Principal Investigator: Prof. Dr. Aydin Sezgin, RUB

Achieved results and methods

Aydin Sezgin: The contributions during the 2^nd phase of the MARIE are as follows. In the context of space-time signaling methods, we have proposed a novel Reinforcement Learning (RL) based approach for multi-target detection under clutter in MIMO radar systems [32]. In this work, we treated the radar scanning as a Markov decision process (MDP) and constructed an RL agent to scan the scene adaptively. At the beginning of the process, the radar sends orthogonal waveforms, just like a classical MIMO radar. After collecting the first echoes, the RL agent investigates each angle bin with a robust Wald-type detector and synthesizes waveforms optimized for each bin that needs to be investigated. The whole process is conducted in a reward-action cycle; therefore, the radar system has smooth operation and high performance even in dynamic targets under high clutter. The approach overperforms the classical omnidirectional MIMO radar operation in localization accuracy under the same power allocation and time-on-target.

In another work related to space-time processing, we have proposed a deep learning framework for the emulation of large virtual antenna arrays from a thinned array [33]. Maintaining high-performance operation under partial array blockage is an important task to maintain robustness in a radar system. This can be achieved by replacing the failed receive channels with their virtual emulations. However, deriving and estimating the signal at the virtual antenna is not a straightforward task. Therefore, in this work, we proposed a deep learning scheme that trains itself on the data collected by antennas in the array in various conditions to learn the mapping between an array with a smaller aperture and a bigger one. We have demonstrated that the proposed method not only successfully synthesizes the large array but is also able to overperform the fully populated physical arrays in angular estimation as a result of the denoising process that inevitably occurs during the synthesis of the large array in the neural network. In that sense, achieving robust and resilient operation in multi-RIS-aided radar setups under partial blockage would be an important topic in the 3^rd phase of S03; therefore, we believe that our experience here would be an asset for realizing the vision of MARIE.

In the context of RIS-aided radar setups, we have conducted one of the first studies on RIS-aided radar setups in the literature, where we investigated the use of RIS in a bi-static radar setup to localize hidden objects in the far field [34]. To this end, we also formulated a fast beam-scanning algorithm where the RIS produces beams with unit modulus and scans the scene by iteratively refining the width of the beams that are suspected to contain targets. With this work, we have demonstrated that RISs can be used as an effective tool for scanning hidden areas in a scene.

Apart from the theoretical investigations, we have investigated the practical viability of employing RF tags in RIS-aided radar configurations through an experimental study, establishing the model and conducting power budget calculations for such configurations, as outlined in [35]. We have observed that although traditional Friis models for power budget calculations are still valid for the mmWave frequencies, the nearfield coverage zones are extended considerably, particularly for indoor scenarios. This, in turn, would considerably affect the beamforming accuracy of the system. A noteworthy challenge in the 3^rd phase of S03 involves accurately tracking motion disturbances during platform movement, particularly in scenarios where RIS is obstructed. To address this challenge, integrating RF tags as supplementary anchor points in RIS-aided radars has been presented. The preceding study has provided us with valuable insights and experience in navigating this aspect of research.

The use of RF tags in the self-localization problem for a mobile platform, necessitates robust and fast autonomous algorithms. Since the location of the platform itself is unknown, classical techniques such as time gating are not sufficient to resolve the RF tags. For this reason, in our study at [36], we developed a method that reduces clutter and detects tags from long distances with weighted nuclear norm minimization. The spatio-temporal characteristics of the clutter and RF tags under RF illumination are different. While the clutter is distributed roughly uniformly, the RF tags are sparsely placed in the scene. Moreover, the tags can be designed to release the received RF pulse in a long, or short period of time, which are called high-Q and low-Q resonators, respectively. With this perspective, the received signal from such a scene can be seen as low-rank plus sparse recovery problem, where the clutter signal is modelled as the low-rank component and the RF tag reflection is modeled as the sparse one. When such a problem is solved, the RF tags can be determined from long distances without the need for any other apriori information. The same technique has been enriched with neural networks and applied to the other sensing problems [40] [41].

Daniel Erni: Regarding the design of scalable realistic RIS concepts applicable in the mm-wave and THz range, the contributions from the 2^nd phase of the MARIE are the following. In a systematic study we focused on liquid crystal (LC) tunable metasurfaces (operated as reflectarrays), where its voltage-controlled unit cells are implemented as double metal layers (Al) within a fused silica/LC/fused silica layer stack. A first test case for our developed technology included a dynamical metasurface with 31x31 unit cells each consisting of a tunable electric-inductive-capacitive (cELC) resonator operating at 28 GHz with a relative tuning range of 8.8% given a voltage swing of 10 V and a response time around 140 ms [42] yielding holographic beam-steering within a range of -40° to 40° [43]. This tuning concept was recently extended to a waveguide-fed, linear 40-element dynamical metasurface antenna (DMA) using a spatial harmonic-type of metasurface coding to achieve a 1D steering range of -32° to 34° at 105 GHz [44], respective, a 36x19 array was yielding -37° to 45° at 96.5 GHz [45], where corresponding structures derived from each of them for 2D beam-steering are investigated and will be ready soon [44,45]. At frequencies in the range of 293 to 304 GHz a preconfigurable metasurface concept using dynamical 1-bit digital metasurface coding (DMC) through optical activation of GeTe phase-change material within 36x36 square-shaped unit cells has been analyzed in [46]. The metasurface was operated from 3 different THz illumination angles, namely 0°, 45° and 75°, respectively, and provided associated beam scanning ranges of -13° to -45°, 0° to 30°, and 15° to 48°, respectively. Further steps included novel concepts to suppress grating lobes 2D leaky-wave antenna (LWA) arrays (and metasurfaces) using proper phase management of the involved space harmonics to inhibit grating lobes [47], as well as a push-pull excitation scheme of adjacent, complementary paired LWA elements for highly-efficient side-lobe suppression [48]. A measure to considerably reduce the metasurface's footprint relies on spoof surface-plasmon polariton (SSPP) propagation in distinct sub-wavelength metal structures. The concept has proven successful in test structures such as a pseudo-periodic holographic metasurface antenna [49] and a SSPP LWA using in addition a super-periodic arrangement of pre-fractal structures, yielding an ultra-wide -90° to 58° frequency beam scanning range [50]. Our contributions have also led to an invited comprehensive joint publication in the MARIE consortium summarizing our collective efforts in THz antenna design, modeling and measurements [51].

Note that in the 3^rd phase PI Erni is no longer PI in S03, since he is strengthening the workforce in C05. This is on the one hand due to the specific request of the DFG reviewers to emphasize C05 in the evaluation during the 2^nd phase, and on the other hand due to the fact that the PIs in C09 are extending their already working RIS with their sensational compact LC-based phase shifters (phase shift of up to 360°!) in the 3^rd phase even further in the direction of 250 GHz and above. Their unique technology is the showstopper for non-LC based RIS approaches like the one investigated by PI Erni in S03a, while a new initiative on fundamentally novel Reconfigurable Intelligent Volumes (RIV) is started by him in M03. To this end, S03 will intensively use and cooperate with C09 within MARIE.

Selected project-related publications

1. M. Ahmed, A. A. Ahmad, S. Fortunati, A. Sezgin, M. S. Greco and F. Gini, "A Reinforcement Learning Based Approach for Multitarget Detection in Massive MIMO Radar," in IEEE Trans on AES, Oct. 2021, DOI: 10.1109/TAES.2021.3061809.
2. M. Ahmed, U. S. K. P. M. Thanthrige, A. E. Gamal and A. Sezgin, "Deep Learning for DOA Estimation in MIMO Radar Systems via Emulation of Large Antenna Arrays," in IEEE Comm. Letters, May 2021, DOI: 10.1109/LCOMM.2021.3053114.
3. El-Absi, A. Abbas, D. Tubail, F. Ilgac, A. Abuelhaija, Y. Zantah, A. Sezgin, T. Kaiser, "Path Loss Modeling of RFID Backscatter Channels with Reconfigurable Intelligent Surface: Experimental Validation," in IEEE Access, 2023, 10.1109/ACCESS.2023.3317892.
4. Sánchez-Pastor, U. Thanthrige, F. Ilgac, A. Jiménez-Sáez, P. Jung, A. Sezgin, R. Jakoby, “Clutter Suppression for Indoor Self-Localization Systems by Iteratively Reweighted Low-Rank Plus Sparse Recovery,” Sensors, Oct. 2021, DOI: 10.3390/s21206842.
5. M. Ahmed, A. Alameer, D. Erni and A. Sezgin, "Maximizing Information Extraction of Extended Radar Targets Through MIMO Beamforming," in IEEE Geoscience and Remote Sensing Letters, April 2019. DOI: 10.1109/LGRS.2018.2876714
6. U.Thanthrige, P. Jung, and A. Sezgin, “Deep Unfolding of Iteratively Reweighted ADMM for Wireless RF Sensing,” Sensors, Apr. 2022. 10.3390/s22083065
7. P.-Y. Wang, B. Sievert, J. T. Svejda, N. Benson, F.-Y. Meng, A. Rennings, and D. Erni, "A Liquid Crystal Tunable Metamaterial Unit Cell for Dynamic Metasurface Antennas," IEEE Trans. Antennas Propagat., 2023, DOI: 10.1109/TAP.2022.3220945.
8. P.-Y. Wang, Y.-L. Lyu, F.-Y. Meng, A. Rennings, and D. Erni, "Grating-Lobe Suppression for Periodic Leaky-Wave Antennas at the Full Array Level," IEEE Antennas Wireless Prop. Lett., 2022, doi: 10.1109/LAWP.2022.3191981.
9. P.-Y. Wang, Y.-L. Lyu, F.-Y. Meng, A. Rennings, and D. Erni, "Sidelobe Suppression For Leaky-Wave Antennas Using A Complementary Paired Configuration," IEEE Antennas Wireless Prop. Lett., 2022, doi: 10.1109/LAWP.2022.3196399.
10. Farokhipour, B. Sievert, J. T. Svejda, A. Rennings, N. Komjani, and D. Erni, "Wide-angle Spoof Surface Plasmon Polariton Leaky-Wave Antenna Exploiting Pre-Fractal Structures With Backfire to Nearly Endfire Scanning," IEEE Antennas Wireless Prop. Lett., 2022, doi: 10.1109/LAWP.2022.3199073.
11. Sheikh, A. Prokscha, J. M. Eckhardt, T. Doeker, N. A. Abbasi, J. Gomez-Ponce, B. Sievert, J. T. Svejda, A. Rennings, J. Barowski, C. Schulz, I. Rolfes, D. Erni, A. F. Molisch, T. Kuerner, and T. Kaiser, "THz Measurements, Antennas, and Simulations: From the Past to the Future," IEEE J. Microwaves, 2023, doi: 10.1109/JMW.2022.3216210, (invited paper).