With the rapid development of smart terminals and infrastructures, as well as diversified applications(e.g., autonomous driving, virtual and augmented reality, space-air-ground integrated networks) with colorful demands, current networks (e.g., 4G and 5G networks) may not be well suited to the requirements of novel applications and services. Recently, efforts from both the industry and academia have been made on the research into 6G networks, artificial intelligence (AI) will play a pivotal role in the design and optimization of 6G networks.

6G will support applications with more stringent requirements beyond the eMBB, uRLLC, and mMTC technologies, and will be designed to employ a wide spectrum of AI-empowered applications and services at various network edge nodes. Edge Intelligence(EI), as an emerging paradigm, has been considered as one of the key enabling technologies for future 6G networks. EI will unleash the full potential of communication, computing, caching, control with AI capability at various network edges for 6G networks.

Despite the great potential, edge intelligence poses many new challenges on its deployment and implementation in 6G networks. For example, since a large number of smart mobile devices and diversified AI services will be deployed in 6G networks, it is very challenging to manage and control the system in an efficient way to meet diverse requirements for these edge devices and services.

The objective of this feature topic is to explore recent advances on the theory, key techniques, and applications for edge intelligence in future 6G networks. After the call for papers, a significant number of submissions have been received. All of the submitted papers are evaluated according to the standard reviewing process of China Communications. Following a rigorous peer-review process, 7 articles are finally accepted including 4 invited papers in this special issue.

The accepted papers cover the topics about edge semantic cognitive intelligence, federated edge learning, federated reinforcement learning, edge intelligence assisted resource management and allocation,etc. We hope this special issue will inspire novel ideas and new research directions in edge intelligence for 6G networks for researchers and developers.

The emerging edge intelligence is anticipated to underlay the pathway to connected intelligence for 6G networks. In the paper “Edge Semantic Cognitive Intelligence for 6G Networks: Novel Theoretical Models, Enabling Framework, and Typical Applications”，the authors discusses and conceives edge semantic cognitive intelligence (ESCI) in the aspect of theoretical models, frameworks, and applications.Besides, this paper discusses the ESCI framework orchestrating deep learning with semantic communication and presents two representative applications to shed light on the prospect of ESCI in 6G networks. At last, some open problems are finally enumerated to elicit the future research directions of ESCI.

Distributed learning approach named federated edge learning (FEEL) is a promising technique for the physical layer designs to reduce the communication overhead and privacy disclosure. In the paper “Federated Edge Learning for the Wireless Physical Layer:Opportunities and Challenges”，the authors provide a comprehensive overview of the application of FEEL to physical layer designs concerning CSI acquisition, and transmitter and receiver designs. This represents a paradigm shift in the deep learning(DL)-based physical layer design. The authors also point out several limitations inherent in FEEL, particularly when applied to the wireless physical layer.Finally, further research on FEEL-based physical layer design is encouraged, including theoretical analyses and system design to facilitate the practical deployment.

Satellite communication has been seen as a vital part of 6G communication, which greatly extends network coverage. In satellite communication, resource management is a key problem attracting many research interests. In the paper “Edge Intelligence Assisted Resource Management for Satellite Communication”, the authors propose a joint beam activation,user-beam association and time resource allocation approach. By optimizing the resource provision, the mismatch between per-user achieved rate and its target rate are minimized. Meanwhile, some beams can be opportunistically turned off for the purpose of saving satellite energy. A multi-agent stochastic learning based algorithm is applied to solve the non-linear and non-convex resource management problem, which has low complexity as well as local optimality. Simulation results show that the proposed approach converges fast and outperforms various baseline schemes in terms of user performance mismatch and satellite energy saving.

Federated learning (FL) is a promising edge learning framework that enables multiple edge devices to collaboratively train a common artificial intelligence model without exchanging raw data. In the paper“Clustered Federated Learning with Weighted Model Aggregation for Imbalanced Data”, the authors propose a model aggregation method for FL based on adaptive clustering, called weighted clustered federated learning (CFL), to deal with data imbalance. At each round of model aggregation, the participating edge devices are first clustered based on the cosine similarity of their local gradients, where the similarity threshold can be adapted to the time-varying divergence of local gradients. Then the weighted per-cluster model aggregation is performed on the edge server. The authors analyze the convergence rate of the proposed learning framework to obtain the weights for model aggregation to balance the contributions of each cluster. Experimental results show that the CFL framework can effectively accelerate the convergence rate of the learning process and improve the generalization performance of the global model in the data imbalance scenario.

With the augmentation of nascent services and user equipment (UE), 6G networks need enhanced service capabilities with the lower request response time. To relieve the backhaul link stress and reduce the content acquisition delay, mobile edge caching has become one of the promising approaches. In the paper “Federated Reinforcement Learning with Adaptive Training Times for Edge Caching”, the authors propose a novel federated reinforcement learning (FRL)method with adaptive training times for edge caching.Through a new federated learning process with the asynchronous model training process and synchronous global aggregation process, the proposed FRL-based edge caching algorithm can mitigate the performance degradation brought by the non-identically and independently distributed (non i.i.d.) characteristics of content popularity among edge nodes.They further analyze the theoretical bound of the loss function difference, based on which the training times adaption mechanism is proposed to deal with the tradeoff between local training and global aggregation for each edge node in the federation. Numerical simulations have verified that the proposed FRL-based edge caching method outperforms other baseline methods in terms of the caching benefit, the cache hit ratio and the convergence speed.

The future development trend of 6G is to provide users around the world with full coverage in the true sense. Due to its coverage and strong resistance to natural disasters, the satellite network has been considered as a powerful supplement and can achieve true seamless coverage in the whole area. Moreover,the combination of satellite communication and Mobile Edge Computing (MEC) can provide users with full coverage on-orbit computing services by deploying MEC servers on satellites. In the paper“Joint Computing and Communication Resource Allocation for Edge Computing Toward to Huge LEO Networks”, the authors study the task offloading problem in the multi-user and multi- satellite scenario, where each satellite is deployed with MEC services. They propose a novel algorithm to achieve the joint task offloading and communication computing resource optimization (JTO-CCRO). Simulation shows that the proposed JTO-CCRO algorithm can converge quickly and reduce the task completion time and user energy consumption effectively.

It is envisioned that pervasive Internet of Everything (IoE) applications will be supported in the future 6G networks. However, due to the limited power and computational capacity, mobile devices(MDs) are unable to support computation-intensive and time-critical applications in future IoE applications. Mobile edge computing (MEC), emerges a promising paradigm, which can meet the requirements of IoE applications. In the paper “Deep Reinforcement Learning Based Joint Partial Computation Offloading and Resource Allocation in Mobility-Aware MEC System”, the authors investigate the problem of joint partial computation offloading and resource allocation (CORA) by integrating energy harvesting technology into the MEC system. The aim of this work is to minimize the weighted sum of the average cost of the system including processing delay, communication delay and energy consumption. The authors propose a deep reinforcement learning (DRL) based algorithm to obtain the optimal scheduling without prior knowledge of task arrival,renewable energy arrival as well as channel condition. Simulation results show that the proposed algorithm can reduce the total cost in terms of energy consumption and delay of MDs, and the efficiency among multi-users compared with other benchmarks.

To sum up, the Guest Editors of this feature topic are most grateful to the authors for their valuable contributions and the anonymous reviewers for their helpful and insightful comments. We would also like to acknowledge the support and guidance from the editorial team of China Communications.