Work

Proximal Policy Optimization-Based Joint Resource Allocation for Quality of Service-Aware 5G Vehicle-to-Everything Networks

Year: 2026
Type: conference-paper
Abstract: This paper presents an adaptive multi-dimensional Proximal Policy Optimization (PPO) framework for joint optimization of power control, numerology selection, and resource block (RB) allocation in 5G Vehicle-to-Everything (V2X) networks. Unlike existing approaches that optimize these parameters independently or in pairs, the proposed framework employs a unified deep reinforcement learning agent that simultaneously controls transmit power (1-20 dBm), subcarrier spacing (15-60 kHz), and dynamic RB distribution across three priority-based service classes: Safety-critical (SSV+), Comfort (C1), and Infotainment (C2). All system-level parameters (bandwidth, RB structure, numerology set, and UE power range) are chosen in accordance with 3GPP NR-V2X specifications in TR 37.885 and TS 38.214 so that the simulation setup remains consistent with standardized assumptions. The system features a discrete-continuous hybrid action space, enabling the PPO agent to learn context-aware numerology switching while maintaining smooth power and RB allocation convergence. Extensive simulations based on realistic V2X scenarios show superior performance, with SSV+ latency reduced to 4 ms (38% improvement), packet reception ratio exceeding 95%, and energy efficiency of 1.95 Satisfaction/Watt at an average power consumption of 13 dBm. The learned policy exhibits intelligent behaviors including adaptive numerology selection based on mobility patterns, targeted power adjustments for reliability assurance, and elastic RB reallocation during congestion periods. These results validate the effectiveness of joint multi-dimensional optimization for next-generation V2X networks requiring both ultra-reliable low-latency communication and energy efficiency.
Source: HAL
Language: English
Open Access status: green
OpenAlex ID: W7214566630
FWCI:
Cites:
Cited by:
Related to: 10