I have been mulling re-building this amp for three or four years and finally got to it. The philosophical approach is to isolate the audio circuits from the power supply to the greatest extent possible without resorting to batteries. In an attempt to broaden the rectifier conduction angle the supplies use small reservoir caps (i.e. do not take all the current at the AC peak). I have to some extent applied this to the heater / filament supplies too. I have not used chokes because they almost always result in a PSU resonance at 10Hz or so and that is not acceptable to me. These wonderful DH triodes can benefit hugely from a more sophisticated approach to power supply design, my approach is to minimise power supply reactance; this is an approach I first dipped into with the MK11 416 phono-stage, took further with the 6AS7G PP design and now, logically, I am taking it as far as I know how with this SE design.
The raw B+ is developed using HEXFRED diodes charging a mere 15µF film reservoir capacitor. The B+ then encounters a slow-start mosfet series regulator/ripple filter. This brings some 60dB of ripple rejection to the game while allowing for slow start through the simple expedient of putting a long time-constant between the reference voltage and the gate. (Talking of reference voltages, I have decided to adopt a precision current source feeding wire-wound resistors instead of zener diodes, much quieter and so easy to adjust!) The voltage-regulated and clean B+ is then current regulated to the required current to feed each stage using the excellent performance afforded by mosfet cascode current regulation*. There is a separate voltage and current regulator combination for each drive and output stage. Part of what I aimed for here is to accomplish sensible levels of power dissipation in the series pass regulators and the current regulators together with both excellent supply noise rejection and minimal - tending to zero - interaction with the PSU by the signal. Each audio stage power signal path is completed by a high quality oil bypass capacitor the value of which is no bigger than that commensurate with the series inductance and resistance of the power signal path for each stage. I calculated values that yield excellent LF phase response; Any bigger simply results in less agility since the loop energy recovery time is increased. This approach is audible, this amplifier is extremely agile, to my ears anyway. There is no need for electrolytic or monster film caps in the B+ supply.
(* See the notes at the foot of the 6AS7G PP project)
The C- set-up is a little different. This is a voltage source and I have used a simple mosfet series pass device to accomplish this, again, using a CCS and resistor to obtain the reference voltage. Unlike the B+ system, this regulator is not time-delayed. Since there is 'only' one stage of active ripple rejection in this supply I included a RC raw C- filter to present low ripple to the mosfet; ripple and noise at the source of the regulator is well below 1mV. Then there is a 540H choke between the C- and the grid, further improving the isolation of the grid from noise. Regarding bias stability, the output stage grid circuit is designed to provide a low DC resistance path from the grid to the cathode, just conforming to the maximum grid resistor recommendation for a 50 of 10k. Of course, the current regulation design prevents the 50 from grid current induced run-away; However the 50 can still draw grid current and I want to ensure a direct, low DC resistance path from the grid to ground to limit reduction of the grid bias voltage if grid current does flow. Having the grid choke terminated directly to the variable mosfet source-follower augments the bias stability by providing a much lower DC bias voltage source resistance than available from a Zener or gas tube stabilised supply followed by a potentiometer.
The 300Bs I am using (Sophia perforated plates) and the 50s both draw 1.25A filament current and I have taken advantage of that by using current regulated filament supplies with enough voltage headroom to supply the 7.5V filament of the 50s. Taking further advantage of the common filament current, both the output stage and the C- are switchable between conditions for 50s or 300Bs. While it all looks complicated, it really comes down to using very effective functional blocks (source followers and cascode current regulators and heater/filament current regulation) everywhere appropriate so as to feed a simple audio circuit with the best juice. I think only batteries can surpass this approach and I am not (yet) willing to consider batteries for anything other than perhaps, a phono-stage. Click on the thumbnail and note the 50/300B switch!
Using fixed bias on the output stage results in a natural "Ultrapath" topology, without the (known) problem of limited hum and noise rejection that results with a non-bypassed cathode resistor. The B+ caps are of course in the signal path, and so I have bypassed them using the same quality caps as the coupling cap (Auricap, not the best but bloody good). If we look from the plate(s) towards the psu, first we have the bypass capacitor loop, then a Z of 1.5 to 2 megohms due to the current source, looking into around 5 ohms at the series pass source connection. That should pretty much stop any signal current from flowing in the PSU, at least that is my intention; to really detach the amplifier from the power supply, leaving the supply voicing controlled by the bypass capacitors, which again is why I am using Auricaps. Such rigorous isolation of the audio circuit from the PSU should also result in almost no sensitivity to line power quality, line cords or rectifiers!
The drive stage was subject to some experimentation. I started with a choke-loaded 5842 at 15mA, with no cathode bypass cap. The sound of the amp was marvellous, very free, agile and detailed, as good as I have ever heard anywhere; however, the gain was slightly more than I wanted and the available voltage swing is marginal to drive a 300B to full output. I spent many hours looking at tube data, what I wanted was a gain of 30 to 35 coupled with low plate resistance and the ability to swing more than ±100V. There were few choices, at the end of the day I decided to give the 6N1P a try. I have played with this tube before and it is both linear and good sounding. The Magnaquest EXO-99 chokes I am using can be connected in series for 600H (limited to 10mA) or parallel for 150H (limited to 20mA). I tried the series option first with the tubes sections connected in parallel at a total of 10mA. Not bad but the high frequency vivacity was not there. I suspected that the plate resistance of the 6N1P (with no cathode bypass capacitor) was inadequate to drive the capacitance of the choke resulting in dulling of the top end. I put in Blackgate Nx HiQ cathode bypass caps and things improved but overall, the sound did not match what the 5842 could deliver and I did not want to rely on cathode bypass capacitors. Next, I reconnected the choke in parallel (as it had been with the 5842) and increased the total current to 16mA. Much better. The final touch was to include LED bias. I went to Radio-Shack looking for something in the 3.5V range and came home with two white LEDs. These work well* and show a forward drop of 3.4V, just about perfect. Now, the sound was fully up to the standard set by the 5842 (and that takes something special in my experience).
*Note, the resistance of these LEDs appears to increase above 10kHz and so I still ended up using small bypass caps. I will try some known supplier of LEDs to see if this issue is associated with the Radio-Shack parts, I hope so.
The surprise came when I discovered that this drive stage could drive the 50s into class A2! I saw 3.5W at 2% and 7W at 5%, moreover, the transition into overload (beyond 7W) is the smoothest I have ever seen. With 300Bs, the drive stage still limits at the onset of grid current.
This amp is the best I have done to date and fulfils my expectations of the 50 DHT. Now, I will have to get to work and wind new output transformers using the amorphous C cores I have stashed away!
Click to see the schematic: