I'm waiting for parts to come in to breadboard this. In the meantime, I'd love to hear any thoughts.
Notes:
Notes:
- Tubes are 6SF5 (half a 12AX7), E1148 (slight variation on 7193/2C22, so half a 6SN7), and 6DY7 (two beam tetrodes in one bottle, shared cathode and screens).
- This is my first design with a CCS that isn't under an LTP. Not sure I'm injecting GNFB in the right place, but LTspice says it's effective there. The CCS is set for about 1.34mA, which comes in at 1.4V at the 6SG5 cathode at idle.
- At a minimum, screens and bias will be regulated. I'll probably regulate the whole thing.
- LTspice says this will generate 22W at 1.2% THD from a 1VRMS input. However, that's probably over-optimistic, because it's got the plates swinging to 30V on the low side and the same distance from 2*B+ on the high side. Oh, and I also don't have series resistance in the OT model.
- The OT inductance ratios are correct for the modeled 16R speaker load. That's an artifact from when I was exploring different OT options, because this wants an oddball impedance (14k). I'll fix up the numbers for the 8R intended load.
- LTspice says the tubes are idling at about 19mA. That's very cold - half of the dissipation limit. The datasheet recommends -20V bias for 400V on the plates, but the only commercial amp to ever use this tube (Stromberg-Carlson ASP-422) indeed sits all the way down at -29V. I'm pretty skeptical - will have to work this out on the breadboard.
You have two CR networks plus the OPT in the forward gain path - that may get ugly for low frequency GNFB stability. Removing one CR network by dc coupling the 1st and 2nd stages would reduce that stability concern.
You have no designed-in means to reduce high frequency forward path gain, so stability will be at the mercy of whatever parasitic roll-offs exist (including in the OPT), and the level of GNFB applied.
You have no designed-in means to reduce high frequency forward path gain, so stability will be at the mercy of whatever parasitic roll-offs exist (including in the OPT), and the level of GNFB applied.
HAHAHAHA YOU FELL FOR IT! I MADE YOU ALL THINK I DON'T KNOW WHAT I'M DOING!
Oh wait, I actually don't know what I'm doing. Oops.
I'll take the L on this one. I had a couple things backward in my mind, and LTspice happily confirmed me. I think I get it now.
Replacing the CCS with a single resistor (see below) gets me almost identical results: 22.5W at 1.5% THD. Maybe I'll play with putting the CCS on the plate, maybe I'll just walk away now.
I tried. I couldn't pull the U1 plate low enough to avoid having U2's grid positive wrt its cathode. (Actually, I did manage, but I had to run a 220k plate load on U1, which made the gain too high for U2 to be able to do anything with the signal.)
Maybe if I put the CCS in the load? Hmm ...
It definitely makes it easier for people who don't quite understand what they're doing (e.g. me) to think that they do. Especially when it confirms the incorrect understanding.
Can't do one. Something in the 6DY7 model makes LTspice very twitchy, and I have to have *just* the right settings to get a transient analysis. No matter what I do, AC analysis fails with a "singular matrix" error.
The thing is, with the GNFB injection point I designed, LTspice was saying it was effective. Probably for the same reason it acted like my infinite-impedance cathode was perfectly fine.
Naturally, with a bias resistor instead, the typical GNFB injection point works fine.
Oh wait, I actually don't know what I'm doing. Oops.
I'll take the L on this one. I had a couple things backward in my mind, and LTspice happily confirmed me. I think I get it now.
Replacing the CCS with a single resistor (see below) gets me almost identical results: 22.5W at 1.5% THD. Maybe I'll play with putting the CCS on the plate, maybe I'll just walk away now.
I tried. I couldn't pull the U1 plate low enough to avoid having U2's grid positive wrt its cathode. (Actually, I did manage, but I had to run a 220k plate load on U1, which made the gain too high for U2 to be able to do anything with the signal.)
Maybe if I put the CCS in the load? Hmm ...
It definitely makes it easier for people who don't quite understand what they're doing (e.g. me) to think that they do. Especially when it confirms the incorrect understanding.
Can't do one. Something in the 6DY7 model makes LTspice very twitchy, and I have to have *just* the right settings to get a transient analysis. No matter what I do, AC analysis fails with a "singular matrix" error.
The thing is, with the GNFB injection point I designed, LTspice was saying it was effective. Probably for the same reason it acted like my infinite-impedance cathode was perfectly fine.
Naturally, with a bias resistor instead, the typical GNFB injection point works fine.
I figured out the right voodoo incantations to get this. It's -0.5dB from 33Hz to 55kHz, and -3dB from 12Hz to 150kHz.
Basically …
Apart from it not being terribly interesting (which really is not a fault!!!), it is a nearly bog-standard amplifier. With decent quality components, fastidious detail in layout, a good star grounding methodology, and a heavy output transformer, i'll produce darn decent results.
Am I missing something?
⋅-=≡ GoatGuy ✓ ≡=-⋅
A straight forward V-amp in stage 1 (with an unbypassed CCS error on cathode)
A similarly straight forward phase inverter, almost unity gain
And a pair of Tetrodes in PP
… with 'the usual' amount of GNFB
A similarly straight forward phase inverter, almost unity gain
And a pair of Tetrodes in PP
… with 'the usual' amount of GNFB
Apart from it not being terribly interesting (which really is not a fault!!!), it is a nearly bog-standard amplifier. With decent quality components, fastidious detail in layout, a good star grounding methodology, and a heavy output transformer, i'll produce darn decent results.
Am I missing something?
⋅-=≡ GoatGuy ✓ ≡=-⋅
Spice should be used to verify and debug a well reasoned design, not to throw s**t at the wall until something sticks
Haha, yeah, guilty, at least partially.
I agree that it doesn't substitute for proper, well-thought-out design. But it's also really useful as a learning tool, to try things out from my other learning resources and see the outcome of different circuits. That's what this was. I obviously didn't (probably still don't) fully understand the operation of a CCS, having only used it in an LTP before. So I tried something in spice to see what would happen, like I've done dozens of times for other things I needed to understand better. And this time it bit me.
Not really. I mean, the dual-tetrode-in-one-bottle thing is a little unusual, but even that is common enough. The only atypical thing I haven't talked much about is that I'm using toroids for OTs - a pair on each channel, with the primaries wired up so as to null out any unbalanced DC current.
Which, incidentally, I finally added into the sim properly tonight, with a disappointing drop in output power to 17W. I thought I had an equivalent mocked up properly, but I guess not. I did have to compromise on the impedance - I've got 11.7k, it should be 14k. It was the closest I could get. If it ends up really not being enough clean power, I'll switch to a 6.6k trafo and put 8R speakers on the 4R tap.
I find that is because there are unconnected nodes, although I always put resistances in for the caps and power etc (Cap ESR etc being important for bode plots).
I find I have to delete any spare unconnected components.
Last edited:
Spice should be used to verify and debug a well reasoned design, not to throw s**t at the wall until something sticks
“Agile” development