This option can also be called class Super A, the current is set automatically, stabilized and maintained at the initial level.
Maybe you can take note of something interesting from here:
Maybe you can take note of something interesting from here:
That is a good point. I must admit, my intellect is really being tested here.
Yes i noticed that was a bad screenshot
Thank you Hennady!
I have not seen that arrangement before. Is that JVC's super class A? Looks beautiful. Ill certianly investigate it.
The super A circuit is interesting, but is an entirely different approach: class A but non-linear achieved through exponential Ic versus Vbe and fast local feedback around the output stages. This is in contrast to Rupopulles' class AB with a slow bias regulator.
Ed
Ed
Yes I want subsonic bias feedback. My favorite funny business audio-band autobias is this beauty here on https://www.data-odyssey.nl/AutoBias_II.html.
It does need two matched quad bjt's and component/current values are quite critical.
It does need two matched quad bjt's and component/current values are quite critical.
This is where I landed:
Using opamp current mirros as current sensors allows Rsense to be exceedingly low. So small one could probably get away with 1W resistors.
Discharge current is as small as can be without causing a large bias surge at startup and is now supplied by a current source.
With the addition of Q36 an inverting integrator can be used. C12 cleans up the opto-led current at high frequencies.
The bias or rather crossover current drops with about 15mA at 20Hz 100W yet rises with about 15mA at 20kHz. (somehow lol)
It sets the DC bias at 106mA per output device.
Slowly starting to breadboard this,,,,
Using opamp current mirros as current sensors allows Rsense to be exceedingly low. So small one could probably get away with 1W resistors.
Discharge current is as small as can be without causing a large bias surge at startup and is now supplied by a current source.
With the addition of Q36 an inverting integrator can be used. C12 cleans up the opto-led current at high frequencies.
The bias or rather crossover current drops with about 15mA at 20Hz 100W yet rises with about 15mA at 20kHz. (somehow lol)
It sets the DC bias at 106mA per output device.
Slowly starting to breadboard this,,,,
Try a square wave. The pulse becomes very narrow at high slew rates (~10ns) and has to be held for a long time (seconds).
Ed
Ed
yes indeed. Attempting to sense the bias this way does allow square wave to ''eat up the bias'. After a handful of seconds it has dropped to zero. Couple seconds after the square wave stops the bias returns, i suppose thats fine,
Late to the party, but are you aware of the 'standard' autobias design by Edmond Stuard (also a member here, also from NL).
That's a proven design and might give you some ideas and alternative perspective.
Ian Hegglun has also done interesting work in that area.
We all stand on the shoulders of giants, and if you don't know what went before you, you are doomed to repeat it instead of building on it. My € 0.02.
Jan
That's a proven design and might give you some ideas and alternative perspective.
Ian Hegglun has also done interesting work in that area.
We all stand on the shoulders of giants, and if you don't know what went before you, you are doomed to repeat it instead of building on it. My € 0.02.
Jan
Op-amps are too slow to detect the overlap. Use diodes and discrete transistors. The speed should ultimately be limited by the output transistors.
Ed
Ed
Veel dank Jan!
Uw input wordt gewaardeerd
.
Ovidiu Popa and Edmond Stuart implemented this approach in their PGP Amp:
https://www.data-odyssey.nl/AutoBias_II.html
Ian Hegglun's approach:
https://www.diyaudio.com/community/...and-non-switching-auto-bias-power-amp.375141/
Both are quite ingenious, Is this the work you are referring to?
Lefts not forget the option of just using a LT1166:
https://www.analog.com/media/en/technical-documentation/data-sheets/1166fa.pdf
All these approaches however use full-bandwidth bias feedback. I found that 'sliding bias' or non-switching circuitry often re-introduces some sort of distortion. I think this is because the amp now has to deal with the non-linear currents in the bias-system. They also complicate the use of emitter to emitter speedup caps.
Hence im trying to develop non-non-switching subsonic bias feedback.
much cheers,
Ruben
Me too i suspect the opamps are too slow. Then I got this result. The total bias voltage rises with a large 20Hz squarewave:
Here V(c) is the output voltage of the sensor circuit. I(RL) the loadcurrent. V(dip) is the 'bias reading'.
V(c) indeed does not slew fast enough dropping to 2.1V on the downslope but is fast enough (too fast?) on the upslope dropping to 0.55V.
Now simplified a bit, time constants are now lower and the input of U1 now got a diode clamp.
Here V(c) is the output voltage of the sensor circuit. I(RL) the loadcurrent. V(dip) is the 'bias reading'.
V(c) indeed does not slew fast enough dropping to 2.1V on the downslope but is fast enough (too fast?) on the upslope dropping to 0.55V.
Now simplified a bit, time constants are now lower and the input of U1 now got a diode clamp.
Unless you have already done so, maybe you should try to get your slow loop to enforce the same class-B current distribution law as the emitter followers.
I am considering whether it wont be better to do some high frequency filtering on Vc, such that the bias always falls with square waves. Seems safer that way.
Yes but the more basic article is from Edmond and was published in Wireless World in 2003.
It's not so much the particular implementation (there are many ways to skin a cat) but the systematic approach to the problems.
Attached.
He is one of the early adopters of looking at the wingspread curves showing gain throughout the crossover region as a measure of linearity.
Even references Marcel's work ;-)
Jan
Last edited:
@Rupopulles I'm not entirely sure what I mean either, but it must be something like this:
Case A.
Suppose the voltage between the bases of Q29 and Q30 were kept constant in the circuit of post #1, and that the temperature of the transistors stayed constant as well. There would then be a nonlinear function defined by the transistors and their emitter resistors that defines how the momentary output current splits between Q33...Q36.
As a simplified but impractical example:
Imagine we only had Q33 and Q34, no Q35 and Q36, and that they had perfectly exponential voltage-to-current characteristics. Also imagine they had no emitter resistors and that the other transistors were ideal voltage followers. The product of the current through Q33 and the current through Q34 would then be the square of the quiescent current (nonlinear class-(A)B control law). Their difference would be the output current of the amplifier, set by the normal signal feedback and the load impedance.
In reality, you have the emitter resistors making the transition less gradual and the equations more complicated, but still, there is some control law.
Case B.
Suppose you made a fast class-(A)B bias loop, or an autobias with funny business, as you call it. There is then a nonlinear circuit somewhere defining the nonlinear function, that is, the nonlinear class-(A)B control law. For example, in "Audio power with a new loop", Electronics World February 1996, pages 140...143, https://worldradiohistory.com/UK/Wireless-World/90s/Electronics-World-1996-02-S-OCR.pdf , Tr21, Tr24, R34...36 and R37...39 define the control law.
Case C.
Suppose you have a complimentary emitter follower output stage with emitter resistors, and you add a class-(A)B loop with a control law that matches the control law of the output devices. It then shouldn't matter much whether the loop is fast or slow, as the loop and the output devices are essentially in agreement with each other.
It could very well be that what I just wrote is impractical, as I remember you had issues due to non-instantaneous behaviour of the output stage at 20 kHz.
Edit:
With a complementary emitter follower stage with emitter resistors, you basically have zero current through one side when the current through the other side is anything above a few times the quiescent current. With a big low-frequency square wave, the loop has practically no chance to sense whether the quiescent current is still OK, as the currents through the output devices just switch between 0 and some large value. I don't see how what I wrote above could help to solve that; with such signals, it would only work with smooth, non-switching control laws.
On the other hand, the autobias circuit for valve amplifiers on https://zelfbouwaudio.nl/forum/ conceptually should have the same issue, but no one ever had any issues with music.
Case A.
Suppose the voltage between the bases of Q29 and Q30 were kept constant in the circuit of post #1, and that the temperature of the transistors stayed constant as well. There would then be a nonlinear function defined by the transistors and their emitter resistors that defines how the momentary output current splits between Q33...Q36.
As a simplified but impractical example:
Imagine we only had Q33 and Q34, no Q35 and Q36, and that they had perfectly exponential voltage-to-current characteristics. Also imagine they had no emitter resistors and that the other transistors were ideal voltage followers. The product of the current through Q33 and the current through Q34 would then be the square of the quiescent current (nonlinear class-(A)B control law). Their difference would be the output current of the amplifier, set by the normal signal feedback and the load impedance.
In reality, you have the emitter resistors making the transition less gradual and the equations more complicated, but still, there is some control law.
Case B.
Suppose you made a fast class-(A)B bias loop, or an autobias with funny business, as you call it. There is then a nonlinear circuit somewhere defining the nonlinear function, that is, the nonlinear class-(A)B control law. For example, in "Audio power with a new loop", Electronics World February 1996, pages 140...143, https://worldradiohistory.com/UK/Wireless-World/90s/Electronics-World-1996-02-S-OCR.pdf , Tr21, Tr24, R34...36 and R37...39 define the control law.
Case C.
Suppose you have a complimentary emitter follower output stage with emitter resistors, and you add a class-(A)B loop with a control law that matches the control law of the output devices. It then shouldn't matter much whether the loop is fast or slow, as the loop and the output devices are essentially in agreement with each other.
It could very well be that what I just wrote is impractical, as I remember you had issues due to non-instantaneous behaviour of the output stage at 20 kHz.
Edit:
With a complementary emitter follower stage with emitter resistors, you basically have zero current through one side when the current through the other side is anything above a few times the quiescent current. With a big low-frequency square wave, the loop has practically no chance to sense whether the quiescent current is still OK, as the currents through the output devices just switch between 0 and some large value. I don't see how what I wrote above could help to solve that; with such signals, it would only work with smooth, non-switching control laws.
On the other hand, the autobias circuit for valve amplifiers on https://zelfbouwaudio.nl/forum/ conceptually should have the same issue, but no one ever had any issues with music.
Last edited:
The autobias circuit for valve amplifiers on https://zelfbouwaudio.nl/forum/ has slow loops that monitor and control each output device individually, clamping the sensed momentary current to twice the set quiescent current to keep everything working in class AB or B. The advantage is that the current never drifts to a too high value. It only gets into trouble when you play large asymmetrical square waves for a long time, and it can inject some (rather limited amount of) distortion into the differential signal path.