Research On Dynamic Clustering Algorithm For Base Station Cooperation

Currently, Cooperative processing for base station (BS) cooperation system is considered as one of the hotspot issues in wireless communication. However, in order to utilize cooperative processing techniques in realistic cellular system, a crucial step is to divide the whole system into several small clusters. The existing clustering algorithms include static clustering and dynamic clustering algorithm, where the dynamic clustering algorithm has been proved performing better than static clustering, but it still faces with four main problems summarized following. First, clustering algorithms are lack of effective guidable clustering model. Second, clustering problem is regarded as a simply BS selection problem, thus the sequence and one-way selection manner in clustering algorithm would destroyed the best clustering structure and result in limited-capacity of system. Third, the precondition of whole channel state information (CSI) in most existing algorithms is impractical in realistic system. Last, several users in clustered system may suffer from severe inter-cluster interference and lead to bad users'fairness. This dissertation would focus the research on these problems of dynamic clustering algorithm.This paper first introduced a general indicator of Degree of Wiliness to Cooperate (DWC) and established a clustering model based on DWC. And then based on this model, two dynamic clustering algorithms were explored and one overlapping clustering modified scheme was proposed. The chief content of this article is as follows.1)A clustering model based on DWC was established. First, we define DWC between two BSs as the SINR gain achieved by cooperative processing of them. And then the clustering objective of maximizing the system sum rate was approximately to maximizing the sum of all the received SINR in BS. After thoroughly analyzing the relationship between received SINR and DWC, we safely concluded that the sum rate maximization could be translated into maximizing the sum of DWC between every two BSs in system.2)A whole CSI based benefit-tree clustering algorithm was proposed. The clustering problem is modeled as constructing benefit-trees of a connected graph with edge costs. Then a Benefit Degree (BD) introduced between two trees was defined by the normalized sums of DWC between every two BSs belong to different two trees. After that, during every iteration round, two trees with biggest BD were selected to combine into one tree. By this way, this algorithm simultaneously generates several clusters of dynamic size which could solve the limited-capacity problem caused by conventional clustering scheme. Simulation results showed that compared to the greedy clustering algorithm, the system spectrum efficiency in this algorithm increases about 0.4bits/Hz and the computational complexity is directly proportional to system size.3)A partial CSI based Affinity propagation clustering algorithm was proposed. In this algorithm, DWC was used to design the input matrix and a cooperation fact was introduced to adjust the number of clusters in clustering results. Assumming every BS has the partial CSI information of whole system, during every iteration rounds, the messages of Responsibility and Availability were propagated to accumulate the coordination incidences among BSs. After that, it constructed several clusters of dynamic size. Simulation results showed that this scheme performs almost the same as the whole CSI based benefit-tree clustering algorithm. Besides, in our algorithm, the convergence iteration rounds increase slowly as the system size enlarging. This exhibits high convergence speed and is well suited for implementation in realistic system.4)An overlapping modified clustering scheme based on joint ZF-THP algorithm was proposed. First, we defined an interference threshold by using the average value of inter-cluster interference in clustered system. Then, the BSs who bring about severe inter-cluster interference over the interference threshold were adjusted into an overlapped cluster. Second, according to the characteristics of overlapping cluster, a novel joint precoding algorithm was designed to reduce the complexity of overlapped BSs. Simulation results showed that this scheme greatly reduces the complexity of overlapped clustered BSs and enhances the capacity of partial users as well as its'fairness.