In the past decade, the unmanned aerial vehicles (UAVs) swarm has become the disruptive force reshaping our lives and work. In particular, advances in artificial intelligence have allowed multiple UAVs to coordinate their operations and work together to accomplish various complex tasks, one of which is Federated Learning (FL). As a promising distributed learning paradigm, FL can be adopted well with the limited resources and dynamic network topology of UAV swarms. However, the current FL's training process relies on homogeneous data paradigms, which require distributed UAVs to hold the same structure data. This ideal hypothesis is can not apply to the heterogeneous UAV swarms. To tackle this problem, in this paper, we design a clustered federated learning (CFL) architecture, in which we cluster UAV swarms based on the similarities between the participants' optimization directions. Then, we formulate the model trading among model owners, cluster heads, and UAV workers as a three-stage Stackelberg game to optimize the allocation of the limited resources. We design a hierarchical reinforcement learning algorithm to search for the Stackelberg equilibrium under the clustered federated learning system. The performance evaluation demonstrates the uniqueness and stability of the proposed three-stage master-slave game under the clustered framework, as well as the convergence and effectiveness of the reinforcement learning algorithm.

Various Industrial Internet of Things (IIoT) devices and their compelling applications have witnessed exponential growth over the past few years. Compared to the consumer internet, the industrial IoT often faces tough communication latency boundaries and low packet loss. Recently, time-sensitive networks (TSN) technique has been considered the most promising solution to tackle these issues, providing deterministic communication guarantees for time-sensitive applications. In TSN, to achieve bounded low latency, it introduces the Cyclic Queuing and Forwarding (CQF) mechanism to shape and regulate the transmission selection to provide the protected transmission window for time-triggered (TT) flows. Currently, a considerable amount of literature has explored traffic scheduling with the CQF mechanism in TSN. However, these methods ignore the impact of routing policy on scheduling optimization. In this paper, we design two algorithms to tackle the problem of joint routing and scheduling optimization with the CQF mechanism for TSN. We propose a TT flow sorting algorithm (JDPS) to measure the scheduling priority weight of each TT flow, which uses the normalization method to calculate the weight based on the flow's deadline, period, and size. Then, we propose a joint routing and scheduling optimization with load balance algorithm (JRS-LB) to improve the scheduling success radio and the network resource utilization, which jointly considers the injection time slot offset, scheduling, and routing of TT traffic. The evaluation results mainly focus on the scheduling success ratio, execution time, and resource utilization for the proposed algorithm compared to the Tabu and Naive algorithms. Extensive simulations indicate that the proposed JDPS and JRS-LB algorithms can improve the scheduling success radio compared with the other algorithms.

Predicting the traffic in metro and access networks has a significant influence on expanding and upgrading the network. However, the intricate temporal and spatial correlations in such networks are affected by the users' behaviour and geographic characteristics. Consequently, predicting network traffic in metro or access networks faces significant challenges. To enhance the accuracy of this prediction, this paper explores a spatio-temporal approach to predicting traffic, based on a transformer model. This model effectively captures the temporal correlations in the traffic through a multi-head attention mechanism, learning long-term correlations resulting from the behaviour of the users of the individual nodes. Additionally, it acquires the spatial correlations of the traffic through a graph convolutional layer to capture short-term bursty correlations between users. Unlike previous single-point predictions, we optimize the model to make collaborative predictions for all the network nodes. Through parameter optimization, the improved model accurately predicts network traffic in small-scale regions. We implemented simulations using OMNET++ and predicted the traffic in various scenarios. Concurrently, we validated the proposed method with real network traffic based on operator measurements, achieving a prediction accuracy of 98.49%.

Benefiting from dense connections in data center networks (DCNs), load balancing algorithms are capable of steering traffic into multiple paths for the sake of preventing traffic congestion. However, given each path’s time-varying and asymmetrical traffic state, this may also lead to worse congestion when some paths are overutilised. Especially in the two-tier hybrid optical/electrical DCNs (Hoe-DCNs), the port contentions and large-grained optical packets of the fast optical switch (FOS) require the top-of-rack (TOR) switch to have microsecond-level load balancing capability for microburst traffic. This paper establishes a leaf-spine Hoe-DCN model to illustrate the principal characteristic of dynamic load balancing in TOR switches for the first time. Moreover, we propose the dynamic distributed multi-path (DDMP) load balancing algorithm that relies on dynamic hashing computing for network flow distribution in DCNs, which dynamically adjusts traffic flow distribution at microsecond level according to the inverse ratio of the buffer occupancy. The simulation results show that our proposed algorithm reduces the TOR-to-TOR latency by 15.88% and decreases the packet loss by 22.06% compared to conventional algorithms under regular load conditions, which effectively improves the overall performance of the Hoe-DCNs. Moreover, our proposed algorithm prevents more than 90% packet loss under low load conditions.

Low-Earth-orbit (LEO) satellites are one of the most promising technologies for allowing mobile communication systems to provide ubiquitous network service. However, an LEO constellation can hardly accommodate the kind of time-sensitive service widely required on various Internet-of-Things and industrial Internet-of-Things devices. The limited onboard resources and dynamic constellation topology significantly reduce the robustness of LEO networks and increase the risk of service disruptions. To realize LEO-based time-sensitive networks (TSNs), we introduce a network-layer-based latency-scheduling architecture for LEO satellite networks. The latency-scheduling architecture leverages multi-hop latency optimization instead of optimizing a single node. Based on the cyclic queuing and forwarding (CQF) mechanism usually used in terrestrial networks, we propose a time-sensitive scheduling algorithm (CoLT-TSA) that uses end-to-end collaborative latency tolerance for LEO constellations. CoLT-TSA tries to use the delay redundancy of time-sensitive traffic to improve the scheduling capabilities for TSN flows. The proposed algorithm improves network throughput without affecting the timeliness of time-sensitive services. Moreover, flow contention can be solved by CoLT-TSA based on the collaborative scheduling of satellites along the routing. We conduct simulations to evaluate the proposed algorithm, focusing on the latency, time-out ratio, packet loss and throughput. Various scenarios are simulated for CoLT-TSA and conventional techniques. These extensive simulations indicate that CoLT-TSA can decrease packet loss by more than 2.72% and prevent most of the packet loss of time-sensitive services, as well as reducing the scheduling time-out ratio by more than 4.5% compared to conventional algorithms

Stochastic nonlinear impairment is the primary factor that limits the transmission performance of high-speed orbital angular momentum (OAM) mode-division multiplexing (MDM) optical fiber communication systems. This Letter presents a low-complexity adaptive-network-based fuzzy inference system (LANFIS) nonlinear equalizer for OAM-MDM intensity-modulation direct-detection (IM/DD) transmission with three OAM modes and 15 wavelength division multiplex (WDM) channels. The LANFIS equalizer could adjust the probability distribution functions (PDFs) of the distorted pulse amplitude modulation (PAM) symbols to fit the statistical characteristics of the WDM–OAM-MDM transmission channel. Therefore, although the transmission symbols in the WDM–OAM-MDM system are subjected to a stochastic nonlinear impairment, the proposed LANFIS equalizer can effectively compensate the distorted signals. The proposed equalizer outperforms the Volterra equalizer with improvements in receiver sensitivity of 2, 1.5, and 1.3 dB for three OAM modes at a wavelength of 1550.12 nm, respectively. It also outperforms a CNN equalizer, with improvements in receiver sensitivity of 1, 0.5, and 0.3 dB, respectively. Moreover, complexity reductions of 67%, 74%, and 99.9% are achieved for the LANFIS equalizer compared with the Volterra, CNN, and ANFIS equalizers, respectively. The proposed equalizer has high performance and low complexity, making it a promising candidate for a high-speed WDM–OAM-MDM system.