Engineer builds mmWave radar that classifies materials to detect asbestos

A recent engineering graduate spent six months building an FMCW millimeter-wave radar capable of classifying materials by their electromagnetic signature, aimed at a practical European problem: detecting asbestos in walls without the costly lab-sampling process that inflates a roughly $1 analysis into a $60 bill. The hardware prototype pairs a Texas Instruments IWRL6432 development board with an ESP32, and the author built a test bench to measure how different materials respond to the radar’s chirps.

The signal chain works by sweeping the transmit frequency (a chirp), mixing the returning echo with the transmitted signal to produce a beat tone proportional to distance, running a range FFT to map reflected energy against depth, and applying Capon (MVDR) beamforming across a MIMO receive array to resolve angle of arrival. The result is a per-range, per-angle density ‘spectre’ — an electromagnetic fingerprint of the surface — which is fed as a tensor into a convolutional neural network that effectively learns each material’s permittivity (epsilon’ and epsilon”) to output a material class. Much of the two months of firmware work went into the usual embedded compile-flash-debug grind.

To avoid the price of commercial tools like Ansys HFSS, the author simulated antenna behavior in the open-source FDTD solver OpenEMS, speeding runs from an hour to two minutes by computing only the TX-to-RX transfer function and convolving it with chirps rather than simulating full propagation. Practical signal-theory tricks feature throughout — substituting a Gaussian pulse for an unstable Dirac impulse to extract transfer functions, and recognizing that an off-center Gaussian produces sinusoidal outputs. The project reached a working proof-of-concept classifier trained on beamforming signatures, but stalled out short of validating consistent asbestos detection at real-world concentrations, mainly due to a lack of funding.