Thermal AM-PM Distortion

Unlike static phase offset, thermal AM-PM introduces a phase modulation that is a function of the signal's amplitude history. As the input envelope power fluctuates, it modulates the junction temperature, which in turn alters the transistor's parasitic capacitances (Cgs, Cgd). This capacitance shift changes the phase of the amplifier's transfer function dynamically, creating a distortion that is nonlinear and history-dependent, not correctable by a simple static phase rotator.

The thermal time constants of a semiconductor device typically range from microseconds to milliseconds, corresponding to a thermal bandwidth of a few kilohertz to megahertz. This means the distortion mechanism is primarily excited by the low-frequency components of the signal envelope, not the RF carrier. Modulated signals with wide envelope bandwidths (e.g., 100 MHz NR) will have their low-frequency spectral components fall within this thermal bandwidth, making the AM-PM distortion a significant contributor to in-band error vector magnitude (EVM).

The root physical cause is the temperature sensitivity of nonlinear capacitances in the transistor. Key mechanisms include:

• Cgs Modulation: The gate-to-source capacitance changes with temperature due to shifts in the Fermi level and carrier mobility.
• Cgd (Miller) Effect: The gate-to-drain feedback capacitance is highly bias-dependent, and its temperature coefficient introduces a dynamic phase rotation.
• Threshold Voltage Drift: Vth decreases with rising temperature, altering the transistor's operating point and its associated small-signal capacitances.

Thermal AM-PM does not occur in isolation. The same self-heating that causes a phase shift also induces thermal AM-AM distortion (gain compression or expansion). These two effects are coupled through the device's electro-thermal physics. A predistorter must correct both simultaneously. A model that only addresses AM-AM will leave a residual phase error, and vice-versa. Joint AM-AM/AM-PM models with thermal memory terms are required for full linearization in modern GaN and GaAs PAs.

A key observable characteristic of thermal AM-PM is asymmetric spectral regrowth. Because the phase distortion is dispersive (frequency-dependent due to the thermal time constants), the upper and lower adjacent channel power ratios (ACLR) become unbalanced. This asymmetry is a direct fingerprint of memory effects and cannot be replicated by a memoryless nonlinearity. Analyzing the ACLR asymmetry provides a diagnostic metric for the severity of thermal AM-PM in a transmitter chain.

Standard memory polynomials often fail to capture the long-duration lag of thermal AM-PM. Effective behavioral models augment the polynomial with low-frequency envelope filtering terms. A common approach is to include a parallel branch that filters the squared envelope magnitude through a low-pass filter (representing the thermal impedance) and uses this filtered signal to modulate the phase of the basis waveforms. This thermal-aware memory polynomial structure explicitly separates short-term electrical memory from long-term thermal memory.

The gain compression or expansion counterpart to thermal AM-PM, where the device's temperature history dynamically modulates the amplitude-to-amplitude transfer characteristic. While AM-PM affects phase, thermal AM-AM alters the instantaneous gain profile. Both arise from the same self-heating mechanism and must be compensated simultaneously in wideband signals where envelope frequency heating falls within the thermal bandwidth. Key distinction: AM-AM is magnitude distortion; AM-PM is phase distortion.

Defines the dynamic relationship between power dissipation and junction temperature rise, measured in °C/W. Represented by Foster or Cauer thermal models as RC ladder networks, thermal impedance captures both the magnitude and time constants of the heat dissipation path. The thermal time constant—typically microseconds to milliseconds—dictates the memory duration of thermal AM-PM distortion. Accurate extraction via transient thermal response measurements is critical for building predictive electro-thermal models.

A co-simulation framework coupling semiconductor device physics with dynamic heat generation and dissipation equations. This approach solves the interdependent electrical and thermal equations simultaneously to predict temperature-dependent nonlinearities including AM-PM conversion. Finite element analysis resolves spatial temperature gradients across multi-finger devices, capturing thermal crosstalk effects that simple lumped-element models miss. Essential for GaN/GaAs MMIC design where self-heating dominates.

An augmented behavioral model extending standard memory polynomials with low-frequency thermal lag terms. While conventional memory polynomials capture electrical memory effects (nanosecond scale), thermal-induced variants add envelope-frequency terms to model the slow junction temperature dynamics. The model structure typically includes:

• Standard polynomial terms for instantaneous nonlinearity
• Electrical memory taps for short-term dispersion
• Thermal memory taps with longer time constants for AM-PM drift

A charge capture phenomenon in Gallium Nitride HEMTs where electrons become trapped in surface states or buffer layers, creating slow-memory distortion that is often thermally activated. Trapping interacts with self-heating in complex ways: increased temperature can accelerate detrapping rates while simultaneously shifting threshold voltage. This combined electro-thermal-trapping dynamic produces AM-PM distortion signatures that pure thermal models cannot fully capture, requiring specialized characterization.

Digital linearization techniques that incorporate real-time temperature sensing or electro-thermal model estimates into the predistorter coefficient calculation. Approaches include:

• Temperature-compensated LUTs indexing correction by amplitude and temperature state
• Model-based thermal de-embedding that subtracts predicted thermal phase shift
• Adaptive coefficient tracking with thermal time-constant-aware update rates These methods address the non-stationary nature of thermal AM-PM during burst-mode transmission.