A stochastic technique for on-line prediction and tracking of wireless packet networks

A stochastic learning technique to predict and track the states of a two state wireless packet network that changes according to a Markov process is proposed. The proposed technique does not use pilot symbols or training data for state estimation. Instead, prediction and tracking the network state information is performed in an on-line fashion using stochastic iterative control suitable for real-time/pseudo real-time applications. The proposed method is computationally simple making it a good candidate for low power wireless applications. The computational power consumption due to the predictor/tracker can be traded-off easily for accuracy and speed. Experimental results are promising for the speed and accuracy with which the time-varying network state is predicted and tracked. Theoretical convergence of the iterative control algorithm is also discussed.