A power amplifier is most efficient when driven close to saturation and most linear when driven well below it. A 5G waveform, with its high peak-to-average ratio, sits in exactly the wrong place on that curve. Crest factor reduction and digital pre-distortion exist to move it — and together they decide what a radio site costs to run.
An OFDM signal is the sum of many subcarriers. Occasionally those subcarriers align in phase and produce a peak far above the average power — typically eight to twelve decibels above it before any treatment.
The amplifier has to accommodate that peak without clipping, which means operating with its average output well below saturation. That backed-off operating point is where amplifiers are least efficient, and efficiency is the single largest determinant of site energy consumption.
This is a recurring operating cost, not a capital one. An amplifier a few percentage points more efficient saves energy every hour of every year across every site in the network, which is why the digital front end attracts engineering attention out of proportion to its size.
Deliberately reduce the peak-to-average ratio before the signal reaches the amplifier, typically from around eight to twelve decibels down to six to nine. The peaks are clipped or cancelled in a controlled way.
Clipping is distortion. It degrades EVM and generates out-of-band emission, so the reduction is bounded by the spectral mask and the EVM budget rather than by how much peak reduction is desirable.
Apply the inverse of the amplifier's transfer characteristic to the signal before it arrives, so the amplifier's own non-linearity cancels it. The amplifier can then be driven harder for the same linearity.
Real amplifiers respond to recent history, not just the present sample, because of thermal and bias-network time constants. A memoryless correction is insufficient at wide bandwidths, which is why practical pre-distorters carry memory terms.
The characteristic moves with temperature, ageing, supply and frequency, so the correction is derived continuously from an observation receiver rather than fixed at calibration.
Higher average output power for the same spectral compliance, which is the same thing as higher efficiency at the same output power. Either way the site draws less.
OFDM sums many subcarriers, which occasionally align to produce peaks eight to twelve decibels above average power. Accommodating those peaks forces the amplifier to operate backed off from saturation, where it is least efficient — and amplifier efficiency dominates site energy consumption.
Deliberately reducing the peak-to-average ratio before the amplifier, typically to six to nine decibels, by clipping or cancelling peaks in a controlled way. It is bounded by the spectral mask and the EVM budget, because clipping is itself distortion.
Applying the inverse of the amplifier's transfer characteristic to the signal beforehand, so the amplifier's own non-linearity cancels it. That allows the amplifier to be driven harder for the same linearity, which is the same as higher efficiency at a given output.
Because real amplifiers respond to recent signal history, not just the present sample, owing to thermal and bias-network time constants. A memoryless correction is inadequate at the bandwidths 5G uses.
The amplifier characteristic moves with temperature, ageing, supply voltage and frequency. A correction fixed at calibration degrades in service, so the coefficients are derived continuously from an observation receiver.
Yes. Crest factor reduction changes the signal statistics the pre-distorter is trained against, so tuning them separately and combining them afterwards gives a worse result than designing the pair as one subsystem.
Written for engineers. Share it with one.
CFR and DPD as one subsystem.
PRODUCTAdaptive pre-distortion.
PRODUCTPeak reduction within the mask.
KNOWLEDGEThe number this is judged by.
PRODUCTThe hardware it runs in.
INDUSTRYWhere the energy goes.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.