Sparse Terahertz Frequency-Domain Sensing with Kilohertz Measurement Rate

Lauri Schwenson, Florian Walter, Alexander Jäckel, Konstantin Wenzel, Lars Liebermeister, Cornelius Mach, Enrique Castro-Camus, Martín Koch, Martin Schell, Robert B. KohlhaasJournal of Infrared Millimeter and Terahertz Waves 2025 · DOI

Abstract

The widespread industrial application of terahertz spectroscopy is currently constrained by system complexity, high costs, and limited measurement rates. Optoelectronic frequency-domain spectroscopy (FDS) offers a promising approach to mitigate the system complexity and costs. To address the need for fast measurement rates, we present the first terahertz spectrometer capable of a flexible sparse sensing measurement mode, achieving a 6.25-fold increase in measurement rate compared to state-of-the-art FDS systems. The cycle through eight discrete frequencies in a 1.5-THz window is acquired in 800 µs and achieves a peak dynamic range exceeding 90 dB in 800 ms. We apply this scheme to a multi-layer thickness evaluation, showing a maximum single-shot standard deviation of 2 µm across all three layers. The sparse sensing method shows equivalent layer thickness accuracy per measurement time when compared to a full spectral scan with a state-of-the-art spectrometer.

Introduction

Terahertz (THz) spectroscopy is a critical tool for non-destructive testing, particularly for measuring the thickness of individual layers in multi-layered materials like automotive paint or plastic pipes. However, industrial adoption is hindered by a "speed vs. cost" trade-off. Time-Domain Spectroscopy (TDS) is fast but relies on expensive, bulky femtosecond lasers. Frequency-Domain Spectroscopy (FDS) uses cheaper, simpler diode lasers but is traditionally slow because it sweeps through a continuous range of frequencies. This research addresses the speed bottleneck of FDS, aiming to reach kilohertz measurement rates—fast enough for real-time monitoring on a factory floor—without sacrificing the cost-effectiveness of the FDS architecture.

Methods

The proposed system introduces a sparse sensing measurement mode. Instead of a continuous sweep, the system utilizes a "jumping" laser that switches rapidly between eight pre-selected discrete frequencies (ranging from 50 GHz to 1.5 THz) within an 800 µs cycle.

Schematic of the sparse sensing THz spectrometer setup, showing optical components, the sensor head, and the signal processing chain.

Key technical components include:

  • Hardware: Two continuous-wave (cw) lasers (one static, one tunable), a phase modulator, and an FPGA (Xilinx Zynq 7010) for high-speed signal acquisition and modulation control.
  • Signal Processing: An offline lock-in evaluation extracts amplitude and phase information from the stable segments of the laser's "dwell time" at each frequency.
  • Algorithm: To calculate layer thicknesses from sparse data, a two-step optimization based on the Transfer Matrix Method (TMM) is employed. A simplified objective function quickly narrows the search space before a final high-precision fit, allowing the algorithm to run in under 1 ms on standard hardware.

Results and Discussion

The system achieved a repetition rate of 1.25 kHz, with one full 8-frequency cycle completed every 800 µs. The dynamic range reached 60 dB in a single shot and over 90 dB with 800 ms of averaging, matching the noise floor of state-of-the-art continuous systems.

Time-domain representation of the THz signal over two measurement cycles, illustrating the rapid switching between eight discrete frequencies.

Validation was performed using a 30-mm-thick high-density polyethylene (HDPE) sample for phase accuracy and a three-layer ceramic sample for thickness evaluation. The sparse method demonstrated a single-shot standard deviation of 2 µm or less across all three layers. Crucially, the sparse sensing approach showed equivalent thickness accuracy per unit of measurement time when compared to a full spectral scan.

Conclusion

The core innovation is that for specific industrial tasks like thickness gauging, a full spectrum is redundant; a few well-chosen frequency points provide sufficient information to solve the underlying physical model. While the current system has a ~20 dB lower dynamic range than continuous systems due to higher electronic bandwidth requirements, future work aims to implement lock-in evaluation directly on the FPGA. This could push measurement rates to 10 kHz, further enhancing real-time industrial monitoring capabilities.

Summary updated

Abstract

Abstract The widespread industrial application of terahertz spectroscopy is currently constrained by system complexity, high costs, and limited measurement rates. Optoelectronic frequency-domain spectroscopy (FDS) offers a promising approach to mitigate the system complexity and costs. To address the need for fast measurement rates, we present the first terahertz spectrometer capable of a flexible sparse sensing measurement mode, achieving a 6.25-fold increase in measurement rate compared to state-of-the-art FDS systems. The cycle through eight discrete frequencies in a 1.5-THz window is acquired in 800 µs and achieves a peak dynamic range exceeding 90 dB in 800 ms. We apply this scheme to a multi-layer thickness evaluation, showing a maximum single-shot standard deviation of 2 µm across all three layers. The sparse sensing method shows equivalent layer thickness accuracy per measurement time when compared to a full spectral scan with a state-of-the-art spectrometer.
BibTeX
@article{schwenson2025sparse,
  author = {Lauri Schwenson and Florian Walter and Alexander Jäckel and Konstantin Wenzel and Lars Liebermeister and Cornelius Mach and Enrique Castro-Camus and Martín Koch and Martin Schell and Robert B. Kohlhaas},
  title = {{Sparse Terahertz Frequency-Domain Sensing with Kilohertz Measurement Rate}},
  journal = {Journal of Infrared Millimeter and Terahertz Waves},
  year = {2025},
  doi = {10.1007/s10762-025-01072-6},
}
Figure