It's basically two Norton lossless feedback amplifiers connected in a push pull arrangement.The winding in the emitter circuit provides negative feedback and the ratio of the windings sets the amount of feedback and therefore the overall gain of the amplifier stage.I'm not at home at the moment so I can't access my list of web links, but try a Google search for Chris Trask's papers on lossless feedback for more information.When you reduce the supply rail it may reduce the noise floor, and also possibly reduce the gain, which also may give the impression of reducing the noise floor. But below about 3v the bias point of the transistors is way off and the output power capability (and therefore IMD performance) is seriously reduced. What you are observing may actually be a secondary artifact resulting from something else changing which is in turn 'fixing' another problem such as FM BC interference, amplifier oscillation or some other out of band parameter which is impacting the overall performance.One final thing - it may be that George's 'reverse engineering' is not 100% accurate, especially regarding the construction and winding of the feedback transformer. I suspect that the two amplifiers maybe should be wound on two separate halves of the binocular core but that the input winding is common to both. This is how Trask has implemented some of his augmented push pull amplifiers, but I can't confirm this is how the Wellgood design works.If you have an opportunity I'd suggest that you also try the LZ1AQ amplifier design http://www.lz1aq.signacor.com/docs/wsml/wideband-active-sm-loop-antenna.htm as a comparison. It's actually easier to build if you leave out the regulator and just feed the output into coax rather than using CAT 5 / 6 cable. I found that the LZ1AQ performed at least as well, if not better, than the Wellgood clone. But as I stated earlier this may be actually be due to an error in the reverse engineered circuit.The Yahoo Groups Loops forum is other very useful source of information and is frequented by Trask and Wellbrook. There has been some discussion regarding the Wellgood copy and Wellbrook don't seem to be that bothered that the design is now out in the open. This could be because they have now moved to a new design, or it could be that George's Wellgood circuit isn't correct and they don't want this to be known.