Modeling RPS and Evaluating Video Repair With VQM

Abstract

Reference picture selection (RPS) is a promising repair technique in lossy networks for delay-sensitive video, whereby the video encoder uses one of several previous frames as a reference frame for predictive encoding. RPS can operate in two different modes: an optimistic policy that uses negative acknowledgements (NACKs) and a more conservative policy that relies upon positive acknowledgements (ACKs). This paper compares RPS NACK and RPS ACK under various network conditions and video contents using two analytical models. The two models characterize RPS NACK and RPS ACK by incorporating the impact of reference distance on video quality, prediction dependency among video frames and Group of Pictures (GOP) length. Given packet-loss rate, round-trip time and capacity constraints, the models predict average video quality for videos with RPS ACK and RPS NACK using the Video Quality Metric (VQM). Using these two models, a series of experiments are conducted to investigate RPS performance under various conditions. The insights derived from our models can help determine appropriate choices for RPS NACK and RPS ACK under various network conditions and video contents.