I've had a breakthrough in the 801A A2 amp design, a new path forward.
But first, a mistake was made. Remember when I said I was getting a 0.38ohm output impedance? That turned out to be false LOL I thought it was suspiciously low at the time given that I had not drastically altered the feedback fraction, those results were not reproducible, so something was off at the time.
I am more consistently getting output impedances around 2.8ohm with various feedback setups I have tried with the resistively loaded pentode stage. This is borderline high, I am seeking a damping factor of 3 or more for 8ohm speakers, so 2.8ohms on the output ain't really gonna cut it.
So what to do. Altering the degree of NFB via my cathode trimmers improves the output impedance, albeit marginally, but increasing it too much and the stage does not have enough gain to drive the 801A to clipping. As I mentioned, I have left the cathode of the EF37A unbypassed / degenerated, meaning there is internal NFB, which reduces the gain. I decided to try bypassing the cathode with a 500uF electrolytic to see what it got me...I also added a film cap to my feedback loop to take the DC current out of the equation, making the bias process of the EF37A much more straightforward.
Open loop mu of the EF37A stage jumped from 40 to 100 with the cathode bypassed. Closed loop measurements in this setup below.
THD @ 1W 1%
Output Z: 2.8ohms
-2.4dB @ 20kHz
So we have some improvement in THD, output Z is largely the same, rolloff at 20kHz is unchanged but improved at 17kHz.
How about the subjective sound? A step back, no question, as there is now an electrolytic in the signal path
less detail, degree of realism and space has diminished, more "smeared".
So now what? I need more gain out of my input stage to increase NFB, lower the output impedance for proper 8ohm damping, and continue to improve bandwidth. At this point though, I am reaching the limits of what can be done with a resistor loaded stage, using 220K on the plate of the EF37A.
I did spend all that time learning about CCS loaded pentodes...
LET'S DO IT
One of the difficulties with using one of these stages is dialing in the CCS to provide the anode current for the pentode in addition to the plate load resistor, which is grounded. The benefit is the ability to use a higher resistor value that is typically practical on the plate of the pentode, continuing to flatten the load line and increase gain of the stage beyond the quoted specification. In addition, the CCS provides the same PSRR you could expect in a CCS loaded triode stage (not that it is needed in this amp, with the Maida regulator + CCS, ripple is being crushed to the tune of >200dB PSRR!).
The disadvantage is DC bias point stability. With a nearly flat load line, small changes in cathode voltage result in big swings in bias point, so there is the potential for bias point drift. I figured if I used a larger plate load resistor without going nuts, I could get away with it. I settled on 470K.
So, the CCS stage was simulated in LTSpice to get ballpark figures, and away we went. Here is what it looks like now. Yup it is an absolute disaster, but hey its a prototype on a piece of wood
I needed to throw a 1K resistor in series with my CCS trimmers to get the current down to EF37A levels.
After some tweaking, dialing-in of voltages / currents, we settled on the following EF37A bias with the CCS in place: CCS current 1.56mA / Va 240 / Vg2 95V / Vk 2.77V.
I measured the open loop mu of the stage
without a bypass cap, it went from 40 to 125!!! Eff off electrolytic, I don't need you
now getting more open loop gain than with the resistively loaded stage WITH bypass. As a result, the input was driving the 801A to clipping and then some, allowing me to further increase the feedback fraction.
So, how about the measurements...
THD @ 1W 0.44%
Output Z 2.44ohm
-1.67dB @ 20kHz
1kHz sine wave at full output, 18.06Vpp into 8ohms = 5.2W
Yeah...big gains across the board. Now will get a >3 damping factor for 8ohm speakers, further HF extension, lower THD, and I got my 5 watts.
The sound is back to its airy, realistic, hyper-detailed state with the electrolytic out of the picture.
I monitored the plate and cathode voltages of the EF37A with music playing at varying volumes, did several shut-downs and start-ups as well - the bias does not drift, it finds its way right back to the same point every time.
So this is the way forward, keeping the CCS, more tweaking and optimization to come, but it is undoubtedly an improvement.
Just to show what further gains could be had
measurements wise if the plate load resistor is further increased (which will in turn increase the gain of the stage and NFB), here are the measurements repeated with the cathode bypassed with the 500uF electrolytic with the CCS in place.
THD @ 1W 0.24%
Output Z 2.15ohm
-1.19dB @ 20kHz
Will continue to improve the design, but it's time to call it a day!