mmWave Beamforming

As the phased array steers the beam, the active impedance seen by each power amplifier changes dynamically due to mutual coupling between elements. This causes the PA's nonlinear characteristics—both AM-AM distortion and AM-PM conversion—to vary as a function of beam angle.

• A PA optimized for broadside may exhibit severe compression at 45° steering angles
• Impedance mismatch degrades power-added efficiency (PAE) and distorts the array pattern
• Traditional single-state DPD cannot compensate for beam-dependent nonlinearity

In dense mmWave arrays with half-wavelength spacing, antenna crosstalk creates parasitic signal paths between elements. A PA's output couples into adjacent elements, generating intermodulation products that combine in the far-field and cannot be corrected by per-element DPD alone.

• Mutual coupling strength increases with frequency and array density
• Crosstalk-induced distortion is beam-angle dependent
• Over-the-air DPD (OTA DPD) captures the combined array nonlinearity but requires far-field feedback

Power amplifier die temperatures vary significantly across a phased array due to non-uniform power dissipation and cooling constraints. Thermal memory effects cause slow variations in gain and phase that differ from element to element, creating a spatially distributed nonlinearity.

• Center elements typically run hotter than edge elements
• Gallium Nitride (GaN) devices exhibit strong trapping effects linked to temperature
• Per-element DPD must track individual thermal states for accurate linearization

mmWave systems operate with signal bandwidths of 400 MHz to 2 GHz for 5G NR. At these bandwidths, the PA's memory effects span many symbol periods, and the array's true-time-delay limitations cause beam squint—where different frequency components point in slightly different directions.

• Frequency-dependent beam patterns create spectral regrowth that varies spatially
• Generalized Memory Polynomial (GMP) models must capture both temporal and spatial memory
• Fractional delay filters are critical for aligning wideband feedback paths

Capturing the true far-field distortion for OTA DPD requires a dedicated observation receiver with a probe antenna in the far-field. This feedback path introduces its own nonlinearities, noise, and loop delay that must be precisely estimated and compensated.

• Far-field probe placement is impractical for deployed base stations
• Direct RF sampling with high-speed ADCs enables wideband capture but increases cost
• Loop delay estimation must achieve sub-sample accuracy for stable coefficient extraction

A 256-element array with per-element DPD requires 256 independent linearizers, each with potentially hundreds of coefficients. The coefficient extraction and real-time predistortion computation scale linearly with element count, challenging FPGA resources and power budgets.

• Coefficient interpolation reduces calibration time by deriving coefficients for uncalibrated states
• Real-valued time-delay neural networks (RVTDNN) offer compact nonlinear models
• Hardware-efficient architectures like look-up table (LUT) adaptation trade accuracy for speed

A multi-element antenna system where the relative phase of each element is electronically controlled to steer the composite beam without mechanical movement.

• Phase shifters at each element adjust the wavefront direction
• Enables sub-millisecond beam switching for user tracking
• Critical for overcoming the Friis free-space path loss at mmWave frequencies
• Typical arrays range from 64 to 256 dual-polarized elements for 5G NR

A partitioned precoding approach that splits beamforming between analog RF domain and digital baseband domain to balance performance with power consumption.

• Analog beamforming: Phase shifters in RF path provide coarse directional control
• Digital beamforming: Baseband processing enables multi-stream MIMO and fine precoding
• Reduces the number of RF chains compared to fully digital architectures
• Dominant architecture for massive MIMO mmWave systems due to ADC/DAC power constraints

The standardized 3GPP NR protocols for establishing and maintaining directional links between gNB and UE in 5G mmWave networks.

• P-1: Initial beam acquisition through wide sector sweep
• P-2: gNB beam refinement using narrower beams within a sector
• P-3: UE beam refinement for receiver-side optimization
• Relies on SSB bursts and CSI-RS reference signals for measurements
• Beam failure recovery triggers when all serving beams drop below threshold

Undesired secondary beams that appear at angles other than the intended steering direction when array element spacing exceeds half-wavelength.

• Caused by spatial aliasing when d > λ/2
• Divert transmitted energy away from the target, reducing effective isotropic radiated power (EIRP)
• Create interference in unintended spatial directions
• At 28 GHz, half-wavelength spacing is approximately 5.4 mm, demanding dense packaging
• Trade-off between thermal management and grating lobe suppression

Electromagnetic interaction between adjacent antenna elements that alters each element's impedance and radiation pattern as a function of scan angle.

• Modifies the active impedance seen by each power amplifier
• Causes scan-dependent gain variation and beam pattern distortion
• Complicates per-element digital predistortion due to varying nonlinear loads
• Modeled through S-parameter matrices and embedded element patterns
• Compensation requires real-time impedance-aware DPD adaptation

A finite set of predefined beamforming weight vectors standardized for NR mmWave to reduce channel state information feedback overhead.

• Type I codebook: Single-panel arrays with regular beam grids
• Type II codebook: Multi-panel arrays with amplitude/phase combining for higher resolution
• UE selects preferred precoding matrix indicator (PMI) from the codebook
• Balances beamforming gain against uplink control channel capacity
• DFT-based codebooks provide uniform spatial coverage