In this paper, we propose a deep learning method to identify which region has failures when clients make complaints. The format of the data we collect includes structured data, unstructured data, and time-series data. Experimental results show that our model can accurately distinguish which site is the root cause for a malfunction or a breakage in the process of the service. The results of this research can speed up maintenance personnel to repair faults.

Fault location estimation is an important part of network management for IPTV services. This work can be achieved by analysing the amount of customer complaints. This article proposes a simple model to analyse these data. The model uses fuzzy logic gates to process the classified data. The final result shows that this model can provide fault location prediction well.

In this paper, we propose two probabilistic modeling methods for IPTV network diagnosis, namely intolerable quality detection and device alarm analysis. The proposed methods can provide auxiliary information during troubleshooting processes, as well as inferring problematic devices automatically. Experimental results show that our models can distinguish problematic devices from normal ones and are more accurate than previous work under two evaluation criteria. Furthermore, we also demonstrate how we utilize these models in our actual network management system. It reveals that our methodologies are useful in finding out the service degradation proactively.

This paper proposes a model to derive an overall assessment index of the quality for IPTV network in order to make the network faults diagnosis more efficiently. This paper also provides a systematic approach to estimate the quality index based on the packet loss events. This model has been applied to the actual operational environment, not only to regularly monitor the customer perceived quality, but also to provide discriminating rules for locating abnormal sites.

This paper proposes an S-model to normalize the parameters measured from MPEG TS [1][2]. Using the additive normalized scores, an overall assessment score of the quality customer perceived is derived. This paper also provides a systematic approach to calculate the coefficients used in this model. This model has been applied to the actual operational environment, not only to detect the streaming quality, but also to provide quality trend analysis and degradation area judgment.

A two-layered model is proposed in this study to describe the error generating process of each component link in a concatenated digital transmission system. The first layer models the arrivals of physical impairments as a continuous time Poisson process. Each occurrence of a physical impairment yields a cluster of errors. The second layer describes the statistics of the intra-structure of each error cluster in discrete time. According to the nature of physical impairments, the length of an error cluster has a general probability distribution and is independent and identically distributed variable. Within an error cluster, the bit errors are generated according to a sequence of Bernoulli trials, This mixed-type description facilitates the evaluation of both global and local information regarding the error performance of the overall concatenated transmission system.>