How Do You Design With a Class-D Audio Amplifier IC?

Detailed Explanation

A Class-D audio amplifier is, functionally, closer to a switching power converter than to a traditional linear audio amplifier: the output stage's transistors are driven fully on or fully off at a high carrier frequency (commonly several hundred kHz to a few MHz), with the audio signal encoded into the switching pattern — typically as pulse-width modulation, where the duty cycle at any instant represents the instantaneous audio amplitude. A low-pass characteristic (either an explicit LC filter or the speaker's own electrical and mechanical inertia) then recovers the audible waveform from the high-frequency switching pattern.

This is the same fundamental principle as a PWM-driven H-bridge motor driver or a MOSFET gate-driver-based switching converter — a switched output stage instead of a linear one — applied to audio rather than motor drive or DC-DC conversion, and it inherits the same core tradeoff: substantially higher efficiency than a linear approach, in exchange for needing deliberate management of switching-frequency energy that a linear design simply doesn't produce.

Half-Bridge vs Full-Bridge Output Stages

Most Class-D amplifier ICs use a full-bridge (bridge-tied load, BTL) output: two switching half-bridges drive the speaker differentially, with the audio signal appearing as the voltage difference between the two outputs rather than referenced to ground. This has two practical advantages over a single-ended half-bridge output: it works from a single supply rail with no large output coupling capacitor (a half-bridge output needs one, or a split supply, since its output swings around half the supply voltage rather than around ground), and it roughly doubles the achievable output voltage swing for a given supply rail, increasing available output power.

A single half-bridge output is simpler and is used in some space- or cost-constrained designs, or where a part combines multiple half-bridge channels flexibly rather than fixing them into full-bridge pairs — but review the specific IC's application circuit for the coupling-capacitor or split-supply requirement this topology usually carries.

Filterless vs LC-Filtered Output Design

Every Class-D amplifier's raw output carries switching-frequency energy on top of the recovered audio signal — the design question is whether that energy needs an explicit LC low-pass filter removed, or whether it's low enough, and the load short enough, to leave unfiltered:

• Filterless (filter-free) design relies on the amplifier IC's own modulation scheme (several proprietary schemes exist across vendors, generally aimed at minimising differential switching energy and keeping most of the residual common-mode energy below what a short cable radiates effectively) plus the natural inductance of the speaker voice coil to attenuate the switching frequency enough that no external filter is needed. This only holds for a limited speaker cable length — beyond that length, the cable itself starts to act as an efficient radiator at the switching frequency and its harmonics, and filterless operation is no longer appropriate. See the FAQ below for how to evaluate this for a specific design.
• LC-filtered design adds an explicit differential low-pass filter (an inductor per leg, in series with the speaker, plus a capacitor across the differential output) between the amplifier output and the speaker terminals, attenuating switching-frequency energy before it ever reaches the cable. This is the safer default whenever the speaker cable run is longer than a short in-enclosure lead, when the product needs to pass formal EMC emissions testing with margin, or when a sensitive RF receiver shares the enclosure or is nearby.

Analog vs Digital (I2S) Input

Class-D amplifier ICs are available with either analog input (a continuous audio voltage, with gain set by external resistors, an internal gain-select pin, or a digital gain-control interface) or digital input, most commonly I2S or TDM, where the IC accepts PCM audio samples directly and performs the digital-to-switching-pattern modulation internally without an intermediate analog stage.

Choosing between them follows the same logic as any digital-vs-analog interface decision: a design whose audio source is already digital (an MCU's I2S peripheral, a DSP, a Bluetooth audio SoC) avoids an unnecessary DAC stage — and the extra conversion noise and latency that comes with it — by using a digital-input Class-D IC directly; a design whose source is genuinely analog (an analog microphone preamp, a legacy 3.5 mm input) is usually simpler with an analog-input part rather than adding an ADC purely to feed a digital-input amplifier.

EMI and EMC Considerations

The switching-frequency energy that makes Class-D efficient is also its main EMC liability, and it shows up in two related but distinct forms:

• Differential-mode switching ripple — the residual switching-frequency signal riding on the recovered audio waveform, attenuated by either the LC filter or the filterless modulation scheme's own design.
• Common-mode noise — driven by the fast switching edges (high dv/dt) coupling through parasitic capacitance to the enclosure, ground, or nearby conductors, largely independent of whether the design is filterless or LC-filtered, since an LC differential filter does relatively little to attenuate common-mode energy on its own.

A speaker cable is, electrically, an efficient antenna for whatever frequency content it carries — this is precisely why cable length is the key variable in the filterless-vs-filtered decision, and why even an LC-filtered design benefits from keeping speaker traces and cable runs as short and tightly coupled (differential-pair style, minimising loop area) as practical. For longer cable runs, a common-mode choke on the speaker output, in addition to the differential LC filter, is a common way to attenuate common-mode radiation the differential filter alone doesn't address — see how to reduce PCB EMI for the general EMI-reduction principles this builds on. Some parts also support spread-spectrum switching (dithering the carrier frequency slightly rather than running at one fixed frequency), which spreads peak emissions across a wider frequency range rather than concentrating them at one point in the spectrum — useful for meeting a narrowband emissions limit, though it doesn't reduce total radiated energy.

Design Considerations

• Confirm the filterless cable-length limit against the specific part's datasheet, not a general Class-D rule of thumb — this varies meaningfully between vendors and modulation schemes.
• Plan for a common-mode choke on the speaker output for any design with an external cable run, even if a differential LC filter is already present, since the LC filter alone does little to attenuate common-mode radiated emissions.
• Match the input type (analog vs I2S/TDM) to the actual audio source, rather than defaulting to whichever type is more familiar — an unnecessary DAC or ADC conversion stage adds cost, board area, and a real conversion-noise source for no benefit.
• Validate emissions with pre-compliance testing before committing to a filterless design for a product that must pass formal EMC certification — a filterless design that measures acceptable on the bench with a short test lead can still fail full compliance testing with the product's actual enclosure and cable routing. Zeus Design designs audio amplifier circuits and validates EMC performance as part of complete product electronics design.

Common Mistakes

• Assuming "filterless" means no EMI consideration is needed at all, rather than understanding it as a modulation-scheme-dependent tradeoff that only holds for a limited cable length and needs its own EMC validation.
• Relying on a differential LC filter alone to solve an emissions problem that's actually common-mode, when a common-mode choke is the component that actually addresses it.
• Adding an unnecessary DAC-then-ADC conversion chain by pairing a digital audio source with an analog-input amplifier IC (or vice versa) instead of choosing an amplifier IC that matches the source's native format.
• Routing speaker traces or cable with a large loop area, increasing radiated emissions regardless of whether the design is filterless or LC-filtered, since loop area directly affects how efficiently a given current radiates.
• Skipping EMC pre-compliance testing on the assumption that a reference design's filterless configuration will behave identically in a different enclosure and cable layout.