A Study Of Fault-tolerant Data Collecting Protocol For Disaster Scenario In Wireless Sensor Networks

Wireless sensor networks plays an important role in the monitoring of emergency or disaster scenarios such as floods, fires, landslides or earthquakes where is dangerous for human. However, these catastrophic environments will pose an fundamental design challenge for a robust and efficient data collection. On the one hand, the sensor nodes used to collecting data may fail to work suddenly or unpredictably as they may melt or get smashed in the harsh environments. On the other hand, in such emergency sensor networks, the rate of data generation usually exceeds beyond the capacity to deliver the data to the sink and some data have to reside temporarily in the nodes waiting for their turns to be delivered to the sink. Hence, unpredictable failures will cause the potential loss of some data. Considering the above problems, how to safely and efficiently deliver sensed data to the sink is a critical and challenging problem in such systems.As the emergence of network coding technique, many researchers have proposed kinds of network coding techniques to imporve data persistence in disaster scenarios. Growth Codes proposed by Kamra in SIGCOMM2006is a typical network coding protocol. By the means of growing encoding and random data distributions, Growth Codes effectively imporve data persistence of network. However, Growth Codes give birth to a large amount of redundant codewords distributed in the network, and the proportion of useful codewords are decreasing sharply as the number of recovered data increasing. Therefore, the collecting performance of Growth Codes protocol is greatly affected, especially in the sparse sensor networks.To solve the above problems of Growth Codes, in this paper, we propose the novel local-configuration network model, distributed buffering mechanism, filtering method, multi-source encoding mechanism and so on. They collaborate together to greatly improve the collecting efficiency in various network connectivity. The main parts of this paper includes:(1) To solve the poor data collecting efficiency of Growth Codes, we propose a local-configuration network model. And then by using the collecting property of single-hop collecting tree, we design distributed buffering mechanism, filtering method and Filtering and Growth Codes based protocol (FGCP). This protocol could effectly raise the probability of successfully decoding a received codeword on sink node, thus the gathering performance in emergency scenarios is greatly improved. Experiment results show that FGCP effectively imporves the collecting efficiency on the base of data persistence. (2) Growth Codes in distributed sense networks is implemented by combining codewords with its sensed data. This kind of implements will cause locality of data distribution. The locality is extremely notable in sparse sensor network. To solve this problem, we devise a Multi-Source Encoding technique which provides more equal probability of being received for all data, and then propose MFGCP protocol (Modified FGCP). Experiment results shows this technique works well in sparse network.(3) In the single-hop collecting tree of FGCP, neighbouring buffer nodes overhear many similar codewords, which will cause the data redundancy among buffer nodes. Hence, we simplify the single-hop collecting tree in local-confiugration model to a single node and then design a SFGCP protocol (Simplified FGCP). This protocol could maintain similar collecting efficiency to FGCP and achieve better fault-tolerance performance in high density network.(4) In Growth Codes network coding scheme, the best time to increasing coding degree (transition points) depends not only on the way the coder are implemented, but also on the topology of the network. This work is of great importance since it is the theoretical foundation of data gathering protocol based on Growth Codes. Combining the new local-configuration network model, we theoretically analyze the optimal transition points of Growth Codes in the local-configuration network. The optimal transition rate is formally derived to maximize the likelihood that the sink can immediately recover new data when it receives a codeword. This theoretical results work well in the protocols we proposed in this thesis.For sensor network in disaster scenarios, the work of studying how to improve data persistence and good collecting efficiency will have a great influence on WSN technique and will promote the development of WSN.