MIMO OTA Done Right: Multi-Probe Calibration, Channel Models & Throughput Correlation

Why MIMO OTA is hard (and why it matters)

Conducted tests don’t capture spatial correlation, antenna efficiency, user orientation, or device desense. Over-the-air (OTA) MIMO testing does—but only if your field synthesis, calibration, and channel modelling are solid. If they’re not, you’ll ship devices that look great on a cable and underperform in the field.

This post distils a practical playbook for MPAC (multi-probe anechoic chamber) MIMO testing that correlates with real-world throughput, BLER and user experience.

Choose the right OTA architecture

• MPAC (multi-probe): Surround the DUT with an array of probes driven by weighted signals to synthesise target channels. Best for throughput and mobility scenarios with short test times.
• CATR: Excellent for plane-wave conformity and beam/pattern diagnostics. Pair with a channel emulator for angular spreads.
• Spherical near-field (SNF): Highest accuracy for radiation patterns; slower for throughput sweeps.

If your KPI is throughput vs channel profile/MCS, MPAC is usually the most efficient.

Calibration: the foundation of believable results

1. Geometry & Origin

• Establish the chamber coordinate system; laser-align the DUT phase centre and turntable axis.
• Verify probe radius and angular placement; record as calibration data.

2. Path/Power Normalisation

• Measure each probe→reference-antenna path.
• Compute per-probe amplitude weights so that equal digital drive yields equal field at the DUT origin.

3. Phase Equalization

• Extract per-probe complex error (cable + probe + instrument).
• Apply phase offsets to flatten the phase at the origin; re-verify after warm-up.

4. Timing Alignment (for wideband/OFDM)

• Time-align paths so target delay spreads are created by the emulator—not by unequal cable lengths.

5. Polarization Purity

• Characterise co/x-pol of each probe. Use polarisation grids/rotators; store per-probe pol matrices.

6. Health & Drift

• Implement a daily quick-check: one probe at a time into the ref antenna to catch gain/phase drifts.
• Log temperature; many “mystery” drifts are thermal.

Deliverables to keep: geometry report, path/phase tables, timing skew, pol matrices, uncertainty budget.

Channel models that matter

Map your use cases to 3GPP-style channels (indoor hotspot, factory floor, urban micro/macro). Typical NR-era families:

• TDL-x (Tapped Delay Line): Fast, simple delay/power taps—good for regression and production.
• CDL-x (Clustered Delay Line): Adds angles (AoA/AoD), spreads, K-factor, and polarisation—better for MIMO rank/correlation studies.

Key knobs you must match in MPAC synthesis:

• Angular spreads (azimuth/elevation of arrival)
• Cluster powers & delays (delay spread)
• Polarisation matrix (XPR, depolarisation)
• Doppler (if you vary turntable speed or emulate mobility)
• Correlation targets (Rx correlation matrix, effective rank)

Ask vendors for channel-synthesis certificates: measured angular spectrum vs target, and correlation matrices at the DUT origin.

Test flow that correlates with the field

1. Define the network context
   Carrier BW/SCS, MIMO (2×2/4×4), rank control, scheduler, reference signals, and traffic model (full buffer vs app-like).

2. Pick anchor profiles
   A small set (e.g., Indoor Office, Factory, UMi NLOS) at two SINR points each gives wide coverage without exploding test time.

3. Run two classes of KPIs

• Radiated power/sensitivity: TRP/TIS/EIRP/EIS in the same setup; they explain SNR headroom.
• User KPIs: Throughput & BLER across MCS ladder, rank adaptation, precoding (PMI) distributions.

4. Create the mapping
   For each anchor profile:

• Compute effective SNR at the receiver (post-OTA, post-rank) from measured EIS/TRS and emulator SINR.
• Build a throughput vs effective-SNR curve (one per profile).
• Validate prediction error across the rest of your profiles. You’re aiming for tight error bands (e.g., within single-digit % across most of the MCS range).

5. Lock correlation
   Freeze the calibration, seeds, profiles, and DUT firmware as a golden setup. Use it to police future changes.

Practical tips that save weeks

• Golden DUT + golden gNB: Keep a stable reference device and baseband build. Every change can shift your correlation.
• Raster-reduction: For TRP/TIS patterning, use optimised sampling (e.g., Clenshaw–Curtis) to cut time without bias.
• Orientation states: Test multiple hand/head/placement states for UE; for CPE/routers, include wall/ceiling/desk mounts.
• Desense/coexistence sweeps: Add controlled Wi-Fi/UWB/Bluetooth interferers; many “throughput” problems are coexistence.
• Thermal plateaus: Run long tests only after the DUT settles thermally; log case temp.

Uncertainty & repeatability (what auditors will ask)

• Contributors: probe gain/phase repeatability, positioner accuracy, ref antenna factor, instrumentation linearity/noise, chamber scattering, temperature.
• Method: Monte-Carlo or GUM-style combination to produce expanded uncertainty (k=2).
• Targets: Publish uncertainty with your KPIs; require the same from suppliers and contract labs.

Cross-site correlation improves dramatically when seeds, profiles, and calibration artefacts are shared—and when both sites run the same golden DUT.

Common failure signatures (and fixes)

• Great TRP, poor throughput: Antenna is efficient but high correlation or desense; check CDL synthesis and coexistence.
• Setup-specific rank collapse: Mis-timed paths or wrong Doppler; re-align timing, check emulator settings.
• Day-to-day drift: Temperature or instrument reference; extend warm-up, lock ref clock, automate daily quick-check.
• Unstable BLER at high MCS: Chamber stray reflections raising fading variance; treat seams/pedestals, verify absorber performance at your bands.

Procurement checklist (for labs scaling MIMO OTA)

• MPAC capability: probe count/geometry, supported CDL/TDL profiles, polarisation control.
• Calibration package: included artefacts, procedures, uncertainty statements, and FAT/SAT acceptance masks.
• Throughput stack: gNB emulator, traffic generator, logging APIs, automation hooks.
• Lifecycle: recalibration tools, drift monitoring, service SLAs, spare probes/cables.
• Upgrade path: FR2 support, channel-emulator capacity, larger QZ or dual-pol feeds, and coexistence injectors.

MIMO OTA that correlates is a system: calibrated probes + defensible channel synthesis + stable network context + disciplined uncertainty. Get those right and your chamber becomes a predictive instrument for real-world user experience—not just a pretty room with absorbers.