Dude Where is My Packet 1 Overview Characteristics

Dude, Where is My Packet? 1

Overview Characteristics of the Internet l General techniques l Error recovery for audio l Effect of loss on MPEG l Error recovery for MPEG l NUS. SOC. CS 5248 OOI WEI TSANG 2

Loss Characteristics of The Internet 3

Characteristics of Internet l 60 -70% of paths do not show any loss l Those with loss have an average of 4. 5 – 6% packet loss [Paxson 97] End-to-end Internet packet dynamics NUS. SOC. CS 5248 OOI WEI TSANG 4

Packet Loss Pattern 1000 30 NUS. SOC. CS 5248 OOI WEI TSANG 5

Characteristic of Internet l Bursts of loss are typically short (2 -3 consecutively loss packets) l Long burst do occur l Burst may occur periodically NUS. SOC. CS 5248 OOI WEI TSANG 6

Wireless Link l Loss rate measured in my office 10 ~ 20% l up to 50% reported! NUS. SOC. CS 5248 OOI WEI TSANG 7

General Error Recovery Techniques Perkins, Hodson and Hardman IEEE Network Magazine 1998 8

Retransmission 1 2 NUS. SOC. CS 5248 OOI WEI TSANG 3 4 3 9

Redundant Data 1 2 2 3 NUS. SOC. CS 5248 OOI WEI TSANG 3 3 4 4 5 10

Error Concealment 1 2 NUS. SOC. CS 5248 OOI WEI TSANG 3 4 4 11

Retransmission for audio 12

Retransmission is Bad Need scalable feedback l NACK consumes bandwidth l May need to retransmit frequently l Retransmission maybe useless l NUS. SOC. CS 5248 OOI WEI TSANG 13

Retransmission is Good Only retransmit when needed l Exact recovery l NUS. SOC. CS 5248 OOI WEI TSANG 14

Scalable Retransmission l On packet loss T = random(0, RTT) l wait for T l multicast NACK l l On receiving NACK from others l suppress own NACK NUS. SOC. CS 5248 OOI WEI TSANG 15

Retransmit when group size is small l loss rate is low l large latency acceptable l NUS. SOC. CS 5248 OOI WEI TSANG 16

Redundant Data for audio 17

Parity FEC 1011001 1000010 XOR 0010010 0001001 NUS. SOC. CS 5248 OOI WEI TSANG 18

Parity FEC 1011001 100001001 0010010 NUS. SOC. CS 5248 OOI WEI TSANG 19

Parity FEC Ordering NUS. SOC. CS 5248 OOI WEI TSANG 20

Parity FEC Ordering NUS. SOC. CS 5248 OOI WEI TSANG 21

Parity FEC Ordering NUS. SOC. CS 5248 OOI WEI TSANG 22

Parity FEC Ordering NUS. SOC. CS 5248 OOI WEI TSANG 23

Parity FEC Ordering NUS. SOC. CS 5248 OOI WEI TSANG 24

Reed-Solomon Code l RS(n, k) RS NUS. SOC. CS 5248 OOI WEI TSANG 25

Reed-Solomon Code NUS. SOC. CS 5248 OOI WEI TSANG 26

Media Specific FEC 1 2 2 3 NUS. SOC. CS 5248 OOI WEI TSANG 3 3 4 4 5 27

Interleaving NUS. SOC. CS 5248 OOI WEI TSANG 28

Pro and Cons l Media Independent FEC Overhead for redundant data l Exact recovery l Could be computationally costly l l Media Specific FEC l Approximate recovery only NUS. SOC. CS 5248 OOI WEI TSANG 29

Error Concealment 30

Effect of Loss on Audio l Speech Human ears can interpolate l Loss up to length of phoneme can still be tolerable l NUS. SOC. CS 5248 OOI WEI TSANG 31

Insertion-based Repair Splice l Silence Substitution l Noise Substitution l Repetition l NUS. SOC. CS 5248 OOI WEI TSANG 32

Other Repair Methods Interpolation l Regeneration l NUS. SOC. CS 5248 OOI WEI TSANG 33

Pro and Cons Could be computationally costly l Approximate recovery only l Does not work well for long burst of packet loss l NUS. SOC. CS 5248 OOI WEI TSANG 34

Colin’s Recommendations 35

Non-Interactive Apps Interleaving l FEC l Retransmission for unicast only l NUS. SOC. CS 5248 OOI WEI TSANG 36

Interactive Applications l Media Specific FEC NUS. SOC. CS 5248 OOI WEI TSANG 37

Error Concealment l Repeat NUS. SOC. CS 5248 OOI WEI TSANG 38

Effects of Packet Loss on MPEG Jill Boyce and Robert Gaglianello ACM Multimedia 1998 39

Measurement-based Study l Need to understand the problem before proposing solution NUS. SOC. CS 5248 OOI WEI TSANG 40

Data Gathering Method From: NYC 13 , Austin 21, London 18 l To: Holmdel, NJ l NUS. SOC. CS 5248 OOI WEI TSANG 41

Data Gathering Method l Video: Two 5 -mins MPEG l 30 fps l 384 kbps and 1 Mbps l IBBPBBPBB l QSIF 176 x 112 and SIF 352 x 240 l One row per slice l NUS. SOC. CS 5248 OOI WEI TSANG 42

Average Packet Loss NUS. SOC. CS 5248 OOI WEI TSANG 43

Packet Loss vs Time NUS. SOC. CS 5248 OOI WEI TSANG 44

How to Fill Packets B 1 frame 1 packet l P 1 slice 1 packet l I 1 slice 1 packet l NUS. SOC. CS 5248 OOI WEI TSANG 45

Packet Size Distributions NUS. SOC. CS 5248 OOI WEI TSANG 46

Packet Size vs Loss Rate NUS. SOC. CS 5248 OOI WEI TSANG 47

Lesson 1 slice per packet even if larger than MTU l If smaller packets means higher loss rate: fill packet until MTU l Else one slice per MTU l l Alt: Change size of slice NUS. SOC. CS 5248 OOI WEI TSANG 48

Frame Effected by Errors NUS. SOC. CS 5248 OOI WEI TSANG 49

Correlation of Packet Loss NUS. SOC. CS 5248 OOI WEI TSANG 50

Overview of Error Recovery for Video 51

Methods Retransmission l Error Concealment l FEC l Limiting Error Propagation l l Reference Frame Selection NUS. SOC. CS 5248 OOI WEI TSANG 52

Reference Frame Selection I P B B IP X NUS. SOC. CS 5248 OOI WEI TSANG 53

Methods Retransmission l Error Concealment l FEC l Limiting Error Propagation l Reference Frame Selection l Changing Temporal Pattern l NUS. SOC. CS 5248 OOI WEI TSANG 54

Changing Temporal Patterns Injong Rhee SIGCOMM ‘ 98 55

Basic Idea “Better Late Than Never!”, or l Late packet is still useful l NUS. SOC. CS 5248 OOI WEI TSANG 56

MPEG Frame Pattern I B B P I P B B NUS. SOC. CS 5248 OOI WEI TSANG 57

H. 261 Error Propagation I P X P X P X P P loss I P X NUS. SOC. CS 5248 OOI WEI TSANG retransmission 58

H. 261 Frame Pattern I P X P P P loss NUS. SOC. CS 5248 OOI WEI TSANG 59

H. 261 Frame Pattern I X P X P P retransmission loss NUS. SOC. CS 5248 OOI WEI TSANG 60

PTDD l Periodic Temporal Dependency Distance l Large PTDD Later Deadlines l Longer Error Propagations l Less Temporal Dependencies l NUS. SOC. CS 5248 OOI WEI TSANG 61

QAL I P P P Enhancement Layer I P P P Base Layer NUS. SOC. CS 5248 OOI WEI TSANG 62

QAL I P X P P Enhancement Layer I P P P Base Layer + FEC NUS. SOC. CS 5248 OOI WEI TSANG 63

QAL + PTDD I P P P Enhancement Layer I P P P Base Layer NUS. SOC. CS 5248 OOI WEI TSANG 64

SR-RTP Feamster and Balakrishnan PV’ 02 65

Overview An analytical model for packet loss l SR-RTP l Post Processing l NUS. SOC. CS 5248 OOI WEI TSANG 66

Analytic Model p : prob. packet loss l f : observed frame rate l : frame drop rate l Si : mean num. of packets in frames of type i l P(i) : prob. frame type is i l P(F) : prob. frame is useless l NUS. SOC. CS 5248 OOI WEI TSANG 67

p f prob. packet loss observed frame rate frame drop rate Si mean num. of packets in frames of type i P(i) : prob. frame type is i P(F) : prob. frame is useless NUS. SOC. CS 5248 OOI WEI TSANG : : 68

Is it accurate? NUS. SOC. CS 5248 OOI WEI TSANG 69

SR-RTP l Extensions to RTP for selective reliability NUS. SOC. CS 5248 OOI WEI TSANG 70

SR-RTP 0 Length ADU Sequence Number ADU Length ADU Offset Priority NUS. SOC. CS 5248 OOI WEI TSANG Layer Number 71

Feamster & Balakrishnan’s ADU l 1 ADU = 1 Frame NUS. SOC. CS 5248 OOI WEI TSANG 72

Loss Detection LEN: 50 ADU: 0 ADU Size: 100 ADU Off: 0 LEN: 20 ADU: 0 ADU Size: 100 ADU Off: 80 NUS. SOC. CS 5248 OOI WEI TSANG LEN: 50 ADU: 1 ADU Size: 150 ADU Off: 40 LEN: 50 ADU: 3 ADU Size: 50 ADU Off: 0 73

Retransmission Request Based on priority l Example: l I-Frame: highest priority l P-Frame: higher priority if closer to prev I l B-Frame: no retransmition l NUS. SOC. CS 5248 OOI WEI TSANG 74

Error Concealment for Video Repeat pixels from previous frame l Interpolate pixels from neighbouring region l Interpolate motion vectors from previous frame l NUS. SOC. CS 5248 OOI WEI TSANG 75

Inferred MV for P Frame I NUS. SOC. CS 5248 OOI WEI TSANG P 76

Inferred MV for I Frame P NUS. SOC. CS 5248 OOI WEI TSANG P I 77

Evaluation l Peak Signal to Noise Ratio (PSNR) l 2552 over average square of differences in pixel values l Calculated in d. B (10 log NUS. SOC. CS 5248 OOI WEI TSANG 10) 78

Performance of SR-RTP NUS. SOC. CS 5248 OOI WEI TSANG 79

Today’s Summary 80

How to recover packet loss Retransmission l FEC l Error Concealment l NUS. SOC. CS 5248 OOI WEI TSANG 81

How to limit the damage of error Interleaving l Key frame selection l Change reference frame l NUS. SOC. CS 5248 OOI WEI TSANG 82