ET-DPD Co-Design

ET-DPD co-design addresses the compounded distortion arising from the interaction of the power amplifier's native nonlinearity and the supply modulator's dynamic behavior. Rather than treating these as separate problems, a joint behavioral model captures the cross-dependencies between RF input, supply voltage, and output distortion.

• Compensates for ET-induced AM/PM distortion caused by supply-dependent phase shifts
• Corrects supply modulator nonlinearity including slew-rate limiting and switching ripple artifacts
• Accounts for supply-dependent gain compression across the full dynamic voltage range
• Eliminates the need for separate PA linearizer and ET compensator blocks

The foundation of ET-DPD co-design is the dual-input behavioral model, which accepts both the baseband RF signal and the instantaneous supply voltage as independent variables. This enables the predistorter to predict the PA's output as a function of both excitation dimensions.

• Augmented Volterra series extend classical memory polynomial models with supply voltage terms
• 3D Look-Up Tables (3D LUTs) index correction coefficients by both input power and supply voltage
• Models capture cross-term memory effects where past supply voltages influence current RF behavior
• Enables accurate prediction of output under arbitrary shaping function trajectories

In ET-DPD co-design, the shaping function that maps signal envelope to supply voltage is not designed in isolation. It is jointly optimized with the DPD coefficients to find the Pareto-optimal trade-off between efficiency and linearizability.

• Iso-gain contour analysis identifies supply voltage trajectories that maintain constant gain
• Shaping functions are designed to avoid regions of high AM/PM sensitivity that are difficult to linearize
• Crest factor reduction (CFR) is co-optimized to prevent supply modulator overdrive from extreme peaks
• The shaping function becomes a tunable parameter in the overall linearization optimization loop

Conventional DPD training assumes a fixed supply voltage. ET-aware training excites the PA across its full dynamic supply range during coefficient extraction, ensuring the predistorter generalizes to all tracking conditions.

• Training signals must span the entire supply voltage range and envelope bandwidth
• Closed-loop adaptation continuously updates coefficients to track thermal and aging effects
• Training data captures envelope-bandwidth mismatch scenarios where the modulator cannot perfectly track
• Coefficients are validated across multiple shaping functions to ensure robustness to ET parameter changes

A critical aspect of ET-DPD co-design is the precise time-alignment of the RF signal path and the envelope tracking supply path at the transistor drain. Misalignment by even fractions of a nanosecond causes severe distortion that no predistorter can correct.

• ET delay alignment must be maintained within tens of picoseconds for wideband signals
• Co-design incorporates digital delay compensation in the baseband processing chain
• Joint optimization accounts for group delay variations in both the RF and supply modulator paths
• Adaptive delay tracking compensates for temperature-dependent propagation changes

ET-DPD co-design optimizes for total system power added efficiency (PAE), not just the PA's standalone efficiency. This holistic metric accounts for the power consumed by the supply modulator, the DPD processing, and the feedback receiver.

• System PAE = (RF output power - RF input power) / (PA DC power + modulator DC power + DPD compute power)
• Co-design identifies the ET efficiency knee where further supply modulation yields diminishing returns
• Trade-offs between linearization complexity (gate count, power) and achievable efficiency are explicitly modeled
• Optimization targets include 5G NR spectral mask compliance and error vector magnitude (EVM) requirements simultaneously

The high-bandwidth power converter that generates the dynamically varying supply voltage tracking the RF envelope. In co-design, its slew rate, output impedance, and switching ripple are not isolated specs but direct inputs to the DPD linearization model. A modulator with insufficient bandwidth creates envelope clipping that no DPD can fully correct, making modulator-DPD co-specification essential.

A deterministic mapping, often implemented as a look-up table (LUT), that translates instantaneous baseband signal magnitude into a target supply voltage. Co-design optimizes this function jointly with DPD coefficients rather than in isolation. The shaping function defines the iso-gain contour operating trajectory, and its derivative directly impacts the ET-induced AM/PM distortion that the predistorter must invert.

The precise time-synchronization of the RF signal path and the envelope tracking supply voltage path at the PA transistor drain. A misalignment of even hundreds of picoseconds creates severe nonlinear distortion that appears as memory effects to the DPD. Co-design frameworks incorporate delay estimation as a joint parameter, often using cross-correlation of the AM/AM and AM/PM profiles during training.

A PA modeling framework that accepts both the RF input signal and the dynamic supply voltage as independent variables. Unlike single-input models, this captures the compounded nonlinearity of the ET-PA system. Common structures include augmented Volterra series and dual-input memory polynomials, which serve as the forward model for training the inverse predistorter in co-design.

Distortion introduced by the supply modulator itself—clipping, slew-rate limiting, and non-flat frequency response—that corrupts the intended supply voltage waveform. In co-design, the modulator is not assumed ideal; its nonlinear transfer function is embedded into the ET-DPD joint model. This prevents the DPD from attempting to correct artifacts that originate in the power supply path.

A coefficient extraction process where the DPD is trained using data that captures the PA's behavior across its full dynamic range of supply voltages. This requires test signals that exercise both RF envelope and supply voltage variations simultaneously. The resulting predistorter generalizes across all tracking conditions, preventing supply-dependent gain compression from degrading EVM at low power levels.