Network Coding Based Data Collection And Neural Network Based Dataaggregation In Sensor Networks

The service that wireless sensor network provides is to sense the physical phenomena and transmit the required data back to the information system. Sensors are often battery powered, therefore energy efficiency has long been considered as the primary disign index in information collecting application. This thesis focuses on two energy efficient information collecting techniques, namely data collection with mobile sinks(DCMS) and data aggregation(DA).DCMS consumes less energy than traditional approaches with static sink. Load balance is not needed. DCMS with netowrk coding(NC) provides additional advantage to DCMS with robustness to sensor failure and packet loss. Since the scalability of the state of art of LT network coding turns out to be deficient, we proposed dual-degree degree-limited LT codes(DDL-LT) and DDL-LT network coding based on packet centric network coding. Besides the excellent scalability, DDL-LT codes performs better also in engery efficiency, coding delay and transmission efficiency in sacrifice of collection integrity.DA saves a huge amount of energy by considerably reduces packets being transmited in the network with static sink. DA is attractive to applications that do not require raw data of each sensor. However, most of researches focused on DA protocol designment instead of the aggregation function to aggregate data packets. Therefore we proposed distributed neural network as the aggregation function which enables DA to recognize and classify complex events before which only simple functions such as max,average etc.can be performed. Distributed neural network has low complexity and lower delay than deta collection. It is also eavesdropping proof and robust to noise.