The wireless network environment sensor: A technology-independent sensor of faults in mobile wireless network links.

摘要

With the proliferation of portable computing devices connected via wireless networks and with the availability of 3G wireless networks in the near future, there is a need for predicting and identifying wireless link failures so that users or systems can take appropriate action. This thesis explores the problem of detecting, predicting, and identifying wireless link faults in a technology-independent manner. We first explore which faults are predictable and which are not. Faults due to signal attenuation from large-scale path loss (free space attenuation and shadowing) are predictable. Faults due to small-scale effects such as multipath are best detected and compensated for within the wireless network device itself. We then propose a sensor, the Wireless Network Environment Sensor (WiNE Sensor) that detects and when possible, predicts link failures. In creating this sensor, rather than using a change detection method of determining link degradation, we develop a set of baselines to model known-quality links. We then extend the baseline concept to model failing links, which allows us to differentiate among the causes of predictable link failures. The creation of fault baselines led to the development of a second sensor, the Match WiNE Sensor, which identifies the cause of predictable link failures.; While there are many products that display wireless link signal strength for a specific device, we know of no system that detects, predicts, and identifies wireless link failures in a technology-independent manner. We developed autoregressive (AR) models of both known-quality and failing wireless links and extended Akaike's Final Prediction Error scheme to determine the optimum window size for AR models over a limited duration. We also adapted a sensor fusion algorithm created by our colleagues Thottan and Ji to determine wired backbone link quality.; As proof of principle, we implemented the WiNE Sensor and Match WiNE Sensor in software and tested them on two different computing platforms using three different wireless networks. The WINE Sensor provides, on average, 42 seconds of warning time for free-space attenuation failures and 31 seconds for shadowing failures. The Match WiNE Sensor correctly identifies the cause of link failures 82% of the time.