General Overview

This collection of preprints explores diverse challenges and advancements in communication and sensing systems, with a notable emphasis on optimization techniques and emerging technologies. Several papers focus on integrated sensing and communication (ISAC), highlighting its potential for spectral efficiency and enhanced performance. Rexhepi et al. (2024) (Rexhepi et al., 2024) tackle peak-to-average power ratio (PAPR) reduction in OFDM-ISAC systems using tone reservation and manifold optimization, achieving comparable PAPR reduction to state-of-the-art methods with lower computational complexity and improved sensing performance. Similarly, Xiu et al. (2024a) (Xiu et al., 2024a) investigate physical-layer security in an ISAC-UAV system, optimizing beamforming and trajectory to maximize the average secrecy rate while maintaining sensing accuracy. Zhao et al. (2024) (Zhao et al., 2024) introduce BUPTCMCC-6G-CMG+, a GBSM-based ISAC channel model simulator, contributing to the standardization and evaluation of ISAC systems. Shi & Liu (2024) (Shi & Liu, 2024) address data association and line-of-sight (LOS) identification for networked device-free sensing in multipath environments, proposing a two-phase localization protocol for improved accuracy in OFDM cellular systems.

Beyond ISAC, several contributions explore advanced signal processing and optimization techniques for diverse applications. Dehkordi & Jabbari (2024) (Dehkordi & Jabbari, 2024) propose a deep Q-network (DQN) based resource allocation framework for UAV-assisted mobile edge computing (MEC), optimizing task offloading and UAV trajectory to balance processed data and communication delay. Lin et al. (2024) (Lin et al., 2024) derive the Bayesian Cramér-Rao Lower Bound (BCRLB) under a probabilistic data association (PDA) fusion framework for multi-target tracking with extended Kalman filter (EKF). Zobova & Krylov (2024) (Zobova & Krylov, 2024) present a novel MEMS resonant accelerometer architecture with switchable electrostatic transmission, enabling self-calibration and sensitivity tuning. Liu et al. (2024) (Liu et al., 2024) introduce a computational ghost imaging system based on low-density parity-check (LDPC) codes, leveraging signal redundancy to improve imaging quality.

Several papers focus on specific applications and challenges within communication systems. Demir & Björnson (2024) (Demir & Björnson, 2024) analyze a user-centric cell-free massive MIMO system with RIS-integrated antenna arrays, demonstrating enhanced data rates through intelligent transmitting surfaces. Xiu et al. (2024b) (Xiu et al., 2024b) propose a robust beamforming design for near-field DMA-NOMA mmWave communications with imperfect position information. Chemingui et al. (2024) (Chemingui et al., 2024) introduce an EMF-aware waveform design for dual-functional radar communication systems, optimizing performance while adhering to electromagnetic field exposure limits. Xiu et al. (2024c) (Xiu et al., 2024c) explore movable antenna enabled ISAC beamforming for low-altitude airborne vehicles, maximizing communication capacity under sensing SNR constraints.

The application of machine learning techniques is prominent across several studies. Tallam Puranam Raghu et al. (2024a) (Tallam Puranam Raghu et al., 2024a) investigate self-supervised representation learning with continuous training data for improved myoelectric control, demonstrating enhanced performance and user experience with VICReg pre-training. He & Chiang (2024) (He & Chiang, 2024) propose TFT-multi, an extension of the temporal fusion transformer (TFT), for simultaneous forecasting of vital sign trajectories in intensive care units. Mehrban (2024) (Mehrban, 2024) examines the use of LDPC codes in cooperative communication for improved reliability. Kuzio et al. (2024) (Kuzio et al., 2024) propose a statistical approach for identifying fault frequency variation in vibration-based local damage detection.

Finally, several papers address specific challenges in diverse domains. Khojastepour et al. (2024) (Khojastepour et al., 2024) analyze the age of gossip in networks with multiple views of a source, deriving recursive expressions for average version age of information. Asgharpoor Golroudbari (2024) (Asgharpoor Golroudbari, 2024) introduces TE-PINN, a transformer-enhanced physics-informed neural network for quaternion-based orientation estimation. Carson et al. (2024) (Carson et al., 2024) investigate interpolation filter design for sample rate independent audio effect RNNs. Leang et al. (2024) (Leang et al., 2024) explore VQ-VAE with prosody parameters for speaker anonymization. Shi et al. (2024) (Shi et al., 2024) present a microwave photonic frequency measurement and time-frequency analysis method achieving high bandwidth and temporal resolution. These diverse contributions collectively advance the state-of-the-art in signal processing, communication, and sensing technologies, offering promising solutions for future wireless systems.

Paper Highlights

RIS-Integrated Antennas Boost Cell-Free Massive MIMO Performance

User-Centric Cell-Free Massive MIMO With RIS-Integrated Antenna Arrays by Özlem Tuğfe Demir, Emil Björnson https://arxiv.org/abs/2409.15765

Caption: This figure presents the cumulative distribution function (CDF) of the spectral efficiency, comparing the RIS-integrated cell-free massive MIMO system with conventional systems. The RIS-optimized system with only 4 antennas per AP outperforms a conventional system with 36 antennas, demonstrating the significant gains achieved by integrating RIS. The plot also shows the performance of a RIS system with random phase shifts, highlighting the importance of the proposed optimization method.

Cell-free massive MIMO is a promising technology for beyond 5G, offering more uniform data rates. However, achieving this often requires numerous access points (APs) with many antennas. This paper proposes a novel architecture integrating a reconfigurable intelligent surface (RIS) into each AP's antenna array. Acting as an intelligent transmitting surface in the uplink, this approach aims to boost data rates cost-effectively by leveraging the RIS's larger aperture, collecting more signal energy with fewer radio frequency chains.

The system model comprises L APs, each with M antennas and an N-element RIS (N > M), serving K single-antenna users. A key assumption is lossless signal transmission through the RIS, achieved by encasing the area between the AP antennas and the RIS with reflectors, creating an "open-box" structure. This setup establishes a theoretical performance benchmark.

The paper focuses on centralized uplink operation using dynamic cooperation clustering (DCC). Channel estimation utilizes pilot sequences, and the RIS employs a long-term phase-shift configuration designed to maximize received signal strength. This configuration is determined by maximizing the following objective function:

$\sum_{k \in D_l} E{||\sqrt{\tau_p p_p} \mathbf{H}_l \mathbf{\Psi}l \mathbf{h}{kl}||^2}$,

where $\tau_p$ represents the number of pilot transmissions, $p_p$ is the pilot transmit power, $\mathbf{H}_l$ represents the channel between AP l and its RIS, $\mathbf{\Psi}l$ is the RIS phase-shift matrix, and $\mathbf{h}{kl}$ is the channel between user k and RIS l. This optimization problem is solved using the power iteration method. An achievable spectral efficiency (SE) expression is derived for this architecture, considering the effective channels created by the RIS and the chosen phase-shift configuration.

Simulation results demonstrate significant SE gains. In a smaller network (L=50 APs, K=10 users), the RIS-integrated system with M=4 antennas achieved a 55% improvement in median SE compared to a conventional system with the same number of antennas but no RIS, even surpassing a conventional system with M=36 antennas. While gains were reduced in larger networks (L=100 APs, K=20 users), they remained substantial, particularly with increased coverage areas accommodating higher AP and user density. The optimized long-term RIS phase-shift configuration consistently outperformed random phase-shift selections. This study showcases the potential of RIS-integrated antenna arrays to enhance cell-free massive MIMO performance, especially with limited active antennas.

Self-Supervised Learning and Dynamic Data Enhance Myoelectric Control

Self-Supervised Representation Learning with Augmentations of Continuous Training Data Improves the Feel and Performance of Myoelectric Control by Shriram Tallam Puranam Raghu, Dawn MacIsaac, Erik Scheme https://arxiv.org/abs/2409.16015

Caption: Box plots comparing the performance of LDA and LSTM classifiers trained with different data types (Ramp-R, Dynamic-D, and VICReg augmented Dynamic-V) across several Fitts' Law metrics. LSTM-V, trained with VICReg augmented dynamic data, consistently outperforms other classifiers, particularly in completion rate, movement time, and instability, highlighting the benefits of self-supervised learning and dynamic data in myoelectric control. The results suggest that using continuous dynamic data and incorporating self-supervised learning techniques like VICReg significantly improves online control performance and user experience.

Myoelectric control, while promising, struggles to translate lab performance to real-world use. This study investigates training myoelectric classifiers with continuous dynamic data, reflecting real-world movements, and employing self-supervised learning. Researchers compared traditional Linear Discriminant Analysis (LDA) and deep learning Long Short-Term Memory (LSTM) classifiers trained with different data: ramp contractions, continuous dynamic contractions, and continuous dynamic data augmented with Variance-Invariance-Covariance Regularization (VICReg). An online Fitts' Law test with 20 participants evaluated usability and effectiveness.

The pattern recognition pipeline involved feature extraction using Low-Sampling Frequency 4 (LSF4) features and Mean Absolute Value (MAV), followed by classification with LDA and LSTM models. Crucially, continuous dynamic data was labeled by incorporating a constant delay based on visual choice reaction time (CRT). Confidence-based rejection (threshold 0.5) mitigated misclassifications. Proportional velocity control, using a sigmoid transfer function f(x) = σ(α(x – x₀)) (α = 10, x₀ = 0.5), allowed modulating cursor speed based on contraction intensity.

Results demonstrated that LSTMs trained with continuous dynamic data significantly outperformed LDA. VICReg pre-training further enhanced online performance and user experience, especially for the LSTM model (LSTM-V). LSTM-V showed large effect size improvements in completion rate (p < 0.02; d > 0.8) and movement time (p < 0.001; d > 0.8) compared to baseline LDA trained with ramp data (LDA-R). All temporal models drastically improved instability (p < 0.0001, d > 1.8) compared to LDA-R. User ratings corroborated this, with LSTM-V receiving the highest scores. This study highlights the importance of continuous dynamic training data for myoelectric classifiers. Temporal models, like LSTMs, effectively leverage this richer data, improving online control. Self-supervised learning techniques like VICReg offer substantial performance and user experience benefits, paving the way for more intuitive prosthetic devices.

Conclusion

This newsletter showcases exciting advancements in communication and sensing systems. The highlighted papers demonstrate the potential of innovative approaches like integrating RIS into antenna arrays for cell-free massive MIMO, significantly boosting spectral efficiency with fewer active antennas. Furthermore, the application of self-supervised learning with dynamic data in myoelectric control showcases a promising avenue for creating more intuitive and user-friendly prosthetic devices. These advancements underscore the ongoing evolution and increasing sophistication of signal processing and communication technologies, paving the way for more efficient, robust, and user-centric systems in the future.