[Vitalstatistix]

The more buffers the merrier?

I'm getting a headsave ultra but I have no idea how much buffers I'd need. The ultra comes with Elantec 2001 buffers (5), but I can add 4 more.

I have the ety ER4P and a future Senn 650

 

[dokebi]

I wonder what the difference of a buffer and an opamp is. According to a diy site you can even make an adapter that converts opamps into buffers. If the object of a buffer is to increase current. Can't you use an opamp to do the same? How about a cha47 topology but with 4 instead of 2. Anyone?

 

[jeffreyj]

Op-amps (operational amplifiers, that is) typically provide lots of voltage gain but due to a number of engineering constraints they invariably have wimpy output stages that can't source or sink a lot of current - 10 to 50mA max is the usual range. This is where a buffer steps in: they don't provide voltage gain but they do provide lots of current gain, freeing the op-amp from the onerous task of supplying high power.

How many buffers one "needs" to drive a particular pair of headphones is the subject of considerable (and usually non-sensical) debate. If you have low impedance cans (less than 64 ohms, say) then you need lots of current gain but not so much voltage gain. If you have moderate impedance cans - around 100 ohms - then you need a little of both voltage and current gain, while high impedance cans, like the Sennheiser HD-600 at 300 ohms, require lots of voltage gain but very little or no current gain.

Theoretically, one BUF634 buffer should be more than adequate to supply all the output current any pair of cans could possibly take without frying, but some argue that paralleling (or "stacking" as they like to call it 'round these parts) multiple buffers is beneficial anyway. The argument that it lowers the output impedance is a bunch of bs because the buffers are invariably enclosed in the op-amp feedback loop so the effective output impedance will be in the milliohm range, anyway. Stacking buffers does reduce the rise in low frequency distortion that occurs from thermal feedback on the silicon die from the output transistors to the drivers. Most modern buffers take this into account (using Vbe cross-cancellation techniques, typically) so this may or may not be a problem with a particular buffer. Generally, if you feel compelled to stack buffers, then you probably need to use a different buffer or else look into a discrete current gain stage - stacking is just so inelegant.

To be fair, my take on this is distinctly in the minority here, and YMMV anyway.

Can you parallel op-amps (with the obligatory current-sharing resistors!) instead of using a buffer? Well, to an extent, yes, but keep in mind that one BUF634 can provide 250mA of output current which would take a minimum of 5 op-amps in parallel to duplicate. The economic advantage of this technique is rendered moot if the op-amp you are paralleling is an AD8620!

 

[aos]

Quote:

the buffers are invariably enclosed in the op-amp feedback loop so the effective output impedance will be in the milliohm range, anyway

Yes, but isn't the factor of reduction the total gain itself? If the gain is only say 5, then a 15 Ohm buffer (EL2001 can get that high) is still 3 Ohms, not milliohms. I will admit I was not able to hear differences between 1 and 2 buffers but I am MOST DEFINITELY able to hear differences between EL2001/2 and EL2009 (which has 1Ohm open loop impedance). You'd think even a single EL2001 should drive headphones let alone a line out just fine, but EL2009 is considerably better for example in tightness of bass.

The other thing is, parallelling most definitely reduces distortion. This is clearly measurable (and has been by several people) so it's not some kind of subjective bs. If you are driving high impedance phones it can be inconsequential but lower the load to 32Ohm (or even 16) and there's a big difference in THD.

 

[ppl]

This time I can't see Jefferyj's point at all........ First off the Output impedance of two or more Paralelled Buffers is less than each on it's own, You Know this......... This Impedance will be reduced by negitive feedback by the feedback factor amount..... The amount of distortion present when an Amp drives a Load is more than when it is unloaded.

Paralelling solves these two problems nicely by lowering the output impedance of the Amp Prior to the application of negitive feedback the end result after feedback is also lower Output impedance. since each buffer is now only handeling part of the Load current the Output transistors will most probaly be in a Higher gain range since Hfe tends to fall as more current is Required. These two factors alone can account for alot lower THD and IMD measurements. Subjectivly you notice more athority and punch. This is also done in alot of High End Loudspeaker Amps by paralelling lots of output transistors.

 

[Vitalstatistix]

So having say 9 Elantecc 2001 buffers is okay right? That is, it's not harmful or anything?

 

[jeffreyj]

Quote:

Originally posted by aos Yes, but isn't the factor of reduction the total gain itself?

It's the inverse, actually. The ratio of closed loop to open loop gain times the output impedance. IOW - the higher the closed loop gain, the less the output impedance is reduced.


Quote:

The other thing is, parallelling most definitely reduces distortion. This is clearly measurable (and has been by several people) so it's not some kind of subjective bs....

Sure, but to what extent? The difference between 4% and 2% is substantial, I'll agree, but I'm not so impressed in going from 0.02% to 0.01%. Then the law of diminishing returns kicks in: each additional buffer after the second provides less and less benefit while simultaneously increasing the role of parasitics in the circuit's operation.

I bought a few EL2001's from Tangent a while back and I think they sound pretty good... for a monolithic buffer. They pale in comparison, though, to a simple complementary common emitter buffer I built using very generic 2n3904 and 2n3906 transistors. This cCE buffer not only provides more current gain than the BUF634, it also provides just a touch of voltage gain and sports a saturation voltage of 200mV to either rail. Oh, and it costs all of about $0.40US per channel.

 

[dokebi]

Quote:

simple complementary common emitter buffer I built using very generic 2n3904 and 2n3906 transistors

Elaborate for this newbie. Thanks

 

[jamont]

I'd like to see it too, thanks Jeffrey!

 

[Squalish]

And for this one.

If you think it will sound/measure better than what is currently commonly used, feel free to publish your designs as if they were the greatest things since sliced bread.

 

[jeffreyj]

This is not a "unique" circuit, folks. The complementary pairs are set up in a tight local feedback loop with gain set by the voltage dividers from output collector to driver emitter and ground. Compensation for the local loop is provided by the RC networks R14/C2 and R15/C3. C1 sets the -3dB point for the op-amp at 1.59MHz. I used the slightly beefier MPSA06/56 complementary pair in the output instead of the 2n3904/6 because they have max collector current ratings of 0.3A instead of 0.1A. The open-loop output impedance is pretty much the value of the emitter resistors (2.2R) and these aren't too critical. No current limiting is provided, so putting a 10-22R resistor in series with the output might be prudent (also a good idea if driving capacitive loads, btw).

This circuit makes maximum use of the available supply voltage because the output transistors are configured as common emitter. Emitter followers (the more usual configuration) not only have Vce[sat] to contend with, but also the Vbe diode drop, meaning at least 0.8V of each rail gets wasted. Of course, fidelity will be suffering by the time saturation approaches, but its still nice to have an extra 1.2Vpp available.

The buffer is biased into Class AB by the input diode string and draws a quiescent current of 1.7mA from each rail (as tested). If more output current (or a lower Zo) is needed, one can merely parallel output transistors (each with their own Re) or perhaps use some of Zetex's very nice devices with over 1A Ic ratings in a modified TO-92 package.

Cheap, but good.

 

[aos]

Quote:

Sure, but to what extent? The difference between 4% and 2% is substantial, I'll agree, but I'm not so impressed in going from 0.02% to 0.01%. Then the law of diminishing returns kicks in: each additional buffer after the second provides less and less benefit while simultaneously increasing the role of parasitics in the circuit's operation.

Well, I am CERTAINLY impressed any time I see 50% reduction in something - and that is what you get when you go from 0.02 to 0.01. Second buffer typically halves the distortion. And while it may not mean much to reduce from 0.02 to 0.01, reduction from 0.2 to 0.1 is probably worthwhile, and that is what you get with EL2001 driving 32 Ohm headphones. I wouldn't pass that up.

I do agree what additional buffers after that start to have diminishing returns. It still might not be crazy to go for 3 or perhaps 4.

I have heard discrete amplifiers as well as opamp-driving-transistor amplifiers and I must say that they fail to compare to great sounding opamp amplifiers / buffers. Don't forget that using EL2001 as audio amplifier is not a common knowledge.

 

[jeffreyj]

Quote:

Originally posted by aos ...And while it may not mean much to reduce from 0.02 to 0.01, reduction from 0.2 to 0.1 is probably worthwhile, and that is what you get with EL2001 driving 32 Ohm headphones. I wouldn't pass that up.

Well I am of the opinion that 0.1% distortion for a SS amp is terrible. A two orders of magnitude improvement is easily obtainable with discrete buffer circuits without exerting too much effort. One of the major, but not-so-obvious advantages that a discrete design has over a monolithic one is that there is no thermal feedback path to cause a rise in low-frequency distortion.

Quote:

I have heard discrete amplifiers as well as opamp-driving-transistor amplifiers and I must say that they fail to compare to great sounding opamp amplifiers / buffers. Don't forget that using EL2001 as audio amplifier is not a common knowledge.

Op-amps and monolithic buffers are incredibly convenient devices, but some of the best designers in the world (Jim Williams, Robert Pease, even that pioneer of analog IC design, Bob Widlar) concur: if you want speed, power and accuracy, you have to go discrete.

 

[Vitalstatistix]

I am certainly most impressed by your advanced discussions on buffers and 0.1% distortions blah blah blah

At the end of the day - my questions remain unanswered.

Don't tell me you guys are too "advanced" to answer a newbie's questions. Heck - if you can't, point me to somewhere where I can find out more. I'm sick of doing a google search looking at websites I don't even comprehend.

 

[tangent]

Quote:

At the end of the day - my questions remain unanswered.

You've gone and trod into the land of engineering, where no solution is universally bad or good. The best solution depends on the circumstances.

So, you start throwing out solutions that just don't work. JeffreyJ's discrete buffers fall into this category. Sure, you could hack a META42 board to accept them, but it's almost academic. Tigger probably won't do it.

EL2009s, same deal.

So we're down to whether 2x or 4x buffers on the outputs is better. If it's a wall powered amp and money is not a major concern, go with 4x. If you want to save a bit of money and not lose out on a major improvement, go with 2x. If it's a battery-powered amp, going with 4x will require heroic battery setup measures, so 2x will make more sense.