Why A Class D Amplifier Is A Switching Amplifier
A switching amplifier treats each transistor as a snap-on, snap-off valve rather than a dimmer. The transistor sits fully open or fully closed at every instant, so it wastes very little energy as heat. The cost: rapid switching creates high-frequency noise that a tuned LC filter must strip before the signal reaches the speaker cone.
- On-or-off, not in between — the transistor is held in hard saturation or hard cutoff at every moment, which is what makes “switching” the accurate name and the source of the efficiency jump.
- The filter sets the ceiling — the inductance and capacitance of the LC output filter determine the upper frequency limit and add a small phase shift that softens the very top of the treble.
- Heat is the lost road — a linear class-A or class-AB stage would deliver lower source impedance and firmer cone damping, but it burns most of its input power as waste heat in a large finned sink.
- Faster parts shift the balance — higher switch frequencies and lower-DCR inductors let the filter run hot enough that audible ripple barely tints the mids, but they tighten the electromagnetic-interference budget.
The Mechanism, Step By Step
Think of a bicycle brake lever. When you squeeze it, the rubber pad is either fully pressed flat against the metal rim or fully lifted clear. There is no half-squeezed middle state. Now picture that pad snapping on and off roughly 400,000 times per second. The fraction of time it stays closed encodes how much braking force the wheel should feel. The wheel still decelerates smoothly, yet the pad itself is not sitting in a half-clamped, half-heating-up condition. A switching amplifier uses the same logic with a transistor instead of a brake pad.
In a linear amplifier, the output transistor acts like a variable resistor. It sits somewhere in the middle of its curve, passing a fraction of the supply voltage while a fraction of the current flows through it. That middle zone is where both voltage and current exist at the same time, and their product is power turned into heat. The transistor must be physically large enough to dump that heat into a finned metal sink, and the sink often becomes the largest part of the whole chassis.
A switching amplifier bypasses the middle zone entirely. The transistor is driven into one of two extreme states. In hard saturation, the device is wide open. The voltage across it drops to a few hundredths of a volt, so almost no heat is generated even though full current flows. In hard cutoff, the device is fully closed. No current passes, so again no heat appears. By flipping between these two states hundreds of thousands of times each second, the transistor wastes only a small fraction of the input power as heat. The catch is that the raw on-off pulse train is not a clean audio waveform. It is a ragged, high-frequency stream that would make the speaker cone buzz with ultrasonic noise if it reached the voice coil untouched.
That is where pulse-width modulation (PWM) enters. A modulator circuit compares the incoming audio signal against a fast triangular carrier wave. The wider the audio level at any given moment, the longer each pulse stays on within one carrier cycle. The result is a train of rectangular pulses whose average height follows the audio. This train still lives at a switch frequency typically between 200 kHz and 1 MHz, well above the 20 kHz ceiling of human hearing. The ear cannot resolve individual pulses, but the speaker coil would still react to them without help.
The final stage is the LC output filter. An inductor, usually a few millihenrys of copper wire, resists sudden changes in current. A capacitor, several microfarads of ceramic or film, resists sudden changes in voltage. Wired in series and parallel around the speaker terminals, they form a low-pass network that soaks up the 200 kHz to 1 MHz ripple and passes only the audio band below 20 kHz. The inductor’s DC resistance (DCR) and the filter’s natural resonance set the upper frequency limit and introduce a small phase shift. That phase shift is why the very top of the treble can read a shade softer than the output of a linear amplifier, whose transistors impose no such passive roll-off on the waveform.
[Audio input] ➔ [PWM modulator] ➔ [Switching transistor pair] ➔ [LC output filter] ➔ [Speaker voice coil]
Side By Side: The Real Differences
| Practical dimension | Switching (Class-D) | Linear (Class-A / Class-AB) |
|---|---|---|
| Transistor state at any instant | Hard-saturated or hard-cutoff; no mid-zone dwell | Operates in the linear mid-region where voltage and current overlap |
| Typical input power lost as heat | 5% or less, even at mid-power | 50% or more of the drawn power turns into waste heat |
| Heatsink requirement | Small fin or none at all | Large extruded-aluminum fin array, often the biggest part of the chassis |
| Output impedance into the voice coil | A few ohms, set by the LC filter inductance and DCR | Well under 1 ohm, set by the negative-feedback loop |
| Treble upper limit | Bounded by the LC filter resonance; small phase shift above 15 kHz | Bounded only by the transistors’ current headroom; no passive roll-off in the output path |
| EMI and noise budget | Tighter as switch rate climbs; gate-drive losses rise with each doubling | Minimal; the output stage radiates far less high-frequency energy |
How It Changes The Sound
Bass. The LC filter’s inductance and the inductor’s DCR set a floor on how cleanly the amplifier drives the woofer’s voice coil. At modern switch rates of 400 kHz or higher, the residual ripple sits far above the 20 kHz hearing limit, so bass notes do not carry a visible or audible wobble. However, the filter’s inductance adds a small source impedance in series with the voice coil. A 4-ohm woofer returning back-electromotive force during a fast transient sees that extra impedance, so the electrical damping grip is looser than a linear stage with a sub-ohm output would provide. In a large room or with a low-sensitivity driver, that looser grip can show up as a bass note that rings a hair longer before it settles. The note does not lose punch; it simply takes a fraction of a millisecond more to stop.
Vocals. The midrange, roughly 200 Hz to 4 kHz, is where the PWM duty cycle tracks the signal most linearly. With lower-DCR inductors and switch frequencies at or above 400 kHz, the filter output is smooth enough that vocal timbre does not shift or thin out. The one subtle change: because the switching output stage presents a source impedance of a few ohms rather than a fraction of an ohm, the midrange driver’s voice coil is damped less firmly. A well-chosen filter keeps the difference below the threshold of most listeners in a quiet, small room. At louder levels, though, the slight loss of damping can make the voice sit a touch behind the tweeter compared with a linear amp. The singer still sounds natural, but the precise moment the note stops is a shade less clipped.
High frequencies. Above roughly 10 kHz, the LC filter’s inductance begins to rise in impedance as frequency climbs, while the capacitor’s impedance falls. The two together produce a gentle roll-off and a phase shift that reaches its peak near the filter’s resonant point. The audible result is that the air and sparkle above 15 kHz soften by a small margin. Cymbal attacks lose a tiny amount of their initial brightness. A designer who picks a higher-Q inductor and a smaller capacitor can push that roll-off past 20 kHz, keeping the treble edge intact. But the electromagnetic-interference footprint grows, and the PCB layout has to keep the switching nodes well away from the filter’s sensitive side. In practice, most well-built Class-D units keep the roll-off just above the top of the audible band, so the difference is subtle.
Common Myths, Checked
“A switching amplifier is just a digital amplifier, so it sounds digital.”
The PWM modulator is digital at the comparison stage, where a comparator decides whether to turn the switch on or off for that cycle. But the output of the transistor pair and the LC filter is a continuous, analog voltage waveform. The speaker cone sees a smooth analog signal, not a stream of 1s and 0s. The “digital” label applies only to the modulation circuit, which happens to use a digital process to shape the pulse widths. The audible result is an analog waveform whose only digital artifact is the residual ripple the filter is designed to suppress. You would hear that artifact only if the filter were poorly designed or the switch rate were far too low for the application.
“The higher the switch frequency, the better the sound, so designers should push it as high as possible.”
Raising the switch rate from 200 kHz to 1 MHz does shrink the inductor and capacitor, and it pushes the ripple further out of the audible band. But the gain in audible flatness levels off quickly. Past roughly 400 kHz, the remaining improvement lands in the far ultrasonic range, well below what the ear can resolve. What does not level off is the cost side: each doubling of switch rate increases gate-drive loss in the transistors and widens the electromagnetic-interference spectrum the PCB layout and filter must contain. Past a certain point the EMI budget gets so tight that adding more inductance just to tame the noise erases the very benefit the higher frequency was supposed to deliver. A well-chosen switch rate sits where the audible improvement is still meaningful and the thermal and EMI costs remain manageable.
Reader Questions, Answered
Why Does The Transistor Have To Avoid Its Linear Region?
In the linear region, the transistor has both a meaningful voltage across it and a meaningful current through it at the same time. Their product is power, and that power shows up as heat in the silicon die. In hard saturation the voltage across the device drops to a few hundredths of a volt, so even at full current the heat is minimal. In hard cutoff the current is zero, so there is no heat at all. By staying in one of those two extreme states at every instant, the transistor draws power from the supply and delivers it to the speaker without wasting more than a small percentage as waste heat. That is why the heatsink can be tiny or absent entirely, and why the amplifier body stays cool to the touch even at high output.
What Exactly Does The LC Filter Do, And Why Does It Limit The Treble?
The switching pair throws out a PWM pulse train at, say, 400 kHz. The inductor resists rapid current changes; the capacitor resists rapid voltage changes. Wired together they form a second-order low-pass network that smooths the rectangular pulses into a continuous audio waveform. But a low-pass filter is not flat to infinity. Its inductance and capacitance set a resonant frequency, and above that point the amplitude drops and the phase shifts. The result is that the very top of the treble, above 15 kHz, is attenuated by a small amount relative to the mids. A well-chosen filter places that roll-off high enough that most listeners do not notice it during normal listening.
If Efficiency Is So Much Better, Why Do Some Engineers Still Pick A Linear Output Stage?
In a small room at moderate listening level, the speaker cone is gently driven and the damping a linear stage provides is rarely the limiting factor. The linear path’s well-under-1-ohm source impedance and firm electrical grip on the voice coil matter most when the amplifier is pushed hard into a low-impetance 4-ohm driver or into a large, bass-heavy room. For a compact desktop amp, a nearfield speaker pair, or a system that will sit on a shelf, the switching path is the well-chosen road: less heat, smaller enclosure, and the damping difference falls below the threshold of the ear at normal volume. The linear stage’s advantage only becomes audible under demanding conditions.
Does A Higher Switch Frequency Make The Amplifier Sound More Accurate?
Higher switch rates let the LC filter use smaller, lower-DCR inductors, which reduces residual ripple in the bass and mids. The audible improvement is real, but it levels off quickly. Past roughly 400 kHz to 1 MHz the remaining gain sits in the ultrasonic range, well below what the ear can resolve. What does not level off is the electromagnetic-interference budget: each doubling of switch rate adds high-frequency spectral content that the PCB layout, the filter components, and the power supply must suppress. The design cost rises roughly in step while the audible reward does not, so there is a practical ceiling where more switch speed buys no perceptible sonic gain. Engineers settle at the rate where the filter is small enough to be practical and the EMI stays within the PCB’s layout rules.
(This text was published on hifireport.com. Unauthorized reuse is not permitted.)