Envelope Memory Effect

The primary electrical origin of the envelope memory effect. As the input signal envelope varies, the transistor's rectified current changes, causing voltage drops across the bias network impedance. This modulates the transistor's operating point dynamically.

• Baseband Impedance: The impedance of the bias feed network at the modulation frequency directly controls the magnitude of the memory effect
• Video Bandwidth (VBW): The frequency range over which the bias network can respond; insufficient VBW causes severe memory
• Decoupling Capacitors: Poorly designed capacitor networks create resonances that trap low-frequency energy, exacerbating the effect

Self-heating in the transistor channel creates a secondary, slower memory mechanism. Instantaneous power dissipation raises the junction temperature, which alters electron mobility and threshold voltage with a thermal time constant.

• Thermal Time Constants: Typically in the microsecond to millisecond range, affecting LTE and 5G NR slot-level dynamics
• GaN vs. GaAs: Gallium Nitride devices exhibit different thermal memory profiles compared to Gallium Arsenide due to superior thermal conductivity
• Pulse-to-Pulse Variation: In radar and TDD systems, thermal memory causes amplitude and phase drift across a burst

The signature observable manifestation of the envelope memory effect in laboratory measurements. When plotting output amplitude or phase shift against input amplitude, the trajectory no longer follows a single-valued curve but opens into a hysteresis loop.

• Loop Width: The separation between the ascending and descending traces quantifies the memory strength
• Frequency Dependence: The hysteresis loop widens as the modulation bandwidth approaches the inverse of the memory time constant
• Model Validation: Accurate behavioral models must reproduce this hysteresis to correctly predict adjacent channel leakage

The envelope memory effect is the fundamental bottleneck limiting the linearization bandwidth of digital predistortion systems. Memoryless DPD cannot compensate for dynamic distortion, leaving residual spectral regrowth.

• Memory Polynomial Depth: The required memory depth in a DPD model scales directly with the duration of the envelope memory effect
• Wideband 5G Challenge: For 100 MHz NR carriers, the memory effect spans multiple symbol periods, demanding high-order memory models
• Cross-Terms: Memory effects interact with static nonlinearity, generating complex cross-term distortion products that require generalized memory polynomial structures to cancel

Capturing the envelope memory effect requires coupled electro-thermal models that solve both the semiconductor transport equations and the heat diffusion equation simultaneously.

• Compact Models: Extensions like the MEXTRAM or Angelov models incorporate self-heating nodes for circuit simulation
• Behavioral Extraction: The memory effect can be extracted from pulsed S-parameter measurements at varying quiescent bias points
• Volterra Kernels: In the Volterra series framework, the memory effect manifests in the diagonal terms of higher-order kernels, representing the interaction of a signal with its delayed versions

Mitigating the envelope memory effect in the transmitter chain involves both circuit-level design and algorithmic correction.

• Bias Network Design: Minimizing baseband impedance across the modulation bandwidth using active biasing or wideband low-impedance terminations
• Envelope Tracking: Dynamically adjusting the drain supply voltage in phase with the envelope can linearize the transistor's intrinsic response, reducing memory
• Memory Polynomial DPD: The standard digital compensation technique, where the predistorter includes delayed taps of the signal envelope to cancel the amplifier's memory

The primary physical origin of the electrical memory effect. As the input signal envelope varies, the transistor's rectified current changes, causing voltage ripple on the DC bias lines due to finite impedance. This dynamically shifts the transistor's operating point, making the instantaneous gain a function of the past signal envelope. Key characteristics include:

• Dominant at low modulation frequencies (below ~10 MHz)
• Caused by insufficient decoupling capacitors or poor bias network design
• Manifests as asymmetric intermodulation distortion (IMD) sidebands

A slower dynamic nonlinearity caused by the junction temperature of the transistor changing with the dissipated power. As the envelope amplitude increases, power dissipation rises, heating the transistor die. This temperature change alters the electron mobility and threshold voltage, modulating the gain. Thermal effects are distinct from electrical effects due to their low-pass nature:

• Time constants typically in the microsecond to millisecond range
• Dominant for signals with wide instantaneous bandwidth but slow power variation
• Critical in GaN HEMT and high-power density technologies

A widely adopted behavioral model that captures the envelope memory effect by extending a static polynomial with a finite number of delay taps. The model expresses the output as a sum of nonlinear functions of the current and past input samples. Its structure is:

• y(n) = Σ_k Σ_q a_{kq} x(n-q) |x(n-q)|^{k-1}
• k is the nonlinearity order, q is the memory depth
• Effectively a Volterra series simplification that ignores cross-terms between different delays
• The standard baseline for digital predistortion (DPD) in modern wireless systems

A tell-tale signature of the envelope memory effect in two-tone tests. In a memoryless nonlinearity, the third-order intermodulation products (IM3) appear symmetrically in amplitude. When memory effects are present, the upper and lower IM3 sidebands exhibit amplitude and phase asymmetry. This asymmetry is a direct consequence of the bias network impedance varying with the envelope frequency (difference between the two tones). Key metrics:

• AM/AM and AM/PM conversions become frequency-dependent
• Asymmetry increases with tone spacing
• Used to quantify memory effect severity

An enhanced behavioral model that captures cross-terms between the signal and its lagging/leading envelope powers. Unlike the standard memory polynomial, the GMP includes terms like x(n) |x(n-q)|^2 and x(n-q) |x(n)|^2, which model the interaction between the instantaneous carrier and the delayed envelope. This structure is necessary for:

• Accurately modeling strong electrical memory effects in wideband signals
• Compensating for bias network impedance variations across the modulation bandwidth
• Providing a more robust basis for DPD coefficient extraction in GaN Doherty amplifiers

A critical system-level challenge where the envelope memory effect of the power amplifier interacts with the dynamics of the envelope tracking (ET) modulator. The ET power supply modulates the drain voltage to follow the signal envelope, but the supply modulator itself has bandwidth and memory limitations. This creates a dual-input memory system:

• The PA exhibits memory to the RF envelope
• The PA gain is also a nonlinear function of the shaped drain voltage
• DPD must linearize a 2D nonlinear dynamic system, often using a joint model of RF input and supply voltage