New class A biasing (with non-switching class AB overflow)

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
The start of this thread is actually more or less a copy of my last post in another thread (http://www.diyaudio.com/forums/showthread.php?postid=280834#post280834), but I had the feeling that it did not get the attention it deserves. It also offers an idea that can be used in many amplifiers and is not limited to the 50W amplifier that was the subject of that thread.

The circuit below shows a new(?) biasing scheme for push pull amplifiers, in particular class A amplifiers. Instead of having a fixed bias voltage (like using a Vbe multiplier), it uses voltage sensing over the emitter resistors of the output transistors. For class AB this is less useful, since the best voltage across the emitter resistors should be far lower (20mV - 50mV) for optimum idle current. For class A it is not a big problem to have 0.6V across the emitter resistors. Often more than one set of output transistors is used in parallel to spread the dissipation and the sensing transistors can be connected to one pair of these, while the rather high 0.6V across the emitter resistors asures almost perfect current sharing.
The advantage of this approach is that no thermal runaway will occur and the bias transistors need not be thermally coupled to the output transistors, even should not be. Furthermore there is no need for bias adjustment. It is easy to select proper resistor values, for the right class A idle current. This approach was proposed by PanzerLord in http://www.diyaudio.com/forums/showthread.php?postid=276001#post276001
Although this works quite well in class A pre-amplifiers, and has been used commercially by Yamaha for that, in class A power amplifiers the loudspeaker impedance can drop dramatically for certain frequencies or waveshapes. This has been shown by Otala c.s. in the early eighties. In that case the output current of the class A amplifier would be limited to twice the idle current and severe distortion occurs. To circumvent this current limiting, it would be nice if the biasing scheme would allow the amplifier to move gradually into class AB if more current is required than the amplifier can deliver in class A. Push pull amplifiers with fixed voltage biasing have this characteristic automatically, although switching distortion may occur due to reverse biasing of the cut-off transistor.

Such an improved biasing scheme is now presented, by adding two resistors R5 and R8 and two diodes D6 and D1 (strange default labels BTW). The diodes are Infra Red LEDs of the GaAs type with a forward voltage of 1-1.2V, like CQY36. Most GaAlAs types have a higher Vf of about 1.3-1.5V and cannot be used, just as other LEDs with visible colors.
For small output currents the voltage across R3 and R4 is determined mainly by the Class A biasing and Q2 and Q4 keeps that voltage around 0.6V. The Vf of the LED is too high to have influence. If the output current is increased by a bigger input voltage or a heavier load the voltage across R3 increases for positive signals and the voltage across R4 decreases, for negative signals the other way around. Without the LEDs the maximum output current would be twice the bias current, since at that moment the voltage across R3 (or R4) becomes zero and the voltage across R4 (or R3) cannot increase beyond VbeQ2+VbeQ4. But with the LEDs the contribution of the voltage across R3 (or R4)to the Q2/4 is gradually clipped, leaving some voltage left for R4 (or R3).
This can be seen on the two graphs. The top one shows the output current through R3 and R4 with 10Vp input sine. The output stage is clearly in class A with a bias current of 70mA. The second graph shows the same but for an input signal of 30Vp. Now the stage is in class AB. But note that the current through R3 and R4 never reaches zero, just as the current through the output transistors will not go to zero. This is a "non-switching" class AB, i.e. the current is never switched off completely, nor reverse biased as you can get with normal fixed voltage biasing (Vbe multiplier).

Instead of using IR LEDs for this trick, you can also use a normal small signal Si diode and a Schottky diode in series, giving a Vf of about 1V. A third method is to use a Vbe multiplier with a multiplication factor of approx 1.6. The bigger the Vf, the smaller the minimum current will be in the output stage. If Vf becomes more than 2Vbe it will have no effect.
Of course, distortion will rise as soon the output current halves are "shaped", actually pre-distorted by the action of the LEDs. But this rise in distortion will not be so dramatic as the currents through R3 and R4 suggest. They will be subtracted from each other. The output voltage will still have a distortion less than 0.1% (open loop!). In this simulation most of the distortion comes from R1 and R9 due to varying base currents of Q1 and Q3. This will improve by using darlingtons for the ouput transistors.

The emitter resistor values are quite high in this example and only small signal transistors are used, but that is only because of my lack of models for power transistors in my free downloaded Circuitmaker program. The circuit can easily be used in power amplifiers. The value of R5 and R8 can be from 20 Ohm - 1kOhm. A small value will keep the output longer in class A without starting to shape the currents, a bigger value will increase the influence of the LEDs. This is simply because R5 and R8 are voltage dividers with the Rd of the LED as second resistor.

Another idea: if you use optocouplers for the LEDs you can use the output current as an indicator whether the amplifier is in class A (no current) or class B (current). Of course these should be IR optocouplers with low Vf for the LED, like MCT5210. The output transistors of both optocouplers should be connected in parallel to make a hardwired OR-function.

Steven
 

Attachments

  • class a bias.gif
    class a bias.gif
    20.6 KB · Views: 3,780
Steven,

the idea seems to be interesting to me. IMHO it should be proven in the real amplifier and listenning tests are to be done. I have not had good experience with class A variable bias, this was worse compared to constant bias.

But very interesting, indeed. Might be the key to solve cross-over distortion residuals, resident at class B and class AB amplifiers.

Pavel
 
Hi,

A lot depends on the I/V character of the LED’s. This can vary a lot from very sharp to rather soft. Anyway the resulting curve will generate still a lot of higher harmonics, but probably less than a straight AB biasing. The best similar approach I’ve seen was published some years ago in WW and Electronics. It was called “D2S” biasing if I remember well. It made use of the difference of 2 square functions, hence the name “D2S”.

Anyway your idea is worth investigating Steven.

Cheers ;)
 
Pjotr said:
Hi,

A lot depends on the I/V character of the LED’s. This can vary a lot from very sharp to rather soft. Anyway the resulting curve will generate still a lot of higher harmonics, but probably less than a straight AB biasing. The best similar approach I’ve seen was published some years ago in WW and Electronics. It was called “D2S” biasing if I remember well. It made use of the difference of 2 square functions, hence the name “D2S”.

Anyway your idea is worth investigating Steven.

Cheers ;)


Hi Piotr,

do you happen to have a scan of the article or schematic you could email me?

Thanks,

Eric
 
dear Steven,
I like to read your posts very much but it seems that you spend too much efforts trying reinvent the wheel :D

S.Tanaka, "New Biasing Circuit for Class B Operation," JAES, 1981, Vol. 29, No 3, p. 148

Fifteen years ago I also was that optimistic and try to check as much as possible. Pity to say but 90% of my prototyping time was waste as somebody else did it before. :(
 
New? bias scheme

The topology has much resembles to one developed by mr Tanaka in Japan in the 80's. It was described in the Journal of the Audio Engineering Society then. I do have a copy at the office but not here now. When more details are required I can post them later.
His topolgy uses an output triplet and normal diodes (no LED).
The VAS is rather normal.
I have designed an amplifier, but not yet build or simulated, with this O/P topology and an input based on J.M. Plantfeve.
http://perso.wanadoo.fr/jm.plantefeve/index.html

Comment is welcome

Rick Ruijsink
 

Attachments

  • tanaka-lapaz.gif
    tanaka-lapaz.gif
    8.3 KB · Views: 2,545
peufeu said:
I have tried in simulation and it does seem to work, but how do you handle HF ? The input 5K resistors will make this stage very slow at high frequencies. Any ideas ?

Nice to see that after a couple of weeks suddenly reactions pop up on this. It looked as if there was no interest on such design ideas, compared to ready to built circuit diagrams.
I'm at work now and have little time to answer everything now, but I would like to say already that this circuit is far from complete and was just to indicate the biasing. The 5k resistors were there just to create some input to the circuit without thinking about input stages and input biasing. So also no HF issues here to be solved. In practice Is1 and/or Is2 will be VAS stages and there will be no 5k resistors.

I hope to be able to react on other comments this evening.

Interesting thing about the Tanaka circuit. I think I have such an old JAES even at home, so maybe this circuit was somewhere buried in my subconscious memory.

Steven
 
PMA said:
...the idea seems to be interesting to me. IMHO it should be proven in the real amplifier and listenning tests are to be done. I have not had good experience with class A variable bias, this was worse compared to constant bias.

You can bias the amplifier at the normal class A idle current. It will behave like a normal class A amplifier quite close to maximum output current (twice the idle current), only then the current shaping kicks in. That doesn't look nice on the graph, but we have to keep in mind that a constant voltage biasing (Vbe multiplier or equivalent) is far worse. That will give sharp edges when cutting off the current in one of the output transistors (although the output voltage will hardly be affected).
Anyway, up to the moment the current shaper kicks in there is no variable biasing.
But if the output is biased at a low idle current (like a class AB amp), then you will get an amplifier with class A variable biasing as most of the time the current shaper is active. But I consider this hardly as class A, more as non-switching class AB, since the current wave forms in each half are not the same anymore as the output current.

Pjotr said:
A lot depends on the I/V character of the LED’s. This can vary a lot from very sharp to rather soft. Anyway the resulting curve will generate still a lot of higher harmonics, but probably less than a straight AB biasing. The best similar approach I’ve seen was published some years ago in WW and Electronics. It was called “D2S” biasing if I remember well. It made use of the difference of 2 square functions, hence the name “D2S”.

I agree, but again I would like to promote this as just a way to shift into class AB if needed in extreme situations. Most of the time it should be a "regular" class A.
I'm curious about the D2S solution. I do not have that WW&E article at home. I know of solutions that use a closed Vbe-loop, where the currents in the output halves have a constant product, so in each half flows the reciprocal current of the other half. None will ever have a zero current, so the output will be non-switching, but there is no linear operating area that could be considered class A.
Well, I will have a look when peufeu posts it on the web.

Originally posted by dimitri and Rick NL
S.Tanaka, "New Biasing Circuit for Class B Operation," JAES, 1981, Vol. 29, No 3, p. 148

Well, yes, you are right. I even found it back in some 25 years of old JAES journals.
I may well have kept it in subconscious. I've read the article (again, I guess). Like Rick is showing in his amplifier with Plantefeve input stage, this article is about non-switching class AB amplifiers. In that case the voltage drop over the emitter resistors should better be low for best performance. Therefor the bias sensing is done on the emitters of the drivers instead of the output transistors. This is OK, but the advantages I mentioned at the start of this thread are not valid for the Tanaka solution. These were that no thermal runaway will occur and the bias transistors need not be thermally coupled to the output transistors, even should not be. Furthermore there is no need for bias adjustment. It is easy to select proper resistor values, for the right class A idle current.

Originally posted by Nelson Pass
I refer you to an old patent # 3,995,228, which uses the
diodes, but without the feedback scheme. It's advantage
is that can be used at the output stage of a power amp.

Yep, I found your patent, Nelson. I think the same comments are appropriate as for the Tanaka circuit. It works with low voltage drops over the output emitter resistors, which is an advantage for power output stages. But because it is acting on the drivers it will need bias voltage adjustment (probably) and thermal tracking of the bias transistors and the output transistors.
As my circuit, that appears to be derived from Tanaka's circuit, is intended for class A amplifiers, I do not care that much for 0.6V voltage drop over the emitter resistors, and enjoy the benefit of no bias adjustment and no risk of thermal runaway.


Thanks everyone for your comments.

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