Evaluating Perceptual Coders

I just added a new page to my website, called Evaluating Perceptual Coders. It's about a unique way to analyze the quality of MP3s, AAC, WMA, FLAC, etc.

Analyzing Class AB Amplifer Bias – Idle Current

I have been trying to get my head around class AB bias and idle current. Looking at various authoritative sources, there seems to be some conflicting schools of thought.

The first is that class AB output devices “turn off” – or need to. While it is true that the top or bottom output devices have to do the heavy lifting for each half of the cycle, I cannot see a mechanism that decisively turns off the non-pulling half. Using Elliott Sound Products Project 3a as an example, the bias is a constant-current network, so it should maintain the same voltage difference between the Q5 and Q6 bases at all times. It is true that under load, the pulling half may “stretch” enough to make the bias insufficient to keep the non-pulling half turned on, but this is an incidental effect. If the amplifier were driving headphones, or maybe just an oscilloscope, the quiescent bias current would be present at all times.

Moreover, this is not a problem, according to W. Marshall Leach, Jr., because gm-doubling is a “fallacy”. The fallacy arises because engineers analyze the top and bottom halves of the AB output stage with separate load resistors. This gives rise to two load lines, which we must splice together somehow, and the drawn-in section has a different slope (gm). However, if we analyze the circuit as a unit – which it is – then there is a single, straight load line, which is asymptotic with the normal transfer functions of the top and bottom output devices. Of course, to avoid a dead zone, we must bias the output devices so that there is some overlap in their individual load lines.

The problem then becomes how to figure out how much overlap is enough. Too little, and the dead zone becomes a source of distortion. Too much, and we start wasting power and overheating the output devices. That seems difficult enough, without adding gm doubling, assuming it is indeed fallacious. Does anyone have any thoughts about this?