Thermal Memory Effect

The junction temperature of a power amplifier transistor fluctuates dynamically in response to the low-frequency envelope of the modulated signal. Unlike static DC heating, the instantaneous power dissipation varies with the signal amplitude, causing the temperature to track the envelope power within the device's thermal bandwidth. This creates a feedback loop where the signal shapes its own distortion environment.

Temperature shifts alter the transistor's threshold voltage, carrier mobility, and parasitic capacitances, producing two distinct distortion components:

• Thermal AM-AM: Gain compression or expansion that varies with temperature history, deviating from the instantaneous amplitude characteristic.
• Thermal AM-PM: Phase shift modulation caused by temperature-dependent junction capacitances, introducing memory-based phase errors.

Thermal memory effects operate on timescales dictated by thermal time constants, typically ranging from microseconds to milliseconds. This is orders of magnitude slower than electrical memory effects caused by bias network impedance or trapping. The thermal impulse response creates a convolutional relationship between past power dissipation and present junction temperature, making the distortion history-dependent over many symbol periods.

A hallmark of thermal memory is asymmetric spectral regrowth in the output spectrum. The dispersive phase response introduced by temperature-induced nonlinearities causes an imbalance between the upper and lower adjacent channel power. This asymmetry cannot be corrected by memoryless linearization techniques and requires memory-capable digital predistortion models that account for the thermal lag.

In Gallium Nitride (GaN) power amplifiers, thermal memory interacts with charge trapping phenomena. Electrons captured in surface states or buffer layers create slow-memory effects that are often thermally activated. The combined electro-thermal dynamics produce complex, coupled nonlinearities that require unified behavioral models capturing both self-heating and trap state dynamics for accurate compensation.

Compensating for thermal memory requires thermal-aware digital predistortion (DPD) architectures. These systems incorporate:

• Real-time temperature sensing or electro-thermal model-based estimation of junction temperature.
• Temperature-compensated look-up tables (LUTs) that index correction coefficients by both instantaneous amplitude and thermal state.
• Augmented memory polynomial models with low-frequency thermal terms to capture the long-duration lag.