Remote environmental sensing · development
Remote Thermal Imaging Monitor
A personal project: a battery-powered monitor for remote sites, built around a small thermal-imaging sensor and a link home over cellular or satellite. The pages here work through the parts that have to be trustworthy before any of it means anything — turning the sensor's raw frames into real temperatures, and knowing how often the data can actually get out. Rough by design; a working phase, not finished things. Day-to-day progress is in the Journal.
Sensor — Grid-EYE
A Panasonic AMG8854M01 8×8 thermopile array, read over I²C by an nRF9151. What it takes to get from ±3 °C datasheet accuracy to something usable — pixel-to-pixel uniformity, the optical window, per-pixel correction against a PT1000 on the bench.
Flat-Field Uniformity Study
How much of the array's ~2.9 °C pixel-to-pixel spread is a fixed pattern that calibrates out, measured over a 2 → 21 °C reference sweep. A per-pixel offset table more than halves it.
Window Calibration
Warm-up sweeps with an IR window glued to the sensor, differenced against the no-window baseline. The window bias is scene-dependent and changes sign, and the window sits at air temperature, not the die — so a warm-up bounds the compound term but not transmission alone.
Production Calibration
Turning the bench method into a fast per-unit routine — discrete isothermal plates, an on-device least-squares fit, coefficients to NVS, and a console command interface for the nRF9151.
Literature
Reference list for the thermal channel: thermopile-array correction, water skin-vs-bulk temperature, IR validation against flowing water, and windows in front of an IR sensor.
Two-Zone Contrast Test soon
Apparent-temperature error versus range and window state, using a Peltier-driven contrast target on an isothermal plate.
Sensor — VL53L8CX Ranging
An ST VL53L8CX 8×8-zone Time-of-Flight ranging array, sharing the nRF9151's I²C bus with the Grid-EYE. Bring-up, and what it takes to put a window in front of it without losing the measurement.
First Bring-Up
Getting a real 8×8 distance frame end to end — two software bugs, a bus signal-integrity fix, and a missing bus-arbitration lock that was the actual cause of the intermittent failures.
Window Transmission
Paired 31-frame series, no window vs the Grid-EYE's ZnS window, at the sensor's actual 940 nm ranging wavelength. Signal drops by roughly half; reported distance does not move.
Static-Scene Noise
254-frame precision characterization against a stable office scene. Flat-region zones hold ~0.3% noise; zones on a real depth edge are far noisier. Repeating the run with the lights off changes neither.
Xtalk & Offset Calibration soon
Per-unit crosstalk and offset calibration via the ST ULD driver, stored on-device.
Connectivity — NTN
A remote site may have no cellular coverage at all, so the link home falls to non-terrestrial-network satellites. How sparse are the connect windows, and how does that bound what the device can send?