Non-complementary class AB linearized amp

[tiefbassuebertr]

Originally Posted by pauljguy

Hi,
I designed, built, tested and listened to this amplifier. Its circuitry is a bit different, but I find it sounds very good (very difficult to tell it apart from a tweaked Linn amp using blind, level-matched AB switching), and it does not generate the excessive heat from a class A design.

Its design is based on a non-complementary BJT circuit. It runs
class AB, and uses current amplifying stages instead of voltage
amplifying stages (VAS) to drive the output transistors. The output
stransistors are in a totem-pole configuration, but run common
emitter rather than the usual common collector (emitter follower)
layout. The front end of the amplifier uses a differential compound
pair (Sziklai).

The first transistor after the differential pair is biased such
that instead of cutting off sharply for class AB, it has an
extended non-linear cutoff to minimize the impact of "crossover"
distortion. This base circuit also compensates
for the inherent non-linearity of the "gm doubling" of the output
transistors when run class AB. The result is that over the entire
operating range, the openloop gain of the amp doesn't vary more than
5-10%. Closing the feedback loop with about 60db of loopgain results
low distortion.

Paul G.

Is there not a great similarity to the circuits about follow URL's
Siliconix & Siemens NMOS amplifiers
http://www.amplimos.it/images/2sk77%...MAHA%20B-1.gif
http://www.amplimos.it/images/2SK1056AMP.bmp
http://peufeu.free.fr/audio/schemas/Kaneda_Mosfet.jpg
‹à“cŽ®_UHC_MOS-FET_Power_Amp
(except the fact, that there are MOSFETs in the output stage in use instead BjT) ??

more URLs you will find by 1) (by post #1) about
Only N-Channel MOSFETs (NMOS); better Audio from non complements by Audio Power?

[pauljguy]

Originally Posted by tiefbassuebertr

Is there not a great similarity to the circuits about follow URL's
Siliconix & Siemens NMOS amplifiers
http://www.amplimos.it/images/2sk77%...MAHA%20B-1.gif
http://www.amplimos.it/images/2SK1056AMP.bmp
http://peufeu.free.fr/audio/schemas/Kaneda_Mosfet.jpg
‹à“cŽ®_UHC_MOS-FET_Power_Amp
(except the fact, that there are MOSFETs in the output stage in use instead BjT) ??

more URLs you will find by 1) (by post #1) about
Only N-Channel MOSFETs (NMOS); better Audio from non complements by Audio Power?

I always enjoy following down your many references to posted circuits. The closest one is the Siemens/Siliconix circuit although it uses MOSFETs. I did use MOSFETs in one of the earlier versions of the circuit I posted, but I could not get the low distortion that the BJT gave me, especially after adding curvature to the transfer characteristics (using Q100,Q200). Looking at the circuit values, I believe all the designs you posted do not deliberately use non-linearity to compensate for gm-doubling and get a "soft" transfer between class A to AB.

I would not be at all surprised if the technique I use is already patented - I have made up circuits like class D amplifiers, and discovered what I thought was new, is really a well beaten path.

The topology I used isn't really all that new. The only thing I haven't seen done before is the trick using the non-linear Vbe-Ic relation to reduce distortion. I tried using the same trick with MOSFET output devices, but I could not get the same low distortion, unless I used ton's of feedback (with the usual HF stability problems). It's actually quite easy to convert my circuit to MOSFET, just run the cascodes into a resistor between gate and source, and adjust biases.

My circuit doesn't really show itself off as being all that unusual. However - I'm quite pleased with its performance. I'll probably be tearing it apart in a few weeks time to try something else! Right now it's on a "special" breadboard with nice big heatsinks and a hefty power supply. The circuit you see is about the hundredth iteration using LTSpice, and about the fifth physically built since this spring.

Anyhow... thanks for your references!

By the way, I rarely see amplifiers that use inverting input feedback, where the first stage has signal and feedback coming together on the same node. This gives some problems with input impedance, but I can't see any other disadvantage. Is it a technique frowned-upon by audiophiles?

Paul G.

[Elvee]

Originally Posted by pauljguy

The topology I used isn't really all that new. The only thing I haven't seen done before is the trick using the non-linear Vbe-Ic relation to reduce distortion. .

The Circlophone uses a similar trick, at Q9/Q11's level, with the 56Ω B-E resistors driven in current mode by the VAS:



But the Circlophone does not only rely on that mechanism to shape the crossover: subtraction of exponentials cannot yield a perfectly linear function, and the "best fit" region is very narrow.
These two issues are addressed by the common-mode servo: it keeps the open-loop transfer function at the optimum bias point and then straightens it further in real time.

Our amplifiers have similar topologies, the stages are arranged somewhat differently though.
I have found that a cascode in front of the OP stage brings no benefit.
This might look counter-intuitive, but I recommend you try to bypass them, you might be surprised.

[tiefbassuebertr]

Originally Posted by pauljguy

Anyhow... thanks for your references!
By the way, I rarely see amplifiers that use inverting input feedback, where the first stage has signal and feedback coming together on the same node. This gives some problems with input impedance, but I can't see any other disadvantage. Is it a technique frowned-upon by audiophiles?
Paul G.

Yes, you are right. This models I know:
1) AVM
for basic schematic go to post #4 about
Has anybody Burmester or AVM schematics?
more URLs:
stereoplay 1988 Monoblock from Günter Mania
Wer kennt sich mit AVM Verstärkern aus?, Verstärker/Receiver - HIFI-FORUM
anywhere on this forum I have post the simulation results of this topology
2) the first amp designs (e. g. 101, 105) from Bob Stuart & Allen Boothroyd (Meridian Audio) uses the inverted modes
dhs Meridian 101 Control Unit
Meridian 101 experts: Help needed! - pink fish media
MERIDIAN INTERGRATED AMPLIFIER MCA-1 BOOTHROYD AND STUART LTD FAULTY | eBay
Beyond The Wall Of Sleep: Meridian Boothroyd Stuart 101/105 Schematics & Fault Finding Guide
There was also an paper from Bob Stuart & Allen Boothroyd, where this was justified in detail, but I can't find it.
3) Musical Fidelity A1-X (only preamp line section and RIAA head amp - go to the attachement)

BTW - low input impedance between 1K and 5K is an advantage and not a disadvantage (at least by modern op amp ICs and discrete line amp solutions). Parasitic lead (cable) effects are less critical. Only in cases of tube preamp line outputs with output impedance between 5K and 10K aren't useful for such input impedances.

[pauljguy]

Originally Posted by Elvee

The Circlophone uses a similar trick, at Q9/Q11's level, with the 56Ω B-E resistors driven in current mode by the VAS:


But the Circlophone does not only rely on that mechanism to shape the crossover: subtraction of exponentials cannot yield a perfectly linear function, and the "best fit" region is very narrow.
These two issues are addressed by the common-mode servo: it keeps the open-loop transfer function at the optimum bias point and then straightens it further in real time.

Our amplifiers have similar topologies, the stages are arranged somewhat differently though.
I have found that a cascode in front of the OP stage brings no benefit.
This might look counter-intuitive, but I recommend you try to bypass them, you might be surprised.

You are correct in that you cannot get a perfectly linear transfer function using an exponential curve (by appropriately biased BJT stage), but it did allow me to obtain a 5-10% variation in openloop gain as compared to the more than 2:1 variation in gain from "gm doubling" and gain droop. That gain variation was over the full amplitude range of the amplifier without feedback. Remember, this is a class AB circuit. It runs a LOT cooler than when I bias it for class A.

Even class A circuits have considerable gain variation when run openloop. Optimizing the circuit for class A gave a bit less distortion. The linearizing circuit would compensate for the gain change at higher collector currents (driver & output). I had a variation of this circuit that used MOSFETs, but they were not as linear as BJTs.

Using LTSpice (or equivalent) plot the derivative of Vout/Vin vs. Vin with no feedback. You may need to do tricks with large inductors to allow for DC servoing to take place. It is enlightening to see how linear (or not) the amplifier is.

With a fairly linear amp, once feedback is applied, it is possible to have quite low levels of distortion, without the stability issues you might have if you use a LOT of feedback with a more less linear output.

The Cascode circuit was quite necessary to maintain balance and reduce the effects of the output signal getting into the previous stages. The even harmonic distortion is much worsened without the cascode. After the differential pair, the circuit is very closely balanced, although it doesn't appear that way.

Paul

[kenpeter]

Imagine your lower output transistor going into thermal runaway.
This overcurrent situation cannot be sensed, because you got an
LED voltage drop on the collector. This LED prevents your sense
transistor from functioning near the negative rail.

[kenpeter]

A summed resistor feedback (like yours) will shape linear class A
if you get rid the lowpass cap and let quadrature feedback have
full-time control of the common mode bias.

You can sum a pair of Schottky diodes, and get Square law curved
Class A. I seem to be the only one promoting this class, as the
actual crossing current is not well defined. Only that it will be less
than linear A. And varies toward a safe tempco if the sensing BJT
is on the hot output sink, and the diodes are in a cooler location.
The circuit remains a simple sum across VBE, just like yours.

Or you can do as Elvee, and use a logic function (like NAND) to
define a near ideal class AB. This logic is not saturated in the
usual sense of on vs off, but jammed in the linear region upon
the logic threshold. You cannot get near B crossing shapes from
the simple sum of linear resistors and curved diodes, but you
can with linearized logic. And thats really the next step up...

I would not rely upon the curves of hot output devices to define
anything nor stay matched to another VBE. You are asking for
thermal runaway , or a dead spot in the crossing ... You have
sensing resistors, I suggest you use them to full advantage...

[pauljguy]

Originally Posted by kenpeter

Imagine your lower output transistor going into thermal runaway.
This overcurrent situation cannot be sensed, because you got an
LED voltage drop on the collector. This LED prevents your sense
transistor from functioning near the negative rail.

You must have a good eye for detail! Yep, the circuit could be a problem. I'll get rid of the LED on the collector (it's the overload indicator). If I need an overload indicator, it would be better to use a separate transistor and LED to do the job. I have current sensors (red/yellow/green LED) in the circuit anyhow, so getting rid of the overload indicator won't be a loss. I have run the amp flat out into a short circuit for about 1/2 hr, it didn't seem to mind it, so the overload indicator is a bit redundant.

Thanks for the tip.

Paul

[pauljguy]

Originally Posted by kenpeter

A summed resistor feedback (like yours) will shape linear class A
if you get rid the lowpass cap and let quadrature feedback have
full-time control of the common mode bias.

You can sum a pair of Schottky diodes, and get Square law curved
Class A. I seem to be the only one promoting this class, as the
actual crossing current is not well defined. Only that it will be less
than linear A. And varies toward a safe tempco if the sensing BJT
is on the hot output sink, and the diodes are in a cooler location.
The circuit remains a simple sum across VBE, just like yours.

Or you can do as Elvee, and use a logic function (like NAND) to
define a near ideal class AB. This logic is not saturated in the
usual sense of on vs off, but jammed in the linear region upon
the logic threshold. You cannot get near B crossing shapes from
the simple sum of linear resistors and curved diodes, but you
can with linearized logic. And thats really the next step up...

I would not rely upon the curves of hot output devices to define
anything nor stay matched to another VBE. You are asking for
thermal runaway , or a dead spot in the crossing ... You have
sensing resistors, I suggest you use them to full advantage...

I had a bit of trouble with your terminology.... I assume the "quadrature" feedback and the lowpass cap consist of C10, R21, Q8. In my design, Q8 is normally cut off until about 2 amps flow thru R50,R51 (or 4 amps into the load). The current drawn thru Q8 diminishes the current source for the long-tailed pair, and drops the current thru the power transistors.

What you propose ( I assume) is to vary the current source for the long-tailed pair dynamically. Whereas I vary the linearity of Q100, Q200 to give a "remote cutoff", you add/subtract a common signal to both sides of the driver circuit to control the switch-off of the power transistors. The push-pull action will largely cancel this common signal. Both schemes should spread out the transition between power transistor on-off.

My primary goal was to reduce the "gm doubling" effect. I could spread that effect out, but I chose a circuit that would "bend" the open loop transfer function to minimize both the gm doubling and gain variations from changing collector currents. I assume that's where your squaring circuitry with the schottky diodes comes in. You are modulating the gain with a common mode signal, in such a way to straighten out the overall response. Perhaps it would be better described as adding a compensatory polynomial term in the transfer function. How did you tailor the circuit for best openloop linearity?

I wanted a very simple way to linearize the overall circuit, hence I avoided any extra stuff like logic. I wasn't concerned with getting precise class B, just an operating point that gives class A for low volume, and a smooth transistion to cutoff to avoid switching effects. If it gets a bit warm - no problem. Full class A is just way too hot. As it is, the circuit is too damn simple - it takes quite a bit of analysis to see that there is any linearization going on. I originally did this years ago with a single-ended tube amp. By varying the bias of the driver stage (6SN7), I could reduce the harmonics by bending the transfer functions so they more or less cancelled the non-linearity. The linearization was not at all apparent when looking at the schematic.

My thermal compensation of the output transistors is quite crude - the diodes in the current source for the long tailed pair sit on the heatsink. One of the nice things about LTSpice is that you can use the step function for temperature, and more or less compensate for thermal effects. The circuit has pretty reasonable current stability with temperature in real life.

The linearization is not perfect, nor is it consistent with temperature. The distortion figures I measured in the real-life circuit were worst-case, ranging from cold at turn-on, and hot after running at full output. The distortion does vary a bit with bias current, especially if I aim for bias currents below 100ma. I chose a bias current of 400ma, it doesn't give me the lowest possible distortion, but it is fairly consistent with bias current, temperature, and signal level. It still has better distortion than an unlinearized full class A amp with similiar components.

Paul