Skip to main content

Table 3 Network management and optimization for 5G networks using ML

From: An architecture and performance evaluation framework for artificial intelligence solutions in beyond 5G radio access networks

Use case

Objective and scenario

Approach

AI architecture

Refs.

User association

Delay minimization under energy consumption constraint in ultra-dense network with edge computing

DRL

Decentralized

[50]

User association

Load balancing with service rate selection in vehicular network with heterogeneous base stations

DRL

Centralized

[51]

Cell selection

Capacity maximization and number of handovers balancing in open access femtocell network

Q-learning

Decentralized

[52]

User association

UE outages minimization by cell range expansion for Picocell with bias value

Q-learning

Decentralized

[53]

Network slicing

Maximize network revenue by intelligently admitting network slice requests

RL

Centralized

[54]

Network slicing

Maximize the utility of individual service provider by joint slicing computing and communication resources

DQL

Decentralized

[55]

Traffic prediction

Short-term traffic prediction for individual users

Meta-learning

Decentralized

[57]

Resource allocation

Minimize service latency in a sliced RAN by computing resource allocation and task transmission scheduling

DRL

Centralized

[56]

Traffic prediction

Predict the maximum service-specific traffic load for each slice based on real-world 5G network data

DL

Centralized

[58]

Content placement

Track and predict time-variant content requests from users

DL

Centralized

[59]

Content delivery

Determine caching and computing offloading decisions to reduce operational cost on edge server

DRL

Centralized

[60]

Rate adaptation

Video quality-aware rate control for real-time video streaming

DRL

Decentralized

[61]

User association

Coverage and capacity maximization for massive MIMO systems

DRL

Centralized

[62]

User scheduling

Sum rate maximization with joint user-cell association and selection of number of beams in 5G mmWave networks

Transfer Q-learning; Unsupervised learning

Hybrid

[63]

User scheduling

Power-efficient resource allocation in cloud RANs

DRL

Centralized

[64]

User scheduling

Minimizing the age of correlated information in IoT systems

DRL

Centralized

[65]