Research And Implementation Of IPTV Soft Probe QoE Evaluation Module Based On Neural Network

In the era of high information development,the demand of interactive network television(IPTV)services is increasing,and high video quality is in urgent demand.The influence of video terminal performance parameters are not considered in the traditional quality of service(QoS)monitoring system.Therefore,the quality of experience(QoE)of video cannot be well reflected.How to better evaluate QoE of IPTV service has become a research hotspot.The QoE evaluation technology of IPTV live broadcast is studied and a video QoE evaluation module based on BP neural network model is established.The IPTV soft probe QoE evaluation module is developed by C++ language.The main work is as follows:(1)Data acquiring and preprocessing.The subjective score is recorded according to the subjective video quality evaluation methods and principles and the objective index is calculated.Data cleansing is firstly accomplished for 20,000 objective indexes.After eliminating the illegal data,the data beyond the normal range and the fixed value,15,000 pieces of data are remained and used in the experiment.Then,the correlation between each index and QoE is analyzed and the MDI_DF is removed because it has little correlation with QoE.At last,the normalization is performed for the data.(2)The BP neural network model and the RBF neural network model are established respectively to evaluate the QoE of the 15,000 live broadcast data.The assessment results are compared and the BP neural network model with smaller mean square error and the average absolute error is finally selected as the algorithm of the soft probe evaluation module.(3)A soft probe evaluation module is developed.The evaluation module is divided into acquisition module,processing module,interaction module and tool module according to the function,and each sub-module that make up the evaluation module is implemented by C++ language.(4)The module function is verified.Use Wireshark packet capture software to evaluate the accuracy of indicators collected,verified data collection,recording,reporting and whether the reporting cycle and other normal functions to achieve.The simulation environment was set up to test the average absolute error percentage of QoE and human subjective score returned by the soft probe evaluation module with BP model and original company parameter model.The mean absolute error of BP model and parameter model reached 3.52% and 10.9%,the BP model satisfies the functional requirements of the evaluation module.