Short Term Wireless Channel State Prediction Using Markov Models and Supervised Learning

Packet error rate is one of the most important factors that determine the quality of service achieved over wireless networks. However, long term characteristics of the error process at the link-layer are not sufficient to guide decisions which may require short-term knowledge of the link state, such as, scheduling, and rate adaptation. We expect that transmitters will benefit from short term link state knowledge since avoiding transmitting during periods where channel quality is poor may save unnecessary transmissions and thus limit bandwidth wastage and energy usage. In this study two types of schemes are evaluated that exploit short-term phenomena of the error process at the link-layer in order to predict link state. The first scheme employs a Naive Bayes Classifier (NBC) while the second one relies on a Markov chain model. Preliminary simulation results derived from a simple topology reveal that the NBC-based scheme can predict failed transmission with an accuracy of up to 84.1%.