| dc.description.abstract | Satellites have for some years made it possible to transfer data over long distances, and from desolated locations. Due to their natural broadcast ability they have some major advantages compared to terrestrial links. This ability has made satellites a useful tool for military, science, multimedia and wireless communication.
Compared to a few years ago, we are today able to make more accurate measurements, in different environments. Sensors have recent years become more intelligent, and can work independently without being controlled by humans or computers. Other improvements are their mobility, and ability to work wireless.
In recent years there has been a lot of research addressed to combining satellite communication and wireless sensors in ad hoc networks. If these two wireless technologies can be incorporated in the same network, it would result in a lot of possibilities. Scenarios where desolated wireless sensors measures independently, and sends data through a satellite link, to a gathering center are now possible. ZigBee Alliance has developed a protocol stack architecture, specialized for small wireless sensors working in ad hoc structured networks. IEEE has developed a standard for the two lowest OSI-layers, IEEE 802.15.4, which the ZigBee layers work upon.
This thesis will study and evaluate the performance of different TCP versions throughout heterogeneous wireless networks. Today is TCP NewReno the most common TCP protocol, but we will also consider other TCP versions, like TCP Westwood+, TCP Hybla and TCP Vegas. The most important aspect of this evaluation is the average throughput. To test the performance of the different TCP versions over the heterogeneous wireless links, we tested over a real satellite link, and a virtual wireless channel. We emulated transmission from a sensor node sending to a sink, through this channel. There are different ways to simulate this channel. We have looked at two models; Characteristic Model and the Two State Markov Model. The latter option is a model which uses four different parameters to define the channel. These parameters are bandwidth, delay, alpha and beta. Alpha and beta defines the transition probability between a good state (perfect transmission) and a bad state (no transmission). The first model, the Characteristic Model, uses IEEE 802.15.4 parameters and a waypoint mobility model to calculate the probability of packet loss. We have extended the already existing ACE Emulator, to emulate an ad hoc network. For our testing we decided to use the Two State Markov Model.
Based on the results we achieved, we can conclude that TCP Vegas performed best over the satellite link. We tested TCP NewReno, TCP Westwood+, TCP Hybla and TCP Vegas for this link. Unexpectedly performed TCP NewReno second best in this test. Over the virtual wireless channel performed TCP NewReno best. We tested TCP NewReno and TCP Westwood+ over this channel. Our overall conclusion is that TCP NewReno performs best in general throughout our heterogeneous wireless networks. We have to take into account the fact that we experienced some problems with both networks during testing, but the results give a strong indication of TCP NewRenos quality and range of use. | en |
