Enhancing Vehicular Cooperative Downloading with Continuous Seeding through Deep Reinforcement Learning

Niebisch M, Pfaller D, Djanatliev A (2023)

Publication Language: English

Publication Type: Conference contribution, Conference Contribution

Publication year: 2023

Pages Range: 1-6

Event location: Panama-Stadt, Panama PA

DOI: 10.1109/LATINCOM59467.2023.10361894

Abstract

Vehicular communication needs are steadily increasing. While some data is unique for every vehicle, other data, like updates, have to be distributed to many cars. However, the sole reliance on cellular networks unnecessarily increases data costs and data usage. A vehicular cooperative downloading scheme utilizes additional communication technologies, like direct communication between vehicles, to exchange data. For an efficient use of this communication channel, enough data has to be present in the vehicles, which can be exchanged. A seeding strategy introduces parts of the update into the vehicles to start the cooperative downloading procedure. However, if too few data chunks are present, the direct communication link is not fully utilized, while over-provisioning the vehicles with data unnecessarily increases cellular usage. In this paper, we first introduce an initial seeding strategy and find its optimal results for a baseline comparison. We then use deep reinforcement learning to develop a strategy which continuously determines how much data needs to be added to the vehicle fleet at any time, optimizing the overall system's performance. Our results consider various update sizes and mobility scenarios and show that proximal policy optimization is a suitable approach to lower the usage of cellular communication in a cooperative downloading scheme. We can reduce the use of cellular networks by more than 98%, while continuously seeding improves the best initial seeding approach by more than 60%.

Authors with CRIS profile

Michael Niebisch Computer Science 7 (Computer Networks and Communication Systems) Anatoli Djanatliev Computer Science 7 (Computer Networks and Communication Systems)

Related research project(s)

Heterogeneous vehicle networks for data transmission in the field April 1, 2020 - Sept. 30, 2023

Involved external institutions

Audi AG DE Germany (DE)

How to cite

APA:

Niebisch, M., Pfaller, D., & Djanatliev, A. (2023). Enhancing Vehicular Cooperative Downloading with Continuous Seeding through Deep Reinforcement Learning. In Proceedings of the 2023 IEEE Latin-American Conference on Communications (LATINCOM) (pp. 1-6). Panama-Stadt, Panama, PA.

MLA:

Niebisch, Michael, Daniel Pfaller, and Anatoli Djanatliev. "Enhancing Vehicular Cooperative Downloading with Continuous Seeding through Deep Reinforcement Learning." Proceedings of the 2023 IEEE Latin-American Conference on Communications (LATINCOM), Panama-Stadt, Panama 2023. 1-6.

BibTeX: Download