Journal
JOURNAL OF COMMUNICATIONS SOFTWARE AND SYSTEMS
Volume 18, Issue 2, Pages 111-123Publisher
CROATIAN COMMUNICATIONS & INFORMATION SOC
DOI: 10.24138/jcomss-2021-0154
Keywords
Computation Offloading; Dynamic Price; System Utility; Deep Reinforcement Learning (DRL); Unmanned Aerial Vehicles (UAVs); MEC
Ask authors/readers for more resources
This paper proposes a method for using dynamic pricing to provide computation services to vehicles through Mobile Network Operators deploying Mobile Edge Computing servers on unmanned aerial vehicles. The optimization problem is solved using a Deep Reinforcement Learning algorithm, and extensive simulation results demonstrate the superiority of the proposed pricing model.
In the fifth-generation (5G) cellular network, the Mobile Network Operator (MNO), and the Mobile Edge Computing (MEC) platform will play an important role in providing services to an increasing number of vehicles. Due to vehicle mobility and the rise of computation-intensive and delay-sensitive vehicular applications, it is challenging to achieve the rigorous latency and reliability requirements of vehicular communication. The MNO, with the MEC server mounted on an unmanned aerial vehicle (UAV), should make a profit by providing its computing services and capabilities to moving vehicles. This paper proposes the use of dynamic pricing for computation offloading in UAV-MEC for vehicles. The novelty of this paper is in how the price influences offloading demand and decides how to reduce network costs (delay and energy) while maximizing UAV operator revenue, but not the offloading benefits with the mobility of vehicles and UAV. The optimization problem is formulated as a Markov Decision Process (MDP). The MDP can be solved by the Deep Reinforcement Learning (DRL) algorithm, especially the Deep Deterministic Policy Gradient (DDPG). Extensive simulation results demonstrate that the proposed pricing model outperforms greedy by 26% and random by 51% in terms of delay. In terms of system utility, the proposed pricing model outperforms greedy only by 17% In terms of server congestion, the proposed pricing model outperforms random by 19% and is almost the same as greedy.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available