General Overview

This newsletter explores cutting-edge advancements in wireless communications, sensing, and signal processing, focusing on emerging technologies like extremely large-scale MIMO (XL-MIMO), reconfigurable intelligent surfaces (RIS), and fluid antennas (FA). We'll delve into the challenges and opportunities these technologies present for localization, beamforming, and interference mitigation.

Zhou et al. (2025) propose a three-step mixed near-field and far-field target localization method for XL-MIMO. This addresses critical issues like half-wavelength antenna spacing limitations and hybrid UPA architecture. Their approach involves signal reconstruction, a modified MUSIC algorithm for angular parameter estimation, and a novel classification method to resolve angular ambiguity and estimate near-field target ranges. Meanwhile, Chrysanidis et al. (2025) introduce deceptive wireless beamforming (DWB). This technique injects deceptive Doppler and range information into the transmitted signal to mislead eavesdroppers, achieving this through a power-efficient optimization problem. Johansson (2025) focuses on complex baseband signal reconstruction from nonuniform bandpass samples, proposing a least-squares optimal time-varying FIR filter design. This offers enhanced reconstruction accuracy and reduced filter order compared to existing windowing-based methods.

The application of deep learning to wireless communications is another key theme. Deng and Han (2025) propose a novel DNN architecture for precoder learning in weighted sum rate maximization, leveraging joint unitary and permutation equivariance for improved learning performance and reduced complexity. Liu et al. (2025) utilize graph neural networks (GNNs) for joint learning of wideband user scheduling and hybrid precoding in scenarios with massive user combinations and shared analog precoders. They also address the "same-parameter same-decision" property that can hinder GNN learning. For automatic modulation classification (AMC), Yang and Sahay (2025) propose a deep ensemble approach for robust uncertainty quantification, enhancing prediction reliability in noisy environments. Liao et al. (2025) introduce SincAlignNet, a novel network combining SincNet and contrastive learning, for auditory attention detection using EEG and audio signals.

Integrated sensing and communication (ISAC) features prominently. Liu et al. (2025) propose an ISAC beamforming method to enhance sensing performance by minimizing the integrated sidelobe level ratio while maintaining communication quality. Miranda et al. (2025) explore joint delay-Doppler estimation using OFDMA payloads in multi-node ISAC networks. Magbool et al. (2025) investigate RIS-aided target obfuscation in ISAC, minimizing the sensing signal-to-interference-plus-noise ratio at an adversarial detector. Gal and Bilik (2025) propose a hybrid approach for NLOS target localization in automotive radar.

Beyond ISAC, the collection explores diverse applications. Chen et al. (2025) analyze the ambiguity function of frequency-hopping MIMO radar with movable antennas. Skog et al. (2025) investigate the connection between magnetic-field odometry aided inertial navigation and magnetic-field SLAM. Yang et al. (2025) introduce adaptive subarray segmentation for near-field channel estimation in XL-MIMO. Guo et al. (2025) propose a deep learning-based joint CSI estimation-feedback-precoding framework for MU-MIMO OFDM systems.

Finally, several papers focus on specific system designs. AlaaEldin et al. (2025) analyze and optimize RIS-enabled multi-user M-QAM uplink NOMA systems. Weinberg and van der Merwe (2025) develop a methodology for passive sonar sensor placement for undersea surveillance. Zheng et al. (2025) investigate enhanced beampattern synthesis using electromagnetically reconfigurable antennas (ERAs). Kim et al. (2025) propose a generative diffusion model-based compression of MIMO CSI.

Paper Highlights

Magnetic-Field SLAM Gets a Boost

On the Connection Between Magnetic-Field Odometry Aided Inertial Navigation and Magnetic-Field SLAM by Isaac Skog, Manon Kok, Gustaf Hendeby, Chuan Huang, Thomas Edridge https://arxiv.org/abs/2503.04286

Caption: Position Error Comparison of Indoor Navigation Systems

Magnetic-field simultaneous localization and mapping (SLAM) offers a compelling approach for scalable indoor localization using readily available and cost-effective inertial and magnetometer sensors. However, the inherent rapid error accumulation in low-cost inertial navigation systems significantly limits the duration of exploration phases, hindering their practical deployment. Recent advancements in processing data from magnetometer arrays have demonstrated the feasibility of extracting odometry information – specifically, displacement and rotation data – from local magnetic field variations. This has led to the development of magnetic-field odometry-aided inertial navigation systems, which exhibit significantly slower error growth rates. This paper investigates the crucial connection between these two approaches, exploring whether a magnetic-field SLAM system can implicitly extract odometry information when utilizing a magnetometer array, without requiring any modifications to its algorithms.

The study employs a comprehensive system model that encompasses both magnetic-field odometry-aided inertial navigation and magnetic-field SLAM systems. This model considers a navigation platform equipped with multiple three-axis magnetometers and an inertial measurement unit. The navigation state, denoted as x_τ, includes the platform's location, orientation, and any necessary auxiliary states. Critically, the ambient magnetic field, represented by m(r), is modeled as a continuous stochastic process, often characterized as a Gaussian process. The paper meticulously analyzes the Bayesian filter recursions for both SLAM (p(x_t, m|Y_1:t, u_1:t-1)) and odometry-aided inertial navigation (p(x_t|Y_1:t, u_1:t-1)) systems. This analysis reveals that the SLAM system can, in fact, extract odometry information through the process of marginalization. A key finding is that, without additional assumptions, the optimal odometry-aided solution can only be achieved by solving the full SLAM problem.

Simulations were conducted to validate these theoretical findings. A synthetic magnetic field, based on real-world data, was generated, and measurements from a moving sensor array were simulated. The results clearly demonstrated that when a magnetometer array is employed, the magnetic-field SLAM system successfully extracts odometry information. This leads to a remarkable reduction in error growth during exploration—by two orders of magnitude compared to using a single low-cost magnetometer. As anticipated, the SLAM system maintained a bounded position error, while the odometry-aided system's error accumulated over time due to the lack of loop closure information.

While these theoretical and simulation results strongly support the potential of magnetic-field SLAM systems with magnetometer arrays for extracting odometry information and extending exploration phases, significant practical challenges remain. A primary challenge is the computational complexity associated with generating the high-resolution magnetic field maps necessary for accurate odometry extraction. Existing SLAM systems typically model field variations at the meter scale, whereas odometry-aided systems require sub-meter resolution. This necessitates a substantial increase in the number of basis functions used in the magnetic field model (m(r) ≈ Σ^n_b_i=1 φ_i(r)θ_i), resulting in a considerable increase in computational cost. Future research should prioritize the development of multi-resolution mapping techniques to address this challenge, potentially by combining local high-resolution maps for odometry with global low-resolution maps for overall localization.

Adaptive Subarray Segmentation Revolutionizes Near-Field Channel Estimation

Adaptive Subarray Segmentation: A New Paradigm of Spatial Non-Stationary Near-Field Channel Estimation for XL-MIMO Systems by Shuhang Yang, Puguang An, Peng Yang, Xianbin Cao, Dapeng Oliver Wu, Tony Q. S. Quek https://arxiv.org/abs/2503.04211

Caption: The diagram illustrates the off-grid subarray segmentation-based assorted block sparse Bayesian learning algorithm under the multiple measurement vectors framework (SS-ABSBL-MMV). It shows how observations from segmented subarrays are processed through multiple blocks (G) with corresponding P and Q parameters, ultimately contributing to angular domain estimation, guided by a weak constraint and influenced by noise variance (σ²). This adaptive segmentation allows for more accurate channel estimation in XL-MIMO systems by accounting for spatial non-stationarity.

Extremely large-scale multiple-input multiple-output (XL-MIMO) systems operating at millimeter-wave/terahertz frequencies hold immense promise for 6G and beyond. However, the inherent challenges of near-field propagation and spatial non-stationarity (SnS) significantly complicate channel estimation (CE). Traditional methods often rely on dividing the array into equal subarrays, which can be suboptimal due to the dynamic and varying nature of SnS. This paper introduces a groundbreaking framework based on adaptive subarray segmentation to overcome these limitations and achieve highly accurate near-field CE.

The authors begin by developing a refined spherical-wave channel model specifically tailored for line-of-sight (LoS) XL-MIMO systems, explicitly incorporating SnS effects. This model accounts for non-ideal LoS paths that may occur in obstructed-LoS scenarios and utilizes VR-based weighting masks to accurately capture the spatially varying channel characteristics. The proposed channel model is given by: h = gLose-infrLos. b (rlos, Los) S, where S represents the SnS properties. The authors then analyze the detrimental effects of both over-segmentation and under-segmentation, highlighting the critical importance of precise subarray division. Over-segmentation, which divides the array into more subarrays than necessary, reduces angular resolution. Conversely, under-segmentation fails to effectively isolate the SnS birth-death points, negatively impacting CE accuracy.

To address these challenges, the authors propose a dynamic hybrid beamforming-assisted power-based subarray segmentation paradigm (DHBF-PSSP). This paradigm cleverly combines cost-effective power measurements with a dynamic hybrid beamforming (DHBF) architecture for joint subarray partitioning and decoupling. The power-adaptive subarray segmentation (PASS) algorithm utilizes the statistical properties of measured power profiles to dynamically partition the array, ensuring optimal segmentation for the given channel conditions. The DHBF architecture, coupled with subarray segmentation-based group time block code (SS-GTBC), enables efficient subarray decoupling even with limited RF chain resources. Finally, a subarray segmentation-based assorted block sparse Bayesian learning algorithm under the multiple measurement vectors framework (SS-ABSBL-MMV) leverages angular-domain block sparsity and inter-subcarrier structured sparsity to perform CE for each sub-channel. Both on-grid and off-grid versions of the SS-ABSBL-MMV algorithm are developed, employing DFT codebooks to reduce computational complexity.

Extensive simulations validate the exceptional performance of the proposed framework. The DHBF-PSSP significantly improves the performance of existing CE algorithms by optimizing pilot allocation. The SS-ABSBL-MMV algorithm achieves superior normalized mean square error (NMSE) performance compared to baseline algorithms, particularly in low SNR regimes. For instance, with a pilot length of 64 and an SNR of 5dB, the SS-OG-ABSBL-MMV algorithm achieves an NMSE of approximately -13dB, significantly outperforming other benchmark algorithms. The results also demonstrate the robustness of the PASS algorithm to varying power fluctuations and the fast convergence of the SS-ABSBL-MMV algorithm. This work presents a valuable new paradigm for near-field CE in XL-MIMO, paving the way for more efficient and robust communication systems in 6G and beyond.

Conclusion

This newsletter highlighted two impactful papers that address key challenges in wireless communication and sensing. Skog et al.'s work demonstrates the potential of indirectly extracting odometry information within magnetic-field SLAM systems using magnetometer arrays, promising significant improvements for indoor localization. The challenge lies in balancing the need for high-resolution mapping with computational feasibility. Yang et al. tackle the complexities of near-field channel estimation in XL-MIMO systems by introducing adaptive subarray segmentation. Their DHBF-PSSP paradigm and SS-ABSBL-MMV algorithm showcase substantial performance gains, particularly in low SNR environments. Both papers contribute significantly to the advancement of wireless technologies, paving the way for more robust and efficient systems in future 6G networks and beyond. They underscore the importance of innovative signal processing techniques and the power of combining different approaches, such as SLAM and odometry, or hybrid beamforming and sparse Bayesian learning, to achieve significant performance improvements.