Paired-band occupancy sensitivity on the AX210 fleet

Current assessment (2026-09-15)

The paired occupied-band comparison remains planned, with no session listed for c-csi-band-paired-hardware. EXP-011 supplies a later empty-room delivery contrast under monitor injection, not a measured occupancy advantage. The original hostapd and office-bench readiness statements are historical.

The protocol needs a feasible transmitter/receiver allocation for each band. Two nodes alone do not establish simultaneous paired illumination and reception. The earlier matched-width plan must also match source identity, usable record class and delivered timing. A verified-empty baseline cannot remove every occupied-traffic confound.

Next evidence needed: Use the current acquisition method from EXP-011, and independent labels validated through EXP-F1. Freeze the band assignment and swap order before acquisition. Keep the public-data result, empty-room delivery result and occupied-band test as separate estimands.

Basis: Thesis Narrative, the September 10 synthesis, and the September 15 campaign index. See the review record.

Earlier protocol and dated results

The plan below records the earlier hostapd/office-bench design. It is not a current fleet runbook; acquisition changes need a dated protocol amendment before execution.

Why this card exists

band-5ghz-occupancy-discriminability is currently strength: supported on one real dataset — our own zero-shot re-analysis of WiMANS raw CSI (Intel 5300): ρ(moving-count, CV) +0.71 at 5 GHz vs +0.28 at 2.4 GHz, presence AUC 0.991 vs 0.879. The ray-traced programme (c-csi-band-calibration) reported the two bands tracking identically, which the hypothesis note attributes to the localised ray-tracer under-resolving 5 GHz fading.

So the claim rests on a single hardware platform, and the sim disagrees with it. The AX210/Pi5 fleet is the second real platform — and, unlike a public dataset, it lets us control the three things a band comparison is usually confounded by: the subcarrier basis, the illumination rate, and the receiver.

What makes this measurable rather than merely another capture

Two nodes on the same floor mean both bands can be observed simultaneously, on the same realisation of the room — not sequentially, as every public dual-band set does. Paired observation removes the between-session channel drift that otherwise sits inside the band contrast.

The design's load-bearing controls are recorded on the campaign brief (c-csi-band-paired-hardware). The one worth stating here too: the campaign does not rely on ambient traffic. The illuminating node runs hostapd on its AX210 — never on wlan0, which carries eduroam and the tailnet and has no cabled recovery — and emits a paced legacy-rate multicast stream on a channel we choose. Because legacy OFDM is 52 tones at 312.5 kHz on both bands, that gives subcarrier parity natively rather than by offline resampling, and it removes the dependence on a band-steered faculty AP whose channel we cannot even read remotely. The second control is the node→band swap, without which the contrast is "monad02 minus monad01".

Geometry caveat

gis_experiment points at ax210-fiit-deployment, whose Rx set is the planned Pareto-knee 9-Rx layout on fiit-library-lidar-floor-0. The two nodes carrying Stage 1 are currently on an office bench, not at those planned positions. Register the actual monad01/monad02 placements (gis_place_device) before any result is quoted with a geometry — until then the campaign's spatial claim is "two receivers, one room, positions fixed across the crossover", which is all Stage 1 needs and all it may assert.

Status

Planned, not executed. Stage 1 (the verified-empty-room arm) is runnable remotely and is explicitly a pilot to estimate the variance component, not a powered test — no AX210 empty-room CV dispersion estimate exists yet in the vault. Stage 2 (the occupancy ladder) requires people in the room and is gated on Stage 1's sigma.

Key references

• band-5ghz-occupancy-discriminability — the hypothesis this card is the standing hardware gate for.
• occupancy-csi-variance — the parent claim the band rule refines.
• huang2025_060d ↗ — the single real dual-band anchor to be replicated.
• li2021_1875 ↗ — idle-AP beacon duty cycle at 0.42% of a transmitting AP; why an empty room starves passive capture.
• wang2025_2c42 ↗ — the auxiliary-Pi illumination pattern this campaign adopts.
• sharma2024_c8a2 ↗ — why ambient-traffic composition, not just rate, perturbs the sensing feature.