Subject: Cutting-Edge Advancements in Wireless Communication, Sensing, and Signal Processing

Hi Elman,

This newsletter covers recent preprints exploring diverse facets of wireless communication, sensing, and signal processing, introducing novel methodologies and addressing critical challenges in next-generation networks.

General Overview

This collection of preprints delves into various aspects of wireless communication, sensing, and signal processing, presenting innovative approaches and tackling key challenges in the development of next-generation networks. A significant focus lies in enhancing physical layer security and channel estimation techniques. For instance, Wang et al. (2025) Wang et al. (2025) propose a tag-based encoding (TBE) scheme for massive MIMO UAV systems, utilizing physical layer fingerprints for both authentication and secrecy in chaotic environments. Complementing this, Lee and Hong (2025) Lee & Hong (2025) introduce a two-timescale channel estimation method for RIS-aided near-field communications, exploiting the asymmetric coherence times of different channels to reduce pilot overhead. Further advancements in channel estimation are explored by Alikhani and Alkhateeb (2025) Alikhani & Alkhateeb (2025), leveraging digital twins as priors for zone-specific subspace prediction and calibration, resulting in improved accuracy and a reduced search space.

Novel antenna technologies and their applications also feature prominently. Lu et al. (2025) Lu et al. (2025) introduce the concept of fluid antennas, which allow for flexible radiation characteristics through manipulation of eigenmodes, opening new possibilities for beamforming and interference management. Zheng et al. (2025) Zheng et al. (2025) propose a rotatable antenna (RA) model, optimizing both beamforming and antenna deflection angles to maximize SINR in multi-user scenarios. Wang et al. (2025) Wang et al. (2025) analyze ESPRIT DoA estimation in large-dimensional regimes, identifying inconsistencies and proposing a correction via random matrix theory (RMT), leading to the development of the G-ESPRIT method. Furthermore, Wang et al. (2025) Wang et al. (2025) present a novel codebook design for unified near-field and far-field precoding in 6G MIMO systems, addressing the limitations of existing 5G codebooks.

Several contributions address specific applications and challenges in wireless systems. Shang et al. (2025) present two papers, one focusing on channel modeling and rate analysis of optical inter-satellite links (OISL) Shang et al. (2025), and another on spectrum sharing frameworks in satellite-terrestrial integrated networks (STINs) Shang et al. (2025). Yang et al. (2025) Yang et al. (2025) propose an efficient pre-processing method for 6G dynamic ray-tracing channel modeling, tackling the computational complexity of high-mobility scenarios. Fishel et al. (2025) Fishel et al. (2025) introduce Adaptive Rate Task-Oriented Vector Quantization (ARTOVeQ) for remote inference over dynamic links, enabling adaptation to varying channel conditions. Zhao et al. (2025) Zhao et al. (2025) investigate UAV-mounted active STAR-RIS for IoT NOMA networks, optimizing beamforming, trajectory, and power allocation for sum-rate maximization.

The application of deep learning is a recurring theme. Oh et al. (2025) Oh et al. (2025) propose a digital DeepJSCC framework with blind training for adaptive modulation and power control, demonstrating robust performance across diverse communication environments. Tian et al. (2025) Tian et al. (2025) develop a two-timescale approach for joint multi-user channel estimation and localization, leveraging spatial characteristics and introducing a scattering environment aware location-domain turbo channel estimation algorithm. Li et al. (2025) Li et al. (2025) present RainGaugeNet, a CSI-based rainfall classification model using sub-6 GHz signals, demonstrating high accuracy in both line-of-sight and non-line-of-sight scenarios. Several papers also explore applications in diverse areas, including EOG-based communication interfaces for quadriplegics (Raj & Kumar, 2025), modeling measurements for quantitative imaging of subsurface targets (Kim & Tsogka, 2025), and a comprehensive multiport network model for stacked intelligent metasurfaces (Abrardo et al., 2025).

Finally, several papers focus on optimization and resource allocation in various communication scenarios. Bahingayi et al. (2025) Bahingayi et al. (2025) investigate achievable rates in SIM-aided MIMO systems, proposing a hybrid optimization framework combining Riemannian manifold optimization (RMO) and weighted minimum mean square error (WMMSE) methods. Jalali et al. (2025) Jalali et al. (2025) address placement, orientation, and resource allocation optimization for cell-free OIRS-aided OWC networks, employing the ε-constraint method for multi-objective optimization. These contributions collectively advance the state-of-the-art in wireless communication and sensing, offering promising solutions for the challenges of future networks. Further research directions include exploring the interplay between these novel techniques, developing robust and scalable implementations, and validating their performance in real-world deployments.

Paper Highlights

LLM-FUESS: A Novel Approach to Soft Sensor Modeling

A Soft Sensor Method with Uncertainty-Awareness and Self-Explanation Based on Large Language Models Enhanced by Domain Knowledge Retrieval by Shuo Tong, Runyuan Guo, Wenqing Wang, Xueqiong Tian, Lingyun Wei, Lin Zhang, Huayong Wu, Ding Liu, Youmin Zhang https://arxiv.org/abs/2501.03295

This paper introduces LLM-FUESS (Few-shot Uncertainty-aware and self-Explaining Soft Sensor), a groundbreaking framework leveraging Large Language Models (LLMs) and In-Context Learning (ICL) for soft sensor modeling in industrial systems. Traditional soft sensors, reliant on supervised learning, face challenges like high development costs, poor robustness, training instability, and lack of interpretability. LLM-FUESS addresses these limitations by bypassing traditional training paradigms.

The framework comprises two key components: LLM-ZAVS (Zero-shot Auxiliary Variable Selector) and LLM-UFSS (Uncertainty-aware Few-shot Soft Sensor). LLM-ZAVS, enhanced by domain-specific knowledge retrieval from an Industrial Knowledge Vector Storage (IKVS), performs zero-shot auxiliary variable selection. It leverages Retrieval-Augmented Generation (RAG) and Chain of Thought (CoT) prompting to provide global and local explanations for variable importance rankings and scores. LLM-UFSS, on the other hand, utilizes structured process data formatted as text-based input-output pairs for ICL. It employs a context sample retrieval augmentation strategy from an Industrial Process Data Vector Store (IPDVS) to enhance the quality of context demonstrations, enabling prediction generation from few-shot examples without model training or parameter updates.

LLM-FUESS incorporates uncertainty quantification through confidence intervals and confidence scores, addressing a key limitation of traditional deterministic soft sensors. Its self-explanation capabilities generate human-readable explanations for predictions, enhancing transparency. Furthermore, the framework exhibits robustness to missing values, handling them as 'N/A' within the text-based input format.

Evaluation on industrial datasets demonstrated LLM-FUESS's superior performance. LLM-ZAVS significantly improved feature selection compared to traditional methods, while LLM-UFSS achieved the lowest MAE and RMSE across datasets, surpassing methods like Random Forest Regression, Multilayer Perceptron, and k-Nearest Neighbors Regression. The integration of LLMs and ICL, coupled with uncertainty quantification and self-explanation, positions LLM-FUESS as a significant advancement in soft sensor technology, offering a code-free, model-free approach that is robust, flexible, and interpretable.

BlindJSCC: Robust DeepJSCC for Adaptive Communication

Digital Deep Joint Source-Channel Coding with Blind Training for Adaptive Modulation and Power Control by Yongjeong Oh, Joohyuk Park, Jinho Choi, Jihong Park, Yo-Seb Jeon https://arxiv.org/abs/2501.02273

Caption: This diagram illustrates the BlindJSCC framework, showcasing the training phase with K encoder-decoder pairs and adaptable bit-flip probabilities (μ), and the communication phase where an optimal pair and modulation/power scheme are selected based on the environment. The transmitter encodes an image (e.g., a bird), and the receiver decodes the transmitted signal after passing through a channel, highlighting the adaptive nature of BlindJSCC for varying channel conditions.

Traditional DeepJSCC methods struggle to adapt to varying communication environments due to the complex relationship between power, modulation, and task performance. This paper introduces BlindJSCC, a novel framework addressing this challenge through error-adaptive blind training and training-aware communication.

Error-adaptive blind training models the communication channel using Binary Symmetric Channels (BSCs) with trainable bit-flip probabilities (μₙ), enabling joint optimization with the encoder and decoder. A continuous relaxation technique approximates the discrete binary error eₙ:

eₙ = -tanh((log(μₙ/(1 - μₙ)) + log(Uₙ/(1 - Uₙ)))/τ)

where Uₙ are random variables and τ is a temperature parameter. A regularized loss function ensures the bit-flip probabilities capture diverse communication scenarios:

L = E[d(u, û)] + λR(μ)

where d(u, û) is a distortion measure, λ is a regularization weight, and R(μ) encourages convergence to higher μₙ values, representing challenging channels.

Training-aware communication dynamically selects the optimal encoder-decoder pair and transmission parameters (power and modulation) based on current channel conditions and constraints on total power (Pₜₒₜ) and target rate (Rₜₐᵣ₉ₑₜ). Adaptive power control (APC) adjusts transmission power for a fixed modulation, while adaptive modulation and power control (AMPC) jointly optimizes both.

Simulations demonstrate BlindJSCC's superiority, achieving up to a 10% improvement in PSNR and reducing power consumption by up to 70% at 20 dB SNR compared to other DeepJSCC frameworks. Remarkably, BlindJSCC achieves this with only three encoder-decoder pairs, while others require nine. This highlights the effectiveness of blind training and adaptive communication for robust performance and efficient resource utilization.

Feedback-Driven Connectivity Enhancement for IoT

Connecting the Unconnectable through Feedback by Yimeng Li, Yulin Shao https://arxiv.org/abs/2501.02335

This paper presents a novel framework for enhancing uplink connectivity in IoT networks, particularly for cell-edge devices, by leveraging real-time feedback from access points (APs). Traditional methods like increased power or multi-antenna diversity are often impractical for resource-constrained IoT devices. This feedback-driven approach leverages the robust downlink coverage of APs to provide real-time decoding status feedback to IoT devices. This feedback, combined with feedback channel codes, enables devices to dynamically adapt their transmission strategies, effectively reducing the critical uplink SNR (Ω) required for successful communication.

The framework exploits the asymmetry between downlink and uplink communication. APs, with greater power and multiple antennas, provide robust downlink feedback, enabling IoT devices to adjust their transmission without increasing power consumption. The critical uplink SNR in feedback mode (Ω_f) depends on both uplink and downlink conditions, creating a dual-channel coupling effect:

Ω_f,dB ≈ 1/(exp(B_1,k + B₂R^-α_D - 1 - U₄) + U₄) + U₅

This coupling leverages the AP's energy to improve uplink coverage for power-constrained devices.

Analytical models quantifying coverage probability (φ_f(R)) and the number of connectable APs (M_f(D)) were developed, incorporating the dual-channel coupling effect:

M_f(D) = 2πλ Σ^K_k=1 *w_k*e^μ_UJ_2,k γ(α_U, μ_UJ_1,kD^α_U)/(μ_UJ_1,k)^α_U

Numerical results validated the models and demonstrated significant coverage improvements, especially at the cell edge. At 200 meters, the number of connectable APs increased by up to 51%, and coverage probability improved by three orders of magnitude compared to forward-only communication. This feedback-aided approach effectively "connects the unconnectable," offering a robust and energy-efficient solution for enhancing IoT uplink coverage in challenging environments.

Caption: Impact of Feedback on IoT Uplink Connectivity: Number of Connectable APs (M) vs. Distance to IoT Device (D).

Conclusion

This newsletter highlights a convergence of innovative approaches across wireless communication, sensing, and signal processing. From leveraging LLMs for soft sensor modeling and developing robust DeepJSCC frameworks to exploiting feedback mechanisms for enhanced IoT connectivity, these papers present significant advancements. The common thread lies in addressing critical challenges in next-generation networks, such as robust channel estimation, efficient resource allocation, and extending coverage to resource-constrained devices. The presented solutions offer promising pathways towards realizing the full potential of future wireless systems. The interplay between these techniques, particularly the integration of AI/ML and novel communication paradigms, warrants further investigation and promises to unlock new possibilities in the field.