Memory Effect

Caused by dynamic self-heating of the transistor channel. As input power varies, instantaneous power dissipation changes the junction temperature, which modulates electron mobility and threshold voltage. This creates a low-frequency pole in the kHz to MHz range, causing slow gain and phase variations that depend on the signal's recent power history. GaN HEMTs exhibit significant thermal trapping due to high power density.

Originates from the impedance of the DC bias network and envelope frequency decoupling. The varying drain current of the PA creates a voltage drop across the bias circuit impedance, modulating the instantaneous drain supply voltage. This is particularly severe in wideband signals where the video bandwidth of the bias network is insufficient. Key contributors include:

• Bias tee inductance causing low-frequency resonance
• Decoupling capacitor ESR limiting charge delivery
• Baseband impedance at the drain terminal

Occurs in compound semiconductors like GaN and GaAs where electrons are captured by deep-level traps in the buffer layer or surface states. These traps have time constants from nanoseconds to milliseconds. When trapped, charge depletes the channel, causing gate lag and drain lag—slow transient responses in drain current following changes in gate or drain voltage. This creates a history-dependent current collapse distinct from thermal effects.

Memory effects are categorized by their time constants relative to the signal envelope period:

• Short-term memory: Time constants comparable to the RF carrier period, caused by matching network reactance and intrinsic device capacitances. Typically handled by quasi-static models.
• Long-term memory: Time constants much longer than the envelope period, caused by thermal, bias, and trapping dynamics. These require explicit memory terms in Volterra or memory polynomial models. Long-term memory is the primary challenge for wideband DPD.

Memory effects break the assumption of a memoryless nonlinearity, making simple AM-AM/AM-PM look-up tables insufficient. The PA output becomes a function of current and past inputs, requiring DPD models with memory depth. Insufficient memory modeling leads to:

• Residual spectral regrowth at wide offset frequencies
• Degraded ACLR for wideband signals
• Poor EVM due to uncorrected dynamic distortion Generalized memory polynomials and Volterra series explicitly capture these effects.

Memory effects are quantified using two-tone envelope measurements and pulsed S-parameters. Key characterization techniques include:

• Envelope load-pull: Measures dynamic impedance at baseband frequencies
• Pulsed IV curves: Isolates thermal and trapping time constants by varying pulse width and duty cycle
• Two-tone IMD asymmetry: Asymmetry in upper and lower intermodulation sidebands directly indicates the presence and phase of memory effects
• Wideband modulated stimulus: Captures real-world dynamic behavior under operational signals

The number of past input samples considered in a behavioral model, determining the temporal span over which memory effects are captured. Memory depth directly impacts model accuracy and computational complexity.

• Short-term memory: Bias network impedance effects (nanoseconds to microseconds)
• Long-term memory: Thermal dynamics and trapping phenomena (milliseconds to seconds)
• Selecting optimal depth involves bias-variance tradeoff analysis

A long-term memory phenomenon caused by self-heating in the power amplifier's semiconductor junction. As the transistor dissipates power, its temperature changes, altering its gain and phase characteristics with a time constant determined by the thermal capacitance of the die and package.

• Dominant in GaN and GaAs power amplifiers
• Time constants range from microseconds to milliseconds
• Requires specialized compensation in DPD systems

The two fundamental nonlinear distortion mechanisms in power amplifiers. AM-AM distortion describes the nonlinear relationship between input amplitude and output amplitude (gain compression). AM-PM distortion describes the input amplitude-dependent phase shift, causing unwanted phase modulation.

• Both exhibit memory effects in wideband operation
• Complex baseband Volterra models capture both simultaneously
• Critical for Error Vector Magnitude (EVM) degradation

An enhanced memory polynomial model that includes cross-terms between the signal and its lagging envelope values. These cross-terms capture the interaction between nonlinearity and memory more accurately than the standard memory polynomial, especially for wideband signals.

• Adds envelope memory terms to the standard model
• Better captures complex memory effects in Doherty amplifiers
• Higher parameter count but improved accuracy for wideband signals

A primary physical cause of short-term memory effects. The finite impedance of the DC bias network at the modulation frequency creates a dynamic voltage envelope that modulates the amplifier's operating point based on recent signal history.

• Video bandwidth of the bias network is the key parameter
• Insufficient video bandwidth causes memory-induced distortion
• Proper bias network design is essential for minimizing memory effects