Research On Data Compression And Resource Allocation In IoT

With the development of embedding terminal and next generation communication technology,the IoT?Internet of Things?,as the support of ubiquitous computing,has draw much attention of research filed on smart home,smart city and intelligent trans-portation.In the near future,IoT will achieve the information sharing among human,machine and things,which promotes deeper fusion between physical world and informa-tion world.Nowadays,most of sensor nodes in IoT are supported by batteries,due to their mobility or the geometrically location.With the growing number of sensor nodes in IoT and richer sensing information of single node,the information in the whole network is sharply increasing.The conflict between limited physical resources and rapid growth of data amount,is an urgent problem to be solved in IoT system design.On the other hand,with the widespread application of the IoT in various scenarios,the service requirements of users are becoming more and more complex and diverse,so the development of the future IoT requires the network to carry more personalized and intelligent services.How-ever,multiple IoTs are deployed in the same area at present to provide different services,leading to repeated construction and low utilization of network.Based on above discussion,this dissertation investigates the following issues:Con-sidering the sensing data characteristics of redundancy and massive amounts,this disser-tation investigates data compression and aggregation,aiming to reduce the data amount of in-network,and prolong the lifetime of IoT.In the meanwhile,to improve the utilization of physical resources in IoT,this dissertation also studies the virtualization of IoT.The main contributions of this dissertation are summarized as follows.?1?Aiming at eliminating the redundancy of sensing data caused by overlapping a-mong sensor nodes,this dissertation adopts distributed source coding?DSC?method to aggregate redundancy data.However,it is challenging to allocate minimum encoding rate for each node,because Spelian-Wolf constraints increase exponentially with respect to the number of sensor nodes.Thus,this dissertation jointly considers encoding rate allocation problem and flow scheduling problem,and propose a cross-layer optimization framework.By analyzing the convexity of constraints,the optimization problem is proved to be con-vex.To solve the problem of constraints exponentially increasing,original constraints are relaxed and relaxed constraints are proved.Finally,dual decomposition is employed to solve original problem separately.Compared with fixed encoding rate scheme,the opti-mal rate allocation scheme can significantly reduce data redundancy and network traffic.?2?In multimedia IoT,traditional video encoding schemes can not be executed on sensor nodes due to limited resources.This dissertation studies the compressed sensing based video encoding/decoding framework and the construction of sparse basis.The pro-posed framework contains two layers encoding/decoding schemes and transfers computa-tion burden to sink node,thus it is suitable for IoT and significantly reduces the amount of sampling data.At the first layer,to successfully decode compressed data,the first group sparse basis is constructed by exploiting the correlation between frames for each sensor node.At the second layer,to obtain the second group sparse basis,dictionary learning is employed by considering the spatial correlation between neighbor nodes.The online learning method decides whether to re-training sparse basis according to recovered qual-ity.The sink node employs two groups of sparse bases to perform l₁norm minimization problem twice and recover original video for each node.Due to the high efficiency of constructed sparse bases,the proposed framework in this dissertation provides high com-pression ratio and better sparse representation.?3?Nowadays,the existing IoT is merely used as a method for information collec-tion and transmission,its services and network functions coexist in the same system.To improve the utilization of IoT network and satisfy personalized service requirements,this dissertation proposes the virtualization of IoT network and studies the virtual IoT map-ping problem.Firstly,the virtualization architecture is presented by adopting light-weight virtualization technology.In this architecture,services and functions are separated,and the Infrastructure is regarded as a service pool rather than a data transmission channel.To solve the virtual IoT mapping problem,this dissertation proposes real-time process-ing models and batch processing model.The goal of two models is to improve resource utilization and maximize IoT service providers'revenue.In two models,multiple types of resources are taken into account.In addition,to enable multiple virtual nodes to co-exist on the same physical node,traditional flow conservation law is improved.Several efficient mapping algorithms are designed to solve these two NP-Hard problems.These heuristic algorithms can find sub-optimum solutions in polynomial time.Finally,results show that proposed models outperform the existing methods in terms of acceptance ratio,revenue,and resource utilization.