frank fitzek,tatiana k. madsen, and patrick seeling ip header compression enabling high quality...

115
Frank Fitzek,Tatiana K. Madsen, and Patrick Seeling IP Header Compression Enabling High Quality Consumer-Oriented Communications Aalborg University, Denmark [email protected] [email protected] Arizona State University, USA [email protected]

Post on 21-Dec-2015

220 views

Category:

Documents


0 download

TRANSCRIPT

Frank Fitzek,Tatiana K. Madsen, and Patrick Seeling

IP Header Compression Enabling High Quality Consumer-Oriented Communications

Aalborg University, Denmark

[email protected]

[email protected]

Arizona State University, USA

[email protected]

2Center for TeleInFrastructure

Content

IntroductionPotential savings due to Header CompressionHeader Compression schemesActivities in the field of Header CompressionTools and voice/ video quality evaluationConclusions – why HC should be applied

About the Tutorial

4Center for TeleInFrastructure

Presenters Prof. Dr. Frank Fitzek

Head of Future Vision group

Dept. of Communication Technology

Aalborg University, Denmark

Email: [email protected]

Dr. Tatiana K. Madsen

Dept. of Communication Technology

Aalborg University, Denmark

Email: [email protected]

Patrick Seeling

Department of Electrical Engineering

Arizona State University

USA

Email: [email protected]

5Center for TeleInFrastructure

Acknowledgements

Prof. Martin Reisslein, ASU, US

Stephan Rein, TU Berlin, Germany

Stefan Hendrata, Carmeq, Germany

acticom GmbH, Germany

Introduction

7Center for TeleInFrastructure

Introduction

2G (GSM) is voice dominated3G (UMTS) is IP basedlarge IP overheadlink bandwidth is limited (25 billion Euro for frequencies)idea: use header compression to reduce IP overheadheader compression has to be robust3GPP has chosen Robust Header Compression (ROHC) open question: Quality of Service?

8Center for TeleInFrastructure

Why do we need header compression?

Large IP overhead for real time services voice serviceaudio servicesvideo service

Smaller packets have smaller delaysSmaller packets are less error-prone on the wireless link

9Center for TeleInFrastructure

Wireless Channel

CharacteristicLimited bandwidthWireless link is very

error prone, BER as high as 1e-3

High error rates are tolerated in order to allow efficient usage of radio resources

10Center for TeleInFrastructure

IP Protocol Suite

11Center for TeleInFrastructure

12Center for TeleInFrastructure

Multiple Description Coding (MDC)

13Center for TeleInFrastructure

MDC with Overhead

14Center for TeleInFrastructure

AMR Speech Codec

15Center for TeleInFrastructure

Finding the redundancy

Redundancy between header fields of the same packet (e.g.

RTP/UDP/IP) are referred to as intra packet redundancy

Redundancy between parts of header between different

packets (RTP) are referred to as inter packet redundancy

The redundancy changes with the IP version: IPv4/6

16Center for TeleInFrastructure

Header Fields RTP/UDP/IPv4

17Center for TeleInFrastructure

Header Fields RTP/UDP/IPv6

Potential savings due to Header Compression

19Center for TeleInFrastructure

Potential Savings for Voice/Audio Services

Four example voice/audio codecsLPCGSMG.711AMR

These values does not depend on the content (no silent detection)

20Center for TeleInFrastructure

Voice

mean bitrate IPv4 savings  IPv6 savings 

codec [kbps] [%] [%]

LPC 5.6 74 81

GSM 13.2 55 65

G.711 60.0 21 29 

21Center for TeleInFrastructure

Voice AMR

mean bitrate IPv4 savings  IPv6 savings 

codec [kbps] [%] [%]

AMR Mode 7 5.6 74 81

AMR 13.2 55 65

AMR 60.0 21 29 

22Center for TeleInFrastructure

Potential Savings for Video Services

H.26L video coder

These values does depend on the video content as well as the codec settings

Different video sequences

Different video quality

QCIF video format

23Center for TeleInFrastructure

Frame Size Trace For Star Wars II

Star Wars IICD 1955 kbit/s

Star Wars IICD 2880 kbit/s

24Center for TeleInFrastructure

Video Format/Quality

QP values 1, 31, 51

QP 1 QP 31 QP 51

25Center for TeleInFrastructure

P. S

eelin

g an

d M

. Rei

ssle

in a

nd F

.H.P

. Fitz

ek a

nd

S. H

endr

ata.

Vid

eo Q

ual

ity

Eva

luat

ion

fo

r W

irel

ess

Tra

nsm

isis

on

wit

h R

ob

ust

Hea

der

C

om

pre

ssio

n. 2

003.

.

26Center for TeleInFrastructure

P. S

eelin

g an

d M

. Rei

ssle

in a

nd F

.H.P

. Fitz

ek a

nd

S. H

endr

ata.

Vid

eo Q

ual

ity

Eva

luat

ion

fo

r W

irel

ess

Tra

nsm

isis

on

wit

h R

ob

ust

Hea

der

C

om

pre

ssio

n. 2

003.

Header Compression Schemes

28Center for TeleInFrastructure

Different Header Compression schemes

Compressed TCP – Van Jacobsen RFC 1144only for TCP/IP for wired networks

Perkins improvement for of CTCP

IPHConly for IP protocolno feedback

29Center for TeleInFrastructure

General Structure of Header Compressors

Two entities: compressor and decompressorCompressor sends initial baseBase is used by compressor and decompressorCompressor removes redundancyDecompressor adds removed informationProposed solution differ in a possible feedback channel

N N

Base

Compressor Decompressor

Base

N*

30Center for TeleInFrastructure

CTCP (Van Jacobsen)

TCP/IP header compressionUsing delta compressionDesigned for wired networksNot robust against error-prone linksBase update with each new incoming packet

31Center for TeleInFrastructure

Loosing synchronization

Synchronization loss = decompressor’s copy of the base is different from the compressor’s copy

Synchronization is lost any time a packet is dropped Detection: using detection of TCP retransmissions. All

retransmissions are sent uncompressed

32Center for TeleInFrastructure

Performance of VJ scheme in case of random errors

0

0,5

1

1,5

2

2,5

3

3,5

4

344 K 328 K 550 K

VJ, random erros

no VJ, random errors

VJ, no errors

no VJ, no errors

When synchronization is lost, the decompressor starts to toss packets base update more often than needed

S. J

. Per

kins

and

M. W

. Mut

ka,

Dep

ende

ncy

Rem

oval

for

Tra

nspo

rt

Pro

toco

l Hea

der

Com

pres

sion

ove

r N

oisy

C

hann

els.

199

7.

Kby

tes/

s Throughput of bulk data transfersFile sizes of 344K, 328K, and 550K

33Center for TeleInFrastructure

Perkins – Refinement of CTCP

Perkins & Mutka – improvement of CTCP in case of noise presence Differentials are sent against a base that changes infrequently packet

loss does not cause endpoints to loose synchronization All packets refer to the first packet of the frame the same mechanisms can be used to detect loss of synchronization

34Center for TeleInFrastructure

Perkins – Refinement of CTCP

Base refresh (sending uncompressed header) – to combat overflow problems

Robustness introduced by periodically repetition of full base information each N packets

N packets define a frame

Larger overhead Less compression due to higher delta values Additionally, 1 byte of CID (connection identifier) is transmitted

35Center for TeleInFrastructure

Performance of Perkins scheme

No errors

2,8

2,9

3

3,1

3,2

3,3

3,4

3,5

3,6

3,7

344 K 328 K 550 K

kbyt

es/s Perkins, no errors

VJ, noerrorsno VJ, no errors

Random errors

0

0,5

1

1,5

2

2,5

3

344 K 328 K 550 K

kbyt

es/s

Perkins, randomerrorsVJ, random errors

no VJ, randomerrors

S. J. Perkins and M. W. Mutka, Dependency Removal for Transport Protocol Header Compression over Noisy Channels. 1997.

Throughput of bulk data transfersFile sizes of 344K, 328K, and 550K

36Center for TeleInFrastructure

IP Header Compression (IPHC)

Provides extensions to VJSupport UDP, IPv6, Additional TCP features

Uses delta encoding

37Center for TeleInFrastructure

TWICE algorithm

TCP header compression reduces throughput over lossy linksBandwidth is wasted when unharmed segments

are retransmitted after a timeout

Possible solutions:Perkins algorithmTWICE algorithm

38Center for TeleInFrastructure

TWICE algorithm

Decompressor can detect loss of synchronization by using TCP checksum

Motivation: totally lossless HC is not possible, make an educated guess

If inconsistency is due to a single lost segment + lost segment increments the compression state in the same way

Apply TWICE the delta of a current segment

39Center for TeleInFrastructure

Compressed RTP (CRTP)

Compressed RTP (RFC 2508) Compresses 40 byte header to 4 or 2 bytesFirst-order changes

Expected changes in the fields that can be predicted, no transmission of differences needed

Second-order changesChanges that have to be compressed

Enhanced Compressed RTP (RFC 3545)Refinement of CRTP in presence of packet loss,

reordering and long delaysLocal retransmissions and repeated context updates are

used

40Center for TeleInFrastructure

Robusr Checksum-based Compression (ROCCO)

Refinement of CRTPIncludes checksum over uncompressed header

facilitation of local recovery of the synchronizationTargeted to cellular usage

41Center for TeleInFrastructure

Robust Header Compression

RTP/UDP/IPUDP/IPIPuncompressed

42Center for TeleInFrastructure

ROHC Modes

43Center for TeleInFrastructure

States of Compressor and Decompressor

44Center for TeleInFrastructure

Unidirectional

45Center for TeleInFrastructure

Optimistic

46Center for TeleInFrastructure

Reliable

47Center for TeleInFrastructure

Decompressor

Activities in the field of header compression

49Center for TeleInFrastructure

ROHC working group

RFC 3095 ROHC

(Framework + RTP. UDP, ESP, uncompressed)

RFC 3242: LLA profile - ”0 byte”

RFC 3408: 0-byte for R-mode

RFC 3241: ROHC over PPP

RFC 3816: ROHC MIB

RFC 3843: ROHC for IP

RFC 3220: SigComp

50Center for TeleInFrastructure

Link-layer Assisted ROCH(0-byte)

Purpose is to efficiently match existing applications to existing link technologies

Air interfaces, as GSM and IS-95, will be used in all-IP networks, but their radio bearers are optimized for specific payload size.

Adding even 1 byte of ROCH header is costlyHeader-free packet format

51Center for TeleInFrastructure

LLA ROCH

Lower layers provide the necessary informationCare should be taken of

Packet type identifierSequence numberCRC

1 byte

Smallest header

in ROCH RTP

Smallest header

in LLA

NHP (No Header Packet)

Header field functionality

provided by other means

52Center for TeleInFrastructure

LLA ROCH

Zero-byte operation for U/O modes (RFC 3242)Zero-byte operation for R-mode (RFC 3408)Periodic context verification is performed

CSP (Context Synchronization Packet) contains only header information, no payload

53Center for TeleInFrastructure

Interfaces towards the Assisting Layer

ROCH RTP

LLA profile

Interface

ROCH to AL

Link technology

ROCH RTP

LLA profile

Interface

ROCH to AL

Link technology

channel

54Center for TeleInFrastructure

Signaling Compression (SigComp)

Motivation3GPP R5 introduces IP Multimedia subsystem (IMS) that

uses Session Initiation Protocol (SIP) for call signaling and session setup

SIP is text-based. SIP message is from a few hundreds bytes up to two thousand bytes. On average 500 bytes

For cellular networks large message size is problematic introduce delays

Compression of signaling messages is desirable

55Center for TeleInFrastructure

Signaling Compression (SigComp)

SigComp RFC 3320

Requirements for Signaling CompressionEfficient compression 1:8 – 1:15

Compress any text based protocol

For bidirectional application protocol, the choice to use SigComp is independent in both directions

Transport independent

56Center for TeleInFrastructure

SigComp Architecture

Local applicationLocal application

Transport layerTransport layer

Compressor

dispatcher

Compressor

dispatcherDecompressor

dispatcher

Decompressor

dispatcher

Compressor 1Compressor 1

Compressor 2Compressor 2State handlerState handler

State 1State 1

State 2State 2

Decompressor

(UDVM)

Decompressor

(UDVM)

SigComp layer

SigComp

message

SigComp

message

Application message and

Compartment identifierCompartment

identifier

Decompressed

message

Tools

58Center for TeleInFrastructure

Tools

NetMeterAudioMeterVideoMeter

59Center for TeleInFrastructure

Netmeter Tool

Bandwidth w/o ROHC

Bandwidth with ROHC

Header compression

Overall compression

60Center for TeleInFrastructure

VideoMeter Tool

PSNR calculation is standard metric for objective video quality measurements

Freezing for lost frames

VideoMeter tool for visualization

61Center for TeleInFrastructure

ROHC testbed

62Center for TeleInFrastructure

Metrics

Peak Signal to Noise Ratio (PSNR) for quality comparison

2

10

25510logPSNR

MSE

63Center for TeleInFrastructure

VideoMeter Setup

YUV Original

Encodedoriginal

YUV Encoded

YUV Transmitted

Transmitted bitstream

Erroneous YUV

1

2

3

4

5

1

2

4

5

1

2

2

4

5

Transmission testbed:

ROHCNormal

Voice Quality Evaluation for Wireless Transmission with ROHC

S. Rein and F.H.P. Fitzek and M. Reisslein

Voice Quality Evaluation for Wireless Transmission with ROHC. 2003. in International Conference on Internet and Multimedia Systems and Applications (IMSA 2003), pages

461-466. Honolulu, USA.

65Center for TeleInFrastructure

GSM Encoder

RTP

UDP

IP

ROHC

link

GSM Decoder

RTP

UDP

IP

ROHC

link

IP RTP GSMUDP

ROHC GSM

Original voice Transmitted voice

Com

munication S

ystem

UMTS link error simulation

Protocol suite with ROHC

66Center for TeleInFrastructure

Methodology for ROHC evaluation

Communication System with ROHC

Communication System without ROHC

Original speechDistorted speech Distorted speech

Predict ROHCspeech quality

Predict speech quality

Calculate gain for ROHC

67Center for TeleInFrastructure

Voice quality evaluation framework

Usually expensive software required (state of the art PESQ software available for 10.000 U.S.$)

Alternative methodology: used a set of elementary objective metrics to predict the subjective voice quality

Metrics represent sensible engineering trade off to networking studies

Performance of the metrics is usually verified by a correlation analysis

68Center for TeleInFrastructure

Voice quality metrics: Correlations to subjective quality

Segmental SNR 0.77

Inverse linear spectral distance 0.63

Delta form spectral distance 0.61

Log area ratio 0.62

Energy ratio 0.59

Log likelihood 0.49

Cepstral distance 0.93

Why use an array of metrics for predicting the subjective voice quality?

69Center for TeleInFrastructure

Every metric covers different distortion typesCoding and noise distortions in the time and

frequency domainGood reliability by including metrics verified

by different authors

70Center for TeleInFrastructure

Varying delay with IP packet voiceReference and transmitted voice file have to

be synchronized Developed segmental cross correlation

(SCC) algorithm in the time domainSCC makes elementary metrics usable for

modern communication systems

71Center for TeleInFrastructure

missing samples

wrongly inserted

Segmental Synchronization

72Center for TeleInFrastructure

Results: Delay jitter measurements

73Center for TeleInFrastructure

Results: SNR measurements

74Center for TeleInFrastructure

Results: voice quality

75Center for TeleInFrastructure

Results: voice quality

76Center for TeleInFrastructure

Results: Mean Opinion Score

77Center for TeleInFrastructure

On top of bandwidth savings: Voice quality is improved

ROHC roughly cuts bandwidth for voice transmission in half

ROHC is a very useful complement to third generation mobile systems

Video Quality Evaluation for Wireless Transmission with Robust Header Compression

P. Seeling and M. Reisslein and F.H.P. Fitzek and S. Hendrata

Fourth International Conference on Information,  Communications & Signal Processing 

Fourth IEEE Pacific-Rim Conference On Multimedia 15-18 December 2003, Meritus Mandarin Hotel, Singapore

79Center for TeleInFrastructure

Video services over wireless networks are gaining more and more interest

Services are needed that convince customers to buy new equipment (3GPP)

3GPP networks support video services such as the IMS and the MBMS entities

Problem: Video services need much more bandwidth than voice calls, which makes them hard to sell

Video services

80Center for TeleInFrastructure

Serving Sara (560x304 pixels)

81Center for TeleInFrastructure

Motivation

Video services are transmitted over IP networks using the RTP/UDP/IP protocols

Video is encoded efficiently, but the protocol stack overhead is not

The protocol overhead comprises a large portion of the traffic (even more for small video formats as for the mobile phones)

Therefore: Compression of the protocol overhead needed

82Center for TeleInFrastructure

Methodology

Uncorrelated bit errors as recently found for UMTS channels

Simulated with and without ROHC implementation

O-Mode

83Center for TeleInFrastructure

ROHC Results

84Center for TeleInFrastructure

Insights

Protocol header compression can be achieved by using ROHC

Higher compression ratios for smaller video formats and higher quantizationLower fraction of packets comprised of actual

video data (i.e., payload)Lower error probability and state changes within

ROHC

85Center for TeleInFrastructure

ROHC Results

86Center for TeleInFrastructure

ROHC is an efficient header compression scheme for multimedia in wireless environments

Compression achieved depends on video format and video content

Utilization of ROHC for multimedia streaming does not introduce additional losses in terms of perceived video quality (PSNR)

Cooperative Header Compression

F.H.P. Fitzek and T. K. Madsen and P. Popovski and

R. Prasad and M. Katz.

Cooperative IP Header Compression for Parallel Channels in Wireless Meshed Networks. In IEEE International

Conference on Communication (ICC). 2005

88Center for TeleInFrastructure

Scenario: multiple channels between terminal and cellular network

89Center for TeleInFrastructure

HC Approach Single Channel

90Center for TeleInFrastructure

HC Approach Multiple Channels

91Center for TeleInFrastructure

HC Approach Multiple channels with channel errors

92Center for TeleInFrastructure

More Robust HC Approach by Exploiting Multiple Channels

93Center for TeleInFrastructure

HC Approach: Possible Implementation

94Center for TeleInFrastructure

Error healing

95Center for TeleInFrastructure

AICs construction for three cooperative channels

96Center for TeleInFrastructure

Packet Error Probability vs Channel Error Probability

97Center for TeleInFrastructure

Efficiency vs Channel Error Probability

98Center for TeleInFrastructure

99Center for TeleInFrastructure

Cooperative HC

Advantageous of the proposed schemeExploit cooperative behavior of parallel channels Bandwidth efficiencyLow memory consumption and low complexityRobustnessNo need for feedback channel (can be used for

multicasting)Only a small number of cooperative channels are

needed to perform efficiently

Analogy: FEC

Why HC should be applied!

101Center for TeleInFrastructure

Network Provider’s view

Increased quality of service for the userDelay (web pages, download)BandwidthCost (?)

CapacityInstalled cells can support more users $Cells that will be installed with larger coverage,

which results $

102Center for TeleInFrastructure

Network Capacity

How can a network provider calculate its potential saving?

A very simplified viewThumb rule

103Center for TeleInFrastructure

Notation

P = mean payload sizeH_u = uncompressed headerH_c = compressed headerRR = Ratio = no. of users with HC / total number of

users.G: Capacity gain. Capacity gain can be defined e.g.

how much more users in percent we can serve if some of them will activate HC (total number of users with HC activated / total number of users without header compression)

104Center for TeleInFrastructure

Equation

105Center for TeleInFrastructure

Some examples

0

2 0

4 0

6 0

8 0

1 0 0

1 2 0

1 4 0

N o n R O H C R O H C G a i n

R R = 3 0 % ; P = 1 3 ; I P v 4 : H C = 6

G a i n

R O H C

N o n R O H C

106Center for TeleInFrastructure

Some examples

0

2 0

4 0

6 0

8 0

1 0 0

1 2 0

1 4 0

1 6 0

1 8 0

2 0 0

N o n R O H C R O H C G a i n

R R = 7 0 % ; P = 1 3 ; I P v 4 : H C = 6

G a i n

R O H C

N o n R O H C

107Center for TeleInFrastructure

Some examples

0

5 0

1 0 0

1 5 0

2 0 0

2 5 0

N o n R O H C R O H C G a i n

R R = 9 0 % ; P = 1 3 ; I P v 4 : H C = 6

G a i n

R O H C

N o n R O H C

108Center for TeleInFrastructure

Some examples

0

2 0

4 0

6 0

8 0

1 0 0

1 2 0

1 4 0

1 6 0

1 8 0

N o n R O H C R O H C G a i n

R R = 9 0 % ; P = 3 3 ; I P v 4 : H C = 6

G a i n

R O H C

N o n R O H C

109Center for TeleInFrastructure

Some examples

0

5 0

1 0 0

1 5 0

2 0 0

2 5 0

N o n R O H C R O H C G a i n

R R = 9 0 % ; P = 3 3 ; I P v 6 : H C = 6

G a i n

R O H C

N o n R O H C

110Center for TeleInFrastructure

Some examples

0

5 0

1 0 0

1 5 0

2 0 0

2 5 0

3 0 0

N o n R O H C R O H C G a i n

R R = 9 0 % ; P = 1 3 ; I P v 6 : H C = 6

G a i n

R O H C

N o n R O H C

Reference List

112Center for TeleInFrastructure

Reference List

F.H.P. Fitzek , S. Hendrata, P. Seeling and M. Reisslein. Chapter in Wireless Internet -- Header Compression Schemes for Wireless Internet Access. CRC Press. 2004.

V. Jacobson, Compressing TCP/IP headers for low-speed serial links, RFC 1144, February 1990.

M. Degermark, B. Nordgren and S. Pink, IP header compression, RFC 2507, February 1999.

S. J. Perkins and M. W. Mutka, Dependency Removal for Transport Protocol Header Compression over Noisy Channels. In Proc. of IEEE International Conference on Communications, Canada, June 1997, pp.1025-1029.

M. Degermak, M. Engan, B. Nordgren, and S. Pink, Low-loss TCP/IP header compression for wireless networks. In Proceedings of ACM MobiCom’96. October 1997, pp.375-387.

113Center for TeleInFrastructure

Reference List

C. Perkins and J. Crowcroft, Effect of interleaving on RTP header compression, in Proceedings of IEEE Infocom 2000, Tel Aviv, Israel, 2000, pp. 111-117.

K. Svanbro, H. Hannu, L.-E. Jonsson, and M. Degermark, Wireless real-time IP services enabled by header compression, in Proceedings of the IEEE Vehicular Technology Conference (VTC), Tokyo, Japan, 2000, pp. 1150-1154.

C. Bormann, C. Burmeister, M. Degermark, H. Fukushima, H. Hannu,

L.-E. Jonsson, R. Hakenberg, T. Koren, K. Le, Z. Liu, A. Martensson, A. Miyazaki, K. Svanbro, T. Wiebke, T. Yoshimura, and H. Zheng, Robust Header Compression (ROCH): Framework and four profiles: RTP, UDP, ESP, and uncompressed, RFC 3095, Tech. Rep., July 2001.

114Center for TeleInFrastructure

Reference List

L-E. Jonsson and G. Pelletier, ROCH: A Link-Layer Assisted Profile for for IP/UDP/RTP, RFC 3242, Tech. Rep., April 2002.

R. Price, C. Bormann, J. Christoffersson, H. Hannu, Z. Liu, and J. Rosenberg, Signaling Compression (SigComp), RFC 3320, Tech. Rep., January 2003.

T. Koren, S. Casner, J. Geevarghese, B. Thompson and P. Ruddy, Enchanced Compressed RTP for Links with High Delay, Packet Loss and Reodering, RFC 3545, Tech. Rep., July 2003.

115Center for TeleInFrastructure

Reference List

F.H.P. Fitzek and T. K. Madsen and P. Popovski and R. Prasad and

M. Katz. Cooperative IP Header Compression for Parallel Channels in Wireless Meshed Networks. 2005. in IEEE International Conference on Communication (ICC).

P. Seeling and M. Reisslein and F.H.P. Fitzek and S. Hendrata. Video Quality Evaluation for Wireless Transmisison with Robust Header Compression. 2003. in Proceedings of the IEEE Fourth International Conference on Information, Communications & Signal Processing and Fourth IEEE Pacific-Rim Conference On Multimedia (ICICS-PCM 03), pages 1346-1350. Singapore.

S. Rein and F.H.P. Fitzek and M. Reisslein. Voice Quality Evaluation for Wireless Transmission with ROHC. 2003. in International Conference on Internet and Multimedia Systems and Applications (IMSA 2003), pages 461-466. Honolulu, USA.