Loading...

Semantic Communications Enabled Multimedia Task Offloading For VANETs Using Machine Learning

Soheilian, Mohammad Hosein | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58591 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Safaei, Bardia
  7. Abstract:
  8. With the rapid growth of the Internet of Things (IoT), vehicular networks have become one of its most significant branches, serving as a key enabler for communication between vehicles and intelligent infrastructures. These networks depend on large-scale and diverse sensory data—from cameras, radars, and especially LiDAR sensors—to perceive their surrounding environment. Fast and accurate processing of this data is essential for critical applications such as autonomous driving, hazard prediction, and traffic management. Due to vehicles’ limited computational resources and the high latency associated with cloud communication, edge computing has emerged as an effective solution to mitigate delays and reduce network load. However, transmitting raw LiDAR data leads to excessive communication overhead and inefficient bandwidth utilization. This study introduces an adaptive semantic communication framework that leverages deep learning to intelligently compress and transmit LiDAR data based on real-time channel conditions. The proposed model includes an SNR-aware adaptive mechanism, which dynamically adjusts the compression ratio according to current channel quality and sends only the most essential semantic information. In contrast to previous approaches that required distinct models for each compression rate, our method employs a single unified model trained for all conditions. This design optimizes data transmission by recognizing the semantic significance of 3D points within the LiDAR data. Experimental evaluations show that, despite a simpler architecture, the proposed model attains 85.5% overall accuracy, only 2% below the best existing method, while cutting network task execution time by up to 23%. Furthermore, compared to other semantic communication techniques, it achieves 2.6× to 3.5× faster processing with markedly lower computational overhead. These results highlight that adaptive semantic modeling for LiDAR-based task offloading can deliver simultaneous high accuracy, minimal latency, and efficient data communication in vehicular networks representing a key advancement toward smarter driving and enhanced road safety systems
  9. Keywords:
  10. Internet of Things ; Vehicular Networks ; Task Offloading ; Semantic Communication ; Machine Learning

 Digital Object List

 Bookmark

No TOC