Application of Neural Networks for Achieving 802.11 QoS in Heterogeneous Channels

Abstract

In error-prone IEEE 802.11 WLAN (Wireless Local Area Network) environments, heterogeneous link qualities can significantly affect channel utilizations of mobile stations and consequently the user-perceived QoS (Quality of Services) of multimedia applications. In this paper we propose a novel optimization framework which provides QoS by adjusting IWSs (Initial Window Size) according to current channel states and QoS requirements. It is a table-driven approach which off- line pre-establishes the table of the best IWSs based on a cost-reward function. Neural networks are utilized to learn the mapping correlation and then to generalize that to other situations of interest. At runtime, the IWS of each user can thus be determined optimally with a simple table lookup rapidly without much time spent on learning about the nonlinear and complicated correlation. A video streaming transmission scenario is used to evaluate the performance of our scheme. The simulation results demonstrate that the proposed mechanism can effectively provide QoS for each user when the capacity of the network is sufficient for the requirements of all users.