Scheduling Real-time Traffic over Hybrid Channels

The emergence of 5G networks has heightened the need for real-time traffic management to ensure timely throughput and meet strict deadlines, maintaining the relevance and effectiveness of transmitted data. Delayed information often loses its practical value, while steady and timely data flow is essential to support high-demand scenarios and meet modern expectations for low-latency, high-reliability communication. Hybrid channel models, combining high-speed but less reliable channels with slower, highly reliable ones, offer a promising approach to enhance network reliability. However, effectively balancing these channels under strict scheduling constraints remains a challenge, as does maintaining the freshness of transmitted data, often measured by the age-of-information (AoI). AoI prioritizes updates to keep data relevant for real-time decision-making, adding complexity to scheduling tasks. Reinforcement learning (RL) provides a powerful method to address these challenges. By learning from dynamic conditions, RL-based approaches can develop adaptive scheduling policies that meet deadlines, ensure steady throughput, and maintain data relevance. This work focuses on leveraging hybrid channels and RL to address the stringent requirements of modern wireless networks. The first study examines scheduling algorithms for deadline-constrained traffic over hybrid channels. Two cases are considered: with and without access to Channel State Information (CSI). For each, the User Selection First (USF) and Channel Selection First (CSF) algorithms are developed, stabilizing virtual queues and enhancing throughput under varying conditions. The second study addresses scheduling real-time traffic with general patterns, aiming to improve both AoI and timely throughput. Formulated as a Markov Decision Process (MDP), an RL-based algorithm, RL-ART, is introduced. Built on a Double Deep Q-Network (DDQN), RL-ART adapts scheduling decisions to network dynamics, achieving notable gains in AoI and timely throughput across diverse conditions. Extensive experiments demonstrate that USF and CSF effectively fulfill stringent throughput requirements, reduce packet-dropping rates, and stabilize virtual queues. RL-ART further enhances AoI and sustains high throughput under varying simulated scenarios, addressing key challenges in hybrid channel scheduling and aligning with the performance demands of 5G and future wireless networks.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.