The deployment of reconfigurable intelligent surfaces (RIS) in millimeter-wave (mmWave) multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems is usually hindered by the dual challenges of high-dimensional channel estimation and prohibitive control phase-shift overhead. To reduce channel estimation complexity and feedback overhead, this paper constructs a unified tensor-based framework. First, by exploiting the sparsity of mmWave channels and the Vandermonde structure of antenna arrays, we formulate the received signals as a canonical polyadic (CP) model characterized by tensor decomposition uniqueness. Subsequently, a closed-form channel estimation algorithm is proposed to fit the constructed CP model and extract channel parameters. To further reduce the feedback overhead, we propose a tensor-based low-rank RIS feedback scheme. By exploiting the inherent Kronecker-structure of the optimal RIS phase-shift vector, the proposed scheme approximates it using a low-rank tensor representation, thereby enabling compression based on the estimated channel state information (CSI). Simulation results validate that the proposed framework achieves high-precision and low-complexity channel estimation while maintaining nearoptimal spectral efficiency with reduced feedback bits, particularly in line-of-sight (LoS)-dominant scenarios.

Ensuring the secure communication of unmanned aerial vehicle-assisted reconfigurable intelligent surface (UAV-RIS) is crucial in maintaining seamless connection in next-generation Internet of Things (IoT) networks. For this purpose, intelligent beamforming is essential to ensure secure data transmission from IoT devices to UAV-RIS and optimize communication while preventing adversarial attacks. This paper proposes a novel framework of federated learning for long short-term memory-based deep deterministic policy gradient with an attention mechanism (F-DDPG-AM). The proposed algorithm aims to improve security and mitigate potential threats in UAV-RIS-assisted IoT networks. The F-DDPG-AM combines the federated LSTM’s power to capture long-term dependencies in sequential data with the attention mechanism to focus on key network states and improve decision-making efficiency. The F-DDPG-AM framework improves learning efficiency, accelerates convergence, and enhances resilience against adversarial attacks by selectively prioritizing crucial network information and focusing on insecure scenarios. In addition, federated learning in the proposal ensures secure decision-making through local training for UAV-RIS-enabled IoT networks. The F-DDPG-AM enhances system scalability, trustworthiness, and compliance with secure machine learning principles by decentralizing the training process. The simulation results demonstrate the superior performance of the proposed F-DDPG-AM framework in defending against attacks, significantly outperforming traditional security approaches and other existing reinforcement learning models.

This paper investigates a secure beam-scanning framework for near-field integrated sensing and communication (ISAC) systems, driven by location division multiple access (LDMA). Specifically, an ISAC base station performs full-map robust beam-scanning within each transmission cycle, aiming to simultaneously detect potential eavesdroppers (Eves) and ensure secure communication for legitimate users (Bobs). Based on the Bobs’ channel state information (CSI) obtained at the cycle’s start and the estimated CSI of Eves sensed in the previous cycle, we formulate a robust optimization problem. This problem jointly optimizes the hybrid analog-digital precoding and time allocation for beam-scanning, with the objective of maximizing the worst-case average sum secrecy rate. To simplify the solution process, we first eliminate or relax the semi-infinite constraints caused by uncertain multipath channels from two perspectives: convex hull and bounded uncertainty. Subsequently, we design a near-field LDMA codebook in both azimuth and distance domains to construct ideal radar beampatterns for covering and partitioning the spatial scanning region. We also develop efficient analog precoders to significantly reduce computational complexity. Based on the convex hull model, we develop a low-complexity alternating optimization (AO) algorithm. In addition, for the bounded uncertainty model, we propose a semidefinite relaxation-based AO algorithm without requiring a rank-one constraint. Simulation results demonstrate that the proposed framework enables effective full-map Eves sensing while guaranteeing secure communication for Bobs. Moreover, the convex hull-based algorithm exhibits superior robustness and scalability compared to conventional bounded uncertainty approaches.

This paper investigates a near-field covert communication system enhanced by an absorptive reconfigurable intelligent surface (ARIS) and a fluid antenna system (FAS), enabling covert transmission to arbitrary receiver locations. By generating near-field spherical waves via large-scale antenna arrays at Alice and ARIS, covert transmission to Bob is enabled while evading detection by Willie. We jointly optimize Alice’s hybrid precoding, ARIS reflection coefficients, and Bob’s active port selection to maximize the worst-case covert transmission rate. We begin by evaluating ARIS’s suitability versus conventional RIS. We demonstrate the asymptotic orthogonality of near-field beam-focusing vectors in the 3D domain for uniform planar arrays, and characterize the beam-focusing behavior in cascaded ARIS-enabled covert transmissions. Additionally, we reveal the channel gain improvement owing to FAS over traditional antenna systems. To solve the coupled non-convex problem, we propose a low-complexity block coordinate descent algorithm. It incorporates Fibonacci search for hybrid precoding, three complexity-performance trade-off strategies for reflection coefficients optimization, and both exhaustive search and linear conic relaxation for active port selection. Finally, we recover precoding via an alternating minimization scheme. Numerical results show that (i) significant improvement of covert transmission is achieved only with both ARIS and FAS, when Bob and Willie are co-located; (ii) the proposed algorithm outperforms near-field and far-field beam alignment schemes without ARIS, as well as beam focusing of full-map zeroing with ARIS, when Bob and Willie share the same reception direction.

Efficient radio resource management for Device-to-Device (D2D) communication is challenging due to the need to satisfy diverse Quality of Service (QoS) requirements for various applications. While machine learning techniques have gained attention for addressing these issues, joint prediction of multiple radio resource parameters, such as transmission power, or bandwidth allocation, often leads to suboptimal performance. Thus, in this paper, we propose a coordinated multi-task learning approach based on deep neural networks for joint prediction of channel quality, power, and bandwidth allocation for D2D communication. Then, we design task-specific loss functions and their mutual coordination to maximize sum capacity, ensure QoS, and satisfy task-specific constraints, such as limits on transmission power and bandwidth. Simulation results demonstrate that the coordinated multi-task learning improves the sum capacity and the ratio of devices satisfied with capacity by up to 54% and 30%, respectively, compared to state-of-the-art techniques.

Reconfigurable intelligent surfaces (RISs) have gained considerable interest because of their inherent passive and energy-efficient design. Integrating unmanned aerial vehicles (UAVs) with reconfigurable intelligent surfaces (RIS), known as UAV-RIS, can significantly improve network performance and serve as a crucial enabler for advancements in 6G mobile networks. However, ensuring security in UAV-RIS systems poses notable challenges, particularly in the presence of imperfect channel state information (CSI) and beamforming complexities. In this paper, we identify the critical security requirements for UAV-RIS beamforming in practical scenarios. To address these challenges, we introduce a novel deep deterministic policy gradient with a distributional critic (DDPG-DC)-based beamforming approach aimed at securing UAV-RIS systems while improving the overall secrecy rate. Our proposed secure beamforming solution achieves up to a 48% performance improvement compared to existing state-of-the-art algorithms.

Unmanned aerial vehicles (UAVs) in integration with reconfigurable intelligent surfaces (RIS) play a crucial role in improving wireless communication coverage and enhancing overall performance. However, optimizing the beamforming for the RIS and base station (BS) is critical in improving coverage and ensuring connectivity in densely populated areas. The primary challenge in this process arises from the diverse Quality of Service (QoS) requirements set by user equipment (UEs). To address this complexity, machine learning algorithms are employed to predict the optimal beamforming configurations for both the BS and RIS. However, the traditional supervised learning methods are becoming less effective due to the ever-changing demands of UEs, as these methods rely on fixed data patterns that struggle to adapt to the fluctuating QoS requirements of UEs. Thus, in this paper, we propose an unsupervised learning-based deep learning (DL) approach to jointly predict the optimal beamforming matrix for RIS and BS, enhancing communication coverage and maximizing QoS satisfaction of UEs. The proposed DL-based beamforming adaptively predicts the beamforming matrix, facilitates efficient data exploration during the initial learning phase, and seamlessly scales as the process advances, thereby enhancing overall performance. Numerical results demonstrate that the proposed DL-based RIS and BS beamforming outperforms by up to 89%, compared to the state-of-the-art methods.

As sixth-generation (6G) networks advance to enable massive connectivity and intelligent services, energy efficiency becomes a vital concern, particularly for battery- and edge-powered devices. Building on previous work in task-oriented semantic communication using deep reinforcement learning, this paper proposes an energy-efficient semantic communication framework based on an enhanced Twin-Delayed Deep Deterministic Policy Gradient (TD3) algorithm integrated with an Adaptive Semantic Compression Policy (ASCP). In the proposed framework, we incorporate a predictive channel-aware semantic forecasting (PCSF) module, which leverages lightweight long short-term memory (LSTM) learning models to predict short-term fluctuations in channel conditions. By proactively anticipating variations in signal quality, the agent adjusts semantic compression levels and transmission strategies in advance, enhancing both energy efficiency and the preservation of task-relevant semantic information, particularly under rapidly changing network dynamics. Simulation results show that our energy-aware TD3-ASCP framework, enhanced with PCSF, significantly improves transmission efficiency and accuracy by up to 32% and 26%, respectively, compared to state-of-the-art algorithms.

Spectral efficiency in mobile networks can be increased by allowing devices communicating directly with each other in a form of device-to-device (D2D) communication to reuse channels assigned to cellular devices, i.e., devices communicating via base station. However, the resources reuse leads to additional interference between the cellular and D2D devices. This interference can be suppressed by a smart selection of channels to be reused and allocation of transmission power to all devices at the reused channels. Since this problem is NP-hard, we propose a solution based on deep deterministic policy gradient (DDPG) for cellular channel reuse decisions combined with deep neural network (DNN) for transmission power allocation to all devices. Both machine learning models (DDPG and DNN) are naturally sub-optimal. Thus, we further extend the work towards coordinated learning of both DDPG and DNN so that a potential performance degradation due to sub-optimal outputs of DNN and DDPG is suppressed via a mutual interaction between DDPG and DNN. Simulation results show that proposed solution boosts sum capacity by up to 63 % compared to the best-performing state-of-the-art work.

In this paper, a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-assisted uplink non-orthogonal multiple access (NOMA) system with semi-grant-free (SGF) transmission in the presence of the illegal user is investigated. Particularly, with the help of SGF transmission, grant-free (GF) users omit the tedious process of requesting authorization from the base station and are able to transmit signals by sharing the resource blocks reserved for grant-based (GB) user. Among theKGF users, the one with the best channel conditions is eligible to share the resource block for GB user, which ensures the quality of service for the GB user and avoids collisions due to too many GF users. For this setup, we consider the covert performance and secrecy performance. Particularly, expressions for the outage probability (OP), detection error probability (DEP), optimal detection threshold, and secrecy outage probability (SOP) are derived to assess the system performance. In addition, we present many special cases to get more intuitive insights. Finally, the simulation results verify the correctness and validity of the theoretical calculations, and the effect of each parameter on the system performance is also investigated.

Vehicle Edge Computing (VEC) assists vehicles in performing latency-sensitive tasks by deploying resources near the vehicle. Designing an incentive mechanism for vehicles and VEC is crucial for realizing an intelligent transmission system. Considering the rationality of resource allocation, we model the utility functions of the VEC and the vehicle, which are used as optimization objectives. Specifically, the VEC allocates resources through pricing to maximize revenue under resource-constrained conditions, and the vehicle weighs payments against energy consumption to determine offloading and resource allocation. Given the vehicle movement and the variable channel state, we use the Deep Reinforcement Learning (DRL) algorithm to solve these optimization problems. To reduce the learning difficulty of the DRL algorithm in complex VEC scenarios with multiple optimization variables, we propose a Pricing-Driven Resource Allocation (PDRA) algorithm that performs mobility-aware task offloading and calculates the optimal values of the optimization variables in the utility function of the vehicle to reduce the decision dimension. Furthermore, we also propose a DRL-based Pricing-Driven Dynamic Resource Allocation (DPDDRA) algorithm to achieve efficient resource allocation. Extensive experimental results show that the proposed algorithms can reduce the learning difficulty while maximizing VEC and vehicle revenue in complex VEC scenarios.

In this paper, we propose a novel architecture for a multi-layer active-passive cascade transmitting reconfigurable intelligent surface (RIS)-aided receiver. We aim at enhancing the scalability of antenna dimensions and energy efficiency for user equipment (UE), as well as improving the amplitude freedom of antenna gain. This architecture integrates reflecting RIS for applications in both jamming and anti-jamming scenarios. We formulate a problem to maximize the worst-case spectral efficiency (SE) of the UE, based on imperfect channel state information (CSI) of the malicious device, thereby ensuring the SE of our UE while disrupting the signal reception of the illegal UE. To address the inherently non-convex nature of the formulated problem, we propose an alternating optimization framework, which decomposes the main problem into several subproblems. Specifically, we uniformly discretize the uncertain domains to obtain robust CSI, allowing us to determine an optimal receiver vector. To balance computational complexity and performance, we propose solutions for the subproblems of base station beamforming and coefficients with different patterns of RIS. Furthermore, to effectively disrupt malicious inter-device communication while avoiding detection and localization, we develop a collaborative interference pattern incorporating silent interference and non-interference protocols. Importantly, the pattern carefully balances performance with the overhead of CSI acquisition. Finally, simulation results validate the efficiency of the proposed algorithms, demonstrating that the proposed architecture achieves better jamming and anti-jamming performance.

Mutual reuse of communication channels among device-to-device (D2D) pairs enhances the spectral efficiency of the mobile networks. However, the interference among D2D pairs mutually reusing the same channels imposes a significant challenge. In combination with allocation of the transmission power of each pair for the reused channels, the problem of joint D2D channel reuse and transmission power allocation becomes NP-hard. Thus, we employ deep deterministic policy gradient (DDPG) to decide how the D2D channels should be reused by the D2D pairs. Then, for the reused channels, we allocate the transmission power of the D2D pairs sharing the channels using deep neural network (DNN). However, combining the DDPG-based channel reuse with the DNN-based transmission power allocation leads to an accumulation of errors introduced by DDPG and DNN. The accumulated errors degrade the overall communication capacity. Thus, we also introduce a coordination between DNN and DDPG to suppress the effect of the error accumulation. Simulation results demonstrate that the proposed DDPG-based channel reuse even without coordination increases the sum capacity by 15% compared to state-of-the-art works. On top of this gain, the coordination of both DDPG and DDN adds another 12% in the sum capacity.

The evolution of Artificial Intelligence (AI) integrated with the Sixth-generation ($\mathbf{6 G}$) framework poses significant challenges to low-latency applications. Recently, semantic communication has emerged as a promising technique for future intelligent applications. However, the resource management problem combined with semantics is not fully explored. In this paper, we present a deep reinforcement learning-based twin-delayed deep deterministic policy gradient (TD3) for task-centered semantic communication. The proposed TD3 algorithm optimizes bandwidth, and semantic information and prioritizes data with maximum signal-to-noise ratio (SNR) for the efficient transmission of useful information. Simulation results demonstrate the effectiveness of the proposed TD3 scheme compared to state-of-the-art work in terms of transmission efficiency by up to $36 \%$ for varying users and up to $33 \%$ for varying SNR.

Unmanned aerial vehicles (UAVs) have been adopted in various applications, including agriculture, public safety, surveillance, and crucial military missions. However, alongside their advantageous nature, UAVs have also been employed for malicious activities, leading to an increased requirement for timely detection and identification. Despite significant progress in UAV detection, challenges persist, particularly concerning various types of UAVs, the payload carried by UAVs, and the traits of their flight. Employing single machine learning for detection and identification has limitations due to the inability to handle diverse datasets and acquire complex relationships. Therefore, in this paper, we introduce a novel integration of the Visual Geometry Group-based convolutional neural network (VGG-CNN) framework employed for detection with the Mask Region-based convolutional neural network (MR-CNN) for identification of UAVs (jointly termed MR-DCNN). For efficient deployment of MR-DCNN, we add diversity to the dataset by performing data augmentation of new images in the training dataset for the detection of various types of UAVs, payload categories, and flight characteristics. The performance evaluation of the MR-DCNN approach was conducted via simulations, revealing superior detection capabilities for malicious UAVs compared to existing methods.

Multihop communication has gained prominence within the realm of the Internet of Underwater Things (IoUT) owing to its exceptional reliability amidst the challenges posed by the underwater acoustic environment. Despite this, the persistence of limitations caused by propagation delay, high collision rate, and limited energy in underwater communication remains, representing the most formidable hurdles in ensuring the successful transmission of data gathered by sensor nodes. To address these challenges, we employ a machine learning (ML)-based optimal cooperating node selection for each hop, considering the Shortest propagation delay, minimal residual Energy, and a low Collision rate (referred to as SEC). For this purpose, we initially assemble the sensor nodes to create a list of cooperative nodes, considering the aspect of SEC. Then, using an assembled list of cooperating sensor nodes, we employ ML-based algorithms, such as reinforcement learning (RL-SEC), deep Q-networks (DQN-SEC), and deep deterministic policy gradient (DDPG-SEC), to predict the optimal cooperating node for each hop. The simulation results of the DDPG-SEC demonstrate a significant improvement of approximately 56% when compared with RL-SEC, DQN-SEC, and other state-of-the-art techniques.

User applications based on the deep neural networks (DNNs), such as object or anomaly detection, image recognition, or language processing, running on computation- and energy-constrained user equipment (UE) can be partially or fully processed in the edge computing servers to reduce a processing time and save an energy in the UE. To further reduce the processing time and the UE’s energy consumption, DNN with multiple exit points can be incorporated. In this article, we address the problem of the decision on whether the computation should be offloaded from the UE to the edge computing server or processed locally by the UE and we solve this problem jointly and “on-the-fly” together with DNN exit selection. Since the formulated problem is very complex, we exploit the deep deterministic policy gradient for the exit selection and the offloading decisions (labeled DDPG-EOD) for the DNN-based applications. To this end, we first convert the problem into the Markov decision process, and then, we employ an end-to-end learning via DDPG with the actor-critic architecture. Second, we use a knowledge distillation-based technique to efficiently select the DNN’s exit to minimize the delay and energy consumption. Simulation results show that the proposal is highly scalable, converges very quickly, and surpasses the best performing state-of-the-art approach by up to 120% and 100% in terms of the overall DNN processing delay and the energy consumption, respectively.