Instability in a new amp.

[DeathRex]

I'm making a PPP 6GB5 for about 5 months now. I got it working except for some instability, when I put in more than about 14db of global feedback, I'm getting a oscillation, ultrasonic waveform. Less feedback and it works fine. The tubes in it now are: 6N2P preamp, 6N1P phase splitter, and 6GB5.
I tried a 6CG7 in place of the 6N1P and I could get about 17db of feedback before the oscillation began.

I haven't drawn up a schematic on the computer yet, but this one is close, except, the outputs are PPP 6GB5 into a 100 watt 2200ohm output transformer with the 16ohm leg hooked into a 8ohm load. R67 and 69 are 187Kohm, might need to lower them. B+ is about 410, 6N1P B+ is about 330. It's the same on both channels. I have several tubes to match into pairs, but I haven't yet.

[Geek]

Put in some zero-pole compensation and limit the bandwidth. You're getting into transformer weirdness at ultrasonic.

Cheers!

[WA4SWJ]

It also might not hurt to put in some low value resistors - 100 ohms or so - in series with the screen grids of the output tubes right at the sockets.

Best of luck!

Regards,

[DeathRex]

It also might not hurt to put in some low value resistors - 100 ohms or so - in series with the screen grids of the output tubes right at the sockets.

Best of luck!

Regards,

I do have those, forgot to draw them in, 200ohm 2 watt. I don't have control grid resistors for the 6N1Ps, would I need those?

[DeathRex]

Put in some zero-pole compensation and limit the bandwidth. You're getting into transformer weirdness at ultrasonic.

Cheers!

Whaaaaat? I don't understand.
C35 is also not there. If I put in a 1nf cap across the feedback resistor, the oscillation starts.

The amp, with no feedback, sounds much better than the GU50 with no feedback. 10db might just do me, but I don't want it to startup later.

[Geek]

ZPC isn't in the FB loop. it's from the 12AX7 anode to ground in series with a resistor. The ST-70's used them, but those values will not work for the 12AX7. Values need to be determined with a function generator and oscilloscope. Basically you want to make an R/C circuit in there for the cleanest tops/bottoms of a square wave at 5KHz.

Cheers!

[DeathRex]

Well I need a function generator. I can only make sine waves.

[mesherm]

Well I need a function generator. I can only make sine waves.

I use a Tenma 72-505. Its small and battery powered. Does Sine and Square waves.
It's not continuous sweep but in steps but that works fine for basic testing of amps. About $50.

[DeathRex]

OK I got a squarewave generator, and have a beautiful squarewave up to the 6N2Ps out where I have a tiny bit of ringing. With no feedback I get a bit more ringing on the output. With a 100pf cap and a pot I adjusted the ringing down to a nub, but with feedback I get more ringing. I can adjust it so I can get about 20db of feedback, but it looks like furry zig zags, with the pot at about 44Kohms.

No feedback and no ZPC


No feedback ZPC added


14db feedback with ZPC


20+db feedback with ZPC

[Gingertube]

Your ZPC (Zero Pole Compensation) is too agressive and is not optimal in design.

Ian's method for stuff like this:

First some basics.

When adding a ZPC or Zobel or Stability Network (depending on what you want to call it) you need to know some very basic things.

In a normal RC circuit you get half the signal and a 45 degree phase shift when the impedance of the capacitor is equal to the impedance(resistance) of the resistor.

You are wanting to adjust the amplitude response at high frequencies while leaving the phase response largely intact since it is the phase shift which is causing the instabilty.

To do this you choose a ZPC resistor of 1/10th the existing gain stage load resistance.
In your case the existing anode load resistor is 200K. So you want to use a 20K (18K or 22K will do).
WHY:
Now the gain will be down to half when the impedance of the RC network = 200K, the same as the existing anode load resistor ( that is, when the impedance of the series cap = 200K - 20K = 180K)
BUT the 45 degree phase shift will be when the impedance of the cap is equal to the 20K it is in series with, this is obviously a much higher frequency (9 times higher).

You can see that for the compensation to do the right thing then that series resistor value cannot be more than say 1/3 of the load resistor (poor choice but may work), 1/5th would be better but I like to go to 1/10th for best results. If the resistor is higher than this then you defeat the whole purpose of trying to adjusting amplitude response (at higher frequencies) while leaving phase response alone.

Having chosen that resistor then the you can choose the series cap next.
Look at you square wave photo. If you know the oscilloscope time base you can work out the period of 1 full cycle of the ringing on the top of the square wave, and from 1/period the frequency of the ringing.

The minimum value cap is then chosen such that at the ring frequency the cap has an impedance equal to that 200K - 20K = 180K. (Xc = 1/[2 pi f c]). That will give you half amplitude response at the ring frequency BUT very little phase shift. You will probably need to go higher in value for the cap (for more attenuation at the ring frequency) but this defines the lower usable limit of the cap value.

Adjust the cap value up till you have just a small overshoot on the square wave leading edge with the ring lasting no more than 1 and 1/2 full cycles.
Sorry missed a step - do all this without C35 fitted.

Once all that is done you can do a final tweak at your desired feedback level by trying various values for C35 till you get a nice squarewave.

Hope this make sense to you.

Before doing any of this you need to make sure that none of your tubes are suffering from parasitic oscillation. All tubes should have grid stops. Add gridstops on both sides of the phase splitter ditto for the input stage.
There is a "magic" formula for calculating grid stop values from the tubes gm value - just can't remember what it is right now. I just recall that when I used the formula on a 12AX7 (gm = 0.0016 Amps/Volt or 1.6mA/V) I arrived at a value of 4K7.
Try 4K7 gridstops on the input stage and on both sides of the differential amp phase splitter.

Cheers,
Ian

[DeathRex]

I haven't installed the gridstoppers on the 6N1Ps yet, but here's the new picture:
This is with a 47pf cap, I tried putting in a 15pf and 27pf in parallel, it only changed it a tiny bit, with a 30Kohm pot in series. I also have a 680pf in parallel with a 3.8K resistor to give 10db of feedback. I changed the ground return feedback resistor to 24ohm. I also found I have a lot of hum coming from an unshielded Triad transformer into the inputs. I'll have to shield them.

[DeathRex]

Some other problems I had:
Had both channels of the oscope hooked up, but had the grounds reversed, had signal all over the place.
6N1P has a much lower bias than a 6CG7, only had 2 volt bias on the phase splitter with 100 volts going in.
Had a pair of testing 6DW4Bs where 6CJ3/6DW4Bs were supposed to go, they were dropping twice as much voltage as were supposed to, because they aren't identical.

[DeathRex]

I found that I have about 2/3 power at higher frequencies 15000+Hz. I can turn it up and the sinewave increases, then stops at about 17 volts RMS and any more increase just distorts. I wonder if it's because the output transformer is reflecting back 1100 ohms to the parallel 6GB5s.