General Overview

This collection of preprints explores advancements across signal processing, communication systems, and medical imaging, with a particular emphasis on enhancing estimation and detection techniques. Several papers focus on refining fundamental algorithms. For instance, Lai et al. (2025) (Lai et al., 2025) introduce a recursive least squares algorithm with fading regularization, achieving finite-time convergence without persistent excitation. Verma and Bernstein (2025) explore trajectory prediction using adaptive input and state estimation (AISE) combined with the Frenet-Serret formulas (Verma & Bernstein, 2025). In a related work, they leverage AISE and the invariant extended Kalman filter (IEKF) for target tracking (Verma & Bernstein, 2025). Shah et al. (2025) (Shah et al., 2025) analyze the asymptotics of the maximum likelihood estimator for generalized linear models with 1-bit measurements.

Novel communication system architectures and algorithms also feature prominently. Sun (2025) (Sun, 2025) investigates near-field integrated sensing and communication (ISAC) with rotatable and movable antennas, proposing alternating optimization algorithms for sensing-centric and communication-centric designs. Heidari et al. (2025) (Heidari et al., 2025) introduce PhaseMO, an adaptive massive MIMO approach for energy efficiency without compromising coverage. Li et al. (2025) (Li et al., 2025) propose a hybrid beamforming design for holographic metasurface-based terrestrial users in multi-altitude LEO satellite networks.

The preprints also delve into specific challenges in signal recovery and reconstruction. Zhang et al. (2025) (Zhang et al., 2025) introduce a frequency-domain structure consistency loss function and a data component embedding strategy for signal data recovery, particularly in magnetic particle imaging. Sun et al. (2025) (Sun et al., 2025) present a support vector approach in segmented regression for map-assisted non-cooperative source localization. Tehrani et al. (2025) (Tehrani et al., 2025) improve quantitative ultrasound parameter estimation using ADMM with constraints on minimum physically feasible values.

Deep learning continues to influence the field, as demonstrated by Ren et al. (2025) (Ren et al., 2025), who advocate for separate source-channel coding (SSCC) empowered by large language models (LLMs) and error correction code transformers (ECCTs). Lee et al. (2025) (Lee et al., 2025) use deep learning for enhanced listened speech decoding from EEG. In medical imaging, Han (2025) presents deep learning frameworks for sparse-view CT reconstruction (Han, 2025) and interior tomography (Han & Wu, 2025).

Finally, several preprints explore applications in wireless communication and security, including terahertz communications (Zugno et al., 2025), interference reduction in MTC (Said et al., 2025), Starlink PNT applications (Qin et al., 2025), joint communications and sensing (JCAS) for 6G satellite networks (Sheemar et al., 2025), radio map construction using transformers (Li et al., 2025), ADHD classification with hyperdimensional computing (Colonnese et al., 2025), holographic metasurfaces for wave computing (Rahimian Omam et al., 2025), jammer detection in massive MIMO (Du et al., 2025), transient signal detection (Bao et al., 2025), secure beamforming (Sun et al., 2025), and electromagnetic signal injection attacks on autonomous vehicles (Liao et al., 2025). Dandapanthula and Ramdas (2025) (Dandapanthula & Ramdas, 2025) address multiple testing in sequential change detection, controlling the error over patience metric.

Paper Highlights

Holographic Metasurface-Based Beamforming for Multi-Altitude LEO Satellite Networks

Holographic Metasurface-Based Beamforming for Multi-Altitude LEO Satellite Networks by Qingchao Li, Mohammed El-Hajjar, Kaijun Cao, Chao Xu, Harald Haas, Lajos Hanzo https://arxiv.org/abs/2501.04164

Caption: This diagram illustrates the proposed hybrid beamforming architecture for a holographic metasurface-based terrestrial user terminal in a multi-altitude LEO satellite network. RF signals from the feed impinge on metamaterial elements within a microstrip waveguide, enabling holographic beamforming towards the serving satellite. Multiple RF chains connected to a digital beamforming unit allow for interference mitigation from other LEO satellites using a low-complexity MMSE algorithm.

This paper presents a significant advancement in addressing the challenges of providing seamless internet connectivity through Low Earth Orbit (LEO) satellite networks. The authors propose a novel hybrid beamforming design for terrestrial users equipped with holographic metasurface-based antennas, operating within a multi-altitude LEO satellite network. This architecture offers a compelling alternative to traditional beamforming designs, which often struggle with inter-satellite interference and rely on computationally intensive full Channel State Information (CSI) acquisition.

The proposed hybrid beamforming architecture strategically decouples the design process into two distinct sub-problems. The first focuses on optimizing the RF holographic beamformer to maximize the channel gain from the serving satellite to the terrestrial user. This optimization leverages the unique properties of holographic metasurfaces, allowing for precise beam shaping and steering. The phase shift of each element in the metasurface is carefully controlled to achieve this maximization, as represented by the formula: β^(m)_n = ∠(Σ^N_i=1q^(m)f^(0,m)_i) + ∠(q^(m) + ∠f^(0,m)_n). Here, β^(m)_n is the phase shift of the nth element in the mth microstrip, q^(m) is the response from the element to its feed, and f^(0,m)_n denotes the channel between the serving satellite and the element.

The second sub-problem addresses interference mitigation. A digital beamformer employing a Minimum Mean Square Error (MMSE) based detection algorithm is designed to minimize the impact of interference from other satellites. Crucially, to reduce the overhead associated with full CSI acquisition, a low-complexity MMSE algorithm is introduced. This algorithm relies on the statistical distribution of the LEO satellite constellation, obtained through stochastic geometry, rather than requiring precise knowledge of each satellite's channel state. This approach significantly reduces computational complexity without compromising performance, especially in dense LEO deployments. The covariance matrix of the baseband channels from interfering satellites, crucial for the MMSE design, is derived as: R_I = ζς²(λ/4π)²(|A| − 1)P₁L(d₀)(Q(C^H + (C^H) ⊗ I_MN)Q^H)/4. Here, ζ and ς are antenna and rain attenuation coefficients, λ is the wavelength, |A| is the number of satellites, P₁ is the probability of interference, L(d₀) is the average small-scale fading, Q is a matrix related to the holographic metasurface configuration, and C is the mutual coupling matrix.

Simulation results presented in the paper demonstrate the superior performance of the proposed architecture, particularly when mutual coupling effects, inherent in dense holographic metasurface deployments, are considered. The low-complexity MMSE algorithm, based on satellite distribution, achieves throughput comparable to its full CSI counterpart, but with significantly reduced overhead. Importantly, the holographic metasurface-based system outperforms traditional antenna array architectures in terms of throughput for a given transceiver size, highlighting the potential of this technology for future LEO satellite communications.

Use Cases for Terahertz Communications: An Industrial Perspective

Use Cases for Terahertz Communications: An Industrial Perspective by Tommaso Zugno, Cristina Ciochina, Sharad Sambhwani, Patrick Svedman, Luis M. Pessoa, Ben Chen, Per Hjalmar Lehne, Mate Boban, Thomas Kürner https://arxiv.org/abs/2501.03823

Caption: This infographic illustrates the potential of Terahertz (THz) communication, highlighting key opportunities like large bandwidth and sensing capabilities, alongside challenges such as free-space loss. It also showcases diverse use cases spanning immersive digital experiences, robotics, and healthcare, emphasizing the transformative impact of THz across various industries. The graphic further depicts the THz spectrum, its regulatory landscape, and its relationship to other frequency bands.

This paper offers a timely and crucial industrial perspective on the burgeoning field of terahertz (THz) communications. Based on the work of the ETSI Industry Specification Group (ISG) on THz, it provides a comprehensive overview of potential use cases and frequency bands of interest, setting the stage for future standardization efforts. The authors argue that the THz frequency band, spanning 0.1 to 10 THz, presents a unique opportunity for future wireless systems due to its vast bandwidth and unique propagation characteristics. Recent progress in RF component fabrication further strengthens this argument, suggesting that mass production of THz devices is within reach.

The ETSI ISG THz has meticulously identified several promising frequency bands within the 100 GHz - 1 THz range. Specifically, regulated bands between 100 and 275 GHz offer a substantial 91.2 GHz of allocated bandwidth for fixed or mobile services. Beyond 275 GHz, twelve bands totaling an impressive 488 GHz have been identified as promising candidates, with 91 GHz specifically earmarked for fixed and mobile services. The inherent advantages of THz communications are clearly outlined: the vast bandwidth enables higher capacity and lower latency compared to existing technologies, the enhanced directionality minimizes interference and maximizes spatial reuse, and the small antenna size facilitates compact transceiver design. Furthermore, the unique properties of THz radiation open doors to enhanced sensing capabilities, making it ideal for applications like e-health, immersive entertainment, and public safety.

The ETSI ISG THz has identified nineteen distinct use cases for THz communications, categorized into application areas such as robotics, environmental awareness, immersive digital experiences, and local area collaboration. These use cases paint a vivid picture of the transformative potential of THz technology. For example, high-speed THz links can enable efficient robot-to-network communication and support high-resolution environmental mapping. In environmental monitoring, THz sensors can detect harmful gases and pollutants with unprecedented accuracy. Immersive XR applications, demanding high data rates and ultra-low latency, can greatly benefit from THz connectivity.

However, the path to widespread THz adoption is not without challenges. The authors acknowledge the limitations of THz communication, including its high sensitivity to blockage, limited propagation range, and the current high power consumption of RF components. To overcome these hurdles, they highlight the importance of enabling technologies such as AI/ML, advanced MIMO, Reconfigurable Intelligent Surfaces (RIS), and energy-efficient device technologies. AI/ML can be leveraged to predict and mitigate blockage events, while advanced MIMO techniques can enhance spectral efficiency. RIS can strategically create additional propagation paths to extend coverage, and energy-efficient device technologies are essential for extending battery life and reducing operating costs.

The paper concludes by emphasizing the ongoing standardization efforts within the ETSI ISG THz. Work items focus on crucial aspects like use case identification, frequency band selection, channel modeling, and RF hardware modeling. The identified use cases present diverse requirements, with some demanding latency below 0.5 ms, data rates exceeding 1 Tbps, and localization accuracy down to 0.5 cm. These findings will inform the development of technical specifications for THz systems, ultimately paving the way for their seamless integration into future 6G and beyond networks.

Is Your Autonomous Vehicle Safe? Understanding the Threat of Electromagnetic Signal Injection Attacks on Traffic Scene Perception

Is Your Autonomous Vehicle Safe? Understanding the Threat of Electromagnetic Signal Injection Attacks on Traffic Scene Perception by Wenhao Liao, Sineng Yan, Youqian Zhang, Xinwei Zhai, Yuanyuan Wang, Eugene Yujun Fu https://arxiv.org/abs/2501.05239

Caption: This diagram illustrates an Electromagnetic Signal Injection Attack (ESIA) targeting an autonomous vehicle's perception system. The attacker injects malicious signals into the camera, distorting the captured image and causing the AI model to misclassify objects, as shown by the altered color and incorrect "motorbike" label in the "Results" section. This highlights the vulnerability of autonomous driving systems to ESIA and the potential for dangerous misinterpretations of the driving environment.

This paper tackles a critical security concern for autonomous vehicles: the vulnerability of camera-based perception systems to Electromagnetic Signal Injection Attacks (ESIA). These attacks can distort camera images, leading to misclassification of objects and potentially dangerous driving decisions. The authors delve into this under-researched area, investigating the robustness of various AI models under ESIA and developing a novel simulation method to facilitate further research.

Recognizing the difficulty in obtaining real-world ESIA attack data, the researchers developed a sophisticated simulation method. This method accurately mimics the adversarial patterns generated by ESIA, simulating the color channel distortions that create the characteristic color strip artifacts observed in real-world attacks. The validity of the simulation was rigorously tested by comparing the performance degradation of several object detection models on both simulated and real attack images. The high consistency in performance degradation between the two datasets, statistically confirmed by t-tests, validates the effectiveness and realism of the simulation method.

Leveraging this simulation method, the researchers constructed a dataset of simulated adversarial patterns applied to traffic images from the BDD100k dataset. This dataset was carefully categorized by attack severity (mild, moderate, and severe) and environmental conditions (weather, scene, and time of day). The performance of three state-of-the-art multi-task models (HybridNets, A-YOLOM, and YOLOP) was then evaluated on object detection and drivable area segmentation tasks. The results reveal a concerning trend: a significant performance drop across all models and scenarios with increasing attack intensity. For example, the mAP50 for object detection in the HybridNets model, under clear conditions, decreased by 10.80%, 31.60%, and 50.50% under mild, moderate, and severe attacks, respectively. These findings highlight the vulnerability of current AI models to even moderate levels of ESIA.

Further analysis uncovered intriguing patterns in model vulnerability across different driving scenarios. Surprisingly, models exhibited greater resilience to attacks in complex city street scenes compared to simpler highway environments, suggesting that scene complexity can partially mitigate the impact of ESIA. The researchers also investigated potential driving risks by visualizing model predictions under attack, identifying two particularly hazardous scenarios: "driving against traffic," where reverse lanes are misclassified as drivable, and "drivable area reduction," where the perceived drivable area shrinks significantly. Visualizing model attention using Grad-CAM revealed that attacks cause the model's focus to shift towards erroneous or less relevant areas, further underscoring the potential dangers of ESIA.

Conclusion

This newsletter highlights a diverse range of advancements and challenges in signal processing, communications, and related fields. From the development of novel beamforming techniques for LEO satellite networks using holographic metasurfaces to the exploration of terahertz communication use cases and the alarming vulnerability of autonomous driving systems to electromagnetic attacks, the papers covered in this newsletter underscore the dynamic nature of the field. The emphasis on incorporating domain-specific knowledge and constraints, as seen in the ultrasound parameter estimation and signal recovery papers, demonstrates a trend towards more robust and tailored algorithms. The continued influence of deep learning is evident in applications ranging from semantic communication to medical image reconstruction. Finally, the research on security threats to autonomous vehicles serves as a stark reminder of the importance of developing resilient and secure systems as we move towards a more connected and automated future. The common thread connecting these diverse research areas is the pursuit of more robust, efficient, and secure systems for a wide range of applications.