Boss Waza-Craft buffer: what's up with that?

[Rob Strand]

You don't happen to have a schematic for that as well, do you?

It's only in my mind.  The differences are fairly trivial and don't change the basic idea:

- D1 not fitted
- C1 moved to other side of C2
- C3 = 47u                           ;  maybe the DM-2W is actually 47u.
- R5 = R6 = 33k
- C4 I couldn't see one fitted. ; I'm not 100% sure it connects to the diode on the DM-2W
(- R1 couldn't read it)

On the SD-1W and the DM-2W C2 is made-up of two SMD film caps in parallel.

I might mark-up the schematic to include both versions.

What you do think about C4 connecting to the diodes?

[Fancy Lime]

...

What you do think about C4 connecting to the diodes?

Well, I think it makes sense in that position for stabilization but I have never seen a cap there in this topology. So I don't really know if it is useful or necessary but I would definitely not put it past the designers to put it in there.

Honestly, I am in awe that you even manage to trace these things at all. I cannot see jack in terms of what is connected to what.

Andy

[antonis]

IMHO, this topology should be better with the use of one diode in series with an extra base to GND resistor..

[Rob Strand]

Well, I think it makes sense in that position for stabilization but I have never seen a cap there in this topology. So I don't really know if it is useful or necessary but I would definitely not put it past the designers to put it in there.

Honestly, I am in awe that you even manage to trace these things at all. I cannot see jack in terms of what is connected to what.

Yes it's a pain.  It definitely helps to have a few possible circuits in your head which match the patterns you see on the board.   I guess that's why asking for a second opinion about the cap connecting.  I had a look this morning and it's fairly clear C4 is connected as shown on the DM-2W.

I can only speculate at the motivations for having the cap there but the truth is you might not need it at all  .
More often than not there's no cap there. 

There's two motivations to put it there:  The first is to prevent supply noise/ripple affecting the current source.   The second is to ground the base of the current source transistor to stop it affecting stability (due to CB capacitance of current source transistor).  The low impedance of the two diodes helps both cases.  However in this case the diode current is low so the impedance isn't *that* low, maybe 400 ohms.  At 400 ohms a 10uF would actually do something to the ripple.  I'd have to look at all the fine details to see if adds something.

Another scenario is, the DM-2 has quite noisy circuits so maybe audio testing showed-up a small amount of junk and the engineers traced it to the supply and added the cap.  The SD-1 is analog so the supplies are clean and don't need the cap.

IMHO, this topology should be better with the use of one diode in series with an extra base to GND resistor..

Without looking at the details there's no way to know what is better.  The temperature stability would improve but the effects I just mentioned would get worse.   If you replace one diode with a resistor the resistor would need to be about 2.8k so that makes those issues about 7 to 8 times worse.

[bool]

The only possible improvement in that *exact* topology is to use a schottky/standard signal diode combo in the ccs - and to up the second bjt loading current (reducing the emitter resistor). This netting the effect of sliiiightly increased negative excursion output swing-ability. (if you really want to use this topology).

Imho a ring-of-two cfp pair dances around this topo all day long, ymmv.

[Fancy Lime]

...

Imho a ring-of-two cfp pair dances around this topo all day long, ymmv.

I've been wondering why not just use a Darlington. Should get us very close to unity when used as a bog-standard emitter follower and if we feel we have to we could always bootstrap it for high impedance. Much simpler than the Waza solution. But would it really be that much worse? It may be less linear or more noisy. Does anyone know?

Bool advocates a Sziklai pair instead. What would the advantages of that be over a Darlington?

Also: If linearity and perfect unity gain are our concerns, why not just use an opamp? Does that really make a difference in terms of noise anymore with modern opamps?

Andy

[bool]

...
What would the advantages of that be over a Darlington?
...

NFB so less THD, better output swing, potentially lower S/Nr if you combine a low-noise transistor with a more powerful output device.

[Fancy Lime]

...
What would the advantages of that be over a Darlington?
...

NFB so less THD, better output swing, potentially lower S/Nr if you combine a low-noise transistor with a more powerful output device.

You mean "low-noise transistor in the first position, high-current transistor in the second position"? Do we truly care about the output current? I mean, we are still feeding a circuit with a moderately high input impedance, in most cases anyway. So I would rather opt for two low-noise transistors, no? Does it matter if one uses the PNP-first or NPN-first configuration? I noticed that in complementary NPN-PNP pairs, the PNP typically has a slightly lower noise figure. On the other hand, the truly low-noise NPNs, like the MPSA18, don't even have complementary PNPs.

EDIT:
I forgot to ask: what's a "ring-of-two" cfb pair?

Andy

[Rob Strand]

I've been wondering why not just use a Darlington.

One decision point is JFET vs Transistor.  It's hard to beat the noise of a JFET with a high-impedance source.  The side effect of the JFET is more attenuation.  So if that's a problem you can use a current source load, like the waza, or the feedback amps I poster earlier.   

JFETs introduce more distortion (if that's an issue) and more buffer loss when compared to transistors.

All transistor versions of the circuit have a limited output load to input impedance ratio set by the gains of the transistor.  If you don't need to drive a low impedance load you can focus on optimizing operating currents for noise.

Notice that the transistor version of the feedback amps with gain=1 basically form a Sziklai pair.  So you can view it as a buffer with a Sziklai pair  or a feedback amp with two transistors.   The Feedback amp structure however lets you come-up with the JFET input version of the circuit relatively easy, you don't need to invent a JFET+BJT Sziklai pair first.

The Sziklai pair/Feedback amp version of the circuit does have low distortion but I think it's hard to beat the transistor buffer with a current source load - I think it beats the Sziklai pair/Feedback amp.   The transistor buffer with the current source load also has the least buffer loss.

Now getting to the Darlington.  So first you have to accept using the BJT input and the trade-off of noise vs input source impedance.   Once you do that a Darlington is an OK solution.   However, it does have more loss than a single stage buffer but it lets you up the input impedance.    On a real Darlington you would probably add a resistor across B and E of the second transistor.   You can tune that resistor to minimize buffer loss and distortion.

Think of the BE resistor on the second Darlington transistor as creating a bias current for the first transistor.   Now imagine instead of having that resistor you place a current source from the emitter of the first transistor  to ground.   What you end-up with is the Waza circuit except it has a BJT input instead of the JFET.   This form of the circuit is slightly better than the Darlington with the BE resistor, both in distortion and buffer loss.

FYI,  you can also make a JFET buffer with a current source load.  This is basically the Waza circuit with the second buffer stage removed.

I forgot to ask: what's a "ring-of-two" cfb pair?

This is a circuit developed by Williams in the mid 1960's.  There was a heap of articles in Wireless World, they were mostly in Letters to the Editor.  Even the famous Peter Baxandall got into it!.

There's a current source version and a voltage source version.   The idea is the output current is virtually independent of the supply voltage - which is what bool was getting at.


See
https://www.americanradiohistory.com/index.htm
Williams, WW Sept 1966, p456, Letters to the Editor ; *** The first appearance of the "ring of two" idea.
Williams, WW July 1967, p318, Ring of 2 Voltage Reference,  Article

http://cdsweb.cern.ch/record/1063330/files/CM-P00068130.pdf
http://web.mit.edu/magic/Public/papers/06131873.pdf

[bool]

A sziklai made with a lownoise fet at input, a speedy pnp and a ccs load (same-ish component count as in this particular boss buffer) would performance wise dance around all here proposed alternatives all day long (and night). (Basically, it's a discrete current-feedback opamp configured for gain=1).

The question is "does it vibe with your tone". Which I think for a guitar or a bass it doesn't.

[Fancy Lime]

A sziklai made with a lownoise fet at input, a speedy pnp and a ccs load (same-ish component count as in this particular boss buffer) would performance wise dance around all here proposed alternatives all day long (and night). (Basically, it's a discrete current-feedback opamp configured for gain=1).

The question is "does it vibe with your tone". Which I think for a guitar or a bass it doesn't.

What do you mean by "vibe"? It's a buffer, it's not supposed to vibe, it's not supposed to alter the tone in any way. As long as the output is as exactly as possible the same as the input in terms of AC voltages in the audio band, but at a much lower impedance, then it's a good buffer. If, on the other hand it makes any input signal sound like a harpsichord, it's an interesting effect but not a (good) buffer. I can't imagine that it would, in that case, be a particularly useful buffer for anything. Granted, most guitar buffers alter the sound a bit but that's a bug not a feature.

Andy

[Rob Strand]

A sziklai made with a lownoise fet at input, a speedy pnp and a ccs load (same-ish component count as in this particular boss buffer) would performance wise dance around all here proposed alternatives all day long (and night). (Basically, it's a discrete current-feedback opamp configured for gain=1).

I just tried doing an all transistor minimum noise design for a 15k source impedance (very rough approximation to a guitar).  The output load was an AC coupling cap to a 100k load.    Three cases:
1) Single transistor buffer with current source load
2) Sziklai/Feedback (two BJT's)
3) Single transistor buffer with current source load followed by Transistor buffer with resistor load; basically the transistor version of the Waza.
So (2) wins on distortion and loss, second is (1), and last is (3).   (1) and (3)  are quite close.  One caveat is the collector current is quite low for minimum noise so case (1) will have a higher impedance to 3.

Noise is about the same in all three cases, largely being dominated by the source resistance.

The other point is the noise minimum is very shallow.    Increasing the collector current by a factor of 10 only increased the noise by 1dB or so.

Designs not optimized for noise gave different orderings of goodness.   Like I could lower the loss or lower the distortion.

[bool]

... hmm yeah nice analysis but we sort of knew that in advance did we..

I'm sure Boss had a good reason to waste smd beans like that. perhaps they had a large stock of semis they had to dispose of.. or they thoroughly -vibed- their buffer paradigm in and out. (like "vibe: the final frontier")

But for the giggles, you could un-IP the boss thing in a semi-same-ish fashion with using just a fet and one bjt: delete the ccs sub-cckt; and stick the 470R between the bjt B-E, making it a pseudo-ccs for the fet. For this to work properly, the bjt load resistor should be also lowered to something like 4k-ish. Or increase the 470R a bit (around 680R would load the fet with cca 1mA).

So here you have plans for a "Waza stupid" buffer laid out in plain sight!

[Fancy Lime]

Sooo, does that mean we'll be using Sziklai buffer now? With a N-Channel JFET first and a PNP second? Or just a regular old BJT-BJT Sziklai? The combines hfe should be large enough for decently high input impedance. If not, there is always bootstraping, Cornish buffer style.

Andy

[Fenton Bresler]

Greetings all!

I've been messing around with Darlington and Sziklai buffers for a while, purely out of interest and for educational purposes. But I don't really see the value in them above that. Surely any op-amp would outperform the lot, and with far fewer components. Am I missing something?

Boss seem to have been throwing parts around on this one, perhaps it could have been improved with the addition of a single BJT. There's already a current sink available, one could simply extend that sink with a BJT and a current setting resistor in place of R4 alone.

[Rob Strand]

... hmm yeah nice analysis but we sort of knew that in advance did we..

Sooo, does that mean we'll be using Sziklai buffer now? With a N-Channel JFET first and a PNP second? Or just a regular old BJT-BJT Sziklai? The combines hfe should be large enough for decently high input impedance. If not, there is always bootstraping, Cornish buffer style.

The decision the make the first transistor JFET or BJT changes things a bit.   At the end of the day making the first stage a JFET is most likely to give the best noise performance.

When the first transistor is a BJT it is best to operate it with a low current to keep the noise down.   So that means the output impedance goes up and it is helped by having a second transistor.   So with the constraint of operating Q1 at low current the Sziklai case looks pretty good.

When the first transistor is a JFET it is best to operate the JFET at a high-ish current to keep the noise down.  When you do that the output impedance of the JFET is relatively low but not great.  If you have a JFET with a current source load, even without the second transistor like the Waza, the performance is actually pretty good ie. low buffer loss and also low distortion.  I compared the simple JFET+CS case to the JFET+BJT feedback amp (ie. the JFET version of the Sziklai) and the JFET+BJT has more distortion but lower loss than JFET+CS.

It's good to keep in mind some of the differences are fairly academic.   There's also some real differences.  We would need to watch out for low impedance loads on the simple JFET + CS.   The JFET+BJT Feedback amp is probably the most likely to have oscillation issues and perhaps will need good supply bypassing.

The thing about the JFET+BJT feedback amp is you have a second variable to pin down: the collector/source resistor at the output.    Since we are operating the JFET at high-ish current and the JFET source current flow from thought the output resistor, if the output resistor is chosen too high the BJT will operate at low current and essentially the circuit will fall back to a JFET + source resistor buffer.  In this case the BJT is kind of benign.  From what I can see there is a value of the output resistor which gives the least buffer loss.  Then decreasing that resistor further increases the buffer loss again but the distortion decreases and the supply current increases - so here we have the option to trade-off performance specs.

The fact we need to operate the transistor at a minimum current and the JFET is operating at a high-ish current means this configuration is likely to draw a little more current.

Notice in the JFET+BJT feedback amp the BE resistor sets the JFET operating current.  We don't need a current source to set the JFET current.   It only sets the current when the BJT current is not so low as to make the BJT benign.

I've been messing around with Darlington and Sziklai buffers for a while, purely out of interest and for educational purposes. But I don't really see the value in them above that. Surely any op-amp would outperform the lot, and with far fewer components. Am I missing something?

As far as getting the most out of the circuit for a guitar I'm not so fond of the single-part Darlington transistor solution since it doesn't allow you to set the bias current of the first transistor.  If you make your own Darlington then you can tune the operating current using the BE resistor.    I only see the Darlington being an advantage if you have a low impedance load.   The Sziklai does seem to have some advantages over the Darlington in general.

You can definitely get lower noise and lower distortion opamps but the good ones quite often operate at higher supply currents.   The discrete designs are pretty good and are probably better than your average opamp.  The thing discrete designs offer is they are likely to operate at lower currents, maybe a bit more swing.  You can also build a single opamp, so no wasted current.

If we go back to the original problem:
BJT buffer has higher noise with the high impedance source of a guitar
==> JFET buffer low noise  but  JFET+source resistor has a volume drop
==> So to fix either of those problems you have to do something.
       Probably best to keep the JFET input from noise perspective.
       After that, all the options have small differences in performance.

The combines hfe should be large enough for decently high input impedance. If not, there is always bootstraping, Cornish buffer style.

Bootstrapping gives you higher input impedance but it doesn't solve the input noise currents of a BJT design.

[Jubz]

Hi.

JFET, BJT and Darlingtons options have been discussed but I have the impression we forgot the MOSFET (and opamps, why not just use opamps?). MOSFET buffer seems very good on paper : has higher input impedance than BJT and lower output impedance then JFET (and loses less signal). But I dont often see it used in stompboxes? What are the tradeoffs (internal capacitance? noise? distortion?)?

[Rob Strand]

MOSFET buffer seems very good on paper

Yes they sure do.  In theory they should have noise specs comparable to JFETs, however, in practice they are much noisier.  The commonly available parts are quite noisy.

[bool]

Bootstrapping gives you higher input impedance but it doesn't solve the input noise currents of a BJT design.

To some extent it's possible to band-limit the bootstrap fb signal, in net effect diminishing the buffers' cur.noise outside of the band of interest, and this is cheap and simple to do, but fails at midrange freqs, which is where the crunch lives...

[Rob Strand]

To some extent it's possible to band-limit the bootstrap fb signal, in net effect diminishing the buffers' cur.noise outside of the band of interest, and this is cheap and simple to do, but fails at midrange freqs, which is where the crunch lives...

I haven't tried that.  Normally I try to get full bandwidth, and largest cap I want to use sets the bandwidth.