Project M04 - 3D Reflectometric Material Characteriza-tionusing Co-Located MIMO Radar

Principal Investigators: Dr. Jan Barowski, Prof. Dr. Nils Pohl, RUB

Project M04 - 3D Reflectometric Material Characteriza-tionusing Co-Located MIMO Radar

Principal Investigators: Dr. Jan Barowski, Prof. Dr. Nils Pohl, RUB

Achieved results and methods

Multi-Distance Calibration of FMCW Radar Sensors and Material Characterization

Continuing our research from the 1st phase of MARIE, FMCW radar sensor calibration [1] has further been a major focus within M04. In contrast to a vector network analyzer (VNA), the self-heterodyne down-conversion principle in FMCW systems results in an intermediate frequency (IF) signal that occupies a certain bandwidth, depending on the range distribution of the relevant target. In contrast, the heterodyne stepped-frequency approach (utilizing two sources) of a VNA always results in a single tone IF signal, measuring its magnitude and phase. Additionally, FMCW transceivers utilize a band-pass filter before digitizing the IF signal. A high-pass contribution allows to block very close reflection components (e.g. bond-wires) while it strengthens the contributions from far remote targets that suffer from the r-4 path-loss term in the radar equation. A complementary low-pass contribution is commonly used to cut the IF-bandwidth to meet the sampling requirements of the analog-digital-converter (ADC).

While this is very basic in FMCW architecture considerations, it causes significant problems in calibrating the devices, since identical targets result in IF-signals at different beat-frequencies, depending on their distance to the sensor. During the 2nd phase of MARIE, we therefore utilized multiple calibration measurements with varying ramp-slopes to shift the IF-signal frequency of the calibration target in a static mechanical measurement setup. We demonstrated the shift of the calibration plane using a novel exceptionally stable FMCW radar [2] and the corrugated waveguide material-characterization-kit (MCK) from SwissTo12 [3] as a reference environment. Within this transmission-line setup, the radar sensor directly measures the reflection from a material sample. Due to the monostatic nature of the sensor, the transmission factor can only be observed in an indirect way by placing a short cap at the end of the opposing waveguide - therefore behind the sample - in approximately doubled distance. The calibrated measurements were performed using a “short” calibration measurement in the object plane and utilizing two measurements while the ramp-slope was doubled in the second measurement to emulate a target at twice the distance and to capture the IF-paths properties for both regions. The calibration result is presented in Fig. 1 using an HR-silicon wafer as a well-known test object. Both, magnitude and phase, are compared to a VNA measurement calibrated by the Through-Reflect-Match (TRM) method and match very well. In the context of material characterization, we have investigated on the extension of our available algorithms that rely on model inversion. For that purpose, machine learning based approaches have been considered [4]. Especially open-set techniques have been considered because these are able to detect and reject measurements that exhibit characteristics that were not part of the training. Therefore, strongly disturbed measurements, e.g. by flawed calibration, misalignment, or blocking objects, can be properly accounted for.

Compensation of Sensor Movements in SAR on fast Moving Mobile Platforms

While experimental SAR measurements in the lab can be well done under quasi-static assumptions [5,6], dynamic indoor and outdoor scenarios within MARIE [6,7] require continuous scanning while the measurement platform is moving. Additionally, this also significantly fastens the SAR imaging as it is usually limited by the mechanical scan speed. In contrast to pulsed radar systems that are usually used in large-distance remote sensing from space or aircraft, the sweep duration of our wideband FMCW transceivers (usually 1ms, covering more than 50 GHz) is comparably long to meet the requirements of the frequency synthesizer. Therefore, the quasi-static assumption in the signal processing cannot be hold, even if the platform moves with moderate speed, e.g. 1m/s. In this case, the position shift over the FMCW ramp duration (1ms) is already 1mm and therefore in the same order of magnitude as the radiating aperture (e.g. WR10 waveguide, 2.54  mm). This causes different problems, which have been investigated [8]. First of all, a significant Doppler-shift is induced into the target’s reflection pulse that needs to be compensated in order to achieve a coherent superposition from all measurements. While for far remote targets, the Doppler shift is nearly constant, i.e. the target hyperbola’s slope can be linearized, targets in close proximity exhibit a varying Doppler shift, due to the nonlinear nature of the range hyperbola around its point of closest proximity. Without a priori knowledge on the true position of the target, this is hard to compensate, and the target is strongly blurred. Additionally, the position shift causes problems regarding the sampling of the synthetic aperture. Usually, the sampling should be done at half aperture size steps, assuming an isotropic radiator. Again, the aforementioned movement of 1mm during the FMCW ramp is already within this order of magnitude. Furthermore, an additional settling and reverse time (0.2 ms) for the FMCW phase locked loop (PLL) needs to be considered between subsequent measurement and further increases the spatial shift.

To overcome this issues, we have demonstrated in [8] that a sub-band processing of the FMCW sweeps while scanning the aperture improves the spatial sampling and properly accounts for the nonlinear Doppler shift. The concept separates the complete FMCW chirp into N sub-bands of identical fractional bandwidth. Assuming a linear movement, N virtual measurement positions are provided for which relaxed quasi-static constraints can be met. This is demonstrated in Fig. 2. The improvement in image quality is shown in Fig. 3. By selecting N equal to the number of bins within a single IF-signal, the approach can be understood as a matched filter SAR processor where each ADC-bin corresponds to a sampling point along the synthetic aperture. Since this is computationally very expensive, it makes sense to keep N reasonably small and thus process larger sub-bands in order to profit from computationally more efficient FFT implementations. We could further show that the range resolution of the SAR image is kept, even though each sub-band only contributes a fraction of the measurement bandwidth. This is due to the coherent superposition of all sub-bands within the SAR processor (e.g. Backprojection) that re-establishes the total signal bandwidth.

Due to high number of measurements that is needed to form a SAR image and due to the superposition of measure-ments within the algorithm, we experienced that the measured reflection properties of the imaging target are strongly coupled to the angular behavior of its radar cross section (RCS). In the case of moving MIMO radars even the multistatic RCS is needed. While SAR algorithms usually assume isotropic scattering from the target, material characterization from single surface reflections aims to invert the measured reflection coefficient to obtain the underlying material parameters. In this case however, the reflection coefficient is strongly depended on the incident angle and thus non-isotropic. To cope with this, engineered and tailored wavefronts like the CSB will be investigated in the 3rd phase.

Near-field effects on micrometer accurate ranging

As accuracy demands increase, traditional far-field assumptions prove inadequate, with radar measurements impacted by phase variations that distort distance estimates—irrespective of the measurement technique used, be it pulse position (time-of-flight) or phase measurement. For the first time, our research [9] identified significant distance variations attributable to near-field effects in free-space measurements using ultrawideband mmWave radar, with these variations reaching up to a quarter wavelength, crucial for applications requiring micrometer accuracy. We introduced two computational solutions: a numerical simulation based on physical optics and an approximate closed-form expression. Through rigorous experimental validation, our methods demonstrated a high degree of accuracy (refer to Fig. 4), offering a practical means to adjust distance estimates accurately without the necessity for extensive and costly calibration beforehand.

Calibration-free distance measurement with micron accuracy

Based on the results of MARIE’s 1st phase, in the 2nd phase, we developed a breakthrough concept for measuring distances accurately without needing calibration [10]. This concept uses an ultrawideband FMCW radar system operating in the D-band, specifically between 126–182 GHz. This frequency range allows for a high range resolution and a quasi collimated beam despite the small size of the antenna, minimizing errors due to near-field effects. Our work represents the first instance of achieving micron-level accuracy in distance measurements over a medium range, using millimeter-wave radar technology, without the need for environmental adjustments or prior calibration. We tackled challenges related to the radar's hardware imperfections, the estimation of various parameters, and the signal's propagation path, including correction for the refractive index of air and near-field effects. In cooperation with C01, we established a radar based measurement setup for the measurement of gas refractivity [11]. The signal processing approach we designed to be used in ranging is robust to noise, micromotions, and interfering reflections. Moreover, it is efficient enough to run on the radar sensor's own embedded system. Extensive experiments, referenced against laser interferometry (see Fig. 6a), demonstrates the system's accuracy: errors were limited to within ±1 micron over a measurement range of 4.8 m (0.8–5.6 m) (see Fig. 6b), where the main source of error is due to near-field effects. Consistent testing under different conditions showed a maximum scale error of less than ±0.4 µm/m (see Fig. 6c), with a random error  as small as 30 nm (see Fig. 6a), providing high sensitivity. To our knowledge, these results set a new standard for accuracy in distance measurements with millimeter-wave radar technology.

In our research on achieving micron-level accuracy in distance measurements [10], we encountered significant challenges due to the humidity of air. The errors attributed to humidity readings were within the range of ±0.05 to ±0.4 µm/m, lower than anticipated due to the imprecision of humidity measurements. The most accurate humidity sensors available today have an accuracy of ±1 to ±2%RH, which under our experimental conditions, could result in an error margin of ±1.1 to ±2.3 µm/m. By employing advanced environmental sensor technology for compensation, we managed to attain the highest possible accuracy within the current technological constraints. Thus, the accuracy of radar-based distance measurements is essentially bounded by the accuracy of the humidity sensors used. Another challenge in accurate distance measurement is beam alignment. The radar's beam (or the direction it's pointed in) needs to match the straight-line path to the target. Misalignment causes a poorer signal quality and systematic errors in measurement, making the radar target appearing farther away than it actually is. We came up with a beam alignment scheme for planar radar targets, which is based on iterative maximization of the received signal strength. Within the 3rd phase we will investigate on MIMO beamforming methods to further address this issue.

In collaboration with C07 we were able to demonstrate the characterization of extremely small thickness variations in material characterization applications. As highlighted in [25], the system stability was shown to be much more important to these kinds of measurements than the absolute system bandwidth. Within this collaboration we scanned silicon wafers with etched Siemens stars to characterize the achievable resolution in both dimensions, cross-range and range. For this purpose, thickness variations of 20µm and 350nm have been fabricated in Darmstadt and measured with a D-band radar in Bochum. The results are presented in Fig. 8, demonstrating that micrometer deviations are resolvable with a very high accuracy and even Nanometer thickness variations can still be measured well.

In collaboration with M01, S04 and S05, first steps into the direction of large scale in-room sensing by means of SAR above 100 GHz and with ultra large bandwidths were done [26]. In a cooperation with C02, C03, C05, and M02, a compact and fully integrated 0.48 THz radar sensor was realized and characterized [27]. The transceiver reaches 53 GHz of bandwidth and feeds a dielectric lens from two on-chip antennas, which have been characterized using the antenna measurement setup from C05. The combined directivity including the lens antenna reaches 39.3 dBi, therefore narrowing the opening angle of the beam below 1°.  

Selected project-related publications

  1. Kueppers, T. Jaeschke, N. Pohl and J. Barowski, "Versatile 126–182 GHz UWB D-Band FMCW Radar for Industrial and Scientific Applications," IEEE Sensors Letters, vol. 6, no. 1, pp. 1-4, Jan. 2022, Art no. 3500204, doi: 10.1109/LSENS.2021.3130709.
  2. Abouzaid, T. Jaeschke, S. Kueppers, J. Barowski, and N. Pohl, “Deep learning-based material characterization using FMCW radar with open-set recognition technique”. IEEE Transactions on Microwave Theory and Techniques., vol. 71, no. 11, pp. 4628-4638, Nov. 2023, doi: 10.1109/TMTT.2023.3276053.
  3. Batra, J. Barowski, D. Damyanov, M. Wiemeler, I. Rolfes, T. Schultze, J. C. Balzer, D. Göhringer, and T. Kaiser, "Short-Range SAR Imaging from GHz to THz Waves", IEEE Journal of Microwaves, vol. 1, no. 2, pp. 574-585, Apr. 2021, doi: 10.1109/JMW.2021.3063343.
  4. Batra, J. Barowski, et al., "Millimeter Wave Indoor SAR Sensing Assisted with Chipless Tags-Based Self-Localization System: Experimental Evaluation," IEEE Sensors, vol. 24, no. 1, pp. 844-857, Jan. 2024, doi: 10.1109/JSEN.2023.3332431.
  5. Schorlemer, C. Schulz, N. Pohl, I. Rolfes and J. Barowski, "Compensation of Sensor Movements in Short-Range FMCW Synthetic Aperture Radar Algorithms," IEEE Transactions on Microwave Theory and Techniques, vol. 69, no. 11, pp. 5145-5159, Nov. 2021, doi: 10.1109/TMTT.2021.3108399.
  6. Piotrowsky, J. Barowski, and N. Pohl, "Near-Field Effects on Micrometer Accurate Ranging With Ultra-Wideband mmWave Radar," IEEE Antennas and Wireless Propagation Letters, vol. 21, no. 5, pp. 938-942, May 2022. doi: 10.1109/LAWP.2022.3152558
  7. Piotrowsky, S. Kueppers, T. Jaeschke, and N. Pohl, "Distance Measurement Using mmWave Radar: Micron Accuracy at Medium Range," IEEE Transactions on Microwave Theory and Techniques, vol. 70, no. 11, pp. 5259-5270, Nov. 2022. doi: 10.1109/TMTT.2022.3195235
  8. Hattenhorst, L. Piotrowsky, N. Pohl, and T. Musch, "A mmWave Sensor for Real-Time Monitoring of Gases Based on Real Refractive Index", IEEE Transactions on Microwave Theory and Techniques, vol. 69, no. 11, pp. 5033-5044, Nov. 2021. doi: 10.1109/TMTT.2021.3092718.
  9. Barowski, L. Schmitt, K. Kother, and M. Hoffmann, "Design, Simulation, and Characterization of MEMS-Based Slot Waveguides," IEEE Transactions on Microwave Theory and Techniques, vol. 71, no. 9, pp. 3819-3828, Sept. 2023. doi: 10.1109/TMTT.2023.3255589
  10. Starke, J. Bott, F. Vogelsang, B. Sievert, J. Barowski, C. Schulz, H. Rücker, A. Rennings, D. Erni, I. Rolfes, and N. Pohl, “A compact and fully integrated 0.48 THz FMCW radar transceiver combined with a dielectric lens,” International Journal of Microwave and Wireless Technology, pp. 1–12, 2023, doi: 10.1017/S1759078723001368.