General Overview

This collection of preprints explores diverse advancements in wireless communications, sensing, and signal processing. Several works focus on reconfigurable intelligent surfaces (RIS), demonstrating their potential in various applications. Ramezani, Girnyk, and Björnson (2024) Ramezani et al. (2024) investigate dual-polarized RIS for cell-specific transmission, proposing Golay complementary array pairs for broad-beam reflection. Demir and Björnson (2024) contribute two papers: one on RIS-assisted integrated sensing and communication (ISAC) for mono-static target detection (Demir & Björnson, 2024), deriving a generalized likelihood ratio test (GLRT) detector and joint optimization of RIS phase-shifts and precoding; and another on point-to-point MIMO channel estimation (Demir & Björnson, 2024), introducing reduced-subspace least squares (RS-LS) estimators that exploit array geometry and clustered multipath propagation. Amiriara et al. (2024) analyze the outage performance of RIS-assisted device-to-device (D2D) communication in the presence of interference, employing a deep neural network (DNN) for real-time outage probability prediction (Amiriara et al., 2024). Finally, Ni et al. (2024) propose architectures integrating multi-functional RIS (MF-RIS) for enhanced communication and sensing in 6G networks, exploring multi-objective optimization for balancing performance trade-offs (Ni et al., 2024).

Beyond RIS, several papers address specific communication challenges and emerging technologies. Zheng et al. (2024) propose a two-timescale design for movable antenna (MA) enabled MU-MIMO systems, optimizing antenna positions and beamforming vectors for improved ergodic sum rate (Zheng et al., 2024). Jere et al. (2024) explore explainable AI (xAI) for MIMO receive processing using reservoir computing (RC), configuring RNN weights based on domain knowledge (Jere et al., 2024). Sivadevuni et al. (2024) investigate energy efficiency optimization with rate-splitting multiple access (RSMA), balancing low and high QoS requirements (Sivadevuni et al., 2024). Several contributions focus on specific applications, including an electronics-free passive ultrasonic communication link for deep-tissue sensor implants (Yener et al., 2024), balanced space- and time-based duty-cycle scheduling for light-based IoT (Botirov et al., 2024), and a mobile IoT device for BPM monitoring (Chuquimarca et al., 2024).

The theme of enhancing communication and sensing performance through advanced signal processing techniques is prevalent. Jiang et al. (2024) analyze the performance of local partial MMSE precoding in user-centric cell-free massive MIMO systems (Jiang et al., 2024). Tominaga et al. (2024) evaluate the spectral efficiency of D-MIMO networks under Rician fading (Tominaga et al., 2024). Bhola et al. (2024) introduce a cooperative UAV-relay based framework for satellite aerial ground integrated networks (Bhola et al., 2024). Han et al. (2024) analyze the performance of pilot-aided simultaneous communication and tracking with multiple drones (Han et al., 2024). Smida et al. (2024) discuss the challenges and methods for in-band full-duplex MIMO systems in ISAC (Smida et al., 2024). Liu et al. (2024) propose a cooperative multi-target positioning scheme for cell-free massive MIMO using multi-agent reinforcement learning (Liu et al., 2024). Ju et al. (2024) introduce a transformer-assisted parametric CSI feedback technique for mmWave massive MIMO systems (Ju et al., 2024). Meng et al. (2024) analyze the performance of network-level ISAC and optimal antenna-to-BS allocation (Meng et al., 2024).

Several papers explore the application of deep learning in various contexts. Xu et al. (2024) advocate for diffusion models in mobile communications, showcasing their potential for channel generation and communication management (Xu et al., 2024). Wei et al. (2024) present a new architecture for neural-enhanced multi-object tracking (Wei et al., 2024). Zhang et al. (2024) propose filtered randomized smoothing for robust modulation classification (Zhang et al., 2024). Chege et al. (2024) introduce a Bayesian framework for PMF tensor estimation and rank detection (Chege et al., 2024). Chien et al. (2024) utilize the differential evolution algorithm for joint active and passive beamforming design in RIS-assisted MIMO systems (Chien et al., 2024). Cummins et al. (2024) investigate spectrally efficient LDPC codes for IRIG-106 waveforms via random puncturing (Cummins et al., 2024). Kim et al. (2024) propose privacy-enhanced over-the-air federated learning via client-driven power balancing (Kim et al., 2024). Shi et al. (2024) derive an improved PCRLB for radar tracking in clutter, considering geometry-dependent target measurement uncertainty (Shi et al., 2024).

Finally, several works address fundamental aspects of signal processing and communication theory. Traonmilin, Aujol, and Guennec (2024) study optimal algorithms for low-dimensional model recovery with linear rates (Traonmilin et al., 2024). Kalla et al. (2024) propose a method for mitigating polarization-induced fading in optical vector network analyzers (Kalla et al., 2024). Xu et al. (2024) present a physics-based perspective for understanding and utilizing spatial resources of wireless channels (Xu et al., 2024). Shiraishi et al. (2024) introduce content-based wake-up for energy-efficient IoT sensing data retrieval (Shiraishi et al., 2024). These diverse contributions collectively advance the state-of-the-art in wireless communications and signal processing, offering valuable insights and novel methodologies for addressing the challenges of future networks.

Paper Highlights

Edge-Guided Inverse Design of Digital Metamaterials for Ultra-High-Capacity On-Chip Multi-Dimensional Interconnect

Edge-guided inverse design of digital metamaterials for ultra-high-capacity on-chip multi-dimensional interconnect by Aolong Sun, Sizhe Xing, Xuyu Deng, Ruoyu Shen, An Yan, Fangchen Hu, Yuqin Yuan, Boyu Dong, Junhao Zhao, Ouhan Huang, Ziwei Li, Jianyang Shi, Yingjun Zhou, Chao Shen, Yiheng Zhao, Bingzhou Hong, Wei Chu, Junwen Zhang, Haiwen Cai, Nan Chi https://arxiv.org/abs/2410.07572

The demand for faster data transmission in compute-intensive applications like AI is driving the need for innovative optical interconnect technologies. This paper presents a breakthrough in on-chip optical interconnects, leveraging the multi-dimensional nature of light to achieve record-breaking data transmission rates. The authors utilize ultra-broadband, multi-mode digital metamaterials designed using a novel edge-guided analog-and-digital optimization (EG-ADO) method.

This EG-ADO method combines the computational efficiency of topology optimization with the fabrication reliability of digital metamaterials. The process begins with topology optimization, utilizing the adjoint method to determine the permittivity distribution within the metamaterial design region. This analog design is then translated into a digital format using edge detection algorithms, addressing the manufacturing challenges often associated with analog metamaterials. A customized direct binary search algorithm further refines the design, ensuring compatibility with standard foundry processes and a minimum feature size of 120 nm. This approach drastically reduces computational complexity compared to traditional methods, particularly for higher-order mode multiplexers.

The researchers designed and fabricated four- and five-mode multiplexers (MUXs) on a silicon-on-insulator (SOI) platform. The five-mode MUX achieved impressive results, with a single-wavelength interconnect capacity of 1.62 Tbit s⁻¹ and a record per-wavelength per-mode channel rate of 324 Gb s⁻¹ using 108 GBaud PAM-8 signals. By combining mode-division multiplexing (MDM) with dense wavelength-division multiplexing (DWDM) across 88 channels in the C-band, they achieved a remarkable multi-dimensional interconnect capacity of 38.2 Tbit s⁻¹, with a spectral efficiency of 8.68 bits s⁻¹Hz⁻¹. The objective function used for optimization is given by:

L = 1 - (1/M) * Σᵢ₌₁ᴹ (Tᵢ(λ)/√Δλ)ᶠ

where M is the number of modes, Tᵢ(λ) is the conversion efficiency for each mode, Δλ is the wavelength range, and ||·||ᶠ is the Frobenius norm.

The EG-ADO method offers substantial advantages over existing methods for designing high-order mode MUXs. It addresses the fabrication challenges of analog metamaterials while maintaining high computational efficiency. The resulting devices exhibit low loss, low crosstalk, and broad bandwidth, making them ideal for high-capacity on-chip interconnects. The researchers suggest that further system optimizations, such as using wavelength and polarization-insensitive edge couplers and expanding the DWDM channel count, could enable on-chip interconnect capacities to reach 0.218 Pb s⁻¹. This work signifies a major step forward in on-chip optical interconnect technology, paving the way for next-generation data centers and optical compute interconnects.

Towards xAI: Configuring RNN Weights using Domain Knowledge for MIMO Receive Processing

Towards xAI: Configuring RNN Weights using Domain Knowledge for MIMO Receive Processing by Shashank Jere, Lizhong Zheng, Karim Said, Lingjia Liu https://arxiv.org/abs/2410.07072

Caption: This paper introduces a novel approach for configuring Reservoir Computing (RC) networks for MIMO receive processing by incorporating channel statistics directly into the untrained RNN weights. This method leverages domain knowledge to improve performance and explainability, offering a signal processing-based understanding of RC's operation in the context of MIMO-OFDM systems. The proposed technique is validated through simulations, demonstrating significant performance gains compared to randomly initialized RC networks.

While deep learning has demonstrated impressive empirical performance in wireless communication tasks like MIMO receive processing, understanding the reasons behind this performance remains a challenge. This paper addresses this "black box" problem by exploring Explainable AI (xAI) within the physical layer, focusing on MIMO-OFDM receive processing using Reservoir Computing (RC), a type of Recurrent Neural Network (RNN).

The authors build upon previous work demonstrating RC's superior performance compared to traditional and other learning-based MIMO detectors. They delve deeper into the why by providing a first-principles understanding of RC's operation, grounded in signal processing principles. The key innovation lies in systematically incorporating domain knowledge, specifically channel statistics, into the design of the untrained RNN weights. This approach moves beyond relying solely on data-driven learning and allows for a more informed and explainable design process.

The paper extends previous frequency-domain weight configuration methods to a novel time-domain approach, particularly suited for minimum-phase channels. A significant contribution is the theoretical explanation linking symbol detection over non-minimum-phase channels to the Windowed Echo State Network (WESN) architecture. This justification for the WESN structure, incorporating input windowing and output skip/delay connections, enhances its explainability. The time-domain weight configuration leverages the Toeplitz structure of the channel matrix and aims to minimize the expected squared error between transmitted and equalized signals. This optimization problem, denoted as P, is formulated as:

min E [||XHΨ(XHΨ)†x - x||²], where H is the normalized channel matrix, Ψ represents the impulse responses of the reservoir neurons, and x is the transmitted signal.

To tackle this complex optimization problem, the authors propose a PCA-based weight configuration method. This transforms P into a more manageable dimensionality reduction problem P* that can be solved using PCA. The solution of P* yields the optimal basis matrix F*, which is then used to configure the untrained RNN weights. The paper also provides theoretical performance guarantees by deriving an upper bound on the approximation error achieved by using F* in a related optimization problem P2. This bound offers insights into the trade-off between the number of neurons in the RNN and the resulting approximation error.

Extensive simulations validate the proposed weight configuration methods for both OFDM and 4x4 MIMO-OFDM systems under 5G/5G-Advanced scenarios. Results demonstrate significant performance improvements for configured WESNs compared to randomly generated weights, particularly at high SNRs where randomly initialized networks encounter an error floor. In strictly minimum-phase CDL-D channels, both frequency and time-domain configuration methods achieve similar BER performance improvements. Similar improvements, albeit smaller, are observed for mixed-phase channels, suggesting the potential need for further configuration or fine-tuning of feedforward weights to handle the non-minimum-phase component. For 4x4 MIMO-OFDM under the 3GPP SCM with CDL-D and CDL-E LOS PDPs, configured weights effectively mitigate the error floor seen with random weights. These results demonstrate the practical benefits of incorporating domain knowledge into RNN design for improved performance and explainability in MIMO receive processing.

Two Birds With One Stone: Enhancing Communication and Sensing via Multi-Functional RIS

Two Birds With One Stone: Enhancing Communication and Sensing via Multi-Functional RIS by Wanli Ni, Wen Wang, Ailing Zheng, Peng Wang, Changsheng You, Yonina C. Eldar, Dusit Niyato, Robert Schober https://arxiv.org/abs/2410.06584

Caption: (a) Depicts a separated antenna deployment for single-objective optimization, focusing on either communication or sensing individually. (b) Illustrates a shared antenna deployment with an MF-RIS, enabling multi-objective optimization for both communication and sensing in an ISAC system. (c) Shows a Pareto front diagram, visualizing the trade-off between maximizing communication rate (f1(x)) and sensing SINR (f2(x)) in multi-objective optimization.

The advent of 6G necessitates advancements in both data transmission and sensing capabilities. Integrated sensing and communication (ISAC) presents a promising approach by sharing resources, but faces challenges in range and beam management. This paper proposes leveraging multi-functional reconfigurable intelligent surfaces (MF-RISs) to overcome these limitations and enhance both communication and sensing performance in 6G networks. MF-RISs, unlike conventional RISs, offer simultaneous signal reflection, refraction, and amplification, enabling full-space coverage and mitigating the double-fading effect.

The paper explores MF-RIS applications across various communication modes (unicast, multicast, and broadcast) and multiple access schemes (OFDMA, NOMA, and RSMA). The dynamic adjustment of amplitude and phase coefficients by MF-RISs allows for strengthening received signal power and mitigating interference, resulting in improved data rates and spectral efficiency. In unicast communication, MF-RISs enhance individual user experience, while in multicast and broadcast scenarios, they improve coverage and reduce network congestion. Integrating MF-RISs with different multiple access schemes provides further performance gains, capitalizing on their ability to manipulate signals in different domains (frequency for OFDMA, power for NOMA, and both for RSMA).

Beyond communication, MF-RISs contribute significantly to wireless sensing, including target detection and localization. By creating virtual line-of-sight (LoS) links and amplifying echo signals, MF-RISs extend sensing range and accuracy, even in challenging non-line-of-sight (NLoS) environments. The paper highlights the synergistic benefits of integrating MF-RISs with ISAC, presenting MF-RIS-aided radar-communication coexistence (RCC) and dual-functional radar-communication (DFRC) systems. In RCC, MF-RISs manage interference between independent sensing and communication signals, whereas in DFRC, they optimize a shared signal for both functionalities.

The paper emphasizes the crucial role of multi-objective optimization in MF-RIS-aided ISAC systems. Given the conflicting objectives of maximizing communication throughput and minimizing sensing error, single-objective optimization is inadequate. Multi-objective optimization, employing techniques like Pareto optimization, enables finding the optimal trade-off between these competing goals. Numerical results demonstrate the superiority of MF-RIS-aided ISAC over benchmarks utilizing passive RIS, STAR-RIS, and active RIS, showing substantial improvements in both communication quality-of-service (QoS) and sensing signal-to-interference-plus-noise ratio (SINR). For instance, as the QoS requirement for communication increases, the sum SINR for sensing decreases less significantly for MF-RIS-aided ISAC compared to passive RIS-aided ISAC. Similarly, increasing the number of RIS elements leads to a more substantial improvement in the sum SINR for MF-RIS-aided ISAC compared to the passive RIS counterpart. The paper concludes by outlining future research directions, including MF-RIS-aided near-field ISAC, AI-based beamforming design, and the integration of sensing, communication, and computing.

Conclusion

This newsletter highlights the ongoing research pushing the boundaries of wireless communication, sensing, and signal processing. The exploration of reconfigurable intelligent surfaces (RIS), particularly multi-functional RIS (MF-RIS), demonstrates the potential for significant performance gains in both communication and sensing, particularly within the context of integrated sensing and communication (ISAC). The development of novel optimization techniques, such as the edge-guided analog-and-digital optimization (EG-ADO) method for metamaterial design, and the incorporation of domain knowledge into deep learning models, as seen in the application of reservoir computing (RC) for MIMO receive processing, are key themes. These advancements pave the way for higher data rates, improved sensing accuracy, and more efficient resource utilization in future 6G networks. The continued exploration of these technologies, along with the integration of deep learning and other advanced signal processing techniques, promises to unlock even greater potential in the years to come.