GaN HF Amplifiers: What a 100 W Evaluation Board Does—and Does Not—Prove

The useful question is not whether “GaN beats an old linear.” It is what each item being compared actually does. A 100 W wireless-power evaluation board, a configurable Class-E board, a 100 W HF/6 m accessory amplifier and an integrated polar-modulation transceiver belong to different product categories. Output watts alone do not make them substitutes.

EPC9083 and EPC9065: Different Boards, Different Jobs

The phrases “100 W,” “92%” and “Class E” do not describe one EPC amateur-radio amplifier. They point to at least two distinct development platforms:

The distinction is explicit in EPC’s own documents. The EPC9083 requires frequency- and load-specific Class-E components. The EPC9065 quick-start guide says the board has no current protection or enhanced controller, is intended for qualified professionals in a laboratory, is not intended for commercial use and is not designed for EMC compliance. Those are appropriate conditions for an evaluation tool—and decisive reasons not to present it as a completed amateur product.

XPA125B: A Finished Accessory with Different Boundaries

The XPA125B English manual identifies it as a solid-state linear power amplifier with an ATU. That manual and the linked English v1.0.2 document both list at least 100 W from 1.8–30 MHz and at least 80 W from 50–54 MHz. Xiegu also maintains a manufacturer download page; use documentation that matches the exact hardware and firmware rather than combining figures across files.

The later manual also lists 13 dB gain, a 12–15 V supply, up to 30 A transmit current and a nominal 14–500 Ω ATU range. Power gain belongs in dB, not dBi: 13 dB is a ratio of about 19.95, so 5 W input predicts roughly 100 W output before compression and loss. The same manual tells operators of unlisted generic QRP radios to start at 1 W; that is an input-safety instruction, not a promise of 100 W output from 1 W.

These specifications describe a much more complete user function than either EPC board. They still leave important comparison gaps:

• The single “spurious suppression ≥50 dB” line does not state detector bandwidth, reference convention, test frequencies, load, power or whether it means harmonics, all spurious products or something else.
• No two-tone IMD curve, occupied-bandwidth result, AM-AM/AM-PM data or adjacent-channel measurement is provided.
• No RF-output/DC-input efficiency curve is given. Multiplying maximum supply voltage by maximum current does not produce a valid operating-point efficiency.
• The manual warns that prolonged high-power use can overheat the PA, but does not publish a complete power-versus-mode, ambient-temperature and duty-cycle table.
• The ATU’s resistance range is not a Smith-chart coverage map and does not prove that every complex impedance between those resistance values can be matched at full power.

These are documentation limits, not proof that the unit fails a test. “Not specified” and “measured badly” are different conclusions.

GaN, Class E, Polar Modulation and Envelope Tracking Are Different Layers

Four terms often collapse into one marketing sentence even though they answer different questions:

A real RF waveform can be written in polar form as:

In a polar transmitter, the phase path generates the carrier with time-varying phase and the amplitude path controls output magnitude. Kahn’s 1952 envelope-elimination-and-restoration paper establishes the central requirement: the amplitude and phase components must be recombined with the correct time relationship. Modern DSP changes the implementation, not that physics.

Envelope tracking is different. EPC’s AN028 reference design uses a dynamic supply that follows a 20 MHz LTE envelope while a separate RF PA is represented by the load. Its reported 94.3% total converter efficiency applies to that envelope-supply demonstration—not to a complete RF transmitter and not to the EPC9083 or XPA125B.

Why a Polar Add-On Is a Transmitter Subsystem

An external polar box is possible. It can receive baseband audio and frequency/mode control, or digitize/decompose an incoming RF waveform. But that box must then recover or generate A(t) and φ(t), align their delays, calibrate amplitude and phase errors, drive the switching PA, select the correct output network, filter harmonics, sequence transmit/receive and handle faults. That is more than conventional linear gain.

The practical implementations make the boundary visible:

• The current QRP Labs QMX operating manual describes EER SSB, per-unit phase predistortion, separate USB/LSB phase-versus-amplitude synchronization and calibration that measures IMD3, IMD5 and IMD7 while adjusting delay.
• Polar Explorer was explicitly a companion transmitter, not an RF-input amplifier. Its developer says standalone development ended when the technology moved into FlexRadio’s integrated Aurora.
• FlexRadio’s current external-amplifier FAQ says Aurora’s polar transmitter needs separate carrier and envelope inputs that ordinary radios do not provide, so it is offered only as an integrated transceiver.

That evidence supports a narrower conclusion than “an external polar amplifier cannot exist.” It can exist, but unless the host exposes suitable synchronized signals, the external unit has to reproduce significant parts of the exciter and transmitter-control chain.

RF Coverage and Modulation Bandwidth Are Separate Problems

A board that operates at 6.78 MHz does not become a 1.8–54 MHz product because the information bandwidth is only a few kilohertz. Three bandwidths have to be kept separate:

• Carrier tuning range: every amateur band on which the RF power stage, driver and control system can operate.
• Instantaneous modulation bandwidth: the RF passband needed to preserve the wanted emission and the envelope-path bandwidth needed to reproduce A(t) with acceptable distortion.
• Output-network and filter behavior: the impedance transformation around the fundamental plus rejection of switching harmonics and other unwanted products.

The EPC9083 guide starts its Class-E design from a specific operating frequency, desired load power and load resistance. It also shows that moving the load above or below the design value changes switching loss and device voltage behavior. The EPC9065 likewise requires ZVS timing and load-range analysis. Neither result means Class E or D can never be multiband; it means the cited boards do not supply the band-switched matching, low-pass filters and controls needed to prove that product.

Efficiency Claims Need a Boundary and a Waveform

The most useful overall definition is:

But many published numbers use a narrower boundary: transistor drain efficiency, PA-board DC-to-RF efficiency, envelope-converter efficiency or power-supply efficiency. A complete transmitter also consumes driver, gate-drive, envelope-modulator, DSP, control, cooling and AC/DC power. State whether the signal is an unmodulated carrier or a modulated waveform, its peak-to-average ratio, the averaging interval, output level, band, supply, load and temperature.

A 92% result at 6.78 MHz into a wireless-power load is valuable evidence for the EPC9065 operating point. It is not the wall-to-RF efficiency of a filtered SSB transmitter. Conversely, the XPA125B’s published maximum 30 A draw is a protection/supply-sizing value, not an efficiency test point. FlexRadio’s 80% nominal Aurora figure is a product-level manufacturer claim with a wider boundary than one transistor measurement, but it still needs band-, mode- and level-specific test data for detailed comparison.

Signal Purity Is a Measured System Result

A switching PA naturally creates harmonic energy that the output network and low-pass filters must control. A polar transmitter adds other error mechanisms: amplitude-path ripple or compression, phase-path error, delay mismatch, PA AM-PM conversion, quantization, carrier leakage, supply noise and load-dependent behavior. Adaptive predistortion can reduce some errors, but only within the bandwidth, power, load and feedback conditions for which it remains stable and calibrated.

A meaningful comparison needs at least:

1. fundamental power, DC input and thermal equilibrium on every band;
2. harmonics and other spurious emissions with stated detector bandwidth and reference;
3. two-tone IMD for SSB at several output levels, including peak-reference convention;
4. occupied bandwidth or spectral regrowth for representative voice processing and data waveforms;
5. carrier and opposite-sideband suppression where applicable;
6. results into a nominal load and documented mismatch points, before and after protective foldback;
7. startup, band-change, keying, overdrive, overtemperature and filter/relay fault behavior.

“Spurious suppression,” “IMD3” and “clean signal” are not interchangeable metrics. Current ITU-R SM.329 covers the spurious domain, while ITU-R SM.1541 addresses out-of-band emissions. National rules control the amateur station; for example, current US 47 CFR §97.307 distinguishes necessary bandwidth, adjacent-channel splatter and spurious attenuation, with different numerical provisions below and above 30 MHz.

Compare Products at the Same Boundaries

Complete Functions Matter More Than Headline Watts

GaN can enable fast, efficient switching. Polar modulation can combine a nonlinear RF power stage with a reconstructed linear waveform. EPC’s boards are useful evidence for those building blocks, and current integrated amateur products show that the architecture can be commercialized.

A one-frequency wireless-power result characterizes its documented operating point; it does not establish a universal external HF amplifier. The 100 W/92% EPC9065 result belongs to a Class-D wireless-power platform, while the Class-E EPC9083 is a configurable 60 W evaluation board rather than a 100 W multiband radio product. The XPA125B provides a broader accessory-amplifier function, although its published documents do not answer every efficiency or spectral question.

The defensible conclusion is therefore conditional: compare complete functions and measurements under the same bands, waveforms, loads and boundaries. Device material, PA class and headline output are inputs to that comparison—not the verdict.

Questions, antenna-factor records or height trials to share? Contact RF.Guru.