An extremely large-scale intelligent reflecting surface (XL-IRS) aided communication system is studied. Although XL-IRS can effectively combat the double path-loss attenuation, the large aperture size introduces significant near-field effects. The complex near-field propagation and large number of XL-IRS elements can lead to optimality and complexity challenges in beam focusing design. Focusing on a spectral efficiency (SE) maximization problem, two unsupervised learning based algorithms are conceived for the joint optimization of base station and XL-IRS beam focusing, which operate without pre-training and exhibit strong robustness. Specifically, a dense-connected dilated autoencoder meta learning (DDAML) algorithm is proposed to achieve high SE by utilizing dense connections, while reasonably designing an autoencoder to reduce computational complexity. Furthermore, considering a need for low execution time in practical applications, a convolutional dilated autoencoder meta learning (CDAML) algorithm is also proposed to further reduce computational complexity. Simulation results show that the proposed DDAML algorithm achieves the highest SE, while the proposed CDAML algorithm significantly reduces computational complexity at the cost of limited SE loss. Moreover, the two proposed algorithms also demonstrate remarkable robustness in XL-IRS-aided near-field communications.

Physical layer security is expected to play an important role in the next generation of wireless networks. The liquid neural network-based low-complexity beamforming (LNNLC-BF) is proposed in this paper to address the computational complexity and robustness of RIS-assisted multiuser secure communication system with imperfect channel state information and resource constraints. First, the differential equation of time-varying channel features is constructed by the continuous-time modeling of liquid neural network. Then, the sparse activation and dynamic path pruning are combined to reduce the dimension of matrix. Finally, a low-dimensional projection is designed to transform the high-dimensional beamforming into a low-dimensional spatial optimization to realize the joint mapping of beam and phase shift. We evaluate the performance of the proposed method through numerous and extensive simulations and show that the proposed LNNLC-BF provides a high security and low computational complexity scheme in the presence of imperfect channel state information.

The maritime joint sensing and communication system (MSCS) has recently emerged as a promising solution to address the maritime spectrum scarcity issue for high-efficiency communication and environmental sensing. To mitigate the significant path loss experienced over the complex sea surface, the intelligent reflecting surface (IRS) is integrated into the MSCS to enhance the signal quality by dynamically adjusting the phases of its reflecting elements. Building upon this foundation, we aim to improve the sensing performance by maximizing the sensing signal-to-noise ratio while maintaining normal maritime communication, which involves optimizing active and passive beamforming vectors. Under the unpredictable environmental information and high-complexity and high-overhead channel information estimation in MSCS, we propose a heuristic algorithm based on genetic evolution to tackle this problem. To ensure the algorithm convergence and the feasibility of available solutions, we introduce an individual processing approach and an elitist reservation strategy in each genetic generation. Numerical results and simulations validate the convergence and efficacy of the proposed algorithm. Additionally, we analyze the effects of critical parameters and IRS structure on the algorithm performance.

Differential privacy is a promising framework for computing on sensitive data while preserving privacy. However, balancing privacy and accuracy remains a significant challenge. In this paper, the DPAdaMod_AGC algorithm is proposed to address the above challenge, which uses adaptive gradient clipping to improve the accuracy of deep learning models without compromising privacy. By reducing the inclusion of nonessential noise during training, the proposed algorithm combines stochastic gradient descent with gradient clipping striking an effective trade-off between privacy and accuracy. Simulation analysis shows that compared with the DPAdaMod algorithm, the algorithm proposed in this paper not only improves the classification accuracy by 2.3%, but also consumes less privacy budget while protecting privacy. The results demonstrate the superiority of the DPAdaMod_AGC algorithm in achieving both privacy and accuracy goals.

The most common symptom of stroke is limb dysfunction, resulting in a reduced quality of life for the patient. Patients can be rehabilitated through specific exercises movements. However, when the exercises are performed without the presence of a healthcare provider, patients and families cannot be informed about the correctness of the exercises movements. Recent graph convolutional neural network (GCN) approaches address this problem by extracting features from a coordinate grid of skeletal data obtained from videos. The traditional GCN network cannot accurately reflect the spatial dependencies between nodes due to the fixed graph structure. In fact, the interactions between nodes may change with time and environment, so the fixed neighbor relationship cannot adapt to the dynamic change of graph structure. To address this problem, a hierarchical decomposed graph structure is proposed in this work for predicting continuous scores instead of discrete labels. In addition, this work introduces an attention-guided hierarchy aggregation module to better extract the features of the main hierarchical edge sets. The results of the simulation demonstrate that the proposed HDGCN-LSTM significantly outperforms the existing algorithms interms of MAD, RMS and MAPE.

With the increasing prominence of trajectory data privacy and security issues, trajectory data privacy protection algorithms based on deep learning have received widespread attention. The core idea of such algorithms is to use synthetic trajectories to replace real trajectories for publishing. However, the generated synthetic trajectories may deviate from the real trajectories excessively, thus affecting the utility of trajectory data. Aiming at the above problems, the MogLSTM-TrajGAN trajectory data privacy protection algorithm is proposed in this paper. Firstly, encoding the real trajectory data to obtain the spatio-temporal distribution of the real trajectories and adding the gating unit on the traditional long short-term memory network to establish a richer interaction space between the input and the output. Then, the improved long and short-term memory network is combined with the generative adversarial network to generate synthetic trajectories. Finally, the trajectory loss function is designed to measure the similarity loss between the synthetic trajectory data and the real trajectory data. Simulation results show that the trajectory data generated by the proposed algorithm retains the spatial, temporal and point-of-interest features of the real trajectory data, and effectively balances the privacy and utility of the trajectory data.

Integrating cognitive radio into Internet of Things (IoT) is conducive to reducing spectrum scarcity for large-scale IoT deployment, where a core technology is the design of spectrum access algorithms for effective assignment of spectrum holes. However, due to the partially observable channels and increased number of users in the cognitive radio Internet of Things (CRIoT) network, the secondary users have difficulty avoiding interferences and accessing the spectrum quickly. This study presents a distributed dynamic spectrum access (DSA) algorithm that employs a priority experience replay deep echo state Q-network (PER-DESQN) for CRIoT networks with multiple users and channels. To accelerate the Q-network convergence, we use an echo state network based on the underlying temporal correlation to estimate Q-values. Then, to resolve the Q-value overestimation and improve prediction accuracy, the estimated Q-value and decision action process are trained using a double deep Q-network (DDQN). Moreover, a priority experience replay mechanism that uses the Sum-Tree combined with importance sampling weights is proposed to optimize the DDQN to address the instability of the Q-value resulting from random sampling. As the simulation results demonstrate, the proposed algorithm can make fast and accurate DSA decisions and boost the network channel capacity significantly.

Hybrid precoding has been recognized as promising and effective for practical 5G communication. It is generally challenging to obtain the sample for deep learning-based hybrid precoding due to its need of massive precoding vector and channel matrix. To effectively solve this issue, a novel coordinated hybrid precoding algorithm based on unsupervised learning graph attention networks (CHP-ULGAT) is first developed by making full use of the underlying topology formed by the channel matrix. Subsequently, a more realistic situation of existing an ultra-low execution time is considered. A sub-optimal coordinated hybrid precoding based on unsupervised learning convolutional neural networks (CHP-ULCNN) is proposed to further reduce complexity. Moreover, we present effective ways to design the multi-matrix operation and the loss function to address the practicability of the algorithm. Extensive simulation results show that the proposed hybrid-precoding algorithms have obvious advantages in spectral efficiency (SE) and energy efficiency (EE) improvement with ultra-low computational complexity, considering the different number of RF chains and deployment scenarios.

Terahertz (THz) massive multiple-input multiple-output (MIMO) is considered as a key technology for future sixth-generation (6G) wireless communications, in which hybrid precoding facilitates an important trade-off of hardware cost and spectrum efficiency. However, the performance of traditional schemes is limited owing to the beam split effect and the non-convex optimization problem as well as the inter-user interference under imperfect channel state information (CSI) in THz massive multi-user (MU)-MIMO systems. To overcome these challenging problems, we propose an unsupervised convolutional neural network (CNN)-based hybrid precoding scheme with attention mechanism. Specifically, we first adopt the true-time-delay (TTD) structure to mitigate beam splitting. Then, to solve the non-convex optimization problem of TTD hybrid precoding and to further mitigate inter-user interference, we propose a robust hybrid precoding scheme by applying the attention mechanism and CNN, which can be trained to generate an optimal analog precoder targeting at an achievable rate maximization under imperfect CSI. Simulation results show that the proposed algorithm has good robustness and can maintain excellent achievable rate performance in the case of imperfect CSI.

Millimeter wave (mmWave) communication combined with massive multiple input multiple output (MIMO) system is one of the most promising technologies in future wireless networks due to the characteristics of high bandwidth and narrow beam. For the strong channel attenuation, the more accuracy channel state information (CSI) is needed to make sure that the signal is received accurately in mmWave system. In this paper, a hardware-friendly channel estimation algorithm, named modular image denoising approximation message passing (MIDAMP), is proposed by combining the real image denoising network (RIDNet) and the learning approximation message passing network. The gap between the estimated channel and the real channel can be greatly reduced through using the powerful denoising ability of MIDAMP. The results of simulation demonstrate that the proposed MIDAMP algorithm has better performance in estimation accuracy and achievable sum rate (ASR) compared with some existing algorithms.

Power consumption caused by radio frequency (RF) chains in millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems can be solved by beam selection. However, the spectral efficiency of traditional beam selection algorithms is unsatisfactory due to the reduction in the number of RF chains and the multiuser interference. This work proposes a differential evolution (DE)-based beam selection algorithm and an improved QR precoder to reduce power consumption and increase the performance of the systems. The proposed algorithm selects the optimal beams with DE-based beam selection for each user, which reduces the power consumption and the interference among each beam. In addition, to greatly decrease the multiuser interference, we propose an improved QR precoder by equalizing diagonals and using Tomlison-Harashima (TH) theory which can greatly reduce the computational complexity and improve the performance. The simulation results show that the proposed scheme outperforms some existing algorithms in the aspects of energy efficiency and spectral efficiency.

In Internet of Things (IoT) applications with multihop networking, not only traditional energy efficiency but also many distinct features should be considered when designing routing protocols, including different security requirements, heterogeneity, and scalability. In this article, an energy-efficient multilevel secure routing (EEMSR) protocol in IoT networks is proposed. Considering that clustering is a reasonable solution of conserving energy, a cluster-based multihop routing protocol is utilized to reduce the high communication overhead due to the scalability of IoT networks. In particular, more reasonable analytic hierarchy process and genetic algorithms are adopted to assign accurate weight and optimize intercluster routing in which heterogeneous IoT networks are considered to support large amount of heterogeneous IoT entities and services. Moreover, multiple trust levels are adopted to defend the different attacks by calculating the trust factor on the clustering and routing, including data perception trust, data fusion trust, and communication trust. It is shown that the proposed algorithm outperforms the existing algorithms in terms of network lifetime, throughput, packet delivery ratio, network energy balance, and adaptability.

Coordinated beamforming (CoBF) for multi-user massive multiple-input and multiple-output (MIMO) heterogeneous networks (HetNets) promises for capacity enhancement. However, challenges of energy efficiency (EE) and ultra-low latency are yet to be addressed due to the circuit power and calculation latency heavily depend on the number of transmit antennas. To solve these problems, a maximizing EE algorithm named coordinated beamforming based on convolutional neural networks (CoBFCNN) is proposed in which the advantages of convolutional neural networks and deep learning are fully exploited. Basing on the results of this study, an optimization problem of maximizing EE with lower complexity and lower calculation latency for the different constraints is formulated and exploited for multi-user massive MIMO HetNets. Simulation and analysis show that the proposed CoBFCNN algorithm can significantly satisfy the performance of maximizing EE for the multi-user massive MIMO HetNets with significantly lower complexity and ultra-low calculation latency, especially when the number of antennas is large.

The number of radio-frequency (RF) chains is limited in millimeter-wave massive multiple-input and multiple-output (MIMO) systems, which brings challenges to channel estimation. To solve this problem, we exploit generative adversarial networks (GAN) and a deep convolutional neural networks (CNN) for a three-dimensional (3D) lens millimeter-wave massive MIMO system, which could learn channel structure and obtain accurate channel estimation from the training data. A novel GAN-CNN blind denoiser (GCBD) based neural networks is proposed in this paper. Based on the analysis and simulations, the GCBD networks enjoys satisfying accuracy even in the low signal-to-noise (SNR) region and significantly outperforms existing algorithms with lower estimation error, including a support detection (SD)-based channel estimation and sparse non-informative parameter estimator-based cosparse analysis approximate message passing for imaging (SCAMPI), Non-Local Means (NLM), Block Method of 3-Dimension (BM3D) schemes. Moreover, we consider a typical blind denoising problem by removing the unknown noise from the noisy channel, which is different from the exiting scheme.