Based On The Theory Of Optimal Filter Estimate Non-stationary Network Link Packet Loss Rate

With the rapid development of network technology and the increase of network types, the topology and variation of parameters of network are becoming more and more complex. For a better network designing, controlling and management, we must understand the network internal features in real-time, have a high-speed perception on the of network parameters which is constantly changing, and the discipline inside these dynamic changes, and achieve the description of multi-level and multi-space of the whole network. The network tomographic technology does not require the coordination of network node. It measures the end-to end path level parameters by sending probing package, then evaluates the internal link network parameters by using statistical methods, in which way overcomes the disadvantages existed in the practical use of traditional network measurement.However, most current researches on network tomography based on the assumption that the state of the network parameters is changeless during the measurement period. Thus it is difficult to accurately describe the non-stationary processes of the network link parameters. The non-stationary network tomography can obtain the network state parameters and their changes more comprehensively and concisely, therefore to optimize the design of the network system and improve the management service level. This paper studies the non-stationary network tomography. By solving the key theoritical issues, we hope to strengthen the capability of troubleshooting the actual networks.To obtain the time-varying link-state accurate description in a small cost, it is necessary to desing a reasonable and reliable non-stationary network model to estimate the link parameters. Considering that the actual network link parameters have continuous and random changes over time, the time-varying link parameters estimation is more a waveform estimates problem than a parameter estimation problem. That means we can use the best filtering theory to solve the problem and provide a new concept for non-stationary network tomography research when studying the key technologies. This paper proposes two estimation methods based on the best filtering theory to estimate the link loss rate:the optimal estimation method based on spatiotemporal correlation, and the estimation method based on kalman filter theory. The optimal estimation method utilizes the state transition matrix to express the spatiotemporal correlation of the packet loss and to forecast. Then it uses the least squares theory to correct the prior estimate and the estimations meet the minimum mean square error. The kalman filter estimation method bases on kalman filter model. It trains the equation coefficients by prior data and recursive estimation link loss rate under a non-stationary network environment in two processes:time-update (a priori estimate) and measurement-update (a posteriori estimate). Both methods do not need to send a lot of stage back-to-back probe packets; they have a good approximation of the real link loss rate curve while reducing the influence by sending packets.We carry the NS2simulation out for the above two estimation methods of the time-varying link packet loss rate and the results of the simulation validate their validity and correctness of the link packet loss estimation under a non-stationary network environment. We also show the further comparison of these estimation methods presented in this paper.