Subject: Wireless Comms, Sensing, and Signal Processing Advancements

Hi Elman,

General Overview

This newsletter explores recent preprints covering diverse challenges and advancements in wireless communication, sensing, and signal processing, with a particular emphasis on enhancing the performance and efficiency of future wireless systems, especially in the context of 6G.

Several papers focus on specific technological advancements. Christen et al. (2025) propose a FIFO scheduling architecture for GRAND, aiming to improve block error rate and address throughput inconsistencies in this code-agnostic decoding method. Sevdiren et al. (2025) analyze the detrimental impact of I/Q imbalance on THz communication, particularly on achievable rates in ultra-wideband systems. The growing field of Integrated Sensing and Communication (ISAC) is represented by Umra et al. (2025), who introduce an RL-driven cognitive MIMO radar for robust multi-target detection in cluttered environments, and Pu et al. (2025), who propose an OTFS-ISAC system with sub-Nyquist ADC sampling for efficient radar estimation and simultaneous communication. Zhang et al. (2025) provide a comprehensive overview of THz-ISAC-empowered UAVs in 6G, focusing on transceiver design and key technologies like hybrid beamforming and waveform design.

Machine learning and deep learning are also prominent themes. Lopez Alcaraz et al. (2025) demonstrate the potential of ECG analysis for neuropsychiatric diagnosis using explainable machine learning. Wang et al. (2025) introduce DULRTC-RME, a deep unrolled low-rank tensor completion network for radio map estimation. Letizia (2025) explores deep learning for physical layer communications, while Ding et al. (2025) propose EEG-PatchFormer, a transformer-based framework for decoding human attentive states from EEG data.

Several contributions focus on specific signal processing and communication techniques. These include investigations into the impact of model mismatch on DOA estimation with MUSIC Emenonye et al. (2025), analysis of UAV-based cell-free massive MIMO R.R et al. (2025), an uplink rate-splitting multiple access (RSMA) scheme for mobile edge computing Xu et al. (2025), an overview of affine frequency division multiplexing (AFDM) Yin et al. (2025), and a novel THz defect detection method Sevdiren et al. (2025).

Further contributions explore advanced signal processing techniques, including analysis of BEM-based channel estimation for OTFS Wu et al. (2025), comparison of pre-optimized irregular arrays (PIA) Irshad et al. (2025), semantic feature division multiple access (SFDMA) Ma et al. (2025), a graph neural network for beamforming (ICGNN) He et al. (2025), a deep learning framework for SWIPT (SWIPTNet) Han et al. (2025), and GNN-enabled fluid antenna systems He et al. (2025).

Finally, specific challenges are addressed, including blind Capon beamforming Koldovský et al. (2025), metamaterial sensor design Hossain & Hannan (2025), UAV cognitive semantic communication Song et al. (2025), time-aware quantization for diffusion transformers (TQ-DiT) Hwang & Lee (2025), pulse shaping for Zak-OTFS Das et al. (2025), an EM channel model for HMIMO Wei et al. (2025), beyond-diagonal RIS (BD-RIS) Khan et al. (2025), and stability analysis of score-driven filters Donker van Heel et al. (2025).

Paper Highlights

Beyond Diagonal RIS: A New Frontier for 6G Internet of Things Networks

Beyond Diagonal RIS: A New Frontier for 6G Internet of Things Networks by Wali Ullah Khan, Chandan Kumar Sheemar, Eva Lagunas, Symeon Chatzinotas https://arxiv.org/abs/2502.03637

Caption: Three BD-RIS architectures are depicted: (a) cell-wise single-connected, where each antenna connects to a single impedance element within its cell; (b) cell-wise fully-connected, with interconnections between all antennas within a cell; and (c) element-wise group-connected, where antennas are grouped and connected through shared impedance elements. These architectures offer varying levels of complexity and control over the electromagnetic wavefront, impacting the performance gains achievable in 6G IoT applications.

Reconfigurable Intelligent Surfaces (RIS) are revolutionizing wireless communications by allowing for programmable control over radio wave propagation. While diagonal RIS (D-RIS) offers some advantages, its limited ability to manipulate waves hinders further performance improvements. This paper explores the emerging concept of beyond-diagonal RIS (BD-RIS), which incorporates non-diagonal elements in its scattering matrix, enabling finer control over electromagnetic wavefronts. The authors discuss the limitations of D-RIS and introduce key BD-RIS architectures: single-connected, fully-connected, and group-connected, along with their operating modes: reflective, transmissive, hybrid, and multi-sector. They analyze the trade-offs between simplicity, control, and scalability offered by these different architectures and modes.

BD-RIS offers several advantages for 6G IoT applications. Enhanced beamforming and flexibility stem from BD-RIS's ability to control both amplitude and phase, unlike D-RIS, which only adjusts phase shifts. This results in improved signal strength and quality, crucial for the diverse connectivity demands of IoT devices. Adaptability to dynamic IoT environments is another key benefit, enabling BD-RIS to reconfigure itself and accommodate changing traffic and mobility patterns. Superior interference management is achieved through joint phase and amplitude control, effectively mitigating interference in complex IoT deployments. BD-RIS also provides extended coverage and connectivity through full-space (360-degree) coverage, which is ideal for large-scale IoT applications. Finally, its versatility makes BD-RIS suitable for a wide range of scenarios, including UAV-assisted IoT and vehicular communications.

A case study on BD-RIS-assisted vehicle-to-vehicle (V2V) communication in an underlay cellular network demonstrates the significant performance gains achievable with this technology. The objective was to maximize the spectral efficiency of V2V communication while maintaining the quality of service (QoS) of the cellular network by jointly optimizing the transmit power of the V2V transmitter and the phase shift design of the BD-RIS. The spectral efficiency maximization problem is formulated as: maximize the rate expression, subject to power constraints and BD-RIS constraints. This problem is non-convex, and an alternating optimization technique is employed to obtain an efficient solution. Simulation results demonstrate that BD-RIS significantly outperforms both D-RIS and conventional systems. With a carrier frequency of 3.5 GHz, cellular RSU transmit power of 10 W, and V2V transmit power of 1 W, the spectral efficiency of BD-RIS-assisted V2V communication with 64 elements was approximately three times higher than that of conventional V2V communication. This highlights the potential of BD-RIS to significantly improve the performance of future wireless networks.

Despite its potential, BD-RIS faces several implementation challenges. Hardware design complexity arises from the intricate impedance networks and the need for large-scale integration. Adaptive channel estimation is crucial but challenging due to the complexity of controlling both diagonal and non-diagonal elements. Non-ideal hardware effects, such as amplifier nonlinearities and phase noise, can degrade performance. Finally, signal detection sensitivity is critical, especially in long-distance, interference-prone IoT environments. Future research directions include integrating AI/ML for autonomous optimization, exploring joint communication and sensing (JCAS) capabilities, and enhancing physical layer security to fully realize the transformative potential of BD-RIS in shaping high-performance 6G IoT networks.

Explainable and externally validated machine learning for neuropsychiatric diagnosis via electrocardiograms

Explainable and externally validated machine learning for neuropsychiatric diagnosis via electrocardiograms by Juan Miguel Lopez Alcaraz, Ebenezer Oloyede, David Taylor, Wilhelm Haverkamp, Nils Strodthoff https://arxiv.org/abs/2502.04918

Caption: This diagram illustrates the study's methodology, which involves training a machine learning model on ECG features and demographics from the MIMIC-IV-ECG database and validating it on the ECG-View II database for neuropsychiatric diagnoses. The diagram also highlights the use of both datasets' demographics and ECG features, including intervals like PR, ST, and QT, as input for the model.

Electrocardiograms (ECGs), traditionally used for cardiac assessments, are emerging as a promising tool for diagnosing and monitoring neuropsychiatric conditions. This burgeoning field of research leverages the interconnectedness of the cardiovascular and neuropsychiatric systems, exploring how subtle ECG changes can reflect underlying neurological and psychiatric disorders. While the precise mechanisms linking these systems remain under investigation, preliminary findings suggest that autonomic nervous system dysfunction, neurodegenerative processes, vascular alterations, and medication effects can manifest as detectable ECG abnormalities.

This study utilized machine learning models trained on ECG markers and demographic data from the MIMIC-IV-ECG database to predict neuropsychiatric conditions based on ICD-10 codes. The models were then externally validated using the ECG-VIEW-II dataset. The study focused on conditions with internal and external Area Under the Receiver Operating Characteristic Curve (AUROC) scores exceeding 0.7. Explainability was incorporated using Shapley values to understand the contribution of individual ECG features to the predictions.

The results demonstrated significant predictive performance for several neurological and psychiatric conditions. For Alzheimer's disease (G30), the model achieved an internal AUROC of 0.813 (95% CI: 0.812-0.814) and an external AUROC of 0.868 (95% CI: 0.867-0.868). For unspecified dementia (F03), the model performed even better, with an internal AUROC of 0.849 (95% CI: 0.848-0.849) and an external AUROC of 0.862 (95% CI: 0.861-0.863). Age emerged as a consistently important predictive feature, with older patients contributing more to most conditions. Specific ECG features, such as QTc interval, T-wave axis, RR-interval, and PR-interval, also played significant roles in predicting different conditions.

These findings confirm existing knowledge about ECG abnormalities in neuropsychiatric conditions while also offering new insights. For instance, the study corroborated previous findings of low QRS duration and low PR interval being associated with Alzheimer's disease. However, it also revealed potentially novel markers, such as high RR interval in Alzheimer's, which could be related to parasympathetic dysfunction. The study acknowledges the potential confounding effects of medications used to treat these conditions, which can also induce cardiac changes. Future research should focus on disentangling these effects and exploring the causal relationships between ECG features and neuropsychiatric disorders. This research paves the way for integrating ECGs into routine clinical practice for the diagnosis, monitoring, and personalized management of neuropsychiatric conditions, potentially transforming both research and clinical practices.

Affine Frequency Division Multiplexing: Extending OFDM for Scenario-Flexibility and Resilience

Affine Frequency Division Multiplexing: Extending OFDM for Scenario-Flexibility and Resilience by Haoran Yin, Yanqun Tang, Ali Bemani, Marios Kountouris, Yu Zhou, Xingyao Zhang, Yuqing Liu, Gaojie Chen, Kai Yang, Fan Liu, Christos Masouros, Shuangyang Li, Giuseppe Caire, Pei Xiao https://arxiv.org/abs/2502.04735

Caption: This figure illustrates the time-frequency resource allocation for different waveforms, showcasing the transition from Single-Carrier Modulation (SCM) to Orthogonal Frequency Division Multiplexing (OFDM) through Affine Frequency Division Multiplexing (AFDM) and Orthogonal Chirp Division Multiplexing (OCDM). The varying slopes of the lines represent different chirp rates (controlled by parameter c₁), highlighting AFDM's flexibility in adapting to channel conditions, with OFDM as a special case where the chirp rate is zero. The different colors represent different symbols/subcarriers within a given waveform.

Next-generation wireless networks face a critical challenge: dynamic, high-mobility environments like V2V, UAV, and satellite communications, which create severe Doppler spreads. These spreads negatively impact the widely-used OFDM waveform, causing destructive inter-carrier interference (ICI) and degrading performance. Traditional solutions struggle to estimate and compensate for this interference in doubly dispersive channels (DDC). This article explores Affine Frequency Division Multiplexing (AFDM), a promising chirp-based waveform designed to address these challenges and offer the flexibility needed for diverse next-generation applications.

AFDM uses the discrete affine Fourier transform (DAFT), $s[n] = \sum_{m=0}^{N-1} x[m]e^{j2\pi n (\frac{f_c}{N}m + c_1\frac{m^2}{N} + c_2\frac{n^2}{N})}$, to modulate information symbols onto orthogonal chirp subcarriers. Unlike OFDM, AFDM's chirp subcarriers retain their characteristics even under delay and Doppler shifts, enabling the system to effectively estimate the channel's delay-Doppler profile. The waveform's two tunable parameters, c₁ and c₂, provide flexibility for various applications. c₁ controls the digital chirp rate and influences the time-frequency distribution of the signal, while c₂ affects the initial phase of each subcarrier and can be used to enhance performance or even as a dynamic secret key for security. Importantly, AFDM offers backward compatibility with OFDM, simplifying future network upgrades.

The article explores several promising applications of AFDM, including space-air-ground integrated networks (SAGIN), underwater acoustic communications, high-frequency band communications, and secure communications. AFDM's resilience to Doppler spread makes it ideal for the challenging channels encountered in these scenarios. Furthermore, the flexibility offered by c₁ and c₂ allows for tailored system design to meet the diverse requirements of different applications. For example, in SAGIN, AFDM can accommodate the varying channel conditions across different network tiers, while in secure communications, c₂ can be used for dynamic encryption.

Despite its potential, AFDM faces several challenges. Channel estimation in highly dynamic environments remains complex, although embedded pilot-aided (EPA) techniques and direct estimation of the effective channel matrix (ECM) offer promising solutions. Pulse shaping is crucial to mitigate symbol spreading caused by fractional delay and Doppler shifts. The article shows that using Hamming or Dolph-Chebyshev windows instead of rectangular pulse shaping significantly improves channel estimation accuracy. Efficient signal detection algorithms are also essential, with sequence-wise detectors like zero-forcing (ZF), linear minimum mean squared error (LMMSE), and message passing (MP) offering a trade-off between performance and complexity. For example, in a scenario with N=512 and six paths with a maximum delay of 3Δt and a maximum Doppler of 2Δf, AFDM with MP detection achieves significantly better BER performance compared to OFDM, OCDM, and OTFS. Finally, optimizing MIMO and multi-user access techniques for AFDM is crucial for achieving high spectral efficiency and supporting massive connectivity. The article suggests orthogonal resource allocation in the DAFT domain as a potential solution for AFDMA systems.

Conclusion

This newsletter highlights the ongoing research pushing the boundaries of wireless communication, sensing, and signal processing. From novel hardware solutions like BD-RIS to innovative waveform designs like AFDM and the application of machine learning in diagnostics and channel estimation, the field is rapidly evolving to meet the demands of future 6G networks and beyond. The highlighted papers showcase not only the potential advancements but also the challenges that remain, paving the way for further research and development in these critical areas. The convergence of hardware innovation, advanced signal processing, and intelligent algorithms promises to transform the wireless landscape, enabling a wide range of applications from enhanced IoT connectivity to improved healthcare diagnostics.