Now, 20k is a tough load for most miniature tubes to drive. Consequently, very buffed triodes like the 5687 or 6BX7 or parallel triodes or maybe the Cathode Follower should be used to buffer the phase splitter from this load.
   
The Feedback Loop
   We know we have to use feedback, but how should it be configured? My preference is to never extend the loop beyond more than two active devices. Having built the same amplifier with one global feedback loop versus two sub-circuits each with its own feedback loop, I definitely found the latter more listenable. The cause might be found in the assumption that the two sub-circuits better handled the clipping of the output stage.

   (A good friend of mine tells me that he prefers the sound of an amplifier with an undersized output transformer, say a pair of 6550s driving a Dynaco ST-35 output transformer. He argues that output transformer clipping is less objectionable to the ear than the tube clipping.)
   My current thinking is that mix of non-feedback sub-circuits with feedback sub-circuits might be best, for example, a hybrid power amplifier that consists of a tube gain stage with a moderate amount of feedback cascading into a no-feedback, Class A, unity gain, push-pull MOSFET output stage.

  By dividing the amplifier into two and using two feedback loops, only the last half of the amplifier clipped. The first half could withstand a much greater input signal than the amplifier as a whole could, so even though the output stage had begun to clip the output signal, the input stage was putting out a clean signal.

  One experiment worth trying is to wrap the feedback loop only over the phase splitter and the output stage as shown above. This circuit could then be fed from an input that used or did not use feedback. Notice that the input circuit would have to swing at least all the voltage that the amplifier put out (say, 40 volts) plus the voltage (say, 5 volts) needed to drive the phase splitter (total, 45 volts).  Still, what vacuum tube do so well is swinging big voltages linearly. 

  My guess is that once the input signal relaxed enough to allow the output stage to break out of clipping mode, the output stage could more readily snap back in line with a signal that was not also greatly distorted. Perhaps it was because each active device must break out clipping, assuming a very high feedback ratio, before the next device can do as well and all  the delays add up to a slower recovery.

pg. 19

<PREVIOUS

NEXT >