Seeing Through Wi-Fi.

1. The sensing pipeline

The system is organized as a deterministic signal chain rather than a black-box classifier:

1. Capture: acquire CSI amplitude and phase measurements from supported Wi-Fi hardware and preserve timestamps, channel metadata and antenna topology.
2. Calibration: remove hardware-specific phase offsets, unwrap phase where appropriate, reject corrupt samples and normalize across subcarriers.
3. Pre-processing: suppress static multipath, impulsive noise and motion artifacts using bounded filters and adaptive baselines.
4. Feature space: derive temporal, spectral and spatial features from subcarrier dynamics and antenna relationships.
5. Inference: estimate presence, movement state and coarse pose using models whose confidence and provenance are recorded with every material output.
6. Validation: compare predictions against controlled ground truth and expose uncertainty rather than silently converting weak evidence into certainty.

2. Why CSI instead of cameras?

CSI does not produce an optical image of the room. That can materially reduce the privacy surface for environments where cameras are inappropriate. It does not, however, make sensing automatically private: RF observations can still reveal behavior and occupancy patterns. A defensible implementation therefore needs explicit retention limits, local processing, access control, audit trails and user-visible system state.

3. Presence, movement and pose

Presence detection is the lowest-risk inference layer. Movement classification can then use temporal energy, Doppler-like signatures and cross-antenna coherence. Pose estimation is substantially harder: it depends on hardware geometry, room topology, multipath, training distribution and calibration stability. AetherSense should therefore expose pose as a confidence-bounded estimate rather than a ground-truth reconstruction.

4. Respiration and heart-rate trends

Fine-grained chest motion can modulate RF propagation at low frequencies. In a controlled environment, those periodic components can be isolated and tracked over time. Heart-rate estimation is especially sensitive to motion, multipath and hardware limitations. The correct engineering posture is experimental trend analysis with validation against independent reference instrumentation—not a diagnostic or clinical claim.

Safety boundary: AetherSense outputs are research telemetry. They must not be presented as medical diagnosis, emergency detection, identity determination or biometric truth without validated evidence, appropriate governance and a clearly defined deployment context.

5. DARPA-grade engineering means measurable controls

• Reproducibility: version every capture configuration, calibration procedure, feature pipeline and model artifact.
• Provenance: bind each inference to source measurements, timestamps, model version and confidence.
• Fail-closed operation: malformed CSI, stale calibration or insufficient signal quality must produce an explicit unavailable/low-confidence state.
• Adversarial evaluation: test false positives caused by fans, pets, HVAC, reflections, channel changes and unrelated RF activity.
• Local-first architecture: default processing stays on the sensing host; network export is explicit, authenticated and auditable.
• Reversible deployment: isolate experimental models from production inference and maintain deterministic rollback paths.

6. AetherSense architecture

The reference platform separates capture, signal processing, inference and presentation:

Wi-Fi CSI source
      ↓
Capture + timestamp integrity
      ↓
Calibration / denoising / quality gate
      ↓
Feature extraction + temporal windows
      ↓
Presence → movement → pose → experimental physiology
      ↓
Confidence + provenance + audit event
      ↓
Local operator console / bounded API

Rust is a strong fit for the low-level capture, streaming and safety-critical boundaries; Python is useful for research, feature engineering, model evaluation and scientific benchmarking. The interface should make signal quality and uncertainty visible instead of hiding them behind a polished visualization.

7. The research agenda

The next engineering milestones are controlled datasets, hardware-specific calibration, reproducible benchmarks, multi-person separation experiments, robustness testing across room geometries, and independent validation of physiological trend extraction. The platform should publish confidence intervals and failure modes alongside headline performance.

Conclusion

AetherSense is an experimental sensing architecture built around Wi-Fi CSI, edge inference and governance. The core technical challenge is not merely classification accuracy. It is constructing a measurement system whose outputs remain reproducible, privacy-conscious, uncertainty-aware and auditable under real-world interference.