Description
Subcarrier features are statistical or spectral descriptors computed independently per OFDM subcarrier from a CSI window — most commonly mean, variance, skewness, kurtosis, entropy, and dominant Doppler bin. Treating each subcarrier as its own channel exposes frequency-selective fading patterns that a single scalar like RSSI averages out, and gives ML pipelines a fixed-shape feature tensor (subcarriers × time) that maps cleanly to CNNs and transformers.
When it's used
- Hand-crafted feature pipelines feeding random-forest / SVM / kNN classifiers
- CNN input planes for HAR and crowd counting
- Sensitivity analyses that pick the most discriminative subcarriers
Limitations
- High-dimensional and correlated; usually needs PCA or subcarrier selection
- Vulnerable to per-subcarrier hardware artefacts on Intel 5300
- Loses cross-subcarrier phase coherence that frequency-domain models exploit