Research On Recursive In-Network Processing In Wireless Sensor Networks

Due to the low cost and ease of deployment,wireless sensor networks(WSNs)are used in many applications.In recent years,we are seeing that WSNs extend the applications in to some domain-specific area,such as structural health monitoring(SHM),volcano tomography and fault diagnosis of machines.In these applications,all the sensor nodes cooperatively monitor one common object(e.g.,the structure or the volcano),each sensor node needs to sample sufficient raw data and deliveries the data to the sink for once centralized computation to exact some global features.In the centralized data processing,the transmission cost is quite large and is not conducive to the life of the WSNs.To reduce the transmission cost,in-network processing is an effective way.However,different from most existing in-network processing works that pre-assume some common computation/aggregation functions,in-network processing of a given centralized algorithm requires to exactly partition the algorithm first and then appropriately assign the partitioned computations into WSNs.We call it as lossless in-network processing,which has not gotten much studied.Lossless in-network processing raises two questions: 1)what pattern should a centralized algorithm be partitioned into so that the partitioned computations can be flexibly assigned into a WSN with arbitrary topology? and 2)for each partition pattern how to design efficient routing for the resource-limited sensor nodes?In this paper,we focus on two kinds of centralized computing(LSE and SVD),which are commonly used in wireless sensor networks,to analyze the recursive computing mode in recursive in-network processing.According to the characteristic of the calculation,recursive lossless in-network processing is divided into constant cost of recursive in-network processing and incremental cost of recursive in-network processing.In view of the characteristics of these two kinds of recursive in-network processing,the paper first studies the optimal in-network processing problem and gives the corresponding in-network processing algorithms.Extensive simulations are conducted,the results show the good performance of the proposed algorithms,they can significantly reduce the network transmission cost under different network conditions.The work in this paper,either the computation partition patterns or the routing ideas,is not limited to lossless in-network processing in WSNs,but can also serve as a guideline for distributed computing of big data where the “raw data” spreads in a large network.