Neural network based mobility prediction for handoff management in wireless MPLS network

Abstract

Growing demand for different services over cellular mobile networks has emphasized the necessity of QOS provisioning. However, nodes mobility jeopardizes the resource allocation process, and decreases the quality of service provided to delay sensitive traffic. So, in such networks, the remote location prediction has significant impact on near-optimum resource management procedure. In the other hand, in recent years, MPLS has been considered as the preeminent technology to incur QOS for integrated services. In this paper we propose a new location prediction method based on Neural Networks, to manage LSPs in an MPLS domain. Proposed predictor uses geographical characteristics of underlying area, in addition to the movement history of that remote. A set of confidence ratios is considered as the output of our predictor. That set is considered as a criterion for establishing and managing LSPs. Each output of the predictor indicates the degree of confidence for the corresponding neighboring cell, showing that how likely the remote may move to that cell. This procedure proposes two types of Pre-Established LSP, called "Simple LSP" and "ConstRaint based LSP". A set of simulations with typical assumptions has been carried out to evaluate the performance and response time of the proposed method on QOS.