Near‑field to far‑field (NF‑FF) transformation is essential for large‑aperture antenna testing, yet millimeter‑wave phase acquisition is costly and offset mounting breaks the centering assumption. Existing approaches handle either the phase problem or the offset‑vector problem, requiring dense full‑field data or known offsets. The key observation is that amplitude fields measured under different offsets are merely coordinate‑transformed views of the same underlying near field. The goal is to recover a center‑aligned field from amplitude‑only data without any phase or offset information. To this end we propose AntennaFlow, a three‑stage framework:
- A contrastively trained encoder that maps offset views to an offset‑invariant embedding;
- A deterministic flow‑matching transport that converts offset amplitudes to centered amplitudes;
- The Simplified Extrapolation Technique (SET), whose Green‑function Taylor expansion is valid only for centered fields. Experiments demonstrate fast, phaseless, offset‑vector‑free NF‑FF reconstruction from sparse amplitude measurements, consistently outperforming prior baselines while preserving physical consistency.
Review: AntennaFlow cleverly combines contrastive representation learning with flow‑based transport, delivering practical phase‑free offset correction and opening new avenues for generative models in antenna metrology.