Discrete (dynamically biased?) buffer, questions

[bas]

Hi Head-fiers, you don't know me, but I know you.

For a while now, I've been itching to build a discrete buffer module to play around with. Looking through some of the information here and elsewhere, a dynamically biased buffer may be a fun project. Also, since it needs to reside outside the feedback loop due to instability issues (IIRC), it should be easy to add to my existing amps. <actually, class A looks more attractive to me now>


I've got questions...

0. Does a dynamically biased buffer sound better than a comparable static bias buffer (in blind A/B)?
<My current opinion: probably not. See below.>

1. Is dynamic bias a good idea in the first place, i.e., is this kind of buffer stable when it's not in the feedback loop or is it still a hit-and-miss?

2. What's the effect of having a buffer inside or outside the feedback loop?
<Part of answer in my second post, I think. In loop: cancels DC offset.>

3. What is a good/great design for a discrete buffer? In particular I would be grateful if somebody pointed me to schematics!
If at all possible, I'd like to avoid using kits or order pcb's since for me, the soldering part is work, but figuring out a decent layout is fun.


Thanks for your help!

Ben

<Edits in pointy brackets.>

 

[BlazerFRS]

The difference between dynamically baised and not is clearly over my head; but I too have wanted to build a discrete buffer for awile.

I've heard good things about this one:

http://www.headphoneamp.co.kr/bbs/vi...esc=asc&no=124

unfotunately most of the site is in a language my computer can't read, but the diagrams have plenty of info to get it done.

 

[bas]

Thanks for the link, Blazer! That looks very useful. Somehow, Sijosae's soldering makes me feel inadequate... dunno why really...

Unfortunately, Google's translation is probably more confusing than the ????? ??? ????? ??? ??? placeholders on the original page. Furtunately though, Sijosae linked to Walt Jung's article on the subject. I had seen that one before, but since my first encounter with Jung's work left me with the impression that his writing is about as accessible as one of those fancy magnetic-key locks [size=xx-small](the kind that can't be picked with normal tools, seriously, why are people still using the serrated crap?)[/size] when you lost the darn key, I hadn't read it carefully before.

As it happens, my initial condemnation was quite mistaken. It's an excellent read and also answers part of question 2: a buffer creates a DC offset; when the buffer is in the feedback loop, this offset is cancelled and thus never reaches the headphones. Since Walt notes that the offset for his buffer is about 20-30 mV, placing it inside the loop is obviously preferable. That makes dynamically biased buffers much less desirable, *assuming* that they have similarly sized offsets.

While searching around I also came across Per-Anders' page. It has very useful design notes, too. Good man!

 

[amb]

When talking about dynamic biasing, one has to understand its raison d'etre. Prior to the invention of dynamic biasing, there was static biasing. A push-pull audio amplifier either operates in class A, class B, or class AB. Most readers here are no doubt familiar with these but the following are articles that go into some depth explaining the differences:

http://www.tpub.com/neets/book7/25e.htm
http://sound.westhost.com/class-a.htm
http://www.music-recording.com/index.cfm?id=1824

It should be obvious that Class A has the benefit of the best linearity and lack of crossover distortion (among other advantages), at the cost of poor efficiency. Class B is just the opposite, having good efficiency but severe crossover distortion. Class AB eliminates the crossover distortion and retains most of the efficiency of Class B, but nevertheless the transistors will cut off at some point and is thus not as "clean" as class A.

The efficiency issue is a problem for high powered speaker amplifiers because a class A amp must pass a lot of current and dissipate a large amount of heat even when it's idling. This means that the power supply, output stage, heat sinking, etc., must all be built to match. A 40W per channel class A amplifier may need to be physically as large as a 400W+ class AB amp. Aside from the high heat and high energy bill factors, a suitably designed and built class A amp will be much more expensive than a class AB amp of comparable power output.

In the 1970s, engineers went to work to design dynamic biasing circuits in an effort to achieve class A performance characteristics without its inefficiency. The first such design to appear was from Nelson Pass under the Threshold brand. Lots of other similar concepts followed from other companies. Basically, instead of applying a high, steady bias to the output transistors as in true class A, these designs vary the amount of bias based on the actual output. The output transistors do not turn off at any time as in true class A, but at idling conditions the bias is scaled back to keep the power dissipation low.

Whether dynamic biasing actually achieves equal performance to true class A is open to debate. There is no doubt, however, that it solves the efficiency problem.

With all that said, let's look at headphone amps rather than high power speaker amps. Since headphone amps are scaled way down in terms of required output power, true class A designs do not carry quite the same penalty. The heat, power and size requirements are still higher than class AB, but quite manageable. True class A is still not battery friendly, but for wall-powered amps, applying dynamic biasing, in a sense, is a solution looking for a problem.

There are some on these forums who would swear to a sonic superiority of dynamic biasing. Technically, I cannot see how it could be any better than true class A.

 

[Glassman]

frankly, the sonic superiority of the dynamically biased circuits used around here is not entirely in the fact that they are dynamically biased, rather it has a cute side effect of introducing cascode, which clearly is technically superior..

btw. bas, YGPM

 

[bas]

Thank you so much, amb, for that *very* informative post! That really cleared things up for me.

Also a big thanks for dropping Mr. Pass' name. Not only are his diy pages pretty cool, he's also got a plethora of articles explaining his amps in some detail. Very nice. Regarding dynamic biasing, I digged up an old comment of his:

"I authored the first patent on the dynamically biased Class A amplifier in 1974, however I have not used the technique for the last 15 years. The reason is that I found the quality of sound associated with an efficient Class A operating mode inferior in depth and less liquid at high frequencies, simply because it operates at reduced bias at low levels. Given the plethora of cool running "Class A" amplifiers on the market, you might say I opened a Pandora's box."

Together with amb's explanation, it would seem that dynamic bias is at best as good as class A and probably worse. Nelson's poison of choice today looks to be single-ended class A.

Quote:

Originally Posted by Glassman frankly, the sonic superiority of the dynamically biased circuits used around here is not entirely in the fact that they are dynamically biased, rather it has a cute side effect of introducing cascode, which clearly is technically superior..

Hm, I am inclined to think that dynamic bias does *nothing* to produce sonic advantages. However, your comment re cascoding is intriguing. If I understand it correctly, cascoding two transistors reduces voltage-induced nonlinearities, whereas class A reduces current-induced nonlinearities. So why should cascoding be more beneficial than class A?

 

[amb]

Quote:

Originally Posted by bas Nelson's poison of choice today looks to be single-ended class A.

Yes, and ironically, the single-ended class A topology is the least efficient mode of amplification.

Quote:

Hm, I am inclined to think that dynamic bias does *nothing* to produce sonic advantages. However, your comment re cascoding is intriguing. If I understand it correctly, cascoding two transistors reduces voltage-induced nonlinearities, whereas class A reduces current-induced nonlinearities. So why should cascoding be more beneficial than class A?

cascoding and class A operation are two different elements to amplifier design that deal with different issues. You can have an amp with cascode but not class A, or a class A amp without cascode, or an amp with both. Cascoding is a technique that not only linearizes the transfer curve characteristics of transistors, it also increases usable bandwidth by avoiding the Miller and Early effects.

Nelson Pass wrote a good article on cascoding.

Also, the excellent textbook The Art Of Electronics by Horowitz and Hill has a section that deals with cascoding.

I don't think it would be possible to blanketly quantify whether class A or cascoding provides more sonic benefit, since there are so many variables.

 

[bas]

I've read his article on cascode amps. At this rate, I'll probably make it through the entire Nelson Pass library within the week.

AMB, you've answered my next question: cascode and class A aren't mutually exclusive. Wouldn't that be the best solution? I must admit that I'm in way over my head at this point. OK, that much is probably painfully obvious.

I'll have to pick up AoE at the library... But what I really need is a schematic with ultra-generous notes.

Like: these are the input transistors - yeah, I figured you didn't know that already - and those are the output transistors, dummy. Hey, now pay attention and look at those constant current sources, oh, and don't forget the cascoding... right there!

 

[ppl]

Dynamic Bias by definition is the Technique for providing the implementation of dynamically biased circuits. The Method remains the same regardless of the technology it is applied to In Dynamically Biased circuits, bias currents are varied according to signal amplitude. Benefits include reduced power dissipation, reduced noise, and increased dynamic range. The techniques can be employed in various types of circuits such as, for example, amplifiers, log-domain circuits, and filters. This paragraph is the definition of the (USPTO) for describing a dynamic Bias generator applicable to any type of Linear Amplifier from Audio power amplifiers to hand held wireless devices where power consumption and linearity are of primary importance.

An amplifier which combines the low quiescent current requirements of a Class B transistor amplifier with the minimal distortion qualities of a highly Biased Class AB amplifier is the goal of a Dynamic bias design as applied to Audio amplifier output stages.

Dynamical biasing ~ Simulation Results show the DC Current and Distortion under different bias conditions. The conventional Class A and class AB Constant bias’s methods obtain average efficiency: 2.4% where as a dynamic bias’s average efficiency is over 5.8% representing a 140% improvement in efficiency, while obtaining the same performance as the conventional constant bias design.

Malcolm Halksford has also done some great work with dynamically biased Audio Output stages see (DISTORTION CORRECTION IN AUDIO POWER AMPLIFIER) http://www.essex.ac.uk/ese/research/...ction%20PA.pdf
If you look at fig.4 and fig. 5 on page 3 you see two output stages showing marked similarity to Mr. Passes. I can truly understand ones frustration with Dynamic Biasing because of stability issues. These circuits can be a real headache to get properly working requiring lots of R and D time on all possible variations of the intended use. These stability issues are the primary reason team PPA went with my standard Class Ab biased Diamond buffer as opposed to the Dyno~Bias version. I could get the Dyno~Bias stable in that Amp and there are plenty of Dyno~Biased Buffers still keeping there PPA owners happy, However in the interest of less hassle for the Builder the Dyno~Bias circuit was not used in PPA V2 however it still uses an IMHO novel bias circuit in its own right. See http://www6.head-fi.org/forums/showp...9&postcount=93


While Dynamic Biasing may have been given up on log ago by some others are still actively perusing the virtues of this technique as evidenced by the almost continuous applications for patents on one form or another of Dynamic Biasing such is the case when Analog devices needed to get lots of bandwidth with small Current consumption in there new Fast fet series of op amps of which this patent describes, http://free.patentfetcher.com/Patent...tch&PN=6262633

Also see the Elantec patent on the now classic EL-20XX series of Buffers http://free.patentfetcher.com/Patent...tch&PN=4833424 Buffers


In conclusion it is my position that Dynamically Biased output stages can obtain the Holy grail of Class A sound reproduction without the heat or power consumption of Class A and thus ideal for portable applications. That unlike Class A the output transistors are never switched off even under abusive load and output conditions since one of the side effects of dynamic biasing is having the current of the Conducting output Transistor dumped into the non-conducting output transistor thus preventing turn off of ether device.

 

[bas]

Thanks a lot for the information, ppl! Guess I'll just have to try them all.