Interference Measurement And Modeling In Wireless Sensor Networks

Wireless sensor networks, the updated promising wireless networks consisting ofsensors spatially distributed to monitor and collect valuable information, are cumulativelyavailable for mission-critical applications. Due to the limited power supply ofsensors and the spatial deployment requirement in these applications, the transmissionperformance of a wireless sensor network is usually seriously influenced by the interferenceamong the complex transmission links along the multiple hops to the sink. To fulfillstringent requirements on transmission performance, it is crucial to understand andfurthermore mitigate the complex wireless interference among sensor nodes, in considerationof the importance of interference modeling for the transmission performance ofnumerous wireless sensor network protocols.Previous works have widely adopted simplistic interference models that fail tocapture the wireless realities, one of which is the probabilistic packet reception performance,as the key issue to the transmission performance. Recent empirical studiessuggest that the physical interference model, also referred to as the Packet ReceptionRatio (PRR) versus Signal-to-Interference and Noise Ratio (SINR) model or PRR-SINRmodel, offers significantly improved realism than other simplistic models, the widelyaccepted disc model included. However, existing approaches of physical interferencemodeling exclusively rely on the use of active measurement packets and synchronization,which impose prohibitively high overhead to bandwidth-limited wireless sensornetworks. Moreover, they are lack of analysis of the relationship between overhead andaccuracy, resulting in overhead imponderable.This thesis investigate several key problems in interference measurement and modelingin wireless sensor networks such as how to decrease overhead, how to increase theusability of interference measurement and how to balance the overhead with measurementaccuracy.The main research works and contributions are described as follows.This thesis firstly describes the background and empirical observations of interfer- ence in wireless sensor networks. As shown in our experimental results, the PRR-SINRmodel yields considerable spatial and temporal variations in reality, posing a majorchallenge for accurate measurement at run time.This thesis then proposes the passive interference measurement (PIM), a low costpassive measurement designed to tackle the complexity of accurate physical interferencecharacterization, based on the above results obtained. PIM exploits the spatiotemporaldiversity of data traffic for radio performance profiling with limited energy expenditurein terms of gathering a small amount of statistics about the network. PIM balancethe workload of sensor nodes based on capability by leaving modeling to the aggregatorwhich is more powerful. We evaluate the efficiency of PIM through extensive experimentson both a 13-node and a 40-node testbeds of TelosB motes. Our results show thatPIM is capable of achieving high accuracy of PRR-SINR modeling with significantlylower overhead compared with the commonly applied active measurement approach.One step further, this thesis extends to study how to balance the overhead andaccuracy while measuring interference model with assured accuracy at run time, by developingthe accuracy-aware interference measurement (AIM). AIM is based on a newdesigned regression-based PRR-SINR model with an accuracy characterized analyticallybased on statistics theory. AIM also adopts new clock calibration and in-networkaggregation techniques to reduce the overhead of interference measurement. Our extensiveexperiments on a 17-node testbed of TelosB motes show that AIM is able to achievethe assured accuracy of PRR-SINR modeling with significantly lower overhead than stateof the art approaches.