More Input Stage Madness!!!

What's the "figure of merit" for a triode?

Transconductance, or gm.

Why?

Because every triode wants to maximize voltage gain (µ) while minimizing dynamic plate resistance (rp).  High µ is desired so that fewer stages are needed to net a given amplification factor, while low rp is desired to minimize voltage loss when the driven impedance drops, which increases the high frequency -3dB point.  (These are especially important considerations in designs employing negative feedback.)

Since µ = gm * rp, and gm = µ/rp, the gm is the parameter which describes the extent to which these ideals are attained.  Hence, it's useful as a "figure of merit" (but only for triodes).

The problem is that gm is primarily dependent on the cathode-grid spacing... the smaller this distance, the higher the gm.  It is also increased, although to a smaller extent, as the ratio of cathode-plate to cathode-grid spacing is decreased and as µ is increased.  There is a limit to how close the cathode can be to the grid; at a certain point it becomes too difficult/time consuming/expensive to assemble tube structures with such high precision.

Transconductance is also dependent on the plate current, and will vary proportionally to the 1/3rd power of plate current.  In other words, if the plate current is increased by 8x, the gm will double.  This relationship holds true for all tubes--pentode, triode, etc.

Let me introduce you to a Russian vixen, the 6S45PI:

This tube has an incredible transconductance of 45 mmho--for comparison, the ubiquitous 12AX7's gm is an anemic 1.5 mmho.  Big difference here.  This gigantic gm is what gives this little firecracker the ability to give half the voltage gain of the 12AX7, but cranking it out at about 1/40th of the plate impedance… around 1.5KW instead of 70KW!  In this particular case the former input stage was an EF86, pentode strapped with a 220KW plate load.  Although actual voltage gain dropped from 180x to around 40x (as measured), the output impedance also dropped from about 180KW to less than 1KW.  As I discovered later, this made a huge difference.

I knew that the main B+ rail in the preamp was fixed at 300VDC by virtue of its MOSFET series-pass regulator.  I knew that I wanted maximum undistorted headroom out of the stage, so I set the quiescent plate voltage at B+/2, or 150VDC.

Since greater plate current would net me higher gm and lower rp (two of my design goals), I desired to run the 6S45PI fairly hot.  I also knew that due to its rather high m I would not need very much negative bias on the grid to pinch off the plate current.  (A quick rule of thumb for pinch-off g1 voltage is Vp/m, which, in this case, calculated out to be 150/52, or 2.88V.  Clearly the operating Vg1 would be somewhere about halfway between that and 0V.)

I picked 30 mA as a point that would get me in the ballpark of the rated gm.  To drop 150V across the load at 30mA would require a load resistance of 5KW, conveniently provided by two 10KW, 5W wirewound resistors in parallel.  Wirewound resistors are also the quietest resistors, which is a good reason to use them in an input stage.

I also wanted to implement a different type of cathode bias voltage generation.  In this case, instead of relying on the old cap-bypassed resistance, I went with an unbypassed reverse-biased silicon junction diode.  Actually, after testing a few combinations out, I ended up using what I had handy—two 1N4007’s in series.  The ultimate voltage drop at 30mA after temperature stabilization was 1.3VDC, which worked out perfectly, dropping the requisite 150VDC across the 5KW load resistance.

Even though the dynamic impedance of a forward-biased diode is low, it’s not zero.  When testing the amplifier, I hooked a scope probe straight to the cathode and watched the voltage across the diodes.  It did have a few mV of signal on it, and the waveform was distorted from the original sinusoidal input signal (strong even harmonic present), which meant that performance would be increased by a lower impedance at the cathode; not only would this maximize voltage gain, but it would also reduce noise, and reduce output impedance.

I then attempted to bypass the diode string with a large value electrolytic cap (3200mF@10VDC) but there was no benefit in terms of sonics, and any gain improvement was imperceptible.  Although the circuit was already very low in terms of noise, I was expecting a greater reduction in “blow” and “hiss” than bypassing the diodes resulted in.  Therefore, instead of adding extra components/work/crowding under the hood, I left out the bypass cap.

To preserve the frequency response I had to maintain the RC (or “time”) constant of the plate’s output circuit.  After a bit of experimentation with the capacitor substitution box (you DO have one of those, don’t you?) I settled on a value I had a few spares of, 470nF@400VDC.

Due to the much lower output impedance of my new input stage, I could also do something I had wanted to do before but could not afford to: lower the gain potentiometer’s resistance value, from 1MA to 10KA—a 100x reduction in value!  By virtue of this fact I could also dispense with the small value “bright” or “speedup” caps across the old 1MA placed from hot to wiper to preserve high frequencies at midrange potentiometer settings.  Having replaced those components, I immediately noticed there was much less of a tonal shift as I rotated the gain pots from stop to stop.

Next was to find out what to do with that EF86 stage I just ripped out.  I knew that pentodes distort a little differently than triodes… because of the effects of the intermediate grids, the pentode often has a softer characteristic as it runs out of steam—instead of providing a very linear response that suddenly transitions into clipping, as triodes tend to do, pentodes exhibit a small degree of compression when pushed hard.  They also have much more potential for huge voltage gains.  I decided to follow up the firecracker input stage with a cranked pentode EF86 stage.

Design of this stage was rather straightforward as well.  I went with generic operating conditions—100KW Rl, 95Vg2, 300VB+, and the double 1N4007 forward-biased diode trick for a cathode bias voltage scheme.  With these two cascaded stages, I had about 75dB of voltage gain… in other words, for a 1mVrms input signal I had > 50Vrms on the plate of the EF86.  Not too  shabby at all!

Noise and bandwidth became my next concerns.  Because of the incredibly high gm and low output Z of the 6S45PI, the low value attenuator pots, the low value of Cgp, and the very high voltage gain of the EF86, I had an OUTRAGEOUS frequency response… One oscillation that popped up had a frequency of about 80MHz, yes, EIGHTY MEGAHERTZ.  

That's about 12 octaves higher than I really needed to be!

So, first thing was up the grid stopper on the input to 100KW.  I had a grid-leak resistance of 100KW on the OUTSIDE of the stopper, which meant that the grid circuit DCR was a bit higher than the rated value of 150KW.  All things considered, you don’t want anything connected to g1 but the stopper, and with a very short lead as well, but I may try moving the grid-leak to the inside and testing for stability.  As I later found out, the noise of this tube can be greatly reduced with a lower DCR g1 circuit, but I have not tested this for myself yet.

That, at least, seemed to quench the spontaneous parasitics, so I plugged in and started to crunch around with the amp.  Nice.  VERY nice.  The overall HF response of the entire preamp had gone way up.  I noticed very sharp response; transients like pick swish and percussive palm muting really jumped out of the amp.

Still there was a large amount of HF hiss that had to be tamed.  That’s when I first tried bypassing the cathode diodes, but that was disappointing.  Then I tried placing small value (10pF) ceramic caps from grid to plate on both the 6S45PI and EF86.  I found the more effective position to be across the EF86 Rl, so that’s where I left it, where it did not interfere with the sparkle I had going on with the clean channel.

Except for a few more very small tweaks, I left the rest of the preamp alone and have been playing with it this way for the last four weeks, which is a good sign.  After the EF86 there's another 30dB of additional voltage gain, then a cathode follower, which marks the end of the "intentionally clipping" stages.  The rest may indeed clip, but not by design.  It is due to this 100 or so dB of total voltage gain that the noise issue is really my biggest one, but I have a few tricks still left to try.

Every once in a while a small microphonic oscillation gets triggered while playing cranked up, but a finger-touch quickly quenches it.  The frame grid construction of the 6S45PI means it resonates at a very high frequency which takes some getting used to hearing... in short it is not at all microphonic like your typical mini bottle tube.  I am going to try dampening the vibrations in various ways, one of which involves a Kevlar fiber wrap right around the envelope of the tube, and will post information as I get it.

All in all I must thoroughly thank Jeremy Epstein once again for feeding my habit--he is a real gentleman and I shall definitely pay him back soon for his generosity in sending me two 6S45PI Russian beauties.  There's a small part of me that enjoys the fact I play through an amp that runs its input stage harder than some people run their OUTPUT stages.  With the latest tweaks, the sound is transparent and incredibly revealing of both musical genius and idiocy, which can be difficult and intimidating at first... with more power of resolution comes more responsibility of precise execution, but I wouldn't have it any other way.