ETSI STQ Workshop “Compensating for Packet Loss in Real-Time Applications”, Feb 2003

Speech/Audio Coding for IP networks

Alan DuricSen. Systems ArchitectSIP/email: alan.duric@globalipsound.com

Agenda

• Introduction• Traditional approach to speech coding for VoIP

applications• New paradigm• iLBC• Questions• Demo

QoS – (endpoints) perspective

• Year after year the same story

• More then 3000 papers since 1984

• Limited ToS support at the end points

• Introduction of new technologies and applications is making situation even more difficult

QoSQoS is already 19 years old is already 19 years old ––is it time to get a real job !? is it time to get a real job !? ☺☺[ ]

Traditional approach to speech coding for VoIP

CELP SPECIFICS

• Current low bit rate codecs: ITU G.729, G.723.1, GSM-EFR, and 3GPP-AMR were developed for circuit switched & wireless telephony and are all based on the CELP (Code Excited Linear Prediction) paradigm.

• CELP coders are stateful, they have memory, error propagation results from lost or delayed packets.

• Long time is needed to resynchronize coder and decoder (often 70-100 ms)

original

iLBC

g729

g723

PLC Staterecovery

iLBC Advantage over CELP

iLBC, like other GIPS codecs treats every packet individually, making it suitable for packet communications.[ ]

New Paradigm

• Approach & diagnose• What can be improved?• How?• Proof of (concept and design)

Approach

We need holistic view/approach for both

• Horizontal (end-to-end) QoS perspective

• Vertical (top-down) QoS perspective

Vertical (Top Down) Perspective

PhysicalPhysical

LinkLink MLPPP/FR/ATM AAL1MLPPP/FR/ATM AAL1

NetworkNetwork IP/WFQ/IPIP/WFQ/IP--precprec

TransportTransport RTP/UDP/RSVPRTP/UDP/RSVP

SessionSession SIP/H.323SIP/H.323

PresentationPresentation Speech Codecs/…Speech Codecs/…

EC

What impacts perceived quality?VoIP End Point

Lost Packets

CODEC

Delay & Jitter

Packet discardsby jitter buffer

VoIP End Point

Needed to mitigate impact of Needed to mitigate impact of delay/jitter and packet lossdelay/jitter and packet loss[ ]

Echo

Playout Controller

What can be improved?

• One side solutions:– Advanced Playout Controller– AEC, NEC with right design

• Both end solution:– Codec

Design principles

• High basic quality• Robustness (e.g. for codec no inter-frame dependency,

MDC)• Low complexity• …• Realistic test methodology and tools during design

phases

MDC

Input SpeechFrame

Packets sent onIP Network

N + 1 N N - 1N + 2 NN + 1D2D1D1D1 D2D2

20 msN+2

20 msN+1

20 msN

Improvements for “one-side” solutions

Source: Lockheed Martin Global Telecommunications (COMSAT)

Adaptive jitter bufferJitter Fixed jitter buffer NetEQ™Adaptive jitter bufferAdaptive jitter bufferJitterJitter Fixed jitter bufferFixed jitter buffer NetEQ™NetEQ™140

120

20

80

100

40

60

0

140

120

20

80

100

40

60

00 200 400 600 800 1000 1200 1400 1600 1800 2000

Packet number

Dela

y (m

s)

0 200 400 600 800 1000 1200 1400 1600 1800 20000 200 400 600 800 1000 1200 1400 1600 1800 2000

Packet number

Dela

y (m

s)Saved approximately 30—80 ms

iLBC Performance

Source: Dynastat

GIPS Ehanced G.711+GIPS NetEQ™

G.711+GIPS NetEQ™

G.711+ITU PLC

G.729A

G.711+No PLC

Matching PSTN QualityTelephony bandwidth speech test result

SOURCE LOCKHEED MARTIN GLOBAL TELECOMMUNICATION (COMSAT)

5

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

NETWORK CONDITION (% PACKET LOSS)

0% 5% 30%10% 15% 25%20%

MO

S

GIPS iPCM™-wb+GIPS NetEQ™-wb

G.722+GIPS NetEQ™-wbG.722.1

Source +no PLC

Better Than PSTN QualityWideband speech

0% 5% 10% 15% 20% 25%

5

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

MO

S

NETWORK CONDITION (% PACKET LOSS)

wide band sound quality

Proof of concept and design (part 3)

telephony band sound quality

iLBC (Internet Low Bitrate Codec)

iLBC (internet Low Bitrate Codec)

• Speech sampled at 8 kHZ,• using a block-independent linear-predictive coding (LPC)

algorithm.• Bandwidth 13.33 kbps (50 bytes per 30 ms)• Frame size 30 ms (support for 20 ms in the next revision)• Complexity and memory requirements are similar to ITU

G.729A• Basic Quality is equal to or better than G.729. Packet loss

robustness is significantly better than G.729. • Packet loss concealment - Integrated example solution

The Core iLBC method

• Gain-shape waveform matching forward in time• Gain-shape waveform matching backward in time

• Start state encoding

• Pitch enhancementOriginal speech segment

Decoded segment representation

• Packet loss concealment

iLBC - IETF work

• IETF deliverables, submitted during February ‘02:– iLBC codec specification draft - experimental

standards track– iLBC RTP Payload Profile - regular standards

track (AVT)– Statement about IPRs in ILBC and its “freeware nature”

Summary

• Accelerate deployment of VoIP technology by using realistic QoS enhancements and solutions that are already available

• VoIP endpoints, focus on both: one side improving solutions and both end improving solution

• Move quality exprience to the next level with wideband coders

Questions ???

Demo

More information

• Web site www.ilbcfreeware.org with:– Info about initiative– Info about codec– Latest iLBC IETF drafts (spec and payload format)– Latest iLBC float point Source code– FAQ list

• IETF drafts:– draft-ietf-avt-ilbc-00.txt - codec spec (exper. stds track)– draft-ietf-avt-rtp-ilbc-00.txt - RTP payload profile (AVT group)

• Web site www.globalipsound.com

• Free demo SIP client available, please request at:SIP/email: alan.duric@globalipsound.com