Hmm... there really isn't a simple way to explain this, but I will take my best shot.
Control systems of all types use negative feedback. This ranges from a thermostat to an amplifier. Here's the basic idea:
- The base control system is designed to generate a particular output function given a particular input function. The relationship between the two is called the control system's transfer function. The transfer function can be very complicated, or very simple, ranging from non-linear non-continuous transfer functions to simple linear relationships.
- Power amplifiers are designed to have a linear transfer function. That is, the output is designed to always be proportional to the input in a linear fashion.
- All control systems have some error, caused by part tolerance or design, that causes the output function and input functions to be related in a way different from transfer function to some degree. In other words, the effective transfer function in practice isn't identical to the theoretical transfer function.
- It is a often a simple task to measure the actual output, calculate the intended output, and derive an "error signal" that is the difference of the two. This error signal is the basis for feedback in amplifiers. "Negative" feedback refers to the typical use of error signals being subtracted from the input function to correct it (I believe... been a while since I had a control systems course...).
- In control system theory, an error signal added to the input function can theoretically correct the errors in the effective transfer function, such that the output and input functions are then related by the theoretical transfer function as is intended.
- The above works quite well for steady state systems. When the system becomes transient, bad things can happen. Just why is difficult to explain, but the true answer comes from the solution of the control system differential equation. What is found is that the magnitude of negative feedback and rate of change of the input function are inversely related such that they both cannot increase without limit, less the sytem become unstable.
- In layman's term, it is possible for an error signal to "overcorrect" the input signal. This happens when the rate of change in the input signal is to rapid for the error signal to track. The result is oscillation in the feedback system.
So, the greater the magnitude of the feedback signal, the closer the output can track the theoretical function (lower distortion). However, during transients the input may change rapidly enough to cause brief periods of an inability to properly correct the transfer function. The result is transient distortion. Were the rate of input change to continue at that level, the system would become unstable.
Since music is transient in nature, high levels of feedback can lead to higher levels of distortion than were present in the original signal, but only during brief transients. Sine wave measurements will not pick up this distortion. However, since humans also have a limited frequency range of hearing, it is possible to design proper signal filtering and feedback levels such that at all times the distortion levels remain either (1) below the level present in the original signal or (2) well beyond the human hearing range and taken care of by proper filtering.
At least, that's my understanding of it. I'm not an electrical engineer by training, but the differential equations describing fluid, mechanical, and electrical systems are essentially the same. All control systems exhibit this same basic behavior.