General Overview

This collection of papers explores diverse advancements in signal processing, communication systems, and machine learning applications, with a particular focus on enhancing future wireless networks. Several works target improved performance and efficiency. For example, Azim et al. (2024) Azim et al. (2024) statistically characterized the radar cross section (RCS) of various indoor factory targets, finding that the Lognormal distribution provides a good fit across all considered objects. This work has implications for radar-based sensing in industrial environments. Meanwhile, Shah et al. (2024) Shah et al. (2024) proposed a Newtonized Orthogonal Matching Pursuit (NOMP) algorithm for high-resolution target detection in sparse OFDM ISAC systems, demonstrating superior performance compared to existing compressed sensing methods. This contribution advances the state-of-the-art in integrated sensing and communication (ISAC). Askerbeyli et al. (2024) Askerbeyli et al. (2024) tackled sum rate maximization in constant envelope MIMO downlink with the RZF precoder, developing power allocation algorithms that significantly improve performance while considering constant envelope quantization. This work addresses practical hardware constraints in MIMO systems.

Another prominent theme is the application of advanced signal processing techniques to real-world challenges. Le et al. (2024) Le et al. (2024) introduced a multi-scale temporal analysis for failure prediction in energy systems using PMU data, demonstrating the superiority of multi-scale analysis over single-window models. This work has significant implications for the reliability of power grids. Elbir et al. (2024) Elbir et al. (2024) presented HANDBALL, a hybrid beamforming design for ISAC with low-resolution DACs, achieving satisfactory sensing and communication performance even with one-bit and few-bit designs. This contribution addresses the growing need for hardware-efficient ISAC systems. Two papers by Monemi et al. (2024) (Monemi et al. (2024) and Monemi et al. (2024)) revisited the Fraunhofer and Fresnel boundaries for phased array antennas, providing a more accurate characterization of these boundaries than traditional single-element antenna models. This work refines our understanding of phased array antenna behavior.

Machine learning, particularly deep learning, plays a key role in several contributions. Tao et al. (2024) Tao et al. (2024) proposed an LLM-based framework for bearing fault diagnosis, demonstrating improved generalization capabilities compared to traditional methods. Meng et al. (2024) Meng et al. (2024) introduced Spectro-Spatial Covariance Vectors (SSCV) and a novel 3D convolutional network, FOA-Conv3D, for blind estimation of acoustic parameters from Ambisonics recordings. Torre-Cruz et al. (2024) Torre-Cruz et al. (2024) developed an unsupervised method for heartbeat detection and classification in PCG signals using the dissimilarity matrix, achieving robust performance in noisy environments.

Further advancements include contributions to theoretical frameworks and practical system designs. Tadipatri et al. (2024) Tadipatri et al. (2024) proposed a convex relaxation approach for generalization analysis of parallel positively homogeneous networks. Bullo et al. (2024) Bullo et al. (2024) investigated energy-aware dynamic neural inference for energy-harvesting devices. Liu et al. (2024) Liu et al. (2024) introduced a user-centric semantic communication system.

Finally, several papers address specific challenges in diverse applications. Park et al. (2024) Park et al. (2024) proposed a hybrid TDMA-NOMA-cooperative transmission strategy for multi-UAV collaborative sensing. Ge et al. (2024) Ge et al. (2024) tackled target handover in distributed ISAC systems. Crilly Jr (2024) Crilly Jr (2024) proposed a signal discovery method for interstellar communication.

Paper Highlights

User-Centric Semantic Communications

User Centric Semantic Communications by Xunze Liu, Yifei Sun, Zhaorui Wang, Lizhao You, Haoyuan Pan, Fangxin Wang, Shuguang Cui https://arxiv.org/abs/2411.03127

Caption: This diagram illustrates the user-centric semantic communication system. The system consists of the user side, the transmitter side (which includes Task Planning and Reflection, Task Execution, and Frame Selection), where user requests drive the process of task planning, execution, and frame selection to deliver relevant semantic information. If the task plan cannot fulfill the user's request, the system selects and transmits the most relevant video frames.

Current semantic communication systems prioritize efficient data compression, often overlooking the user's specific needs. These systems typically extract semantic information based on predefined criteria at the transmitter, such as PSNR in image transmission, without truly understanding what the user wants. This can lead to transmitting irrelevant information, rendering the communication process inefficient and ultimately meaningless from the user's perspective. Think of a scenario where a user wants to extract a specific detail from a video, like a license plate number. Traditional semantic communication might compress the video to preserve the overall scene but lose the crucial detail the user requires.

This paper proposes a paradigm shift – a user-centric semantic communication system. This system puts the user in control. The communication begins with the user sending a specific request for the semantic information they need. The transmitter, equipped with a large language model (LLM) like GPT-4 and specialized AI tools, interprets this request. The LLM and tools work together – the LLM understands the user's intent and formulates a plan of action, breaking down the request into smaller tasks that the specialized tools can execute. A key innovation is the task reflection process. This process verifies if the planned actions can actually fulfill the user's request. If not, the system tries alternative plans, or, as a last resort, selects the video frames most relevant to the user's request and transmits those.

The effectiveness of this user-centric approach was demonstrated using traffic surveillance videos and a diverse dataset of 700 user requests. The system achieved a remarkable 83.90% success rate in fulfilling user requests. More impressively, it achieved this while significantly reducing resource consumption. Compared to traditional systems that transmit entire videos, the user-centric system reduced the number of transmitted frames by 81.70% and the overall data size by 66.33%. This showcases the potential for substantial gains in efficiency without sacrificing the relevance of the transmitted information. The higher success rate for fulfillable requests compared to unfulfillable ones highlights the need for further refinement in the frame selection process when the system cannot directly fulfill the user's request. The relatively lower reduction in data size compared to frame count is attributed to the nature of video compression, where individual frames still carry substantial data. Future work will focus on improving frame selection and addressing this data size challenge.

Diffusion-based Generative Multicasting with Intent-aware Semantic Decomposition

Diffusion-based Generative Multicasting with Intent-aware Semantic Decomposition by Xinkai Liu, Mahdi Boloursaz Mashhadi, Li Qiao, Yi Ma, Rahim Tafazolli, Mehdi Bennis https://arxiv.org/abs/2411.02334

Caption: This diagram illustrates the intent-aware generative semantic multicasting framework. The transmitter decomposes the source signal into semantic classes and transmits only the intended classes to each user, along with a compressed semantic map. Users then leverage generative diffusion models to synthesize the non-intended classes locally, minimizing resource usage and enhancing privacy.

This paper introduces a groundbreaking approach to semantic multicasting, leveraging the recent advancements in generative diffusion models (GDMs). Semantic communication (SemCom) is becoming increasingly important for applications like the metaverse and XR/MR, where efficient multicasting of rich multimodal signals is essential. Traditional multicasting approaches send the entire signal to all users, regardless of their individual needs, leading to wasted resources and potential privacy concerns. This paper proposes an intent-aware generative semantic multicasting framework that addresses these limitations.

The core idea is to decompose the source signal (e.g., an image) into distinct semantic classes (e.g., "car," "building," "sky") based on each user's specific interests. The transmitter then sends only the intended classes to each user over orthogonal wireless resources, minimizing unnecessary data transmission. Crucially, a highly compressed semantic map, containing information about all classes, is multicasted to all users over shared resources. Users then utilize pre-trained GDMs to locally synthesize the non-intended classes using this semantic map. This results in a signal at each receiver that is partially reconstructed from the transmitted intended classes and partially synthesized from the semantic map.

This approach offers significant advantages. First, it dramatically improves resource utilization by transmitting only the necessary information to each user. Second, it enhances privacy by avoiding the transmission of non-intended details to users who don't require them. This eliminates the need for separate anonymization techniques, streamlining the process. The framework employs a deep dual-resolution network (DDRNet) for semantic segmentation at the transmitter and semantic diffusion models (SDM) with classifier-free semantic guidance for synthesis at the receivers. To further optimize performance, a communication/computation-aware scheme dynamically adapts transmission power and compression rate for each class, minimizing overall latency while meeting user-specific reconstruction/synthesis quality requirements. The latency at user k is given by:

T_k = max {T^g + T_ko(p₀, r₀), [I]_klT_kl(p_l, r_l), l ∈ [L]}

where T^g is the generation time, T_ko is the latency of the semantic map, T_kl is the latency of intended class l, p represents power, r represents compression rate, and I is the intent matrix.

Simulations using the Cityscapes dataset showcase the framework's effectiveness. Compared to traditional multicasting, the proposed method significantly reduces per-user latency while maintaining high perceptual quality. For a scenario with 10 users receiving street scene images, the framework achieved a 15.4% reduction in per-user latency at a fixed power budget, or a 50% reduction in transmission power for a fixed latency target, compared to non-generative multicasting. The framework also scales well to larger numbers of users and adapts to varying generation latency, quality requirements, and overlapping user intents. This framework holds immense promise for efficient and privacy-preserving multicasting in future wireless networks, particularly for immersive applications like the metaverse and XR/MR, extending beyond images to video and 3D content.

From Theory to Reality: A Design Framework for Integrated Communication and Computing Receivers

From Theory to Reality: A Design Framework for Integrated Communication and Computing Receivers by Kuranage Roche Rayan Ranasinghe, Kengo Ando, Giuseppe Thadeu Freitas de Abreu https://arxiv.org/abs/2411.02016

Caption: BER and MSE Performance of the Proposed ICC Algorithm

Integrated Communication and Computing (ICC), or over-the-air computing (AirComp), is a key concept for 6G, aiming to seamlessly integrate computation with communication. While theoretical studies and beamforming designs for AirComp have progressed, practical receiver designs capable of simultaneous data detection and computation remain a challenge. This paper presents a novel, flexible, and scalable framework for designing such receivers, leveraging the power of Gaussian Belief Propagation (GaBP).

Existing approaches often struggle with generalization to different computation tasks, scalability to large numbers of devices, and supporting multiple computation streams. This framework addresses these limitations by providing a system generalizable to any nomographic function, scalable to massive numbers of user equipments (UEs) and edge devices (EDs), and capable of supporting multiple computation streams. The system considers an uplink multi-user SIMO scenario where UEs/EDs transmit both communication symbols (d) and computing signals (s) to a multi-antenna base station (BS)/access point (AP). The received signal y at the BS/AP is:

y = Hx + w = H(d + s) + w

where H is the channel matrix, x is the combined transmit signal, and w is additive white Gaussian noise. The receiver aims to detect the communication symbols d and minimize distortion over the computing signals s, ultimately estimating the target function f(s). In this case, the target function is the sum of individual computing signals:

f(s) = Σₖ₌₁ᴷ sₖ

The receiver design utilizes the GaBP framework, which employs element-wise scalar operations, making it inherently scalable. The GaBP algorithm iteratively refines estimates of the data and computing symbols through message passing between factor nodes (representing the received signal at each antenna) and variable nodes (representing the transmitted symbols from each UE/ED). The algorithm is initialized with a closed-form solution to an optimal combiner design problem with successive interference cancellation (SIC).

Simulations demonstrate the framework's effectiveness under various loading conditions (underloaded, fully loaded, and overloaded) with up to 200 antennas and 200 UEs/EDs. Impressively, the framework achieves near-theoretical performance in underloaded and fully loaded scenarios. Furthermore, it's remarkably efficient, requiring only a small fraction (1%) of the total transmit power for the computing signals, effectively achieving computation "for free" without significantly impacting communication performance. For instance, in a massive MIMO setup with 200 antennas and varying numbers of users, the proposed method achieves BER performance close to the theoretical linear bound, particularly in underloaded and fully loaded cases. This work represents a significant advancement towards practical ICC receiver design, paving the way for future research into more complex computation tasks, integration with sensing (ISCC), and enhanced performance in overloaded scenarios.

Conclusion

This newsletter highlights a collection of papers that significantly advance the state-of-the-art in signal processing, communication, and machine learning. A common thread throughout is the focus on efficiency and practicality. The user-centric semantic communication framework reimagines how we approach data transmission, prioritizing user needs and dramatically reducing resource consumption. The generative multicasting framework leverages the power of generative AI to achieve efficient and privacy-preserving multicasting, opening doors for immersive applications like the metaverse. Finally, the novel ICC receiver design provides a practical and scalable solution for integrating computation with communication, bringing the promise of AirComp closer to reality. These advancements collectively represent significant steps towards more intelligent, efficient, and user-focused communication and computing systems.