Class A
100% of the input signal is used (conduction angle Θ = 360° or 2π; i.e., the active element remains conducting (works in its "linear" range) all of the time.
Where efficiency is not a consideration, most small signal linear amplifiers are designed as Class A. Class A amplifiers are typically more linear and less complex than other types, but are very inefficient. This type of amplifier is most commonly used in small-signal stages or for low-power applications (such as driving headphones). Subclass A2 is sometimes used to refer to vacuum tube Class A stages where the grid is allowed to be driven slightly positive on signal peaks, resulting in slightly more power than normal Class A (A1; where the grid is always negative), but incurring more distortion.
Class B
50% of the input signal is used (Θ = 180° or π; i.e., the active element works in its linear range half of the time and is more or less turned off for the other half). In most Class B, there are two output devices (or sets of output devices), each of which conducts alternately (push–pull) for exactly 180° (or half cycle) of the input signal; selective RF amplifiers can also be implemented using a single active element.
These amplifiers are subject to crossover distortion if the transition from one active element to the other is not perfect, as when two complementary transistors (i.e., one PNP, one NPN) are connected as two emitter followers with their base and emitter terminals in common, requiring the base voltage to slew across the region where both devices are turned off.
Class AB
Here the two active elements conduct more than half of the time as a means to reduce the cross-over distortions of Class B amplifiers. In the example of the complementary emitter followers a bias network allows for more or less quiescent current thus providing an operating point somewhere between Class A and Class B. Sometimes a figure is added (e.g., AB1 or AB2) for vacuum tube stages where the grid voltage is always negative with respect to the cathode (Class AB1) or may be slightly positive (hence drawing grid current, adding more distortion, but giving slightly higher output power) on signal peaks (Class AB2); another interpretation being higher figures implying a higher quiescent current and therefore more of the properties of Class A.
Class C
Less than 50% of the input signal is used (conduction angle Θ < 180°). The advantage is potentially high efficiency, but a disadvantage is high distortion.
Class D
These use switching to achieve a very high power efficiency (more than 90% in modern designs). By allowing each output device to be either fully on or off, losses are minimized. The analog output is created by pulse-width modulation; i.e., the active element is switched on for shorter or longer intervals instead of modifying its resistance. There are more complicated switching schemes like sigma-delta modulation, to improve some performance aspects like lower distortions or better efficiency.
Class E
The class E/F amplifier is a highly efficient switching power amplifier, typically used at such high frequencies that the switching time becomes comparable to the duty time. As said in the class-D amplifier the transistor is connected via a serial-LC-circuit to the load, and connected via a large L (inductance) to the supply voltage. The supply voltage is connected to ground via a large capacitor to prevent any RF-signals leaking into the supply. The class-E amplifier adds a C between the transistor and ground and uses a defined L1 to connect to the supply voltage.
The following description ignores DC, which can be added afterwards easily. The above mentioned C and L are in effect a parallel LC-circuit to ground. When the transistor is on, it pushes through the serial LC-circuit into the load and some current begins to flow to the parallel LC-circuit to ground. Then the serial LC-circuit swings back and compensates the current into the parallel LC-circuit. At this point the current through the transistor is zero and it is switched off. Both LC-circuits are now filled with energy in the C and the L0. The whole circuit performs a damped oscillation. The damping by the load has been adjusted so that some time later the energy from the Ls is gone into the load, but the energy in both C0 peaks at the original value, to in turn restore the original voltage, so that the voltage across the transistor is zero again and it can be switched on.
With load, frequency, and duty cycle (0.5) as given parameters and the constraint that the voltage is not only restored, but peaks at the original voltage, the four parameters (L, L0, C and C0) are determined. The class E-amplifier takes the finite on resistance into account and tries to make the current touch the bottom at zero. This means the voltage and the current at the transistor are symmetric with respect to time. The Fourier transform allows an elegant formulation to generate the complicated LC-networks. It says that the first harmonic is passed into the load, all even harmonics are shorted and all higher odd harmonics are open.
Class F
In push–pull amplifiers and in CMOS, the even harmonics of both transistors just cancel. Experiment shows that a square wave can be generated by those amplifiers and theory shows that square waves do consist of odd harmonics only. In a class D amplifier, the output filter blocks all harmonics; i.e., the harmonics see an open load. So even small currents in the harmonics suffice to generate a voltage square wave. The current is in phase with the voltage applied to filter, but the voltage across the transistors is out of phase. Therefore, there is a minimal overlap between current through the transistors and voltage across the transistors. The sharper the edges the lower the overlap.
While class D sees the transistors and the load as two separate modules, the class F admits imperfections like the parasitics of the transistor and tries to optimise the global system to have a high impedance at the harmonics. Of course there has to be a finite voltage across the transistor to push the current across the on state resistance. Because the combined current through both transistors is mostly in the first harmonic it looks like a sine. That means that in the middle of the square the maximum of current has to flow, so it may make sense to have a dip in the square or in other words to allow some over swing of the voltage square wave. A class F load network by definition has to transmit below a cut off frequency and to reflect above.
Any frequency lying below the cut off and having its second harmonic above the cut off can be amplified, that is an octave bandwidth. On the other hand, an inductive-capacitive series circuit with a large inductance and a tunable capacitance may be simpler to implement. By reducing the duty cycle below 0.5, the output amplitude can be modulated. The voltage square waveform will degrade, but any overheating is compensated by the lower overall power flowing. Any load mismatch behind the filter can only act on the first harmonic current waveform, clearly only a purely resistive load makes sense, then the lower the resistance the higher the current.
Class F can be driven by sine or by a square wave, for a sine the input can be tuned by an inductor to increase gain. If class F is implemented with a single transistor, the filter is complicated to short the even harmonics. All previous designs use sharp edges to minimise the overlap.
Class E uses a significant amount of second harmonic voltage. The second harmonic can be used to reduce the overlap with edges with finite sharpness. For this to work energy on the second harmonic has to flow from the load into the transistor, and no source for this is visible in the circuit diagram. In reality, the impedance is mostly reactive and the only reason for it is that class E is a class F amplifier with a much simplified load network and thus has to deal with imperfections.
In many amateur simulations of class E amplifiers, sharp current edges are assumed nullifying the very motivation for class E and measurements near the transit frequency of the transistors show very symmetric curves, which look much similar to class F simulations.