OFDM (Wi-Fi, LTE/NR, broadcast) has a high peak-to-average power ratio. Link budgets quote average power; the PA dies on the peaks. That mismatch is why a "passing" average Pout still fails ACLR or EVM.

1. What PAPR means on the bench
Instantaneous envelope spikes several dB above average. Complementary CDF (CCDF) traces on a VSA show how often peaks exceed average by X dB. A waveform with 10 dB PAPR at 0.01% probability needs headroom the datasheet CW P1dB never promised.

2. PA consequences
- Back off from P1dB / Psat by roughly PAPR plus margin if you want linear amplification without DPD.
- Efficiency tanks in back-off for Class AB; Doherty / envelope tracking / DPD exist because of this.
- Compression on peaks splats spectral regrowth → ACLR fail even when average power meter looks legal.

3. System knobs
- Crest factor reduction / clipping / windowing in baseband (trades EVM)
- DPD with enough memory depth for your PA
- Back-off policy tied to MCS: higher QAM needs cleaner AM/AM and AM/PM
- Front-end loss after the PA still burns average power budget but peaks were already clipped upstream

4. Measurement that matches the standard
Use the demod / ACLR personalities for your RAT, not a CW channel-power guess. Set the trigger and gate to the burst. Confirm the reference level and attenuator so the analyzer's own IMD does not invent shoulders.

5. Design checklist
- Get a CCDF of the exact waveform (FFT size, MCS, RBs)
- Size PA sat and supply for peaks, efficiency for average
- Budget feeder / filter loss after the PA separately from PA back-off
- Prototype with the real duty cycle; TDD heating differs from CW soak

6. When CW characterization still helps
Gain flatness, stability, hot S22. Just do not ship a CW-only compression curve as proof for OFDM compliance.

Average power is the billing meter; PAPR is the stress test. Design for both.