Discrete Buffer Design

I'm in need of a nice discreet buffer to put inside a feebback loop that's a little more conservative with power than typical designs that rely heavily on Class A biasing to achieve high performance. I use Bryston amps and I'm very happy with their sound. They are also fairly efficient. So I figure, why not emulate their design for this buffer?

The problem is, I have very little experience designing discreet circuits. But I'm willing to dive right in nonetheless!

Bryston calls their output stage "quad complimentary" and a brief overview is given here: http://bryston.ca/chrismemo/index.html I really like the fact that the buffer can have some mild gain (say 2X) so the previous stage doesn't have to make the full output swing.

So to start, I just took that basic design and stuck the input stage of the Jung Super Buffer http://www.groupdiy.com/index.php?topic=2491 in front of it. All the component values and transistor types in the output stage are purely guesses. It's not really clear to me how and if the input stage biasing will properly bias the output stage. I'm shooting for an efficient design with fairly low quiescent dissipation ? just like the Bryston SST.



I'm aware that the complimentary pairs will need to be matched and I'm also willing to take that step.

Any guidance would be greatly appreciated!

Thomas

That sure is a complicated way to save power.

If gain is not essential, reduce the bias on the Jung diamond. This circuit can run into class AB very smoothly. Change the 221Ω resistors to 1K. Change the 22K-33K LED bias to 47K. Use just one pair of outputs, and make them the same type and lot as the inputs. Total current becomes 3mA. Ideally you would change all the 10Ω resistors to 27Ω for best AB action, but this may be a frill.

Output impedance will be 15Ω. This is a fundamental flaw with all such plans, lacking a real voltage-gain stage to take overall feedback. Zout is a function of output stage current, and low current means "high" Zout.

> Bryston calls their output stage "quad complimentary"

It is cute (and much older than Bryston). Gain is only approximate, linearity is flawed (Bryston's version less so than others, if done right), bias is uncertain. Bryston (et al) use it inside overall feedback to linearize it, and still faces "increase in the complexity of the biasing considerations".

How low is your power budget? How much output current do you need? How much output impedance can you stand? Some days, a 5532 is really the best answer.

I'd be discreetly complimentary of your chutzpah here, but these are treacherous waters. Note well Bryston's remarks about bias stability and the use of multiple sensors to control.

To begin with, look at where you can calculate and do so---see how far you can get. You knew enough to size the LEDs' resistor to fit the 24 volt rails compared to the Jung circuit (not that this is a particularly sensitive point in the circuit provided the LEDs get enough current to run at a reasonably low impedance). How did you go about selecting devices and resistor values for the Bryston portion? Also, what loads do you expect to drive with this setup?

Having said that the Bryston circuit is an intriguing albeit complex topology that bears investigation, and prefacing it with the part of the Jung circuit is not unreasonable. However, with the values shown the Bryston stage will hardly be conducting: the output of the upper half of Jung for example is one Vbe plus the voltage due to the current source above, roughly (1.6-Vbe)/220 (for a GaAsP red LED, the variety that is a decent match to the tempco of a Si B-E junction). So I get about 4.3mA through the 10 ohm R + the Vbe. The following input, the base of the BC337, has a roughly matching Vbe characteristic (a bit lower threshold, being an NPN) so one is left with some 40-odd mV across the 337's emitter R in parallel with the feedback R, or an emitter current of less than 600uA (actual Vbe's will change and modify this a bit but this is the general idea). So, the only drive is 337's collector current, essentially the same as the emitter current, developing to begin with a voltage across the upper 150 ohm R, or less than 90mV, not near enough to turn on the two series B-E junctions. So the output stage is mostly non-conducting at this point.

Without doing a through analysis I can't say for sure what the thermal performance will be like, once a suitable change in first-stage bias is effected to begin to get some current in the output. I would recommend decent heatsinks on all of the MJE parts though. My instincts say the 1 ohm emitter R's are a little too small.

When I have a chance I may throw this into sim to see what the phase-gain response is when biased in some reasonable area. Another question is what sort of circuit this is to be embedded in. Although pretty speedy-looking, there are a lot of taus accumulating from that many stages.

Good luck!

Correction: It's not as bad as I thought as I was neglecting the essentiallyunity-current transfer of the semi-matched Vbe's. The thermal concerns apply still I think. It still feels like the output stage would be a bit starved though---I will see in a bit.

Not discrete, but how about an LT1010? It runs on 24V bipolar rails fine.

Yep. As I recover from my gaffe and with the help of sammy sim, I get the output MJE devices as being off, essentially. The current in the 2nd stages of the Bryston section clock in at around 700uA, thus only providing about 100mV across any of the output Q B-E junctions.

To get the thing to begin to function the emitter R's in the Jung input Q's can be raised, although this also raises the output Z going forward. There is more work to do.

The more I look at this thing the trickier it gets. I have high respect for the folks at Bryston, and I am sure they have implemented the idea in a sound fashion. However, sims show severe high frequency peaking once the current begins to flow enough to work.

So, I suspect there are some additional compensation components not shown in their basic white paper. As well, their transistors may be better suited to the configuration.

> The more I look at this thing the trickier it gets.

Indeed. As soon as I saw that it was Dan Mayer's Tiger output stage plus added complication, I quit any hard analysis. And skimming the surface, it has Dan's faults "fixed" with extra gain, sort of a Super-Tiger. Dan's Tigers could bite, in analysis and in real life. Bryston has picked a hard path to walk.

> only providing about 100mV across any of the output Q B-E junctions.

My thumbs-only analysis suggests it needs 4 Vbe at the input, not 2 Vbe as the diamond-buffer front-end provides. This ignores some non-ignorable marginal factors. Designing a good Class AB stage is all about controlling milliVolts of bias under many-mV of thermal drift and many-Volts of signal. So you will bring it up to 3.9Vbe and still be underbiased, to 4.1 Vbe and it will suck current like electricity was free. If Jung's current sources are low tempco, then raising the input emitter resistors will bring it up to active current and then run-away.

If we really wanted this level of complexity: the Bryston is scaled for something like 100mA idle, 10 Amp peaks. If we are looking for 10mA peaks, we might start with all resistors 1,000 times higher. But that certainly does not look right. Or saying the Bryston drives 8Ω, we drive 600Ω, we might scale ~100 times higher. But certainly 1Ω output emitter resistors don't seem to fit 600Ω loads. Usually the bigger the better for stability and linearity. For efficiency, low is better. ~10% of load is a good target, hence 0.47Ω-1Ω resistors into 8Ω loads. And 30Ω-50Ω for 600Ω loads.

> I suspect there are some additional compensation components not shown

The Tiger, and indeed any quasi-complementary output's CE-CE side, wants to be violently unstable due to two CE stages in cascade with low/unity feedback. It can be very unpredictable in class AB: the wide swing of current causes wide swing in working Ft and device impedances, so it may idle clean and radiate CB waves at large output. The classic fix is some resistance in the input emitter, and a honking big cap C-B on the output device. That in effect sets output stage Ft low and constant (though at low bias currents the working Ft still falls).

However if the CE-CE stage can be tamed, it can give an output Z much less than the ~30Ω at 1mA that a CC stage gives open-loop. (OK, 13Ω in stacked push-pull, but probably double counting emitter resistors, so 30Ω/mA is about right.)

There is a 4-transistor push-pull gain stage that works sweet: Borbeley's Hafler DH-101 phono stage. But it only works Class A, and is current-feedback so it gets a little odd at low gain. To work predictably it has to have very high Beta in the second stage, since the feedback loop closes ports with current gain of just 1*Beta.

> the emitter R's in the Jung input Q's can be raised, although this also raises the output Z going forward.

A lot of interaction there. Indeed the Diamond Buffer wants those resistors small, OTOO 26mV. Even that is enough to complicate things... I think on paper these resistors should be zero, but in practice (with discrete parts) that can't work.

Without voltage gain, I still vote for a modified Jung. The LED bias current can be much less. Some LEDs won't like that; we can select LEDs or simply use a couple 1N914s. Straight diodes will drop idle current as temp rises, but that may not be a problem in a studio.

Another mod on the Jung is small resistors in the output collectors driving the B-E junction of a pair of Class-B boosters. If the Class-A output runs 1mA idle, you can touch 2mA peak, which will drive 10K loads to full voltage, and handle all low-level audio details in 600Ω, while still Class-A, and shift to A-B for high levels in 600Ω.

With a need for gain: it is hard to beat the better ICs for good device matching and insane complexity, thus potentially good performance for the supply current.

- - - - - - -
Hmmm... this hasty hack -



- seems to have point-oh THD, 0.01% at 1V peak in 600Ω and only a little higher at 20V peak. Temperature stability seems too good to be true (output stage current actually declines a bit at 80C). Output impedance is quite low for ~2.2mA idle current: 5Ω-6Ω.

Output impedance actually wanders all over the place with signal level: if I reverse things and drive the output through 600Ω I get 2% THD with a large dose of 5th and significant out to the 11th. The THD is low in the forward direction only because Zout is so very low compared to the load. From working with a basic Diamond, I know the odd-order distortion is in the nature of the beast. There is a magic value to null the 3rd, but then the 5th stands out. This can be nulled but that leaves a 7th, and you run out of places to correct. I am amazed that this very-B biasing is not worse than it is. Remember it gets 33mA peaks from 2.2mA idle.

Frequency response is absurd: ~100MHz. Minor peaking there: gain rises a bit above unity, a bad sign for a fancy cathode follower.

Yes I think the unity gain buffer is safer and saner, even if we sacrifice the last bit of output swing. The hybrid bryston-based is pretty good in this regard as expected, although near clipping the common-collector outputs go off (not that anyone would notice). It is not efficient however: to get things to work even reasonably, and with the strong warnings about thermal runaway, I've got the thing pulling over 40mA per rail at idle! And note that, as constituted the feedback dividers pull about 4 times the current of a 600 ohm load!

I'm sure there is an optimization that will make all of this better but one questions the value.

As far as going to the rails in alternative designs, there is always capacitive bootstrapping, although the presence of 'lytics may induce sneers from some quarters---and they don't work at d.c. There is also the need to avoid deep saturation of the devices if one is going to drive things into clipping. But then, as PRR said in another post, who clips?

[quote author="barefoot"]I'm in need of a nice discreet buffer to put inside a feebback loop that's a little more conservative with power than typical designs that rely heavily on Class A biasing to achieve high performance. [/quote]
How much output power do you need? Maybe my little SMD buffer will be enough? I have also a class A current source at the input for setting opamps in class A, see t13 and T14 at left.

The output transistors are BCP53 and 56 and they are similar to the famous BD139/140. 100 MHz and 1 A.

Schematic here http://www.sjostromaudio.com/hifi_files/qrv/qrv05r0schema_p1.pdf

peranders, that's a cute little board. I'm glad you explained the current load at the input---I jumped to the schematic before I read your entire post and was a mite puzzled. Similarly, I assume the resistors R2 and R3 are placeholders for a tweak on net input current.

And with your SMD layout the parasitic inductances shuld be nice and low too.

Brad

R2, R3 if you don't have any ground connection and you don't always have the input connected. This was just for flexibility.

barefoot, why don't you consider the TPA6120, all stuff is integrated?

Looks like a nice part. What are those rectangular blocks on the end of the board?

Those are jumpers if you mean J1 and J2 and they are for changing output impedance.

You can find all facts here:
http://home.swipnet.se/~w-50719/hifi/qrv07

Thanks. I was talking about what turn out to be your transformers---clear now from looking at your website.

Brad

Barefoot, have you come to any decision? You could also use a BUF634, an integrated Jung or diamond buffer. I have used BUF634 (the big IC) with bent legs for surface mount but you can get this also as 5-pin TO220 for hole mount.

Have you checked these ready made buffers?

http://www4.head-fi.org/forums/showpost.php?p=1295321&postcount=550

Hey, thanks for the great info everyone! I've been extremely busy trying to get some speaker orders out the door, so I haven't had time to concentrate on this lately.

One quick thought regarding the Jung biasing scheme. Would there be any advantage in replacing the 33K resistor that drives the LEDs with an NTC thermistor that is thermally coupled to the output transistor heat sink? Since the thermistor resistance increases with temperature the LED drive current will drop, lowering the forward voltage and subsequently lowering the bias current. Seems like a simple way to ensure temperature stability, no? Is this a common technique?

Thanks!

Thomas

Sorry, got it backwards. Should be a PTC thermistor.