A refined random gossiping in wireless sensor networks

This project studies random gossiping in wireless sensor networks. Sensors in wireless sensor networks generate measurement data and communicate it with each other such that the desired aggregation involving the measurements of all sensors is achievable. Random gossiping is a decentralized communication paradigm for wireless sensor networks. When random gossiping is applied in the network, a sensor wakes up in a random manner and exchanges messages with its neighbor sensors. Critical problems of using random gossiping for the aggregation are the unmeasurable convergence, the bias of the aggregation, the convergence speed measured by the number of communications in the network, and the support of multiple applications, potentially.

In this project, a sensor is modeled as the integration of sensing, transmission, computation, and storage. The enabling of the communications among sensors requires a cross-layer design to meet the efficiency and low power consumption requirements. To facilitate the cross-layer design, the concept of indicating-header is proposed. The indicating-header serves as the shared information containing the aggregation status of the measurement of a particular sensor in the message of another sensor. Therefore, a straightforward metric of the convergence is given. To overcome the bias of the aggregation, the storage capacity at each sensor is explored with the help of the indicating-headers. A sensor can use the previously received messages stored in the memory to cancel the bias in the aggregation. An improvement of the bias cancellation is shown to be achievable by selecting a subset of the neighbor sensors of a sensor to perform the communications.

To improve the convergence speed, the indicating-headers are communicated in the random gossiping before the transmission of the messages containing the aggregation data. The information in the indicating-header enables the sensor to decide on the necessity of message communications. When it communicates with multiple neighbor sensors, the sensor uses the indicating-header to select only a subset of neighbor sensors for communications. A reduction in the number of communications is achieved while the efficiency of the aggregation is maintained. A further method to improve the convergence speed is proposed to coordinate sensors that are multiple hops away from the sensor in the random gossiping. When the constraint to the network topology is made that the sensor and its neighbor sensors remain static, the random gossiping can be improved by reducing the indicating-header communications. Moreover, when sensors are at topological bottle-neck positions of the network, these sensors may defer their message communications waiting for the groups of sensors that they are ”bridging”to have aggregation locally achieved. Such transmission deferment applied to these sensors reduces further the number of communications in the network.

When multiple applications are running in the network, a difference in terms of the number of communications to perform shall be made between the sensors that are involved in a specific application and those that are not. A refinement of the random gossiping is proposed by considering six different scenarios with respect to the involvement of a sensor and its neighbor sensors in an application. The indicating-header is used to enable sensors to distinguish between the six different scenarios. The sensors which are not involved in the application require fewer communications after the refinement while more communications are performed by the sensors that are involved in the application. Meanwhile, the total number of communications in the network is maintained.