Views

Download Presentation
## Rate Mutilation Improvement for Cross section based P2P Video Gushing

Download Now

**Rate Distortion Optimization for Mesh-based P2P Video**Streaming Tareq Hossain, Yi Cui, Yuan Xue Vanderbilt Advanced Network and Systems Group Vanderbilt University, USA Presenter: Dr. Sachin Agarwal Deutsche Telekom Laboratories**Outline**• Motivation • Video Broadcast • Can P2P Help? • Rate Distortion for P2P Mesh • Rate Optimization • Simulation • Results • Conclusion**Motivation**• Video Broadcast • Increasing popularity due to wide use of internet • Can P2P Help? • Cost effective resource utilization • CPU cycles • Storage space • Uplink bandwidth • Instant deployability • Almost ubiquitous network coverage in the absence of CDN services and IP multicast**Rate Distortion for P2P Mesh**• Mesh based P2P can fully utilize the network resources of its peers compared to a tree based network • We use distributed algorithm – each peer adjusts its own streaming rate to reach the global optimum by satisfying: • Capacity constraint • Relay constraint • Double Pricing Solution • Simultaneous incorporation of capacity constraint and relay constraint significantly reduces the aggregate rate distortion • Single Pricing Solution • Relay constraint is applied after rate distortion algorithm converges • We present rate-distortion optimization for P2P mesh network • Double pricing solution performs better than single pricing solution**Outline**• Motivation • Rate Optimization • Performance Evaluation • Problem Formulation • Distributed Algorithm • Simulation • Results • Conclusion**Performance Evaluation**• Video quality is measured as the Mean-Square-Error (MSE) averaged over all frames • PSNR is used to quantify video quality, defined by • D represents the overall Mean-Square-Error (MSE) averaged over all frames of an encoded video sequence • The distortion D as a function of streaming rate xf is given by • The variables (θ, x0 and D0 ) depend on encoded video sequence as well as on the percentage of intra coded macroblocks.**Problem Formulation**Rate optimization is a convex function of the allocated rate Here f represents a flow between two peers, x is the rate vector and c is the capacity vector A is an L x F (link, flow) matrix of links and flows such that Alf = 1 if flow f goes through link l and 0 otherwise B is an F x F sparse matrix, where ((hk – 1)H + hi)th row is active only if there is a flow from peer hk to peer hi. Formally,**Distributed Algorithm**• Each receiving peer ( ) calculates the rates of its incoming flows in a mesh • Network price: • Net relay price: • Source Rate update for each peer: • Rate is updated based on the minimum of network and net relay price available among the all incoming flows • Rate update for incoming flows: • Rate update for incoming flows with minimum network and net relay price: link price relay price**Outline**• Motivation • Rate Optimization • Simulation • Configuration • Input Data • Multicast Tree Construction • Results • Conclusion**Configuration**• To determine the actual allocated rate, we choose the highest quantized rate that is immediately less than the rate achieved by our solution • The ITU-T test sequences used are: foreman, akiyo, hall, mother-daughter • The server has a fixed rate of 2Mbps • The maximum number of peers ~160 • The uplink bandwidth of each peer is randomly assigned between 0.6Mbps and 2Mbps**Input data**• The PSNR-Rate video input data (a) and Number of peers-Time data (b): Rate (Kbps)**Multicast Mesh Construction**• Peers join the streaming network one-by-one • Joining peer uses the spare capacity of existing peers to determine a suitable parent. The spare coefficient is defined as • Here xf(h)is the incoming flow rate of the peer h • Implementation • At the end of each rate update cycle, peers send their spare coefficient value to parents • The ID of the best suitable parent propagates to the server**Outline**Motivation Rate Optimization Simulation Results Conclusion**Results**• The average PSNR gain over all the videos for the double pricing solution is 1.86 dB (PSNR is 0 when all peers leave ~720s)**Results**• The average gain for the double pricing solution represented in terms of rate**Conclusion**We present an optimal rate allocation solution for P2P mesh network We use non-linear optimization framework Minimize aggregate distortion Maximize the overall PSNR among all peers in a P2P mesh Simultaneously apply peer relaying constraint along with capacity constraint Double pricing solution consistently performs better than single pricing solution**Thank You**VANETS (Vanderbilt Advanced Network and Systems) Group http://vanets.vuse.vanderbilt.edu QUESTIONS?