Experimental performance evaluation of wireless local area networks using machine learning

The performance of a wireless network is often much lower than what one might expect. Although the data rate is defined in the IEEE 802.11b standard as 11 Mbps, it can be only 2.5--4 Mbps in practice. A major reason for this poor performance is that most devices, particularly wireless access cards, use default settings that, while they might "work" in most environments, are not guaranteed to work well. Environmental conditions, such as access point traffic or interference, greatly affect which wireless access card settings are optimal. Thus, an access card might be able to improve its performance if it could adjust its parameters according to what environmental conditions it detects. For example, if the access card detects a low signal to noise ratio, it could increase its transmitting power to avoid packet loss. On the other hand, if the access card determines that the signal to noise ratio is high, it could decrease its transmitting power to increase battery life.; In this project, we examine the application of machine learning to improve the performance of wireless network. First, we define a case study and select variables to randomly generate a number of scenarios. Next, we collect information for these scenarios using a wireless network simulator, Opnet(TM). Finally, we apply machine learning to learn models for predicting the throughput of a wireless card. Our hypothesis is (1) that machine learning can be used to learn models for predicting throughput given environmental conditions and access card settings and (2) that the learned models can be used to select access card settings that will improve performance in a variety of environmental conditions.