mmWave Beam Tracking for V2I Communication Systems Based on Spectrum Environment Awareness

Lulu Zhang, Weizhi Zhong, Junjie Zhang, Zhipeng Lin, Zhuoming Yang, Junzhi WangSymmetry 2022 · DOI

TL;DR

This paper tackles the challenge of real-time mmWave beam tracking in dynamic vehicle-to-infrastructure (V2I) environments. They propose a novel Convolutional Neural Network (CNN)-based beam search strategy that leverages "spectrum environment awareness" by encoding the surrounding environment into a 3D grid. This approach significantly improves beam alignment accuracy and efficiency compared to existing methods.

What Problem Does This Paper Address?

The rapid growth of connected vehicles (Internet of Vehicles - IoV) demands higher data rates and lower latency, making millimeter wave (mmWave) communication a promising solution. However, mmWave signals suffer from high path loss and require highly directional beams. In V2I scenarios, vehicles are constantly moving, making it difficult to maintain precise alignment of these narrow beams in real-time. Traditional beam search methods are often computationally intensive and fail to account for the dynamic environmental factors that influence signal propagation, leading to degraded communication quality.

Methods

The core of this paper is a novel CNN-based beam search strategy that incorporates spectrum environment awareness. Here's a breakdown of their approach:

  • 3D Grid Encoding: Instead of just using vehicle location, they encode the surrounding environment into a 3D grid. This grid captures information about vehicle positions, heights, lanes, and even the presence of buildings. This creates a richer feature representation of the environment, which they call "symmetry-based environmental coding."
  • Label Iterative Optimization: To improve the accuracy of the training data, they employ an iterative optimization process. This involves training the CNN model, evaluating its predictions, and then using those predictions to refine the "labels" (optimal beam pairs) for the training data. This process is repeated until satisfactory accuracy is achieved, making the model more robust to real-world variations.
  • CNN Model: A Convolutional Neural Network is trained using the 3D grid encoded environmental features. The CNN learns to predict the optimal beam pair index (BPI) for a given environmental state.

Datasets

The paper describes a simulation environment for V2I communication. They generated a dataset of 1000 simulated scenes for training and testing. The specific details of the dataset generation involve a ray-tracing simulator to capture channel state parameters. They compare their proposed 3D grid coding method against three other coding methods:

  • Ordered Location Coding Method (OLCM)
  • Natural Order Coding Method (NOCM)
  • Two-Dimensional Occupancy Grid (2DOG)

Key Insights and Ideas

  • Environmental Context is Crucial: The paper strongly emphasizes that simply knowing a vehicle's position isn't enough for effective mmWave beam tracking. The surrounding environment (other vehicles, buildings, etc.) plays a significant role and needs to be incorporated.
  • 3D Encoding for Richer Features: Representing the environment in a 3D grid, rather than simpler 2D methods, provides a more comprehensive feature set for the CNN, leading to better performance.
  • Iterative Label Refinement Boosts Accuracy: The iterative optimization of label data is a key innovation. It allows the model to learn from its own predictions, effectively improving the quality and representativeness of the training data, especially when real-world data is scarce or complex.
  • Robustness to Noise: The proposed 3D grid coding method demonstrates good immunity to positioning errors from sensors, which is a critical factor in real-world V2I deployments.

Key Results

  • The proposed 3D grid coding method, combined with the CNN and iterative label optimization, achieved the highest alignment probability compared to OLCM, NOCM, and 2DOG.
  • The alignment probability improved by approximately 0.4%, 0.5%, and 1% over OLCM, 2DOG, and NOCM, respectively.
  • The iterative label optimization further boosted the beam search accuracy by about 3% compared to using unoptimized theoretical labels.
  • The method showed good performance even with small data volumes and demonstrated robustness to positioning errors, with the y-axis error (lane deviation) having a more significant impact than the x-axis error (positional shift).

Limitations and Future Work

  • Simulation-Based: The results are based on simulations. While the simulation environment is detailed, real-world testing would be necessary to fully validate the approach.
  • Specific Scenario: The study focuses on a specific urban two-lane scenario. Future work could explore the effectiveness of the strategy in more complex and diverse environments.
  • Computational Cost: While the paper aims to reduce beam search overhead, the 3D grid encoding and CNN training can still be computationally intensive.
  • Future Work: The authors suggest performing further tests in other communication scenarios to verify the strategy's effectiveness. They also imply that exploring more advanced CNN architectures or optimization techniques could yield further improvements.

Summary updated

BibTeX
@article{zhang2022mmwave,
  author = {Lulu Zhang and Weizhi Zhong and Junjie Zhang and Zhipeng Lin and Zhuoming Yang and Junzhi Wang},
  title = {{mmWave Beam Tracking for V2I Communication Systems Based on Spectrum Environment Awareness}},
  journal = {Symmetry},
  year = {2022},
  doi = {10.3390/sym14040677},
}
Figure