I got inspired to do some simulations to compare:
I ran them all with a 1kHz sine wave input so that they made 1V rms into a simple 50k ohm resistive load (no shunt capacitance in its load). What I found was:
Substituting a 12AU7 in place of the 6922 doubles the THD (the SLCF with 12AU7 made 0.002% THD, compared to 0.001% with the 6922).
- a standard fixed bias 6922 cathode follower with resistor cathode load and 150V PSU
- the same cathode follower but with a simple DN2540 CCS in its cathode
- the cathode follower with DN2540 CCS in its cathode made into a Super Linear Cathode Follower (SLCF) with an IRF610 in its plate
- that SLCF with an IRF610 replacing the 6922, making it a 'Super Linear Source Follower'.
I ran them all with a 1kHz sine wave input so that they made 1V rms into a simple 50k ohm resistive load (no shunt capacitance in its load). What I found was:
- the 6922 cathode follower with resistive load yielded 0.015% THD (that's quite good)
- the 6922 cathode follower with DN2540 CCS for its cathode load made 0.006% THD (better)
- the 6922 SLCF made 0.001% THD (even better)
- the 'Super Linear Source Follower' made 0.0003% THD, but needed a larger 1.5k build out resistor to tame a peak in its output at 3MHz.
Substituting a 12AU7 in place of the 6922 doubles the THD (the SLCF with 12AU7 made 0.002% THD, compared to 0.001% with the 6922).
Interested to know what a plain resistor loaded source follower (IRF610) gives. Given the flat curves of a mosfet (above a few volts Vds) the SLCF mod might not be doing much.
Interesting comparisons, what to say...this term 'super linear source follower' is new to me ( I am less familiar with the transistor field). Regarding the tubes, better behavior of the 6922 was somehow expected.
If you feel like sharing the .asc file for simulation it would be good.
If you feel like sharing the .asc file for simulation it would be good.
When you wrote "the SLCF mod might not be doing much". did you mean that adding the MOSFET in the plate of the 6922 doesn't do much, or did you mean changing the 6922 to a MOSFET might not be doing much?
I won't be able to run a new simulation until I'm home for the evening, but I'll get back to you on that.
I have four different .asc files, one for each of the above circuits (triode CF w/ resistor, triode CF w/ DN2540 in cathode, triode SLCF, and MOSFET 'SLSF').
It would be interesting to see if the performance was much effected by harder loads. Many amps have 10k input impedance and some have capacitive elements.
I always test with 10k ohm load and 500pF or so parallel capacitance, to simulate a solid state amp with lower input impedance and a long run of cheap interconnect cable.
I can repeat the simulations with a 10k||1nF load. I'll do that this evening. Will be interesting to see how much things change from the 50k load, and also if different loading affects how prone to oscillation the SLCF is. Simulations are suggesting that the SLCF wants to oscillate at around 3 to 6 MHz.
Of course the quality of the models used has a lot to do with whether these findings are relevant. I try to update the models in an attempt to use the best available. I'm using Adrian Immler's 12AU7 and 6922 models, but I don't know if I'm using the best available models for IRF610, DN2540. etc. MOSFETs.
I can repeat the simulations with a 10k||1nF load. I'll do that this evening. Will be interesting to see how much things change from the 50k load, and also if different loading affects how prone to oscillation the SLCF is. Simulations are suggesting that the SLCF wants to oscillate at around 3 to 6 MHz.
Of course the quality of the models used has a lot to do with whether these findings are relevant. I try to update the models in an attempt to use the best available. I'm using Adrian Immler's 12AU7 and 6922 models, but I don't know if I'm using the best available models for IRF610, DN2540. etc. MOSFETs.
I would be interested in the third and fourth: the latter being practically all S.S. it would also deviate from this forum but out of mere curiosity I would like to see the circuit arrangement.
Abouut the second one, if you want to take a look here for some additional inspiration (a BJT was used):
http://www.valveradio.net/audio/low-distortion-cathode-follower.html
The matter is not completely clear to me, especially in the ways it was implemented and the choice for the particular voltage setting of the C.F.
Thanks in advance
I simulated with 100pF and 1nF: only small difference in bandwidth, none in distortion. The fall-off with 1nF is different, less gradual; with 1nF is extended within 0,2 dB a bit longer till 66kHz. With 100p it is 45 kHz 0,2 dB.
Has any of you coparisons with the CARY-AE1 circuit, using 6SN7? Is transposable to ECC82 . . .
Has any of you coparisons with the CARY-AE1 circuit, using 6SN7? Is transposable to ECC82 . . .
I took a look at the CARY-AE1, in practice if I am not mistaken it matches in DC a common cathode amplifier with a cathode follower (via double triode), with the help of an amount of feedback to stabilize. It's correct? I didn't quite understand the role of C = 0.1uF, perhaps compensatory...
I meant that if you use a mosfet as a source follower, adding another one on top to keep Vds constant ("SLSF" as in your fourth test) might not improve distortion much. Unlike for a triode cathode follower where Vak has a much larger effect on operation.
Let me see if I understand that correctly...
The first stage is a compound grounded grid amplifier with the first 6SN7 (V1) being a cathode follower driving the cathode as the input of the second 6SN7 (V2, grounded grid amplifier).
The third (V3) and fourth (V4) 6SN7 sections are paralleled and configured as a plain old cathode follower, operating at a fairly low plate current.
NFB is taken from the cathode of V3||V4 to the grid of V2.
Since V1/V2 doesn't have much gain, and V3||V4 is a cathode follower, there's probably not a whole lot of gain available for NFB from the cathodes of V3||V4 back to the grid of V2.
It looks like the PSU uses separate capacitor multipliers for V1/V2 and V3/V4.
I suppose this circuit could be implemented using 12AU7s, with the same parts values.
The following circuit is the Broskie Aikido, which uses exactly the same number of triodes per channel (shown here using 12SN7s):
This one would be interesting because using 12AU7s it would end up with a gain of about 8X.
The first stage is a totem-pole 'half-mu' stage. It will have exactly half the mu of the 12AU7, so approx. 16/2 = 8.
The second stage is a cathode follower with triode active load.
The 47uf and two 100k resistors from B+ to ground before the output is the Aikido PSU noise feed-forward network (the "Aikido" concept).
The first stage is a compound grounded grid amplifier with the first 6SN7 (V1) being a cathode follower driving the cathode as the input of the second 6SN7 (V2, grounded grid amplifier).
The third (V3) and fourth (V4) 6SN7 sections are paralleled and configured as a plain old cathode follower, operating at a fairly low plate current.
NFB is taken from the cathode of V3||V4 to the grid of V2.
Since V1/V2 doesn't have much gain, and V3||V4 is a cathode follower, there's probably not a whole lot of gain available for NFB from the cathodes of V3||V4 back to the grid of V2.
It looks like the PSU uses separate capacitor multipliers for V1/V2 and V3/V4.
I suppose this circuit could be implemented using 12AU7s, with the same parts values.
The following circuit is the Broskie Aikido, which uses exactly the same number of triodes per channel (shown here using 12SN7s):
This one would be interesting because using 12AU7s it would end up with a gain of about 8X.
The first stage is a totem-pole 'half-mu' stage. It will have exactly half the mu of the 12AU7, so approx. 16/2 = 8.
The second stage is a cathode follower with triode active load.
The 47uf and two 100k resistors from B+ to ground before the output is the Aikido PSU noise feed-forward network (the "Aikido" concept).
Last edited:
I think that Cary circuit also sports one of Broskies acronyms 😎
The Aikido circuit is nice; I think it amplifies close to 20 dB; it can run on a lower Vb too because it is always "symmetric" in distortion top/bottom.
There is an older implementation from around 1984 of this two stage, where the second lower part is in parallel with the input *); the top is driven like here by the CCS. This results in a very low output impedance and lineairity; that circuit from 1984 about was once ised to drive a 211 power tube using ECC83: 200V peak-peak with less than 1% distortion ...
*) So the .47uF is not there and the grids are interconnected.
The Aikido circuit is nice; I think it amplifies close to 20 dB; it can run on a lower Vb too because it is always "symmetric" in distortion top/bottom.
There is an older implementation from around 1984 of this two stage, where the second lower part is in parallel with the input *); the top is driven like here by the CCS. This results in a very low output impedance and lineairity; that circuit from 1984 about was once ised to drive a 211 power tube using ECC83: 200V peak-peak with less than 1% distortion ...
*) So the .47uF is not there and the grids are interconnected.
Sorry, I couldn't follow that.
"the second part is in parallel with the input"?
"the top is driven like here by the CCS"?
"the grids are interconnected"?
OK, I took simple fixed bias source follower using an IRF610 with B+ of 150V, 1M/1M voltage divider from input to gate for 75V at the gate, about 70V at the source. Source resistor = 13k, for Ids of 5.44mA. This was driving a 10k load. Gain was as expected, around 0.95x. THD was 0.0048%.
One interesting thing is that the frequency response shows a rounded bump up a couple of dBs centered at 4MHz. I had to increase the value of the build out resistor to 470R, which reduced the gain, but did not change the THD.
Next, I put a DN2540 MOSFET in the cathode, set to 5.3mA. I had to increase the value of the build out resistor to 680R to tame that 4MHz bump. That took the gain down to 0.935dB, but the THD went down to only 0.0016%. That's really low distortion.
Finally, I put a second IRF610 into the drain circuit of the IRF610 source follower, to make it a 'Super Linear Source Follower'. A couple of things had to change a lot. First, to fit three MOSFETs into that 150V B+, I had to change the input voltage divider from 1M/1M to 2.2M/1M, so that there's now 42.5V at the gate of the source follower MOSFET. This allows me to put between 40V to 60V across each MOSFET (Vds). Another thing that changed was that the bump in ultrasonic response up at 4MHz became much more pronounced and with a much sharper peak (the resonance has a much high Q). After all that was in place, The gain was again about 0.95x and the THD at 1Vrms out was down at 0.0017%. That's five times worse than into 50k ohms, but it's still op-amp low.
That's what I found.
Last edited:
OK, back to the cathode follower circuits, but this time with a more difficult load of 10k and 1nF.
- the 6922 cathode follower with cathode resistor load yielded 0.044% THD -- about three times worse than with the 50k load. Response at 20Hz was -0.15dB.
- the 6922 cathode follower with DN2540 CCS for its cathode load made 0.026% THD -- about four times as much as with the 50k load.
- the 6922 SLCF made 0.0147% THD -- almost 15 times worse than with a 50k ohm load, but still quite low.
- the 'Super Linear Source Follower' made 0.0017% THD. That's five times worse than into a 50k load, but still really, really low. Gain was reduced to below 0.85x (point zero eight five).
Transistors make much better buffers than triodes, because of their much higher transconductance.
I went back and looked at the 12AU7 Bootstrapped Pair with Shunt NFB (no MOSFETs) with 300V B+. working into a 10k ohm load. Gain is 2.3x, THD is 0.009%. That is better than the 6922 SLCF with 150V B+, and you get some overall gain too.
If you take that 12AU7 Bootstrapped Pair with Shunt NFB and use a MOSFET source follower instead of the cathode follower, things should get even better (the MOSFET will drive both the load and the feedback loop better).
Redesign the Boostrapped Pair with Shunt NFB to use a 12AT7 input tube and a MOSFET source follower and you'd get even better results (the 12AT7 has somewhat higher gm and much higher mu than the 12AU7).
Maybe try it with a 6DJ8, since that has higher gm than 12AT7.
Last edited:
Here are the .asc files.
6DJ8 Cathode Follower-Buffer_+150V_03 is the plain cathode follower with cathode load resistor.
6DJ8 Cathode Follower-Buffer_+150V_03A is the cathode follower with DN2540 MOSFET as the cathode load.
6DJ8 Cathode Follower-Buffer_+150V_03B is the Super Linear Cathode Follower.
6DJ8 Cathode Follower-Buffer_+150V_03C is the Super Linear Source Follower (all MOSFETs, no tube).
Your models probably won't be the same as the ones I used. Substitute the 6922, 6DJ8, E88CC, ECC88 of your choice. Also, you may need to supply your own DN2540 model.
6DJ8 Cathode Follower-Buffer_+150V_03 is the plain cathode follower with cathode load resistor.
6DJ8 Cathode Follower-Buffer_+150V_03A is the cathode follower with DN2540 MOSFET as the cathode load.
6DJ8 Cathode Follower-Buffer_+150V_03B is the Super Linear Cathode Follower.
6DJ8 Cathode Follower-Buffer_+150V_03C is the Super Linear Source Follower (all MOSFETs, no tube).
Your models probably won't be the same as the ones I used. Substitute the 6922, 6DJ8, E88CC, ECC88 of your choice. Also, you may need to supply your own DN2540 model.
Thanks for that. Playing around with the fourth simulation, the CCS is a big improvement over resistor loaded source follower, but the upper mosfet (SLSF) actually makes it slightly worse compared to CCS loaded (similar results to those in post #195). All of them good compared to tubes though 😉
The source follower version has to many advantages not to use. Put a jfet up front and it's the ultimate preamp circuit.
I compared the previous #67 (which I have running now) with the 6DJ8 Cathode Follower-Buffer_+150V_03B - the Super Linear Cathode Follower. For comparison, I shifted the output line by dividing it in LTSpice. Both loaded with 1 nF. [I shorted the constant voltage source with R=1 ohm, so what is in the picture is not in the sim, for your comprehension]
Last edited: