Wi-Fi CSI Experiments Got Better When I Split Collection from DSP

Keeping the ESP32 focused on reliable CSI capture and moving heavier analysis elsewhere made the sensing pipeline easier to debug.

By Reaz Romen✓ Site verified

Lab period August 2026 Published 2024-09-09 1 min read

Lab Note. Reconstructed in September 2026 from current engineering work and lab notes. The archive date indicates the period covered; the current site publication date is shown separately.

Trying to collect CSI and do every stage of signal processing on the same small device made the experiment harder to reason about.

I got a cleaner architecture by making the ESP32 responsible for timestamped CSI collection and transport, while a larger machine handled filtering, windowing, feature extraction and visualization. That kept radio timing and data capture close to the hardware without forcing the microcontroller to carry the analysis workload.

The split also improved reproducibility. Raw or lightly processed samples can be replayed through different DSP pipelines without repeating the RF experiment every time. If a classifier behaves strangely, I can determine whether the problem started in collection or analysis.

For sensing work that boundary is valuable. The radio endpoint should be boring and reliable; the experimental part can evolve quickly on the analysis side.