Abstract
Internet of Vehicles (IoV) is expected to become the central infrastructure to provide advanced services to connected vehicles and users for higher transportation efficiency and security. A variety of emerging applications/services bring explosively growing demands for mobile data traffic between connected vehicles and roadside units (RSU), imposing the significant challenge of spectrum scarcity to IoV. In this paper, we propose a cooperative semantic-aware architecture to convey essential semantics from collaborated users to servers for lowering the data traffic. In contrast to current solutions that are mainly based on piling up highly complex signal processing techniques and multiple access capabilities in terms of syntactic communications, this paper puts forth the idea of semantic-aware content delivery in IoV. Specifically, the successful transmission of essential semantics of the source data is pursued, rather than the accurate reception of symbols regardless of its meaning as in conventional syntactic communications. To assess the benefits of the proposed architecture, we provide a case study of the image retrieval task for vehicles in intelligent transportation systems. Simulation results demonstrate that the proposed architecture outperforms the existing solutions with fewer radio resources, especially in a low signal-to-noise-ratio (SNR) regime, which can shed light on the potential of the proposed architecture in extending the applications in extreme environments.
Introduction
Traditional communication systems are "syntactic"—they focus on the perfect reconstruction of every single bit (0s and 1s) sent over the wire. However, as we approach the Shannon capacity limit, this approach is struggling to keep up with the data explosion in the Internet of Vehicles (IoV). Autonomous driving and traffic management require massive amounts of data from cameras and sensors. Currently, if multiple vehicles observe the same event, they might all attempt to upload high-resolution data, leading to massive redundancy and spectrum scarcity. Furthermore, in "extreme environments" where the signal-to-noise ratio (SNR) is low, traditional bit-perfect transmission often fails completely. This research argues that for many IoV tasks, we do not need the bits; we need the information or "meaning" relevant to the task at hand.
Methodology: The Co-SC Framework
The proposed Cooperative Semantic Communication (Co-SC) framework is a deep learning-based architecture designed for multi-user cooperation. Unlike standard systems that separate source and channel coding, Co-SC integrates them to prioritize the transmission of "meanings" over raw data.
The architecture consists of several key components:
- Semantic Encoder/Decoder (Sem-Codec): These modules use deep neural networks (e.g., ResNet-50) to extract high-dimensional features relevant to a specific goal, such as identifying a vehicle, while discarding irrelevant background data.
- Joint Source and Channel (JSC) Codec: This maps semantic features directly to channel symbols, making the transmission inherently more robust to wireless noise compared to traditional digital separation.
- Cooperative Task Performer: Located at the receiver (e.g., an Edge Server), a fusion module combines features from multiple users. By leveraging the spatial correlation between different views of the same object, the system can achieve higher accuracy than any single user could alone.
- Knowledge Base: A shared "understanding" between the transmitter and receiver, trained offline, which allows the system to interpret the extracted features correctly.
Case Study: Cooperative Vehicle ID-Retrieval
To evaluate the system, the researchers focused on the task of cooperative ID-retrieval—identifying a specific vehicle across multiple camera views. They utilized the VeRi-776 dataset, a standard benchmark containing over 50,000 images of 776 vehicles captured by 20 cameras in real-world settings.
In this scenario, multiple vehicles or RSUs capture different angles of a target vehicle. The Co-SC system extracts and transmits only the essential semantic features required for identification.
Results and Performance Analysis
The simulation results highlight significant gains in both efficiency and robustness:
- Extreme Efficiency: The proposed semantic method required only 16 symbols to achieve superior results compared to traditional digital methods (JPEG + LDPC + BPSK), which required approximately 310,000 symbols. This represents an orders-of-magnitude reduction in bandwidth requirements.
- Robustness at Low SNR: At a very low SNR of -3 dB, the Co-SC architecture outperformed non-cooperative deep learning baselines by 28.9% in Rank-1 accuracy. This demonstrates "graceful degradation"; unlike traditional systems that fail completely when noise is high, Co-SC maintains performance because the core "concepts" often survive interference.
- Semantic Focus: Visualization using Grad-CAM (Gradient-weighted Class Activation Mapping) confirms that the model successfully ignores "irrelevant" data.
As shown in the visualization, the Co-SC model focuses its "attention" on the vehicle itself, effectively filtering out background elements like trees or roads that do not contribute to the identification task.
Conclusion and Future Work
The Co-SC architecture represents a paradigm shift from "how to send" to "what to send." By leveraging multi-user correlation and task-oriented filtering, it addresses the spectrum scarcity and noise sensitivity of modern IoV.
However, several challenges remain for future research. The Knowledge Base is currently static; developing mechanisms for it to evolve as environments change is critical. Furthermore, the field lacks a formal "Semantic Shannon Theory" to mathematically quantify semantic noise and capacity. Finally, Security is a major concern, as poisoning a shared knowledge base could compromise the entire communication chain. Addressing these hurdles will be essential for the real-world deployment of semantic-aware IoV infrastructures.
Summary updated