Krell Class A/B power amps, do 'anticipator circuits' work?


My thread 2

In my hi-low speaker sensitivity thread, 8th-note mentions his Krell 300S power amps.
He shares my passion in this department.  I have run Krell Reference KRS200s, upgraded to 400wpc since 1990.
Like all Krells from the first decade of production these are 'pure class  A' all the way up.  The 300S runs class A/B

At this point, the Stereophile review of the 300S by Thomas J. Norton is very instructive and as a subscriber for many decades I acknowledge this source:
          https://www.stereophile.com/audaciousaudio/krell_ksa-300s_power_amplifier/index.html

[Isn't the Rikki Lee Jones hard to find 'Girl at Her Volcano' 10 inch just wonderful?  And the huge drum shots just the stuff we speak of below.]

By the early 1990s Krell felt under pressure from the climate change lobby to reduce the huge power consumption of full Class A operation.  The KRS200s draw more than 1kW per side.  So they abandoned it and moved to Class A/B which applies a sliding bias according to the exigencies of the music signal.

But Krell still highly valued the benefit of instantaneous power availability to cope with peaks in the musical output.  So they created 'anticipator circuits' that they said continually analyse the input signal and instantaneously increase the bias to one of four successively higher levels to accommodate peaks.

I never believed this is possible as the reaction cannot be fast enough to increase the bias before the moment is past.
How can it be implemented the moment the skin of that big bass drum is deflected by the first fraction of a millimeter.  Mr Norton covers the same issue in his review.  It seemed to me the only way to do it is to buffer the signal for at least as long as the amplifier takes to react to the input (this would of course have the massive downside of subjecting LPs to the clock and dither problems of digital).  Or perhaps beforehand to create a log of the programme that would be fed to the amplifier and applied to adjust the biasing in advance.

My KRS200s were in for full refurb in the early 2000s.  Since I was considering changing to newer Krells, I took the opportunity to make these points to the engineer doing the work.  He was not able to explain to me how it is possible but said there is no buffering.

So I have always considered the 'anticipator circuits' to be a pig in a poke or, to put it more politely, advertising flannel.

I note that in nearly 30 years no other amplifier manufacturer has sought to make such a claim.

So I retain my KRS200s as keepers; relics of a past age now gone forever in a dull world of digital amps and Class D.

I find them to be superbly dynamic and generally of extremely high SQ, if perhaps rather warm during the summer as a/c would interfere with the music.

My questions are:
Without buffering how can it be done?  Was/is Krell speaking truth?
Would buffering create the clock and dither problems I anticipate?
Has any designer tried buffering and what was the outcome?

I do rather like the concept of applying advance programme logging.  It would be a bit burdensome but, subject to the step changes of bias not being discernable, allows almost the full benefits of continuous Class A operation while keeping the Greens at bay.






C85b9041 52a0 4fa7 ac78 d902149a2d82clearthinker
As a total guess a tracking peak-hold with delay circuit that drives the bias stage. The bias may rise with the waveform and then stays there slowly decaying down if the music quiets. Won’t be instantaneous but for slow moving signals it may me good enough and for higher frequencies you need a few waveforms to pick up characteristics.
Thanks for the response.
Yes a2d, at the outset Krell said that once raised the higher bias will be held 15 to 20 seconds before it is dropped back if there is no further hi-level signal.
But that doesn't answer how the first rise is triggered in time and what happens if there is a sudden new peak after 25 seconds.
One might postulate a single heavy stroke on a large bass drum every 25 seconds.
And Krell informed me there is no buffer=delay circuit.
The bias cannot be adjusted from the signal alone. Any circuit that slides the bias has to take into account the load presented by the speaker, for that is the only thing that draws current through the output devices. The buffer would, in theory, apply to only one type of speaker impedance vs frequency because it is a known that can be the reference of the buffering scheme. If the buffer is set to maintain the bias voltage to equal the voltage across the emitter resistors at, say, a 6 ohm load @ 1khz and the amplifier is hooked up to a speaker that presents 2.5 ohms at that frequency, what does the amplifier do? What about the other frequencies? In other words, the buffer can never tell what the load presented to the amplifier is going to be and the amplifier will do what it would have done without the buffer at low impedance -- switch to class a/b or b.

The problem with class A amplifiers is that the bias voltage has to be set so the quiescent current does not overheat the devices. However, when the load impedance drops, the higher output current causes a higher voltage drop across the emitter resistors. If this voltage drop exceeds the bias voltage, one of the complementary pair transistor (current sink) shuts off and the amp goes class B and shoots out a lot of distortion from the hard shut down. In order to keep class A at the lower impedance, the bias voltage has to be increased above the emitter voltage at that impedance. Now there is a higher quiescent current and more heat sinking is required, not to mention the increase in power supply filtering in order to keep the input/driver stages from being affected. So the trick to sliding bias is to not anticipate the load current but to compare it to a reference, a reference that allows for a manageable Q current.

I don’t know what Krell uses, but sliding bias can easily be done by monitoring the voltage across the emitter resistor by setting up a reference voltage with diodes in conjunction with resistors. The output across the resistor is compared with the reference voltage and if there is a high current draw, the resulting voltage drop is sensed by the diodes which then open a transistor whose collector steals current from the output transistor base, removing some of the quiescent current gradually, allowing the conducting transistors to operate in class a instead of being reversed biased, shutting down and operating in class B.
You will note I said "waveform", not specifying which one. You are assuming emitter degeneration resistors, but it would be just as easy to pull the value from a high side or low side current sense on the power supply output.  I could see a hybrid that also takes into account output voltage since distortion is around the crossover point.
It is possible to linearize, with the feedback, any amount of output non-linearities.  The difference between class A and AB is in the amount of feedback (tenfold).  Increasing bias without lowering negative feedback won't achieve anything.



We used to build quite a lot of very high biased push pull Class-A’s 100w> with around 150w in B

My old boss (rip) Steven Deratz of Deratz Electronics back in the 80-90’s in Brookvale Sydney, was the first to have a patent on a "Sliding Bias" type circuit, just to save on the amount of heatsinking were had to use with the normal Class-A we built. My personal A amp was a beast 3 man lift self contained water cooled transistor jacket, pump, radiator/fan.

Anyway his patent sliding class-A worked fine, but the problem was the initial leading edge of a larger transient was in B after which all else stayed in A until the bias rolled itself back, or if there was a second leading edge transient larger than the first, it too was B.

We could never make it sound as good and just high biased Class-A, it just wasn’t fast enough to keep up with music’s varying transients. And if you held the A up for a longer period, you had the same heat problem as normal high biasing.

He let the patent lapse and then Technics bought out something called Quarter-A and AA, and then Krell with "Plateau Bias"

Moto? Nothing beats the real thing.

Cheers George
Thanks George.

Quite.  That's rather what I thought.  Your second para describes exactly the issue I am highlighting.
Douglas Self stated that increasing bias often leads to increased distortions. He also measured it. The reason for that is changing transconductance. Transconductance is output current to driving voltage gain, that is rapidly changing when two transistor current (biased region) changes into one transistor current (outside of biased region). He stated:

It is not generally appreciated that moving into Class-AB, by increasing the quiescent current, does NOT simply trade efficiency for linearity. If the output power is above the level at which Class-A operation can be sustained, THD increases as the bias advances into AB operation. This is due to so-called "gm-doubling" (ie the voltage-gain increase caused by both devices conducting simultaneously in the centre of the output-voltage range, in the Class-A region) putting edges into the distortion residual that generate high-order harmonics much as under-biasing does. This vital fact is little known, presumably because gm-doubling distortion is at a relatively low level and is obscured in most amplifiers by other distortions.
You can find it here (5.3): http://www.douglas-self.com/ampins/dipa/dipa.htm

The main reason of going into class A is to reduce feedback. Once feedback is set damage is already done (TIM distortions) and playing with bias won’t change it. Perhaps there is a way of compensating for the change in transconductance (gm doubling) or adjusting feedback dynamically but it is very complicated. My Benchmark AHB2, class AB amplifier, uses separate "error output stage" to avoid recursive feedback and TIM distortions (AAA patent).

Here is a link to Dan D'Agostino's patent for plateau bias. This describes, in general terms, how his circuit works. 

https://patents.google.com/patent/US5331291


Thank you Jaytor, that is extremely helpful.  I had not read it previously.  It may be seen to settle the question I put.

I have read the patent.  It relates to a methodology of adjusting the bias current in response to measured changes in current flowing in the load.
At no point does the patent claim the bias adjustment can be triggered so fast that when a sudden high current event occurs, that event can be presented in either pure Class A or employing a higher bias current than that being employed immediately prior to the occurrence of the event.

The system can adjust the bias in quite complex ways but it cannot provide a raised bias current in time to catch sudden dynamic increases in the programme.

Such effect is advantageous to SQ and can only be achieved by running full-time in Class A.

I am satisfied the natural order is restored.  The application should not have been described by some as 'anticipator circuits'.  That would entail time travel.
Yeah, I think it has more to do with setting the bias based on the current requirements (as in amps) of the current situation (e.g. listening level and speaker load requirements). For many speakers and most listening conditions, a fully biased class A amp is just wasting power, but it's nice to have it in reserve. 

I owned a pair of Krell MDA-300 monoblocks for a while that were room heaters. I think these were made in the days when Krell class A amps were fully biased. I replaced these with an FPB-300 (which used the plateau biasing technology) which I owned for more than 20 years. There was a couple years separating my ownership of these amps, as well as different speakers, so I never really got a chance to directly compare them, but they were both excellent amps.

I don't think I would hesitate (at least from an SQ perepective) to buy one of the newer D'Agostino designed Krells (not sure about the ones after he left). I finally got rid of mine after it failed for a second time with bad caps and it was more expensive to recap it then it was worth. In hind-sight, I wish I had kept it and done the work myself since I've recently gotten into doing a lot of DIY audio electronics.


Yeah.  When you've got all that heat the caps don't last too long.
I had my KRS200s totally rebuilt 10 or 15 years ago.  Perhaps they will need another cap replacement before I'm done with them.  I'm keeping them, they are the daddy of them all.