Subject: Cutting-Edge Advancements in Wireless Communications for 6G and Beyond

Hi Elman,

This newsletter dives into the latest breakthroughs in wireless communications, covering everything from fundamental signal processing techniques to sophisticated system designs for 6G and beyond.

General Overview

This collection of papers explores cutting-edge advancements in wireless communications, spanning from fundamental signal processing techniques to sophisticated system designs for 6G and beyond. Several papers focus on enhancing the performance and efficiency of over-the-air (OTA) computation and communication systems. Li et al. (2025) introduce a channel-aware constellation design for digital OTA computation, dynamically adapting to channel conditions for improved reliability and reduced power consumption (Li et al., 2025). Complementing this, Tang et al. (2025) propose a semantic analog aggregation method for low-complexity remote state estimation over MIMO channels, eliminating the need for CSI and reducing computational burden through a constant-gain filtering algorithm optimized via CSSCA (Tang et al., 2025). Meanwhile, Ortin et al. (2025) present a channel-independent precoder for OFDM systems, leveraging information redistribution and a modified ZF equalizer to mitigate the effects of fading channels (Ortin et al., 2025).

Another prominent theme is the integration of reconfigurable intelligent surfaces (RIS) in future wireless systems. Ma et al. (2025) investigate joint precoder and reflector design for RIS-assisted multi-user OAM communication, aiming to mitigate inter-mode interference and enhance spectral efficiency (Ma et al., 2025). An et al. (2025) explore STAR-RIS enabled multi-path beam routing, exploiting the full-space reflection and transmission capabilities of STAR-RIS to create diverse LoS paths and enhance signal coverage (An et al., 2025). Furthering the exploration of RIS capabilities, Ye et al. (2025) propose a fluid RIS (fRIS) for ISAC systems, optimizing element positions and phase shifts to minimize sensing beampattern mismatch and communication error (Ye et al., 2025). Paulino et al. (2025) demonstrate the practical application of a 6.5 GHz RIS for human activity recognition, showcasing its potential for enhancing RF-based sensing (Paulino et al., 2025).

Several contributions focus on specific signal processing and machine learning techniques for wireless applications. Pinilla et al. (2025) introduce WaveMax, a convex optimization approach for radar waveform design based on FrFT phase retrieval, enabling efficient waveform recovery from the ambiguity function magnitude (Pinilla et al., 2025). Dheerendra and Derzsi (2025) propose a root extension to the Propagator algorithm for direction of arrival estimation, significantly reducing computational complexity while improving angular resolution (Dheerendra & Derzsi, 2025). For EEG emotion recognition, Feng et al. (2025) develop an adaptive progressive attention graph neural network (APAGNN) that dynamically captures spatial relationships among brain regions (Feng et al., 2025). In the realm of spectrum cartography, Timilsina et al. (2025) introduce a domain-factored untrained deep prior, leveraging UNNs to represent radio maps without the need for training data (Timilsina et al., 2025). Similarly, Xu et al. (2025) propose a latent domain plug-and-play denoising approach for radio map estimation, utilizing denoisers trained on natural images to avoid the need for radio map training data (Xu et al., 2025).

Addressing practical challenges in wireless systems, Fernandez et al. (2025) present a study protocol for characterizing exposure to non-ionising electromagnetic fields (Fernandez et al., 2025), while another work by the same lead author investigates the performance of wearable slot antennas, analyzing efficiency, frequency shift, and body absorption (Fernandez et al., 2025). Fernandez et al. (2025) also contribute to the standardization of radio impulsive noise measurement methods (Fernandez et al., 2025). Yoo et al. (2025) introduce meta-learned context-dependent conformal prediction for calibrating wireless AI applications, addressing the distribution shift between calibration and runtime data (Yoo et al., 2025). García-Veloso et al. (2025) propose a novel synchrophasor estimator robust against DC offsets and self-interference (García-Veloso et al., 2025). Nikolaidis (2025) presents a parameterized hardware architecture for frame synchronization at all noise levels, achieving high accuracy and bit rates on FPGAs (Nikolaidis, 2025).

Looking towards the future of wireless communications, Hossain and Vera-Rivera (2025) provide a comprehensive overview of the evolutionary path of 6G, discussing its vision, use cases, and technology enablers (Hossain & Vera-Rivera, 2025), while Latreche et al. (2025) delve into the potential applications and services promised by 6G (Latreche et al., 2025). Xiang et al. (2025) propose OpenSC, a semantic communication system combining scene understanding, LLMs, and open channel coding for improved adaptability and efficiency (Xiang et al., 2025). Saleem et al. (2025) present a deep multi-modal neural receiver for 6G vehicular communication, demonstrating its superior performance across diverse data flows (Saleem et al., 2025). Zhang et al. (2025) introduce the ROMA (rotary and movable antenna) architecture for multi-user MIMO systems, optimizing antenna positions and rotation angles for enhanced spectral efficiency (Zhang et al., 2025). Several papers address specific challenges in 6G, including interference prediction and management (Shah et al., 2025), integrated TN and NTN localization (Saleh et al., 2025), data freshness optimization in crowdsensing networks (Shi et al., 2025), and explainable AI for resource allocation in V2X communications (Khan et al., 2025). Finally, Zhu et al. (2025) propose GF-OTFS, a novel modulation technique combining SC-FDMA and UFMC for improved spectral containment and interference mitigation in dynamic channels (Zhu et al., 2025).

Paper Highlights

Scene Understanding Enabled Semantic Communication with Open Channel Coding

Scene Understanding Enabled Semantic Communication with Open Channel Coding by Zhe Xiang, Fei Yu, Quan Deng, Yuandi Li, Zhiguo Wan https://arxiv.org/abs/2501.14520

Caption: This diagram contrasts traditional semantic communication, which relies on local and shared knowledge, with the novel OpenSC system. OpenSC leverages scene understanding, open channel coding, and a public knowledge base to enhance adaptability and interpretability. The system utilizes structured semantic encoding and open channel encoding at the transmitter, and open channel decoding and LLM-semantic decoding at the receiver, for efficient and robust semantic communication.

Sixth-generation (6G) networks are shifting towards semantic communication, prioritizing meaning over symbols. However, current semantic communication systems face limitations due to static coding strategies, poor generalization capabilities, and a reliance on task-specific knowledge bases. These limitations hinder their adaptability to diverse communication scenarios. This paper introduces OpenSC, a novel system designed to overcome these challenges by leveraging scene understanding, Large Language Models (LLMs), and open channel coding.

A key differentiating factor of OpenSC is its use of a public knowledge base for encoding and decoding, in contrast to traditional systems that rely on fixed, domain-specific knowledge bases. This reliance on publicly available knowledge enables more flexible and adaptive encoding, improving the system's ability to generalize across different tasks and environments. This dynamic approach reduces the dependence on static, task-specific data, a significant advancement in semantic communication.

OpenSC further enhances its semantic encoding capabilities by utilizing scene graphs. Scene graphs provide a structured representation of semantic information, capturing not only the objects present in a scene but also the relationships between them. This contextual information is particularly beneficial for tasks like Visual Question Answering (VQA), where understanding the relationships between objects is crucial for accurate responses.

The system's architecture consists of a semantic transmitter and a semantic receiver. The transmitter performs structured semantic encoding, represented as α = S(I; ζ), to extract semantic features from images. These features are then processed by the open channel encoder, X = C(α; γ), which compresses the information for transmission. The receiver performs open channel decoding, â = C¯¹(X; θ), to recover the semantic features. These features are then fed into an LLM-semantic decoder, q = S¯¹(â; δ), which leverages the structured knowledge from scene graphs and the capabilities of LLMs to generate a corrected and enriched version of the data, ultimately performing the downstream task, such as VQA.

OpenSC also introduces innovative techniques for scene graph generation and channel coding. It uses prototype-based embedding networks to learn class-specific prototypes and instance-specific transformations for subjects, objects, and predicates. This approach allows the system to capture both the unique attributes of individual instances and the shared characteristics of different classes. A matching function, F(s, o) = ReLU(s + o) – (s – o)², aligns subject-object pairs with corresponding predicates in a shared semantic space, facilitating accurate predicate prediction. Furthermore, prototype regularization is employed to address semantic overlap between predicates. The channel coding component utilizes a WordPiece algorithm for tokenization and maps these tokens to QAM symbols for transmission. The receiver employs either ZF-LMMSE detection (with CSI) or symbol demodulation based on Euclidean distance (without CSI) to recover the transmitted information.

Calibrating Wireless AI via Meta-Learned Context-Dependent Conformal Prediction

Calibrating Wireless AI via Meta-Learned Context-Dependent Conformal Prediction by Seonghoon Yoo, Sangwoo Park, Joonhyuk Kang, Petar Popovski, Osvaldo Simeone https://arxiv.org/abs/2501.14566

Caption: This figure illustrates the ML-WCP framework. It depicts the selection of a calibration context ($c^{cal}$) from available datasets ($D_c$) and the subsequent use of a meta-learned estimator within the weighted conformal prediction (WCP) process to generate reliable prediction intervals ($\Gamma$) for a new test context ($c^{te}$). The resulting prediction intervals are guaranteed to contain the true output ($y^{te}$) with a user-defined probability ($1-\alpha$).

Reliable AI applications are essential for the effective operation of modern software-defined networks, like Open Radio Access Network (O-RAN) systems. However, ensuring this reliability is challenging due to the dynamic nature of wireless environments. Traditional calibration methods often fall short because they fail to account for the distribution shifts that occur due to changing traffic patterns, network conditions, and other contextual factors. This paper introduces ML-WCP (meta-learned context-dependent weighted conformal prediction), a novel approach designed to address this challenge by leveraging readily available contextual information for robust calibration.

ML-WCP utilizes a meta-learning approach to learn a zero-shot estimator of distribution shifts. This means that the system can adapt to new, unseen contexts without requiring any data from that specific context. During an offline training phase, ML-WCP uses data from multiple past contexts to train a model that predicts the covariate likelihood ratio: w(x, c_te, c_cal) = p(x|c_te)/p(x|c_cal), where x is the input, c_te is the test context, and c_cal is the calibration context. This learned estimator is crucial for adapting to the dynamic nature of wireless environments. This ratio is then incorporated into the weighted conformal prediction (WCP) framework to generate prediction sets that are guaranteed to contain the true output with a user-defined probability. The architecture of the estimator is specifically designed to ensure the necessary symmetry property: w(x, c₂, c₁) = 1/w(x, c₁, c₂).

The effectiveness of ML-WCP is demonstrated across various wireless applications, including traffic slice prediction, scheduling app profiling, and interference-limited communication. In each case, ML-WCP achieves reliable calibration by meeting target coverage levels while minimizing inefficiency, even under previously unseen distribution shifts. The paper also explores multi-context calibration, proposing two schemes: ML-WCP-MV (majority vote) and ML-WCP-Mix (mixing). ML-WCP-Mix, which models the calibration data as coming from a mixture distribution, generally provides better performance.

Channel-Aware Constellation Design for Digital OTA Computation

Channel-Aware Constellation Design for Digital OTA Computation by Zeyang Li, Chen Chen, Carlo Fischione https://arxiv.org/abs/2501.14675

Caption: This figure presents the constellation points for a channel-aware digital over-the-air computation system. Different markers represent various functions (f1-f10) being computed, showcasing the system's ability to support both symmetric and asymmetric functions. The distribution of constellation points along the I and Q axes reflects the channel-aware design, enabling accurate decoding at the central computation point.

Over-the-air (OTA) computation offers a powerful mechanism for efficient data aggregation in wireless networks by exploiting the superposition property of the multi-access channel. While analog OTA systems suffer from limitations in function adaptability, existing digital OTA methods often lack generality, restricting their application to specific functions or modulation schemes. This paper proposes a novel digital OTA computation system featuring a channel-aware constellation design for demodulation mappers that addresses these limitations.

The core innovation of this system lies in its dynamic adaptation to channel conditions. The constellation is adjusted based on the estimated channel conditions of each participating node (hₖ), incorporating channel randomness directly into the constellation design. This dynamic adjustment inherently prevents overlap between constellation points representing different function values, which simplifies the constellation design and reduces computational complexity.

The system employs two types of transmit coefficients: **Type I (bₖ = √Pₜhₖ/|hₖ|) and **Type II (bₖ = √Pₜ), allowing for performance evaluation in both cellular and cell-free systems. This channel-aware constellation design ensures that distinct function values map to unique combined constellation points at the central computation point (CP), enabling accurate decoding. Notably, this system supports both symmetric functions (e.g., sum, product, maximum, sum of squares) and asymmetric functions, broadening the range of potential applications compared to traditional OTA systems, which are often limited to sum operations.

The performance of the proposed system is evaluated in both cellular massive MIMO and cell-free massive MIMO configurations. In cell-free scenarios, two signal processing approaches are examined: fully centralized processing (FCP) and local processing & centralized voting (LPCV). LPCV incorporates a weighted voting mechanism where weights (w_c) are assigned to votes from each CP based on channel conditions using the formula: w_c = S_c / Σ S_i, where S_c represents a metric reflecting the overall strength and uniformity of channel conditions at CP c. This weighted voting scheme offers a balance between performance and overhead.

Conclusion

This newsletter highlights a convergence of innovative approaches aimed at improving the efficiency, reliability, and adaptability of future wireless communication systems. From channel-aware constellation designs for OTA computation to meta-learning techniques for calibrating AI models in dynamic environments, the research presented here addresses critical challenges in the evolution towards 6G and beyond. The development of OpenSC, with its integration of scene understanding, LLMs, and open channel coding, represents a significant step towards realizing the potential of semantic communication. Similarly, ML-WCP provides a robust framework for ensuring the reliability of AI-powered applications in the face of distribution shifts, a crucial requirement for the deployment of intelligent wireless networks. These advancements, along with the ongoing research in RIS technology and novel signal processing techniques, pave the way for a future of enhanced wireless connectivity.