Assessing dependability of wireless networks using neural networks

Abstract

Critical infrastructures such as wireless network systems demand dependability. Dependability attributes addressed in this paper include availability, reliability, maintainability and survivability. This research uses computer simulation and artificial intelligence to introduce a new approach to assess dependability of wireless networks. The new approach is based on the development of a neural network, which is trained to investigate availability, reliability, maintainability, and survivability attributes (ARMS) of a wireless network. In this work, given a variety of reliability and maintainability attribute scenarios for a wireless infrastructure, the resulting impact on network availability and survivability are determined. Component mean time to failure (MTTF) is used to model reliability, while the mean time to restore (MTR) is used for maintainability. Here, unavailability, the complement of availability, is defined as the fraction of time the entire network system is down, while survivability is the fraction of network user capacity up over time. Both availability and survivability can be instantaneous or averaged over some period. The data set, which is used to train the neural network, is obtained from simulation experiments with a range of component MTTF and MTTR. In addition, the number of times a new regulatory outage reporting threshold is surpassed is also determined. This research also focuses on the relative performance of neural network modeling compared to analytical and simulation techniques for assessing the ARMS attributes of a wireless network, and the additional insights that can be obtained from NN modeling.