Packet Loss Differentiation Over Manet Based on a BP Neural Network

Abstract

An adaptive distributed routing algorithm is essential in MANETs, since there is no central routing system. Actually, there is no central point of coordination; each node is responsible for forwarding data packets to other nodes, thereby acting as router and host. A packet might travel through multiple intermediary ad hoc nodes in order to arrive to its destination, while the nature of wireless multi-hop channel is bringing in various types of packet losses. This paper focuses on three main reasons of online packet losses in MANETs: (1) losses due to wireless link errors; (2) losses due to congestion; and (3) losses due to route alteration. It proposes a deep learning-based algorithm for packet loss discrimination. The algorithm uses the backpropagation neural network (BPNN) concept. We performed simulation experiments for evaluating the performance of the proposed loss discrimination algorithm under different network configurations. Through simulation results, we confirmed that the proposed algorithm improves packet loss discrimination and route alteration in the network. It also reduces congestion and increases network throughput.