Rate-distortion optimized client side rate control for adaptive media streaming

Media streaming over unreliable networks such as the Internet is growing in popularity, but presents unique challenges when trying to get the user experience to be on par with classical mediums such as cable television. These networks have variable network conditions which not only vary between a set of points in the network, but also change over time. In this paper, we present a rate-distortion (R-D) optimized algorithm for adapting the bitrate of streaming media from a chunked encoding, where each chunk is available at multiple bitrates (or quality levels) or is scalable coded. The optimization takes into account the desired startup latency, the desired client buffer size, the current client buffer size, and the estimated network bandwidth. The problem is formulated as a distortion minimization problem subject to multiple rate constraints. For video content, the solution gives a gain of up to 3-4 dB in PSNR in difficult portions of the video when compared to commonly used adaptation techniques and can achieve an arbitrarily small desired startup latency.