Valve Shunt Regulator Design Walk-Through

[Mike H]

Tried 6N1P again, but sim hangs up again with stupid Voltages around those two triodes, so something is not right with that model. Or it just can't cope with them being connected like that, and falls over.

I wouldn't worry too much that ECC88 & 6N1P are not identical, I doubt the models are that close to real life anyway, as it is you will still have to jiggle the R2 value in the real circuit to bias it right.

We still need to find out what this oscillation is all about.

I think it's a time constant somewhere in the signal path, so that the KT88 response lags behind the control amp, so that it keeps overshooting back and forth.

[Mike H]

Sussed it -

The oscillation was about 6.5mV pk-pk @ 3.5 Hz.

Change C6 to 10uF and it stops.

The ripple, @ 100 Hz, is now about 30 uV pk-pk !

[Mike H]

Output impedance -

Had a play with this, I think it's mostly just the AC impedance of C3.

I applied 100mA AC in parallel to the load of 1.6k.

If C3 = 100uF, 16 Ω @ 100 Hz, 1.6Ω @ 1 kHz, 220mΩ @ 10 kHz

If C3 = 220uF, 7.2 Ω @ 100 Hz, 0.7 Ω @ 1 kHz, 116mΩ @ 10 kHz

[IslandPink]

Sussed it -
The oscillation was about 6.5mV pk-pk @ 3.5 Hz.
Change C6 to 10uF and it stops.
The ripple, @ 100 Hz, is now about 30 uV pk-pk !

Nice work !
I'm just trying it in my version of the PS to see what happens there. I'm trying a few values though . 2.2uF wasn't enough to change anything, 6.8uF did something very odd... but will report back later.
I also intend to try swapping in some different models on the valves to see what does & doesn't make a difference.

[IslandPink]

No, I think it takes too long to charge up the 10u through the 50k, the B+ volts are still rising after 5 sec or so :

I'll have a look at your circuit again, and see why it would be different .

( On the output impedance thing, I'll come back to that - but there is some interest in making sure most of the AC current prefers to go through the shunt valve, rather than the final cap, which sparked my interest in this circuit )

[IslandPink]

Did a analysis of Zout for the circuit, and it now seems more like 140ohms at 10Hz , dropping as you go up . Not sure why it should be so different . There are reasons why I want low Zout , to do with return current path.
Will investigate

This was an error on my part, leaving in the AC source, rectifier section for the Zout analysis . It just confuses the program - best to put a plain DC voltage source at the start of the first choke to do this and analyse as I described before. Comes out OK .

Interestingly the 100uF cap in the shunt section can be reduced & reduced to eg. 1uF , then you find the Zout of the KT88/cascode combo is about 4.5R at 10Hz and 1.25R at ~80Hz upwards, which is pretty impressive. What Nick and I discussed initially about this PS is that if the shunt Zout is low enough , it will dominate the return AC current path, vs. the 100uF cap - which should give a cleaner tonal character ( I think ! ) . Normally shunt regs are more like 5R-10R , in which case the character of the cap dominates.

ps. I put in Bench's 6N1P now, which works OK . Still need to recap on your comments about the diodes and your 'fix' with the new file you sent.

[Mike H]

No, I think it takes too long to charge up the 10u through the 50k, the B+ volts are still rising after 5 sec or so :

That's still your old circuit though isn't it? That doesn't work. You should be using my last new one.

[IslandPink]

Sir Yes Sir !!
I know, I was trying to hang on to my 6N1P and the voltage divider current-adjust & suchlike.
But also when I ran your file the B+ voltage hadn't stabilised even after 10sec, so I couldn't reproduce the really low ripple figures you say you've achieved.

Let me see if I can have a fiddle ....

[IslandPink]

Here's something i wanted to do - analyse the gain of the cascode on its own . Here's that section of the circuit . I fed the valves from a 440V DC source. The signal 'drive' via the 100uF cap to the cathode of the lower 6N1P section is driven with 0.01V amplitude AC at 100Hz .
The results on the graph show about 3.3V AC on the anode of the upper 6N1P , a gain of 330 . Nice !

I might do it for another frequency though, too .... but for a much shorter time-span, as this plot took an age !

[IslandPink]

Around x340 at 1kHz and 5kHz .
Only x160 at 10Hz : so I guess this is one mechanism for the higher Zout at 10Hz compared to 1kHz .

The 1uF output cap doesn't affect this .
Aha ! - If I increase the feed cap from the source to 220u instead of 100u, I get a gain of x260 at 10Hz .

[Mike H]

I might do it for another frequency though, too .... but for a much shorter time-span, as this plot took an age !

That's because I sure there's something wrong with that 6N1P model


If you used ECC88 it would run like lightning

[IslandPink]

He he !
Yes, you have a point there .
I wonder why both models of the 6N1P are troublesome ? - I'm using the Bench one at the moment . It seems such an innocuous valve, nothing too odd - should be just a minor variation on the ECC88 data ...

Output impedance -
Had a play with this, I think it's mostly just the AC impedance of C3.
I applied 100mA AC in parallel to the load of 1.6k.
If C3 = 100uF, 16 Ω @ 100 Hz, 1.6Ω @ 1 kHz, 220mΩ @ 10 kHz
If C3 = 220uF, 7.2 Ω @ 100 Hz, 0.7 Ω @ 1 kHz, 116mΩ @ 10 kHz

Ah, but a cap of that value will likely have an ESR of about 1.5R , if it's an electrolytic .

[Mike H]

Alright OK, so I'd got 100m for the ESR.

So it could be worse in fact.


As for the models, they're written in completely different ways, it's not just because one is 6N1P and the other is ECC88.

The triodenh one is much more complicated than the Koren version, nine E and two G elements compared to one of each for the Koren.

What this needs is a Koren version of 6N1P

[Mike H]

OK well I've just looked at two datasheet graphs for loadlines for 6N1P (Russian datasheet from http: // www.tubes.ru/techinfo/HiFiAudio/6n1p.html) and ECC88 (Philips datasheet), the data points Ia vs Va @ Vg are identical. As far as a LTspice model is concerned, all you can do is try to put these data points at the same places as the graph of the valve you're trying to model, as close as you can.

Mind you I haven't verified the accuracy of this ECC88 model against a datasheet.

[IslandPink]

Ah. I had a feeling Stephie's models would be more sophisticated. However, looks like it's horse for courses , as Nick suggested to me - Some of the analyses can't run properly if the model is too complex .

It's not essential to do everything with one circuit . I'll use your PS model with the ECC88's and mine with the 6N1P's depending on what I want.

The remainder of what I want to do is -
1) Run a DC operating point analysis on the circuit with Stephie's 6N1P's - to get final stab at the correct currents and resistor values for the 6N1P's to match the T-Rex sketch
2) Repeat the above for a lower voltage input ( and 320V output ) for the 2A3 supply case. This'll need some different cathode resistors & divider, I expect .
3) Run the circuit ( probably yours ) to assess some step functions or pulse functions, some with big current changes, to see if I can find out what happens when the load varies more than the KT88 current .

One loose end is why I can't replicate that 30uV residual noise, with no meandering, that you say you got with the full circuit, with ECC88 cascode. Let me check & try again .