Source-linked AI summary
Envisioning Device-to-Device Communications in 6G
Shangwei Zhang, Jiajia Liu, Hongzhi Guo, Mingping Qi, Nei Kato
TL;DR
The paper addresses how intelligent D2D communication can support emerging applications in complex, ultradense, dynamic 6G networks. It envisions AI-driven D2D approaches for mobile edge computing, network slicing, and NOMA cognitive networking, and presents these as guidelines for future D2D development.
Problem
Emerging applications exceed current 5G capabilities, while 6G densification creates severe interference and complex resource-management challenges.
Method
The paper envisions AI-driven D2D solutions for mobile edge computing, intelligent network slicing, and NOMA-based cognitive networking.
Results
The article outlines potential intelligent D2D implementations that use idle UE capacity, dynamic D2D-cluster resources, distributed AI, and NOMA-based cluster management.
Takeaways & Limitations
The discussion is intended to serve as guidelines for future D2D development in 6G communication systems.
Abstract
from arXiv · showhide
To fulfill the requirements of various emerging applications, the future sixth generation (6G) mobile network is expected to be an innately intelligent, highly dynamic, ultradense heterogeneous network that interconnects all things with extremely low-latency and high speed data transmission. It is believed that artificial intelligence (AI) will be the most innovative technique that can achieve intelligent automated network operations, management and maintenance in future complex 6G networks. Driven by AI techniques, device-to-device (D2D) communication will be one of the pieces of the 6G jigsaw puzzle. To construct an efficient implementation of intelligent D2D in future 6G, we outline a number of potential D2D solutions associating with 6G in terms of mobile edge computing, network slicing, and Non-orthogonal multiple access (NOMA) cognitive Networking.
I. INTRODUCTION
Emerging applications exceed current 5G capabilities, motivating an AI-driven 6G network and intelligent D2D communication. The paper envisions D2D solutions spanning mobile edge computing, network slicing, and NOMA cognitive networking.
- Motivation: 5G cannot fulfill emerging Internet of Everything applications requiring converged communication, sensing, control, and computing functionalities.Examples include augmented, virtual, and mixed reality, holographic communications, high-precision manufacturing, and smart environments.
- 6G vision: 6G is envisioned as a space-air-terrestrial-sea integrated, AI-driven, higher-frequency, ultradense heterogeneous, green, secure, and privacy-preserving network.Its architecture targets ubiquitous connection for trillion-level objects and ultra-high data rates with extraordinarily low latency.
- Challenges: Further densification creates severe interference, complex resource management, vast signaling, and prohibitive cost and energy consumption.AI techniques are proposed to support intelligent automated network operation, management, and maintenance.
- AI capabilities: Powerful mobile UEs can act as personal workstations supporting light-level on-device AI processing for intelligent networking.Their processors, storage, batteries, and sensors enable local network sensing and parameter adjustment.
- Contribution: The paper envisions intelligent D2D solutions for mobile edge computing, network slicing, and NOMA cognitive networking in future 6G.These solutions associate AI techniques with D2D communication to address future network requirements.
II. ARCHITECTURES AND DEVELOPING TREND OF 6G NETWORKS
6G extends terrestrial networking into a space-air-terrestrial-sea integrated architecture to provide ubiquitous connectivity and broad coverage. AI-enabled D2D supports low-latency, high-speed transmission across diverse network scenarios.
- SATSI architecture: Current terrestrial networks alone cannot provide the extremely broad coverage and ubiquitous connectivity targeted by 6G.6G therefore integrates sea and underwater networks with space-air-terrestrial networks.
- SATSI architecture: A typical SATSI network comprises space, air, terrestrial, and sea or underwater network components.The space segment includes interconnected satellites, while the air segment includes high- and low-altitude platforms.
- Applications: SATSI combines global coverage, full-frequency transmission, and broad application support for services including automated driving, manufacturing, smart healthcare, AR, VR, and MR.The terrestrial network remains the main provider of wireless mobile services for most human activities.
- D2D scenarios: With AI participation, D2D is envisioned to support low-latency and high-speed transmission in scenarios such as intra-plane, intra-ship, and vehicular networks.Examples include two-hop, multihop, and fast accurate beamforming D2D configurations.
B. Extremely Heterogeneous and Ultradense Terrestrial Network
Shorter 6G transmission distances and growing UE diversity make D2D central to an ultradense heterogeneous terrestrial network. However, densification intensifies interference, mobility, energy, and beam-management challenges.
- D2D and densification: D2D is envisioned to evolve with shorter 6G transmission distances to support more numerous and diverse UEs and applications.Ultradense networks integrated with D2D are considered part of the 6G terrestrial architecture.
- Densification challenges: Further densification can degrade performance through severe interference, prohibitive cost and energy consumption, and frequent handover for high-speed UEs.These effects complicate deployment of ultradense terrestrial networks.
- Densification challenges: THz signals require transmitter beams to point precisely at receiver antennas, making network management challenging.Interference management, channel allocation, and spectral-efficiency approaches remain essential.
C. Innately Intelligent and Highly Dynamic Network
The complexity and dynamism of 6G make traditional network management untenable for diverse QoS demands. The paper points to multilevel distributed AI spanning core networks, edge infrastructure, and UEs.
- Intelligent management: Traditional network management methods are considered untenable for 6G because the network will be more complex and dynamic than preceding generations.Intelligent management and optimization are envisioned to fulfill different QoS demands.
- Machine learning: Machine-learning approaches including supervised, unsupervised, and reinforcement learning have been adopted for intelligent resource allocation and network optimization.These methods have also been used for functions such as small base-station management.
- Distributed AI: Multilevel distributed AI is envisioned with a global AI center in the core network and local AI centers embedded in base stations or MEC servers.This arrangement supports global and local network management.
- Distributed AI: Smartphones and other UEs may support on-device data training by sensing and learning local channel, traffic, and mobility patterns.This extends intelligence toward the network edge.
A. From Mobile Phone to Mobile Workstation
Future 6G user equipment is envisioned as personal mobile workstations with powerful processing, storage, sensing, and battery capabilities. These capabilities support short- and long-range connectivity and real-time environmental sensing.
- A. From Mobile Phone to Mobile Workstation: Future UEs are envisioned with high-end processors, rich storage, many sensors, and ultralong-lifetime batteries, making them equivalent to personal mobile workstations.The paper links this evolution to growing computational power and future 6G communication requirements.
- A. From Mobile Phone to Mobile Workstation: Multiple radio interfaces and full-frequency access will let UEs connect to nearby devices through short-range links or to distant systems through satellite communications.The envisioned connectivity supports ubiquitous access across different network systems and QoS requirements.
- A. From Mobile Phone to Mobile Workstation: Powerful processors and abundant sensors enable future UEs to collect real-time data for environmental and wireless-reality sensing.Small antennas at higher frequencies may also allow multiple antennas to be embedded in compact UEs and IoT devices.
B. From Smart mobile Phone to AI-driven mobile Phone
6G devices are envisioned as AI-driven phones with substantial computing, communication, caching, sensing, and local training capabilities. On-device and local AI can support proactive network management, but local AI alone cannot manage large dynamic networks.
- Device capabilities: Future 6G phones may combine powerful computation with multiple antennas, full-duplex transmission, high-speed data transfer, caching, MIMO, cognition, and sensing.The paper cites smartphone computational capabilities approaching the order of the human brain.
- AI capabilities: On-device AI uses local data training, while local AI trains on local network data for network-related intelligence.The supplied passage introduces both on-device and local AI as distinct capabilities.
- Network role: AI-driven phones can support proactive 6G network management and automated network configuration to improve users’ QoE and QoS.The passage frames these functions as applications of device-based AI techniques.
- Network role: Relying solely on on-device AI is insufficient for predicting and managing large-domain dynamic networks.The paper therefore introduces local AI for local network data training.
IV. AI-DRIVEN D2D COMMUNICATION IN 6G
AI-driven D2D communication is presented as a 6G approach for extending devices’ environmental awareness and supporting computation under constrained edge resources. The architecture uses idle devices and D2D clusters to distribute tasks when individual devices or clusters cannot handle them alone.
- AI-driven D2D communication: Intelligent D2D uses shared local and on-device AI information to extend UEs’ view scope and awareness of the local network environment.The paper identifies this as a key feature of D2D communication in 6G.
- AI-driven D2D communication: Edge servers may become congested or overloaded as 6G applications demand intensive computation, storage, and transport resources.The passage also notes that resource management and allocation will be difficult.
- Mobile edge computing: In the D2D-enhanced MEC architecture, tasks are assigned first to idle UEs and then to local D2D clusters through D2D or cellular links.This allocation sequence applies when a single UE cannot support the task.
- Mobile edge computing: If neither the UE nor D2D clusters can support computation-intensive tasks, the architecture further assigns the task beyond those resources.The supplied passage ends before specifying the final destination.
B. D2D-Enabled Intelligent Network Slicing
D2D-enabled intelligent network slicing integrates operator, private-party, and D2D-cluster resources at the network edge. Distributed AI predicts and exposes dynamic resources for integration and supports model-update exchange among UEs.
- D2D-Enabled Intelligent Network Slicing: Dynamic D2D clusters can provide computation, network, memory, and storage resources as physical or virtual infrastructure for network slicing.These resources complement those owned by public mobile network operators and private third-party actors.
- D2D-Enabled Intelligent Network Slicing: The approach federates and integrates PLMN, private third-party, and D2D-cluster resources at the network edge.The supplied figure caption identifies the accompanying structure as D2D-enabled intelligent network slicing.
- D2D-Enabled Intelligent Network Slicing: Distributed AI in the integration layer predicts and exposes available dynamic resources as candidates for integration.The architecture also merges AI-model updates from UEs and exchanges learning parameters.
C. NOMA and D2D Based Cognitive Networking
The paper proposes combining NOMA, D2D clustering, cognitive networking, and AI to support efficient access and relay selection in future 6G networks.
- C. NOMA and D2D Based Cognitive Networking: Massive D2D clusters can be formed according to human aggregation and social behavior, with NOMA serving UEs within each cluster and OMA separating clusters.This hybrid access design supports simultaneous communication across different D2D clusters.
- C. NOMA and D2D Based Cognitive Networking: D2D-aided cooperative NOMA selects nearby UEs as relays to establish D2D connections with UEs in a NOMA group.Local AI and on-device AI can support relay selection through wireless cognition, user pairing, channel estimation, and position location.
- C. NOMA and D2D Based Cognitive Networking: D-OMA can be applied to secondary-network D2D clusters to improve performance through accurate beamforming.The beamforming uses multiple narrow-beamwidth antennas or high-frequency signals with strong directional characteristics.
V. CONCLUSION
The article envisions intelligent D2D communications as part of future 6G systems and discusses implementations spanning edge computing, network slicing, and cognitive networking. It aims to provide guidelines for future D2D development in 6G.
- V. CONCLUSION: The article focuses on efficient implementation of intelligent D2D in future 6G communication systems.Its scope includes D2D-enhanced mobile edge computing, D2D-enabled intelligent network slicing, and NOMA- and D2D-based cognitive networking.