Device-Free Indoor Localization with ESP32 Wi-Fi CSI Fingerprints: Grid-Based Regression, Feature Modeling, and Multi-Link Experiments

Device-free indoor localization (DFL) estimates the position of a person who carries no radio device, by monitoring how their body perturbs wireless channels.[6,7] Wi-Fi Channel State Information (CSI) is especially attractive for DFL due to its ubiquity and rich multipath information.[1] However, many CSI-based DFL systems rely on PC-class network interface cards and require laborious radio-map surveys.[2, 3] This paper presents a mathematically grounded and experimentally validated framework for 2D device-free localization using low-cost ESP32 modules that expose CSI. We consider a discretized grid of locations in a room and model the CSI fingerprints as random vectors whose distributions depend on the occupant’s coordinates. A feature mapping from raw complex CSI to amplitude–phase statistics across multiple ESP32 links is defined, and localization is formulated as (i) a multiclass classification problem over grid cells, and (ii) a regression problem for continuous coordinates. We design lightweight neural architectures for both settings and analyze their error in terms of mean distance error (MDE) and cumulative distribution functions. A testbed with three ESP32 transceivers deployed around a 6 m × 4 m laboratory collects CSI at 100 Hz for 35 grid points and two occupants, resulting in over 1.2 million labeled CSI snapshots. Experiments show that the proposed fingerprinting approach achieves a median localization error of 0.45 m and 90th percentile error below 0.9 m with only three links, while maintaining sub-10 ms inference latency on a Raspberry Pi 4 edge node. We compare against a k-nearest-neighbor (kNN) baseline and analyze the impact of grid resolution, number of links, and feature choices. The results demonstrate that ESP32-based CSI DFL can provide sub-meter accuracy with modest deployment effort, making it suitable for smart home and ambient assisted living applications.