Crest Factor Reduction
Reshaping Signal Statistics to Reclaim Amplifier Headroom
The motivation for crest factor reduction comes directly from how multicarrier modulation behaves statistically. When hundreds or thousands of OFDM subcarriers add in phase for an instant, the composite envelope spikes far above its mean. These peaks are infrequent (the complementary cumulative distribution function shows that a 10 dB peak occurs only about 0.01 percent of the time), yet the amplifier must reproduce them linearly or the resulting distortion splatters into adjacent channels. Sizing the amplifier and its back-off for those rare peaks is enormously inefficient, so the designer instead reshapes the signal to make the peaks smaller while keeping the average and the demodulated constellation intact.
The classic approach is clipping and filtering. The complex baseband samples are limited to a magnitude threshold, which immediately reduces the peak but produces broadband clipping noise. A band-limiting filter then removes the out-of-band portion of that noise, but filtering allows the clipped peaks to partially regrow, so the clip-and-filter cycle is iterated two to four times until the residual peaks settle at the target. More sophisticated methods replace hard clipping with peak windowing, which scales each peak by a smooth window so the spectral spreading is far lower, or with peak cancellation, which subtracts a bandwidth-matched cancellation pulse only at the offending samples.
Every CFR method spends part of the system EVM budget. Because the clipping distortion is effectively in-band noise that lands on the constellation, the engineer treats the PAPR target as a slider between efficiency and signal quality. A base station running 64-QAM LTE can usually accept a 7 dB target; a 256-QAM 5G NR carrier with a tighter EVM limit must stop closer to 8 dB. CFR is almost always cascaded ahead of digital predistortion, since reducing the dynamic range first lets the predistorter linearize the amplifier with less peak headroom.
Governing Relationships
CF = Vpeak / Vrms → PAPR (dB) = 20·log10(CF) = 10·log10(Ppeak / Pavg)
Clipping Ratio (threshold vs. rms):
γ = Aclip / σ where σ2 = mean envelope power
Reclaimed Output Power:
ΔPout ≈ PAPRraw − PAPRtarget (dB of back-off recovered)
Where Vpeak is the instantaneous envelope peak, Vrms the rms envelope, γ the clipping ratio, and σ the rms level. Example: a raw 10.5 dB PAPR clipped to a 7.0 dB target recovers ≈ 3.5 dB of average output power while EVM rises from ≈ 1% to ≈ 5%.
CFR Technique Comparison
| Technique | PAPR Reduction | Out-of-Band (ACPR) Impact | In-Band EVM Cost | Complexity | Typical Use |
|---|---|---|---|---|---|
| Hard clipping only | 3 to 4 dB | Poor (wideband splatter) | Low | Very low | Rarely used alone |
| Clipping & filtering | 3 to 4 dB (2 to 4 iterations) | Moderate (regrowth managed) | Moderate | Low to medium | Macro base stations |
| Peak windowing | 2.5 to 4 dB | Good | Moderate | Medium | Multi-carrier transmitters |
| Peak cancellation | 3 to 4.5 dB | Excellent | Low to moderate | High | 5G NR / wideband PAs |
| Tone reservation | 2 to 3 dB | Excellent (no in-band) | None in-band | High | DVB / standards with spare tones |
Frequently Asked Questions
How much PAPR can crest factor reduction remove before EVM fails the spec?
For 64-QAM LTE downlink with a raw PAPR near 10 to 11 dB, CFR can usually clip to a 6.5 to 7.5 dB target while keeping composite EVM under the 8% limit. Pushing below 6 dB drives EVM past 4 to 5% and breaks the in-band requirement. For 256-QAM, where the EVM limit tightens to about 3.5%, the practical CFR floor rises to roughly 7.5 to 8 dB because the denser constellation tolerates far less clipping distortion.
What is the difference between hard clipping, peak windowing, and peak cancellation?
Hard clipping truncates the magnitude at a threshold and is spectrally messy, generating wideband distortion that needs filtering, which then causes peak regrowth over several iterations. Peak windowing multiplies peaks by a smooth window (Hann or Kaiser) instead of truncating, giving much lower out-of-band emissions for the same peak reduction. Peak cancellation subtracts a bandwidth-matched cancellation pulse only at the peaks, offering the best ACPR control but the highest complexity.
Why is crest factor reduction usually combined with digital predistortion?
CFR and DPD are complementary. CFR reshapes the signal statistics to lower PAPR so the amplifier can be driven harder, but does nothing about the amplifier's own AM-AM and AM-PM nonlinearity. DPD linearizes that nonlinearity but cannot reduce peak excursions. Running CFR first lets the predistorter work with a smaller dynamic range, so the combined chain can reach 40 to 50% average efficiency on a Doherty PA while meeting a minus 45 dBc ACPR mask. The order matters: CFR must precede DPD.