Voice

and Multimedia

over IP

Voice over IP scenarios

Digital voice

Elements of the sampling theory PCM and TDM, Classical PSTN architecture Quality, Codec, Technical issues

RTP - Transport of voice

and other multimedia, soft realtime applications

The H323 architecture SIP signalling

PC to PC and PC to phone VoIP

Software allows voice calls from one PC to another - Softphone Voice is converted to IP packets at PC or at PSTN Gateway Works best on broadband connections Carried like normal data traffic by ISP (best effort)

Phone-to-Phone VoIP

Use existing telephone with an Analogue Telephone Adapter (ATA) or use an IP phone Both connect to Broadband modem Called party may be another VoIP user Or, via a gateway, a traditional PSTN customer

VoIP in the PSTN

Many traditional PSTN calls are carried as VoIP in part For efficiency reasons they may travel with other IP traffic Different from normal VoIP Invisible to end customer Private IP network, not internet Controlled quality

Corporate VoIP

Increasingly growing popularity Inter-site voice carried as IP over leased line or VPN (huge moneysaving especially for international communications) Additionally, a single desk wiring infrastructure (LAN) may carry both data and voice (as VoIP) Voice may stay as IP or be converted to PSTN

Digital voice

Digital voice

Sine wave Analog Signal

Sampling Nyquist Fs=2Fc

Quantizing Quantization Noise

Sampling

8KHz Sampling Rate

One sample each 125 s (8KHz), 8 bits per sample x 8KHz 64Kbps

Nyquists theorem

To be able to restore without loss a signal with a cut frequency fc, we need to sample it at a frequency of at least 2fc. Sampling at 8KHz we loose frequency components above 4KHz. For music audio this is not acceptable (MP3:44.1 KHz)

Quantization

We choose the quantized value closest to the Analog signal

Logarithmic Quantization Better SNR

We give more levels (more precision) to small values => better SNR

A-law (Europ) and -law (USA)

PCM - Pulse-Code Modulation

DS0 (Digital Signal level 0

single digital voice channel in the PCM system Synchronously one sample each 125s 64 Kbps total throughput
Analog only at the local loop ADC and DAC at the toll office (a word about Echo here)

A hierarchy of TDM multiplexes (PDH & SDH) used in PSTN core

Synchronous Digital Hierarchy

E1 (or DS1 - Digital Signal level 1)

Multiplexes synchronously 32 DS0 channels, but channels 0 and 16 reserved for signaling (remains 30 voice channels) Sample rate must remain 8000 samples (Bytes) /s for all channels => the frame time must always remain 125s (at all the levels DS2 DS3 etc) E1 bit rate: 8bits x 32 / 125s = 2.048 Mbps

Synchronous Digital Hierarchy E carriers

Similarly, E2 multiplexes 4 E1 channels, E3=16E1, etc..

Ex: T1 = 24 DS0

Codec The Encoder-Decoder

G.711 The classical codec

Simply what we saw so far.. 8000 Byte samples, 64 Kbps encoded in the DS0 format

Can we achieve the same quality with a smaller bandwidth?

Silence suppression Compression

Many encoding algorithms and corresponding Codecs have been invented and standardized, all seeking the reduce bandwidth with a minor (or no) loss of quality.

Drawback: adding some computational delay

This is minor compared to bandwidth gain

Some standardized Codecs

Voice Quality

Subjective: Mean Opinion Score (MOS)

5 4 3 2 1

Perfect. Like face-to-face conversation or radio reception. Fair. Imperfections can be perceived, but sound still clear. Annoying. Very annoying. Nearly impossible to communicate. Impossible to communicate

Objective

Derived from bandwidth, delay and packet loss rate

Toll (payable) service must be in the 4 to 5 range

Delay

Delay should be <150ms for toll quality

Human sensitive above 250 ms

Sources of delay

Jitter

Voice is a CBR service

A packet that arrives too late must be dropped Continuous playback

What is Jitter?

Different packets experience different delays in the network

Varying queue sizes Different paths introduce jitter, but load balancing algorithms work on a micro-flow level, not on packet level (coarse grain) TCP transport not suitable

Congestion window management increases jitter retransmitted packets will be dropped anyway. VoIP uses RTP/UDP

Solution

Jitter buffer at the receiver Deliberately adding some delay to ensure continuous playback

Jitter Buffer

The bigger the playback delay, the lesser the jitter It is important to have low jitter from the network, otherwise we need a big jitter delay.

Packet Loss rate

IP network (Best effort service)

Routers might drop packets (RED: Random Early Detection)

Packets arriving too late are deliberately droped by receiver Loss rate tolerable up to 5%

10% if we use Interleaving

RTP - The Realtime Transport Protocol

RTP - The Realtime Transport Protocol

Transport mechanism designed for Soft Realtime applications The most important feature is Timestamping

Manage jitter Synchronize multiple streams, i.e. audio and image in a movie

RTP

P: padding (multiple of 4 bytes) X: exist extension headers (unused) M: app-specific marker (ex: start of video frame) Payload Type: specifies type and encoding algorithm (ex: G.711 voice) Sequence numbering to detect loss of packets Timestamp: jitter management and synchronization of streams SSRC: identifies a stream (many streams may be multplexed on a RTP stream, such as video and audio) CSRC combined with CC: permits many sources to be mixed (CC is the count and SSRC is a list). The mixer would be the SSRC

H.323 architecture

H323 Architecture and Protocol Stack

H.323 call sequence

11.
12. 13. 14.

PC discovers Gatekeeper (broadcast discovery request packet) Gatekeeper responds Client registers in the Gatekeeper zone (RAS protocol H225) Client requests bandwidth (this is to manage call admission and QoS!) Client establishes TCP session with Gatekeeper for signalling Client sends SETUP message (Q.931 signalling) Gatekeeper contacts Gateway and sends CALL_PROCEEDING to client Gateway Calls destination number Callee rings, Q.931 ALERT message (GWclient) Callee responds, Q.931 CONNECT message (GWclient) H245 messages to negotiate capabilities (codec, video, teleconf) Data flow proceeds using RTP Call terminates (callee hangs), GW alerts client (Q.931) Client releases Bandwidth to gatekeeper (RAS message)

SIP
The Session Initiation Protocol

SIP - Session Initiation Protocol

LOOKUP and REPLY methods not specified in SIP (free implementation) REGISTER method allows clients to register to proxy and inform of location Another option to use SIP is P2P. Caller and callee exchange directly INVITE/OK/ACK through a TCP connection and then start exchanging data

SIP Methods

The encoding of messages is based on HTTP

Comparison of H.323 and SIP

MP3 (for music audio)

Threshold of audibility as function of frequency

The masking effect

Video Analog Systems

The scanning pattern used for NTSC video and television.

The JPEG Standard

The operation of JPEG in lossy sequential mode.

JPEG Block Preparation

RGB input data Y=.3R+.59G+.11B

After block preparation I=.6R-.28G-.32B Q=.21R-.52G+.31B

JPEG -- DTC

One block of the Y matrix

The DTC coefficients

JPEG

Computation of the quantized DTC coefficients.

The JPEG Standard

The order in which the quantized values are transmitted Run-length encoding: exploiting repetitions

The MPEG Standard

Synchronization of the audio and video streams in MPEG-1.

MPEG Frame Types

I (Intracoded) frames: Self-contained JPEG-encoded still pictures. P (Predictive) frames: Block-by-block difference with the last frame. B (Bidirectional) frames: Differences between the last and next frame. D (DC-coded) frames: Block averages used for fast forward.