Electronics
Class A, AB or D amplifier?
In short: The class letter tells you how the output transistors work, not how the amplifier sounds. Class A is simple and clean, but turns most of its energy into heat. Class AB is the proven compromise behind most classic amplifiers. Class D switches its transistors fully on and off at high speed, so it is small, light and runs cool. The first switching amplifiers had real weaknesses, and their bad reputation was earned. Today's good modules (Hypex, Purifi, ICEpower), however, measure as cleanly as the best class AB amplifiers. When choosing, power, low output impedance, low noise and the ability to drive your speaker matter more than the letter.
In this article:
- What the letters mean
- Class A: always on, always hot
- Class B and crossover distortion
- Class AB: the classic compromise
- Class D: the switching amplifier
- Classes G and H
- Efficiency in a chart
- Example: how much heat at 2 × 50 W
- Why nobody used to trust class D
- How today's modules measure
- Heat, size, weight, price
- What matters more than the class
- Which one when
- Amplifiers and building speakers
- Key takeaways
- Mini glossary
What the letters mean
A power amplifier has a few power transistors (or valves) at its output. They work like taps: guided by the weak input signal, they let a strong current flow from the power supply into the speaker.
The class tells you how much of the waveform each output device conducts, in other words when the tap is open. Picture a sound wave as a hill above zero and a dip below zero. One full cycle is one hill and one dip.
| Class | What the output device conducts | In one sentence |
|---|---|---|
| A | the whole wave, all the time | the tap never closes, current flows even in silence |
| B | exactly half the wave | one device handles the hills, the other the dips |
| AB | a little more than half | like B, but with a small current even in silence so the handover overlaps |
| C | less than half | only in radio transmitters, unsuitable for music |
| D | the device is either fully on or fully off | a switch that clicks a hundred thousand times a second |
| G, H | like AB | AB that adjusts its supply voltage to the volume |
The letters are simply the order in which the circuit types were named. D does not stand for “digital”. It was just the next free letter after C.
Class A: always on, always hot
In class A, a constant bias current flows through the output transistors, at least as large as the biggest current going to the speaker. The music only nudges this current up and down. The transistor never switches off and never has to “take the baton” from another transistor. That makes class A, in principle, the simplest route to low distortion.
The price is heat. The amplifier draws the same power from the supply all the time, whether music is playing or not. Whatever does not go to the speaker turns into heat in the heatsink. And most of all in silence.
- Single-ended class A (one device, fed through a resistor or a current source) has a theoretical efficiency of at most 25 % at full power.
- Push-pull class A (two devices working against each other) or single-ended with an output transformer or choke has a theoretical maximum of 50 %.
- In practice it is lower, because the transistor cannot swing all the way to the supply voltage. At low listening levels efficiency is close to zero.
That is why class A amplifiers have huge heatsinks, heavy transformers and fairly modest power. A well-known example is the 10 W amplifier John Linsley Hood published in Wireless World in 1969. Today class A is offered by companies such as Pass Labs and Accuphase. Many valve amplifiers run in class A too; more on that in Valves or transistors.
The efficiency of ideal class A with a sine wave rises in direct proportion to power: η = 50 % × P / Pmax (push-pull) or 25 % × P / Pmax (single-ended). At a tenth of full power, push-pull class A is only 5 % efficient.
Class B and crossover distortion
Class B solves the heat problem simply. Two transistors split the work: the upper one pushes current into the speaker during the hills, the lower one pulls it back during the dips. In silence nothing flows. Ideal class B reaches an efficiency of π/4, about 78.5 %, at full power.
The trouble is the handover. A bipolar transistor only starts to conduct at around 0.6 V between base and emitter. Near zero, when one transistor is finishing and the other has not yet started, neither conducts properly. A small “step” appears in the wave. This is called crossover distortion.
Crossover distortion is nasty for three reasons:
- It is worst at low volume. The step is always the same size, so with a weak signal it makes up a bigger part of the wave. Ordinary distortion, by contrast, is highest at full power.
- It creates high-order harmonics. A sharp kink in the wave adds tones far above the original, and the ear hears those as harshness much more readily than a gentle rounding of the peaks.
- It is easy to miss when measuring at full power. An amplifier with a nice “0.1 % at 50 W” figure can have much worse distortion at 0.1 W.
That is why pure class B is almost never used in hi-fi.
Class AB: the classic compromise
Class AB is class B with a small bias current added, typically tens to a few hundred milliamps. Both transistors are slightly on around zero and hand over smoothly. The step disappears or shrinks to a level that feedback easily suppresses.
- At low volume the amplifier is effectively working in class A, at high volume in class B.
- Efficiency at full power is typically around 50 to 70 % in practice, depending on the design and the load.
- The bias current has to be set and kept stable as the transistors warm up. A circuit with a transistor mounted on the heatsink takes care of that. Too little current brings crossover distortion back; too much wastes heat and, with bipolar transistors, can paradoxically make distortion worse. Douglas Self covers this in detail in his book on amplifier design.
Most integrated amplifiers and receivers from the 1970s to today are built on class AB. A well-designed class AB amplifier measures excellently, and it is a proven, robust technology.
Class D: the switching amplifier
Class D takes a completely different approach. The output transistors are never “a little bit on”. They are either fully on or fully off. A transistor that is on has very low resistance, so there is almost no voltage across it. A transistor that is off passes almost no current. In both states the product of voltage and current, in other words the heat, is small.
So how does clicking turn into music? The most common method is pulse-width modulation (PWM). The amplifier compares the music with a fast triangle wave, typically at a few hundred kilohertz. When the music is above the triangle, the output connects to the positive supply, otherwise to the negative one. The result is a train of rectangular pulses whose width follows the music: on a hill the positive pulses are wide, in a dip they are narrow.
After the transistors comes an output filter, usually an inductor and a capacitor (an LC filter) with a cutoff somewhere between 30 and 100 kHz. It passes the music and removes the fast switching, so the speaker receives a smooth signal. It is a bit like a dimmer with a light bulb: the bulb flickers very fast, but your eye only sees the average brightness.
Importantly, class D is still an analogue amplifier. The pulse width changes continuously, not in steps. Some amplifiers accept a digital input and modulate directly from the numbers, but that is a separate matter from class D itself.
Good modules reach around 90 % efficiency or more at full power. Losses come from the switching itself (the transistor spends a moment passing between states), from the on-resistance of the transistor and from the filter inductor. Switching continues even in silence, so the amplifier draws a few watts at idle.
Classes G and H
These are improved class AB designs that save heat in a different way than switching:
- Class G has two or more supply voltage rails. At low volume it runs from the low rail and wastes little. On peaks it briefly switches to the higher rail.
- Class H varies the supply voltage continuously so that it tracks the music with a small margin.
Sonically the result matches class AB, it just runs cooler. They are used mainly in high-power professional amplifiers. Note that manufacturers do not always use these two letters consistently, and sometimes swap them.
Efficiency in a chart
The best way to compare the classes is efficiency: how much of the power drawn from the wall ends up in the speaker. The chart shows how it changes with volume. The left edge is what matters: in normal listening an amplifier spends most of its time at a few percent of its power. For how many watts you really need, see How many watts do you really need?
At 10 % power, ideal class B is about 25 % efficient, push-pull class A 5 % and typical class D around 80 %. At 1 % power, which is normal for quiet evening listening, the figures are about 8 %, 0.5 % and 30 %. Even class D is less efficient at low volume, because its fixed idle loss is then larger than the output power itself. In absolute watts, though, it is still very little.
Example: how much heat at 2 × 50 W
Take a stereo amplifier rated at 2 × 50 W into 8 Ω. Let's work out how much heat it produces at full power, at low volume (2 × 1 W) and in silence.
Peak voltage at 50 W into 8 Ω: U = √(2 × 50 × 8) ≈ 28.3 V, so a ±28.3 V supply (ideal, no headroom).
Push-pull class A: the supply always delivers 2 × Pmax = 100 W per channel, regardless of the music.
Class B: power drawn from the supply P = (2/π) × Usupply × Ipeak. At full power (I = 28.3 / 8 ≈ 3.54 A) that is about 63.7 W; at 1 W (U = 4 V, I = 0.5 A) about 9 W per channel.
Heat = power drawn − power to the speaker.
| Class | Heat at 2 × 50 W | Heat at 2 × 1 W | Heat in silence |
|---|---|---|---|
| A (push-pull, ideal) | 100 W | 198 W | 200 W |
| B (ideal) | 27 W | 16 W | 0 W |
| AB (real, approximate) | 40 to 60 W | 20 to 30 W | 5 to 20 W |
| D (typical, 90 % at full power) | about 11 W | about 2 to 5 W | about 2 to 5 W |
What this tells you:
- Class A runs hottest in silence. The ideal version turns 200 W into heat even when nothing is playing. A real 2 × 50 W class A amplifier needs even more, often 250 to 400 W from the wall all the time, which is like a small radiator. That is why such an amplifier weighs 20 to 40 kg, mostly because of the heatsinks and the transformer.
- Class B and AB run hottest somewhere in the middle. Ideal class B has its highest losses at about 40 % of full power, namely 4/π² × Pmax, or about 20 W per channel. That is what the heatsink is sized for.
- Class D produces around 11 W of heat at 2 × 50 W, ten times less than class A. An aluminium plate is enough instead of a finned heatsink.
The figures for real class AB and D are approximate. They depend on the bias current, voltage headroom, power supply and the specific design.
Why nobody used to trust class D
The idea of a switching amplifier is old; it was described as early as the 1950s. The first attempts to sell it in the 1960s, such as the Sinclair X-10 kit, did not work out technically: transistors of the time switched slowly and distortion was high. Cheap, fast enough MOSFETs only came later, and class D went mainstream around the turn of the millennium. The first generation had real problems that enthusiasts heard and measurements confirmed:
- Speaker-dependent response. The output LC filter is tuned for a specific load, for example 8 Ω. But a speaker is not a fixed resistor; its impedance changes with frequency. If the feedback was taken before the filter, the filter was left uncontrolled. Into a high impedance it could show a treble rise of several dB, into a low one a drop. The same amplifier therefore played the treble differently with different speakers.
- Higher output impedance in the treble. The filter inductor sits in series with the speaker. Its impedance rises with frequency, so the amplifier has a weaker “grip” on the speaker exactly where the tweeter works.
- High-frequency residue and interference. A switching residue remained at the output, tens to hundreds of millivolts at hundreds of kilohertz. You cannot hear it, but it could disturb measuring equipment, radio reception or downstream devices. Cheap measurements then showed nonsensical numbers.
- Distortion rising with frequency and volume, when the feedback was not enough at high frequencies.
That is where the reputation “class D is for subwoofers, not music” came from. For some of the early products it was true.
How today's modules measure
The breakthrough came with designs that take the feedback after the output filter. The feedback then “sees” what actually reaches the speaker and corrects the filter's errors. The result:
- Hypex (Netherlands): the UcD and later NCore modules by designer Bruno Putzeys. The FusionAmp is also built on them; see the article Hypex FusionAmp.
- Purifi (Denmark): the Eigentakt modules designed by Bruno Putzeys and Lars Risbo. They are among the best-measuring power stages of all.
- ICEpower: spun off from Bang & Olufsen, its modules power many active speakers and amplifiers.
For good modules from these brands, independent measurements show:
- distortion and noise across normal listening levels in the thousandths of a percent or lower, far below the threshold of audibility,
- a frequency response that changes by only fractions of a dB with different loads across the audible band,
- output impedance in the hundredths of an ohm, similar to good class AB.
When measuring, you need a filter in front of the analyser that removes the switching residue (for example the Audio Precision AUX-0025). Without it, class D distortion comes out needlessly bad.
That does not mean every class D amplifier is good. Cheap boards with chips such as the TPA3116 or TPA3255 from Texas Instruments can sound decent, but some have a load-dependent response, more noise or a weak power supply. The same applies as with class AB: what counts is the specific design and its measurements, not the letter.
Heat, size, weight, price
| A | AB | D | |
|---|---|---|---|
| Efficiency at full power | 20 to 50 % | 50 to 70 % | 85 to 95 % |
| Heat at low volume | maximum | medium | low |
| Heatsink | huge | medium | small or none |
| Typical weight for 2 × 100 W | rare, tens of kg | 10 to 20 kg | 1 to 5 kg |
| Typical power | up to 2 × 50 W | up to hundreds of W | from a few W to thousands of W |
| Standby consumption | high | medium | low |
| Main risk of a poor design | too little power, supply noise | crossover distortion | load-dependent response, interference |
On price (roughly): a small class D amplifier with a TI chip starts at around €50 to €150. Ready-made amplifiers with Hypex or Purifi modules cost from about €500 to several thousand euros. You can buy decent class AB from a few hundred euros. Class A with sensible power is inherently expensive because of the heatsinks, transformer and chassis, usually thousands of euros.
With class AB and A, a heavy transformer really is necessary. Class D is often paired with a switch-mode power supply, which is why it is light. So weight is no longer a measure of quality, only of how the amplifier handles energy.
What matters more than the class
As long as an amplifier works within its limits, well-designed class A, AB and D are usually indistinguishable in a blind test. The differences you can actually hear mostly come from these things:
1. Enough power
An amplifier that overloads and clips the peaks sounds bad regardless of its class. Musical peaks are 10 to 20 dB above the average, so headroom matters. The calculation is in the article How many watts do you really need?
2. Low output impedance
The speaker and the amplifier's output impedance form a voltage divider. The speaker's impedance changes with frequency, so with a high output impedance the level of individual frequencies changes too.
The speaker's impedance ranges from 4 Ω (minimum) to 30 Ω (woofer resonance).
Amplifier with 0.05 Ω output impedance: 4 / 4.05 → −0.11 dB, 30 / 30.05 → −0.01 dB. A difference of 0.1 dB, inaudible.
Amplifier with 2 Ω output impedance (typical of some valve amps): 4 / 6 → −3.5 dB, 30 / 32 → −0.6 dB. A difference of about 3 dB, a clearly audible change in tonal balance.
This is one of the main reasons why some valve amplifiers can sound different with the same speaker. The figure is often given as the damping factor = 8 Ω / output impedance. Above roughly 20 to 50, raising it further makes practically no difference.
3. Low noise
With sensitive speakers, such as horns or full-range drivers above 95 dB sensitivity, amplifier noise can be heard as hiss from a metre away. Look for a signal-to-noise ratio referenced to 1 W, or a noise figure in microvolts. A good amplifier has output noise in the order of tens of microvolts.
4. Stability and current into a real load
A speaker is not a resistor. It has inductance, crossover capacitance and an impedance that dips below the nominal value at some frequencies. A good amplifier must:
- deliver enough current into 4 Ω, or even 3 Ω at the impedance minimum,
- stay stable with a capacitive load, for example electrostatic speakers or long cables,
- have a frequency response that does not depend on the load.
Measurements “into an 8 Ω resistor” will not show this. What helps are measurements into 4 Ω, into a simulated speaker load, or a check of whether the power nearly doubles at 4 Ω.
Which one when
- You want the most music for your money, little heat and a small box: class D with a good module (Hypex, Purifi, ICEpower) or a quality class D integrated amplifier. At today's prices it is the sensible default choice.
- You are building an active speaker or subwoofer: class D. The amplifier sits inside the cabinet, where there is little space and little air for cooling.
- You own or are buying a classic integrated amplifier: good class AB is perfectly fine. There is no need to replace it just because of the letter.
- Class A: if you enjoy its simplicity, want to build one yourself, have sensitive speakers and do not mind the heat and the electricity bill. It is a choice made out of love for the circuit, not a requirement for good sound.
- Valves: a separate topic with its own compromises, see Valves or transistors.
Whatever the class, check independent measurements before you buy: distortion at both low and high power, power into 4 Ω, noise and response into different loads. If you want to know whether you can really hear a difference, compare amplifiers blind at the same volume, for example following the guide on the ABX blind test page. Levels must be matched to a tenth of a dB, otherwise the louder one always “wins”.
Amplifiers and building speakers
When you design a speaker in Loudsmith, the amplifier class only shows up indirectly:
- Passive speaker: check the calculated impedance. If the minimum drops below about 3 Ω, you need an amplifier that handles low loads. That applies to AB and D alike. More on passive and active speakers in the article Active or passive.
- The app's calculations assume a voltage source, meaning an amplifier with almost zero output impedance. With a valve amplifier at 1 to 3 Ω, the real response will differ, especially around the woofer resonance and the crossover.
- Active speaker: a built-in class D module with DSP (such as the Hypex FusionAmp) saves you both the crossover and an external amplifier. Each driver gets its own channel, which also makes the power requirements easier.
- Cooling in the cabinet: class D gets warm too. A plate amp in the back panel needs free space outside and should not be covered by damping material on the inside.
- Measuring: when you measure a speaker driven by a switching amplifier, the switching residue can upset some sound cards. Usually it does not matter, but if you measure impedance through a sound card, it is better to use a linear amplifier or a card with a good input filter.
Key takeaways
- The letter is about heat, not sound. The class determines how the output transistors work and how much energy they turn into heat.
- Class A runs hottest in silence. A 2 × 50 W class A stereo amplifier is like a small radiator all the time.
- Class D's bad reputation is a thing of the past. Modules with feedback taken after the output filter measure as cleanly as the best class AB.
- Power, output impedance, noise and handling 4 Ω matter more than the letter.
- Compare blind and at the same volume. Otherwise the louder one always wins.
The good news for speaker builders is that you do not have to tie yourself to a particular amplifier. Design a cabinet in Loudsmith with a sensible impedance that does not dip too low, and it will play well from any decent class.
Mini glossary
- Amplifier class: the way the output transistors or valves operate.
- Bias current: the current that flows through the output devices even in silence.
- Efficiency: the percentage of the power drawn that goes to the speaker. The rest is heat.
- Crossover distortion: a step in the wave where the transistors hand over in class B.
- Push-pull: two output devices working against each other, one on the positive half, the other on the negative half.
- PWM: pulse-width modulation. The information is carried in the width of the pulses.
- Output filter: the inductor and capacitor after a class D stage that remove the switching.
- Feedback: part of the output is fed back to the input so the amplifier corrects its own errors.
- Output impedance: the amplifier's internal resistance, which sits in series with the speaker.
- Damping factor: the speaker's nominal impedance divided by the amplifier's output impedance.
- MOSFET: a type of transistor that can switch very fast. The basis of today's class D amplifiers.
Sources: D. Self, Audio Power Amplifier Design (Focal Press/Routledge, several editions); B. Cordell, Designing Audio Power Amplifiers (McGraw-Hill 2011, 2nd ed. Routledge 2019); J. Linsley Hood, Simple Class A Amplifier (Wireless World 1969); B. Putzeys, Simple Self-Oscillating Class D Amplifier with Full Output Filter Control (AES Convention 2005); K. Nielsen, Audio Power Amplifier Techniques with Energy Efficient Power Conversion (PhD thesis, DTU 1998); datasheets from Hypex, Purifi, ICEpower and Texas Instruments. The efficiency, heat and output impedance calculations are our own; the class D curve is a typical model.
Free tools: Room planner · Speaker measurement · Blind ABX test