General Overview

This collection of papers explores cutting-edge advancements in wireless communication, signal processing, and AI, with a particular emphasis on optimizing resource management and enhancing performance in complex network environments. A significant focus area is integrated sensing and communication (ISAC), with several papers proposing novel approaches to joint optimization. Ma et al. (2025) introduce an antenna selection scheme for active RIS-aided ISAC systems, aiming to reduce the number of energy-consuming RF chains while maintaining performance. Leyva et al. (2025) propose a two-stage distributed beamforming design for cell-free massive MIMO ISAC, enhancing scalability by distributing signal processing tasks. Shen and Chiu (2025) explore the potential of RIS-aided fluid antenna array-mounted UAV networks, demonstrating significant rate improvements through joint optimization of beamforming, UAV deployment, and antenna position adjustment.

Beyond ISAC, contributions address resource allocation and optimization in evolving network architectures. Mhatre et al. (2025) introduce an explainable AI (XAI) framework for deep reinforcement learning (DRL) based resource management in 6G ORAN architectures, emphasizing intent-based action steering for improved adaptability and KPI-based rewards. Mohsin et al. (2025) also leverage hierarchical DRL for adaptive resource management in integrated terrestrial and non-terrestrial networks, demonstrating significant improvements in spectral efficiency and throughput compared to traditional methods. These works highlight the growing importance of DRL and XAI in managing the complexity of future networks.

Enhancing communication efficiency and reliability is explored through novel signal processing techniques and modulation schemes. Zhou et al. (2025) propose an efficient optimization framework for beyond-diagonal RIS (BD-RIS) aided multi-user communication, addressing the computational complexity challenges of joint active and passive beamforming. Dong et al. (2025) introduce a novel modulation scheme based on the Kramers-Kronig relations for optical IM/DD systems, demonstrating improved BER performance and receiver sensitivity compared to PAM-4 and CAP-16. Valadares et al. (2025) investigate complex-valued neural networks for ultra-reliable massive MIMO, combining quasi-orthogonal space-time block coding (QOSTBC) with SVD for CSI correction and neural network-based decoding.

Specific applications and emerging technologies are also addressed. Führling et al. (2025) present a robust approach to egoistic rigid body localization, addressing the challenges of incomplete observations and unknown target shapes. Majumdar et al. (2025) explore the implementation of finite impulse response (FIR) filters on quantum computers, bridging the gap between classical signal processing and quantum computing. Greatorex et al. (2025) introduce a scalable event-driven spatiotemporal feature extraction circuit for event-driven sensors, enabling low-latency and low-power real-time sensing. Wu et al. (2025) propose a wideband amplifying and filtering RIS for wireless relay, addressing the limitations of traditional RIS in terms of operating distance and band interference.

Finally, the integration of AI and emerging communication paradigms is explored. Soltani et al. (2025) propose VERITAS, a framework for verifying the performance of AI-native transceiver actions in base stations. Meng et al. (2025) investigate secure semantic communication with homomorphic encryption. Arfeto et al. (2025) introduce GenSC-6G, a prototype testbed for integrated generative AI, quantum, and semantic communication. Filo et al. (2025) demonstrate NL-COMM, a non-linear MIMO processing framework for enhanced video streaming. Anido-Alonso and Alvarez-Estevez (2025) apply multi-task deep learning for sleep event detection and stage classification. Singh et al. (2025) provide an overview of physical layer design for ambient IoT, while Jamshed et al. (2025) discuss the integration of AI, ambient backscatter communication, and non-terrestrial networks for 6G. Spieker et al. (2025) introduce PISCO, a self-supervised k-space regularization method for dynamic MRI. Chen et al. (2025) investigate near-field XL-MIMO systems. Salazar-Peña et al. (2025) utilize LSTM and Bi-LSTM models for intra-day solar and power forecasting. Rostami Ghadi et al. (2025) analyze physical layer security in FAS-aided wireless powered NOMA systems.

Paper Highlights

Hierarchical Deep Reinforcement Learning Optimizes Spectrum Sharing in Integrated Terrestrial and Non-Terrestrial Networks

Hierarchical Deep Reinforcement Learning for Adaptive Resource Management in Integrated Terrestrial and Non-Terrestrial Networks by Muhammad Ahmed Mohsin, Hassan Rizwan, Muhammad Umer, Sagnik Bhattacharya, Ahsan Bilal, John M. Cioffi https://arxiv.org/abs/2501.09212

Caption: This diagram illustrates the hierarchical deep reinforcement learning (HDRL) framework for dynamic spectrum allocation in integrated terrestrial and non-terrestrial networks. The framework consists of global, regional, and local policy agents (π_gⁱ, π_rⁱ, π_lⁱ) operating on different timescales and considering various factors like spectrum availability, user distribution, and channel conditions to maximize network performance. Each level utilizes specific input parameters (A_spec, D_beam, G_avg, D_reg, P_u, P_v, H, I) to determine actions (a_gⁱ, a_rⁱ, a_lⁱ) impacting overall network reward functions (F_g, η_g, R_avg, F_r, η_r, QoS, η_l, I_ici).

The increasing demand for seamless wireless connectivity, driven by the surge in connected devices and the emergence of non-terrestrial networks (NTNs), presents significant challenges for spectrum sharing. Traditional methods struggle with the complex, hierarchical nature of these integrated networks and the dynamic characteristics of NTNs like varying orbital dynamics and coverage patterns. This paper introduces a novel hierarchical deep reinforcement learning (HDRL) framework designed to address these challenges by dynamically allocating spectrum across integrated terrestrial and non-terrestrial networks.

The proposed HDRL framework breaks down the network into three hierarchical layers: satellite, high-altitude platforms (HAPs), and a combined layer of unmanned aerial vehicles (UAVs) and terrestrial base stations (TBSs). Each layer has a dedicated DRL agent operating on different timescales. Higher-level agents provide guidance to lower-level agents through metacontrol signals, enabling coordinated spectrum allocation.

The satellite agent focuses on global spectrum allocation, distributing spectrum across beam cells based on the total available spectrum, beam distribution, and average channel gain. The HAP agents then refine this allocation within their respective coverage areas, considering global allocation constraints, user distribution, and local channel conditions. Finally, the UAV and TBS agents make real-time decisions on spectrum access and power allocation for their associated users, factoring in user and UAV positions, channel gains, and interference levels.

The HDRL framework aims to maximize a cumulative reward function formulated as:

max E [Σ ( w₁R_avg + w₂η + w₃F + w₄P_UAV)],

where R_avg represents the average data rate, η denotes spectral efficiency, F represents fairness, P_UAV is a penalty term related to UAV operation (e.g., energy consumption), and w_i are the corresponding weights assigned to each factor.

Benchmarking against exhaustive search, random access, single-agent DRL (SADRL), and multi-agent DRL (MADRL) reveals the effectiveness of the HDRL approach. The framework achieves 95% of the optimum spectral efficiency attained by exhaustive search while being significantly faster, boasting a 50x speed improvement. Compared to MADRL, commonly used for spectrum sharing, HDRL demonstrates a 3.75x faster execution speed and a 12% higher overall average throughput. The HDRL framework also exhibits 10-18% higher performance improvements compared to SADRL and random access across various scenarios.

Further analysis reveals the framework's stability and learning progression. In a single 500-step episode, HDRL achieves 5%, 11%, and 25% higher average throughput compared to MADRL, SADRL, and random access, respectively, with minimal throughput fluctuations. Evaluation across space-air-ground (SAG), air-ground (AG), and UAV-aided networks demonstrates the framework's adaptability to different network complexities, achieving the highest cumulative reward in the SAG network, followed by AG and UAV-aided networks. Ablation studies confirm the robustness and adaptability of HDRL under varying reward formulations and environmental conditions.

Conclusion

This newsletter highlights the convergence of advanced techniques in wireless communication, signal processing, and AI to address the growing complexities of future networks. The highlighted paper on hierarchical deep reinforcement learning (HDRL) for spectrum management in integrated terrestrial and non-terrestrial networks showcases a promising approach to optimizing resource allocation in the face of increasing demand and heterogeneous network architectures. The hierarchical structure of the DRL framework allows for efficient decision-making across different timescales and network layers, resulting in significant improvements in spectral efficiency and throughput compared to traditional and existing DRL methods. This underscores the potential of AI-driven solutions to manage the dynamic and complex nature of future 6G networks, paving the way for seamless and efficient connectivity.