ET-Aware DPD Training

Unlike static-supply DPD, ET-aware training requires dual-input behavioral models that accept both the RF input signal and the instantaneous supply voltage as independent variables. Training datasets must span the full supply voltage dynamic range to characterize the PA's nonlinear behavior at every operating point along the shaping function trajectory.

The predistorter coefficients are extracted as a function of the instantaneous drain voltage, not just the input envelope. This requires 3D look-up tables (3D LUTs) or augmented Volterra models indexed by both input power and supply voltage. The extraction algorithm must solve for coefficients that invert the PA's supply-dependent gain compression and ET-induced AM/PM distortion simultaneously.

ET-aware training must account for supply modulator nonlinearity—including slew-rate limiting, switching ripple artifacts, and non-flat frequency response—that corrupts the intended supply voltage waveform. A joint ET-DPD model captures the compounded nonlinearities of both the PA and the modulator in a single unified structure, enabling a single predistorter to compensate for the entire transmitter chain.

The training process is critically sensitive to ET delay alignment—the precise time-synchronization between the RF signal path and the envelope tracking supply voltage path at the PA's transistor drain. Even sub-nanosecond mismatches introduce severe distortion that the DPD model must characterize and compensate for, requiring delay estimation algorithms integrated into the coefficient extraction loop.

ET-aware DPD training typically employs a closed-loop architecture using a feedback observation receiver to continuously monitor the transmitter output. This enables real-time coefficient updates that track changes in thermal memory effects, GaN trapping phenomena, and modulator behavior over temperature and aging, ensuring sustained linearity without periodic recalibration.

When the required supply voltage bandwidth exceeds the modulator's tracking capability, envelope-bandwidth mismatch causes clipping and residual distortion. ET-aware training incorporates this limitation by modeling the ET modulator slew rate constraint and extracting predistortion coefficients that are robust to the resulting waveform truncation, preventing spectral regrowth at the modulator's performance boundary.

A modeling framework that accepts both the RF input signal and the dynamic supply voltage as independent variables. Unlike single-input models that only see the RF envelope, this structure explicitly captures supply-dependent gain compression and ET-induced AM/PM distortion. Training data must span the full two-dimensional input space—varying instantaneous power and drain voltage simultaneously—to populate a valid 3D predistortion look-up table.

The precise time-synchronization of the RF signal path and the envelope tracking supply voltage path at the power amplifier's transistor drain. A misalignment of even a few nanoseconds causes the PA to operate at the wrong supply voltage for a given signal sample, generating severe distortion that no DPD model can correct. ET-aware training must either assume perfect alignment or incorporate delay estimation as a preprocessing step.

A deterministic mapping—often implemented as a look-up table—that translates the instantaneous baseband signal magnitude into a target supply voltage. The shaping function defines the trajectory through the PA's iso-gain contours. ET-aware DPD training data must be collected while the PA operates under the exact same shaping function that will be used in deployment, as different mappings produce different nonlinearity profiles.

An extension of the classical Volterra series that incorporates dynamic supply voltage terms into the basis function set. Standard memory polynomial models fail under envelope tracking because they cannot distinguish between memory effects caused by RF history and those caused by supply voltage history. The augmented formulation adds cross-terms that capture the nonlinear interaction between the signal envelope and the time-varying drain voltage.

Distortion introduced by the supply modulator itself, including:

• Slew-rate limiting when the envelope rises faster than the modulator can track
• Switching ripple artifacts that intermodulate with the RF carrier
• Non-flat frequency response that filters the intended voltage waveform

ET-aware DPD must treat the PA and modulator as a single cascaded nonlinear system, training on the combined response rather than assuming an ideal supply.

An adaptive system where an observation receiver continuously monitors the transmitter output and feeds back the error signal to update predistortion coefficients in real-time. This is critical for ET-aware DPD because the combined PA-modulator system drifts with temperature, aging, and channel frequency. Closed-loop training captures these variations dynamically, maintaining linearity without requiring periodic offline recalibration.