Compound and Darlington Buffers

[N9]

I've been playing around with various BJT configurations for a while, and have wandered into something I don't quite grok.

I'm comparing the performance of Darlington and compound pairs as emitter followers, and, in simulation at least, they seem practically identical.

Is there a notable difference in headroom or other audible characteristics between compound and Darlington configurations? I feel like I'm missing something...

Here are the two circuits I'm looking at:

[R.G.]

They are quite similar in most characteristics. The Darlington has a touch less headroom, as it will not saturate to as low a voltage as the Sziklai pair, but we're only talking about a diode drop or so less. Both have wide frequency responses and good current gain. The Sziklai pair has a higher voltage gain, but this is the difference between maybe 0.999 and 0.980; not worth writing home about.

The Szilai pair is prone to a peak in response ...way... out in high frequencies, half to several MHz or more. The Darlington doesn't do that, but the Darlington tends to be touchy about capacitive loading.

For pedal purposes, the biggest issue will be whether you have two NPNs or one of each kind.

[bool]

1) for a sim, you need to get the both schems straigh to get a meaningful result. For starters you have two different CCS .. no bueno

2) you can get a darlington in a single package, a sziklay costs you two packages

3) with both you need to be aware of loading and self-oscillation issues. You need to decouple both inputs and outputs properly

4) personally I prefer the "tone" and response of a sziklay, but there's a narrow window when it's best. A darlington is pretty much a plug-and-play thing, no fine-tuning needed

[N9]

They are quite similar in most characteristics. The Darlington has a touch less headroom, as it will not saturate to as low a voltage as the Sziklai pair, but we're only talking about a diode drop or so less. Both have wide frequency responses and good current gain. The Sziklai pair has a higher voltage gain, but this is the difference between maybe 0.999 and 0.980; not worth writing home about.

The Szilai pair is prone to a peak in response ...way... out in high frequencies, half to several MHz or more. The Darlington doesn't do that, but the Darlington tends to be touchy about capacitive loading.

For pedal purposes, the biggest issue will be whether you have two NPNs or one of each kind.

Thanks for this information!

A couple more questions:

-- Am I correct in thinking that a base resistor will reduce peaking in the Szkilai pair?

-- Darlington pairs seem much more common in the effect realm; is this simply because it's easier/cheaper to use two of the same transistor rather than a complementary pair?

1) for a sim, you need to get the both schems straigh to get a meaningful result. For starters you have two different CCS .. no bueno

2) you can get a darlington in a single package, a sziklay costs you two packages

3) with both you need to be aware of loading and self-oscillation issues. You need to decouple both inputs and outputs properly

4) personally I prefer the "tone" and response of a sziklay, but there's a narrow window when it's best. A darlington is pretty much a plug-and-play thing, no fine-tuning needed

-- Am I wrong in thinking that the current source needs one less diode for the Szkilai pair?

-- What additional decoupling would be needed?

[R.G.]

-- Am I correct in thinking that a base resistor will reduce peaking in the Szkilai pair?

If you mean a series "base stopper resistor, no. The peaking comes from the 100% voltage feedback from the second collector to the first emitter. The peaking is a result of the parasitic junction capacitances of the transistors around that loop, not from damping on the input base. So a base stopper can't help.

If you mean a base-to-emitter resistor, maybe. Especially a base-emitter resistor on the second transistor, as this lowers the small signal voltage gain of the first transistor.

Darlington pairs seem much more common in the effect realm; is this simply because it's easier/cheaper to use two of the same transistor rather than a complementary pair?

I don't really know. NPNs have always been cheaper with silicon, but that's so long ago that I'm not sure it matters. I suspect that most people don't know much about the complementary pair connection. The complementary pair does have some internal gotchas that can be mysterious if you're not aware of them, darlingtons are generally not as prone to issues. Maybe that's it.

-- Am I wrong in thinking that the current source needs one less diode for the Szkilai pair?

Yes.

If the current source is a current source, and not tied into the two upper transistors' functions, you just put in a current source. The number of diodes and hence the voltage across the transistor and stabilizing resistor makes a difference in how close to ground the CCS will go. But you could also use two transistors as a current mirror and feed the mirror input from a resistor to V+. The current will vary with V+, but you're not really interested in **constant** current in the macro sense, but a high impedance CCS for small signal stuff. A slowly varying CCS would be fine here.

[N9]

Looks like I have a long way to go in understanding these circuits! I've done some reading and am focusing on the Szkilai pair, since I don't understand it as well and haven't found many examples of it.

This is the new schematic I've drawn up:


Now for the questions:
-- Should I expect every Szkilai pair to peak at some very high frequency? Is there a typical workaround for it?

-- I've seen a huge number of variations on current mirrors, and I'm struggling to grasp which version is most acceptable for this circuit. Does a simple buffer circuit like this need anything other than a basic mirror?

[bool]

a CCS is much better really

[R.G.]

Forgot to mention:
Here's a Sziklai pair buffer with CCS loading already worked out for 6V power and optimized for low current drain.
http://geofex.com/FX_images/Onboard_Preamp.pdf

[N9]

That's a very nice little circuit, RG! Do you folks know of an online resource for learning about basic CCS design? Google hasn't been very helpful so far...

[PRR]

My initial observation: if you indeed have a 10K source (R9), and a 10K load (R2), *why* do you need a double-transistor buffer?

One 30-cent transistor worked unity voltage gain will give an impedance ratio over 100. Unless you need >1% absolute precision, that would be ample for the job in your sketch.

> thinking that the current source needs one less diode for the Szkilai pair?

Why?

The *current* in the current-source is selected according to your *load*. If you need to drive 1K to 10V peak, you need 10V/1K= 10mA at-least in this current cource. For typical 9V guitar-cord stuff, figure 3V peak and usually >50K load, 0.06mA. Unless the upper (signal) half is exceptionally crude, that's your main thought. Since upper-halves often have some "crudeness" (Gm varies with device current), you may look to double this current, say 0.15mA. (Your CCSes are running 1.5mA or 3mA.) 

Note that for guitar-cord 50K loads, and all but the smallest battery (R.G.'s onboard buffer with coin-cells), you can get fine drive with a simple resistor "current source". On a nearby thread I show a 10K emitter resistor driving 3.5V peak into 50K. Standing current is 0.4mA, not as low as a sharp-pencil CCS design, but in most situations another tenth-mA or so won't blow a fuse or noticably run-down a battery. And the simplicity is, IMHO, compelling. In emitter-follower work, the distortion is "zero" for nearly any reasonable purpose.

This plan has a third the number of parts and I doubt the difference is audible. Input Z happens to be higher, output Z is lower (lacks your R6, maybe needed to stabilize compounds with many transistors), current consumption is lower. Also much wider bandwidth (>2MHz), though I suppose you added R6 C5 to roll-off supersonics.

[N9]

Note that for guitar-cord 50K loads, and all but the smallest battery (R.G.'s onboard buffer with coin-cells), you can get fine drive with a simple resistor "current source". On a nearby thread I show a 10K emitter resistor driving 3.5V peak into 50K. Standing current is 0.4mA, not as low as a sharp-pencil CCS design, but in most situations another tenth-mA or so won't blow a fuse or noticably run-down a battery. And the simplicity is, IMHO, compelling. In emitter-follower work, the distortion is "zero" for nearly any reasonable purpose.

This plan has a third the number of parts and I doubt the difference is audible. Input Z happens to be higher, output Z is lower (lacks your R6, maybe needed to stabilize compounds with many transistors), current consumption is lower. Also much wider bandwidth (>2MHz), though I suppose you added R6 C5 to roll-off supersonics.

Thanks for your sketch and extra info! I should have clarified; my primary purpose for this circuit is to ask "what if?" rather than building it to use every day. You're quite right to say that there are much simpler ways to handle this task.

With that said, I've got a couple more questions about your example schematic:

-- Aren't all emitter followers prone to oscillation when driving a capacitive load, such as a guitar cable? I realize you may be omitting an output resistor for simplicity's sake, but I'm still curious.

-- After thinking this over a little more, it seems like most guitar buffers are best implemented as either A: an op-amp or B: a cheap and cheerful single transistor arrangement like you've shown. Are the more complex arrangements best suited for specialist purposes, such as RG's onboard circuit?

[bool]

You need to isolate both input and output (with resistors). Without deep explanations, a simple buffer will "reflect back" to the input all anomalies that are seen downstream the output (this will also "modulate" the input impedance). The simplest way is to include a "buildout" resistor to reduce the impact of the downstream anomalies.

The "input" resistor will also isolate the transistor base from some of the parasitics of the source impedance. Another way to foolproof your buffer is to form a LPF with the input resistor and a small cap to GND (from the transistor base), somewhere around 100pF or a bit more for RFI rejection.

The really smart way is to use a single-package darlington because it reduces the "reflected" influence from the output, due to its high compound hFe. You can also run it with much higher current (smaller emitter resistor) for better output drive ability, say 2mA. All that for a marginal price difference. If you go SMT, a mmbta13 and a couple of passives is all you practically need.

[PRR]

> Aren't all emitter followers prone to oscillation when driving a capacitive load

Well, then aren't multi-stage connections even more prone?

The simple one-BJT EF is not unstable without help from what is driving it.

Firstly, because being under-unity voltage gain, and it being hard to find a passive network with significant voltage gain, it is unlikely to get in trouble.

I'm going out on a limb and say that resistive sources are no-problem, and capacitive sources may be no-problem also.

Inductive sources may be a problem. The EF's capacitive input (from both hFE fall-off and any caps you tie to its tail) against an inductor make a tuned circuit which, in many cases, will love to ring. Interestingly we don't find many explicit inductors in audio these days. (Any wire is an inductor but with reasonably compact build and rational-performance BJTs, we don't often make trouble.)

The Darlington does a double-twist on the load impedance and is more complicated. In short, the "hFE" falls twice as fast as frequency, so the input is not a simple reactance.

The Sziklai is much more fun. It is two common-emitter stages. Each offers high voltage gain. No amplifier has gain to infinite frequency, they always pass through unity gain somewhere up there. At the corner between flat and falling, there is a 45 degree phase-shift; when gain is far down the fall-off the phase-shift approaches 90 degrees. Per stage. We have two such stages, so it approaches 180 degrees. If LF gain is high, it will get very close to 180 deg before all gain is lost.

Now we connect the output back to the input. We are *sure* to have a point where gain is more than unity and phase-shift is approaching 180 deg.

Just like that, it can't oscillate (we need 180.00 deg). However it sure can peak-up. For "ordinary" BJTs where hFE is flat to 1MHz and falls to unity at 300MHz, we typically find a tall narrow peak about halfway out, often 10MHz-50MHz.

Your guitar does not make 10MHz. However a passing taxi-radio might. (And the stories of some guitar amplifiers playing at truck-stops and being blasted by 27MHz CB signals...)

Also- this is just the transistors. There's capacitance everywhere and it all sucks. At these frequencies the "negligible" inductances have some effect. If your loop has 175 degrees of phase-shift that you know about, and another 5 deg of "negligible" phase-shit, it's gonna sing full-power out in the taxi/police/CB/SW bands.

I think the Sziklai is like sleeping with serpents. I need a very good reason to do that. It has a big role in Power Amplifiers where the extra volt of headroom costs real money, and where the driver and output devices are of *different* speeds. (A 300MHz driver and a 3MHz 2N3055 are less-likely to add-up to nearly-180-deg phase-shift.... until you burn a '3055 and go to find a replacement which happens to be "better" {faster}.)

You can "tame" the Sziklai by *laming* one of its transistors, so it does not contribute enough gain to get into trouble. Or enough gain to do much good. Massive emitter resistance may reduce the MHz peak but also reduce LF performance. Massive shunt or Miller caps typically clobber the top of the audio band before they reduce the peak.

[merlinb]

-- Aren't all emitter followers prone to oscillation when driving a capacitive load, such as a guitar cable? I realize you may be omitting an output resistor for simplicity's sake, but I'm still curious.

Yes, emitter followers are more likely to oscillate into capacitive loads -the exact opposite of common-emitter amplifiers. When a capacitive load combines with stray base-emitter capacitance you can end up with an overall negative input capacitance, which will oscillate with any stray source inductance.

[N9]

Thanks again to everyone for their detailed answers!

Now that I know a little more about this subject, I'm wondering what scenarios DO call for a more complex buffer arrangement... Is it usually better to jump straight to op-amps?
RG's low-power circuit is the only example I've seen that makes use of a Sziklai pair.

[PRR]

> When a capacitive load combines with stray base-emitter capacitance you can end up with an overall negative input capacitance,...

I quickly found a peak with substantial (not "stray") B-E capacitance and load capacitance. The "best" (worst) I could find quickly is +5.98dB-- going off these particular values gave smaller peaks.



The inductive input case is also peaky but seems to require pretty specific values (enormous L and tiny C), unless I have been exceptionally "lucky". (True, a breadboard or a final-build will find instabilities never seen in a simple simulator because of the intrinsic bad-luck of real amplifiers.)