Integrated sensing and communication (ISAC) enables reliable transmission by sensing targets in scattering environment. However, a single base station (BS) struggles to accurately detect three-dimension (3D) moving targets. Meanwhile, asynchronicity caused by spatially separated transceivers introduces timing offsets (TOs) and carrier frequency offsets (CFOs), which impair the accuracy of sensing parameter. To address this, this paper studies an orthogonal frequency division multiplexing (OFDM) networked ISAC system, where multiple base stations (BSs) cooperatively sense multiple unmanned aerial vehicles (UAV) through a joint active and passive sensing framework. The sensing signals received from the BSs are transmitted through a backhaul-limited link to a fusion center (FC) for sensing parameter estimation. In this context, we propose a novel cooperative sensing scheme for UAV parameter estimation based on quantized signals. The joint active and passive sensing problem is formulated as a problem with imperfect parameters (such as TO and CFO). An efficient algorithm is proposed to solve the resulting problem. Simulation results indicate that the proposed cooperative sensing design achieves higher estimation accuracy than other baselines. The results also confirm the performance advantage of multi-BS cooperative sensing over conventional single-BS sensing.

Multi-domain multivariate time series (MTS) forecasting is increasingly important for large-scale and transferable time-series modeling. However, heterogeneous datasets usually contain different numbers of variables, making scalable inter-series dependency modeling challenging. Existing scalable forecasting paradigms often rely on channel-independent (CI) strategies to accommodate variable-dimensional datasets. Nevertheless, by sharing global parameters across independently processed channels, CI models may confuse heterogeneous dependency structures under multi-domain joint training, leading to the dependency confusion problem. In this paper, we argue that effective multi-domain MTS forecasting requires variable-number-agnostic channel-dependence modeling that can explicitly capture fine-grained inter-series dependencies (FID), including both local inter-series dependencies and cross-temporal inter-series dependencies. To this end, we propose the Fine-grained Inter-series Dependency Enhanced (FIDE) framework, a plug-and-play module for patch-based CI forecasters. FIDE introduces dependency prototypes to dynamically perceive input-specific inter-series dependency patterns and employs dependency transmission to propagate dependency information across temporal segments while remaining agnostic to the number of variables. Extensive experiments on eight real-world benchmarks with six representative CI forecasting models demonstrate that FIDE consistently improves forecasting accuracy, effectively mitigates dependency confusion, and achieves statistically significant gains over baseline methods.

To support the intelligent evolution of Internet of Things (IoT) systems toward enhanced comprehensiveness and sophistication, this paper proposes a distributed training and inference oriented toward continual learning. By integrating federated continual learning with inference offloading, the system addresses key challenges in IoT scenarios, including large-scale data processing, catastrophic forgetting during incremental updates, and resource constraints. A joint optimization framework of training-inference is established to analyze model accuracy, latency, and energy consumption. The optimization objective and strategy are formulated as a Markov Decision Process (MDP) with a hybrid action space and customized reward mechanism. The Soft Actor-Critic (SAC) algorithm enables action grouping and the transformation between discrete and continuous actions, achieving unified optimization of training and inference. Simulation results show that compared to existing approaches, the proposed method improves node selection, resource allocation, and offloading strategy by jointly considering communication and computation costs.

With the advancement of wireless communications toward higher user densities and increasingly complex environments, near-field communication has emerged as a critical research focus. Supporting multi-user access in this regime with manageable complexity remains a key challenge, particularly due to the coupling between the spherical wavefront effect and the spatial non-stationarity (SnS) property, which complicates the exploitation of spatial correlation among user channels. To address this, we propose a unified multi-user channel model that incorporates joint support to capture the structured sparsity shared across users. Additionally, we introduce a two-dimensional (2D) Markov prior to model both local sparsity pattern continuity across adjacent grid points and joint burst sparsity in the shared support structure. Based on this, we develop a spatial information-sharing-aided framework that alternately estimates model parameters. Specifically, an inverse-free variational Bayesian inference (IF-VBI) algorithm is employed in the channel estimation module to avoid high-dimensional matrix inversion while enabling information exchange among users via joint support. In the common grid update module, joint updates across users are performed to achieve a full spatial-domain optimum, whereas in the joint visible region (VR) matrix detection module, the user-sharing structure is exploited to decouple and efficiently solve the VR matrix. Simulation results validate the effectiveness of the proposed approach, demonstrating improved estimation accuracy and computational efficiency in multi-user near-field extremely large-scale multiple-input-multiple-output (XL-MIMO) systems.

Integrated sensing and communication (ISAC) has drawn great attention for its capacity to simultaneously support wireless communication and environmental sensing. However, in dynamic ISAC systems, environmental mobility poses critical challenges in both dynamic channel estimation and continuous sensing parameter tracking across multiple time slots under severe Doppler effect. To address these challenges with a unified framework, we propose a novel base station-user cooperative sensing approach for joint dynamic channel estimation and sensing parameter tracking, including target/scatterer locations and actual velocities, which is formulated as a maximum a posterior (MAP) problem. In cases where some moving radar targets also serve as communication scatterers, the spatial overlap induces an underlying partially common sparsity between the location-domain sensing and communication channels. Based on this, we develop a two-dimensional Markov Model (2D-MM) based on dynamic location grids to capture the spatio-temporal correlations and partially common sparsity, thereby enhancing both sensing and communication performance. To alleviate the high-complexity matrix inverse in the E-step of sparse Bayesian inference, we propose a dynamic subspace-constrained variational Bayesian inference (D-SCVBI) algorithm with the aid of prior information obtained from the two-dimensional discrete Fourier transform (2D-DFT) localization and state evolution model. Finally, simulations show that the proposed D-SCVBI algorithm attains remarkable performance gains over various baselines.

As microservice architecture gains widespread adoption in cloud computing due to its scalability, effective resource management becomes increasingly crucial for cloud service providers. However, evaluating end-to-end (E2E) latency and performing real-time container allocation remains challenging due to the difficulty in decoupling performance impacts caused by service execution variability and complex execution logic. In this work, we propose a lightweight resource scaling framework for microservices based on latency decoupling (MiLD), enabling efficient resource allocation based on container scaling. MiLD recursively traverses the execution time graph and decouples potential linear relationships by analyzing trace data. To tackle the uncertainty brought by the predicative decoupling, we design a Value-at-Risk(VaR) optimization method to mitigate cumulative latency risks, balancing performance and resources. Experiments demonstrate that MiLD can decide in seconds, reducing container usage by 1.3x and tail latency by 2.03x compared to baselines in real microservice applications.

With the increase of carrier frequencies and the expansion of large antenna array sizes, wireless communication systems are progressively entering the near-field region. Characterized by spherical wave characteristics, near-field channels exhibit more complex spatial structures and higher parameter dimensions. To mitigate these issues, we propose a near-field channel tracking framework based on a Markov model-based prior. Specifically, by leveraging the spherical wave propagation characteristics, we construct a three-dimensional polar-domain sparse representation, i.e., azimuth, elevation, and distance dimensions, enabling high-resolution channel modeling. Incorporating a Markov model-based prior, we can effectively track variations in spherical wave channels. For channel estimation, we develop a Turbo subspace-constrained variational Bayesian inference (Turbo-SC-VBI) algorithm that avoids high-dimensional matrix inversion, significantly reducing computational complexity. Numerical results demonstrate that the proposed method achieves superior channel tracking performance, with lower pilot overhead and acceptable computational complexity.

With the progressive evolution of large language models (LLMs) and the increasing need of computing for$\mathbf{6 G}$, it becomes crucial for multiple network nodes with limited computing resources to share the need for large model inference. Model partition methods have been proposed to enable computation-intensive artificial intelligence (AI) services by splitting an AI model across multi-edge and cloud nodes. In this paper, a distributed inference optimization framework for transformer decoder-only based LLMs (DIO-LLMs) is proposed in edge-cloud collaborative networks. The partitioning and offloading strategy is determined based on the computing workload and network status. DIO-LLMs specifically accounts for the parallel execution capabilities of the transformer architecture. It employs a two-phase model partitioning strategy, comprising inter-layer and intra-layer partitions, to effectively distribute LLMs across edge and cloud nodes. Additionally, to mitigate inference latency under resource limitations, a Greedy Proximal Policy Optimization (GPPO) based algorithm has been developed to devise optimal strategies. Simulation results indicate that under memory constraints, the proposed algorithm can reduce inference latency more effectively than other baseline algorithms.

The growing demand for edge services calls for an efficient microservice placement strategy to ensure optimal deployment in dynamic edge-cloud computing environments. In this work, we propose a heterogeneous workload-aware microservice placement framework that optimizes deployment by maximizing edge throughput while minimizing costs. To account for the diverse deployment requirements of microservices under heterogeneous workloads, we explicitly address the placement and resource constraints for both light and heavy workloads—an aspect rarely addressed in existing studies. The formulated problem is inherently non-continuous, non-convex, and involves fractional operations, posing significant computational challenges. To overcome this, we develop a Sparsity-promoting Fractional Programming (SFP) algorithm that relaxes and reformulates the problem into a sparse-promoting linear program using l0-norm approximation. Extensive evaluations demonstrate the effectiveness of our framework in improving edge throughput, reducing deployment costs, and efficiently managing workload heterogeneity.

Knowledge distillation transfers knowledge from a complex network (teacher) to a lightweight network (student). Existing knowledge distillation methods typically employ a loss function comprising task and distillation losses, and they use a hyper-parameter to balance the two losses. However, in this paper, we observe an inconsistency between the gradient directions of these two losses, which introduces a trade-off between two gradients, hindering the student from learning the entire knowledge. To overcome this challenge, we propose a Universal Knowledge Distillation Framework Based on Decoupling Gradients (DGKD). DGKD breaks the trade-off by decoupling the gradients of the two losses and distributing these to separate branches of the student. Additionally, we introduce a Knowledge Interaction Module (KIM) and Inference-stage Simplifications to optimize our DGKD, enhancing its flexibility and simplicity. Extensive experiments validate the superiority of DGKD. For example, DGKD achieves a +2.80% accuracy improvement on CIFAR-100 for the VGG13-MN-V2 pair.

Time series forecasting plays a pivotal role in edge intelligence. Current research predominantly focuses on exploring complex model architectures, such as Transformer and Graph Neural Network (GNN), which demonstrate remarkable advantages in capturing high-dimensional complex features. However, these models suffer from inherent limitations in computational efficiency and deployment resource, resulting in significant constraints in temporal efficiency and edge computing compatibility. To address these challenges, this paper proposes a lightweight linear model architecture termed ELinear. By introducing the Linear Channel Fusion module (LCF), the Frequency-domain Multi-Period Awareness mechanism (FMPA) and the Residual Period Fusion module (RPF), we enhance the prediction accuracy of linear models. Experimental results demonstrate that ELinear achieves a 4.7% reduction in Mean Absolute Error (MAE) and a 3.1× improvement in compute speed compared to state-of-the-art (SOTA) models on the widely adopted ETT benchmark dataset.

With the development of virtualization technology, serverless computing has been gaining significant attention in recent years, primarily due to its advantages in scalability and a pay-as-you-go pricing model. In edge networks, the deployment of fine-grained function instances to handle massive request data makes it more difficult to optimize the quality of user experience. This paper addresses the scheduling of heterogeneous function processing requests generated by users in serverless edge computing scenarios, considering the constraints of resources on edge nodes, and making decisions regarding the warm and cold start during the scheduling process. The problem is modeled as a constrained multi-objective optimization issue aimed at minimizing latency and energy consumption. A deep reinforcement learning strategy, grounded in Multi-Agent Proximal Policy Optimization (MAPPO), is introduced to address this challenge, with each user being represented as an autonomous agent. Simulations were conducted to assess the impacts of the learning rate, the request volume, and the size of the input data of the function. The experimental results indicate that, compared with Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), under different scenarios of request scales, the average system delay is reduced by at most 31%, and the average energy consumption is reduced by at most 24%.

Millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems enable extraordinarily high data rates and necessitate precise channel state information (CSI) for effective beamforming. To explore the structural sparsity feature, we propose an advanced low-rank matrix factorization technique for channel estimation. Our algorithm treats channel estimation as a problem of low-rank matrix completion by exploiting the simultaneous sparsity and low-rank nature of the channel. First, we decompose the channel matrix into two low-rank factor matrices, refining them through projected gradient descent to satisfy the constraints of low-rankness and sparsity simultaneously. For improved results, we introduce a deep unfolding network, named Deep Unfolding Low-Rank Factorization Net for Channel Estimation (DULRF-CE), which integrates the low-rank matrix factorization model into the neural network design. This network includes a module capable of nonlinear sparse transformations and a denoising module, giving it the ability to learn a more accurate low-rank decomposition and a sparser channel representation. Our experimental results show that the DULRF-CE algorithm provides a significant advance, offering up to 4.3 dB improvement in average normalized mean square error performance over existing matrix-completion-based channel estimation techniques in various measurement scenarios.

Split Learning (SL) is a promising collaborative machine learning approach, enabling resource-constrained devices to train models without sharing raw data, while reducing computational load and preserving privacy simultaneously. However, current SL algorithms face limitations in training efficiency and suffer from prolonged latency, particularly in sequential settings, where the slowest device can bottleneck the entire process due to heterogeneous resources and frequent data exchanges between clients and servers. To address these challenges, we propose the Heterogeneous Split Federated Learning (HSFL) framework, which allows resource-constrained clients to train their personalized client-side models in parallel, utilizing different cut layers. Aiming to mitigate the impact of heterogeneous environments and accelerate the training process, we formulate a latency minimization problem that optimizes computational and transmission resources jointly. Additionally, we design a resource allocation algorithm that combines the Sample Average Approximation (SAA), Genetic Algorithm (GA), Lagrangian relaxation and Branch and Bound (B&B) methods to efficiently solve this problem. Simulation results demonstrate that HSFL outperforms other frameworks in terms of both convergence rate and model accuracy on heterogeneous devices with non-iid data, while the optimization algorithm is better than other baseline methods in reducing latency.

Containerized microservice (MS) architecture has emerged as the preferred solution for increasingly complex IoT applications in edge computing (EC). However, containerized MS’s runtime requires frequent pulling the layer-based container image and data transfer between dependent nodes, which may incur huge traffic and latency overhead, especially in resource-limited EC-enabled IoT. Additionally, focusing only on reducing such overhead potentially harms load balance, thus degrading service performance. To address these issues, we propose an innovative layer-aware concurrent containerized MS scheduling framework to jointly trade off both consumers’ QoS (latency) and service provider’s profit (load balance). Specifically, we first model the scheduling problem as a multi-objective Markov Decision Process, fully accounting for the latency of each phase in MS’s lifecycle and various perceptible affinities related to the IoT application’s Directed Acyclic Graph. Secondly, a multi-objective DRL algorithm (MO-PPO) is proposed by reconfiguring the Actor-Critic network and devising an empirical trajectory sampling balance mechanism, enabling only a single trained model to approximate the Pareto front. Finally, extensive experiments based on real-world data traces show MO-PPO reduces the service latency by 31.8% and load imbalance by 38.7% at the optimal trade-off point compared to the baselines.

Federated Edge Learning (FEL) emerges as a pioneering distributed machine learning paradigm for the 6 G Hyper-Connectivity, harnessing data from the IoT devices while upholding data privacy. However, current FEL algorithms struggle with non-independent and non-identically distributed (non-IID) data, leading to elevated communication costs and compromised model accuracy. To address these statistical imbalances, we introduce a clustered data sharing framework, mitigating data heterogeneity by selectively sharing partial data from cluster heads to trusted associates through sidelink-aided multicasting. The collective communication pattern is integral to FEL training, where both cluster formation and the efficiency of communication and computation impact training latency and accuracy simultaneously. To tackle the strictly coupled data sharing and resource optimization, we decompose the optimization problem into the clients clustering and effective data sharing subproblems. Specifically, a distribution-based adaptive clustering algorithm (DACA) is devised basing on three deductive cluster forming conditions, which ensures the maximum sharing yield. Meanwhile, we design a stochastic optimization based joint computed frequency and shared data volume optimization (JFVO) algorithm, determining the optimal resource allocation with an uncertain objective function. The experiments show that the proposed framework facilitates FEL on non-IID datasets with faster convergence rate and higher model accuracy in a resource-limited environment.

With the prevalence of real-time intelligence applications, online edge learning (OEL) has gained increasing attentions due to the ability of rapidly accessing environmental data to improve artificial intelligence models by edge computing. However, the performance of OEL is intricately tied to the dynamic nature of incoming data in ever-changing environments, which does not conform to a stationary distribution. In this work, we develop a data importance-aware collection, communication, and computation integration framework to boost the training efficiency by leveraging the varying data usefulness under dynamic network resources. A model convergence metric (MCM) is firstly derived that quantifies the data importance in mini-batch gradient descent (MGD)-based online learning tasks. To expedite model learning at the edge, we optimize training batch configuration and fine-tune the acquisition of important data through coordinated scheduling, encompassing data sampling, transmission and computational resource allocation. To cope with the time discrepancy and complex coupling of decision variables, we design a two-timescale hierarchical reinforcement learning (TTHRL) algorithm decomposing the original problem into two-layer subproblems and separately optimize the subproblems in a mixed timescale pattern. Experiments show that the proposed data integration framework can effectively improve the online learning efficiency while stabilizing caching queues in the system.

Mobile Edge Computing (MEC) holds a promising paradigm shift, facilitating the migration of cloud resources from the network core to the network edge to bolster the support for high-quality time-sensitive applications. However, due to the dynamic and resource-constrained nature of MEC systems, there is a desirability for exploring service migration issues. In this paper, we investigate the pre-copy-based migration of stateful containerized services, emphasizing the diverse user requirements and the unique characteristics of containers. Traditional pre-copy schemes often rely on fixed downtime thresholds or predefined iteration rounds to terminate the iterative copying process, which could potentially lead to a slow migration process and ultimately diminish migration efficiency. To address this issue, we propose a dynamic model formulation for the service migration problem tailored to accommodate various user requirements, and devise an enhanced pre-copy migration strategy based on the Particle Swarm Optimization (PSO) named PSO-PCM. Moreover, for seamless migration of services, the container’s Union File System (UFS) is also considered in this paper. Simulation results indicate that our proposed scheme outperforms the baseline schemes, significantly reducing total migration time and enhancing migration efficiency.

Abstract Human action understanding (HAU) is a broad topic that involves specific tasks, such as action localisation, recognition, and assessment. However, most popular HAU datasets are bound to one task based on particular actions. Combining different but relevant HAU tasks to establish a unified action understanding system is challenging due to the disparate actions across datasets. A large‐scale and comprehensive benchmark, namely SkatingVerse is constructed for action recognition, segmentation, proposal, and assessment. SkatingVerse focus on fine‐grained sport action, hence figure skating is chosen as the task object, which eliminates the biases of the object, scene, and space that exist in most previous datasets. In addition, skating actions have inherent complexity and similarity, which is an enormous challenge for current algorithms. A total of 1687 official figure skating competition videos was collected with a total of 184.4 h, exceeding four times over other datasets with a similar topic. SkatingVerse enables to formulate a unified task to output fine‐grained human action classification and assessment results from a raw figure skating competition video. In addition, SkatingVerse can facilitate the study of HAU foundation model due to its large scale and abundant categories. Moreover, image modality is incorporated for human pose estimation task into SkatingVerse . Extensive experimental results show that (1) SkatingVerse significantly helps the training and evaluation of HAU methods, (2) the performance of existing HAU methods has much room to improve, and SkatingVerse helps to reduce such gaps, and (3) unifying relevant tasks in HAU through a uniform dataset can facilitate more practical applications. SkatingVerse will be publicly available to facilitate further studies on relevant problems.

In the forthcoming sixth-generation (6G) era, edge-network-cloud collaboration is needed to support artificial intelligence as a service (AIaaS) with a strong demand for computing power. However, how to guarantee the Quality of AI Service (QoAIS) and utilize the edge-network-cloud collaboration to enhance the performance of AI service is a big challenge. In this paper, we propose an AI service management and network resource scheduling architecture based on human-like networking. Considering the Quality of Service (QoS) requirements and AI tasks, we propose a joint AI agent placement with deep neural network (DNN) deployment and dynamic bandwidth resource allocation algorithm (JAAPD-D). JAAPD-D is proposed to solve the short-term and long-term joint resource allocation problem which includes communication, computation, and memory resources in the network. We adjust the agent placement, DNN deployment, and schedule routing path to ensure effective service transmission in the long time interval and dynamically allocate bandwidth resources in the short time interval. We use Lyapunov optimization to ensure the system stability of the whole network, meet the QoS requirements of various services, and minimize the average end-to-end delay of services. Simulation results show that JAAPD-D outperforms existing algorithms in terms of delay, traffic accepted rate, network system throughput, and cost.

Multi-human parsing is an image segmentation task necessitating both instance-level and fine-grained category-level information. However, prior research has typically processed these two types of information through distinct branch types and output formats, leading to inefficient and redundant frameworks. This paper introduces UniParser, which integrates instance-level and category-level representations in three key aspects: 1) we propose a unified correlation representation learning approach, allowing our network to learn instance and category features within the cosine space; 2) we unify the form of outputs of each modules as pixel-level results while supervising instance and category features using a homogeneous label accompanied by an auxiliary loss; and 3) we design a joint optimization procedure to fuse instance and category representations. By unifying instance-level and category-level output, UniParser circumvents manually designed post-processing techniques and surpasses state-of-the-art methods, achieving 49.3% AP on MHPv2.0 and 60.4% AP on CIHP. We have released our source code, pretrained models, and demos to facilitate future studies on https://github.com/cjm-sfw/Uniparser.

Anomaly detection plays an essential role in network security and traffic classification. Many studies have focused on anomaly detection to improve network security, including machine learning and deep learning methods. These methods often require numerous samples and must obtain the results by classifying the entire data set, thereby limiting their inflexibility. Although transfer and multitask learning have achieved some results in the model’s transferability, these methods must manually label or reprocess the test set. These problems limit the application of previous methods in network security management. To solve these problems, we propose a transferable and adaptable network intrusion detection system (TA-NIDS) based on deep reinforcement learning. The interaction process between the agent and the environment varies every time. A small-scale data set can be used to produce many interactive processes. Therefore, robustness is guaranteed when there are few samples. Then, a reasonable reward function allows the agent to learn how to first choose outliers without classifying the entire data set. This makes the TA-NIDS more adaptable to the scene when we prioritize apparent outliers. More importantly, the original features are transformed into the state of the environment, so no requirement exists for the feature dimension. Furthermore, the general rather than the specific state of one data set makes the model transferable to other data sets. The experimental results for IDS2017, IDS2018, NSL-KDD, UNSW-NB15 and CIC-IoT2023 show that the proposed framework maintains good accuracy when prioritizing outliers and transferability are prioritized simultaneously.

Most existing downlink channel estimation methods rely on channel sparsity in the angular domain to reduce pilot overhead for massive multiple-input multiple-output (MIMO) systems. Compared with channel estimation methods based on predefined basis or offline dictionary learning, in this paper, we design an online two-timescale joint uplink/downlink dictionary learning and channel estimation (TTS-JDLCE) algorithm for dynamic scenarios, where the channel sparsity and angle reciprocity between uplink and downlink transmissions are both exploited to reduce the pilot overhead. The downlink channel estimation is constructed as a TTS stochastic optimization problem with a constraint coupled by the long-term dictionary and short-term sparse channel representations. Treating the dictionary as a learnable parameter, the proposed algorithm can capture dynamic spatial information for enhancing performance. By introducing a relaxed TTS primal-dual decomposition (PDD) framework, the original problem is decomposed into a long-term online dictionary learning subproblem and a family of short-term sparse channel estimation subproblems. Besides, the deep unfolding technique is employed to extract gradient information from short-term problems, which circumvents the non-closed form and non-convexity of long-term subproblem by constructing a convex surrogate problem. Finally, simulations show that the proposed method remarkably reduces the pilot overhead and achieves significant performance gains over various baselines.

Federated Learning (FL) is a novel distributed machine learning approach to leverage data from Internet of Things (IoT) devices while maintaining data privacy. However, the current FL algorithms face the challenges of non-independent and identically distributed (non-IID) data, which causes high communication costs and model accuracy declines. To address the statistical imbalances in FL, we propose a clustered data sharing framework which spares the partial data from cluster heads to credible associates through device-to-device (D2D) communication. Moreover, aiming at diluting the data skew on nodes, we formulate the joint clustering and data sharing problem based on the privacy-preserving constrained graph. To tackle the serious coupling of decisions on the graph, we devise a distribution-based adaptive clustering algorithm (DACA) basing on three deductive cluster-forming conditions, which ensures the maximum yield of data sharing. The experiments show that the proposed framework facilitates FL on non-IID datasets with better convergence and model accuracy under a limited communication environment.

Online learning has been widely applied in real-time artificial intelligence (AI) applications to learn new classes from the dynamic environment. Although the deployment of AI model training over the edge can facilitate faster processing of real-time data, the learning efficiency is plagued by the limited capacity of distributed data acquisition. In fact, not all data samples are equally important, and the random data selection strategy is not beneficial to accelerate training due to redundant data processing. In this paper, we present an importance-driven data collection framework, which leverages the usefulness of important data to improve the learning efficiency over the wireless edge. Specifically, the novel model convergence metric (MCM) is constructed to evaluate the data importance dynamically for model learning. Moreover, considering the constraint of limited network resources on learning efficiency, we establish an MCM maximization problem of joint data collecting, scheduling, and feeding in an edge computing system. A two-timescale hierarchical reinforcement learning (TTHRL) algorithm is designed to decouple the original problem into two-timescale two-level subproblems, where the top-level agent is responsible for data feeding strategy in the long term and the low-level agent learns data scheduling and collecting strategy in the short term. Simulation results show that our proposed scheme can achieve better performance improvements over the baseline schemes.

Network pruning is one of the most dominant model compression methods today, which can be broadly divided into filter pruning and weight pruning. Unlike filter pruning that deletes the whole filters thus prone to cause unrecoverable accuracy loss, weight pruning removes the single weights at a fine-grained level, which effectively avoids this problem. However, weight pruning leads to unstructured sparsity and is thus incompatible with general platforms. To address such limitation, we propose Splittable Pattern-Specific Weight Pruning(SPWP) to achieve both compression and compatibility, consisting of Patterned Weight Searching(PWS) and Kernel Equivalent Splitting(KES). Specifically, we study the intrinsic features of convolution kernels and devise PWS to prune weights in regular shapes based on such features. During inference, KES equivalently splits the pruned sparse kernel into parallel branches according to the linear additivity of convolution, allowing the network to be accelerated on general platforms. Extensive experiments of different models on various datasets demonstrate the superior performance of our method. For example, SPWP can prune 60.1% total FLOPS of ResNet-56 on CIFAR-10 with even a 0.08% of top-1 accuracy increase.

Network pruning is effective in reducing memory usage and time complexity. However, current approaches face two common limitations. 1) Pruned filters cannot contribute to the final outputs, resulting in severe performance drops, especially at large pruning rates. 2) It requires time-consuming and computationally expensive fine-tuning to recover accuracy. To address these limitations, we propose a novel filter pruning method called Cluster Pruning (CP). Instead of directly deleting filters, CP reconstructs them based on their intra-similarity and removes them using the proposed channel addition operation. CP preserves all learned features and eliminates the need for fine-tuning. Specifically, each filter is distinguished by clustering and reconstructed as the centroid to which it belongs. Reconstructed filters are updated to prevent erroneous selections. After convergence, filters can be safely removed through the channel addition operation. Experiments on various datasets show that CP achieves the best trade-off between performance and complexity compared with other algorithms.

This work studies the multi-human parsing problem. Existing methods, either following top-down or bottom-up two-stage paradigms, usually involve expensive computational costs. We instead present a high-performance Single-stage Multi-human Parsing (SMP) deep architecture that decouples the multi-human parsing problem into two fine-grained sub-problems,i.e., locating the human body and parts. SMP leverages the point features in the barycenter positions to obtain their segmentation and then generates a series of offsets from the barycenter of the human body to the barycenters of parts, thus performing human body and parts matching without the grouping process. Within the SMP architecture, we propose a Refined Feature Retain module to extract the global feature of instances through generated mask attention and a Mask of Interest Reclassify module as a trainable plug-in module to refine the classification results with the predicted segmentation. Extensive experiments on the MHPv2.0 dataset demonstrate the best effectiveness and efficiency of the proposed method, surpassing the state-of-the-art method by 2.1% in AP50p, 1.0% in APvolpsup>, and 1.2% in PCP50. Moreover, SMP also achieves superior performance in DensePose-COCO, verifying generalization of the model. In particular, the proposed method requires fewer training epochs and a less complex model architecture. Our codes are released in https://github.com/cjm-sfw/SMP.

Weight pruning is a well-known technique used for network compression. In contrast to filter pruning, weight pruning produces higher compression ratios as it is more fine-grained. However, pruning individual weights results in broken kernels, which cannot be directly accelerated on general platforms, leading to hardware compatibility issues. To address this issue, we propose Shift Pruning (SP), a novel weight pruning method that is compatible with general platforms. SP converts spatial convolutions into regular 1 X 1 convolutions and shift operations, which are simply memory movements that do not require additional FLOPs or parameters. Specifically, we decompose the original K X K convolution into parallel branches of shift-convolution operations and devise the Differentiable Shift Operator (DSO), an approximation form of the actual shift operation, to automatically learn the crucial directions for adequate spatial interactions with the designed shift-related loss function. A regularization term is proposed to prevent redundant shifting, which is beneficial for low-resolution situations. To further improve inference efficacy, we develop a post-training transformation that can construct a more compact model. The introduced channel-wise slimming allows SP to prune in a hybrid-structural manner, catering for both hardware compatibility and a high compression ratio. Extensive experiments on the CIFAR-10 and ImageNet datasets demonstrate that our proposed method achieves superior performance in both accuracy and FLOPs reduction compared to other state-of-the-art techniques. For instance, on ImageNet, we can reduce 48.8% of total FLOPs on ResNet-34 with only 0.22% Top-1 accuracy drop.

Multi-person pose estimation in crowded scenes remains a very challenging task. This paper finds that most previous methods fail to estimate or group visible keypoints in crowded scenes rather than reasoning invisible keypoints. We thus categorize the crowded scenes into entanglement and occlusion based on the visibility of human parts and observe that entanglement is a significant problem in crowded scenes. With this observation, we propose DecenterNet, an end-to-end deep architecture to perform robust and efficient pose estimation in crowded scenes. Within DecenterNet, we introduce a decentralized pose representation that uses all visible keypoints as the root points to represent human poses, which is more robust in the entanglement area. We also propose a decoupled pose assessment mechanism, which introduces a location map to adaptively select optimal poses in the offset map. In addition, we have constructed a new dataset named SkatingPose, containing more entangled scenes. The proposed DecenterNet surpasses the best method on SkatingPose by 1.8 AP. Furthermore, DecenterNet obtains 71.2 AP and 71.4 AP on the COCO and CrowdPose datasets, respectively, demonstrating the superiority of our method. We will release our source code, trained models, and dataset to facilitate further studies in this research direction. Our code and dataset are available in https://github.com/InvertedForest/DecenterNet.

As the key technology of artificial intelligence(AI), Deep Neural Networks (DNNs) have been widely used in mobile applications, such as video analytics in autonomous driving. However, due to the constrained computation capabilities on mobile devices (MDs), it is challenging to meet the critical accuracy and real-time demand of DNN tasks, which would result in a serious drop in quality of service (QoS). A popular alternative is to offload DNN tasks to edges for intelligence inference, nevertheless, this results in a heavy communication burden due to large amounts of raw data. In this paper, we propose an adaptive DNN co-Inference (ADCI) strategy which obtains the flexible computation division among devices and edge servers with elastic execution by combining the early exit and model partition policies. Establishing a balanced utility function, we jointly optimize dynamic offloading and model adoption while taking into account the multi-user and multi-server edge computing system. To tackle the high coupling among mixed variables, we propose a two-stage deep reinforcement learning (DRL) algorithm. The early-exit and model partition decisions are tracked using the Lagrange method as a soft option. Results from simulations show that the ADCI strategy performs well with timely accuracy

In this paper, we design a two-timescale approach for joint uplink/downlink (UL/DL) dictionary learning and channel estimation (TTS-JDLCE) for frequency division multiplexing (FDD) massive multiple-input multiple-output (MIMO) systems in dynamic scenarios. With channel sparsity and angle reciprocity between UL and DL transmissions, the joint UL/DL dictionary is regarded as a learnable parameter to capture dynamic spatial information at a slower timescale than instantaneous downlink channel estimation. By introducing the primal-dual decomposition (PDD) framework, the original non-convex two-timescale stochastic optimization problem is decomposed into a long-term online dictionary learning subproblem and a family of short-term sparse channel estimation subproblems, and then solved in a divide-and-conquer manner, which can converge to the stationary solutions of the original problem over time. Finally, simulations show that the proposed method remarkably reduces the pilot overhead and achieves significant performance gain over various baselines.

Recently, the applications of deep neural network (DNN) have been very prominent in many fields due to its superior feature extraction performance. However, the high-dimension parameter model and large-scale mathematical calculation restrict the execution efficiency, especially for the Internet of Things (IoT) devices. Different from the previous cloud/edge-only pattern that brings significant pressure for uplink communication and device-only fashion that undertakes unaffordable calculation strength, we highlight the collaborative computation between the device and edge for DNN models, which can achieve a good balance between the communication load and execution accuracy. Specifically, a systematic on-demand co-inference framework is proposed to exploit the multi-branch structure, in which the pre-trained Alexnet is right-sized through early-exit and partitioned at an intermediate DNN layer. The integer quantization is enforced to further compress transmission bits. As a result, we establish a new Deep Reinforcement Learning (DRL) optimizer-Soft Actor Critic for discrete (SAC-d), which generates the exit point, partition point, and compressing bits by soft policy iterations. Based on the latency and accuracy aware reward design, such an optimizer can well adapt to the complex environment like dynamic wireless channel and arbitrary CPU processing, and is capable of supporting the 5G URLLC. Real-world experiment on Raspberry Pi 4 and PC shows the effective performance of the proposed solution.

Based on the trustless feature of blockchain, this article designs a general configurable blockchain-enabled smart manufacturing system to achieve flexible manufacturing in response to large-scale manufacturing services. With a transaction pool containing all the pending manufacturing tasks but aligning with the logic flow, the complex manufacturing structure can be uniformly tackled. Furthermore, in virtue of the contradiction between large-scale manufacturing and limited blockchain throughput, we formulate a joint optimization of the block size, task scheduling, and the supply-demand configuration to maximize the customers’ net profit with the probabilistic delay requirements, which addresses the critical issue of efficiency and latency in the blockchain-based live manufacturing process. Meanwhile, the production quality and price preference are involved. For solution, a mixed online bipartite matching-based DQN algorithm is proposed, which circumvents the high dimensionality by separating the task-manufacturer matching from the time-correlated problem. Simulation results show that the proposed flexible framework can well adopt to dynamic customer population, and achieves better convergence.

Immutability, decentralization, and linear promoted scalability make the sharded blockchain a promising solution, which can effectively address the trust issue in the large-scale Internet of Things (IoT). However, currently, the throughput of sharded blockchains is still limited when it comes to high proportion of cross-shard transactions (CSTs). On the other hand, the assemblage characteristic of the collaborative computing in IoT has not been received attention. Therefore, in this article, we present a clustering-based sharded blockchain strategy for collaborative computing in the IoT, where the sharding of the blockchain system is implemented in two steps:K-means-clustering-based user grouping and the assignment of consensus nodes. In this framework, how to reasonably group the IoT users while simultaneously guaranteeing the system performance is the key point. Specifically, we describe the data transactions among IoT devices by data transaction flow graph (DTFG) based on a dynamic stochastic block model. Then, formed as a Markov decision process (MDP), the optimization of the cluster number (shard number) and the adjustment of consensus parameters are jointly trained by deep reinforcement learning (DRL). Simulation results show that the proposed scheme improves the scalability of the sharded blockchain in the IoT application.

In massive multiple-input multiple-output (MIMO) systems employing hybrid analog-digital precoder, there are two commonly used antenna structures that have their own pros and cons, namely, the fully-connected antenna structure (FCAS) and the partially-connected antenna structure (PCAS). The FCAS achieves better spectrum efficiency (SE) even with reduced radio frequency (RF) chains, but the hardware cost and power consumption are still grievous. Contrarily, the PCAS has less hardware cost and power consumption, but suffers from severe performance loss. In this paper, by combining the advantages of both structures, we first propose a sparse adaptive antenna structure (SAAS) for the implementation of the hybrid precoder, which can jointly control the on/off state of all phase shifters (PS) and RF chains through a switch network. Then, a sparse hybrid precoding (SHP) optimization problem based on the proposed SAAS is established aiming at minimizing the total power consumption under individual average data rate requirements. To tackle the challenging non-smooth non-convex stochastic optimization (NSO) emerged with the SHP design and reduce the power consumption of PSs and RF chains, we propose a sparse smooth approximation based an online algorithm to find a stationary point of the NSO problem and establish its convergence. Simulations verify that the proposed antenna structure and algorithm achieve a better balance between power consumption and system throughput than the existing schemes.

Nowadays, blockchain has become a promising tamper-evident and tamper-resistant distributed ledger technology that achieves the security and privacy through the cryptography, consensus mechanism and chained data structure. In this paper, we propose a general blockchain-based smart manufacturing system (BSM) that utilizes the decentralization, immutability, auditability of blockchain to achieve flexible manufacturing that responds to on-demand services in time. For management, manufacturing services are divided into tasks and for unified scheduling, these tasks are queued along the logical flow in the transaction pool. Considering the contradiction between large scale manufacturing and limited transactional throughput, a joint task scheduling over blockchains and supply-demand configuration design is proposed to obtain the maximum customers' net profit while balancing the timeliness as well as the production and blockchain payoff. Moreover, a maximum weight matching based Alternating Optimization framework (MWMAO) is proposed as the solution. Simulation results show that the proposed framework has superiority on the profitability.

In massive MIMO systems, the effective channel state information (CSI) is an essential prerequisite for the beamforming (BF) design. While there is a handful of previously proposed compressive sensing (CS)-based channel estimation algorithms in the literature, the large BF gain provided by the massive MIMO array has not been fully exploited in the channel estimation stage. In order to obtain higher BF gain and better estimation performance with less pilot overhead, we study the joint design of the transmitting and receiving hybrid pilot BF as well as the associated channel tracking scheme. Specifically, a Markov prior is used to model the temporal correlation in massive MIMO channels over different time slots. Then, the hybrid pilot BF is optimized by maximizing the mutual information between the channel measurements and the corresponding downlink sparse channels with the Markov prior. Following, we derive an efficient channel tracking algorithm called Turbo Bayesian Inference (Turbo-BI) to solve the resulting CS problem and generate the channel prior information required to calculate the mutual information for the optimization of hybrid pilot BF in the next time slot. The proposed Turbo-BI can exploit both the sparsity and the temporal correlation of massive MIMO channels to enhance the estimation performance. Finally, simulations show that our proposed algorithm can achieve significant gain over the existing state-of-the-art baselines.

To utilize the potential commutating resources of parked vehicles (PVs) in the large parking lot, we design a large-scale parked vehicular computing system via software defined blockchain. However, the parking time for PVs is uncertain and some computational services have delay requirements. Therefore, in this paper, we propose a delay-sensitive joint blockchain parameters and resource optimization framework including block size and block generation time, as well as the offloading strategy and computing frequency adjustment. Such a design causes the problem to be highly coupled and non-convex, for which we use an alternating optimization (AO) strategy and perform multiple transformations to ensure convexity. Finally, the simulation results show the effectiveness of the proposed scheme.

The boom of video streaming industry has resulted in the increasing demands for transcoding services from heterogeneous users. Recent advances of blockchain technology allow some startups to realize decentralized collaborative transcoding through device-to-device (D2D) networks, where a group of transcoders are selected to perform transcoding cooperatively. For the blockchain-enabled D2D transcoding systems, it's imperative to jointly design transcoder selection, task scheduling and resource allocation schemes in order to provide efficient and trustworthy transcoding services. In this paper, viewing the involved multi-dimensional complex factors and channel fluctuation, we propose a novel deep reinforcement learning (DRL) based transcoder selection framework for blockchain enabled D2D transcoding systems where both the platform dynamics and channel statistics are captured. To reduce the action space size, we adopt a two-stage decision approach to first select the transcoders through a normal DRL based framework and then obtain the optimal task scheduling, power control, and resource allocation scheme by solving a stochastic optimization problem with the constrained stochastic successive convex approximation (CSSCA) approach. Simulation results show that our proposed framework can achieve high transcoding revenue while meeting the quality of service (QoS) requirements, and it can well handle dynamic cases.

In this paper, joint estimation for channel and Inphase/Quadrature imbalance (IQI) is investigated in the downlink Frequency Division Duplexing (FDD) massive multiple-input multiple-output (MIMO) systems. First, exploiting the sparsity of massive MIMO channels and the timescale separation of channels and IQI, we derive a two-timescale sparse maximum a posterior (MAP) formulation for the joint estimation, where the IQI parameter is the long- term variable and the sparse channel is the short- term variable. Then we propose a two-timescale online joint sparse estimation (TOJSE) algorithm to solve the problem, which can converge to the stationary solutions of the original two-timescale non-convex stochastic optimization problem over time. Finally, simulations show that our proposed TOJSE algorithm can achieve significant gain over various baselines.

The rapid development of Internet of Vehicles (IoV) necessitates a secure and reliable infrastructure to store and share the massive data. Blockchain, a distributed and immutable ledger, is widely considered as a promising solution to ensure data security and privacy for IoV. To deal with the massive IoV data, the scalability of blockchain becomes a critical issue, which should maximize transactional throughput as well as handling the dynamics of IoV scenarios. Therefore, this paper proposes a novel deep reinforcement learning (DRL) based performance optimization framework for blockchain-enabled IoV, where transactional throughput is maximized while guaranteeing the decentralization, latency and security of the underlying blockchain system. In this framework, we first carry out the performance analysis for blockchain systems from the aspects of scalability, decentralization, latency and security. Further, DRL technique is adopted to select block producers and adjust block size and block interval to adapt to the dynamics of IoV scenarios. Simulation results show that our proposed framework can effectively improve the throughput of blockchain-enabled IoV systems without affecting other properties.

In this paper, we proposed a D2D assisted hybrid framework, where the computation-intensive task can be offloaded to the cloud or neighboring users. Aiming at minimizing the total energy consumption under delay constraints, the joint optimization of the task offloading, task scheduling and computing resource allocation problem is formulated. Additionally, we model the duration of users' intermittent connections to study the effect of user mobility on the task success rate. Then, the original problem is divided into two sub-problems to solved separately considering the coupling multiplicative variables. Further, a flexible proximal alternating direction method of multipliers (ADMM) based algorithm is proposed to solve the nonconvex sub-problem in a distributed way. Numerical results reveal the effectiveness of the algorithm on convergence and complexity reduction, and the proposed scheme achieves excellent performance when compared with other conventional schemes.

In this study, a deep reinforcement learning (DRL) method was employed to solve the joint optimization problem for user association, resource allocation, and power allocation in heterogeneous networks (HetNets), which is an NP-hard problem. Existing studies have taken various optimization objectives into account. The heterogeneous network-deep-Q- network frame-work (HetDQN) is proposed to solve this type of optimization problem in HetNets. Based on maximum spectral efficiency, we designed a 6- layer deep neural network. The state space, objective function, and reward function are presented. In comparison with the existing solution, HetDQN can achieve a higher spectral efficiency. The simulation results revealed that HetDQN has better performance in term of convergence.

Recent advances in the industrial Internet of things (IIoT) provide plenty of opportunities for various industries. To address the security and efficiency issues of the massive IIoT data, blockchain is widely considered as a promising solution to enable data storing/processing/sharing in a secure and efficient way. To meet the high throughput requirement, this paper proposes a novel deep reinforcement learning (DRL)-based performance optimization framework for blockchain-enabled IIoT systems, the goals of which are threefold: 1) providing a methodology for evaluating the system from the aspects of scalability, decentralization, latency, and security; 2) improving the scalability of the underlying blockchain without affecting the system's decentralization, latency, and security; and 3) designing a modulable blockchain for IIoT systems, where the block producers, consensus algorithm, block size, and block interval can be selected/adjusted using the DRL technique. Simulations results show that our proposed framework can effectively improve the performance of blockchain-enabled IIoT systems and well adapt to the dynamics of the IIoT.

Blockchain-based video streaming systems aim to build decentralized peer-to-peer networks with flexible monetization mechanisms for video streaming services. On these blockchain-based platforms, video transcoding, which is computationally intensive and time-consuming, is still a major challenge. Meanwhile, the block size of the underlying blockchain has significant impacts on the system performance. Therefore, this paper proposes a novel blockchain-based framework with an adaptive block size for video streaming with mobile edge computing (MEC). First, we design an incentive mechanism to facilitate collaboration among content creators, video transcoders, and consumers. In addition, we present a block size adaptation scheme for blockchain-based video streaming. Moreover, we consider two offloading modes, i.e., offloading to the nearby MEC nodes or a group of device-to-device (D2D) users, to avoid the overload of MEC nodes. Then, we formulate the issues of resource allocation, scheduling of offloading, and adaptive block size as an optimization problem. We employ a low-complexity alternating direction method of the multipliers-based algorithm to solve the problem in a distributed fashion. Simulation results are presented to show the effectiveness of the proposed scheme.

Although there has been a wide investigation on channel estimation in frequency-division duplex (FDD) massive multiple-input-multiple-output (MIMO) systems, the effect of imperfect radio frequency (RF) chains have been largely ignored. In this paper, we consider a downlink massive MIMO system with in-phase/quadrature imbalance (IQI) at the base station (BS). Focusing on the joint estimation for channel and IQI, we model the joint estimation problem as a two-timescale non-convex optimization based on maximum a posteriori (MAP) estimate, where the IQI parameter is treated as the long-term variable and the sparse channel vector is short-term. We propose a batch algorithm and a two-timescale online joint sparse estimation (TOJSE) algorithm to solve the problem. The proposed batch algorithm utilizes all the previously received signals to update the current long-term variable, which can achieve better performance but with increasing computational complexity over time. In contrast, the TOJSE algorithm solves the short-term problem related to the current system state and constructs a recursive convex approximation to update the long-term variable in each iteration. Thus, the memory requirements and computational complexity of the TOJSE are remarkably reduced. Moreover, for the low mobility regime, a dynamic TOJSE algorithm is further presented to exploit the temporal correlation of channel support. Finally, the simulations show that our proposed algorithms can achieve significant gain over various baselines.

In this paper, we propose a joint dynamic pilot and data power allocation scheme for time division duplex (TDD) massive multiple-input multiple-output (MIMO) systems, so as to both adaptively mitigate pilot contamination and balance the mutual interference. Due to the unknown of instant channel state information before pilots, we exploit the Gauss-Markov process of temporally-correlated channels and use the Kalman filter to not only filter out the pilot contamination but also provide the priori estimation values. Subsequently, the deterministic approximation of the rate is derived as a function of the priori channel estimation and the priori estimate errors, and accordingly the rate-profile maximization to achieve max-min fairness is formulated. To deal with this optimization coupled across the pilot power and data power as well as the users, we give an iterative alternating rate-suboptimal algorithm composed of two sub-problems, both of which are further solved by introducing the successive convex approximation (SCA) methods and slack variables. Numerical results confirm the improved rate provided by the proposed scheme.

Recently, a two-timescale hybrid precoding (THP) scheme has been proposed to reduce the implementation cost of massive MIMO systems. In THP, the MIMO precoder consists of a high- dimensional RF precoder adaptive to the channel statistics and a low-dimensional baseband precoder adaptive to the instantaneous effective channel state information (CSI). Since the channel statistics changes at a slow timescale and is approximately the same for different subbands, only a single RF precoder is required to cover all subbands over a long term, which helps to reduce the hardware cost and implementation complexity of RF precoder. Moreover, the CSI signaling overhead can also be reduced. In this paper, we consider power minimization for massive MIMO systems with THP and individual average rate constraints. Due to the two-timescale design and individual average rate constraints, the problem is a challenging non-convex stochastic optimization problem. We propose an online constrained stochastic successive convex approximation (CSSCA) algorithm to find a stationary point of the power minimization problem. Simulations show that the proposed algorithm achieves significant gains over various baseline algorithms.

Mobile edge computing (MEC) is an emerging paradigm that mobile devices can offload the computation-intensive or latency-critical tasks to the nearby MEC servers, so as to save energy and extend battery life. Unlike the cloud server, MEC server is a small-scale data center deployed at a wireless access point, thus it is highly sensitive to both radio and computing resource. In this paper, we consider an Orthogonal Frequency-Division Multiplexing Access (OFDMA) based multi-user and multi-MEC-server system, where the task offloading strategies and wireless resources allocation are jointly investigated. Aiming at minimizing the total energy consumption, we propose the joint offloading and resource allocation strategy for latency- critical applications. Through the bi-level optimization approach, the original NP-hard problem is decoupled into the lower-level problem seeking for the allocation of power and subcarrier and the upper-level task offloading problem. Simulation results show that the proposed algorithm achieves excellent performance in energy saving and successful offloading probability (SOP) in comparison with conventional schemes.

The robust beamforming problem in multiple-input single-output (MISO) downlink networks of simultaneous wireless information and power transfer (SWIPT) is studied in this paper. Adopting the time switching fashion to perform energy harvesting and information decoding respectively, we aim at maximizing the sum rate under imperfect channel state information (CSI) and the chance constraints of users' harvested energy. In view of the fact that the constraints for minimal harvested energy is not necessary to meet from time to time, this paper adopts chance constraint to model it and uses the Bernstein inequality to transform it into deterministic constraints equivalently. Recognizing the maximum sum rate problem of imperfect CSI as nonconvex problem, we transform it into finding the expectation of minimum mean square error (MMSE) equivalently in this paper, and an alternative optimization (AO) algorithm is proposed to decompose the optimization problem into two sub- problems: the transmit beamformer design and the division of switching time. The simulation results show the performance gains compared to non-robust state of the art schemes.

Caching the popular contents near users is an effective way to release the burden of the wireless networks and reduce the energy consumption of content service for delivery. In this paper, we consider a distributed way to cache content with different user preference in heterogeneous cellular networks (HetNets). Aiming at minimizing the energy consumption of the whole network, a joint content cache and delivery optimization problem is proposed. Considering the coupling multiplicative variables, we utilize the alternative optimization (AO) algorithm to decompose the original problem into the content cache and delivery problems, which are solved separately through the knapsack solution and message passing (MP) algorithm. Numerical results reveal that the proposed scheme achieves more in energy saving than conventional schemes.

Wireless sensor networks (WSNs) are an important part of the Internet of Things (IoT). In WSNs, the sensors may act as an information collector as well as relay. This trait will cause the sensors that around the Sink easily run out of battery due to heavy traffic-flow, which is called energy hole problem and limits the network lifetime dramatically. To cope with this problem, we derive an optimal hop distance in multi-hop environment and get a conclusion that the optimal receive signal-to-noise is fixed under the large scale fading model. Then, we get a relationship between transmission power and the optimal sensor's coverage. We proposed a new method to improve WSNs' lifetime by using the optimal hop distance and changing the sensor's coverage according to the distance between sensors and the Sink. Simulations show that the proposed method can improve the lifetime and energy efficient of WSNs. Our research can also be used in SDN to help the controllers make decisions.

In this paper, multi-cell power allocation approach is researched. Different from the traditional optimization decomposition method, Deep Reinforcement Learning (DRL) method is employed to solve the power allocation issue which is an NP-hard problem. The objective of our work is to maximize the overall capacity of the entire network in the scenario where the base stations are randomly and densely distributed. We propose a wireless resource mapping method and a deep neural network for multi-cell power allocation named as Deep-Q-Full-Connected-Network (DQFCNet). Compared with the water-filling power allocation and Q-learning method, DQFCNet can achieve a higher overall capacity. Furthermore, the simulation results show that DQFCNet has significant improvement in convergence speed and stability.

This paper considers the queue‐aware optimal energy minimisation sparse beamforming design (QESB)in a downlink cloud radio access network (C‐RAN) system where multi‐RRHcommunicates with multi‐user through a central computing cloud via digitalfront‐haul links. The problem is formulated as the joint optimisation problem ofthe transmission energy consumption, system queue length and front‐haul costwith sparse beamforming design. As we know, the beamforming design adaptive toboth QSI and CSI is challenging because of the high complexity. Apart fromprevious works that take queue length as constraints, in this paper we directlyminimise the queue length state involved joint optimisation problem with SINRconstraints. A smooth function is proposed to approximate the ‐norm function which is discrete andnon‐convex. To overcome the challenge due to the non‐convexity of theoptimisation problem, the semidefinite relaxation (SDR) technology is utilisedto convert the primitive problem into the difference of convex (DC) programmingproblem, and convex and concave procedure (CCP) algorithm is used to induce thesparsity of the beamforming control. The simulation results show that the schemeproposed by this paper can obtain a good tradeoff between system energyconsumption, queue length and front‐haul cost with SINR constraints in C‐RANsystem.

The asymmetry traffic between downlink (DL) and uplink (UL) in massive machine‐type communication (mMTC) systems is so prominent that it makes the traditionally fixed frame protocols insufficient to handle. Meanwhile, the dynamic time‐division duplexing (D‐TDD) is a promising and attractive technology since its number of time slots for the DL and UL can be asymmetric and adjusted dynamically. In the cellular and mMTC co‐existing network, to balance the discrepancy of the UL/DL ratio and alleviate the interference as well, in this study, the authors design a D‐TDD‐based transmission frame structure to first fulfil the basic transmission requirements of the human‐type communication (HTC) users with a low power almost blank subframe. Herein, the stochastic geometry methods are adopted to calculate spectral efficiencies of the HTC user equipment. Then, focusing on the worst queue state in mMTC, they devise the slot allocation problem with the min–max objective of the UL/DL queues and utilise the sub‐gradient descent (SGD) method for a solution. Simulation results show that the proposed traffic aware sub‐frame configuration is more appropriate for the dynamical asymmetry environment. Meanwhile, the adopted dynamic step size SGD algorithm can achieve a trade‐off between the worst‐case queue and the network throughput.

Considering the expensive radio frequency (RF) chain, huge training overhead, and feedback burden issues in massive MIMO, in this letter, we propose a mixed-timescale per-group hybrid precoding scheme under an adaptive partially connected antenna structure, where the RF precoder is implemented using an adaptive connection network (ACN) and M analog phase shifters (APSs), where M is the number of antennas at the base station. Exploiting the mixed time stage channel state information (CSI) structure, the joint-design of ACN, and APSs is formulated as a statistical signal-to-leakage-and-noise ratio maximization problem, and a heuristic group RF precoding algorithm is proposed to provide a near-optimal solution. Simulation results show that the proposed design advances at better energy efficiency and lower hardware cost, CSI signaling overhead and computational complexity than the conventional hybrid precoding schemes.

To satisfy the enormous data transmission demands, caching popular files into memories at user terminals (UTs) is a promising solution, which can alleviate the heavy burden on backhaul links and shorten the transmission delay. In this paper, we study the user caching strategy by exploiting the effect of user mobility. The contact of mobile users is modeled as the Poisson process. Both Zipf distributed and uniform distributed file demands are considered in the caching strategy. To improve user's quality of experience (QoE) for delay sensitive services, we define the user satisfaction metric in terms of the delay time, and maximize it through the proposed caching placement strategy. The mixed equality-inequality constrained optimization problem is solved by the multiplier penalty function (MPF) method. Numerical results reveal that the maximal average user satisfaction is achieved when the file caching coordinates with the file demand.

This paper conducts handover probability analysis in Coordinated multipoint (CoMP) based ultra-dense networks (UDNs) where ri access points (APs) jointly serve the user. To thoroughly investigate the effect of mobility on handover probability, we compare several handover schemes of CoMP coordinating set (CCS) handover and serving cell handover. From the perspective of CCS handover, three schemes are considered including “best ri”, “half best ri” and “none of the best ri”. Meanwhile, from the perspective of the serving cell handover, this paper discusses two schemes: “not the best one” and “not any of the best ri”. With stochastic geometry methods, theoretical expressions of CoMP handover probability with different schemes are derived. Further, simulation results illustrate that scheme “best ri” provides an upper bound while scheme “none of the best ri” exhibits as a lower bound of CoMP handover probability and other schemes lie between them. To sum up, such analysis may provide a good performance reference for the design of CoMP handover.

Next-generation cellular network may encounter unprecedented challenges, while cloud radio access network (C-RAN) is regarded as a promising network architecture to meet the explosive growth of data traffic. In C-RAN, all the signal processing is shifted to the cloud data center to realize joint resource allocation and interference management, which highly relies on the accuracy of channel state information (CSI). However, the assumption of perfect CSI can hardly be secured, especially in dense network scenario. Meanwhile, for dense CRAN, the network optimization will be further confined by curse of dimensionality. In this paper, we propose an adaptive RRH switch-off mechanism through an iterative reweighted sparse beamforming (IrSBF), which only depends on imperfect CSI and has certain endurance to the outage of S!NR. However, the non-convex ℓ0-norm optimization and probabilistic SINR constraints make the beamforming design highly intractable. Utilizing the Bernstein-type inequality to transform the chance constraints to conservative static constraints, we are able to solve the robust sparse beamforming problem through semidefinite relaxation (SDR) technology. Besides, ℓ1-norm approximation method is used to design the sparse beamforming vector for users iteratively. Simulation results suggest that the proposed IrSBF algorithm significantly outperforms the conventional coordinated beamforming (CB) solution and group sparse beamforming (GSBF) algorithm ignoring CS! errors, especially in the case of higher network density.

Deploying femtocells in macrocell network is an economical and reliable way to increase network coverage and efficiency. However, such deployment poses great challenges for the resource allocation of two tier femtocell networks due to the existence of inter-tier and intra-tier interference. We propose a uplink transmission resource allocation scheme with the goal of maximizing the total power efficiency. Nevertheless, the resource allocation optimization problem is NP-hard and non-convex, thereby, we adopt message passing (MP) method which approaches the optimality by iteratively passing the messages between each user equipments (UEs) and access points (APs). To reduce excessive number of iterations, we present two improved MP algorithms, i.e. best selection message passing (BSMP) algorithm as well as UE priority message passing (UPMP) algorithm. BSMP algorithm is optimal allocation algorithm which could avoid excessive number of iterations. Based on BSMP algorithm, UPMP algorithm take the UEs' priorities into consideration to ensure the fairness of resource allocation. Compared to existing schemes, simulation results show that improved MP algorithms have excellent convergences. Meanwhile, the vast improvements on power efficiency as well as weighted power efficiency have been demonstrated.

The cellular heterogeneous networks (CHNs) are of great interest for the great potential of improving the network capacity by employing low-power, easy- deployment short-range mini-base stations (BSs). To cope with the challenge of inter-layer/inner-layer interference environment in such heterogeneous architecture, we introduce cognitive Radio (CR) in CHNs, and named it as CR Enabled cellular heterogeneous networks (CCHNs). However, the transmission protocol deign is crucial for the implement of CCHSs. For one thing, the division of sensing and access phase is vital for the sensing accuracy and spectrum efficiency; For another, the power spend on the spectrum sensing cuts the transmission power allocation budget. In this paper, such coupled problem is solved by a Bi-level optimization method, by which the joint problem is decoupled to the upper level power allocation subproblem and lower level slot partition subproblem. Simulation results show that the proposed algorithm achieves the optimal jointing optimization of spectrum sensing and access, which alleviates the inter-layer interference between heterogeneous cells and improve the performance of entire network.

The renewable energy is expected as one of the promising option to reduce the CO2 emissions for the future wireless communication. This paper considers the power allocation of a multi-cell network where the base station powered by hybrid energy source, i.e., the energy is supplied by a constant energy source and an energy harvester. Due to the features of the renewable energy and multi-cell network, we need to develop an efficient power allocation scheme to utilize the renewable and constant energy efficiently. In order to make full use of hybrid energy, we formulate the power allocation to minimize the energy drawn from the constant source and to maximize the amount of transmitted data per energy consumption, which is shown to be a non-convex optimization problem. It is transformed into an equivalent convex optimization problem by exploiting the properties of fractional programming. Finally an optimal offline iterative algorithm is proposed to solve this problem. Simulation results show that the proposed algorithm can make full use of the harvested renewable energy and all the energy can be utilized in a more efficient mode, which can help to provide some valuable insights for more practical efficient online schemes.

Due to irregular deployment of small base stations (SBSs), the interference in cognitive heterogeneous networks (CHNs) becomes even more complex; in particular, the uncertainty of spectrum mobility aggravates the interference context. In this case, how to analyze system capacity to obtain a closed-form expression becomes a crucial problem. In this paper we employ stochastic methods to formulate the capacity of CHNs and achieve a closed-form expression. By using discrete-time Markov chains (DTMCs), the spectrum mobility with respect to the arrival and departure of macro base station (MBS) users is modeled. Then an integral method is proposed to derive the interference based on stochastic geometry (SG). Also, the effect of sensing accuracy on network capacity is discussed by concerning false-alarm and miss-detection events. Simulation results are illustrated to show that the proposed capacity analysis method for CHNs can approximate the conventional sum methods without rigorous requirement for channel station information (CSI). Therefore, it turns out to be a feasible and efficient way to capture the network capacity in CHNs.

Due to the spectrum-temporary challenges in the cognitive radio networks (CRNs), the pure divide-and-conquer strategies by layered principles strand. In this paper, our study presents a framework for cross-layer modeling design to jointly optimise the power and spectrum allocation, multi-path routing, data traffic, and QoS requirements in overlay CRNs. Because of the complexity of the optimization problem, vertical decomposition and distributed subgradient methods are applied. Firstly, the optimization problem is decoupled into two subproblems - a master problem related to QoS constrained flow control and routing selection, and a subproblem related to spectrum selection and physical power allocation - with dual decomposition method. Secondly, the master problem is further decomposed with the primal decomposition method, while the subproblem derived by a Lagrangian optimum with dual factors. Finally, simulation results show the implementability and efficiency of our proposed framework.

This paper presents a framework for coverage and capacity optimization (CCO) in Heterogeneous Cellular Networks (HCNs). Unlike conventional approaches, we propose a concept of effective capacity (EC) as the optimization objective of CCO, which involves the index of coverage in the form of truncation function. The optimization objective is a NP problem due to that the adjusting parameters are mixed with discrete and continuous, i.e., antenna electronic down-tilt is discrete, while the transmission power and bandwidth are continuous. In this paper, we improve bacterial foraging (BF) algorithm based on taking the HCNs performance analysis of coverage and capacity with percentile level to find an effective increasing direction. According to simulation, the results show that the optimization objective is feasible and the adjustment of percentile level achieves flexible solutions for CCO.

Carrier Aggregation (CA) is one of the promising techniques to support the wider bandwidth and fulfil the higher data rate requirement in LTE-Advanced system. Combined with Cognitive Radio (CR), the packet scheduler is more intelligent to assign the radio resources dynamically. To the best of our knowledge, Carrier Scheduling (CS) scheme in the cognitive LTE-A system with CA has not thoroughly been studied yet till now, even the CS scheme based on multi-service. In this paper, we design an efficient multi-user multi-service carrier scheduling scheme in such scenario, named as QoS-based Separated Random User Scheduling (QSRUS), which considers multi-service QoS requirements and fairness among different services simultaneously. From the simulation results, the proposed scheme can achieve remarkable performance improvements in terms of the QoS performance, the mean switching frequency and the fairness when compared with the traditional scheme.

In a multi-secondary user (SU) and single primary user (PU) cognitive radio (CR) system, each terminal is equipped with multi-antenna. We propose an optimal beamforming design method aiming for the minimal power budget of SUs' network, where both the perfect and imperfect knowledge of PU's precoder are considered, e.g. optimal beamforming vector with perfect PU's precoder (OBV-perfect), optimal beamforming vector with imperfect PU's precoder (OBV-imperfect). In a spectrum sharing based CR network, SUs are allowed to coexist with the PU, provided that the interference power from the SUs to the PU is less than an acceptable value, such that the quality of service (QoS) of PU is guaranteed. Meanwhile, the QoS requirement of minimizing signal to interference and noise ratio (SINR) at the secondary receiver is also included as the constraint. Simulation results show that the system power of OBV-perfect is smaller than the imperfect scenario, since OBV-perfect knows of both the PU's channel state information (CSI) and precoder information, thus contributing to minimize the power under the minimal SINR constraints. However, the system rate of OBV-imperfect is larger than OBV-perfect scenario due to the reason that in our presumed resolution of OBV-imperfect, the employed signal to jamming and noise ratio (SJNR) is a much stricter constraint than SINR for the system. Therefore, both the schemes are optimal beamforming design, and the knowledge of PU's precoder is favorable for access capability in the cognitive MIMO system.

In this paper, we focus on an energy efficient resource allocation in the multi-secondary user (SU) cognitive radio networks with network coding based cooperative transmission (NcCT). We set up a framework for multi-SU resource allocation game with Nash bargaining solution (NBS) under the cognitive radio scenario (CR-MSU-NBS game) where the sum of pairwise NBS function with pairing strategy is exploited as the network optimization objective and context conditions as constraints. Thereby, the network performance is improved by pairwise SUs' win-win cooperation on both system throughput and fairness. Since the CR-MSU-NBS game is NP-hard, we resolve it by a heuristic method. After testifying the global network NBS optimization pairing method, we propose the energy efficient suboptimal resource allocation scheme for multi-SUs. The simulation results show that the proposed scheme achieves a good tradeoff between fairness and efficiency and outperforms the familiar distance-pairing schemes; meanwhile, it achieves a 33.4% energy efficiency improvement but only 7.8% fairness loss averagely to the high computation complexity optimal method with ergodic search.

In conventional multicast scheme (CMS), the total throughput of multicast group is constrained by the user with the worst channel quality. In order to overcome this problem of limited throughput, we consider a resource allocation algorithm based on layered coding, when a limited feedback scheme is considered. A novel subcarrier and power algorithm is exploited for targeting the maximum throughput (MT) of enhancement layers while at the same time guaranteeing the quality of services (QoS) requirements of all users. In this paper, a three-step subcarrier allocation algorithm and two power allocation algorithms are proposed. Simulation results show that the proposed algorithm significantly outperforms CMS. Moreover, it obtains more throughput than another existing algorithm.

In this paper, we investigate resource allocation in the OFDMA cooperative networks with Best effort (BE) services and Real time (RT) services. Both Amplify-and-forward (AF) and Decode-and-forward (DF) schemes are considered adaptively in the two-hop relay transmission. We resolve the relay node, relay strategy selection combined subcarrier and power allocation in mixed services resource allocation problem by first satisfying the QoS requirements of RT services with the least power assumption and then maximizing the overall throughput of BE services. Simulation results reveal that our proposed resource allocation method outperforms previous works for mixed services in terms of QoS satisfaction and maintains a relatively high system throughput.

In this paper, focusing on the spectrum sharing and competition, we propose a double auction based spectrum trading (DAST) scheme which resolves the spectrum access between the primary network (PN) and secondary networks (SNs) subtly. Two different utility functions for primary users (PUs) and secondary users (SUs) are designed basing on a supply-and-demand relationship between them. Also, we adopt expectation and learning process in the module formulation, which takes consideration of the variance of channels, transmission forecasting, afore trading histories and etc. Numerical results with four bidding strategies are presented to reinforce the effectiveness of the two proposed utility evaluation based decision modules in supply falling short of demand cases. Meanwhile, the proposed DAST maintains comparable frequency efficiency with traditional centralized cognitive radio (CR) access approaches.

This paper presents a novel Q-learning based auction (QL-BA) algorithm for dynamic spectrum access in a one primary user multiple secondary users (OPMS) scenario. In the auction market, the secondary user provides a bidding price dynamically and intelligently using a Q-learning based bidding strategy to compete for current access opportunity; meanwhile primary user decides to whom to release the unused spectrum according to the maximal bidding principle. To obtain the limited and time-varying spectrum opportunities, each bidder presents a preference utility through Q-learning, considering the current packet transmission and future expectation. Simulation results show that the proposed QL-BA can significantly improve secondary users' bidding strategies and, hence, the performance in terms of packet loss, bidding efficiency and transmission rate is improved progressively.

An adaptive resource allocation scheme for QoS oriented OFDMA system, which schedules two different utility functions for the Real-time/Non Real-time traffic simultaneously, is proposed in this paper. Instead of partial consideration of uniform kinds of QoS, we introduce an updating ratio factor to schedule users of heterogeneous traffic. Due to the complex optimization objective, the general convex optimal methods are no longer feasible. We are motivated to study a heuristic natural genetic approach to solve this problem. Due to the weak convergence of Genetic Algorithm (GA), we improve it by a well-selected initial population. Numerical results are presented to illustrate that our scheme not only tackles the diverse QoS requirement but also alleviates the unfairness between real-time and non-real-time services under various traffic loads.