[mb]

thanks hugh,
my only problem with that is that i had wanted a higher input impedance.  one of the advantages of this buffered pre is that i can have an extremely high input impedance, and an extremely low output impedance.  so i kind of feel that using a 22K or so pot is going to negate this advantage.
jesse

Hi Jesse,

In electrical terms, I agree. In my listening experience, I'm not 100% convinced. My initial attenuator (the TKD 41-step) was 10k, as that was Hugh's recommendation in earlier days. For different applications I have also used 100k pots, 10k, 22k and 47k good quality metal film resistors, and the verdict it still out. My overall impression is that lower impedance (10k) followed by a buffer sounds better (microdynamics, fluidity). I'll bet it varies from one source device to another, and it also affected by cable.

short, low capacitance cable -> 10k atten -> buffer (best, if source has reasonable o/p impedance)
short, low cap cable -> 100k atten -> buffer (better for tube source? more susceptible to noise?)
.... lots of other combinations??
long, high capacitance cable -> 47/100k passive (worst)

In the end, imho only listening and component matching can confirm what's "best".

[Larry]

When comparing the pot/attenuator in preamps, please take the note that the pot in GK-1 is between the input and the output, which has different impact on input/output impedance of the GK-1.

[AKSA]

Folks,

I should explain how the volume control is implemented in the GK-1 as its operation in this circuit is a little unusual since most preamps have their attenuator right at the input.

It is fully buffered.  The signal is taken direct from source and passed through the GK-1 solid state gainblock which has a Zout of 32R.

It is then presented to a 22K volume potentiometer.  The output of the pot, the wiper arm, is presented to the grid of the output tube.  This grid has a Zin on the order of a megohm, so the pot 'see' a very low source impedance driving it, and a very high input impedance to the tube.  This is the optimum for any potentiometer used in audio, and reflects MB's observation that a volume control should be fully buffered.  This is the real reason that the GK-1 has an inverted, hybrid arrangement;  most have a tube input, and SS output, and this topology feature is partly responsible for, I believe, the astonishing clarity and detail.

I should emphasize that the output of the GK-1 is direct from the tube at all times, and carries a Zout conservatively rated at 140R for 1Vpp into 20K at 1KHz.  This will drive VERY long cables with impunity.

Cheers,

Hugh

[mb]

Folks,

I should explain how the volume control is implemented in the GK-1 as its operation in this circuit is a little unusual since most preamps have their attenuator right at the input.
....

Thanks Hugh,

Your explanation makes me really want to re-audition EchiDna's GK-1, now that it should be nicely run-in...

[AKSA]

Ah, Mervin,

There is always method in my Aussie madness!    

Cheers,

Hugh

[EchiDna]

Folks,

I should explain how the volume control is implemented in the GK-1 as its operation in this circuit is a little unusual since most preamps have their attenuator right at the input.
....

Thanks Hugh,

Your explanation makes me really want to re-audition EchiDna's GK-1, now that it should be nicely run-in...

drop me a line Mervin, I'm playing with wires in the GK-1 at the moment, but given an hour, it will be back together... the trick is... to find THAT hour *sigh*

[mb]

drop me a line Mervin, I'm playing with wires in the GK-1 at the moment, but given an hour, it will be back together... the trick is... to find THAT hour *sigh*

Thanks Glen. Possibly in a week or two... A couple of small items are running in on my system, and I'll leave them to stabilize before listening to the GK-1. Looking forward to it!

[jesserparker]

hugh,
given the layout of the GK1 i am now thinking of a FULLY buffered gain pot as well.  in the design i have sketched out now, the signal would come in directly to the selector switch, and proceed directly from there to the gain pot, and then the buffers (BUF-3 with an input impedance of 500 trillohms, and an output impedance of just 2 ohms), and then on to the power amp.  however....
like i said, now you have me thinking of fully buffering this.  so here's a few questions:

1- would (in your opinion) a circuit this simple (signal only passes through ONE component more than a purely passive design (though granted, it is an IC)) benfit more from keeping the # of parts to a minimum, and keeping the signal path short, or from the addition of buffering of some kind on the input?

2-if input buffering would then be preferable (and why?) what methods could/should (of course in your opinion again) be used to do so, and what are the advantages/disadvantages of this?

3- what are your general thoughs on this particular design?  here is a link to the schematic minus the PSU which i have modified and don't have the design handy, but suffice to say that it will be dual mono, and more than is probably necessary for this IC, but i want to make sure that i'm getting ALL i can out of this lit'l buffer.  

http://www.stereophile.com/archivesart/scan35.html

thanks,
jesse

[jesserparker]

i should also add that the "optional" resistor in the schematic will not be present in mine, nor will the rec out switch and output.  though there may be some provision for a headphone jack, as this design should drive cans fairly well, and that way i can use it as such until i build myself a "proper" headphone amp, and i can do so without making any kind of adapter for my phones to connect to the signal out, and i won't have to unplug the interconnects to the power amp to use it either.  but i'm still deciding if i want to put that switch in the way of the outgoing signal.  if i do go for it i'll probably simply place a two position switch of some nature after the buffer that switches between the output RCAs and the headphone jack, but we'll see,  if anyone has any thoughts on this...  feel free,
thanks,

jesse

[mb]

Hi Jesse,

I'm looking forward to Hugh's response too, but meanwhile here are some comments regarding your project:

- the current favoured buffers appear to be the Elantec EL2001 and Intersil 5002. Some SOTA headphone projects are using them instead of the BUF634, which is slightly old, and expensive
- since you're building, it should be interesting to try two versions:

1) attenuator -> buffer
2) buffer -> attenuator -> buffer

and make the final decision by ear, as to which works best. I've tried (1) and the inverse (buffer -> attenuator), and (1) is definitely better for me. I've avoided (2), simply because of the extra effort to build

FYI, another AKSA has just built a meta42 headamp, and has written to me with very positive input when using it as a preamp. It uses AD8620 for gain, and EL2001 for output buffering. You can pick up lots of tips reading the meta42 site. The info will be 99% relevant for your project.

[AKSA]

Hi Jesse, Mervin,

This is one helluva question, and probably goes to the core of high end design.

As Bill Shakespeare, the high end audio linguist, once remarked, 'To buffer or not to buffer, that is the question.....'  

1- would (in your opinion) a circuit this simple (signal only passes through ONE component more than a purely passive design (though granted, it is an IC)) benefit more from keeping the # of parts to a minimum, and keeping the signal path short, or from the addition of buffering of some kind on the input?

Jesse, you actually address two pivotal issues here.  Do we use a buffer on the input, and do we use an IC?  There are two facts to consider.  Firstly, we know interconnects have quite considerable impact on sonics, and as modern equipment steadily improves, many of us notice this more and more.  This should not be so (and has spawned religious cults!), but clearly is, and relates to parasitic reactances, mostly capacitance but also inductance, and attendant effects on the following stage, often tied up with short term stability.  RF technology calls attention to the issue of 'terminating impedance', which becomes important at around 500KHz, which happens to be close to the primary pole on most audio equipment using global negative feedback, and which might thus be expected to have cascading effects down into the audio band.  All my work with lag compensation on voltage amplifiers tells me that anything which impacts on this pole frequency, and hence stability, is immediately audible.  But that's another story.......

Secondly, sources are not perfect, and like software, suffer from a myriad of standards, many of which are ancient, irrational, and just plain wrong. (The design of the RCA plug, for one.)  Thus source impedances vary markedly, and so do the phase margins of their output stages.  Stability, as commented earlier, is profoundly affected by terminating reactances.


On consideration, and given that pots tend to be chock full of parasitics (which partly explains why there is such huge sonic variance from pot to pot), it seems logical to use a buffer before the attentuation function.  This buffer would need highish input impedance (not too high because this makes the interconnect highly susceptible to hum intrusion), high dynamic range, minimal intermodulation, restricted bandwidth to remove RF which might have injected itself into the interconnect, and vanishingly low output impedance.  Clearly it does not need gain, so an opamp need not be used.  That's some relief, I guess!

This leads you to the buffer IC, which is relatively cheap, but which generally uses at least three cascaded emitter followers, and sometimes even a voltage gain stage, throttled back to unity with nfb.  As Mervin has found, these are damn good for all their complexity, but their output stages are normally Class AB emitter followers in push pull, with all the problems found in conventional amps along these lines and without the powerful charge suckout technology employed in the better SS amps to eradicate the crossover disjunction.  So, this deficiency leads to my next point.

The conflicting requirements of very low output impedance, some bandwidth limitation, single ended class A circuitry throughout and highish input impedance leads inevitably to a differential pair input, a conventional, carefully selected transistor with very low parasitics for a voltage amplifier operating in constant current, and a global negative feedback loop.  Like it or not, the Bailey configuration used in the AKSA is one of the oldest and best configurations for audio that has ever been discovered.  So, we can use it as a buffer, and by dint of nfb extract vanishingly low Zout.  A current source can drive the voltage amp, and since we have Bode/Nyquist stability criteria to meet with any global feedback arrangement, it will be inherently band-limited to around 50KHz.
Another boon to this configuration is low parts count, since just two transistors effectively constitute the signal path.  That's pretty cool, and impossible to beat with an IC, since the manufacturers appear to feel a powerful obligation to use at least five active devices in the signal chain, and this, IMHO, often compromises things sonically.

Jesse, I believe the foregoing answers your questions in para 2.

3- what are your general thoughts on this particular design? here is a link to the schematic minus the PSU which i have modified and don't have the design handy, but suffice to say that it will be dual mono, and more than is probably necessary for this IC, but i want to make sure that i'm getting ALL i can out of this lit'l buffer.  

The design would work fine, but would be quite susceptible to quality variations in the interconnects.  It might also load down the source beyond its comfort zone, since you say you'd like to use 10K pots, or even lower, and while CD players have a Zout around 100R, very few are comfortable with 10K loads.  And the vast majority of tube designs are also uncomfortable with loads this low;  from decades ago most tube amps used a Zin of 47K to give the tube pre an easy life.

I hope this gives you a brief tour of the management of compromise which, like most technologies, dominates high end!    
Cheers,

Hugh

[jesserparker]

than you,
thank you for the speedy and thourough answer.  it seems i can always count on that from you
also, once again i am so glad you actively (very!) participate in this forum because- while i could simply e-mail these q's to you- here everyone who's interested can gain from the answer.

a few follow ups now:

1) as far as interconnects go, i will have quite a bit of flexibility and variety here to try as i "roll my own".  also, i completely agree about the antiquated RCA connector and hence i use bullet plugs (damn you aussies are clever!)  to try and mitigate some of the inherrent pitfalls of the connector (and thus also avoid putting massive amounts of metal in the signal path as the much better BNC connectors will still do).  so quality and flexibility in that area is not a question.  i can tailor the construction of them to  get the exact impedance, capacitance, shielding, whatever i need so i should be able to match whatever characteristics i want to exploit with that.

2)as far as the pot goes i plan on using the vishay/sfernice type for this, law faked (which i initially considered somewhat of a compromise- law faking that is- but upon investigation, some people, thorsten loesch eg, consider a law faked linear pot to give superior performance to a log pot anyhow ), and to correct you on one point;  i did not want to use a 10k pot, but actually wanted to go the other route and take advantage of the fact that the buffer will give me such vanishingly low output impedance and use a 100K pot (upon further consideration i'm thinking between 20 and 50k may be better, but at any rate, i'll be using an easier load than the 10k so my sources shouldn't have any problem at all driving that.)

3) as far as the AB classification of my buffers, apparently they are heavily biased to class A (and consequently run quite hot), so that is alright.  i did hear mention of someone building a discreet component equivalent which (though it would lengthen the signal path considerably) would allow a degree of flexibility here. but to be honest, i have no idea how to do this (point me in the right direction hugh??? )

4)and winding up for today...  hugh, what would you say is the advantage a fully buffered gain stage (input>BUF>gain>BUF) over input>gain>BUF?  this in particular i'm interested to hear.

thanks again for your participation and immense help in this thread, cheers,

jesse

[jesserparker]

hi again Hugh,
last thing i want to do is be a nag, but my guess is that this post somehow slipped between the cracks and all i'm trying to do is bring it back up to surface where hopefully you'll see it  

cheers,
jesse

[AKSA]

Hi Jesse,

Yes, did slip through the cracks.  My apologies.

2

)as far as the pot goes i plan on using the vishay/sfernice type for this, law faked (which i initially considered somewhat of a compromise- law faking that is- but upon investigation, some people, thorsten loesch eg, consider a law faked linear pot to give superior performance to a log pot anyhow  ), and to correct you on one point; i did not want to use a 10k pot, but actually wanted to go the other route and take advantage of the fact that the buffer will give me such vanishingly low output impedance and use a 100K pot (upon further consideration i'm thinking between 20 and 50k may be better, but at any rate, i'll be using an easier load than the 10k so my sources shouldn't have any problem at all driving that.)

I would go around 50K here;  the higher you go, however, the less the loading on the previous stage, but the more susceptibility to hum intrusion.

3

) as far as the AB classification of my buffers, apparently they are heavily biased to class A (and consequently run quite hot), so that is alright. i did hear mention of someone building a discreet component equivalent which (though it would lengthen the signal path considerably) would allow a degree of flexibility here. but to be honest, i have no idea how to do this (point me in the right direction hugh???  )

It can be done, but if your existing buffer is Class A, there may not be much improvement, although the myriad devices in the signal chain worry me a little.

4)and winding up for today... hugh, what would you say is the advantage a fully buffered gain stage (input>BUF>gain>BUF) over input>gain>BUF? this in particular i'm interested to hear.

I think the former is preferable, but at some complexity, which I'd ameliorate with discrete buffer circuitry.  If you go straight from the input to the attenuator and then to the buffer, you might risk making the system susceptible to hum and interconnect choice.  You'd need around a 50K pot at the input;  this is a healthy compromise which does not overlay load the previous stage.

Hope this is helpful, and sorry about the delay!

Cheers,

Hugh