Class-D audio amplifiers convert an audio signal into a high-frequency square wave whose switching times are modulated according to that audio signal. Thus, they operate using so-called pulse-width modulation. When the high-frequency components of the output square wave are removed by filtering, the remaining lower frequency components can reproduce the audio signal with very little distortion. This low distortion, coupled to the high efficiency of class-D amplifiers, is responsible for the increasing use of these devices in a wide variety of applications. However, to date there has been very little predictive mathematical modelling of those class-D amplifiers that employ negative feedback, as is usually the case in practice. Here, we derive and investigate a mathematical model for a novel design of the output stage for a class-D audio amplifier, which employs negative feedback to remove the effects of power supply noise on the amplifier's performance.