Another PS and grounding issue

[sam9]

Self recomends a separate ground path for rail decoupling and another for input signal+NFBreturn. This seems like a reasonable precaution but raises a question in my mind. In the beta enhanced (or Darlington) VAS topology, the collector of the upstream transistor goes to ground. Which of the two ground paths is the better one? It seems to me that if it ties to the input+NFB ground you may be inviting oscillation but if you tie it to the decoupling ground you are inviting any noise that has been decoupled back into the VAS.

Using LT-Spice I placed a signal at this point and sure enough it appears at the output.

Is a third ground path called for here? Any thoughts.

[sam9]

Guess I'll just have to build one each way an see what happens.

[piertr]

I wouldn't put anything signal-related on decoupler ground. Because the bulk rail decouplers are just that (large), substantial ripple voltages can be imposed on them and they can inject power supply noise into the ground that they're connected to. Remember that just because it's 'ground' doesn't mean it doesn't 'move'; it has a finite impedance and large decouplers can push a lot of current through it, creating voltage drops. Any noise on signal or feedback ground will find its way to the output.

I would think that for a similar reason to that above (finite ground impedance), you would want to tie the VAS buffer (common collector) transistor's collector to the same ground that the signal and feedback grounds are connected to. Like I said above, the feedback and signal grounds need to be one and the same because any voltage dropped between them will appear at the output. The same effect might happen with the VAS buffer, but since the collector 'looks' like a current source the effect may be manifested in a different manner. I don't think a third ground is necessary, and as long as the VAS buffer collector is at the same potential as the feedback and signal ground you should be fine.

[carlmart]

Originally posted by sam9
Self recomends a separate ground path for rail decoupling and another for input signal+NFBreturn. This seems like a reasonable precaution but raises a question in my mind. In the beta enhanced (or Darlington) VAS topology, the collector of the upstream transistor goes to ground. Which of the two ground paths is the better one? It seems to me that if it ties to the input+NFB ground you may be inviting oscillation but if you tie it to the decoupling ground you are inviting any noise that has been decoupled back into the VAS.

Using LT-Spice I placed a signal at this point and sure enough it appears at the output.

Is a third ground path called for here? Any thoughts.

I'm not sure if Spice can or will separate between power ground and signal ground.

But it's a fact that you should only join them at star ground.

A friend of mine changed the ground arrangement on a Doug Self board, bought from him, and had tremendous listening improvements.

As far as I know Self, who was told on this improvements, did not correct the pcb. Maybe he says one thing but does another one.


Carlos

[sam9]

Since making the original post, I came to the uncertain conclusion reached above, namely to leave it at two ground networks and have one exclusivly for decoupling and the other for everything else. The responses above have convinced me to go ahead on this basis for future board layouts.

I noticed that between books Slone (who seens to be influanced by Self) switched grounding methods. Similarly, it may be that Self laid out some PCBs before comming to the conclusion that the grounds should be separated. Anyway, I do my own boards so it doesn't matter if he has updated his or not.

To-date I've been fortunate in that a single gound network have caused no audible problems for me. But no need to rely on luck forever.