I finished my first SRX+ build yesterday. It is overall quite satisfying. It doesn't have the breathtaking bass or unforgiving resolution of KGSSHV Carbon, but is rather non-fatiguing for long-term listening.
I took the SRX+ Gerber file Kevin created (Thanks, Kevin!), added larger footprints in order to use the cheaper tube sockets, TO-92 footprints for the DN2540s and moved the heater-lift resistor dividers to the amp board so the wiring between this and the PSU board can be more streamlined. I also changed the small tubes to use 6.3V heater in order to match the surplus power transformer.
The PSU board from Kevin follows JimL's original shunt PSU schematic. I added a normal bias tap in the zener diode string with a copied low pass filter, different lead spacing support on the output film caps in case a certain part number is out of stock. It is also cut short 1/2" from the original.
Got too aggressive in fitting these boards into the small chassis. The internal space is barely large enough for the two boards to lay side-by-side. Had to mill off internal supports here and there, and ended up not having enough space for a real volume pot (the place holder is a rotary encoder waiting for a future attenuator board.
The build started with a faint hum. Swapping tubes can reduce it to a certain level. Then I found a 55mVp-p saw-tooth ripple on the power, which is not supposed to be there (the shunt PSU has >120dB ripple rejection at power line frequency according to simulation). It turns out, the bias circuit voltage doubler drops its leg on the virtual ground and injects >1mA of ripple current, because the bias has to be ground-referenced. The virtual ground is not really a low-impedance node (about 60 ohm @ 60Hz with two 22uF caps). Any noise on the virtual ground is considered common mode to the shunt regulator, and is pretty much out of the control loop. Besides, with transformer HV voltage suitable for this design (I used 600V center-tapped), the voltage doubler doesn't have enough juice such that the 10M90 bottoms out at the low voltage points. So I modified the bias circuit to have its return tied to the output B+ and let the shunt regulator deal with the ripple current. Pro-bias is still easy to obtain, but the normal bias would have to come from dividing the B+ like in the original SRX circuit.
The final assembly has 2.5mV hum in one channel and 5mV in the other. I don't think I can reduce them significantly beyond that without using DC heater and take care of the common mode noise from the header windings, perhaps also need to shield the small tubes. There is always some coupling from the heater to the cathode and it varies from tube to tube. With multiple tubes sharing the same AC heater I don't think you can balance it out completely using a pot, either.
This simple amp performs well, THD+N is as low as 0.01% between 35dBV and 45dBV (40dBV being the rated operating voltage for the SR Lambda Pros to output 100dBSPL, which is pretty loud). One thing I'm not too happy about is the -2dB drop @ 20kHz when loaded with the AP (200k ohm + 66pf including cables). So I played a little with the open-loop performance. The driver stage has only -1.4dB drop @ 20kHz when disconnected from the final tubes, but the Miller cap of the 6SN7 is killing the high output impedance common-grid driver stage, dropping it to -9.5dB @ 20kHz. Negative feedback helped but didn't bring it back to ruler flat. It might help with a cathode follower stage before the 6SN7 but then there goes the simplicity. A rolled-off top end could explain the more forgiving sound, though.
Thanks again to Kevin and Jim to make this such a fun project. If you are going to build it, I'd suggest using Jim's Revised shunt power supply for SRX Plus or any other dual voltage regulators. And don't use a chassis too small!