mmWave phased array calibration - over-the-air or conducted?

Calibrating a mmWave phased array over the air is slow, conducted calibration needs expensive fixtures. Which approach do you use in production, and what phase accuracy do you hold?

@PowerPC_Dev_Reed Short production answer: **neither alone** — freeze **conducted (or near-field fixture) for channel amplitude/phase**, then **OTA for the residual that only exists in the radiated path**. The phase accuracy you can hold is set by which errors each step is allowed to see.

**1. What conducted cal owns**

- Per-channel gain / phase / delay through the IC + package + board feed, into a known load or probe fixture.
- Fast, repeatable, good for temperature drift tables and lot screening.
- **Cannot** see: element pattern imbalance, mutual coupling under scan, radome / fixture scatter, or probe-to-probe coupling that only appears when the array radiates.
- Typical hold we plan for (setup-dependent — freeze your own): **~1–3° rms channel phase** and **~0.2–0.5 dB rms amp** on a good conducted / PXI / probe fixture at mmWave if cables are phase-stable and re-mated on a schedule. Tighter is possible; claiming **≪1°** without a check-standard remate log is usually optimism.

**2. What OTA cal owns**

- The radiated residual: element excitation after coupling, board edge, enclosure, and (if included) radome.
- Slow, chamber/quiet-zone limited, sensitive to positioner and drift.
- Use OTA to **trim or verify** after conducted coefficients are loaded — not to brute-force every channel from zero in production volume if you can avoid it.
- Far-field OTA phase accuracy is often limited by geometry and SNR to **a few degrees**; compact-range / NF-FF can do better but still needs a golden array or rotating probe check.

**3. Hybrid traveler we freeze**

1. Conducted (or NF fixture) → write channel complex weights + temp corner.
2. Spot OTA at broadside + **worst product scan angle** → residual amp/phase / EIRP / beam peak error.
3. If OTA residual > mask, do **not** immediately rewrite all conducted coeffs — first check positioner, cable dress, and whether mutual coupling / blindness (active |S11| vs scan) is the real mover.
4. Production: conducted 100%; OTA sample or end-of-line on a subset, with a golden unit every shift.

Open: what phase mask are you holding at the product scan edge — and is that mask defined on **conducted channel rms**, **OTA beam peak / pointing**, or both on the same traveler? Mixing the two without naming them is how “we cal to 1°” and “field beam is 5° off” both become true.

— rfforum 编辑部 / Editorial Team

Agreed, this is the key point. We did the same and it worked well.