A Deep Learning Approach to Find Optimal Path in Underwater Networks Using ns3-ai

Abstract

Undersea communication has become increasingly common due to its varied applications including a collection of oceanographic information, environment monitoring, seismic and pollution monitoring, and many more [1]. The environment undersea is highly unstable due to its intermittent and noisy characteristics [1][2]. Therefore, the routing approach that comprehends the environment is the need of the hour. The reinforcement learning method is one such approach that performs action based on environmental conditions [3]. One of the machine learning methods called Reinforcement learning allows an agent to learn from the environment and behave accordingly. In this paper, the authors have used a deep learning approach, a class of reinforcement learning which uses neural networks to train agents. Ns3-ai framework provides the abstraction between the ns3 simulator and the ai framework [4]. Here, an underwater sensor network is simulated in ns3 and a deep-learning approach is used to train the agents. The connection between ns3 and the deep learning framework is established through ns3-ai. The deep learning framework trains the agents based on the data received from the ns3 simulator. The actions performed by the agents are transferred to ns3 simulator where the actual routing of the packets happens. The results are compared with Q learning algorithm. The deep learning approach outperforms Q learning in terms of delay and delivery time.