I spent last week characterizing a 3.5 GHz PA for a small-cell prototype, and the back-off efficiency numbers forced me to rethink which architecture actually earns its complexity. The setup: a GaN device rated 40 W peak, tested with a 100 MHz modulated signal at 7 dB PAPR. At peak power the PA hit 55% drain efficiency, but at 7 dB back-off it collapsed to around 18%. That is where the real operating point lives, so that is where the architecture choice matters. Envelope tracking first: I paired the PA with a supply modulator tracking the envelope. Back-off efficiency climbed to roughly 32%, and linearity stayed manageable with DPD. The catch was the modulator itself — its bandwidth and efficiency at low envelope voltages ate into the gains, and the supply modulator added a whole new stability problem to debug. Doherty second: a classic two-way Doherty with the same device gave about 28% at back-off, with no extra active circuitry. Simpler, cheaper, and the load modulation behaved well across the band. But the peaking amplifier's turn-on point drifted with temperature, and the efficiency peak was narrower than the signal's actual power distribution. What I actually learned: matching the architecture to the signal statistics matters more than the headline peak efficiency. For a signal that spends most of its time 6–8 dB below peak, the Doherty's simpler efficiency curve was easier to keep in its sweet spot. For wider PAPR signals, envelope tracking pulled ahead despite the extra complexity. My rule of thumb now: if the PAPR is under 8 dB and the band is narrow, start with Doherty; if the signal is wideband with high PAPR, budget the effort for envelope tracking. How do you choose between them on your projects? Have you seen the Doherty turn-on drift bite you in production?