KT88 amp

[hylaphone]

Hi folks, I've built an amp similar to the Ampeg B15, but with a KT88 output stage. Sounds nice, but I was anticipating more clean wattage. It starts clipping before reaching 50W, and peaks around 75W squarewave if I raise the input signal slow enough.
Output transformer is the 5K/100W Hammond 1650R. Power supply is 525V plates, separate screen supply at 360V. Biased at 60%.)
The output signal compresses a lot on transients. Plenty of voltage swing from the PI, with the KT88s removed the signals are big and clear. Thinking it might be a loading issue with the 100K grid leaks I tacked in some source followers a la Mosfet follies (with +/-150V supply) to no avail. In fact, this lost a few watts of headroom.
Just wondering if anyone could suggest a next step, I was hoping for 70W clean perhaps that was naive or perhaps I've overlooked something?
Attached is a schematic. Thank you!

[Helmholtz]

What is your B+ (OT CT) at the onset of clipping?

With a bias of -46V you need a clean signal of 92Vpp at the KT88 grids for full output

Does the upper 6SL7 grid have a DC ground reference?

[hylaphone]

Thanks for the swift reply.

What is your B+ (OT CT) at the onset of clipping?

507V

With a bias of -46V you need a clean signal of 92Vpp at the KT88 grids for full output

With the KT88 removed, I have 120Vpp from the PI. With the KT88 installed, 70Vpp.

Does the upper 6SL7 grid have a DC ground reference?

Yes, 470K. Forgot that in the quick schematic

It's interesting, even to get to 50W output I have to "feather" the signal - back it off a little, then add it slowly. Otherwise the waveform shrinks.

[uneumann]

>> With the KT88 removed, I have 120Vpp from the PI. With the KT88 installed, 70Vpp.

Did you try removing the NFB connection? Does that changes the behavior with the KT88's installed?

[g1]

I'm wondering if the screen voltage is sufficient.

[pdf64]

What happens to the voltage at the screen grid and paraphase HT supply nodes, idle to max unclipped output, eg perhaps it’s sagging too much when the output valves are drawing current, hence the reduction in signal output?

[hylaphone]

Did you try removing the NFB connection? Does that changes the behavior with the KT88's installed?

No change.

I'm wondering if the screen voltage is sufficient.

Should be! According the the GEC datasheet, 560Va and 300Vg2 should make 100W in fixed bias..

What happens to the voltage at the screen grid and paraphase HT supply nodes, idle to max unclipped output, eg perhaps it’s sagging too much when the output valves are drawing current, hence the reduction in signal output?

Both nodes drop 20V at maximum clean output, certainly not enough to lower the signal?

[Leo_Gnardo]

With the KT88 removed, I have 120Vpp from the PI. With the KT88 installed, 70Vpp.

Such a reduction in drive voltage suggests KT88 control grids are drawing current, which happens when the drive voltage exceeds bias voltage, then the drive circuit is attempting to drive the grids into positive territory and the grid resistance drops precipitously. If I'm not mistaken, this is called Class B operation. Unless there's a very strong drive circuit, the signal voltage becomes loaded down and sags significantly. Besides suspected sinking of screen grid voltage, does the bias voltage also sag when the power approaches 50W?

[Helmholtz]

Both nodes drop 20V at maximum clean output, certainly not enough to lower the signal?

Well, 20V less PI supply may lower max clean PI output by 40Vpp or more.

For a quick test replace your 100k grid leaks by 220ks.

[Helmholtz]

Such a reduction in drive voltage suggests KT88 control grids are drawing current, which happens when the drive voltage exceeds bias voltage, then the drive circuit is attempting to drive the grids into positive territory and the grid resistance drops precipitously. If I'm not mistaken, this is called Class B operation. Unless there's a very strong drive circuit, the signal voltage becomes loaded down and sags significantly. Besides suspected sinking of screen grid voltage, does the bias voltage also sag when the power approaches 50W?

Grid conduction won't start before the pp grid signal exceeds twice the grid bias voltage.
It would be called class AB2 operation.

[hylaphone]

Such a reduction in drive voltage suggests KT88 control grids are drawing current, which happens when the drive voltage exceeds bias voltage, then the drive circuit is attempting to drive the grids into positive territory and the grid resistance drops precipitously. If I'm not mistaken, this is called Class B operation. Unless there's a very strong drive circuit, the signal voltage becomes loaded down and sags significantly. Besides suspected sinking of screen grid voltage, does the bias voltage also sag when the power approaches 50W?

At 50W, the bias supply "sags" from -42V to -41V. As mentioned, I did tack in Mosfet source followers to provide current drive to the grids. It didn't help so I removed them, but maybe I should revisit that. They were set up as shown here.

[hylaphone]

Well, 20V less PI supply may lower max clean PI output by 40Vpp or more.

For a quick test replace your 100k grid leaks by 220ks.

Just tried this, no difference.
I have also tried higher supply voltage to the PI, upwards of 450V and it didn't help.

[g1]

Have you tried altering your load resistance?
And that GEC datasheet seems overly optimistic to me. Plugging your numbers into Nick's calculator gives around 60W with your values and 70V p-p signal applied.
Not 70W, but more than 50 anyway. I had it set for -46V bias and didn't re-try with the lower bias number (-41V).

[Helmholtz]

Just tried this, no difference.
I have also tried higher supply voltage to the PI, upwards of 450V and it didn't help.

Doesn't seem to make sense.
Something is missing. Maybe oscillation involved?

Can you post scope pictures of grid and output signals? And don't forget the scope settings.

[Helmholtz]

Have you tried altering your load resistance?
And that GEC datasheet seems overly optimistic to me. Plugging your numbers into Nick's calculator gives around 60W with your values and 70V p-p signal applied.
Not 70W, but more than 50 anyway. I had it set for -46V bias and didn't re-try with the lower bias number (-41V).

When I feed the numbers into Nick's tool, I get 72W.
But it's essential that the grid Vpp is at least twice the grid bias.

In my experience those original tube datasheets were very reliable. Contemporary KT88 clones may be inferior, though.