In short: A Class-D amplifier reproduces audio not by burning the difference in a linear transistor, but by switching the output rail fully on or off at hundreds of kilohertz with a PWM duty cycle that tracks the signal, then reconstructing the waveform through an LC low-pass filter — yielding 85–95% efficiency against the ~50–78% ceiling of linear Class-AB.
Why linear amplifiers waste power — and why that matters
Consider a classic Class-AB amplifier driving an 8 Ω speaker at 20 W from a ±25 V rail. The output transistor sits in its active region, dropping whatever voltage the load does not. At a mid-scale output the pass device might carry 1 A while dropping 12 V — dissipating 12 W as heat just to deliver 20 W of music. That is the fundamental problem: a linear device is a controlled resistor, and a resistor carrying current in a voltage drop makes heat. Peak theoretical efficiency for Class-B is 78.5%, and real Class-AB with its bias current lands closer to 50–65% at typical listening levels.
Class-D sidesteps this entirely. The output stage has only two states: the high-side switch is on (output = +rail) or the low-side switch is on (output = −rail). In either state the transistor ideally drops ~0 V while conducting, or carries ~0 A while blocking. Power = V×I → ~0 in both. The only losses left are conduction loss through the MOSFET's RDS(on), switching loss during the finite transition, and gate-drive charge. This is the same insight that makes a buck converter efficient — and indeed a Class-D amp is close kin to the switching regulator topology discussed here, only with the duty cycle modulated by an audio signal instead of a feedback loop chasing a fixed voltage.
Key fact: The "D" in Class-D does not stand for "digital." The output is a time-continuous, analog PWM waveform. The letter simply follows A, B, and C in the amplifier-class alphabet.
Pulse-width modulation: encoding amplitude as time
The heart of a Class-D amp is a comparator that compares the audio input against a high-frequency triangle (or sawtooth) carrier. When the audio exceeds the triangle, the output goes high; otherwise low. The result is a square wave whose duty cycle is proportional to the instantaneous audio amplitude — 50% at silence, swinging toward 90% on loud positive peaks and 10% on negative peaks. The average value of that pulse train, recovered by filtering, is the original audio.
A typical carrier (switching) frequency sits at 300 kHz to 1 MHz. The choice is a trade-off: higher fsw pushes the switching residue farther above the 20 kHz audio band (easing the filter) but raises switching losses and EMI; lower fsw is efficient but demands a sharper, bulkier filter.
The output stage and the output filter
Most audio Class-D amps use a full H-bridge (BTL, bridge-tied load) across the speaker, so the load sees twice the rail swing and the common-mode switching cancels. The critical detail is dead time: both switches must never conduct at once (shoot-through would short the rail), so the driver inserts a few nanoseconds of gap. That gap is a primary source of distortion, because during dead time the inductor current freewheels through a body diode and the exact output voltage becomes load-dependent.
The LC filter is typically 2nd-order Butterworth. For an 8 Ω load you might pick L = 22 µH and C = 0.68 µF, giving a corner near f = 1/(2π√(LC)) ≈ 41 kHz — comfortably above 20 kHz audio yet rolling off the switching residue hard. Undersize the inductor and ripple current soars; use a lossy ferrite and you add distortion and heat.
| Class | Topology | Peak efficiency | Typical real-world | Use case |
|---|---|---|---|---|
| A | Always-on linear | 25–50% | ~20% | Hi-fi, small signal |
| AB | Push-pull linear | 78.5% | 50–65% | General audio |
| D | PWM switching | ~100% | 85–95% | Mobile, car, subwoofers, TWS |
Key fact: At 10 W output a Class-AB amp may waste 8–10 W as heat, needing a large heatsink; a Class-D amp wastes under 1 W, which is why your phone and earbuds can run amps at all.
EMI: the price of fast edges
Switching 25 V in a few nanoseconds creates dV/dt and dI/dt that radiate. An unfiltered or poorly laid-out Class-D amp will fail CISPR 25 / FCC Part 15 radiated-emissions limits, with harmonics of fsw smearing across the AM and FM bands. Mitigations engineers rely on:
- Spread-spectrum modulation — dithering fsw by ±10% smears the energy over a band instead of a sharp spectral line, cutting peak emissions by 10–15 dB.
- Edge-rate (slew) control on the gate drive to soften dV/dt without wrecking efficiency.
- Tight LC-filter placement and short, wide power loops to shrink the radiating antenna.
- Filterless modulation schemes (ternary / BD modulation) that let short speaker leads act as the filter in earbud applications.
For board and chip teams, these are not afterthoughts — they are simulated up front. Spectre and Cadence's EMIR flows, or Siemens EMI analysis, let designers predict radiated spectra from the switching-node layout before tape-out, and many teams build behavioral models in Synopsys Saber to co-simulate the modulator, bridge, and filter.
Closed-loop feedback and distortion
Open-loop Class-D is cheap but its THD is dominated by rail noise (PSRR is poor — rail ripple modulates the output directly) and dead-time error. High-fidelity designs close a feedback loop from the switching node (post-bridge, pre-filter) or from the speaker terminal, giving PSRR > 70 dB and THD+N under 0.01%. Self-oscillating topologies such as UcD and Hypex's NCore push the loop gain high across the audio band so the amp corrects its own errors in real time — the reason modern Class-D now rivals the best linear amps in measured transparency.
Want to design one yourself — from the modulator math to the gate driver and EMI-compliant layout? CourseTron's analog and mixed-signal track walks you through it with real tool flows. Browse the CourseTron course catalog →
Where Class-D wins — and where it still struggles
Class-D dominates anywhere battery life, heat, or density matters: TWS earbuds, phones, laptops, soundbars, car audio, and especially subwoofers (where the narrow bandwidth makes filtering trivial and the power is high). It struggles at the extreme high-frequency edge where filter group-delay and residual switching noise matter to golden-eared listeners, and in ultra-low-power idle where quiescent switching loss can exceed a tiny Class-AB stage. For a broader look at how switching vs. linear trade off in the power domain, see our companion piece on power-supply decoupling and the switching node. If you need custom analog/mixed-signal IP designed and verified, Avecas offers VLSI design services covering exactly these flows.
FAQ
Is Class-D a digital amplifier?
No. The output is an analog PWM waveform whose timing is continuous. There is no sampling, quantization, or binary word at the output — only a two-level signal whose duty cycle is a continuous analog quantity. True digital-input amps add a PCM-to-PWM modulator, but the power stage is still analog.
Why does my Class-D amp need an output filter but my earbuds do not?
Speaker-level amps use long cables that would radiate the switching frequency as EMI, so an LC filter removes it. Earbuds use very short leads and filterless BD/ternary modulation, where the tiny voice-coil inductance itself averages the pulses and the short wire radiates little.
What sets the efficiency ceiling in a real Class-D amp?
Conduction loss (I²×RDS(on)), switching loss (∝ fsw × V × Q and edge time), gate-drive charge loss, and dead-time body-diode conduction. Lowering RDS(on) and fsw helps efficiency but trades against distortion and filter size.
Why is PSRR a bigger deal in Class-D than Class-AB?
In open-loop Class-D the output swings between the rails, so any ripple on the rail passes almost directly to the speaker. Closed-loop designs that sense the switching node reject this; it is the single biggest reason cheap open-loop amps sound noisy.