Video quality estimation for internet streaming

Abstract

As broadband access connectivity becomes more prevalent, more users are able to stream video over the Internet. However, the besteffort service model and shared infrastructure of IP networks means network impairments (such as delays, jitter, congestion, and loss) have the potential to hamper viewing experiences. Network service providers would like an online performance monitoring tool in order to create a real-time understanding of video streaming applications. This can aid in monitoring compliance of service-level agreements (SLAs) between Internet Service Providers (ISPs), hosting centers, and content providers; alert operators to potential performance problems; and help in root-cause analysis and debugging. In this paper, we consider the problem of predicting applicationlevel video quality by monitoring video packets within the network. One particular focus is to be able to identify when the encoded bitstream has been modified (by either server or network) enough to impair the displayed quality. We denote all such cases as “bitstream impairments”. Another aspect that we consider is the “original encoded” quality of the video when there are no network induced impairments. We focus on the widely used proprietary Windows media streaming format for this study. Our work is unique in studying the video quality of Internet streaming content by extracting quality related information from different levels in the protocol stack, and by analyzing the quality of streaming content observed at a broadband access network. A number of earlier experimental studies have addressed the network level quality of service (delay, loss, jitter etc.) experienced by multimedia streams transmitted across networks [1, 2]. A number of earlier works explore computing video quality using information gathered from video pixels [3, 4]. But these works either rely on access to the original video and/or require expensive decoding of the received stream, requirements which make them unsuited to network environments where the original video is not available and computation scalability is a must. The contributions of this paper are twofold. First, we develop techniques to extract information related to video quality from video packets observed at an in-network monitoring point. Our approach involves processing different levels of information in the packet stream, encompassing the network, transport and application payloads. Second, we apply our techniques and framework to perform a measurement study of the performance experienced by streaming video sessions observed at a broadband access network.