Terahertz Nondestructive Testing with Ultra-Wideband FMCW Radar

Research publication · Sub-terahertz nondestructive testing

Terahertz Nondestructive Testing with Ultra-Wideband FMCW RadarFMCW radar transmits a continuously swept frequency and measures the beat signal produced by delayed reflections, enabling distance, thickness, and depth-resolved imaging with compact coherent hardware. More

Barnabé Carré, Adrien Chopard, Jean-Paul Guillet, Frederic Fauquet, Patrick Mounaix and Pierre Gellie · Sensors · 2022 · Volume 23, issue 1, article 187 · DOI: 10.3390/s23010187

This paper describes the design and laboratory validation of a frequency-modulated continuous-wave radar sweeping from 128 to 160 GHz. Its homodyne harmonic-mixing architecture combines a 32 GHz bandwidth, measurement rates up to 7.62 kHz and a dynamic range that exceeds 100 dB only under long averaging conditions. The system was characterized for lateral and depth resolution, stability and range precision, then used to image polymer structures, adhesive regions, food samples, granite and pharmaceutical packaging. These demonstrations show that one non-contact platform can reveal interfaces or foreign objects in selected dielectric materials. They do not establish an inline inspection product or verified performance across all materials in the illustrated sectors.

FMCW ranging converts propagation delay into a low-frequency beat signal. A transmitted chirp and its delayed reflection have slightly different instantaneous frequencies; mixing them yields a tone whose frequency is proportional to target distance. A broad chirp separates nearby interfaces more effectively, and repeated sweeps can be acquired much faster than a mechanically scanned time-domain waveform. At approximately 150 GHz, the free-space wavelength is around 2 mm, offering a useful compromise between penetration into low-loss dielectrics and millimetre-scale focusing.

The radar begins with a phase-locked voltage-controlled oscillator sweeping from 16 to 20 GHz. Schottky multiplier stages first quadruple and then double this signal, producing the 128-160 GHz output with peak power reported near 50 mW. A WR-06 horn and an aspheric lens form the beam. The return re-enters through a directional coupler and is combined with the low-frequency oscillator in a harmonic mixer. The difference involving the eighth harmonic carries the target delay; filtering, amplification and digitization recover the range profile. This architecture keeps transmitter and receiver coherent without requiring a second high-frequency swept source.

Featured visual: Image 1 from the Airtable record associated with this publication. Consult the original paper for the authoritative figure caption and interpretation. Source publication.

Visuals are drawn from the Airtable research archive. Figure numbering, rights and interpretation should be checked against the original publication before republication outside this site.

Characterizing what bandwidth and averaging really provide

A USAF target measured the lateral response at approximately 2 mm with numerical aperture 0.5. Along the range axis, 32 GHz bandwidth gives a theoretical resolution near 4.6 mm in air. Processing changes the apparent peak width and sidelobes. A sinc-squared apodization produced a measured main-lobe width around 4 mm, while a Blackman window suppressed sidelobes below about -60 dB but broadened effective resolution to roughly 7.4 mm. Zero-padding can interpolate the displayed range profile and improve peak localization; it does not create the ability to separate two physical interfaces more closely than the bandwidth permits.

Dynamic range also depends strongly on acquisition time. At the native cadence of 7.62 kHz, corresponding to a chirp every 132 microseconds, short measurements produced about 60 dB. Averaging very large numbers of chirps over several seconds reduced the noise floor and allowed values above 100 dB. The highest dynamic range and fastest independent update are therefore not simultaneous specifications. An inspection process must choose integration based on target reflectivity, required speed and acceptable noise rather than treating both maxima as one operating point.

Repeated range measurements tested stability and precision. With 4.2 seconds of integration, the reported repeatability reached approximately 3.6 micrometres for a stationary reflector. This precision concerns locating a peak repeatedly, not resolving two interfaces separated by 3.6 micrometres. A calibrated step target assessed relative ranging accuracy over millimetre-scale displacements. Thermal stabilization and a controlled laboratory environment helped maintain performance during measurements extending to seven hours, so field use would need to determine how vibration and temperature changes affect the same metrics.

Imaging diverse samples without claiming universal inspection

The radar was raster-scanned over areas up to 15 by 15 cm with 1 mm steps. Volumetric data sets were acquired in less than fifteen minutes, with motion of the translation stage rather than chirp generation setting much of the total time. In polymer test pieces, internal surfaces and an adhesive gap produced distinguishable returns. Such contrast arises where dielectric properties change; its visibility depends on layer thickness, incidence, absorption and reflection coefficient. The examples demonstrate interface mapping in the chosen polymers, not an assured response for every plastic or bonded joint.

Food tests examined chocolate containing glass fragments. The reported glass-to-chocolate contrast, around 8 to 10 dB in a selected thin region, was sufficient to localize the inclusion in those samples. A crushed-granite cylinder produced substantial scattering and attenuation, yet voids and a metal rod generated features in the radar volume. Multiple scattering created speckle-like backgrounds that would complicate threshold-based detection. Material-specific calibration and representative sets of harmless and defective products would be necessary before false-positive and detection rates could be known.

In a pharmaceutical package, the radar distinguished cardboard layers, the presence or absence of a leaflet and the positions of tablets in a blister. The metal backing supplied a strong reference whose apparent range changed where a pill altered the propagation path. This provides an indirect count under the demonstrated geometry. It does not verify tablet chemistry, package sterility or compliance with an industrial inspection standard. Likewise, the non-ionizing character of 150 GHz radiation is a physical advantage for contactless sensing, not by itself a complete safety or regulatory assessment of an installed system.

The diversity of samples is useful because it tests the available dynamic range against weak polymer interfaces, scattering mineral structures and strongly reflecting metal-backed packaging. It also exposes why no single acquisition setting serves every case. Strong sidelobes may mask a weak nearby layer; windowing can suppress them but reduces depth resolution. Longer averaging reveals weak returns but lowers throughput. Mechanical raster scanning generates detailed volumes but is unlikely to be the final architecture for a fast conveyor or large component.

Further collaboration could pair the transceiver with line scanning, arrays, robotic motion or synthetic-aperture processing, then evaluate it on well-defined defect populations. Environmental stability, calibration transfer, curved surfaces and product motion would need to be measured explicitly. The paper supplies a strong laboratory instrument and an unusually broad set of feasibility examples. Its most precise contribution is the connection between architecture, bandwidth, averaging and observed contrast, which gives future developers enough information to choose realistic operating compromises rather than extrapolate from headline specifications.

This publication also connects with our broader work on FMCW radar expertise and non-destructive testing.

Bibliographic reference