In Hierarchical Split Federated Learning, the efficiency and quality of model training face challenges due to the varying data quality across nodes and the dynamic changes in their computational and communication capabilities over time. To address this issue, this paper investigates the node selection problem for minimizing system delay under model anomaly constraints and proposes an online node selection algorithm based on Contextual Combinatorial Multi-Armed Bandit (CS-MAB). The algorithm evaluates node data quality through model anomalies and dynamically selects nodes based on their computational and communication capabilities to optimize training delay and improve model quality. Compared to existing algorithms like CS-UCB and FedAvg, CS-MAB demonstrates significant advantages in reducing delay and improving model accuracy. Experimental results show that the proposed algorithm reduces delay by 53.72% and 49.45% on MNIST, FashionMNIST, and CIFAR-10 datasets, respectively, and achieves up to a 41.49% improvement in accuracy.

In traditional model training, data from certain nodes are reused, which not only wastes computational resources but also may lead to insufficient model generalization capability. This chapter addresses this problem by optimizing the quality of SL models without increasing the training latency by proposing a node selection method that requires initial nodes not to reuse local data and takes over the training task by introducing new nodes. Defining this problem as a quality optimization problem, it can be shown that it is an NP-hard problem. To ensure that the training delay does not exceed the initial limit, the training time of the slowest training node among the original nodes is used as a delay constraint. To solve this problem, the model quality optimization problem under the delay constraint is proposed and an approximate node selection algorithm ANS is designed to select the node that satisfies the delay constraint and maximizes the amount of data. In order to demonstrate that the proposed node selection scheme can not only effectively optimize the quality of the BSL model, but also achieve better data volume maximization while ensuring the delay constraint. The designed ANS algorithm is compared with RTNS and GCNS algorithms, and the experimental results show that the approximate node selection ANS algorithm improves the model quality by 25% over the RTNS algorithm and 15% over the GCNS algorithm.

During the training process of Large Language Model, Federated Learning is introduced to solve the trade-off between data-sharing requirements and privacy protection. Due to the large number of model parameters in the large language model and the distributed architecture of federated learning, training delay cost produced by nodes during the training process is large and is an important issue to address. This paper focuses on minimizing the total training delay by balancing the local training delay of participating nodes through dynamic node scheduling. We propose a dynamic scheduling algorithm LTQA based on Bi-directional LSTM model and reinforcement learning theory. During the training process, we find the node with the highest delay cost in each round using a Bi-directional LSTM model. Reinforcement learning theory is used to schedule resources and tasks with other nodes for cooperation, balancing the local training delay of each node according to dynamic environment changes. We evaluate the performance of the algorithm through several experiments on two widely-used large language model. It is found that compared with state-of-the-art FL algorithms, LTQA can save training delay cost by 7.17%-25.8%, and the model accuracy is improved by 2.59%-13.62%.

Computing power measurement faces the challenges of heterogeneity of devices, poor scalability with workloads, and discrepancies between measurement methods and models. In this paper, we propose a computing power measurement framework called AdaptPerf to address these issues. AdaptPerf uses the technique of Neural Architecture Search to build a hypernetwork of benchmark models for heterogeneous devices and adaptively selects the ones suitable for them. It achieves equivalency in computing power measurement through kernel-level latency analysis. We further prove that the Adaptive Benchmark Measurement Model Selection Problem (ABMSP) is an NP-hard problem and introduced a deep reinforcement learning approach to solve it. The theoretical analyses show that the proposed method can guarantee that the obtained solution approximates the Pareto frontier. We implement AdaptPerf in a real-world environment, constructing a hypernetwork of over 4.2×1020 models for heterogeneous devices to autonomously select the benchmark model adapted to the devices. We use different tools, such as Nsight Compute, Nsight System, to analyze and evaluate the feasibility and performance of AdaptPerf. It is found that the models generated by AdaptPerf increase computational throughput by 46.63% and improve data reuse rate by 93.7%, compared with more than 60 traditional models. AdaptPerf also exhibits superior load adaptability and model complexity compared with existing methods such as MLPerf.

Video is a dominant traffic type in mobile networks. D2D communication, one of the most promising technologies for 6G, enables efficient localized distribution of content and thus reduces latency and bandwidth consumption for content delivery. Discovering potential mobile user devices and caching promotional content or short video content in D2D self-organizing communication-based social networks requires overcoming challenges such as heterogeneous structural features of user interactions and user state changes. We propose a graph neural network based GSVIA framework that integrates user information from the perspective of user interaction behavior and social features, and also propose an InfTMC algorithm for maximizing the weights of a set of caching nodes using aggregate function submodularity. Experimental results under two different weighted IC models show that the InfTMC algorithm improves the influence spread by an average of 46.80% and 45.44% compared to the existing algorithms.

Computing power measurement is critical for effective resource utilization in computing power networks. However, it faces the challenges of heterogeneity of devices, poor scalability with workloads, and discrepancies between measurement methods and models. In this paper, we propose a computing power measurement framework called AdaptPerf to address these issues. AdaptPerf uses the technique of Neural Architecture Search to build a hypernetwork of benchmark models for heterogeneous devices and adaptively selects the ones suitable for them. It eliminates the impact of memory access factors among heterogeneous models by conducting kernel-level latency analysis, achieving equivalency in computing power measurement. We implement AdaptPerf in a real-world environment, constructing a hypernetwork of over 4.2 × 1020models for heterogeneous devices to autonomously select the benchmark model adaptive to the devices. We use different tools, such as Nsight Compute, Nsight System, and etc., to analyze and evaluate the feasibility and performance of AdaptPerf. It is found that the models generated by AdaptPerf increase parameter volume by 5.19 times and improve by 4.99% in data reuse rate, compared with more than 50 conventional models. AdaptPerf also exhibits superior load adaptability and model complexity compared with existing methods such as MLPerf.

In order to reduce the delay and bandwidth consumption of content delivery, we investigate how to uncover effective edge nodes in short video social network and cache promotional content or high-quality short video content in them. In this paper, we define the edge caching node selection problem for influence maximization under budget constraint, which proves to be an NP-hard problem. We propose an InfT algorithm based on influence spread tree to uncover edge nodes with high influence capacity in short video social network. During the selection of edge nodes, the cost changes dynamically with the selection of nodes. Experiments on scale-free network show that InfT makes the influence spread of edge caching nodes respectively improve by 21.82%, 438.21%, and 612.04%, when compared to some existing classical algorithms.

To optimize short video data dissemination in edge networks, we propose a novel 3-layered network model. Based on the model, we investigate the optimal edge caching node selection problem, which is proved NP-hard. We also analyzed the upper and lower bounds of the data dissemination range in our proposed model. We developed the graph-attention-based cache propagation for degree calculation (GACPD) to predict the dissemination scale of short video data for each caching node, and the graph embedding and GAT-based caching node selection (GEG) algorithm to select the optimal caching nodes. We implement the GEG algorithm and evaluate its performance on both real and simulated data sets. It is found that GEG can reduce the backbone network traffic by 30% to 50%, as well as with 18% to 30% network bandwidth utilization improvement, compared with the existing ICS and CDA algorithms.

In hierarchical federated learning (HFL), edge computing is introduced for partial model aggregation to reduce latency. High energy cost is an important issue to be solved in the process of parameters uploading. Our study focuses on the issue of minimizing energy cost with delay constraint through node cooperation in HFL, and the decision problem for this is NP-hard. We introduce a cost-efficient HFL (CE-HFL) framework, where nodes not only participate in model training but also transmit and aggregate model parameters for neighbors. A parameter aggregation tree is first generated, and parameter updates can be delivered to edge servers along paths in the tree while being aggregated simultaneously. Through theoretical analysis, it is proved that CE-HFL can achieve energy optimization with delay constraint. We also evaluate its performance through thorough experiments. In comparison with HierFAVG, CFL, and HFEL, it is found that CE-HFL can save energy cost up to 24.58%, 22.02%, and 6.60%, respectively.

With the rapid development of Internet of Things (IoT) and 5G wireless communication technology, a large amount of data is generated at the edge of the network. The combination of mobile edge computing (MEC) and federated learning has become a key technology to improve performance and protect users' privacy data in mobile networks. The selection of nodes for Hierarchical Federated Learning (HFL) affects the quality of model training. In this paper, we investigate the optimization problem of node selection accuracy in HFL. In order to improve the quality of model training, we design an algorithm of node selection based on reputation (NSRA). In NSRA, the edge server selects the node with high reputation prediction value to participate in the model training, and the node selects the neighbor node with high transmission capacity to cooperate. D2D communication is adopted for node cooperation. Through extensive simulations, it is verified the performance of NSRA. The mutual trust between nodes is enhanced, so the ideal prediction effect is achieved. We also observe that compared with RSA, the accuracy is improved by 11.48% and 19.38% in MNIST and CIFAR-10, respectively.

Summary The proliferation of mobile phones and webcams has led to an exponential increase in video data. One of the key technologies of video surveillance systems is Person Re‐identification (Re‐ID). The Re‐ID is used to identify whether the target pedestrian is the same person, and through scene matching, cross‐field tracking and track prediction of suspected pedestrians can be achieved. The edge computing has become the first choice for video analysis and processing, because of shorter response time and more efficient processing. In this paper, we propose a deep square similarity learning (DSSL), which considers the difference correlation, first‐order correlation, and two‐order correlation of image pairs. The training data automatically adjusts the network parameters and the weights of the three correlations to minimize the loss of the training set. Moreover, we conducted experiments on the challenging Re‐ID databases CuHK03 and Male1501. Compared with algorithm IDLA and DHSL, the first recognition rate is increased by 18% and 40%, respectively, in CuHK03, and 22% and 80% in Male1501. Then, we propose an online deep square similarity learning (ODSSL) algorithm to solve problem of data updating after the model is established by DSSL strategy. Meanwhile, ODSSL shows shorter update time and more efficient processing.

In recent years, there is an increasing demand for indoor localization services with the aim to locate people and objects inside buildings. However, localization accuracy is susceptible to inaccurate and high variant sensor measurements due to the unpredictable fluctuations of received wireless signals and the sensitivity of hardware devices. To address this issue, in this paper, we establish a new Bluetooth indoor localization system, whose architecture can be basically decomposed into two parts: the internet-of-things (IoT) framework and the localization module. Concretely, the IoT platform uses the state-of-the-art light weight Spring Boot microservice framework consisting of multi-layer structure. In the localization module, it follows the general process of trilateration but significantly distinguished from it. A set of measures are adopted to strengthen the system’s robustness when obtained measurements cannot be fully trusted. Specifically, in the first place, rather than using conventional propagation model to predict the distance between Bluetooth transmitter and receiver, we design a bran-new LSTM-based distance estimator which can better depict the nonlinearity of attenuation characteristics of radio signal. Moreover, we also employ a series of self-adaptive mechanisms, including elastic radius intersecting, multiple weighted centroid localization and self-adaptive Kalman tracking, to make the system robust against inaccurate measurements and unpredictable sudden variation of received wireless signal. A bunch of tests are conducted in both ideal lab environment and Alibaba’s large-scale warehouse, and experimental results show our indoor localization system outperforms the state-of-the-art benchmarks by a large margin in both localization accuracy and stability.

Summary Narrow Band Internet of Things (NB‐IoT) is a cellular‐based low power wide area network (LPWAN) radio technology, which can provide highly reliable services and wide coverage for IoT devices. Software defined networking (SDN) as an emerging network architecture can realize flexible resource allocation and network management. We introduce SDN into NB‐IoT and investigate the traffic modeling and performance evaluation of SDN‐based NB‐IoT access network. To evaluate the network performance in different environments, we introduce the Beta/D/1, Uniform/D/1, and M/D/1 queuing models, respectively. The proposed queuing models are suitable for different scenarios, in which NB‐IoT devices access the network in a highly synchronized, unsynchronized, or stochastic manner. We use the general solution to the G/G/1 and the M/G/1 queuing model to solve the proposed modeling problems. Through simulations, we investigate the influence of different network parameters. The analysis and simulation results can be used in the SDN controller to dynamically allocate resources and make network management decisions to satisfy different performance requirements of NB‐IoT applications.

The enhanced Mobile Broadband (eMBB) and ultra-Reliable Low Latency Communications (URLLC) are two main scenarios of 5G networks. There is an obvious difference in service requirements between the two different scenarios. When multiple heterogeneous services coexist in the network, it is important to explore optimal resource scheduling and allocation strategies. In this paper, we study the Quality of Service (QoS) optimization problem in eMBB and URLLC coexisting scenario. Considering the services' characteristics of heterogeneity and dynamics, we first introduce the flexible numerology structure which defines a set of flexible transmission time interval (TTI) to satisfy different QoS requirements of heterogeneous services, and then, we formulate a Markov decision process (MDP)-based dynamic resource optimization problem with the flexibilities of time and frequency domains. Next, we prove this optimization problem to be NP-hard and propose an innovative joint scheduling strategy DRSA based on flexible numerology and deep reinforcement learning method to allocate dynamic resources. Through experiments, the flexible numerology significantly outperforms the non-flexible ones. Comparison experiments with Sequence, Greedy and Random strategies show that the average throughput of DRSA is 7.1%, 14.8% and 23.9% higher than them, and DRSA can reduce URLLC services' loss rate by 43.7%, 28.6% and 53.8%.

With the rapid growth of Internet of Things (IoT), network slicing is regarded as an important technology to support the multi-users' needs for 5G mobile network. Network slicing allows network operators to provide services to different users, which can improve the rational utilization of network and hardware resources. In order to ensure the quality of service and build low-cost network infrastructure services, it is a challenging problem to find an appropriate resource allocation mechanism. In this paper, we discuss resource allocation in 5G radio access network (RAN). Considering the real-time resource request of the slice user, we propose a semi-Markov decision system model, which enables the virtual network provider to effectively satisfy the different user demands in real time. Then, we propose a resource slicing algorithm based on deep reinforcement learning (RS-DRL), which aims to improve the long-term benefits of virtual network providers and the utilization of slicing resources. We evaluate the performance of the RS-DRL through evaluations and comparisons. The results show that the proposed RS-DRL algorithm can effectively improve the performance and achieve the long-term benefits quickly.

Limited capacities in computation and battery of Internet of things (IoT) devices are two main bottlenecks for quality of service. Emerging mobile edge computing (MEC) and radio frequency based wireless power transfer (WPT) can help alleviate the issues. Incorporating WPT into MEC, IoT devices can get sustainable energy supply by WPT, and offload computation tasks to MEC to improve the computing ability. In this paper, we jointly consider the radio resource and task allocation for the wireless powered MEC system. To capture the high dynamics in task arrival and wireless network, a stochastic optimization problem which minimizes the energy consumption while guaranteeing queue stability is formulated. By exploiting the stochastic optimization theory, we transform the original problem into a deterministic optimization problem. A radio resource and task allocation (RRTA) algorithm is designed to acquire the optimal solutions of this problem. Theoretical analysis shows that RRTA can achieve arbitrary tradeoff between the energy consumption and queue length. Moreover, the close-to-optimal energy consumption can be reached by RRTA while bounding the queueing length. Experiment results reveal that RRTA can effectively decrease the energy consumption and maintain a small queue length.

With the rapid development of the mobile Internet and the Internet of Things, the number of mobile communication services has grown rapidly. When multiple different types of networks cover the same region, it is important to decide which one users connect to, known as the network selection problem. In this paper, we explore the optimal network selection problem in 5G ultra-dense network. We consider several different types of transmission data such as session, media, background and interactive, which conclude different QoS requirements. And then, we formulate the network selection problem as an MDP model in ultra-dense system, and propose NS-MDP algorithm which aims to obtain best target network by calculating the benefits of the utility function. NS-MDP algorithm takes into account user data requirements, current system status, and network load conditions. Comparison experiments with Best-Rate, Random and Greedy_AHP strategies, show that NS-MDP algorithm's average throughput is 8.6%, 17.8% and 20.5% higher than them, and NS-MDP can reduce blocking rate by 33.3%, 17.6%, and 37.7%.

With the rapid growth of bandwidth-intensive applications, Carrier Aggregation (CA) has been introduced in Long Term Evolution-Advanced (LTE-A) Networks to provide higher data rates. In this paper, we investigate the joint resource block allocation and link adaptation problem with CA in LTE-A downlink. The problem is formulated as an Integer Programming problem aimed at maximizing the cell throughtput while guaranteeing Quality of Service (QoS) of each User Equipment (UE), in the form of the minimum transmission rate. We also consider the proportional fairness of radio allocation among UEs. Due to the NP-hardness of the problem, we propose an efficient algorithm QA-PFRA. QA-PFRA consists of two phases. In the first one we assign radio resource to UEs successively for QoS requirements according to UEs' priority, and in the second one we assign CCs to other UEs to maximize the cell weighted throughput. We develop a simulator to evaluate the performance of QA-PFRA. For comparison, we also implement GA algorithm and ERAA algorithm. It is found that the cell throughput obtained by QA-PFRA is about 70 higher than that obtained by ERAA, and QA-PFRA can also achieve higher throughput compared with GA when the UEs are sparsely distributed in the cell. Furthermore, we can observe that Jain's fairness index obtained by QA-PFRA is about 10 higher than that obtained by GA and 40 higher than that obtained by ERAA.

Location privacy preservation has become an important issue in providing location based services (LBSs). When the mobile users report their locations to the LBS server or the third-party servers, they risk the leak of their location information if such servers are compromised. To address this issue, we propose a Location Privacy Preservation Scheme (LPPS) based on distributed cache pushing which is based on Markov Chain. The LPPS deploys distributed cache proxies in the most frequently visited areas to store the most popular location-related data and pushes them to mobile users passing by. In the way that the mobile users receive the popular location-related data from the cache proxies without reporting their real locations, the users' location privacy is well preserved, which is shown to achieve k-anonymity. Extensive experiments illustrate that the proposed LPPS achieve decent service coverage ratio and cache hit ratio with low communication overhead.

An Incentive Compatible two-hop Multi-copy Routing Protocol (ICMRP) is proposed for disruption-tolerant networks (DTNs), which takes both the encounter probability and transmission cost into consideration to defend the misbehaviors of selfish nodes. ICMRP ensures that nodes can maximize their profit when they report their encounter probability and transmission cost honestly. Meanwhile, the protocol adopts the theory of optimal stopping to select optimal relay nodes. A signature technology based on bilinear map is introduced to ensure the selected relay nodes can get the payment securely, which can prevent the malicious nodes from tampering the messages. Through enough simulations on the ONE simulator, it is proved that ICMPR can effectively stimulate nodes to transmit messages and achieve a higher packet delivery rate with lower cost.

Vehicular Ad Hoc Networks (VANETs) enable car-to-car communication without the support of network infrastructure, which introduce diverse application possibilities and have drawn much attention from academy and industry in the past years. Unlike other ad hoc networks, nodes in VANETs are restricted to move in streets and have limited communication ranges. Intuitively, vehicle-to-vehicle communication somehow has similarity to human-to-human interaction, which lead to an interesting question of exploring the social properties of VANET nodes. To address the question, we consider encounters of vehicles as their social relationships and model VANETs as social graphs. Based on the social graph model, we use two traces of mobile vehicles from San Francisco and Shanghai to explore their social properties. Our analysis show that several universal laws of social network are hold for VANETs. The social graphs forming by vehicles are scale-free networks with power-law like distribution of node degrees. Small world phenomenon is also observed in our experiments: the nodes in VANETs have high cluster coefficient and there exist short paths between node pairs less than 3 hops on average. The implication of our analytical results is of benefit to develop large scale software system for mobile applications such as VANETs, as well as helps to facilitate inter-device wireless communications in pervasive environment.

Femtocell network is an efficient configuration to improve coverage and quality of service (QoS) in cellular networks. However, due to dense deployment, users in a femtocell may be interfered by the base stations nearby, resulting in a deteriorated throughput of the femtocell. In this paper, we investigate the resource allocation problem targeting at max-min throughput of the femtocells. Under the assumption that a number of discrete power levels are available for each device, a joint optimization problem over power control and channel allocation is formulated. We show its hardness and propose a particle swarm optimization based algorithm PCASO. We demonstrate the efficiency of PCASO by thorough numerical experiments.

Opportunistic spectrum access, which allows secondary users opportunistically access unused licensed channels to exploit instantaneous spectrum availability, is a promising approach to achieve efficient spectrum utilization and mitigate spectrum scarcity. To address the challenge of dynamic spectrum access, one of the most important issues is cooperative spectrum reallocation among secondary users to minimize spectrum handoffs. In this paper, we present a low-complexity approach based on conflict graph to optimize spectrum reallocation by local coordination. In order to reduce communication overhead, we propose two heuristic spectrum selection methods named local observation and metric maximum. Experimental results show two benefits of the proposed schemes. On one hand, the coordination approach can dynamically improve the total system throughput (approximately doubled at most). On the other hand, our heuristic strategies decrease the number of spectrum handoff by up to 20% compared with existing strategies.

Delay/Disruptive Tolerant Network (DTN) differs from the traditional networks in that it has no continuous or contemporaneous connections but only intermittent connections among wireless nodes and thus it is viewed as an opportunistic networks. DTNs emerge as a good alternative to provide services to a variety of applications in highly challenged environments. However, the characteristics of DTNs make existing solutions infeasible to be applied directly and new solutions are required to be explored, such as multicast which has been extensively studied before in internet and mobile ad hoc networks. Network coding has been proved been proved as an efficient way to improve the performance of multicast in traditional networks. In this paper, we use simulations to study how multicast with network coding performs in DTNs in terms of delivery delay under various application requirements (i.e., the amount of content to distribute and the number of multicast group members) and the network settings (i.e., the popularity of the network and the contact rate). Some empirical results are provided in this paper as well.

Interactive wireless applications, like VoIP over wireless networks, desire high-quality links and smooth connectivity during user movement. In order to support seamless roaming in wireless networks, handoff optimization has attracted a lot of attention recently. Most existing approaches aim at reducing handoff latency in communication protocols. While these methods provide significant savings in handoff latency, frequent handoffs could still be crucial and problematic for interactive applications. In this paper, we propose a new perspective for handoff optimization by introducing navigation guidance to minimize the handoff frequency. We first formulate the navigation-driven handoff minimization problem, then propose an optimal algorithm and a localized algorithm to solve it. The optimal algorithm assumes global knowledge of AP locations and uses a navigation graph to find a minimal handoff frequency path. The localized algorithm, however, only uses neighbor AP locations for route selection, which is more practical in real applications. Implementation issues of the proposed algorithms are discussed and simulations based on real world AP deployment are used to evaluate their performance. Experiment results show that our algorithms reduce handoff frequency by at most 42% compared to existing strategies.

Sleep scheduling protocols are widely used in wireless sensor networks for saving energy in sensor nodes. However, without considering the special requirements of object tracking, conventional sleep scheduling protocols may lead to intolerable degradation of tracking qualities when they are used in object tracking applications. To handle this problem, sleep scheduling protocols tailed for object tracking have been proposed recently. For saving energy while maintaining satisfactory tracking qualities, these protocols pro-actively awaken sensors according to the prediction of objects' movement. Such sleep scheduling protocols are called the prediction-based sleep scheduling protocols. Most existing prediction-based sleep scheduling protocols require sensor nodes to know the locations of themselves, which may not always be available. In this paper we propose a Location-free Prediction-based Sleep Scheduling protocol (LPSS) for object tracking in sensor networks. LPSS guarantees the coverage level, an important tracking quality in most applications, which is defined as the number of sensors simultaneously detecting the object. In LPSS, when a sensor detects the object, it will emit a signal, namely the sensing stimulus. Sensors decide to wake up or not based on only the received sensing stimulus, the prediction models and the required coverage level, without the requirement of location information. We implement LPSS with two most popular prediction models: the Circle-based and the Probability-based prediction models. Experiment results show that LPSS not only provides qualified coverage levels, but also saves about 40% to 70% energy compared with existing location-free protocols. Moreover, the energy cost of LPSS is close to the ideal approach using accurate location information in terms of the number of awakened nodes.

Power control plays an important role in wireless communications. However, most existing power control protocols are not efficient enough to be applied in real ad hoc network systems. In this paper, we propose PIPC (Proportion-Integral Power Control), a novel closed-loop power control algorithm that can be used in conjunction with any routing protocol as long as the routing table of a node provides enough knowledge of local network topology. PIPC adaptively adjusts the transmission power for each node through an improved version of the classical PID (Proportion-Integral-Differential) control. The topology derived under PIPC is also fed back to the routing table that is used as its input. Both theoretical analysis and simulation study show that PIPC has a good performance.