Network Access And Spectrum Sensing In Heterogeneous Networks

At present,the wireless mobile devices are skyrocketing in both number and variety.The massive number of mobile devices and the ever increasing data services requirement have imposed a high demand with respect to delay and bandwidth of the wireless access network which cannot be fulfilled by the traditional single access network scheme.Therefore,the present access network environment has developed into a heterogeneous network composed of the macro base stations(MBSs)satisfying the need for wide-range network access and the small base stations(SBSs)and WiFi access points satisfying the need for small-range high-density network work access.Due to the pervasiveness of the heterogeneous networks,the access network selection and user admission control problems in the heterogeneous networks have attracted great research interest.At the same time,to increase the efficiency in spectrum reuse between the MBSs and the SBSs and to reduce the management complexity in the spectrum management center,some SBSs in the heterogeneous network adopt the cognitive radio(CR)scheme to dynamically access the spectrum used by the MBSs.As the enabling technology of such cognitive dynamic spectrum access scheme,the technology of spectrum sensing has attracted great attention.In the light of their importance,the main research topics in this paper are focused on the access network selection problem in the heterogeneous networks and the spectrum sensing problem in the cognitive heterogeneous networks.With respect to the access network selection problem in heterogeneous networks,the contributions are:1)In heterogeneous networks,we consider the access network selection problem from the perspective of the mobile devices.By resorting to the Markov decision process(MDP)to model the dynamics of the user number in each network,an intelligent dynamic network access strategy is obtained that maximizes the long term reward.At last,through sufficient simulations,the effectiveness of the MDP based intelligent access network selection algorithm is proven.2)In the heterogeneous networks composed of vehicular networks and cellular networks,the throughput of the vehicular networks is usually affected by the placement of the vehicular network base stations and the level of congestion on certain road sections.Therefore,in the heterogeneous networks composed of the cellular networks and the vehicular networks,from the perspective of a vehicle,the access network selection problem is highly related to the vehicle route planning problem.In this thesis,we jointly consider the access network selection and route planning problem and resort to the semi-Markov decision process(SMDP)to model the movement,turning and network access process of a target vehicle.The proposed algorithm is able to maximize the total network throughput obtained by the target vehicle in movement while guaranteeing that the vehicle is able to reach the destination on time and the network access expenditure is below certain threshold.At last,the proposed joint network access selection and route planning algorithm is proven by sufficient simulations.On the spectrum sensing problem in heterogeneous networks,the contributions are:3)In heterogeneous networks,to increase the efficiency of spectrum reuse and to lower the complexity of spectrum management,some SBSs in the heterogeneous networks adopt the scheme of cognitive dynamic spectrum access.In this scheme,SBSs need to resort to the technology of spectrum sensing to periodically sense the spectrum to prevent interference to the primary users(PUs)(MBSs and other non-cognitive SBSs).By resorting the currently widely researched deep learning technology,we propose a variational auto-encoder-Gaussian mixture model(VAE-GMM)based spectrum sensing algorithm.This algorithm is based on unsupervised deep learning.While possessing higher detection probability compared with the traditional spectrum sensing algorithms,the proposed algorithm require less labeled training data compared with the supervised learning based spectrum sensing algorithms.At last,through sufficient simulations,the effectiveness of the proposed algorithm is proven.4)Currently,most of the spectrum sensing algorithms merely use the sensing samples of the current sensing period to detect the existence of the PUs.In real situation,the transmission and idle states of the PUs have continuity.If we can learn the pattern of PUs' sojourn time at the transmission state and the idle state and combine such pattern with the sensing samples from the present sensing period,the detection accuracy can be promoted.In this thesis,we propose a convolutional neural network(CNN)based algorithm for spectrum sensing.This CNN based algorithm can simultaneously exploit the sensing samples from the present sensing period and the past sensing periods and learn the sojourn time pattern of the PUs' transmission/idle state to promote the detection accuracy.At last through sufficient simulations,the effectiveness of the proposed algorithm is proven.