An interesting question for designers of audio power amplifiers is how far distortion can be reduced, and where the ultimate limit would be found. To try to reach some conclusion about this I started from my best design so far and looked for ways to improve it. What follows is a 'thought experiment' rather than a serious plan for a new design.
Taking my intermodulation figure of -109dB as a starting point: The remaining sources of distortion are fairly easy to track down, the greatest being the mosfet output stage nonlinearity. To reduce the effect of this we can do several things. One is to increase the quiescent current so that there is less variation in gm over the output range. Another is to use as many mosfets as possible in parallel so that the output impedance is smaller and a given percentage variation will generate less distortion. Six mosfets in parallel should reduce this source of distortion by a factor of 6. Combined with increased Iq a reduction by a factor of 10 could be achievable, giving distortion at -129dB. Making these changes to my basic design would involve some redesign to increase drive current available for the mosfets. A further 'improvement' can possibly be achieved just by specifying the distortion at a higher power level, rather than my usual method of testing at 100mV or 300mV input, chosen to reveal crossover effects. Most of the crossover distortion effect is at low power levels, while at higher levels this makes a smaller relative contribution. Unlike most bjt output stages distortion does not rise dramatically at high output current levels with mosfets.
One approach I have considered but rejected is known as 'error feedback' which I described in another article in this series. As pointed out there conventional error feedback is just high negative feedback in disguise. A positive feedback loop with close to unity gain in effect gives very high gain, (theoretically infinite at exactly unity loop gain), round a local negative feedback loop. I have seen reports of serious stability difficulties encountered by some designers trying to use this method, which comes as no surprise. Negative feedback loops need to have accurately controlled and predictable gain and phase characteristics over a frequency range from zero to in some cases well beyond 10MHz to guarantee unconditional stability, and 'infinite gain' positive feedback loops I suggest are not helpful in this respect.
The next step in my amplifier 'redesign' could be to increase overall feedback. The feedback at 20kHz could easily be increased by another 12dB or more by using a faster rate of attenuation between 200kHz and 2MHz. The present 6dB per octave is far slower than necessary, and adding a series of steps it can be increased, perhaps up to the 10dB per octave average rate of attenuation suggested by Bode in his classic 1945 book on feedback amplifier design. A -140dB distortion figure seems quite attainable. Is this the limit? Not yet!
There is still one very powerful distortion reduction method not mentioned, and this is feedforward error cancellation. The simple distortion extraction method possible with the mosfet amplifier also makes distortion cancellation relatively easy and effective. Having extracted a signal including the output distortion D reduced by a factor of 20 all we need to do is amplify it by a factor of 20 and subtract the result from the output signal. This is not really so easy, but it is certainly possible in principle to trim this sort of arrangement to give well over 60dB cancellation. So does this mean that distortion figures better than -200dB are possible?
Probably not. We are now well into the region where even the best resistors and capacitors add their own distortion. In the case of metal film resistors the distortion is approximately proportional to signal voltage squared, so just using several in series would be expected to bring down the nonlinearity. There are high priced 'audiophile' components available, but it seems unlikely that their distortion has been measured down to -200dB, and the claims made are more likely based on what someone believes they can hear, or some theoretical improvement of uncertain validity. The ability to measure distortion at this sort of level will also of course be limited by passive component nonlinearity.
Of more practical importance is the limit beyond which there is no genuinely audible gain in sound quality to make it worthwhile, and some of us believe this was passed long ago, and that most published descriptions of amplifier sound are pure fantasy. The Quad tests over 30 years ago already showed that distortion extracted while using music and driving speakers could be below audibility when reproduced alone, and similar conclusions from my simple 6 transistor mosfet amplifier merely show how easy this is to achieve.
HOME.
