The Research And Application Of DLMS Algorithm In An Unsecure Environment

Distributed adaptation over networks has emerged as an attractive and challenging research area with the advent of multiagent networks. Distributed adaptive networks are well-suited to perform decentralized information processing and optimization tasks and to model various types of self-organized and complex behavior encountered in nature. This thesis mainly concentrates on the problem of distributed estimation. Distributed adaptive networks consist of a collection of nodes with processing and learning abilities. The nodes are linked together through different connection topologies, and they cooperate with each other through local iterations to estimate the same unknown parameters. The key technology of distributed estimation is distributed estimation algorithm, which is composed of adaptive estimation algorithm and distributed strategies.Firstly, this thesis briefly describes the problems and challenges in distributed estimation, and gives an overview of distributed estimation algorithm of its application and development status at home and board. Then a basic adaptive estimation algorithm at single node is introduced, that is Least Mean Square (LMS) algorithm. On the basis of LMS algorithm at single node, the diffusion strategy, which is one of the important distributed strategies, is described in detail, and diffusion LMS (DLMS) algorithm is given. Compare LMS algorithm of non-cooperation with DLMS algorithm of cooperation by simulation. The results show that the DLMS algorithm has much better performance than LMS algorithm, which means that the diffusion of information through network brings improvement on performance.Secondly, the actual network is not absolutely safe, malicious node will tamper its own values and damage the estimation performance of entire network through diffusion. Moreover, the attack is difficult to predict because of the diversification of the attack model. Therefore, the thesis analyzes a variety of attack models and summarized two main kinds of attacks, which are Same-direction attack and Gaussian attack respectively. In order to reduce the impact of Same-direction attack, the diffusion LMS algorithm based on the median filtering mechanism (MF-DLMS) is proposed. The algorithm sorts the values received from its neighbors and judge the values that at both ends of them as malicious, and then the sorted values are processed by median filtering. But MF-DLMS algorithm is not effective against the Gaussian attack, and MF-DLMS algorithm itself is easily influenced by network topology and the number of malicious nodes, which leading to limited application. To find more robust algorithm, a new method of outlier data detection and evaluation and warning mechanism is presented before fusion step, and the method is applied into DLMS algorithm as an improvement to obtain DLMS algorithm based on the Outlier Data Detection method (ODD-DLMS). The algorithm enables every regular node accurately identifies malicious values and refuses them into fusion step during each iteration, of which the evaluation and warning mechanism can update safety indicators and give an alarm timely. The simulation results shows that the two algorithms mention above can greatly improve the performance of the entire network under attacks and effectively reduce the impacts of malicious nodes when compare with DLMS algorithm. However, ODD-DLMS algorithm has better performance and robustness that MF-DLMS algorithm.Finally, using the distributed estimation to solve problems in different fields is a focus of this study. The thesis combines the distributed estimation with spectrum sensing, using LMS algorithm to estimate the amplitude of transmitted signal in real time, and the estimation is employed as the test statistic to detect the existence of primary user. Both the theoretical and simulation results show that the proposed method has better performance for weak signal detection, which also greatly outperforms the classical energy detection method and effectively overcome the influences of noise uncertainty by estimating the noise variance.