Measurements in wireless networks and its applications
Wireless networks are heavily deployed in offices, homes and campuses all around the world. The ease of deployment and cheap wireless card equipment are the prominent reasons for the success of IEEE 802.11 based wireless network. The wireless channel is fundamentally different to its wired counterpart and exhibit characteristics which are difficult to model. The simulation of wireless networks gives valuable insight into the wireless network fundamental limits but fail to fully incorporate the real practical scenarios. Recently the measurement of wireless network characteristics has enhanced the understanding of these networks in practical situations.
The motivation behind this project is to understand and characterize the behavior of wireless links. The real implementation of measurement infrastructure, and analysis of the developed techniques in real wireless networks. This includes, development of bandwidth estimation techniques which can take into account dynamic nature of wireless links, extraction of wireless link properties like physical PER and development and testing of more robust rate adaptation algorithms.
In this project: first we develop a wireless network testbed by using off-the-shelf wireless cards; second we provide an algorithm SPEEDO, which measures the available bandwidth for a single hop wireless network; third we investigate two methods of isolating the physical errors from contention ones.
Available Bandwidth in Wireless Networks
Our main concern in the wireless network measurements is to study the causes effecting the throughput of users in the network. We developed an algorithm SPEEDO which estimates the available bandwidth (AB) for a user in the wireless network. Our approach is to find the system utilization from the immediate surroundings of the user by monitoring the channel. This will give us the used proportion of the channel, while the unused proportion will give us the AB. Apart from utilisation the other major factor which determines a user throughput is the data rate at which a user can send information. We measure the relationship between data rate and packet error rate (PER), which can be used to select the data rate for a particular user in a stationary environment. By combing the data rate selection and channel utilization we implemented SPEEDO on our testbed and show that our approach indeed gives us a good estimate of AB. We also compared it against probegap, an active probing based technique to estimate available bandwidth in wireless networks. A paper related to this work will be published in the proceedings of IEEE LCN 2008.
Isolating Physical and Contention Errors
The major factor in the estimation of available bandwidth is the data rate (or raw capacity) at which a user can transmit information. Current rate selection algorithms perform badly in a congested wireless network because they base their rate selection decision on the packet error rate (which consists of physical and contention error components). In a congested network the contention based errors dominate the physical wireless channel errors and thus result in under utilisation of the network capacity caused by the lower data rates selection by the users. We investigate two approaches to isolate the physical errors from contention errors. First we try to find the errors where channel utilisation is low, which correspond to physical errors as it will have very small component of contention errors. Second we divide a packet into two parts using 802.11 wireless fragmentation technique, where only the first part has to face the contention based errors and the second part of the packet (fragment) only incur physical transmission loss. We implemented both of these approaches on our testbed and show that fragmentation based technique performs better than utilization based approach. A paper related to this is published in proceedings of the third ACM International Workshop on Wireless Network Testbeds, Experimental evaluation and Characterization (WiNTECH 2008).