Hybrid CNN-LSTM-Enhanced AOMDV for Performance Optimization of AOMDV Routing Protocol in MANETs

Mobile Ad Hoc Networks (MANETs) are characterized by their dynamic and decentralized nature, frequently leading to unstable routing paths and diminished network performance. Among the existing solutions, the Ad hoc On-demand Multi-path Distance Vector (AOMDV) routing protocol provides multi-path fault tolerance; however, it still encounters challenges related to route stability, packet delivery efficiency, and adaptability in response to changing topologies. Various efforts have been undertaken to improve routing in MANETs by applying machine learning techniques. While reinforcement learning and supervised learning methods have demonstrated potential, they often lack temporal context. Long Short-Term Memory (LSTM) networks address this limitation by retaining memory of past network behaviours. Conversely, Convolutional Neural Networks (CNNs) extract hierarchical spatial features, rendering them suitable for analyzing topological patterns. The integration of CNN and LSTM can capitalize on the strengths of both, thereby enhancing route prediction accuracy. This paper proposes and develops a hybrid CNN-LSTM deep learning model with an accuracy of 99.4% to augment the AOMDV protocol by predicting more reliable and efficient routing paths based on historical network metrics. The model is trained using simulated routing data in Python on Google Colab and is integrated into the AOMDV decision-making process. Evaluation across varying node densities reveals the enhanced AOMDV protocol achieved up to a 73% improvement in throughput (from 140 kbps to 242 kbps for 50-node scenarios), a 40% reduction in routing overhead, and a 57% increase in PDR (from 0.07 to 0.11), all without increasing end-to-end delay. The results confirm that the proposed deep learning-enhanced AOMDV protocol surpasses the traditional version, particularly in high-mobility and high-density scenarios.