With the rapid growth of emerging network transmission services, communication networks face increasing challenges. Network slicing addresses these challenges by enabling resource isolation and service customization through multiple virtual slices on a shared physical infrastructure. However, resource allocation and path selection in slice management are NP-hard problems that must adapt to dynamic network states and service demands, making static optimization methods inadequate. Moreover, existing reinforcement learning approaches for slice management still face limitations in scheduling control, system modeling, and state encoding. To address these issues, this paper proposes a novel network slicing resource allocation framework that integrates Graph Attention Networks (GAT) for dynamic state encoding and a Deep Deterministic Policy Gradient algorithm for continuous control. The framework jointly optimizes bandwidth allocation factors and path weights, effectively capturing dynamic topologies and link attributes. Experimental results demonstrate that the proposed method achieves superior performance over baseline approaches, improving resource fairness and reducing delay by 14\%. The proposed scheduling framework enhances service differentiation by prioritizing traffic in real time, thereby improving reliability, responsiveness, and overall user experience.

In this work, we propose a cooperative spectrum sharing scheme based on probabilistic and partial edge caching in cognitive radio networks. The primary system contains primary users (PUs) and base stations (BSs), and the secondary system contains secondary transmitters (STs), secondary receivers (SRs), and secondary helpers (SHs). Each PU is associated with the nearest BS and owns the licensed spectrum. SHs cache the required files of PUs with certain probabilities and ratios, and actively assist the transmission to PUs in exchange for spectrum resources for SUs. In each cell, an SH adaptively cooperates with BS to transfer the requested file contents to the PU; thereby, the communication performance of the primary system can be more easily satisfied. As a reward, spectrum can be released to the secondary system in frequency and time domains. An optimization problem is formulated to maximize the area throughput of the secondary system while satisfying the performance requirement of the primary system. An algorithm is proposed to jointly determine file caching probabilities and ratios, and spectrum sharing time and bandwidth. Numerical results show that, compared with the most popular and uniform content placement schemes, our proposed scheme can greatly improve the system area throughput.

Global navigation satellite systems (GNSS) receivers are vulnerable to serious intentional jamming, leading to a loss of positioning capability. Interference recognition technology can assist receivers in selecting appropriate mitigation algorithms. However, traditional interference detection techniques are limited by their reliance on statistical assumptions and fixed parameter settings, resulting in poor adaptability when facing multiple types of interference. Recently, deep learning (DL) has made progress in interference recognition. However, most existing deep networks achieve higher accuracy at the cost of increased complexity, limiting their practical deployment. Moreover, existing methods mainly rely on single-domain frequency features and pay inadequate attention to local texture features, leading to insufficient feature representation and limited recognition performance under low interference power conditions. To this end, we propose an angle-optimized aided dual-stream harmonized network (AO-DSHN) for interference recognition. Specifically, AO-DSHN introduces discriminative time-domain autocorrelation features and frequency-domain features to enhance classification performance. Meanwhile, a dual-stream time/frequency feature cooperative learning network is designed to capture the fundamental shared features across time- and frequency-domain features. Furthermore, a novel dual-domain angle-optimized loss (DAL) is proposed to strengthen the intraclass compactness and interclass separability. Experimental results on simulated, real-world, and public datasets demonstrate that AO-DSHN outperforms several other state-of-the-art methods in terms of recognition accuracy, complexity, and real-time performance, demonstrating its application potential.

The baleful jamming attacks are destructive to the Global Navigation Satellite System (GNSS) receivers. Detecting the jamming type accurately at a GNSS receiver facilitates to select the corresponding mitigation method. Unfortunately, the recognition method based on the manual selection of the signal features relies on the independence of features. Considering the information of similar interference signals in the frequency domain is comparable, the existing deep learning methods, such as convolutional neural networks (CNNs) with local receptive fields, have difficulty in modeling and processing the global correlation between signals. To this end, a dual graph convolutional network (GCN) with adaptive weight learning (AWL-DGCN) method is proposed in this article. The interference signals are converted into graph data, and the GCN, which has demonstrated advantages in processing graph data, is employed for their analysis. More specifically, three kinds of time- and frequency-domain features are extracted to establish different homogeneous graphs between signals. Furthermore, the current graph aggregation method simply adds the homogeneous graphs that contain different information, making it difficult to dynamically balance the contributions of different information, which affects the effectiveness of message passing. Therefore, a dual-GCN structure is proposed for adjusting adjacency matrix weights adaptively, which is composed of the weight learning module and the classification module. The experimental results on simulation, public, and real-world datasets indicate that the proposed method possesses lower complexity, shorter reasoning time, and higher recognition accuracy, which holds potential for application in scenarios, where a timely response to GNSS interference attacks is necessary.

Federated learning (FL) enables collaborative model training across multiple parties while preserving data privacy. However, FL remains vulnerable to privacy leakage through model updates. Differential privacy (DP) has been incorporated into FL by adding noise to model updates to ensure robust privacy protections. Traditional DP methods set a fixed sensitivity limit, resulting in excessive noise addition and performance degradation, especially in complex models such as RS, where the high-dimensional parameters of the model pose a major challenge to effective noise addition. The higher the parameter dimension of the model, the greater the effect of including all the noise will be on the performance of the recommendation system. This study presents an adaptive local differential privacy technique grounded in Fisher information for reinforcement learning, which dynamically modifies the privacy budget by assessing the significance of model parameters during each training iteration. Specifically, Fisher information is used to assess the significance of each parameter layer, with more noise added to less important layers and less noise added to more critical layers. This approach optimizes the noise allocation while maintaining model performance under the DP guarantee. At the same time, we decouple the federated recommendation system (FRS) from the DP mechanism, enabling seamless integration with a variety of recommendation models. We assess the effectiveness of the proposed method through theoretical analysis and experiments on multiple benchmark datasets. The results demonstrate that the Fisher-based adaptive DP method significantly improves model performance compared to traditional fixed-sensitivity DP methods in FL environments, particularly addressing the challenges posed by the complexity of RS parameters.

In this paper, we investigate a secure integrated sensing and communication (ISAC) system aided by a reconfigurable intelligent surface (RIS), wherein a base station (BS) jointly delivers the communication services to legitimate users (LU) and performs radar target detection through unified signal transmission. The deployment of a RIS in the ISAC system enables the creation of virtual line-of-sight (LOS) links for communication and detection to overcome blockages and enhance security. We formulate an optimization problem to maximize the signal-to-noise ratio (SNR) of the radar output by jointly designing the beamforming in the BS and the RIS reflection coefficient while satisfying several constraints of transmit power budget, communication performance requirement, and secure transmission requirement. However, due to the dynamics and complexity of the system, traditional convex optimization methods are extremely computationally intensive. Thus, we adopt deep reinforcement learning (DRL)-based algorithms. Simulation results verify the deployment of RIS enhances system performance and demonstrate the DRL-based algorithms outperform the traditional convex optimization algorithms.

Self-homodyne coherent detection, which meets the transmission requirements of short-reach communication scenarios such as data center interconnects, is limited by carrier fading in the local oscillator (LO) branch and high-power consumption in the digital signal processing (DSP) unit. To address the aforementioned challenges, a comprehensive optimization mechanism is proposed $-$ A polarization control method based on Kalman filter is proposed to address the carrier fading issue; A modulation-format-independent, one-step compensation scheme for polarization demultiplexing and carrier phase recovery based on extended Kalman filter is proposed to mitigate the issue of high-power consumption in DSP. To the best of our knowledge, it is the first time using the Kalman filter for polarization control in the optical domain. The effectiveness of the proposed comprehensive optimization mechanism is validated under both linear and random polarization drift models. The results demonstrate that the proposed polarization control method offers advantages in stabilization performance, convergence speed, and adaptivity. Under the constraint of the adaptive polarization controller, the proposed scheme ensures that the system's bit error rate remains below the forward error correction threshold after 10km of fiber transmission, even with a rotation of the state of polarization (RSOP) of 100 krad/s. Furthermore, the proposed Joint-extended Kalman filter (J-EKF) scheme demonstrates lower power consumption and enhanced tolerance to broad linewidth lasers.

This paper proposes a cognitive relaying scheme based on nonlinear energy harvesting, in which the licensed spectrum of the primary system is shared with the secondary system. The primary system comprises of an access point (AP) and multiple primary users (PUs). An energy-constrained secondary transmitter (ST), multiple secondary relays (SRs) and a secondary destination (SD) constitute the secondary system. ST uses the interference signal from AP to harvest energy. The transmit power of ST and relays should be strictly controlled to not exceed the interference constraint of all the PUs. The relay that has correctly decoded the ST’s data in the first-hop and brings the strongest signal power to the SD in the second-hop is selected to help ST transmit data to SD. Over the statistical properties of Rayleigh fading channels, we analyze the outage probability and throughput of the secondary system. We also present several benchmark systems for the performance comparison. Numerical results show that the protocol greatly improves spectrum efficiency and communication performance, and we emphasize the influence of various parameters on system performance.

With the widespread application of various battery-powered devices, it is very challenging to sustainably meet the tremendous wireless transmission requirements over limited spectrum. We propose a wireless energy harvesting (EH) based uplink transmission scheme for the K-tier heterogeneous cellular network (HCN) with all the channels undergoing Nakagami-m fading. The locations of base stations (BSs) in each tier, mobile users (MUs), and power beacons (PBs) are properly modeled as independent homogeneous Poisson point processes. Each MU can be associated with a BS in any tier that provides the maximum averaged received signal power. MUs can harvest wireless energy from all the PBs in a nonlinear fashion, and then transmit data to the associated BS by using the harvested energy. We properly model the energy statuses of MUs by defining a series of unequally spaced energy levels according to the probability distribution of EH amount. Through properly modeling the aggregate interference, we analyze the outage probability and the area throughput of the multi-tier HCN. Numerical results show that our proposed scheme can greatly outperform the benchmark scheme in terms of outage probability and area throughput. The impacts of key parameters are revealed through extensive simulations, which can guide the network deployment.

We consider an energy harvesting (EH)‐based cooperative relaying system, where the relay operates with EH and forwards source data to the destination according to the decode‐and‐forward protocol. Two intelligent reflecting surfaces (IRSs) are deployed to assist the EH of the first‐hop as well as the data transmissions of both hops. Along with the source transmitting data to the relay at the first‐hop, the first IRS reflects the incident signal to the relay, who can harvest energy and decode information from the received signal using power‐splitting or time‐switching method. After successfully decoding the source data, the relay will forward data to the destination by using the harvested energy with assistance from the second IRS. We analyze and derive the success probability, the throughput, and the ergodic capacity of the proposed scheme. Numerical results show that our proposed dual‐IRS‐aided relaying scheme can achieve much better performance than relay‐only and one‐IRS‐aided relaying schemes.

In heterogeneous cellular networks (HCNs), neighboring users often request similar contents asynchronously. Based on the content popularity, base stations (BSs) can download and cache contents when the network is idle, and transmit them locally when the network is busy, which can effectively reduce the backhaul burden and the transmission delay. We consider a two-tier HCN, where macro base stations (MBSs) and small base stations (SBSs) can cooperatively and probabilistically cache contents. Each user is associated to the BS with the maximum average received signal power in any tier. With the cooperative content transfer between MBS tier and SBS tier, users can adaptively obtain contents from BSs or remote content servers. We properly model both wired and wireless delays when a user requests an arbitrary content, and propose the concept of effective delay. Content caching probabilities are optimized using the Marine Predators Algorithm via minimizing the average effective delay. Numerical results show that our proposed cooperative caching scheme achieves much shorter delays than the benchmark caching schemes.

Nonorthogonal multiple access (NOMA) can facilitate simultaneous data transmissions towards multiple users by using the superposition coding and successive interference cancelation techniques, which can greatly improve the spectrum efficiency. Cooperative relaying and space-time coding can promisingly improve the communication robustness of poor-quality links by achieving the space-diversity gain. Energy harvesting (EH) can prolong the lifetime of energy-limited terminals and make them work continuously. In order to enhance the spectrum efficiency as well as the communication quality, we enable a cluster of EH relays to assist the data transmissions from a base station (BS) to two far-users by using the Alamouti coding based cooperative NOMA strategy. The relays are capable of harvesting wireless energy from a power beacon as well as BS by using time-switching or power-splitting method. According to the energy status and the data decoding status, one relay is selected in a distributed manner according to either Max-min, Max-sum, or Random criterion. We analyze the transmission success probability and the system throughput performance. Extensive simulations are performed to compare the performance of different EH-based space-time coded cooperative NOMA with various relay selection schemes as well as the counterpart orthogonal transmission schemes.

In recent years, the number of patients with depression has grown rapidly. The traditional diagnosis of depression includes mental scales, clinical inquiry etc., which is time consuming and lacks objective confirmation of relevant physiology indicators. In order to overcome the drawback of traditional methods, brain imaging techniques such as electroencephalogram (EEG) have provided new tools for diagnosing depression and shown excellent performance. In this paper, a major depressive disorder (MDD) detection framework is proposed based on parallel spatiotemporal convolution network and mix-multilayer perceptron. First, the wavelet entropy and differential entropy features of EEG were extracted and then parallel spatial temporal convolutional network and mix-multilayer perceptron were employed for further feature representation and extraction. In this process, mmd-loss was creatively added to shorten the gap between the training dataset and the test dataset. Further extracted features were fused and multilayer-perceptron (MLP) was used to perform binary classification. This experiment was evaluated on the MODMA dataset and achieved an accuracy of 0.7832. The experimental results show that the model proposed in our paper is effective in MDD detection and provides better performance compared with the baseline systems.

In order to improve the spectrum efficiency and communication robustness, we propose a cooperative non-orthogonal multiple access (NOMA) scheme in large-scale cellular networks with Device-to-Device (D2D) communication and opportunistic relaying. The distribution of base stations (BSs) is assumed to follow homogeneous Poisson Point Process (PPP). Each BS communicates to both a near-user (CU1) and a far-user (CU2) in its coverage area. A collaborative region is defined between each BS and CU2, wherein there are multiple D2D transmitters (DTs) uniformly distributed and intending to communicate with a common D2D receiver (DR). Among all the DTs that has correctly decoded the data of CU2, one is selected to perform the single-hop decode-and-forward relaying. The selected DT superimposes its own data onto CU2's data, and broadcasts the composite signal. If there is no DT available to forward the cellular data, the DT with the best channel quality towards DR will be selected to transmit the pure D2D data. Through properly modeling the aggregate interference, we analyze the throughput of each terminal. Numerical results show that with specific parameter settings, the NOMA scheme with optimal relay selection can greatly outperform random relay selection in terms of the far-user's throughput and the system throughput. With optimal relay selection, the proposed NOMA scheme can achieve much higher throughput than the orthogonal multiple access scheme.

Driven by the demand for high reliability and low latency applications, the IEEE 802.1 TSN working group proposed Time-Sensitive Networking (TSN) to support time-triggered (TT) traffic for real-time applications and meet compatibility and interoperability requirements. However, when solving the TT traffic scheduling problem in TSN, direct routing using the shortest path or co-design routing for traffic scheduling does not consider the transmission capacity differences caused by different link states, which can result in increased link conflicts and poor scheduling results. This paper takes into account the impact of different link states on routing and scheduling to solve the scheduling problem and minimize the average end-to-end delay of traffic. Specifically, the joint design problem is split into two sub-problems: routing and traffic scheduling. Candidate paths are selected based on delay during routing and added one-by-one using a greedy strategy. When solving the scheduling problem, Optimization Modulo Theory (OMT) is used to model the problem, and Z3 solver is utilized to solve it. Numerical simulations demonstrate that compared to the benchmark methods, the proposed algorithm can effectively reduce the average end-to-end delay of traffic, improve the success rate of scheduling, and optimize the link load distribution.

Human activity recognition (HAR) technology based on wearables has received increasing attention in recent years. The traditional methods have used hand-crafted features to recognize human activities, resulting in shallow feature extraction. With the development of deep learning, an increasing number of researchers have focused on studying deep learning methods. To achieve higher recognition accuracy, the majority of the current HAR research involves multisource and multimodal sensors (MMSs) data. However, due to the limitations in the receptive fields of single-dimensional convolutional kernels, these networks are still infeasible for extracting spatiotemporal features. In this study, a multidimensional parallel convolutional connected (MPCC) deep learning network based on MMS data for HAR is proposed that fully utilizes the advantages of multidimensional convolutional kernels. Moreover, multiscale residual convolutional squeeze-and-excitation (MRCSE) modules are proposed to enrich the diversity of feature information by combining squeeze-and-excitation (SE) blocks. A daily home activity (DHA) data set is constructed based on the requirements for HAR in certain scenarios, such as smart home, and we conduct experiments on the optimal combination of sensor locations on the DHA data set according to a weighted $\text{F}1~({\mathrm{ F}}_{\mathrm{ W}})$ -score. Both tenfold and leave-one-subject-out (LOSO) cross-validations (CVs) are used to evaluate the performance of the proposed network. The MPCC-MRCSE network achieves ${\mathrm{ F}}_{\mathrm{ W}}$ -scores of 98.33% and 95.42% on the physical activity monitoring for aging people (PAMAP2) and OPPORTUNITY data sets using tenfold CVs, respectively, and achieves ${\mathrm{ F}}_{\mathrm{ W}}$ -scores of 81.47% on the PAMAP2 when applying an LOSO CV.

Good sleep quality is very important for everyone to protect physical and mental health. People’s sleep behavior at night reflects their sleep status. In this paper, we propose a method to detect people’s sleep behavior at night by adopting Pseudo-3D (P3D) convolution neural network with attention mechanism. In particular, we propose a new structure, which integrates Squeeze-and-Excitation (SE) blocks into P3D blocks, named P3D-Attention. For the input video, we use P3D blocks to extract spatial-temporal features, and use SE blocks to pay more attentions to the important channel features. The proposed network is tested on the Sleep Action (SA) dataset, which consists of five different actions, namely turn over, get up, fall off bed, play mobile phone, and normal sleep. Experimental results show that the proposed network achieves reasonably good detection results, and the accuracy rate on the test set can reach 90.67%. Compared with 3D convolutional neural networks (C3D), our proposed network can increase the accuracy by about 6% with only 1/6 model parameter size, and achieves an average prediction speed about 1.75 item/s. Compared with the residual spatiotemporal convolution network (R(2+1)D), our proposed network can increase the accuracy rate by about 1.5% with less than 1/2 model parameter size.

Aspect information mining from user comments has become an important means to improve the performance of recommendation systems (RSs). This is because aspect information in comments is fine-grained and tends to reflect the interactions and preferences of users over items in multiple dimensions. These interactions are different from ratings, which are often explicit and linear. Most current RSs based on aspect information learn the contribution of explicit interactions of aspects in a linear manner, while ignoring the implicit features and non-linear interactions of aspects. Since Chinese grammar is greatly different with English grammar, there are few recommendation models based on Chinese movie comment aspects. In this work, we propose an architecture, named aspect-based neural collaborative filtering (ANCF), to extract comment aspect terms based on rules formulated in Chinese dependency parsing. The proposed ANCF integrates a generalized tensor factorization and a tensorized multi-layer perceptrons into the neural network to capture user-item-aspect interactions in a mixed linear and nonlinear way. The aspect potential interaction vector and the actual interaction vector are layered and fused into tensor processing, which can reduce the tensor sparsity and solve the cold start problem of collaborative filtering to a certain extent. Performance results show that the proposed model outperforms some of the traditional ones in terms of recommendation accuracy and effectiveness.

In this paper, we consider a wireless powered cognitive relaying system with a secondary relay (SR) capable of harvesting wireless energy. Along with an access point (AP) continuously transmitting the primary data to a primary user (PU), a secondary source (SS) can transmit the secondary data to a secondary destination (SD) with the help of SR using the decode-and-forward (DF) protocol. SR can harvest energy from both SS and AP in both time and power domains using time-splitting and power-splitting techniques. The interference from primary data transmissions can help boost the amount of harvested energy at SR. The transmit power of SS is regulated by the interference threshold at PU and the allowable peak power. Despite the above two constraints, the transmit power of SR is further constrained by the amount of harvested energy. Once SR successfully decodes the data from SS, it will forward the data to SD using a constrained power. We analyze the approximate outage probabilities for both primary and secondary systems. Simulation results are provided to verify the effectiveness of our theoretical analysis and reveal the impacts of various parameters to the outage performance.

Full frequency reuse with nonorthogonal multiple access (NOMA) in heterogeneous networks (HetNets) can promisingly satisfy the fast-growing wireless transmission requirements, as the data of different users can be simultaneously delivered over the same frequency band. In HetNets, the transmissions from a micro base station (BS) to its cell-edge users often suffer from strong interference from the macro BS, which severely degrades the communication quality. The macro BS and the micro BS can transmit simultaneously to a common far-user using the distributed Alamouti coding technique to improve the communication robustness. Meanwhile, each BS also transmits data to a near-user using the NOMA protocol. The common far-user adopts the maximal ratio combining (MRC) technique to decode its desired data by treating the data of near users as interference. Each near-user decodes the data of far-user using the MRC technique and cancels it to decode its own data by treating the remaining undesired data as interference. Numerical results show that the Alamouti coding based NOMA transmission can greatly improve the throughput of HetNets.