ABSTRACT
This paper investigates the effectiveness of Proximal Policy Optimization a deep reinforcement learning technique, Autoencoders for energy-efficient data transmission, and Gated Recurrent Unit based approach for enabling dynamic spectrum allocation in Mobile Ad-Hoc Networks. After conducting the experiments, it can be observed that the performance of all the techniques is highly competitive. Specifically, PPO had an average throughout of 150 Mbps, while that of Autoencoders and GRU was 143 Mbps and 156 Mbps correspondingly. Latency averaged at 9 ms for PPO, 11 ms for Autoencoders, and 8 ms for GRU, meaning that data transmission was highly efficient. Energy consumption was also low, with an average of 500 mJ for PPO, 485 mJ for Autoencoders, and 490 mJ for GRU. In addition to that, the results consistently observe high packet delivery ratios for all of the deep-learning-based approaches used, with a PD of 0.98 for PPO, 0.97 for Autoencoders, and 0.985 for GRU. It implies that using deep-learning technology might optimize the way frequencies are allocated, which in turn can boost the performance and reliability of MANETs.
