What Is Predistortion?

Predistortion is a signal processing technique used to counteract the nonlinear behavior of electronic components, most commonly radio-frequency (RF) power amplifiers, by intentionally distorting the input signal in a complementary manner so that the combined effect at the output approximates a linear transfer characteristic. It is one of the principal methods for linearizing amplifiers that must operate near saturation, where power efficiency is highest but distortion is most severe. Predistortion is foundational to modern wireless communications, where spectral efficiency standards leave little tolerance for out-of-band emissions caused by amplifier nonlinearity.

The need for predistortion arises from the fundamental tradeoff in amplifier design between power efficiency and linearity. Operating an amplifier at or near its saturation point yields the greatest DC-to-RF conversion efficiency, but in that region the device compresses gain and generates harmonic and intermodulation distortion products. Modern communication waveforms such as OFDM, used in LTE and 5G NR, carry high peak-to-average power ratios and are especially sensitive to these distortion effects.

Digital Predistortion

Digital predistortion (DPD) is the dominant implementation in contemporary base-station and satellite transmitter design. In a DPD system, a behavioral model of the power amplifier's nonlinear transfer function is identified, typically using a memory polynomial, a Volterra series expansion, or a machine learning surrogate, and the inverse of that model is applied to the baseband digital signal before it reaches the digital-to-analog converter and the RF chain. The predistorted signal then passes through the amplifier and emerges with the distortion largely cancelled. Because amplifier characteristics shift with temperature, aging, and load, practical DPD implementations include a feedback path that continuously re-identifies the model and updates the predistortion coefficients. IEEE Xplore publications on digital predistortion of RF power amplifiers document the performance gains achievable with adaptive DPD versus fixed-coefficient approaches.

Analog and RF Predistortion

Before digital signal processing became fast enough to implement DPD in the baseband chain, analog predistortion was the practical option. Analog predistorters insert a nonlinear circuit element, such as a diode expander or a FET biased in a complementary nonlinear region, directly into the signal path ahead of the amplifier. The approach is simpler and adds no latency to the signal path, but it is far less accurate than DPD because the analog circuit cannot be readily adapted and its correction is inherently wideband without frequency selectivity. Analog predistortion remains relevant in millimeter-wave and optical systems where the data rates exceed what digital feedback loops can track in real time.

Neural Network and Machine Learning Approaches

As behavioral models grow more complex, particularly for wideband amplifiers with significant memory effects, researchers have applied neural network architectures as the predistortion function. A network trained on input-output pairs from the amplifier can capture highly nonlinear relationships and memory effects that polynomial models approximate poorly. Neural network predistortion techniques for nonlinear RF amplifiers have shown improved linearization in scenarios where the amplifier exhibits pronounced memory, such as in wideband multi-carrier transmission. The tradeoff is higher computational cost during training and a larger hardware footprint for real-time inference compared with a compact polynomial model. A survey of digital predistortion for RF power amplifiers from the University of Leeds reviews the evolution of DPD architectures from fixed polynomial models through neural-network-based approaches.

Applications

Predistortion has applications in a wide range of systems and domains, including:

• Cellular base stations (4G LTE, 5G NR), where spectral mask compliance requires tight control of out-of-band emissions
• Satellite communications uplink amplifiers, which must operate at high efficiency over long periods
• Cable television (CATV) headend equipment, for maintaining signal fidelity across hundreds of channels
• Optical fiber transmission systems, where electro-optic modulators exhibit nonlinear transfer curves
• Radar transmitters, where waveform fidelity directly affects range and Doppler accuracy