Description
CSI fingerprinting is a pattern-matching localisation and inference technique that uses the multipath profile in CSI as a location- (or class-) specific signature. A radio map is captured offline by labelling CSI windows with their ground-truth state; online queries are matched against this database with kNN, learned embeddings, or probabilistic models. It generalises classical RSSI fingerprinting to richer subcarrier-level features.
When it's used
- Indoor positioning when ranging fails (NLOS-heavy spaces)
- Discrete state classification (room, zone, occupancy bucket)
- Baseline that learning-free WiFi sensing benchmarks against
Limitations
- Database goes stale on environment change; requires re-survey
- Cross-device transferability is poor without explicit calibration
- Memory footprint scales with site size