at around 2.5 Hz. As I see it, the slow performance of the servo is not a problem at all when up and running. Luckily tubes don't age that quickly. But how does it perform at startup of the amp? My guess is that it would take about a minute for the bias to settle.
What bothers me most is that the bias still shifts with a big transient, but that could also be the rest of the design. Have a look at it, maybe that's where I go wrong.
The amp design seems well thought out with "RCA" style feedback from output tube anodes to opposite side driver cathodes plus "SCHADE" style feedback from output tube anodes to grids. You could (if you like) vary that 2nd feedback to get some of the self balancing action of the Baby Huey.
Rob at rmsacoustics also commented on low frequency stability of the bias servo and he came up with a variation of the menno/guido scheme. Unfortunately he seems to have taken down his "hobby" tube audio stuff from his business website. I think I have a copy of what he did somewhere. Send me a PM if you want it.
The Morgan Jones circuit copes with the Class AB stuff by clipping the sensed cathode current at 2 x idle current thus balancing the natural clip at zero tube current.
The Menno/Guido scheme is similar but clips the sensed current signal in a much narrower window about the idle point.
The Koren scheme biases a single output tube with standard fixed bias, senses that tubes average current and uses that as a reference for the simple servos on the other tubes to follow. Quite elegant, but if that "master" tube has a problem then it will force all the "slave" tubes to follow the fault condition.
Broskie has multiple schemes including a "sliding bias" scheme. He is worth reading.
Here is some of his stuff:
Aikido Christmas Sale and More MS Zune review and thoughts
Amplifier auto bias circuits: Alan Dower Blumlein's garter circuit
Amplifier auto bias circuits: Class-AB!
Cheers,
Ian
Rob at rmsacoustics also commented on low frequency stability of the bias servo and he came up with a variation of the menno/guido scheme. Unfortunately he seems to have taken down his "hobby" tube audio stuff from his business website. I think I have a copy of what he did somewhere. Send me a PM if you want it.
The Morgan Jones circuit copes with the Class AB stuff by clipping the sensed cathode current at 2 x idle current thus balancing the natural clip at zero tube current.
The Menno/Guido scheme is similar but clips the sensed current signal in a much narrower window about the idle point.
The Koren scheme biases a single output tube with standard fixed bias, senses that tubes average current and uses that as a reference for the simple servos on the other tubes to follow. Quite elegant, but if that "master" tube has a problem then it will force all the "slave" tubes to follow the fault condition.
Broskie has multiple schemes including a "sliding bias" scheme. He is worth reading.
Here is some of his stuff:
Aikido Christmas Sale and More MS Zune review and thoughts
Amplifier auto bias circuits: Alan Dower Blumlein's garter circuit
Amplifier auto bias circuits: Class-AB!
Cheers,
Ian
Thanks for your suggestions Ian. I've read through all of Broskie's blogs about DC servos and how they are supposed to work in a class AB amplifier. Essentially he clips the signal at 2x quiescent current too, like more designs do. What still makes my head spin is how he uses a integrator and pnp transistor to proces the reference and measured signals. The pnp is part of the integrator and has some considerable gain. How is he keeping things stable hf wise? I get loads of oscillations when swimming his circuit, which I can only tame by adding a strong low pass filter (470k -> 100u) in the output of the integrator but before the feedback cap. There's got to be a better way to do it.
I did something similar to the circuit you posted some time ago and which I posted again two posts ago. The low frequency oscillations show up as a push pull signal at the outputs of U7 and U8 and the attached transistors. A 22uF cap between the collectors of Q7 and Q10 kills the oscillation. My limited experience with this circuit makes me wonder if there are simpler ways to stabilise it. Have you tried this servo in real life Ian?
Took a look at the Blumlein Garter circuit. That's a really neat circuit indeed. The biggest downside seems to be the loss of some voltage in the cathode circuits, which could or could not be a problem depending on the design goals (efficiency, character on overload) and e.g. transformers used. I plan on using 600Vct pt's so not much margin on the high voltage if you ask me.
Oh and I tried balanced feedback to the EF184's ala Baby Huey, but did not save the results to compare them with the design I posted here. Will look at it again.
I did something similar to the circuit you posted some time ago and which I posted again two posts ago. The low frequency oscillations show up as a push pull signal at the outputs of U7 and U8 and the attached transistors. A 22uF cap between the collectors of Q7 and Q10 kills the oscillation. My limited experience with this circuit makes me wonder if there are simpler ways to stabilise it. Have you tried this servo in real life Ian?
Took a look at the Blumlein Garter circuit. That's a really neat circuit indeed. The biggest downside seems to be the loss of some voltage in the cathode circuits, which could or could not be a problem depending on the design goals (efficiency, character on overload) and e.g. transformers used. I plan on using 600Vct pt's so not much margin on the high voltage if you ask me.
Oh and I tried balanced feedback to the EF184's ala Baby Huey, but did not save the results to compare them with the design I posted here. Will look at it again.
I made a PCB for myself with the van der Veen circuit. I used a higher precision op-amp in the first stage and achieved bias balance of four tubes within less than 0.25mA deviation of each other.
I also measured bias drift with large music signals and can tell you that the drift is so small that it is not a concern. With the windowing that the diodes do on AC signals, it is very slow and very small. It is certainly orders of magnitude better than the bias drift that occurs in a cathode-biased amp.
I have a board from Guido Tent in my Unity-Coupled amp and I believe it does take about a minute for the servo to get the bias to the correct point. I used smaller caps in the board I made myself and bias adjusted within 10 seconds or so and I didn't have any stability issues. Of course, I made sure that I didn't have overly big coupling caps that I was trying to charge. I think that is where you can get into trouble is if you have a slow RC time constant on the stage you are trying to bias, then you need an extra, extra slow servo to prevent oscillations.
I also measured bias drift with large music signals and can tell you that the drift is so small that it is not a concern. With the windowing that the diodes do on AC signals, it is very slow and very small. It is certainly orders of magnitude better than the bias drift that occurs in a cathode-biased amp.
I have a board from Guido Tent in my Unity-Coupled amp and I believe it does take about a minute for the servo to get the bias to the correct point. I used smaller caps in the board I made myself and bias adjusted within 10 seconds or so and I didn't have any stability issues. Of course, I made sure that I didn't have overly big coupling caps that I was trying to charge. I think that is where you can get into trouble is if you have a slow RC time constant on the stage you are trying to bias, then you need an extra, extra slow servo to prevent oscillations.
Thanks a lot Spreadspectrum. I was wondering: is that servo you used exactly the same as the one I posted above? .25 mA deviation seems more than good to me and if the reference voltage is variable to each servo, I'm sure you can get them exactly the same. Either way, it is good enough for using a toroidal transformer for the output.
The time constant before the bias in my schematic of the amp above is rather long. 22n to 1meg would give me a -3 dB at 7 Hz, if memory serves me well. I'll test that to see if that gets rid of the low frequency oscillation.
Last question: would you be so kind to share your bias servo with me/us? I'd like to make my own pcb's with two servo's so they can be used in mono blocks. Couldn't find it on your website/blog.
The time constant before the bias in my schematic of the amp above is rather long. 22n to 1meg would give me a -3 dB at 7 Hz, if memory serves me well. I'll test that to see if that gets rid of the low frequency oscillation.
Last question: would you be so kind to share your bias servo with me/us? I'd like to make my own pcb's with two servo's so they can be used in mono blocks. Couldn't find it on your website/blog.
I have attached an image as I built it. I used .1% resistors for R1, R2, R9, and R10. I used 20uF for C1 in the only one of these that I have installed in an amp so far. I also used a different transistor stage. I did away with the follower, I didn't see much point in it. I used an MPSA92 transistor with a 1k degeneration resistor and a 270k load for a 220V negative supply.
I experimented a bit with values for C1 to get low frequency stability. I started with 10uF and noticed a bit of overshoot in the servo response at power-up. I changed to 20uF and noticed that overshoot was less than a volt. I played loud music and the servo output was well behaved. More capacitance probably could have cured the overshoot but I think I got a response that was a good compromise between power-up adjustment time and good behavior. The module from Guido Tent is more conservative (slower) so it will drift less and work in pretty much every amplifier.
If you have stability issues, you have a few options to solve them. Change C1 to a higher value, change the emitter degeneration resistor to lower open-loop servo gain, lower input stage gain, or lower the gain of the stage that you are trying to bias. You just need servo gain to be low enough by the time it starts to interact with the gain of the stage being biased.
My amp output stage was a .1uF/1M coupling network (1.6Hz cutoff) but that worked fine with 20uF C1 and 1k transistor degeneration because my output stage gain was only 3.3 due to the local feedback designed in the amp.
I experimented a bit with values for C1 to get low frequency stability. I started with 10uF and noticed a bit of overshoot in the servo response at power-up. I changed to 20uF and noticed that overshoot was less than a volt. I played loud music and the servo output was well behaved. More capacitance probably could have cured the overshoot but I think I got a response that was a good compromise between power-up adjustment time and good behavior. The module from Guido Tent is more conservative (slower) so it will drift less and work in pretty much every amplifier.
If you have stability issues, you have a few options to solve them. Change C1 to a higher value, change the emitter degeneration resistor to lower open-loop servo gain, lower input stage gain, or lower the gain of the stage that you are trying to bias. You just need servo gain to be low enough by the time it starts to interact with the gain of the stage being biased.
My amp output stage was a .1uF/1M coupling network (1.6Hz cutoff) but that worked fine with 20uF C1 and 1k transistor degeneration because my output stage gain was only 3.3 due to the local feedback designed in the amp.
Attachments
Thanks Spreadspectrum! Your values differ some from the ones I used, but the schematics are pretty similar otherwise (no big surprise). 100k -> 22uF is stable in your build. The closed loop gain in the output is somewhere around 10 in my build IIRC. 470k -> 47uF should do the trick I think. Takes a bit longer to stabilise, but if I'm not mistaken, the bias begins at the max negative value at power on (non-inverting input starts at 0V and needs to go down to adjust the bias more positive, right?).
Anyway, I've drawn the servo boards and power supply for it. That power supply will also provide the negative voltage to the CCS of the drivers/LTP and negative and positive voltages for source followers after the drivers. Pretty simple bipolar supply combined with a doubler for the bias etc. Zener/transistor regulator for the servo op amps.
Filtering in the supply is only moderate. I'm thinking that the differential setup of the amp nulls that out, but please correct me if my thinking is flawed.
Btw, I left the cascodes in the servo. It's form factor is small anyway. Here are the schematics, layout and a quick mockup (not accurate). Please, please let me know what you think of it.
Anyway, I've drawn the servo boards and power supply for it. That power supply will also provide the negative voltage to the CCS of the drivers/LTP and negative and positive voltages for source followers after the drivers. Pretty simple bipolar supply combined with a doubler for the bias etc. Zener/transistor regulator for the servo op amps.
Filtering in the supply is only moderate. I'm thinking that the differential setup of the amp nulls that out, but please correct me if my thinking is flawed.
Btw, I left the cascodes in the servo. It's form factor is small anyway. Here are the schematics, layout and a quick mockup (not accurate). Please, please let me know what you think of it.
Attachments
No, I think I was misinterpreting. You are right that it should provide isolation, but I haven't noticed a big improvement in ripple rejection. Bias voltage is effectively isolated from the signal though and stays very stable with large signals.
Saw any other mistakes than me getting fooled by the look of two transistors on top of each other?
Saw any other mistakes than me getting fooled by the look of two transistors on top of each other?
If you use some kind of limiter to limit variations of idle current, why not to use just stable bias adjusted by pots? It would be better in terms of absence of dynamic distortions caused by bias variations?
What is really needed, a sample and hold device in the servo loop that will open feedback when output is below some threshold, and disconnect the loop holding the voltage stable while the amp delivers more power.
Search for some analog synthesizer schematic, it has a S/H device to keep notes when you release a key.
What is really needed, a sample and hold device in the servo loop that will open feedback when output is below some threshold, and disconnect the loop holding the voltage stable while the amp delivers more power.
Search for some analog synthesizer schematic, it has a S/H device to keep notes when you release a key.
Thanks Wavebourn. I see the servo as a means to cope with ageing tubes without having to readjust, a sort of plug and play. And of course getting more in depth about servo's is an intellectual challenge to me, being a non EE. Let's also not forget that Spreadspectrum already uses the servo which I posted.
I swiftly went over some sample and hold examples. So I should store the bias value ( actually an analog voltage) in a large cap and hold it there with a blocking jfet switch? Looks like a simple job for a uC. Sense no signal -> measure voltage drop over cathode resistor -> store value -> adjust analog out to value to tune bias circuit to desired voltage. Could be done every time at start up of the amp, but requires hv and filaments to be settled and monitor bias adjustment changes by the uC. Gets rather complex though.
Will look better at the analog version of a sample and hold circuit, sounds interesting to me.
The driver board will have provisions for a pot meter bias adjustment circuit, thus allowing me to always choose for the simplest option.
I swiftly went over some sample and hold examples. So I should store the bias value ( actually an analog voltage) in a large cap and hold it there with a blocking jfet switch? Looks like a simple job for a uC. Sense no signal -> measure voltage drop over cathode resistor -> store value -> adjust analog out to value to tune bias circuit to desired voltage. Could be done every time at start up of the amp, but requires hv and filaments to be settled and monitor bias adjustment changes by the uC. Gets rather complex though.
Will look better at the analog version of a sample and hold circuit, sounds interesting to me.
The driver board will have provisions for a pot meter bias adjustment circuit, thus allowing me to always choose for the simplest option.
But isn't the first transistor stage the one that would have much worse PSRR? Adding a follower with a CCS load can't make that any better.
I just didn't see enough tangible benefit to include that second transistor stage. Of course, I am making sure not to run the (single) transistor stage with resistive load near cutoff or anything and all of the stages I have been biasing were mosfet follower grid drivers, so I really didn't have to worry about breaking a max bias resistor spec. For biasing a power tube, you may well want an extra stage running higher current to stay within specs for max bias source resistance.
Sure, that would be the best system but this works a lot better than you might think due to the windowing circuit. I watched bias drift with loud music playing and it seemed negligible.
But isn't the bad PSRR in the voltage amplifier stage, the one with the resistive load? That would be the stage that could use some help, but a CCS load placed there would open other cans of worms...
Followers are already pretty good at rejecting PS noise, so while a CCS load makes them even better, it seems like this is one of those diminishing returns things. I think the more important benefit of the CCS would be to help ensure that your bias driving circuit has a reasonable current at all voltages. That might be important if you are directly biasing a big power tube. I don't have this problem as I almost always am biasing mosfet followers that are direct-coupled to the power tube grid. The mosfet gate pulls almost no current so I don't have to worry about that.
Followers are already pretty good at rejecting PS noise, so while a CCS load makes them even better, it seems like this is one of those diminishing returns things. I think the more important benefit of the CCS would be to help ensure that your bias driving circuit has a reasonable current at all voltages. That might be important if you are directly biasing a big power tube. I don't have this problem as I almost always am biasing mosfet followers that are direct-coupled to the power tube grid. The mosfet gate pulls almost no current so I don't have to worry about that.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.