Peak-to-Average Power Ratio (PAPR)

PAPR is formally defined as the ratio of the instantaneous peak power to the average power of a signal over a given time interval. For a complex baseband signal x(t), it is expressed as:

• Formula: PAPR(dB) = 10 log₁₀( max|x(t)|² / E[|x(t)|²] )
• Peak Power: The maximum instantaneous envelope power occurring during the observation period
• Average Power: The mean signal power computed over the same interval
• Complementary CDF (CCDF): Used to statistically characterize PAPR, showing the probability that the PAPR exceeds a given threshold

A constant-envelope signal like GMSK has a PAPR of 0 dB, while an OFDM signal can exhibit PAPR exceeding 12 dB.

The PAPR of a signal is fundamentally determined by its modulation format and multiple access scheme. Different waveforms exhibit vastly different peak-to-average characteristics:

• OFDM (Orthogonal Frequency Division Multiplexing): High PAPR due to the superposition of many independent subcarriers that can constructively interfere, creating large amplitude spikes
• SC-FDMA (Single-Carrier FDMA): Lower PAPR than OFDM, making it preferred for the uplink in 4G LTE where handset power efficiency is critical
• DFT-s-OFDM: Used in 5G NR uplink to combine the benefits of OFDM with reduced PAPR through discrete Fourier transform precoding
• WCDMA: Moderate PAPR due to code-domain multiplexing and pulse shaping
• Constant Envelope Modulations: GMSK and FSK exhibit 0 dB PAPR, enabling the use of highly efficient saturated power amplifiers

PAPR directly forces a critical trade-off between linearity and power efficiency in the transmitter design. The PA must be operated with sufficient back-off to accommodate signal peaks:

• Output Back-Off (OBO): The PA's average operating point must be reduced below its 1 dB compression point by an amount proportional to the PAPR to avoid clipping and distortion
• Efficiency Penalty: A signal with 10 dB PAPR forces a Class-A PA to operate at less than 10% efficiency, wasting significant DC power as heat
• Doherty and Envelope Tracking: Advanced PA architectures are specifically designed to maintain efficiency over a wider dynamic range, partially recovering the efficiency lost to PAPR back-off
• Crest Factor Reduction (CFR): A complementary technique that intentionally reduces PAPR before the PA, allowing operation closer to saturation at the cost of controlled in-band distortion

PAPR is the primary driver of nonlinear distortion that DPD must correct. The higher the PAPR, the more severe the PA nonlinearity that must be linearized:

• Dynamic Range Requirement: The DPD model must accurately characterize the PA's nonlinear behavior across the entire signal envelope range, from average power to peak excursions
• Model Complexity Scaling: Higher PAPR signals require higher-order nonlinear terms and longer memory depth in the predistorter model to capture the extended nonlinear operating region
• Peak Clipping Interaction: When CFR is used alongside DPD, the clipping distortion introduced by CFR must be accounted for in the DPD training to avoid over-correction
• Training Signal Considerations: The PAPR of the training signal used for DPD coefficient extraction should match the operational waveform to ensure the model is valid over the correct dynamic range

Accurate PAPR characterization is essential for system design and uses both time-domain and statistical methods:

• CCDF Curves: The Complementary Cumulative Distribution Function is the standard tool, plotting the probability that the instantaneous power exceeds a given PAPR threshold. A typical specification is the PAPR at the 10⁻⁴ probability point
• Peak Detection: Real-time peak detection circuits in the transmitter chain monitor for instantaneous peaks that could damage the PA or violate spectral masks
• Vector Signal Analyzers: Modern VSAs compute PAPR, CCDF, and crest factor directly from captured IQ samples
• Statistical vs. Deterministic: While PAPR is often discussed as a single number, it is inherently a statistical quantity; the absolute maximum PAPR is unbounded for Gaussian-like signals, making the CCDF the practical design metric

Several signal processing techniques are employed to reduce PAPR before the signal reaches the PA, trading off complexity, distortion, and data rate:

• Clipping and Filtering: The simplest method, hard-limiting the signal amplitude and then filtering to reduce out-of-band emissions, at the cost of in-band distortion and EVM degradation
• Selected Mapping (SLM): Generates multiple candidate signals with different phase sequences and transmits the one with the lowest PAPR, requiring side information to be communicated to the receiver
• Tone Reservation: Reserves a subset of subcarriers that do not carry data and uses them to generate a peak-canceling signal, avoiding in-band distortion on data subcarriers
• Active Constellation Extension (ACE): Intelligently extends outer constellation points to reduce peaks without increasing the bit error rate
• Companding: Applies a nonlinear companding function (like μ-law) to compress the signal dynamic range at the transmitter and expand it at the receiver

A baseband signal processing technique applied before the power amplifier to deliberately reduce the PAPR of the transmitted waveform. CFR algorithms, such as peak windowing and clipping-and-filtering, limit signal peaks to allow the PA to operate closer to its compression point with higher average efficiency. The trade-off is a controlled increase in Error Vector Magnitude (EVM) and spectral regrowth, which must be managed alongside the DPD system.

A critical metric quantifying in-band signal distortion, measured as the vector difference between the ideal reference constellation points and the actual transmitted symbols. High-PAPR signals driven into a PA's nonlinear region suffer severe constellation compression and phase rotation, degrading EVM. DPD systems aim to minimize EVM by pre-compensating for these nonlinear effects, ensuring modulation accuracy.

The primary regulatory metric for spectral regrowth caused by nonlinear amplification. When a high-PAPR signal is compressed by a PA, intermodulation distortion generates power in adjacent frequency channels, potentially interfering with other operators. ACLR measures the ratio of in-channel power to leaked power. Effective DPD must suppress ACLR below stringent 3GPP limits, a task made more difficult as the signal's PAPR increases.

The amount by which the PA's average output power is reduced below its saturated output power (Psat) to operate in a more linear region. High-PAPR signals require significant back-off to avoid peak compression, which drastically reduces power-added efficiency (PAE). The economic and thermal cost of back-off is the primary motivation for using CFR and DPD to linearize the PA while operating it closer to compression.

The standard statistical tool for characterizing a signal's PAPR. The CCDF curve plots the probability that the instantaneous signal power exceeds a given threshold above the average power. It provides a complete view of the signal's peak distribution, not just the absolute maximum. PA designers use the CCDF to determine the required back-off and to evaluate the effectiveness of CFR algorithms in suppressing rare, high-amplitude peaks.

A power supply modulation technique that dynamically adjusts the PA's drain or collector voltage to track the instantaneous envelope of the transmitted signal. By supplying only the voltage needed for the current amplitude, ET dramatically improves efficiency for high-PAPR signals. The dynamic supply modulation introduces its own nonlinearities, requiring a joint DPD solution that compensates for both the RF PA distortion and the ET modulator's non-ideal behavior.