A Classification-Based Path Selection Scheme for Video Streaming over Multi-Hop Networks

A classification-based method is proposed to select an optimal path for video streaming over multi-hop mesh networks. Our main contribution is translating of path selection over multi-hop networks to a standard classification problem. The classification is based on minimizing average video packet distortion at the receiving nodes. We also consider long range dependence (LRD) characteristics of variable bit rate (VBR) video encoders by modeling the link as a fractional Brownian motion queue. In our method, a support vector machine is applied at each hop to optimally assign the next hop (partial path) for each video packet received at that hop. Wireless channel conditions (WCC) including packet loss probability of the channel and maximum achievable rate are used as class prototypes. Sample feature vectors include video content features (VCF) and video encoding parameters (VEP) extracted from video sequences. The classifiers are trained offline using a vast collection of video sequences and wireless channel conditions in order to yield optimal performance during real time path selection. Our method substantially reduces the complexity of conventional exhaustive optimization methods and results in high quality (low distortion). Simulations are conducted over an elementary multi-hop structure and cascades of such structure which not only proves the superiority of our method in terms of low complexity and high peak signal to noise ratio (PSNR) performance, but also provides important insights that can guide the design of network infrastructures and streaming protocols for video streaming.