Abstract. Class-D audio amplifiers are particularly efficient, and this efficiency has led to their ubiquity in a wide range of modern electronic appliances. Their output takes the form of a high-frequency square wave whose duty cycle (ratio of on-time to off-time) is modulated at low frequency according to the audio signal. A mathematical model is developed here for a second-order class-D amplifier design (i.e., containing one second-order integrator) with negative feedback. We derive exact expressions for the dominant distortion terms, corresponding to a general audio input signal, and confirm these predictions with simulations. We also show how the observed phenomenon of "pulse skipping" arises from an instability of the analytical solution upon which the distortion calculations are based, and we provide predictions of the circumstances under which pulse skipping will take place, based on a stability analysis. These predictions are confirmed by simulations.Key words. class-D amplifier, mathematical model, pulse skipping, total harmonic distortion AMS subject classifications. 34E10, 37N20 DOI. 10.1137/1007883671. Introduction. Class-D amplifiers are used widely in modern electronic devices, such as laptops, mobile phones, and hearing aids, because they are particularly efficient. Their output takes the form of a high-frequency square wave whose switching times are modulated in such a way that the low-frequency components of the output correspond to the intended audio signal. The high-frequency switching components of the output are filtered out.For an open-loop class-D amplifier (i.e., a class-D amplifier without feedback), it is readily shown that no distortion is introduced into the audio range of the output signal by the pulse-width modulation (PWM) process [4,6,12,14]. However, due to nonidealities in practical designs (e.g., output-stage noise, component variation, or distortion in the carrier signal that drives the switching), it is desirable in practice to add a negative feedback loop to the design, but this inevitably introduces some distortion into the audio components of the output (see, for example, [15]). In this paper, we develop a mathematical model for a second-order class-D amplifier, in which a loop filter is configured as a second-order integrator. (For the purpose of our analysis, this second-order integrator is modeled as two first-order integrators connected in series.) We have previously derived similar models for first-order class-D amplifier designs [6], but the presentation here is more systematic and is extended to allow consideration of the stability of the analytical solutions that we develop; the question of their stability proves to be essential in explaining the observed behavior of the amplifier. A further motivation for the design considered here is that the secondorder integrator is now becoming much more prevalent than its first-order counterpart in state-of-art audio amplifier design, as it provides higher loop gain, which allows it to better overcome the nonlinearities inhe...