Telephony is perhaps the most innovative invention of the twentieth century, and the voice network is still the largest network in the world. Telephone and hardcopy mail have been the most important communication tools until the advent of the Internet. The Internet has reshaped the way a corporation communicates with its employees, partners and customers. The growth of the Internet has spawned huge data network backbones connecting the remote sites of global corporations. Even with the advent of the Internet, telephony is still an integral tool for conducting business, with thousands of calls being made daily by corporations around the world. Corporations have invested in two different networks, namely voice and data, requiring two different departments and two different budgets.

In today's networking there is a cost-effective and viable alternative to this traditional model. As an alternative, Voice over IP allows corporations to utilize existing data networking infrastructures in a more cost-effective manner by rerouting telephony traffic over the Internet and/or corporate backbones. Combining the telecommunication and data traffic onto a single medium, companies can reach these savings. The Voice over IP concept builds on this data network infrastructure, using the TCP/IP protocol as its base. There is a variety of components, which make up a Voice over IP infrastructure. Not all solutions will use all components. The clients’ components can be a normal PSTN telephone and a multimedia PC with microphone and speakers, running VoIP client software or a LAN-based telephone. Any client should be able to communicate with any other client, given the correct switching and server components.

Voice Characteristics                                                                                              

• Voice traffic typically occupies a frequency range of 300 Hz to 3300 Hz and is analog in nature. This is as much as a human vocal chord can vibrate to create an audible signal.

• Voice traffic is delay-sensitive. An end-to-end delay of over 250 ms is unacceptable.

• Voice traffic is not loss-sensitive. We can tolerate a loss of a syllable or two in a conversation.

Voice networks or Public Switched Telephone Network (PSTN) are circuit switched and digital in nature and to a large extent, deterministic. Analog voice from a telephone is usually converted into digital format at the CO, or locally at the PBX. Circuit switched networks are those which allow a circuit between subscribers to be established, either for the duration of the call or on a permanent basis. This path may traverse intermediate switching points. When the communication session is over, the switching equipment then breaks down the path, and the switching equipment that was in use is once again available to other users. Circuit switched networks therefore can easily offer a guarantee in terms of bandwidth, voice quality and delay which is often termed as QOS in the data networking language.

A typical voice signal has an approximate bandwidth of 4 KHz. To convert an analog signal to a digital signal, the analog signal has to be sampled at twice the maximum frequency to maintain quality; in other words, at 8000 times a second (Nyquist criterion). Typically, the present day voice networks use at least 8 bits of encoding per voice sample. Hence the total bandwidth per voice channel is 64 Kbps (8 bits multiplied by 8000 samples per second). This means that every quality voice call that is made today typically occupies 64 Kbps of bandwidth and is totally dedicated for the duration of the call.

 Data Characteristics  

 Data traffic:

• Is digital in nature end-to-end.

• Is not delay-sensitive to a very large extent.

• Is loss-sensitive.

Data networks are predominantly packet switched and probabilistic in nature. Probabilistic networks operate on best-effort delivery unlike deterministic networks, which guarantee delivery. Each message to be transmitted through the network is first divided into a number of smaller, self-contained message units known as packets that are then routed to their destination. Each packet may take a different route from origin to destination, traveling along network circuits that are shared with packets from other messages. Packet switching mandates that each packet should have its own header information, to ensure proper routing and to reconstruct the message in its correct sequence at the destination. Transmission of data through the network may be on a virtual circuit basis, or on a datagram basis, depending on the implementation. Packet switched networks are known for economic utilization of network resources. One of the largest data networks known today is the Internet. Other examples are X.25, Frame-Relay and corporate IP networks.

 Business Drivers for Convergence

The most important factors that contribute to the success of any leading-edge technology are commercial viability and technical feasibility. It should make good business sense and be able to be implemented with minimal disruption to the existing network

 There are cost reduction benefits in terms of toll bypass like overcoming long distance charges. Packetizing voice communications into IP packets and sending them over the existing corporate WAN infrastructure can significantly reduce telephone bills. This is accomplished by interleaving the voice packets into the pockets of bandwidth created by the "bursty" nature of data traffic. One of the largest segments of a corporation's telecommunications bill is international fax charges. These costs can be greatly reduced by sending the fax traffic over the corporate Voice over IP infrastructure.

Most corporate WANs are provisioned to support the high-end of the normal traffic load. Most of the times there will be an excess of capacity in the WAN environment that would normally be sitting idle, waiting for the "peak periods" of traffic. Leveraging the existing IT infrastructure to take advantage of its excess of resources can provide clear, crisp voice communication that is cost effective.

Collaborative Computing/Unified Messaging-voice, data and multimedia messages can be a part of a communication. All three media types can be stored in a common message store or database, enabling easy voice annotation of e-mail or inclusion of a video clip into an e-mail message. A single interface to the message store, either a desktop device or a phone, can access either type of message. Now voice messages can be retrieved from your e-mail user interface and played over your PC's speakers. Or, you can have your e-mail read to you over the phone as you are rushing to catch a plane. With a simple phone command, you can forward a critical e-mail message to an assistant to respond to in your absence, providing faster closure of key business issues.

Reliable phone to phone, PC to phone and PC to PC communication voice conversations can be established from a PC to a PC, a PC to an internal or external phone, or between two phones with access to a Voice over IP gateway. Some service providers (ISPs and telcos) are beginning to offer these gateway services.

For any technology to be widely implemented, endorsement from standards bodies is an absolute necessity. VoIP has the blessings of ITU (International Telecommunications Union), IETF (Internet Engineering Task Force), VoIP (Voice Over Internet Protocol) Forum, ETSI (European Telecommunications Standards Institution) and IMTC (International Multimedia Teleconferencing Consortium).

 Points to Ponder                                                                                                       

Ever since the 1970s, there were attempts to converge voice and data networks. The evolution of the digital PBX in the early '70s and CTI in the late '80s, and all the lessons learned thereafter through the triumphs and failures, has culminated in an open and a standards-based VoIP solution.

Contemporary LANs had shared backbones with probabilistic access mechanisms. Today's switched backbones are an absolute Star topology with dedicated bandwidth to the desktops. Voice Local Area Networks (PBX wiring) are also absolute Star topology. Conclusion: Similar topologies and more deterministic LAN.

Every Wide-Area data Network is connection-oriented, or if connectionless, is built on reliable media. As an example, X.25 has elaborate error correction mechanisms up to Layer 3, and Frame-Relay is built under the assumption of total reliable error free media. ATM is totally connection-oriented. Conclusion WAN links were always made to be more deterministic.

Consider closely a T-1/ E-1 line that is provisioned by your carrier. A T1 line has 24 channels of 64 Kbps. Have you wondered why T1 is 1.544 Kbps (E1 is 2.048 Kbps)? Every line that is provisioned and/or conditioned by your carrier has something to do with voice. The underlying premise for sampling at 8000 times a second (that is, every 125 microseconds) comes from the Nyquist criterion for voice bandwidth. For T1, 193 bits are used per sample, 8000 times a second, which is nothing more than 1.544 Mbps. E1 uses 256 bits per sample every 8000 times a second, which is 2.048 Mbps. Conclusion: Every data trunk that exists today is built on voice based parameters.  

The ability to make a voice call over an IP-based network poses a lot of challenges. This section elaborates all the challenges.

•  High bandwidth required for a toll quality call (64 Kbps/call).

•  Low end-to-end delay (<250 ms).

• Connection-oriented voice call on connection-less network.

Quality of voice is a highly subjective parameter. MOS (Mean Opinion Score) is a widely used methodology to define quality of voice. MOS is measured on a scale of 1 to 5, where 5 is the best quality. MOS between 4 and 5 is considered toll quality, and 3.5 to 4 is considered acceptable communication quality.

Compression Techniques:

Solving the Bandwidth Problem  

To achieve toll quality, dedicated bandwidth of 64 Kbps per voice channel is required. This type of coding technique is called PCM (Pulse Code Modulation). This bandwidth is almost impossible to dedicate in a data network. Various compression algorithms are being used to reduce this stringent bandwidth requirement. ADPCM (Adaptive Differential Pulse Code Modulation) further reduces the bandwidth required to 32 Kbps. Digital Signal Processing (DSP) based Code Excited Linear Prediction (CELP) can further reduce the bandwidth required per voice channel to 5.6 Kbps. Various standards as defined by ITU exists today. By far the most popular standard is the G.723.I. This technique is part of H.323 (Video on IP) standard. But solving the bandwidth problems creates another major Issue.

Compressing voice to lower bandwidths results in a lower MOS rating and higher delay. Acceptable delay should be less than 250 ms. Various sources of delay are listed in the next section.

 Delay Issues

Delay can be categorized into two components: namely, fixed and variable.

Fixed/Controllable delay:

• Compression Delay

• Processing Delay

• Buffering/Queuing Delay

• Transmission Delay

• LAN Network Delay

• Decompression Delay

Compression and decompression delay is small, and is a maximum of 50 ms. The use of higher memory and better processors can reduce processing and buffering delays in a router. The LAN delay that comes into picture for a Web-based voice application can be kept to a minimum limit by using point-to-point switched topology. The end-to-end sum of fixed delays is usually less than 130 ms. Variable or network delay is the delay in the IP/WAN/Internet cloud, which is out of our control. The next section deals with solving this problem.

 RSVP: Solving the Delay Problem

Network delay was by far the most hindering factor in the voice-data convergence. This section analyzes some of the contributing factors for convergence. Running real time voice to voice applications on a probabilistic IP data network has its own challenges. However, the number of enabling factors far outweigh the disadvantages.

Technically speaking, advances in the field of QOS, which have contributed to the IP stack, tend to make a probabilistic data network more deterministic. The most promising solution to this problem was developed and standardized by the IETF. Resource ReSerVation Protocol (RSVP) can prioritize and guarantee latency to specific IP traffic streams. RSVP enables a packet-switched network to emulate a more deterministic circuit-switched voice network. (It should be noted that RSVP still works on best effort delivery.) RSVP is an OSI Layer3 protocol with routing support, and thus can be seamlessly integrated into an IP-based router/switch/network. With the advent of RSVP, VoIP has become a reality today. Most of the VoIP product vendors support RSVP. Details on this protocol are described in RFC 2208. In summary, with RSVP enabled for a voice stream, we can accomplish voice communication with tolerable delay on a data network.

Summary

The recent advance in data networking over the past decade has made VoIP a feasible technical alternative to PSTN for corporate voice communications. There are still some legal and regulatory hurdles to be crossed in some countries. It is at least safe to say that companies can employ VoIP for PBX-to-PBX networking. Reducing the bandwidth required per voice channel and having a greater control over the delay factor on the network, combined with significant cost benefits, have finally culminated into a unified voice-data network. Most large enterprises and service providers are investigating VoIP as a way to maximize utilization of existing TCP/IP infrastructure, minimize international telecommunications charges and enable the use of other converged applications, such as video and data conferencing, Web-based call centres and Unified Messaging.