Deep-Learning-Based Resource Allocation for Time-Sensitive Device-to-Device Networks.

Area of Science:

• Wireless Communication Engineering
• Network Resource Management
• Artificial Intelligence in Communications

Background:

• Ultra-reliable and low-latency communication (URLLC) is a key 5G use case for mission-critical applications.
• Device-to-device (D2D) networks offer improved latency and reliability due to device proximity.
• Resource management in D2D networks presents a complex, non-convex combinatorial optimization challenge.

Purpose of the Study:

• To address the computational complexity of D2D network resource allocation for URLLC.
• To investigate a deep learning (DL) based framework for efficient resource management.
• To model latency and reliability using achievable rates in the short blocklength regime.

Main Methods:

• Developed a game theory-based algorithm as a baseline for comparison.
• Proposed a resource management framework utilizing a deep neural network (DNN).
• Evaluated performance in a time-sensitive D2D network context.

Main Results:

• The proposed DL-based method achieved performance nearly equivalent to the baseline algorithm.
• The DL approach demonstrated superior time efficiency compared to traditional iterative methods.
• End-to-end structure of the DL framework contributed to its efficiency.

Conclusions:

• Deep learning offers a viable and efficient solution for resource management in URLLC D2D networks.
• The DL-based framework effectively balances performance and computational complexity.
• This approach facilitates the deployment of demanding mission-critical applications over 5G networks.