RF Power Amplification
High-Efficiency Class-E RF Amplifiers at 13.56 MHz
A linear RF stage can throw away a third or more of its input as heat. A switch-mode Class-E stage does not. This explains how Class-E works conceptually, what limits its efficiency in practice, and why 13.56 MHz is a comfortable operating point for GaN.
On this page
Why linear stages waste power
In a linear amplifier class, Class A, Class AB or Class B, the transistor acts as a controlled current source. At every instant it carries current while a voltage stands across it, and the product of the two is dissipated inside the device. Class A is the worst case, with a theoretical drain efficiency ceiling of 50 percent and often far less in practice. Class AB and B do better by cutting conduction angle, but the fundamental problem remains: current and voltage are both non-zero at the same time, so the device is always burning power.
Switch-mode classes, Class D, Class E and Class F, break that overlap. The transistor is driven as a switch, either fully on with almost no voltage across it, or fully off with almost no current through it. In the ideal case the instantaneous device dissipation is zero throughout the cycle, so all the DC input becomes RF output. The price is that the output is a switched waveform, not a faithful copy of the input, so these classes suit a fixed-frequency carrier rather than a modulated linear signal. An RF plasma thruster runs on a fixed 13.56 MHz carrier, which is exactly the case switch-mode amplification is built for.
How Class-E works conceptually
Class-E adds one idea to the switch-mode picture: it uses a carefully tuned reactive load network to shape the voltage across the switch so that switching happens at the right moment. Picture the sequence over one RF cycle:
- The transistor turns on and pulls its drain node close to ground. Current from the DC feed choke and the load network flows through the closed switch with very little voltage across it, so on-state loss is small.
- The transistor turns off. The drain node is now driven by a shunt capacitor and a series inductor-capacitor branch feeding the load. The network is tuned so the drain voltage rises smoothly from zero, peaks well above the supply, then falls back.
- The network brings the drain voltage back to zero, and ideally its slope to zero as well, just as the transistor is about to turn on again. The switch then closes with no voltage across it and no stored capacitor energy to dump. This is zero-voltage switching, and it removes the turn-on loss that dominates hard-switched stages.
The shunt capacitor is partly or wholly the transistor's own output capacitance, so the device parasitic that hurts a linear design is absorbed into the Class-E network instead of fighting it. That is why Class-E scales gracefully to high frequency.
The efficiency ceiling and what erodes it
With an ideal switch and ideal reactive components, Class-E drain efficiency approaches 100 percent. Real hardware sits below that, and the losses are predictable:
| Loss mechanism | Cause | Mitigation |
|---|---|---|
| Conduction loss | Non-zero on-resistance carrying the drain current | Low on-resistance device, GaN over silicon |
| Finite switching time | Device cannot change state instantly, so a brief voltage-current overlap remains | Fast device, moderate frequency, clean gate drive |
| Output capacitance non-linearity | Device capacitance varies with drain voltage, detuning the ideal waveform | Design the network around the effective capacitance |
| Passive component loss | Series resistance and core loss in the inductor and capacitors | High-Q air-core or low-loss components |
| Gate drive power | Energy to charge and discharge the gate every cycle | Efficient resonant or optimised gate driver |
| Load mistuning | Actual load differs from the design impedance, so switching is no longer at zero voltage | Impedance matching network, adaptive if the load moves |
A well-executed 13.56 MHz GaN Class-E stage can realistically reach high conversion efficiency, which is a large part of why the topology is used for space-grade RF generators. MHz Labs does not publish an efficiency figure for its own unit while it is unmeasured.
Why 13.56 MHz suits GaN Class-E
13.56 MHz is an ISM band allocation used worldwide for industrial RF plasma work, so the operating frequency is regulatory-friendly and shares an ecosystem with mature industrial generators. It is also a sweet spot for Class-E. It is high enough that the reactive network components are small and can be low-loss air-core parts, and the transistor output capacitance is a useful fraction of the shunt capacitance rather than a nuisance. It is low enough that switching transitions are a small part of the RF period even for a mid-power device, so finite switching time costs little efficiency. Modern GaN transistors switch far faster than 13.56 MHz demands, which leaves margin for the gate drive and the layout rather than pushing the device to its limit.
Sensitivity to load impedance
Class-E buys its efficiency by tuning the network to one specific load impedance. Move the load away from that design point and the drain voltage no longer returns to zero at the switching instant, so turn-on loss reappears, the peak drain voltage can rise, and efficiency falls. A plasma thruster is a difficult load in exactly this way: the antenna impedance jumps sharply between the unlit and lit states at ignition and then drifts with power, flow and magnetic field. That is why a Class-E stage for a thruster is paired with an impedance matching network, and why the MHz Labs design uses a real-time adaptive matching network that keeps the amplifier seeing its design impedance as the plasma changes.
Frequently asked questions
What is the difference between Class-D, Class-E and Class-F?
All three run the transistor as a switch. Class-D uses two devices in a push-pull pair with a filter. Class-E uses a single device and a reactive network that shapes the drain waveform for zero-voltage switching. Class-F shapes harmonics at the load to square up the voltage or current waveform. Class-E is popular at HF for its simplicity and its tolerance of device output capacitance.
Can Class-E amplify a modulated signal?
Not directly, because the output is a switched waveform locked to the drive. Amplitude information has to be added by modulating the supply or the drive. For a fixed-frequency plasma-thruster carrier this limitation does not apply.
Why not just use a linear amplifier and accept the loss?
In space the loss is the problem. Every watt the amplifier dissipates has to be conducted to a baseplate and radiated away, which adds radiator area and mass to the whole spacecraft. A 20 point efficiency difference is a large mass penalty.
What happens to a Class-E stage if the plasma load changes suddenly?
Efficiency drops and peak drain voltage rises because switching is no longer at zero voltage. A matching network holds the presented impedance near the design value; an adaptive one tracks the change in real time and also protects the device during the ignition transient.
Does MHz Labs use Class-E?
MHz Labs is developing its RF power processing unit around a GaN switch-mode power stage in the Class-E family. The unit is on a TRL 4 to 6 roadmap and has no flight heritage yet.
Sources and further reading
- Goebel and Katz, Fundamentals of Electric Propulsion, NASA JPL DESCANSO, for the propulsion power-budget context.
- MKS Instruments, ELITE 13.56 MHz RF plasma generators, an industrial reference for the operating frequency.
- Takahashi, Helicon-type radiofrequency plasma thrusters and magnetic plasma nozzles, review of RF thruster loads.
Designing a switch-mode RF stage for a thruster?
MHz Labs builds the high-efficiency RF power processing unit so your propulsion team does not have to. Ask for the current design targets and integration guidance.
Get in touch