160 likes | 297 Views
Streaming To Mobile Users In A Peer-to-Peer Network. Jeong h un Noh , Mina Makar, and Bernd Girod September 8, 2009 Image, Video and Multimedia Systems Group Information Systems Laboratory Stanford University. Transcoding adapts video to mobile devices. Heterogeneous Mobile Devices.
E N D
Streaming To Mobile Users In A Peer-to-Peer Network Jeonghun Noh, Mina Makar, and Bernd Girod September 8, 2009 Image, Video and Multimedia Systems Group Information Systems Laboratory Stanford University
Transcoding adapts video to mobile devices Heterogeneous Mobile Devices • Supporting heterogeneity • Screen size: 160x120, 320x240, 480x320, … • Channel characteristics : GPRS, EDGE, 3G, WiFi, … • Video capabilities: H.263, H.264, …
Streaming to Mobile Device server Peer-to-peer streaming Client-server streaming Transcoding ? Distributed transcoding mobile phone
Outline • Transcoding at multiple parents • IDT: Interleaved Distributed Transcoding • Encoding • Decoding under lossy condition • Advantage of multiple parents • Experimental results for distributed transcoding
Transcoding at Multiple Parents Parent 1 Parent 1 Parent 1 • Regular transcoding Parent 2 Parent 2 Parent 2 Parent 3 Parent 3 Parent 3 • Multiple Description Coding (MDC) • Interleaved Distributed Transcoding (IDT)
Distributed Transcoding: Encoding Original video 4 1 2 3 5 6 10 11 12 13 14 7 9 8 I1 B2 B3 P4 B5 B6 P7 B8 B9 P10 B11 B12 P13 B14 5 5 9 1 1 1 5 9 9 13 Parent 1 9 1 5 13 6 6 6 2 2 2 10 10 Parent 2 10 1 2 6 13 14 1 3 3 3 7 7 7 11 13 Parent 3 1 3 7 11 13 4 8 8 1 1 4 4 8 13 Parent 4 8 1 4 12 13 I1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11 P12 I1 P2
4 12 • parent loss packet loss Distributed Transcoding: Decoding Substream 1 5 5 9 1 1 1 5 9 9 9 1 5 Substream 2 6 6 2 2 10 6 2 10 10 2 6 Substream 3 Substream 4 1 3 3 3 3 7 7 7 7 11 11 1 4 8 8 8 1 4 4 8 Stream assembled at decoder
parent loss parent loss parent loss parent loss parent loss • Trade-off: Severity and frequency of video degradation Advantage of Multiple Parents 1 parent Effective frame rate Effective frame rate time time 4 parents 4 parents
Experimental Setup • Video: Foreman and Mthr & Dthr (H.264 main profile, CIF, 30 fps) • Distributed transcoding (H.264 baseline profile) • Downsampling (CIF to QCIF) • Number of parents: 1 ~ 4 peers • Lossy scenario • Peer lifetime ~ Exp(1/90) • Reconnection time ~ Exp(1/3) • Gilbert model-based channel 0.01 0.99 0.85 Good Bad 50% packet loss No packet loss 0.15 [Konrad , Zhao, Joseph, Ludwig 2003]
End-to-End Rate-Distortion Performance Foreman sequence
Distribution of Transcoding Load 60% load reduction • Encoder: x264-based software encoder • CPU: Pentium 4 with 2.8GHz clock speed
Demo Setup SPPM server SPPM peers SPPM mobile SPPM: Stanford Peer-to-Peer Multicast streaming system [Baccichet, Noh, Setton, Girod 2007]
Conclusions • Video transcoding to meet mobile device heterogeneity • Distributed transcoding in a P2P network • Graceful video degradation against peer/packet loss • Distribution of transcoding load • Extensions • Rate control based on wireless channel condition • Application to fixed-line peers with limited downlink speed
Thank you! Jeong-hun Noh jhnoh@stanford.edu