ET-DPD for Handset PAs

Handset DPD must operate within an extremely tight computational envelope, typically < 50 MOPS (Million Operations Per Second) to avoid draining the battery. This forces the use of low-order memory polynomial models (e.g., MP with M=3, K=5) rather than full Volterra series. Key constraints:

• DSP cycles shared with protocol stack and application processor
• Fixed-point arithmetic preferred over floating-point to reduce power
• Look-up table (LUT)-based predistortion often replaces real-time polynomial evaluation
• Coefficient estimation runs infrequently (seconds to minutes) rather than continuously

The envelope tracking supply modulator in a handset cannot track the full RF envelope bandwidth. For a 100 MHz 5G NR carrier, the envelope bandwidth can exceed 150-200 MHz, while handset modulators typically achieve only 40-80 MHz. This envelope-bandwidth mismatch creates:

• Residual distortion from untracked high-frequency envelope components
• The need for shaping function optimization to reduce tracking bandwidth requirements
• DPD models that must compensate for modulator-induced clipping and slew-rate limiting
• A fundamental trade-off between efficiency improvement and linearity degradation

Handset DPD coefficient tables must fit within tight SRAM constraints, typically < 16 KB for the entire predistorter. This limits:

• 3D LUT dimensions: Quantization of input power and supply voltage axes must be coarse (e.g., 32×16 entries)
• Memory polynomial depth: Taps limited to 2-3 to keep coefficient count manageable
• Multi-band support: Separate coefficient sets for each band-carrier combination consume additional memory
• Compression techniques like piecewise-linear interpolation between LUT entries reduce storage at the cost of interpolation logic

The observation receiver used for DPD training in handsets is severely cost-constrained compared to base station equivalents. Limitations include:

• Reduced dynamic range: Typically 8-10 ENOB (Effective Number of Bits) vs. 12+ in infrastructure
• Coupler directivity: Limited to 15-20 dB, introducing forward signal leakage into the feedback path
• IQ imbalance: Uncorrected gain/phase mismatch in the feedback demodulator corrupts coefficient estimation
• These impairments force the use of robust estimation algorithms (e.g., least squares with regularization) that tolerate noisy training data

Handset PAs experience rapid temperature changes (from idle to full-power transmission) and long-term aging effects that alter their nonlinear characteristics. DPD must adapt without continuous training:

• Thermal memory: Self-heating during transmission changes gain and phase response within milliseconds
• Aging drift: Parameter shifts over months/years require periodic recalibration
• Solutions include temperature-indexed LUT banks that store coefficient sets for different thermal states
• Background training during idle slots or low-traffic periods minimizes disruption to active transmission

Unlike infrastructure where ET and DPD can be designed independently, handset implementation demands tight co-design to meet system-level targets:

• Shaping function and DPD LUT must be jointly optimized to balance efficiency and linearity
• Delay alignment between RF and envelope paths must be calibrated to within < 1 ns to prevent catastrophic distortion
• Supply modulator noise (switching ripple at 10-100 MHz) couples into the PA output and must be included in the DPD behavioral model
• Single-chip integration of ET modulator, DPD engine, and CFR on the same RFIC reduces PCB parasitics but constrains individual block optimization

A dynamic power supply technique that modulates the PA's drain voltage in real-time to track the instantaneous RF envelope amplitude. Efficiency gains of 20-30 percentage points are achievable compared to fixed-supply operation.

• Replaces static battery voltage with a waveform-following supply
• Requires supply modulator bandwidth of 1.5-2x the RF signal bandwidth
• Introduces supply-dependent nonlinearities that DPD must correct
• Primary enabler for extending smartphone battery life in 5G NR handsets

The precise time-synchronization of the RF signal path and the envelope tracking supply voltage path at the PA transistor drain. Sub-nanosecond alignment accuracy is required to prevent severe distortion.

• Mismatch causes spectral regrowth and EVM degradation
• Typical alignment tolerance: < 1 ns for 100 MHz signals
• Implemented via programmable delay lines or digital interpolation filters
• A primary source of residual distortion in practical ET systems

A PA modeling framework that accepts both the RF input signal and the dynamic supply voltage as independent variables. This structure accurately predicts the nonlinear behavior of an ET PA for DPD coefficient extraction.

• Captures supply-dependent gain compression and phase distortion
• Enables a single predistorter to compensate the entire ET-PA chain
• Often implemented using augmented Volterra series or neural networks
• Training data must span the full dynamic range of supply voltages

The maximum rate of change of the supply modulator's output voltage, measured in V/µs. The modulator must track the fast-rising envelope of wideband signals without introducing tracking errors.

• Insufficient slew rate causes clipping and residual distortion
• 5G NR 100 MHz signals demand slew rates exceeding 100 V/µs
• A fundamental bottleneck in handset ET-DPD performance
• Drives the need for crest factor reduction co-optimization

A memoryless predistortion structure indexed by instantaneous input power and supply voltage to apply a complex gain correction. Compensates for the static nonlinearities of an ET PA with minimal computational overhead.

• Suitable for handset implementations with tight power constraints
• Does not address memory effects from thermal or trapping phenomena
• Often combined with a memory polynomial for complete linearization
• Table entries updated via closed-loop adaptation or offline calibration