TC-PAA: Deep Learning-Enabled QoS Enhancement Scheme for Cooperative Internet of Vehicles

Muhammad Adil, Houbing Song, Neeraj Kumar, Mian Ahmad Jan, Amiya Nayak, Ahmed Farouk, Zhanpeng JinIEEE Trans. Vehicular Technology 2024 · DOI

Abstract

As Intelligent Transportation Systems (ITS) evolve, the Cooperative Internet of Vehicles (C-IoV) faces significant challenges in maintaining Quality of Service (QoS) due to network congestion and "broadcast storms." This paper presents the Traffic Congestion and Priority-Aware Algorithm (TC-PAA), a hybrid framework designed to prioritize emergency vehicle communications. By utilizing Deep Neural Networks (DNN) for real-time traffic classification and a dual-path routing strategy involving micro base stations (μBS) and the Dijkstra algorithm, TC-PAA significantly reduces latency and energy consumption compared to traditional routing protocols.

Introduction

The Internet of Vehicles (IoV) is the foundational infrastructure for modern road safety and traffic efficiency. However, high-speed mobility and dynamic network topologies often lead to data congestion. In standard protocols like AODV or DSDV, critical emergency data—such as ambulance alerts—often compete for bandwidth with non-essential background data, leading to potentially life-threatening delays. The TC-PAA framework addresses this gap by introducing an intelligent, priority-aware routing scheme that ensures emergency services receive a dedicated digital path.

The TC-PAA Framework

The TC-PAA methodology is structured into three distinct operational stages: intelligent categorization, dual-path routing, and client-side management.

1. Intelligent Traffic Categorization

The system first processes real-time video feeds from roadside units using computer vision techniques, specifically OpenCV’s background subtraction, to identify vehicles. These identified entities are then passed to a Deep Neural Network (DNN) for classification. The DNN categorizes traffic into two primary classes: Emergency (high priority) and Traditional (ordinary).

The internal processes of the Deep Neural Network used for vehicle classification within the TC-PAA framework.

As shown in Figure 1, the DNN architecture processes raw data samples through multiple hidden layers and a SoftMax function to determine the posterior probability of a vehicle belonging to a specific class. This high-accuracy classification is critical for the subsequent routing decisions.

2. Dual-Path Routing Strategy

Once categorized, data packets follow different network trajectories:

  • Priority Traffic: Emergency data is routed through a micro base station (μBS). This provides a direct, low-latency link to the destination, effectively bypassing the multi-hop delays inherent in standard mesh networks.
  • Ordinary Traffic: Regular vehicle data is managed via the Dijkstra algorithm. This approach calculates the most efficient multi-hop path based on hop counts to minimize overall network strain without interfering with priority channels.

3. Client-Side Management

By decentralizing the classification and decision-making processes to the vehicles or roadside units (the "client-side"), TC-PAA reduces the signaling overhead typically required for centralized network coordination. This shift enhances the system's responsiveness in high-speed highway environments.

Algorithmic Logic

The classification logic is governed by a deterministic set of rules that translate DNN outputs into routing actions. The following algorithm outlines the transition from raw data to prioritized classification:

The algorithmic flow for classifying traffic into priority and ordinary classes based on feature matching.

Performance Evaluation and Results

The TC-PAA framework was benchmarked against industry-standard protocols, including AODV, DSDV, and LEACH. The results demonstrate substantial improvements across several key metrics:

  • Energy Efficiency: TC-PAA consumed only 12.288 mW for emergency vehicle communications, a dramatic reduction from the 75.776 mW required by the AODV protocol.
  • Latency and Hop Count: By utilizing the μBS, the hop count for emergency messages was reduced to just 1 or 2. In contrast, traditional ad-hoc routing often required 17 or more hops, leading to significant end-to-end delays.
  • Throughput Stability: Simulations involving up to 1,000 messages showed that the μBS path maintained a stable and significantly higher throughput (bits per second) compared to Dijkstra or LEACH, which suffered from performance degradation as message volume increased.

Conclusion and Future Work

TC-PAA offers a robust solution for enhancing QoS in C-IoV environments by combining deep learning-based classification with specialized hardware utilization. While the simulation results are promising, future research must address real-world challenges such as extreme weather conditions and physical signal interference. Additionally, securing these systems against "priority spoofing" and managing the maintenance costs of roadside infrastructure remain critical hurdles for large-scale deployment.

Summary updated

BibTeX
@article{adil2024tc,
  author = {Muhammad Adil and Houbing Song and Neeraj Kumar and Mian Ahmad Jan and Amiya Nayak and Ahmed Farouk and Zhanpeng Jin},
  title = {{TC-PAA: Deep Learning-Enabled QoS Enhancement Scheme for Cooperative Internet of Vehicles}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2024},
  doi = {10.1109/tvt.2024.3396691},
}
Figure