Inferensys

Glossary

AM-PM Conversion

Nonlinear distortion where the phase shift introduced by a power amplifier varies as a function of the instantaneous input signal amplitude, degrading modulation accuracy and causing spectral regrowth.
ML engineer managing model versions on laptop, version history visible, technical Git-like workflow.
PHASE NONLINEARITY IN POWER AMPLIFIERS

What is AM-PM Conversion?

AM-PM conversion is a nonlinear distortion mechanism in power amplifiers where the phase shift introduced to the output signal varies as a function of the instantaneous input signal amplitude, degrading modulation accuracy and spectral purity.

AM-PM conversion is the amplitude-dependent phase distortion inherent in nonlinear power amplifiers (PAs), where the insertion phase through the device changes with the instantaneous envelope power of the input signal. Unlike AM-AM distortion, which compresses the amplitude, AM-PM conversion rotates the signal constellation, introducing phase errors that directly degrade Error Vector Magnitude (EVM) and cause spectral regrowth in digitally modulated waveforms. This effect is particularly severe in Gallium Nitride (GaN) PAs operating near saturation to maximize Power-Added Efficiency (PAE).

In mmWave beamforming arrays, AM-PM conversion becomes a dominant linearization challenge because the phase distortion varies per element due to active impedance mismatch and antenna crosstalk. Digital Predistortion (DPD) must therefore correct both amplitude and phase nonlinearities simultaneously, often using complex baseband models like the Generalized Memory Polynomial (GMP) or neural network linearization architectures that learn the joint AM-AM/AM-PM characteristic. Uncompensated AM-PM conversion results in constellation rotation, increased bit error rates, and failed Adjacent Channel Leakage Ratio (ACLR) compliance.

NONLINEAR PHASE DISTORTION

Key Characteristics of AM-PM Conversion

AM-PM conversion is a critical nonlinear effect in power amplifiers where the phase shift introduced by the device varies as a function of the instantaneous input signal amplitude. Unlike AM-AM distortion, which affects magnitude linearity, AM-PM conversion degrades phase modulation accuracy and is particularly severe in mmWave GaN amplifiers operating near compression.

01

Physical Origin in Semiconductor Devices

AM-PM conversion arises from amplitude-dependent capacitance variations in the transistor's depletion regions. As the input drive level changes, the nonlinear input capacitance of the gate-source junction shifts, altering the phase of the forward transfer function.

  • GaN HEMTs: Exhibit strong AM-PM due to trapping-induced dynamic capacitance modulation
  • GaAs pHEMTs: Show moderate AM-PM from varactor-like gate capacitance behavior
  • CMOS PAs: Experience AM-PM from drain-bulk junction capacitance nonlinearity
  • LDMOS devices: Display relatively benign AM-PM characteristics compared to III-V technologies

The effect intensifies as the amplifier approaches gain compression, where the input capacitance becomes strongly signal-dependent.

02

Impact on Modulation Quality

AM-PM conversion directly degrades Error Vector Magnitude (EVM) by rotating constellation points in a signal-envelope-dependent manner. This is especially damaging for high-order modulation schemes.

  • QPSK: Tolerates moderate AM-PM due to wide decision boundaries
  • 16-QAM: Shows measurable EVM degradation from phase rotation of outer constellation points
  • 64-QAM: Requires AM-PM compensation to meet 3GPP EVM limits
  • 256-QAM and 1024-QAM: Extremely sensitive; even 1-2 degrees of AM-PM can cause symbol errors
  • OFDM signals: High PAPR causes symbols at different amplitudes to experience different phase shifts, creating phase noise-like distortion
03

Relationship with AM-AM Distortion

AM-PM and AM-AM distortion are coupled phenomena that occur simultaneously in real power amplifiers. The complex gain can be expressed as:

G(|x|) = G_AM(|x|) · e^(j·φ(|x|))

Where G_AM represents amplitude nonlinearity and φ(|x|) represents the AM-PM characteristic.

  • AM-AM dominates near saturation where gain compression is severe
  • AM-PM is often significant even in the linear region before hard compression
  • The two effects interact: phase distortion can appear as amplitude error after demodulation
  • Memory effects cause both AM-AM and AM-PM to become frequency-dependent
  • Joint compensation requires complex-valued predistortion addressing both magnitude and phase
04

Measurement and Characterization

AM-PM conversion is characterized by measuring the phase deviation between input and output as a function of instantaneous input power. Key measurement parameters include:

  • Degrees per dB (°/dB): The slope of phase change versus input power, typically 1-5 °/dB for GaN PAs
  • Total phase variation: The peak-to-peak phase shift across the operating power range, often 10-30 degrees
  • AM-PM at P1dB: Phase shift at the 1 dB compression point, a standard figure of merit
  • Dynamic AM-PM: Phase shift variation due to memory effects, measured with modulated signals
  • Vector network analyzer (VNA) power sweeps provide static AM-PM
  • Wideband modulated measurements using vector signal analyzers capture dynamic AM-PM behavior
05

Compensation Through Digital Predistortion

AM-PM conversion is corrected by complex-valued digital predistortion that applies an inverse phase rotation before the power amplifier. The predistorter must generate a phase advance that exactly cancels the PA's phase lag at each amplitude level.

  • Memory polynomial DPD: Models AM-PM with complex coefficients that capture both magnitude and phase corrections
  • Generalized memory polynomial (GMP): Adds cross-terms to handle AM-PM that varies with signal envelope frequency
  • Neural network DPD: Learns the inverse AM-PM characteristic directly from I/Q waveforms without explicit model structure
  • LUT-based predistortion: Stores complex gain corrections indexed by instantaneous amplitude
  • Adaptive coefficient tracking: Updates AM-PM compensation in real-time as temperature and bias conditions change

Effective AM-PM correction can reduce phase variation from 20+ degrees to less than 1 degree, enabling high-order modulation at mmWave frequencies.

06

mmWave-Specific Challenges

At millimeter-wave frequencies, AM-PM conversion becomes more severe and complex due to several compounding factors:

  • Doherty amplifier architectures: The load modulation mechanism introduces additional AM-PM from the peaking amplifier's phase discontinuity
  • Antenna array interactions: Active impedance mismatch from beam-steering causes element-specific AM-PM variations
  • GaN trapping effects: Slow charge capture/release creates long-term memory in the AM-PM characteristic
  • Thermal dynamics: Self-heating at high mmWave power densities modulates the phase response over millisecond timescales
  • Wideband signals: 400 MHz and 800 MHz bandwidths in 5G NR require AM-PM correction across the entire signal bandwidth
  • Over-the-air DPD: Must compensate for combined AM-PM of the entire array, including mutual coupling effects
AM-PM CONVERSION ESSENTIALS

Frequently Asked Questions

Addressing the most common engineering questions about amplitude-to-phase conversion in power amplifiers, its impact on signal integrity, and mitigation strategies for mmWave systems.

AM-PM conversion is a nonlinear distortion mechanism in power amplifiers where the phase shift introduced by the amplifier varies as a function of the instantaneous input signal amplitude. Unlike an ideal amplifier that maintains constant phase regardless of input power, real PAs exhibit input-amplitude-dependent phase variations caused by the nonlinear capacitance of the transistor junction. As the input drive level changes, the transistor's internal parasitic capacitances—particularly the gate-to-source and gate-to-drain capacitances in FET-based devices—modulate, altering the device's S21 phase response. This effect is especially pronounced in Class AB and Doherty amplifiers operating near compression, where the device impedance state transitions significantly with signal envelope. In Gallium Nitride (GaN) devices, additional contributions arise from trapping effects that dynamically shift the knee voltage and intrinsic capacitances.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.