Variable operating biass output ?

[Bigun]

This isn't quite a coherent idea yet, but would be interested to hear if there are already some others who have thought along these lines. My apologies if this is 'old news' as my Search skills are not yet great.

The problem is how to avoid cross-over distortion in AB but still have good power efficiency.

I was thinking if we start with Class A but adjust the operating point (dc bias point ?) of the output device to suit the signal. A big signal means moving the operating point up so that negative swings don't clip. Lower signal means the operating point moves down to conserve power.

How I think that this can be done - the bias circuit needs to know in advance what the signal is going to look like so it can adjust the biass. Well, how about the biass circuit 'sees' the input signal before the power devices. You introduce a 'delay' to the signal between the input and the power stage. A 'delay line' so to speak. The non-delayed signal is fed through LP filter (below amplifier LF cut-off) and the output used to drive the biass. Perhaps this can be done symmetrically for a Class A push-pull output.

Now the output biass tracks the signal level in Class A but with greater power efficiency.

Am I crazy ???

 

[Williams Audio]

hi ,
i think the idea as idea is fine but:
1- you will have to deal with feedback that is delayed so you add some distortions
2- there are some big names are trying to work so they call it in different names, please check up with KRELL, CARVER and so
3- the idea is good but not easy to make it work, if you success let me know too
4- NEVER GIVE UP with innovating ideas
best regards
Williams

 

[PaulBysouth]

Gareth,
You idea is a good one, but old (most ideas usually are). While there are many variations of this theme, the most interesting group is for `Non Switching Class B Amplifiers’. The most useful references on these are:

P. Blomley `New Approach to Class B Amplifier design’
Wireless World, Feb. 1971

S. Tanaka `New Biasing Circuit for Class B Operation’
AES 65th Convention, paper 1615, Feb. 1980

E. Margan `Crossover Distortion in Class B Amplifiers’
Electronics & Wireless World, July 1981

Marcel van de Gevel `Audio Power with a New Loop’
Electronics World, Feb. 1986

Huon, Dower US Patent 6-630-865
Oct. 2003

I think the most interesting of these is the Gevel and Huon ideas (Huon and Dower’s patent is an extension to van de Gevel’s work), in which the bias is constantly adjusted during each cycle so that the `Off’ output transistor never actually turns off, but settles to a minimum current. I’ve spend a lot of the last 18 months carefully looking at the Huon / Dower scheme, and while it fixes a number of problems (bias stability in particular) it also introduces some more (higher order distortion principally). Incidentally I went to University with Graeme Huon and we are currently building loudspeakers together, and Wal Dower and I spend 11 years working together designing, building and installing radar equipment all over Australia.

I think there is potential to monitor the output bias current, and continuously adjust the bias circuit using a small microprocessor. Now that most inputs are digital, this gives the opportunity of implementing a small digital delay, so that a look ahead bias scheme could be implemented. But this is far to complex for me!

I haven’t given up on the Huon / Dower scheme as I think it has potential. I also have an extension to the patent that makes it simpler to implement the bias scheme, but stability is an issue. In effect the bias scheme has to operate at about 5 to 10 times the frequency response on the actual amplifier, so 250 KHz plus – not easy!

There is also an interesting variation of the van de Gevel scheme by Edmund Stuart (Electronics World, Dec. 2003) for a Class B amplifier, in which the bias scheme operates at very low frequencies to set the bias current only, and hence eliminates the need for thermal feedback.

I hope all this is of interest.
Paul Bysouth, May 2009.

 

[Bigun]

I think the most interesting of these is the Gevel and Huon ideas (Huon and Dower’s patent is an extension to van de Gevel’s work), in which the bias is constantly adjusted during each cycle so that the `Off’ output transistor never actually turns off, but settles to a minimum current.

Flipping Heck!, this was the 2nd idea I had, I woke up very early this morning because it was running through my head and I was thinking of how to implement this. Darn frustrating to reinvent something old


- but seriously, thank you for responding to my post, I'd never have found those references myself and it saves bigger heartache down the road

Do you have any prototypes on the go, anything you can share ?

 

[Bigun]

I think there is potential to monitor the output bias current, and continuously adjust the bias circuit using a small microprocessor. Now that most inputs are digital, this gives the opportunity of implementing a small digital delay, so that a look ahead bias scheme could be implemented. But this is far to complex for me!

Paul,

I was thinking of analogue delays, but I believe this route could produce lots of distortions. The digital option seems much better.

Mr. Williams - I don't see feedback as an issue, the 'advanced notice' of what the signal looks like can come from the pre-amp or dedicated buffer at the front, the power amplifier can include a feedback loop that stays within the scope of the 'delayed' signal.

I fear I need a partner in crime to try out these things or they will likely never get off the drawing board.

 

[Bigun]

There is also an interesting variation of the van de Gevel scheme by Edmund Stuart (Electronics World, Dec. 2003) for a Class B amplifier, in which the bias scheme operates at very low frequencies to set the bias current only, and hence eliminates the need for thermal feedback.

I hope all this is of interest.
Paul Bysouth, May 2009.

Anybody have a copy of the Edmund Stuart article ??

 

[latala]

To be honest you are flogging a dead horse !
its been tried since the early sixties with sliding bias etc but has never taken off!
If you google improved "class b amplifier" by Renardson you will see that he has done this very successfully He uses a method of feedforward over the output pair to prevent Gm doubling on a class AB output stage thus avoiding any trace of crossover distortion,All it really means in practical terms is a pair of extra resistors over a normal output stage!
Most annoying that he has done this!!! 40 years I have beeen looking into this problem without success
Regards Trevor

 

[Ian Finch]

It may have slipped the memory somewhere but an obvious and proven alternative, "class XD", substantially avoids the crossover problem. If Self's documentation doesn't prove it, perhaps we may only be considering earlier attempts, such (as you say) the sliding bias concept.

As I understand Blomley's method, it didn't....not in an OP stage anyway!

 

[Ian Finch]

Thanks for ref. Iatala, Renardson (easiest to just Google his Surname) has considereable documentation on his project site at Angelfire. A lot of reading there. Bon apetit.

 

[Edmond Stuart]

Anybody have a copy of the Edmund Stuart article ??

Of course I have one! See: Auto Bias
BTW, this design is a bit obsolete. In a few month or so I will update my website with a new version: fully symmetrical and high speed (>1MHz).

Regards,
Edmond.

 

[Mooly]

http://www.diyaudio.com/forums/solid-state/154396-new-idea-low-dissipation-class-amplifier.html

Post #1 "You read it here first"

 

[Sebastiaan]

Dear,

I think that Krell did this best still with their "Sustained Plateau Biasing". I think one should not try to dynamic adjust the bias to the music signal. It is very hard to keep track, avoid delays enz.

The way krell does it, it has multiple bias "presets" (7 if memory serves me well) It looks at the input what kind of dynamic range is there to be expected, and according to that information it set's the bias preset. Does the input exceed this threshold, the system switch to a higher bias preset. This can be done analogue with some relays and comparators and CMOS.

With kind regards,
Bas

 

[Bigun]

To be honest you are flogging a dead horse !
its been tried since the early sixties with sliding bias etc but has never taken off!
If you google improved "class b amplifier" by Renardson you will see that he has done this very successfully He uses a method of feedforward over the output pair to prevent Gm doubling on a class AB output stage thus avoiding any trace of crossover distortion,All it really means in practical terms is a pair of extra resistors over a normal output stage!
Most annoying that he has done this!!! 40 years I have beeen looking into this problem without success
Regards Trevor

Well maybe, but this is DIY so we are unbound by common sense or commercial constraints !

The angelfire website shows a design where the 'top' output has to remain in ClassA in order to generate the error signal for the 'bottom' output. This is not what I'm seeking, as clever as it looks. I want both 'top' and 'bottom' of a push-pull output to remain in Class A, but without a fixed and high biass current. I need it for my aplty named 'Botch Up' project and if it works I'll likely take it further to develop my home 'reference' amp along similar lines.

It may have slipped the memory somewhere but an obvious and proven alternative, "class XD", substantially avoids the crossover problem. If Self's documentation doesn't prove it, perhaps we may only be considering earlier attempts, such (as you say) the sliding bias concept.

As I understand Blomley's method, it didn't....not in an OP stage anyway!

Class XD doesn't do it for me - it shifts the cross over region away from the zero signal level with the hope that low signals will see fewer cross overs. I'm far from convinced that this is a good approach. I wonder, for example, if larger signals will potentially see even more cross overs as the generally lower amplitude but higher frequency peaks ride on the back of the generally larger low frequency waveform. Without XD this low frequency waveform pushes the hf peaks away from the zero line and they see fewer cross overs, but XD moves the cross over point higher to meet them ? Anyhow, it's still mostly a Class AB approach and what I want is Class A all the time but with less power dissipation.

Blomely's method looks very attractive, but as you say, it's still Class AB because there are still cross overs - only they are taking place before the output stage power devices. So it doesn't meet my needs for Class A.

Of course I have one! See: Auto Bias
BTW, this design is a bit obsolete. In a few month or so I will update my website with a new version: fully symmetrical and high speed (>1MHz).

Regards,
Edmond.

Thanks Edmond, nice to be introduced to the article by the author !!!
It looks ingenious and we'll want to see your next version as soon as possible.

The general approach looks relevant to the goal of variable biass Class A - sampling the output stage biass current, integrating it and adjusting the output biass at subsonic frequencies. I wonder if it could be simplified for use where accuracy was less important, perhaps because we are running Class A only and not optimal ClassAB ?

http://www.diyaudio.com/forums/solid-state/154396-new-idea-low-dissipation-class-amplifier.html

Post #1 "You read it here first"

Thanks Mooly - I had almost forgotten about that one, a good thread to re-read. Did you build something like this in the end ?

I am still looking (in this thread at least) for something more automated. Perhaps some fixed bias points controlled not by the music but by the temperature of the heatsink would actually be the solution to the current problem I'm trying to solve. I'm not looking for minimum power dissipation (regular EF in Class AB is perfect for that) but rather I'm looking to avoid cooking my amplifier because it's heatsinks are inadequate.

Perhaps a combination, one control input is the heatsink temperature, another control input is the volume control as you suggested - and if we want to be fancy we could give the user the option to select 'auto' which uses the signal as a control input.

 

[Mooly]

Thanks Mooly - I had almost forgotten about that one, a good thread to re-read. Did you build something like this in the end ?

I didn't, for two reasons.
First, truth be told I am more than happy with the sonics from my present design, and would probably pursue a more refined version based on that, and secondly, I'm not a huge fan of Class A for power stages tbh, both from the heat produced and the wastefulness of the approach.

That's just my personal reasons though

I thought the idea original because there is no processing of the signal, and amplifier requires no "knowledge" of the signal history... the bias is "fixed" but at a point that can accomodate all possible outputs for any given volume setting.

 

[forr]

Hi Mooly,
>I am more than happy with the sonics from my present design.

Can you tell a few words about it ?

 

[Ian Finch]

Class XD doesn't do it for me - it shifts the cross over region away from the zero signal level with the hope that low signals will see fewer cross overs. I'm far from convinced that this is a good approach. I wonder, for example, if larger signals will potentially see even more cross overs as the generally lower amplitude but higher frequency peaks ride on the back of the generally larger low frequency waveform. Without XD this low frequency waveform pushes the hf peaks away from the zero line and they see fewer cross overs, but XD moves the cross over point higher to meet them

As I read it, this method "sees" only less crossovers because an arbitrary level of about 7 V from signal zero is applied to limit dissipation to near AB levels. Is this not also the reality of of all those "high bias" high end designs? There doesn't seem to be any, other than dissipation, reason that the displacement can't be continued to the sensible limits of the output stage. Class A status then becomes no less valid than with sliding bias. The heat output and thermal cycling of the output devices will be significantly greater, but the aesthetic comfort of knowing that the amplifier is in class A conditions will be reinforced. The real point is, this one works.

BTW The significant drop in THD when XD is applied to a given OP stage has to prove something, I think. The fact that the benefit reduces beyond normal domestic listening levels is intentional and can be addressed as above.
If you are suspicious of "blamess" topology, you could probably apply this technique to any favourite AB design. For designs with intentional distortion or sonic "enhancements" in earlier stages, this would seem an ideal approach.

 

[Mooly]

Hi Mooly,
>I am more than happy with the sonics from my present design.

Can you tell a few words about it ?

Hi forr,
It's all here,
http://www.diyaudio.com/forums/solid-state/119151-my-mosfet-amplifier-designed-music.html

 

[Trevor White]

This isn't quite a coherent idea yet, but would be interested to hear if there are already some others who have thought along these lines. My apologies if this is 'old news' as my Search skills are not yet great.

The problem is how to avoid cross-over distortion in AB but still have good power efficiency.

I was thinking if we start with Class A but adjust the operating point (dc bias point ?) of the output device to suit the signal. A big signal means moving the operating point up so that negative swings don't clip. Lower signal means the operating point moves down to conserve power.

How I think that this can be done - the bias circuit needs to know in advance what the signal is going to look like so it can adjust the biass. Well, how about the biass circuit 'sees' the input signal before the power devices. You introduce a 'delay' to the signal between the input and the power stage. A 'delay line' so to speak. The non-delayed signal is fed through LP filter (below amplifier LF cut-off) and the output used to drive the biass. Perhaps this can be done symmetrically for a Class A push-pull output.

Now the output biass tracks the signal level in Class A but with greater power efficiency.

Am I crazy ???

Krell does this in its amplifiers. A few others do it as well.

Evolution Bias Control Systems
The new Evolution Bias system features 12 individual bias levels for optimum performance and efficiency. Evolution Bias constantly compares the audio signal at the input stage with impedance at the output stage, and adjusts the bias current accordingly. In addition, a proprietary bias monitor continuously measures and adjusts the output stage bias, eliminating the need for future adjustment or calibration.

 

[Edmond Stuart]

Adaptive bias control

[snip]
Thanks Edmond, nice to be introduced to the article by the author !!!
It looks ingenious and we'll want to see your next version as soon as possible.

Hi Bigun,
You're welcome!

The general approach looks relevant to the goal of variable bias Class-A - sampling the output stage bias current, integrating it and adjusting the output bias at subsonic frequencies. I wonder if it could be simplified for use where accuracy was less important, perhaps because we are running Class A only and not optimal Class-AB ?
[snip]

Class-A?! Weren't we talking about class-AB with an adaptive bias control?
Anyhow, my new circuit doesn't use an integrator. The bias level is continuously adjusted and with high speed, up to 1MHz. It's also simpler: two transistor arrays (THAT300/320) and two discrete trannies.
As for accuracy, we do need precision components. Remember, the voltage drop across the output emitter (or source) resistors is only few tens of mV. So Vos of all the transistors (that is, NPN to NPN respectively PNP to PNP) of the bias generator must be kept small.
BTW, compared to the LT1166 (which is supposed to do the same job), the distortion at 1kHz is about 25 times lower.

Regards,
Edmond.

 

[tiefbassuebertr]

Dynamic Bias - Active Bias - Automatic/Self biased - URL collection

about this thread you will find an Overview of various approaches:
http://www.diyaudio.com/forums/pass...-self-biased-overview-wanted.html#post2212036