NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 1 Error Recovery

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 2 Problems with Internet Limited Bandwidth Varying Conditions Delay Jitter Packet Loss Delay Heterogeneity :

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 3 Network Encoder Sender Middlebox Receiver Decoder You Are Here

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 4 Overview Characteristics of the Internet General techniques Error recovery for audio Effect of loss on MPEG Error recovery for MPEG

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 5 Loss Characteristics of The Internet

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 6 Characteristics of Internet 60-70% of paths do not show any loss Those with loss have an average of 4.5 – 6% packet loss [Paxson97] End-to-end Internet packet dynamics

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 7 Packet Loss Pattern Number of Occurrences Burst Length

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 8 Characteristics of Internet Bursts of loss are typically short (2-3 consecutively loss packets) Long burst do occur Burst may occur periodically

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 9 Wireless Link Loss rate measured in my office 10 ~ 20% Up to 50% reported!

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 10 Gilbert Model GOODBAD

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 11 A Survey of Packet-Loss Recovery Techniques for Streaming Audio Perkins, Hodson and Hardman IEEE Network Magazine 1998

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 12 Retransmission X

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 13 Redundant Data

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) Error Concealment

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 15 Retransmission for audio

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 16 Pros/Cons of Retransmissions

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 17 Scalable Retransmission On packet loss T = random(0, RTT) wait for T multicast NACK On receiving NACK from others suppress own NACK

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 18 Retransmit when Group size is small Loss rate is low Large latency acceptable

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 19 Selective Retransmission Retransmission based on priority of packets Important/urgent packets are retransmitted first Packets are only retransmitted when there is enough time [Papadopoulos and Parulkar, NOSSDAV 1996]

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 20 Redundant Data for audio

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 21 Parity Forward Error Correction XOR

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 22 Parity Forward Error Correction

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 23 Parity FEC Ordering

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 24 Parity FEC Ordering

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 25 Parity FEC Ordering

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 26 Parity FEC Ordering

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 27 Parity FEC Ordering

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 28 Reed-Solomon Code RS(n,k) RS n k (popular: n = 223) (popular: k = 32) Parity symbols Data symbols

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 29 Reed-Solomon Code Data block (length n+k, e.g., 255) Can correct up to 16 corrupted symbols per block (32/2)

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 30 Media Specific FEC

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 31 Pros/Cons of Redundant Data

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 32 Interleaving

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 33 Pros/Cons of Interleaving

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 34 Error Concealment

Recreate Lost Information NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 35

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 36 Effect of Loss on Audio Speech Human ears can interpolate Loss up to length of phoneme can still be tolerable

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 37 Insertion-based Repair Splice Silence Substitution Noise Substitution Repetition

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 38 Other Repair Methods Interpolation Regeneration

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 39 Error Concealment More complex error concealment algorithms provide better performance (i.e., playback quality)

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 40 Colin’s Recommendations

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 41 Non-Interactive Apps Interleaving FEC Retransmission for unicast only

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 42 Interactive Applications Media Specific FEC

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 43 Error Concealment Repeat

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 44 Packet Loss Effects on MPEG Video Sent over the Public Internet Jill Boyce and Robert Gaglianello ACM Multimedia 1998

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 45 Measurement-based Study Need to understand the problem before proposing solution

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 46 Data Gathering Method From: NYC 13, Austin 21, London 18 To: Holmdel, NJ

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 47 Data Gathering Method Video: Two 5-mins MPEG 30 fps 384 kbps and 1 Mbps GOP: IBBPBBPBBPBBPBB QSIF 176x112 and SIF 352x240 One row per slice

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 48 Average Packet Loss

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 49 Frames Affected by Errors % Frames in Error Packet Loss Rate

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 50 Overview of Error Recovery for Video

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 51 Methods Retransmission Interleaving Error Concealment FEC Limiting Error Propagation Reference Frame Selection

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 52 Reference Frame Selection I P B B P X I

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 53 Methods Retransmission Interleaving Error Concealment FEC Limiting Error Propagation Reference Frame Selection Changing Temporal Pattern

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 54 Case Study: HD video conferencing Korean Air, Inha University (Seoul), University of Southern California, Pratt & Whitney

© PWICE, USC-Inha Two-way Experiments USC (Powell Hall of Engineering) July 2005

© PWICE, USC-Inha Two-way Experiments Inha University (Memorial Library) July 2005

© PWICE, KAL Network Infrastructure Korean Air network infrastructure between Incheon, Kimpo, and Pusan KAL network & external ISP (KT) 1 Gbps: ICN-GMP 50 Mbps: GMP-PUS December 2005

© PWICE, Network Measurements End-to-end packet loss rates of the network path between Kimpo (GMP) and Incheon (ICN) observed during November 15-16, 2005 Network is shared with other business traffic Loss rates are quite low November 2005

© PWICE, Hardware Preparation 3 computers purchased in Korea Computers sent to USC with KAL Software setup in the IMSC laboratories Linux installation RCS software installation Retransmission configuration End-to-end equipment test with local JVC cameras and displays Computers sent back to KAL HQ December 2005

© PWICE, Hardware Test at KAL Re-configuration for Sony HDR-HC1 cameras December 2005 Local area network tests

© PWICE, Experiments: ICN ◄▬► GMP Asymmetric environment: 88 Mb/s & 50 Mb/s Conclusion: ICN - GMP: sufficient BW for HD video Sufficient headroom for data applications December 2005 Bandwidth measurements Visual quality Required by RCS

© PWICE, Experiments: ICN ◄▬► PUS Asymmetric environment: 6 Mb/s & 1 Mb/s GMP - PUS: initially insufficient BW for HD video Investigated network routers with KAL IT help Conclusion: replaced router to improve throughput; to be re-tested December 2005 Bandwidth measurements Visual quality

© PWICE, Measurements: Packet Loss Rate Path: ICN - PUS Two packet sizes: 564 bytes and 940 bytes Conclusion: GMP - PUS loss rate very high December 2005 One-way packet loss rate Two-way packet loss rate

© PWICE, Packet Loss Error Recovery Selective retransmission protocol is used to recover from lost data Effectiveness of protocol is tested in lab environment Gilbert Model is used to induce losses into transmission Transmission delay is chosen to be 10 ms (expected latency in Korea)

© PWICE, Packet Loss Error Recovery Selective retransmission protocol 10% Loss Rate 5% Loss Rate 1% Loss Rate With Retransmissions Sept. - Dec (ICN-GMP Link)

Current Hardware Codecs NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 66 Hauppauge HD-PVR Cavium PureVu™ CNW31XX

HD-PVR H.264 Encoder NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 67

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 68

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 69 Error Control Techniques for Interactive Low Bitrate Video Transmission over The Internet Injong Rhee SIGCOMM ‘98

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 70 Basic Idea “Better Late Than Never!”, or Late packet is still useful

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 71 MPEG Frame Pattern IBBPBBP IPBBPBB

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 72 H.261 Error Propagation IPPPPPP XXXXXX IPPPPPP XXXX retransmission loss

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 73 H.261 Frame Pattern IPPPPPP X loss

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 74 IPPPPPP H.261 Frame Pattern X loss X X retransmission

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 75 PTDD Periodic Temporal Dependency Distance Large PTDD Higher Chance of Repair Longer Error Propagations Less Temporal Redundancy

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 76 QAL IPPP IPPP Base Layer Enhancement Layer

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 77 QAL IPPP IPPP Base Layer + FEC Enhancement Layer X

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 78 QAL + PTDD IPPP IPPP Base Layer Enhancement Layer

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 79 Error Propagation

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 80 Recovery from Error Propagation

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 81 Frame “quality” PSNR =

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 82 PSNR vs. Loss Rate Packet Loss Rate PSNR

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 83 Today’s Summary

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 84 How to recover packet loss Retransmission FEC Error Concealment

NUS.SOC.CS Roger Zimmermann (based in part on slides by Ooi Wei Tsang) 85 Limit the damage of error Interleaving Key frame selection Change reference frame