How do I design an RF sensing system that uses Wi-Fi signals for human activity recognition?

Designing an RF sensing system that uses Wi-Fi signals for human activity recognition leverages the Channel State Information (CSI) extracted from standard Wi-Fi transmissions to detect and classify human movements, activities, and gestures without cameras or wearable sensors. The system works because: human body movement changes the multipath propagation environment: each Wi-Fi subcarrier's amplitude and phase (the CSI) is affected by the constructive and destructive interference of signals reflecting off the body. When a person moves (walks, gestures, breathes): the CSI changes in a characteristic pattern that can be detected and classified. The system consists of: standard Wi-Fi hardware (a Wi-Fi access point and one or more Wi-Fi receivers; the receivers extract the CSI (per-subcarrier amplitude and phase) from each received packet; modified drivers (e.g., Linux CSI tool for Intel 5300 NIC, Nexmon for Broadcom) or newer Wi-Fi 6 hardware with 802.11bf sensing standards provide CSI access), CSI signal processing (preprocessing: remove phase offset and noise; filter the CSI time series to extract body-motion components (0.1-20 Hz) from static multipath; compute features: PCA, spectrogram, velocity profile), and activity classification (ML model: SVM, random forest, or deep learning (CNN, LSTM) trained on labeled CSI data to classify activities: walking, running, falling, sitting, gesturing, sleeping, and even specific gestures or typing patterns).