Live status · Read the latest
monad-knowledge Wi-Fi sensing lab · FIIT STU
designed EXP-014

Prospective CSI temporal dependence and night alarm calibration

In plain words

Does accounting for temporal correlation reduce night-time CSI alarms while preserving response to known feature changes?

Why it matters
A drift trigger needs an interpretable baseline alarm rate before it can support recalibration decisions.
How it is done
Two unattended qualification nights, then twelve prospective nights after the analysis is frozen; synthetic feature perturbations provide responsiveness controls.
Where it stands
Planned only. CR-141 is open and CR-142 is held; no sessions have been claimed or run for this experiment.

Asks, formallyrecalibration-trigger-from-drift: on future nights, AR(1) innovations reduce the typical receiver CUSUM alarm rate relative to raw detrended CSI chain-ratio spread, while retaining sensitivity to preregistered synthetic feature shifts. Night alarms are not human false positives without independent occupancy truth.

StandingDesigned, not yet run

RoomFIIT library Pi 5 AX210 fleet, unattended night controls; no participant recruitment or booked site visit

Setup

Existing lib-empty-night arm, night-empty window, ch48 at 250Hz; expected monad01 transmitter and monad02–10 receivers. No participant recruitment or booked visit. Actual roster, profile/decoder hashes, capture health and storage must be checked before a later plan is approved.

Protocol and publication work: writing/signal-processing-submissions/Submission Plan. Requests: experiments/capture-requests/CR-141 lib-empty-night (two qualification nights) and experiments/capture-requests/CR-142 lib-empty-night (twelve validation nights, held).

Procedure

Each requested night is 5.5h, preserving warmup, a two-hour chronological calibration prefix, and at least two observed scoring hours. Primary feature: schema5 known-injector RX-chain-ratio amplitude spread. Compare raw detrended residuals and AR(1) innovations with fixed CUSUM k=0.5/h=5. Aggregate paired receiver contrasts within night; use night clusters for uncertainty. Full gates, exclusions, stopping rule, synthetic shifts and matched-alarm analysis are in the linked protocol.

Qualification sessions must be claimed as pilot with a reason, including correction of any automatic primary-role projection before analysis. Validation release requires passed instrument gates and a dated analysis freeze. Resolve archive ownership through planned run_id and verify transmitter as well as receivers. Do not claim historical schema5 pilot sessions to this prospective card.

Results

None. This card has no capture or simulation claims. The historical four-night OU / three-night whitening evidence motivates the protocol but is not prospective evidence for this card.

Limits

Clock-defined nights are not independently certified empty. Report night-time alarm rates unless external site evidence supports a narrower empty-reference claim. Synthetic feature changes are not people. Twelve nights are an operational sample cap, not a demonstrated power calculation; consecutive nights and simultaneous receivers are dependent. One-site results cannot establish multi-building performance.

Next steps

Compile/review qualification through the existing fleet workflow; no arming was performed when this card was created. After two qualification nights, pin implementation and evaluate coverage before releasing CR-142. Stop after twelve attempted validation nights without outcome-based extension. Approximate receiver archive budget: 158.4GB qualification + 950.4GB validation, plus transmitter/metadata overhead. Audit after claims and before moving status.

Provenance

—
not recorded