Linear audio amps (Class AB) waste 50–70% of their power as heat. Class-D amplifiers sidestep this by treating the output stage as a switch: the MOSFETs are either fully on or fully off, dissipating almost nothing in between. Efficiencies of 90%+ are routine, which is why every Bluetooth speaker, soundbar, and modern car stereo uses them.

The core idea: compare your audio input against a high-frequency triangle wave (typically 250 kHz to 2 MHz) using a comparator. The output is a PWM signal whose duty cycle tracks the instantaneous audio amplitude. Drive a half-bridge (two MOSFETs) with this PWM, then pass it through an LC reconstruction filter to recover the audio and reject the switching frequency.

Half-bridge vs full-bridge (BTL):

• Half-bridge needs a split supply or a DC blocking cap. Simpler, but the cap must be large for good bass response.
• Bridge-tied load (BTL) uses two half-bridges driving opposite ends of the speaker. Doubles output voltage swing, quadruples power, and cancels common-mode noise. Standard for anything above a few watts.

The output filter is critical — it's a second-order LC low-pass tuned about a decade below the switching frequency. For a 4Ω speaker with 400 kHz switching, target a cutoff around 40 kHz:

L = R/(2πf_c√2) ≈ 4/(2π·40k·1.41) ≈ 11 μH

C = 1/(2πf_cR√2) ≈ 1/(2π·40k·4·1.41) ≈ 700 nF

Use a shielded ferrite inductor rated well above peak speaker current, with low DCR to avoid losses. The cap should be film (polypropylene), not ceramic — ceramic's voltage-dependent capacitance causes distortion.

Real-world example: the TI TPA3116D2 is a ubiquitous 50W stereo Class-D chip. It runs from a single 24V supply, uses BTL output, and integrates the modulator, gate drivers, and MOSFETs. With proper LC filtering and ground layout, you'll get THD+N below 0.1% — better than many Class-AB designs.

Gotchas:

• EMI is brutal. Those fast switching edges radiate. Keep the loop from MOSFETs through filter inductors back to the bypass cap as small as possible. Many designs add a small RC snubber across each MOSFET.
• Dead time matters. Both high-side and low-side MOSFETs off briefly during transitions to prevent shoot-through. Too much dead time causes crossover distortion; too little destroys MOSFETs.
• Idle losses from gate drive and switching are not zero — efficiency drops at low volumes. Modern chips use "modulation schemes" (BD, 1SPW) to mitigate this.