T o support intelligent Internet of Things(I o T)applications，such as autonomous driving，smart city surveillance，and virtual reality(VR)/augmented reality(AR)，cloud services are expected to be pushed to the proximity of IoT devices for quality performance.For instance，to facilitate safe autonomous driving，the service delay of most vehicular applications is required to be within milliseconds，and any information delay may result in dangerous on-road conditions.Edge intelligence aims at processing data/computing-intensive IoT tasks at the edge of network，where a set of IoT devices can work cooperatively for data collection，processing，model training，caching，and data analytics via edge caching，edge training，edge offloading，etc.It empowers intelligent IoT services at the network edge.However，it is challenging to achieve satisfying performance due to the following reasons.On the one hand，as the edge nodes are constrained by storage/computing resources，it is essential to conduct end-edge-cloud resource orchestration and resource sharing，taking into account device mobility，burst & stochastic service requests，and heterogeneous resources.On the other hand，better quality of service/learning of IoT applications is difficult to be guaranteed as the task execution/model training is contributed by distributed IoT devices.As a result，individual quality characterization，participating node selection，multi-level collaboration，robustness against malicious attacks are crucial.At last，to enable IoT intelligence，frequent communication and coordination among IoT devices，edge nodes and cloud servers are required，which can raise significant overhead，delay，and potential disclosure of sensitive information.Overcoming those challenges calls for further in-depth research.

The primary goal of this feature topic is to organize cutting-edge research related to end-edge-cloud orchestration and its application in intelligent IoT，including novel ideas on data collection and analytics，models，methodologies，architecture and system design，performance measurements and test experiments.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，13 articles are finally accepted in this special issue.

The accepted papers cover the topics about cost-efficient and context-aware caching for IoT services，mobile edge computing(MEC)enabled cooperative sensing and resource allocation，RFID authentication protocol for end-edge-cloud collaborative environment，data encryption scheme，and privacy-preserving methods for intelligent IoT system，etc.We hope this special issue can inspire the existing and important future research works in various IoT intelligence empowered by end-edge-cloud orchestration.

Link flooding attack(LFA)is a fresh distributed denial of service attack and one of the most important threats to IoT devices.It manipulates legal low-speed flow to flood critical links so that traditional technologies are difficult to resist such attack.The first paper entitled “ReLFA:Resist Link Flooding Attacks via Renyi Entropy and Deep Reinforcement Learning in SDN-IoT” proposes a new LFA mitigation scheme ReLFA，aiming at the LFA in the software defined IoT.By calculating the Renyi entropy，the congested link is located in the data plane，and the target links are determined according to the alarm threshold.When LFA is detected on the target links，the control plane adopts deep reinforcement learning based methods to carry out traffic engineering.Simulation results show that ReLFA can effectively alleviate the impact of LFA in SDN IoT with shorter rerouting time compared with the latest schemes.

Biometric key is generated from user's unique biometric features，and can effectively solve the security problems in cryptography.The randomness，non-linear and non-stationary characteristics of Electroencephalographic(EEG)signals render it secure to generate biometric keys via EEG data.In the paper“Never Lost Keys:A Novel Key Generation Scheme based on Motor Imagery EEG in End-Edge-Cloud System”，the authors study the biometric key generation based on motor imagery EEG signals in collaboration with end-edge-cloud computing.Using sensors to measure motor imagery EEG data，the key is generated via pre-processing，feature extraction and classification.Specifically，motor imagery EEG signal is collected at the end side，CSP and SVM are applied to data feature extraction and classification respectively at the edge side.The results show that the total time consumption of the key generation process is about 2.45 second.

IoT technology has the vision of the Internet of everything so that frequent communication and coordination are required between IoT devices，edge nodes，and cloud servers.In such a context，it is significant to ensure the communication security of the IoT system.In the paper “A One-Time Pad Encryption Scheme Based on Efficient Physical-Layer Secret Key Generation for Intelligent IoT System”，a joint secret key generation(SKG)and one-time pad(OPT)encryption scheme with the aid of a reconfigurable intelligent surface to boost secret key rate is proposed.The OPT is an application-layer encryption technique to achieve the information-theoretic security，and SKG is a promising candidate to provide the random keys for OTP.The authors divide the process of secure transmission into two stages to maximize the efficiency of secure communication and design an optimal algorithm for allocating time slots for SKG to maximize SKG efficiency without security risk.Furthermore，a robust key updating protocol based on the SKG scheme for OTP encryption is also proposed.This work provides a solution to overcome the system overhead and potential disclosure of sensitive information caused by IoT devices and edge nodes in communication.

IoT terminals and sensors have caused security and privacy challenges due to resource constraints and exponential growth.As the key technology of IoT，Radio-Frequency Identification(RFID)authentication protocol tremendously strengthens privacy protection and improves IoT services.However，it inevitably increases system overhead while improving security.In “UAP:Random Rearrangement Block Matrix-Based Ultra-Lightweight RFID Authentication Protocol for End-Edge-Cloud Collaborative Environment”，the authors design an ultra-lightweight encryption function and propose an RFID authentication scheme based on this function for the end-edge-cloud collaborative environment.The security of the protocol is then proved through BAN logic.Tested on the FPGA hardware platform，the results show that the proposed protocol balances low computing costs and high-security requirements.In industrial IoT systems，state estimation plays an important role in multi-sensor cooperative sensing.However，the state information received by remote control center experiences random delay，which inevitably affects the state estimation performance.Moreover，the computation and storage burden of the remote control center are very huge，due to the large amount of state information from all sensors.To address this issue，a layered network architecture and an MEC-enabled cooperative sensing scheme are proposed in “MEC Enabled Cooperative Sensing and Resource Allocation for Industrial IoT Systems”.The authors firstly characterize the impact of random delay on the error of state estimation and then optimize the cooperative sensing and resource allocation to minimize the state estimation error.An improved marine predators algorithm is designed to choose the best edge estimator for each sensor to pretreat the sensory information.The simulation results show the advantage and effectiveness of the proposed scheme in terms of estimation accuracy.

The next paper entitled “Efficient Multi-User for Task Offloading and Server Allocation in Mobile Edge Computing Systems” is also on MEC systems，where a multi-user with multiple tasks scenario is firstly considered and a multi-server case is later studied.In this paper，the authors propose a distributed unsupervised learning-based offloading algorithm for task offloading and employ distributed parallel networks to guarantee the robustness of the algorithm.Moreover，a memory pool is exploited to store input data and corresponding decisions as key-value pairs.Based on the experience mechanism，the proposed algorithm can omit the step of data calibration compared with the supervised learning method.The results show that the proposed algorithm performs better than discrete particle swarm and fully connected distributed unsupervised learning-based algorithms，which can achieve a near-optimal decision in a short time less than 0.01 seconds.

Peer-to-peer computation offloading is also promising，which enables resource-limited IoT devices to offload their computation-intensive tasks to idle peer devices in proximity.In “Stochastic Learning for Opportunistic Peer-to-peer Computation Offloading in IoT Edge Computing”，the authors investigate the opportunistic peer-to-peer task offloading problem among multiple offloading requestors.Each requestor makes decisions on both local computation frequency and offloading transmission power to minimize its own expected long-term cost on task completion，taking into consideration its energy consumption，task delay，and task loss due to buffer overflow.The dynamic decision process among multiple requestors is formulated as a stochastic game and a decentralized online offloading algorithm is proposed by constructing the post-decision states.Simulation results demonstrate that the proposed online algorithm achieves a better performance compared with existing algorithms，especially in the scenarios with large task arrival probability or small helper availability probability.

Cloud-edge computing serves as an emerging paradigm that exploits the computing，storage，and communication capacities of edge devices.In the paper“Achieving Fuzzy Matching Data Sharing for Secure Cloud-Edge Communication”，the authors propose a fuzzy matching data sharing scheme with a pairing-based cryptosystem for cloud-edge communications.This scheme enables the matching holds with a certain distance of error and allows the policies from both sides to be checked simultaneously without revealing any additional information except the matching holds or not.The formal security proof is also given to show the security，privacy，and authenticity.By comparing with the existing works，the results indicate that the proposed data sharing is practical.

The emerging intelligent mobile applications(e.g.autonomous driving)require the awareness of surrounding environments to make intelligent decisions.Accordingly，context information，which depicts the real-time environment status，has been emerging as a new network traffic type.An efficient way to accommodate contest information is to leverage mobile edge caching.In “Age-constrained Dynamic Content Replacing and Delivering for UAV-assisted Context Awareness”，the authors propose a fine-grained age-constrained cache replacing scheme，wherein the cache replacing and the content delivery are jointly designed.In addition，the UAV dynamically decides which content items to replace based on both the user requests and the content dynamics.An optimization problem is formulated based on the Markov Decision Process and a sufficient condition of cache replacing is found in a closed form，whereby a dynamic cache replacing and content delivery scheme is proposed based on the Deep Q-Network.Compared with the conventional popularity-based and the modified Least Frequently Used schemes，the UAV can offload around 30%traffic from the ground network by utilizing the proposed scheme in the urban scenario.

Platoon assisted vehicular edge computing is a promising paradigm of implementing offloading services.In “Privacy-Preserving Incentive Mechanism for Platoon Assisted Vehicular Edge Computing with Deep Reinforcement Learning”，the authors model interactions among the requester(leader)and multiple performers(followers)as a Stackelberg game，and the requester incentivizes the performers to accept workloads.Furthermore，the incentive problem is tackled via deep reinforcement learning while keeping the performers’ information private and each game player becomes an agent that learns the optimal strategy by referring to the historical strategies of the others.Numerical results show that the proposed scheme has the advantages in quickly converging to the equilibrium solution and reducing the total monetary cost of the requester.

In IoT applications，large volumes of data are collected，however，they might carry massive private information that users do not want to share.The paper“Towards Task-Free Privacy-Preserving Data Collection” proposes a task-free privacy-preserving data collection method to protect private attributes specified by users while maintaining data utility for unknown downstream tasks.The authors first propose a privacy adversarial learning mechanism to protect private attributes by optimizing the feature extractor to maximize the adversary’s prediction uncertainty on private attributes，and then design a conditional decoding mechanism(ConDec)to maintain data utility for downstream tasks by minimizing the conditional reconstruction error from the sanitized features.With the joint learning of PAL and ConDec，a privacy-aware feature extractor where the sanitized features maintain the discriminative information except privacy can be learned.

Multi-mode power IoT combines various communication media to provide spatio-temporal coverage for low-carbon operation in smart park.In our last paper “Adaptive Learning-Based Delay-Sensitive and Secure Edge-End Collaboration for Multi-Mode Low-Carbon Power IoT”，the authors focus on how to support multi-mode heterogeneous networking in low-carbon PIoT.An adaptive learning-based delay-sensitive and secure edge-end collaboration algorithm is proposed to optimize multi-mode channel selection and split device power into artificial noise transmission and data transmission for secure data delivery.This scheme can achieve multi-attribute QoS guarantee，adaptive resource management and security enhancement and access conflict elimination with the combined power of deep actor-critic，“win or learn fast” mechanism and edge-end collaboration.Simulation results demonstrate its superior performance in queuing delay，energy consumption，secrecy capacity，and adaptability to differentiated low-carbon services.

To sum up，the Guest Editors of this feature topic would like to thank all the authors for their valuable contributions and the anonymous reviewers for their helpful comments and suggestions.Also，we would like to acknowledge the guidance from the editorial team of China Communications.