Subject: Cutting-Edge Advancements in Wireless Communications and Signal Processing

General Overview

This newsletter explores the latest breakthroughs in wireless communications, signal processing, and integrated sensing and communication (ISAC) systems. Several noteworthy papers introduce novel approaches to channel estimation and signal detection. Rodriguez Linares and Johansson (2025) present a low-complexity digital linearizer based on 1-bit quantizations, significantly reducing complexity compared to neural network-based linearizers while outperforming conventional methods. For near-field channel estimation in 6G, Lam et al. (2025) propose RACNN, a Residual Attention Convolutional Neural Network, demonstrating superior performance, particularly in mixed far-field and near-field conditions. Dong et al. (2025) address the complexities of fluid antenna systems by employing group sparsity methods for compressive space-frequency channel estimation and spatial equalization, achieving robust recovery and efficient hardware deployment. Duong et al. (2025) introduce a sparse orthogonal matching pursuit method for enhanced parameter estimation in mmWave MIMO ISAC systems, leveraging shared angle of arrival information between sensing and communication channels.

Innovative beamforming techniques also take center stage. Jin et al. (2025) present a general optimization framework for resource allocation in movable antenna-aided systems, tackling complex distance constraints through a penalty optimization framework and alternating optimization. Liu and Liu (2025) investigate distortion-aware beamforming for cell-free mMIMO systems, proposing distributed designs based on ring and star topologies to mitigate nonlinear power amplifier distortion. Yao and Zhang (2025) explore optimal beamforming for multi-target multi-user ISAC, examining the trade-off between communication and sensing beams. Zhang et al. (2025) investigate joint bistatic positioning and monostatic sensing, proposing beamforming optimization strategies for flexible performance trade-offs and demonstrating the advantages of mutual information fusion.

Several papers address specific challenges in diverse communication systems. Pan et al. (2025) propose a game-theoretic framework for interference avoidance in automotive FMCW radar, using Nash Equilibrium and Coarse Correlated Equilibrium concepts for frequency band allocation. Kim et al. (2025) introduce a low-complexity frequency domain nonlinear self-interference cancellation technique for flexible duplex systems. Mazokha et al. (2025) present MobRFFI, an AI-based device fingerprinting and re-identification framework for WiFi networks. Solenthaler et al. (2025) focus on identifying Orbcomm satellite RF fingerprints.

Further contributions include novel approaches to signal processing and optimization. Yin et al. (2025) propose a joint ML-Bayesian approach for adaptive radar detection. Garde et al. (2025) introduce the Age of Detection (AoD) metric. Huang et al. (2025) present an integrated communication and learned recognizer with customized RIS phases and sensing durations. Cao et al. (2025) investigate parameter learning under deficient excitation.

Finally, several papers explore applications of machine learning and optimization. Mohapatra et al. (2025) introduce TReND. Yu et al. (2025) propose a framework for uplink ISAC receiver designs. Tajja et al. (2025) present S-R2D2. Das et al. (2025) propose a learnable state-augmented policy for opportunistic routing. Wang et al. (2025) introduce an ambiguity-free broadband DOA estimation scheme. Kostrzewska and Kryszkiewicz (2025) propose a new nonlinear characteristics model for wideband radio amplifiers. Liu and Liu (2025) present a tri-timescale beamforming design. Tu et al. (2025) introduce a determinantal learning framework. Park and Seo (2025) explore the integration of fluid antenna systems into AirComp-based federated learning. Mylonopoulos et al. (2025) investigate integrated communication and RIS-aided track-before-detect radar sensing. Duan et al. (2025) compare Transformers and Graph Transformers for learning precoding. Bera et al. (2025) utilize recurrence plots for high impedance fault identification. Zhang et al. (2025) propose ExposNet. Abele et al. (2025) present experimental data on mechanical jitter. Perović et al. (2025) investigate sensing rate optimization. Bereyhi et al. (2025) explore MIMO transmission via pinching-antenna systems. Finally, Benoit and Asef (2025) investigate Google OR-Tools and machine learning for global path planning.

Paper Highlights

Deficient Excitation in Parameter Learning

Deficient Excitation in Parameter Learning by Ganghui Cao, Shimin Wang, Martin Guay, Jinzhi Wang, Zhisheng Duan, Marios M. Polycarpou https://arxiv.org/abs/2503.02235

Caption: Convergence of Parameter Estimation Errors Under Deficient Excitation

This paper introduces a groundbreaking online algorithm that excels in parameter learning under deficient excitation (DE), a more realistic and less restrictive condition compared to the traditional persistent excitation (PE). DE acknowledges that complete excitation might be unavailable, concentrating instead on leveraging available information. Mathematically, DE is defined as the existence of positive semi-definite matrices Φ_a and Φ_b of rank n - q such that:

Φ_a ≤ ∫_t^t+T* φ^T(τ)φ(τ) dτ ≤ Φ_b, ∀ t ≥ 0

where φ(t) is the regressor vector, n is its dimension, q is the order of deficiency (0 ≤ q ≤ n), and T is a positive real number.

The proposed algorithm dynamically calculates identifiable and non-identifiable subspaces, adapting to the specific excitation conditions. This allows it to generate an optimal parameter estimate in the least-squares sense. Impressively, even without persistent excitation, the learning error within the identifiable subspace converges exponentially to zero in noise-free scenarios. In the presence of noise, the estimate converges exponentially to the least-squares solution. This robustness is achieved by minimizing a cost function incorporating a forgetting factor, enabling the algorithm to track slowly varying parameters.

Furthermore, the paper extends this approach to distributed parameter learning, a crucial aspect of large-scale and networked systems. Local estimators often suffer from insufficient excitation due to limited measurements. The proposed distributed algorithm allows local estimators to operate within their identifiable subspaces and achieve consensus with neighbors in their non-identifiable subspaces through a cooperative learning strategy. This enables the entire network to estimate unknown parameters effectively, even when individual estimators lack sufficient excitation. Theoretical analysis guarantees exponential convergence of the overall parameter estimation error. The algorithm's effectiveness is validated through system identification examples, where it accurately learns system dynamics even with limited excitation. This highlights its potential for real-world applications.

A General Optimization Framework for Tackling Distance Constraints in Movable Antenna-Aided Systems

A General Optimization Framework for Tackling Distance Constraints in Movable Antenna-Aided Systems by Yichen Jin, Qingfeng Lin, Yang Li, Hancheng Zhu, Bingyang Cheng, Yik-Chung Wu, Rui Zhang https://arxiv.org/abs/2503.02344

Caption: This figure illustrates the concept of movable antennas (MAs) with minimum separation distance constraints. The blue dots represent MA positions (rₘ), confined within their respective regions (circles), while the orange lines depict the connections between MAs and a user (green dot). The red star and lines represent the auxiliary variables (zₘ) introduced to decouple the distance constraints, enabling efficient optimization.

Movable antennas (MAs) offer a new paradigm in wireless communication by dynamically adjusting antenna positions for enhanced performance. This paper addresses a key challenge in MA systems: the minimum antenna separation distance constraint (||rₘ - rₗ||₂ ≥ D). This constraint, coupled with the requirement that antennas remain within a designated region (rₘ ∈ C), creates a complex non-convex optimization problem.

The authors propose a novel penalty optimization framework to tackle this challenge. By introducing auxiliary variables (zₘ) and a penalty term, the framework decouples the distance and region constraints. The modified optimization problem (P1) becomes:

min_{{rₘ,zₘ}ₘ=₁ᴹ,X} f ({rₘ}ₘ=₁ᴹ, X) + p∑ₘ=₁ᴹ ||rₘ - zₘ||₂²

s.t. X ∈ Χ, rₘ ∈ C ∀m, ||zₘ - zₗ||₂ ≥ D ∀m ≠ l

where f(.) is the original utility function, X represents other variables, Χ is the feasible set of X, and p is a penalty factor. This decoupling allows the optimization to be split into two manageable subproblems. The first subproblem, concerning the original variables (X and {rₘ}), resembles a conventional resource allocation problem and can be solved using existing methods. The second subproblem, for the auxiliary variables ({zₘ}), admits a closed-form solution despite its non-convexity.

The framework's efficacy is showcased through case studies on capacity maximization, latency minimization, and regularized zero-forcing precoding. In each case, the framework outperforms existing methods, particularly as the number of MAs increases or the antenna panel region size decreases. The framework's rapid convergence (typically within 25 iterations) and superior performance make it a promising solution for optimizing MA-aided wireless systems.

Determinantal Learning for Subset Selection in Wireless Networks

Determinantal Learning for Subset Selection in Wireless Networks by Xiangliu Tu, Chiranjib Saha, Harpreet S. Dhillon https://arxiv.org/abs/2503.03151

Caption: This diagram illustrates the Determinantal Point Process-based Learning (DPPL) framework for subset selection in wireless networks. It shows the training phase, where a training set is generated and used to train the DPPL framework to obtain parameters and the likelihood function L(X), and the testing phase, where L(X) is sampled to produce the near-optimal schedule.

Subset selection, a fundamental problem in wireless resource allocation, often involves computationally demanding optimization problems. This paper introduces a scalable and efficient determinantal point process-based learning (DPPL) framework to address this challenge. The core idea is to model the optimal subset as a realization of a DPP, which naturally balances the trade-off between quality (e.g., signal strength) and similarity (e.g., mutual interference). The probability of a subset Y under a DPP is proportional to $\prod_{i \in Y} g_i \det(S_Y)$, where $g_i$ is the quality of item i and $S_Y$ is the similarity matrix of the subset Y.

This paper makes a significant contribution by introducing a novel method for constructing valid similarity matrices based on the Gershgorin Circle Theorem. This overcomes limitations of conventional methods that impose potentially unrealistic decomposability and symmetry constraints. The new method allows DPPL to capture more complex correlation structures, expanding its applicability to various wireless network settings.

The effectiveness of DPPL is demonstrated in two case studies: link scheduling in a 2D ad hoc network and a 3D cellular drone network. In both scenarios, DPPL achieves near-optimal sum-rates comparable to traditional optimization methods like geometric programming (GP), but with significantly reduced computational complexity. This efficiency is particularly pronounced in larger networks where traditional methods become computationally prohibitive. The DPPL framework’s ability to learn the underlying quality-similarity trade-off and its flexibility in constructing similarity matrices make it a promising solution for large-scale wireless networks.

Conclusion

This newsletter highlights significant advancements in wireless communications and signal processing. A common thread across these papers is the innovative use of optimization and learning techniques to address complex challenges. From tackling deficient excitation in parameter learning to developing scalable subset selection methods and optimizing resource allocation in movable antenna systems, these contributions demonstrate the potential for significant performance gains in future wireless networks. The development of new metrics like Age of Detection (AoD) and the exploration of novel system architectures like fluid antenna systems further underscore the dynamic nature of this field. The convergence of these advancements promises to shape the future of wireless communication and sensing technologies.