KT88 amp

[hylaphone]

Wait, this should work better:

Set both scope channels to DC coupling and connect probes to the KT88 grids with no or very low PI signal (same 2V/div setting as before).

Adjust zero lines to center of screen, increase PI signal until clipping starts and post scope pics.

Ok here we are, I did double check the PI signals were perfectly balanced before installing the KT88s

[hylaphone]

Replace 100k grid resistors with 220k, 22nF coupling in pi stage to second pi grid is way to small, make it as large as practical at least 100 nF . Stuck 2x 470k resistors on the PI plates and add another series resistor till perfect balance is done .Hope it help. Cheers

Thank you for the tips. I tried these changes and the imbalance is still there, relative to frequency.

Can you scope the signals over the PI plates without power tubes ? Apply 1Vpk at the PI input and report what you get over the PI plates please. I think you get too much voltage into non inverted side. Make the feedback built resistor as well to reduce the second plate voltage as large as the first stage or less. You can make individual cathodes, it doesn't matter as time it is bypassed. It is a simple gain stage circuit, should work as simple as it is.There are lots of Acrosound and Heath amps who use it.
This is identical circuit as yours but built with a different tube. Is not perfect but good enough for a guitar amp

Probes on the PI plates, correct? Sorry, I don't understand "Make the feedback built resistor as well to reduce the second plate voltage as large as the first stage or less." can you clarify?

[Helmholtz]

Ok here we are, I did double check the PI signals were perfectly balanced before installing the KT88s

One more test, as some of your previous posts were a bit contradictory:

Assuming the signals in the pic were taken directly at the KT88 grids using DC coupling, using the same settings: Are you getting zero lines centered to the screen when you reduce the grid signals to zero?
I need to make sure there is no DC offset involved.

[uneumann]

.... the PI signals were perfectly balanced before installing the KT88s

The balance issue is likely due to the feedback (through the 0.022u cap) used to stabilize the lower triode gain in the PI. Without the output tubes, that feedback keeps the PI balanced. Once you add the output tubes and they clip, the grid clipping on the lower tube is not isolated from the PI. It feeds back through the 1k grid stopper and the .22u coupling cap to the plate of the PI. Once that plate sees clipping, it impacts the feedback that is trying to stabilize the PI and it leads to asymmetric clipping. You can fix this (temporarily or permanently) by making those 1k grid stoppers bigger. Try 100k. I'm pretty sure the PI will remain more balanced since it will be less impacted by the clipping at the power tube grids.

That may not solve all your problems, but it can't hurt, and it could address the freq dependent behavior as well.

[hylaphone]

One more test, as some of your previous posts were a bit contradictory:

Assuming the signals in the pic were taken directly at the KT88 grids using DC coupling, using the same settings: Are you getting zero lines centered to the screen when you reduce the grid signals to zero?
I need to make sure there is no DC offset involved.

Sorry about that, I had re-centered the traces as the waveforms clipped. Here is another picture just as you asked.

[hylaphone]

The balance issue is likely due to the feedback (through the 0.022u cap) used to stabilize the lower triode gain in the PI. Without the output tubes, that feedback keeps the PI balanced. Once you add the output tubes and they clip, the grid clipping on the lower tube is not isolated from the PI. It feeds back through the 1k grid stopper and the .22u coupling cap to the plate of the PI. Once that plate sees clipping, it impacts the feedback that is trying to stabilize the PI and it leads to asymmetric clipping. You can fix this (temporarily or permanently) by making those 1k grid stoppers bigger. Try 100k. I'm pretty sure the PI will remain more balanced since it will be less impacted by the clipping at the power tube grids.

That may not solve all your problems, but it can't hurt, and it could address the freq dependent behavior as well.

Just tried this with 100K stoppers, stays balanced until roughly 55W output, but as it clips the inverter side can't keep up. Here is a picture.

Also, issue of imbalance at different frequencies is there with the KT88 removed... not sure if I was clear about that

[Jazz P Bass]

I would ditch the 6SL7 in favor of a 12AT7.

[uneumann]

Just tried this with 100K stoppers, stays balanced until roughly 55W output, but as it clips the inverter side can't keep up. Here is a picture.

OK - it does look more balanced than the pic in post 50. If you look at the PI outputs, rather than the kt88 grids, you'll likely see it's more balanced looking too.
The effect of the clipping is reduced as the grid stoppers get bigger. It's a "feature" of the self-balancing paraphase PI. If the PI output gets clipped, it impacts the balance. That's just the way the circuit works. To completely eliminate this problem you need to add a buffer (follower) between the PI and output tubes. The suggestion to use low-z output tubes (12at7) is also a good one for the same reason. It will be less impacted for the same load (grid stopper values).

[Helmholtz]

Sorry about that, I had re-centered the traces as the waveforms clipped. Here is another picture just as you asked.

Ok, this makes more sense.
Both PI signals get clipped by grid conduction just below 40V, the value should correspond to the bias voltage at full power.
So that's normal. It seems the PI provides enough signal for the KT88s.
You may still try to lower bias voltage (i.e. making it less negative) and see if this helps.

The different PI output waveshapes (and changed duty cycle) are quite normal for a clipping paraphase, which is far from being a perfect PI.

Still wondering about the rounded positive tops of one of the grid signals - and that PI balance would depend on frequency.
Does increasing the values of the 22µ and 22nF caps in the PI make a difference?

The reason for not getting the theoretical ~ 70W clean doesn't seem to be the PI.
Other possible reasons for lowered output might be:
- OT losses of maybe 5W
- Higher "saturation voltage" of your KT88s compared to original GEC types.

[hylaphone]

Ok, this makes more sense.
Both PI signals get clipped by grid conduction just below 40V, the value should correspond to the bias voltage at full power.
So that's normal. It seems the PI provides enough signal for the KT88s.
You may still try to lower bias voltage (i.e. making it less negative) and see if this helps.

The different PI output waveshapes (and changed duty cycle) are quite normal for a clipping paraphase, which is far from being a perfect PI.

Still wondering about the rounded positive tops of one of the grid signals - and that PI balance would depend on frequency.
Does increasing the values of the 22µ and 22nF caps in the PI make a difference?

The reason for not getting the theoretical ~ 70W clean doesn't seem to be the PI.
Other possible reasons for lowered output might be:
- OT losses of maybe 5W
- Higher "saturation voltage" of your KT88s compared to original GEC types.

Ok, well that explains why the MOSFET followers didn't really help

I raised the 22nf cap to 220nf, and paralleled the 22uF cathode cap with a 100uF. No change in balance.

I guess I'm running out of options to reach the target power... I'm surprised at such a big discrepancy, as I did do a lot of head scratching prior to building this and all the math all pointed to 70W+.

Any other wild ideas?

[Helmholtz]

For what it's worth, the sound level difference between 70W and 60W is less than 0.7dB.

[catalin gramada]

Any other wild ideas?

Yes. Forget you 360V supply. Use 12at7, supplied at 400V or more, stuck 47k resistors on the plates and 470 ohm shared cathode (or separate bypassed 1k better) Imagine a 10k or some series/100 ohm shunt nfb network, supply the screen grids at 500V but don't forget some hefty 1k stoppers for each tube, and be happy with that...

[hylaphone]

For what it's worth, the sound level difference between 70W and 60W is less than 0.7dB.

Yes of course. I still haven't given up on the PI balance though. It gets better by bypassing the 470K side of the divider with 100pF or so, which also corrects a small phase shift I was seeing. Though once the power tubes are back in, it doesn't seem to matter much.
I also tried "balancing" the output impedances with a series resistor on the inverter side, but didn't see much of a change.

I reworked the PI for 6SN7, center biased with 100K plate loads and 500V supply. Still enough output swing to drive the KT88s, which are now making 70W clean w/o NFB connected. Added benefit is I can lower the grid resistors, allowing for 6550 - which may buy me a few more watts, we'll see.

This thread has been really enlightening and I want to thank you guys for your help!

[catalin gramada]

Yes of course. I still haven't given up on the PI balance though. It gets better by bypassing the 470K side of the divider with 100pF or so, which also corrects a small phase shift I was seeing. Though once the power tubes are back in, it doesn't seem to matter much.
I also tried "balancing" the output impedances with a series resistor on the inverter side, but didn't see much of a change.

I reworked the PI for 6SN7, center biased with 100K plate loads and 500V supply. Still enough output swing to drive the KT88s, which are now making 70W clean w/o NFB connected. Added benefit is I can lower the grid resistors, allowing for 6550 - which may buy me a few more watts, we'll see.

This thread has been really enlightening and I want to thank you guys for your help!

6sn7 have far little gain to make any relevant global nfb usable in you circuit. You may apply it in previous stage with all consequences regard the phasing issue. If you still open to an experiment use 6sl7 as paraphase as you did but insert a 6sn7 driver stage in between

[Helmholtz]

I reworked the PI for 6SN7, center biased with 100K plate loads and 500V supply. Still enough output swing to drive the KT88s, which are now making 70W clean w/o NFB connected.

Not quite sure what exactly you did, but it seems you're driving the KT88s into class AB2 (some grid current).
I wouldn't want to exceed the plate voltage limit (450V) of the 6SN7, though.