Non - Referenced Objective Streaming Video Quality Evaluation in Cellular Networks of 3 rd and 4 th Generation

This paper presents a novel methodology for quality evaluation of streaming video data services. The methodology requires no reference and it predicts subjective experience of the video quality. The predictions are based on a nonlinear mapping between objective technical metrics collected by the user equipment and subjective scores given by human evaluators. The objective metrics may be taken from various levels of the protocol stack. In the current implementation, the nonlinear mapping is accomplished through a neural network. The performance of the methodology is tested using data UDP streaming video services over LTE (4G) and HSPA (3G). The agreement between the predictions and subjective quality evaluation scores is excellent.