Power Amplifier Non-Linearity

The non-linear relationship between input signal amplitude and output signal amplitude, causing gain compression or expansion. As the amplifier approaches saturation, the gain curve flattens, compressing the output waveform.

• Gain Compression: Output amplitude fails to track input linearly near saturation
• 1 dB Compression Point (P1dB): The input power where gain drops by 1 dB from ideal linear response
• Spectral Regrowth: Amplitude distortion generates intermodulation products that broaden the signal bandwidth

Example: A 20 MHz LTE signal processed through a Class AB amplifier near P1dB may exhibit 3-5 dB of gain compression at peak envelope excursions.

The non-linear relationship where the phase shift introduced by the amplifier varies as a function of the instantaneous input amplitude. This phase modulation distorts the signal constellation.

• Phase Rotation: Higher amplitude signals experience different phase delays than lower amplitude signals
• Constellation Warping: QAM and PSK constellations exhibit amplitude-dependent angular spreading
• EVM Degradation: Phase errors directly increase Error Vector Magnitude

AM-PM distortion is particularly problematic in Doherty amplifiers, where the peaking amplifier's turn-on characteristic introduces significant phase discontinuities.

The dependence of the amplifier's current output on past input values, caused by thermal dynamics, biasing network impedance, and semiconductor trapping effects.

• Thermal Memory: Junction temperature changes modulate gain over microsecond to millisecond timescales
• Electrical Memory: Bias circuit capacitors and inductors create frequency-dependent impedance at the modulation envelope bandwidth
• Trapping Effects: Charge traps in GaN and LDMOS transistors cause slow drain current transients

Memory effects make static LUT-based DPD insufficient, requiring Volterra series or memory polynomial models with temporal depth.

The broadening of a signal's bandwidth caused by intermodulation products generated when a non-linear amplifier processes a modulated signal. This creates out-of-band emissions that violate regulatory masks.

• Third-Order Intermodulation (IM3): The dominant regrowth mechanism, creating products at 2f₁-f₂ and 2f₂-f₁
• ACLR Violation: Spectral regrowth directly increases Adjacent Channel Leakage Ratio
• Bandwidth Expansion: A signal with bandwidth B can regrow to 3B or 5B depending on non-linearity order

Regulatory bodies like the FCC and ETSI impose strict ACLR limits (typically -45 dBc for LTE), making spectral regrowth a primary driver for DPD adoption.

The physical saturation of the transistor's drain current as the input drive increases, fundamentally limiting the amplifier's linear operating range.

• Clipping: Hard saturation where the output voltage cannot exceed the supply rail
• Soft Compression: Gradual gain reduction before hard clipping, common in Class AB biasing
• Back-Off Requirement: Amplifiers must operate 6-12 dB below P1dB to meet linearity specifications without DPD

Modern DPD systems can recover 3-5 dB of back-off, directly translating to power-added efficiency (PAE) improvements of 10-15 percentage points.

The amplifier's non-linear characteristics vary dynamically with the instantaneous envelope of the modulated signal, not just its average power.

• PAPR Impact: High peak-to-average power ratio signals (8-12 dB for OFDM) force the amplifier to handle extreme envelope excursions
• Crest Factor Sensitivity: Amplifiers with sharp gain roll-off near saturation distort high-PAPR signals more severely
• Envelope Tracking Compatibility: Amplifiers designed for envelope tracking exhibit supply-voltage-dependent non-linearity

This envelope dependence is why DPD models must operate on complex baseband samples rather than simple power-averaged metrics.

The two fundamental manifestations of non-linearity. AM-AM distortion describes the non-linear relationship between input amplitude and output amplitude, causing gain compression at saturation. AM-PM distortion is the unwanted phase shift that varies with instantaneous input power. Together, they distort the constellation diagram and generate spectral regrowth. These are the primary impairments that DPD must invert.

The power amplifier's output depends not only on the current input but also on past values. Thermal memory effects arise from transistor heating and cooling over microsecond timescales. Electrical memory effects stem from bias network impedance and trapping phenomena in semiconductor materials. These effects break the static AM-AM/AM-PM assumption and require models with temporal depth, such as the Generalized Memory Polynomial or recurrent neural networks.

Two competing architectures for identifying DPD coefficients. Indirect Learning Architecture (ILA) swaps the PA input and output to estimate the postdistorter, then copies it as the predistorter—simple but sensitive to measurement noise. Direct Learning Architecture (DLA) iteratively minimizes the error between the desired linear output and actual PA output, offering superior performance at the cost of higher computational complexity. Neural network DPD typically employs DLA.

When a modulated signal passes through a non-linear amplifier, intermodulation products cause the signal's bandwidth to broaden—a phenomenon called spectral regrowth. This spills power into adjacent channels, violating regulatory masks. Adjacent Channel Leakage Ratio (ACLR) quantifies this leakage and is the primary metric DPD aims to minimize. Typical targets require ACLR below -45 dBc, demanding 20-30 dB of linearization improvement.

High-efficiency amplifier architectures that exacerbate non-linearity. Doherty PAs combine a main and peaking amplifier with an impedance inverter, achieving high back-off efficiency but introducing severe gain discontinuities. Envelope Tracking dynamically modulates the supply voltage to match the signal envelope, improving Power-Added Efficiency (PAE) but creating complex dynamic distortion. Both architectures demand advanced, neural network-based DPD to be commercially viable.