Wi-Fi CSI Thermal Tomography with ESP32 Arrays: Contactless 2D Indoor Temperature Field Mapping for Smart Buildings

Indoor thermal comfort and energy efficiency depend not only on average room temperature but also on spatial temperature gradients that arise from solar gains, fa¸cade leakage, stratification, and non-uniform HVAC delivery. Conventional thermostats and point sensors cannot capture these variations without dense deployment, which increases installation and maintenance cost. This paper introduces a low-cost Wi-Fi Channel State Information (CSI) thermal tomography framework that reconstructs 2D temperature fields in indoor spaces using arrays of ESP32 nodes without any physical contact sensors in the monitored zone. We extend refractive-index-based temperature models to multiple Wi-Fi links and formulate a tomographic forward model in which each link provides an approximate line integral of temperature-induced phase and amplitude variations. A regularized inverse problem, solved using Tikhonov-regularized least squares and physics-guided basis functions, recovers the spatial temperature distribution on a grid from multi-link CSI features. A prototype testbed with 6–10 ESP32-WROOM-32 nodes deployed around a 6 m × 4 m room is evaluated under controlled heating scenarios, with ground-truth temperature fields obtained from a sparse thermistor grid and a reference infrared (IR) camera. Experiments show that Wi-Fi CSI thermal tomography achieves an average per-cell mean absolute error of 0.9◦C on a 12 × 8 grid, with peak errors below 1.8◦C and stable reconstruction under moderate human motion. Compared to interpolation from a limited set of thermistors, the proposed method improves field reconstruction accuracy by 27% using the same number of wired sensors, while requiring only boundary-mounted ESP32s. The entire reconstruction pipeline runs in 310 ms on a low-power edge computer (Raspberry Pi 4) receiving CSI from ESP32 nodes over UDP. These results demonstrate that commodity Wi-Fi infrastructure can be repurposed for coarse thermal imaging of rooms, enabling spatially aware HVAC diagnostics and control without dense physical sensor instrumentation.