Antenna Crosstalk

The fundamental physical origin of crosstalk is mutual coupling, where surface currents on a driven element radiate near-fields that excite neighboring elements. This creates a reactive electromagnetic field concentrated in the near-field region, with coupling strength inversely proportional to element spacing. In tightly-packed mmWave arrays where spacing approaches λ/2, mutual coupling can exceed -10 dB, meaning 10% of transmitted energy couples directly into adjacent channels. The effect is reciprocal—coupling from element A to B equals coupling from B to A—and frequency-dependent, intensifying at band edges where impedance matching degrades.

Crosstalk fundamentally alters the load impedance seen by each power amplifier as a function of beam-steering angle. As the array scans, the phase relationship between elements shifts, changing the constructive or destructive interference of coupled signals at each PA output. This active impedance mismatch causes the PA to operate at a different point on its Smith chart than calibrated, directly modulating its AM-AM and AM-PM characteristics. The result is a beam-dependent nonlinear distortion profile where the optimal DPD coefficients for broadside operation fail at scan angles beyond ±30°.

Unlike single-antenna nonlinearity, crosstalk-induced distortion exhibits spatial selectivity—the distortion products radiate in different directions than the main beam. The coupled signal components combine with the intended signal at each element with phase shifts determined by the array geometry, creating grating lobes of distortion in the far-field. This means adjacent channel leakage ratio (ACLR) measured at boresight may pass compliance while off-axis ACLR violates regulatory masks. Effective linearization must therefore consider over-the-air (OTA) far-field metrics, not just conducted measurements at individual PA outputs.

Crosstalk breaks the assumption of identical PA behavior across an array. Each element experiences a unique combination of:

• Self-distortion from its own PA nonlinearity
• Cross-distortion from coupled signals of adjacent elements
• Beam-dependent loading that shifts the PA operating point

A single shared DPD lookup table or coefficient set becomes suboptimal. True crosstalk-aware linearization requires either per-element DPD with individual feedback paths or a MIMO-DPD architecture that models the coupled nonlinear system as a multi-input multi-output function, dramatically increasing coefficient count and estimation complexity.

Crosstalk corrupts the DPD training signal itself. When a single element transmits a training sequence, the observation receiver on that element captures not only the intended PA output but also coupled contributions from all other active elements. This contaminated feedback causes the coefficient extraction algorithm to converge to a biased solution that attempts to linearize the composite coupled signal rather than the individual PA. Mitigation strategies include time-multiplexed training where only one element transmits at a time, or de-embedding algorithms that mathematically subtract estimated coupling contributions from the observed waveform.

Crosstalk coupling strength is frequency-selective, typically increasing at higher frequencies within the operating band due to electrical spacing changes. For wideband 5G signals occupying 100 MHz or more, this creates frequency-dependent crosstalk where the distortion profile varies across subcarriers. Additionally, as instantaneous bandwidth increases, the memory effects of crosstalk become more pronounced—the coupled signal arrives with a small but non-zero time delay, creating a dispersive interference pattern that requires memory-polynomial or neural network DPD structures with sufficient temporal depth to capture.