Deep Reinforcement Learning Assisted Beam Tracking and Data Transmission for 5G V2X Networks

Area of Science:

• Wireless Communications
• Network Engineering
• Artificial Intelligence

Background:

• Beam tracking is critical for 5G vehicle-to-everything (V2X) networks, especially for vehicle-to-infrastructure (V2I) communications at higher frequencies (e.g., 5G FR2).
• Challenges include path loss, shorter time slots, high vehicle velocities, and localization errors, creating a trade-off between beam tracking accuracy and data rate.
• Existing methods struggle to maintain performance within the stringent time constraints of high-frequency V2I communications.

Purpose of the Study:

• To develop an efficient beam tracking method for 5G V2X networks that balances tracking accuracy and data rate.
• To address the challenges of high-frequency V2I communications, including short time slots and vehicle dynamics.
• To improve the temporal efficiency of beam tracking within the constraints of 5G FR2 communications.

Main Methods:

• Proposed a reinforcement learning (RL) assisted, high-resolution codebook-based beam tracking method.
• Evaluated and selected the twin delayed deep deterministic policy gradient (TD3) framework for its efficiency in determining beam patterns.
• Integrated recurrent neural networks (RNNs), informed by Hurst exponent analysis, to enhance RL framework performance.

Main Results:

• The proposed RL-assisted method effectively determines proper beam patterns within short durations.
• Demonstrated significant improvements in beam tracking accuracy compared to conventional approaches.
• Achieved enhanced data rates and superior temporal efficiency for V2I communications.

Conclusions:

• The TD3-based RL framework, enhanced with RNNs, provides a robust solution for beam tracking in 5G V2X.
• The method successfully navigates the trade-off between tracking accuracy and data rate in high-frequency V2I scenarios.
• The proposed approach offers a promising direction for optimizing V2I communication performance in dynamic vehicular environments.