Towards a wireless lexicon
- The wireless links in a network encounter variations in received signal power due to fading, shadowing and movement in the environment. These variations introduce correlations in packet reception over time and across multiple links. These correlations affect network performance. Through measurements from several wireless networks, this dissertation presents metrics that quantify the degree of these correlations. Many protocols do not consider these correlations. Using real experiments on different networks, this work shows by how much network performance degrades when protocols ignore such correlations. This work further shows how to tune and modify protocols, based on the metrics it presents, to achieve network performance improvements. Temporal correlation of packet reception on a single link has implications to protocols that determine when a packet is to be transmitted on that link. For example, medium access control (MAC) protocols retransmit a packet immediately after a failure. If losses on a link are highly temporally correlated, then an immediate retransmission is also likely to fail. This work presents a temporal correlation metric, $\beta$ derived from a function of probabilities of packet success conditioned on the history of packets on that link. It computes $\beta$ of a link by comparing this function to the case when packet reception is independent. $\beta$ shows that wireless networks can have many temporally correlated links: a measured network observed nearly 85\% of its links to be temporally correlated. $\beta$ is indicative of performance of MAC protocols. On highly temporally correlated links, backing off after a failure improves the packet success rate. $\beta$ is useful in deriving this backoff: modifying link backoff, based on $\beta$, for a collection protocol reduced the average number of transmissions per node by 35\%. Similarly, a packet's reception correlation across multiple links -- spatial correlation -- has implications to routing protocols that use multiple links to forward a packet. For example, after a forwarding node fails, routing protocols choose an alternate forwarding node without taking into account how correlated the two forwarding nodes are. If the losses on both forwarding nodes are highly correlated, then the new forwarding node is also likely to fail. This work presents a spatial correlation metric, $\kappa$ for every link pair. $\kappa$ is indicative of performance of protocols like opportunistic routing and network coding protocols. $\kappa$ is also useful in showing when network coding protocols are beneficial. Choosing a no-coding dissemination protocol over its network coding counterpart, based on $\kappa$, reduced the total time for dissemination by up to 45\%. This dissertation focusses on two fundamental aspects of correlations and evaluates their importance to many protocols. There are other correlations that exist in a wireless network such as the reception correlation across different transmission channels. This inter-channel correlation has implications to protocols that use multiple channels. This dissertation leaves such correlations as future work. We believe that metrics such as the ones presented in this dissertation form a wireless lexicon that the wireless community can use to explain wireless protocol performance. We hope that such a lexicon will also allow us to develop more efficient future wireless protocols.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Srinivasan, Kannan, 1978-
|Stanford University, Department of Electrical Engineering.
|Tobagi, Fouad A, 1947-
|Tobagi, Fouad A, 1947-
|Statement of responsibility
|Submitted to the Department of Electrical Engineering.
|Ph.D. Stanford University 2010
- © 2010 by Kannan Srinivasan
Also listed in
Loading usage metrics...