Voice Over IP - Tackling Challenges

Improvement in telecom network infrastructure, evolution ofdigital technologies and standardization, have together made it viable to useexisting data networks for voice applications. This clearly reverses the trendof the past. The major technologies are Voice over IP (VoIP) and Voice overFrame Relay (VoFR). However, in future we might see VoATM and many more.

VoIP and VoFR form the foundation of computer telephony andrelated applications, besides promising substantial cost savings. 

However, behind these benefits are the technology challengestransmitting voice over these networks while maintaining quality and reliabilityof traditional telephone networks. With data networks we are more concernedabout reliability of data and delay is secondary. Voice over data network bringsthe need of minimum possible delay. The quality of service is another importantaspect.

Circuit Vs Packet Network Technologies

PSTN is a circuit-switched network. By this we mean that atelephone call reserves an end-to-end physical circuit between the origin anddestination for the duration of the telephone call. The circuit is fullyavailable to the caller, and it is not available to any other network users.

Packet-switched networks such as the Internet, corporateintranets, and most other data-oriented networks, do not reserve a circuitbetween endpoints. Instead, they break up messages or files into many smallpackets. Each packet may take a different route from origin to destination. Thisrequires each packet to include header information to ensure routing to theproper destination, and to reconstruct the message in its correct sequence atthe endpoint. In VoIP we have a 40-bytes header and in VoFR we have a 4-byteheader for the same purpose.

Technical Challenges

Decades of PSTN experience have set some key qualityparameters for the end user. These are

• Virtually all calls (> 95 percent) completed

• Once speaking, no disconnection

• High "toll" quality voice

• Low delay

• Full PBX dialing, call management and fax functions

Though VoIP has been technically demonstrated, and evencommercialized since 1995, it has not been able to convince the enterprise user.To accomplish this, the packet voice system must meet expectations that havebeen set by PSTN user experience.

Let us have a look at some of these challenges in detail, whythey are important and how they are handled in VoIP/VoFR.

Congestion and Delay

Delay (also referred to as latency) can be defined as thetime required for transmission from origin to destination. Congestion, andresulting delay, are much bigger challenges for packet voice networks because ofthe shared nature of network resources, and because of the processing thatoccurs at routers and at endpoints within the network. In the case of e-mail, adelivery time of several hours is common and tolerable. Even more real-timeapplications such as database queries can tolerate several seconds of delay.

By contrast, two-way voice telephony is delay intolerant.Telephone network designers have determined that humans can endure only up toabout 250 milliseconds (ms) delay before the interval becomes annoying tocallers. Delay is introduced at many points in a network; depending on thescenario it may be very easy to exceed the 250 ms limit.

It is helpful to categorize delay into two types: fixed andvariable. Fixed delay can be thought of as the end-to-end delay for any voicepacket even if it did not encounter congestion points in the network. Variabledelays are incremental delays that are caused primarily by congestion in thenetwork or in the gateway voice routers.

For analog voice waves to be transmitted across a datanetwork they must first be compressed and digitized. Compression delay takesfrom 20 to 45 ms, depending on the compression algorithm. At the receiving end,decompression takes about 10 ms or less, regardless of the algorithm. The chartbelow shows a detailed analysis of various fixed delays normally encountered.

Fixed Delays in Data Networks

The sum of all these fixed delays is between 70 and 130 ms.Taking the worst case, the call can tolerate up to 120 ms variable delay, beforeexceeding the 250 ms threshold.

Since queuing delay may be up to 20 ms, the network delaymust not exceed about 100 ms. Frame Relay networks based on current-generationaccess devices and switches do not add measurable delay above what is alreadyaccounted for under the fixed delay. Therefore, it is well under the 250 msthreshold. However, if we use VoIP in Internet scenario, this factor mainly isdependent on ISP and is not under any control. That is the reason that featureslike RSVP/Prioritization which reserve bandwidth for voice are important when wespeak of VoIP.

Variable delays, as the name suggests, change in real-time asa function of traffic and congestion on the network. Queuing to send the dataonto the WAN can take anywhere from 10 to 20 ms. A number of schemes existwithin Frame Relay to minimize queuing delay. However, queuing can addsignificant delay if voice is competing with other applications, especially onIP networks where the technologies for prioritization and congestion control areless mature.

On the Internet, delays are highly variable. Generallyone-way Internet delays fall anywhere between 50 and 400 ms, but it is not atall uncommon to far exceed these levels. Hence, quality in VoIP is based to alarge extent on end-to-end bandwidth available with ISP, which is a verydifficult factor to control.

Packet Loss

Network congestion also results in dropped packets, or packetloss. Data application protocols, such as TCP, automatically retransmit losspackets but retransmission is not possible in telephony due to its real-timerequirement. In speech transmissions, packet loss of less than 5 percent isusually imperceptible to the user. Intermittent losses in the 5-10 percent rangemay be perceptible but tolerable, assuming proper buffering and errorconcealment methods have been implemented within the gateway routers. Greaterthan 10 percent packet loss is generally not tolerable in an enterprisetelephone solution.

Well-engineered Frame Relay and private IP networks arevirtually lossless, although IP will require higher bandwidth than Frame Relayto achieve comparable packet loss levels. The Internet, again, exhibits widevariability. During peak usage hours, loss rates in excess of 20 percent arecommon, even within the networks of reputed ISPs.

Fortunately, a number of technologies exist to minimize thepotential problems. For example, Motorola has been providing voice capability inits Vanguard series access routers since 1995.

Minimizing Congestion and Delay

Prioritization is most effective when coupled with othertechniques that also speed up the flow of voice packets. For example, Motorola’sVanguard router gives voice packets priority over other packets containinglegacy and LAN data.

Prioritization algorithms can also dynamically limit the sizeof data frames when voice is present, and further the number of data frames thatcan be queued up in front of any voice frame.

Congestion control in IP networks is a less mature science,but methods such as UDP packet prioritization are emerging to help networkmanagers keep voice traffic flowing smoothly.

Buffering and error concealment–prioritization andcongestion control greatly diminish the delay and packet losses. However,gateway routers must employ error concealment and buffering technologies tocompensate for the delay and packet loss that will inevitably occur. Buffers atthe receiving end queue up a small amount of packets prior to playout,effectively eliminating the variability of delays that may have occurred in thenetwork.

Voice compression and silence suppression–voice compressionalgorithms make it possible to provide high quality audio while using bandwidthefficiently. Pulse Code Modulation (PCM)–the standard for digital voice coding–isoptimized for speech quality but consumes 64 Kbps, a full DSO or B channel.Other voice compression algorithms try to model PCM more efficiently by usingfewer bits.

More recently, the ITU has specified the G.729 and G.723.1voice compression standards. G.729 is an 8K "toll" quality compressionmethod, which has been adopted by the Frame Relay Forum for its VoFR standards.G.723.1 is a 5.3 or 6.3 near toll-quality algorithm that has been adopted aspart of ITU H.323, an umbrella standard for simultaneous voice, data, and videoover IP networks.

Use of standards-based voice compression may or may not beimportant, depending on the network and the application. However, the emergenceof these standards is a promising development because it will lead to greaterchoice and interoperability among vendors of networking equipment.

Voice compression algorithms employ silence suppression. Thistechnique relies on the pauses between speech bursts to provide additionalcompression. During normal conversation, speech typically occurs 40 percent ofthe time. The rest is silence. Silence suppression takes advantage of the pausesand only transmits during voice spurts. During silent periods, other activevoice or data packets can use the bandwidth.

Echo cancellation–another issue that can emerge whenplacing voice over data networks is echo, a phenomenon in which the transmittedvoice is reflected to the point from which it was transmitted. Depending on itsseverity, echo can be very annoying. In fact, if the delay time between thespeech and the echo return is significant–45 ms or more–the echo can bringthe conversation to a halt.

The most sophisticated method of eliminating echo is with anecho canceller, which builds a mathematical model of a speech pattern andsubtracts it from the transmit path. To work effectively, echo cancellation musttake place in the same gateway router that performs the voice compression andpacketization. Otherwise, expensive external echo cancellers will be required.

Other enabling technologies–the prior section focused onissues and technologies specifically relating to the preservation of voicequality over IP and Frame Relay packet networks. However, there are additionaltechnologies, which are important to preserve or expand the voice and faxfunctionality of existing PBX-based voice networks.

Voice signaling features–in most situations, organizationswill want to take advantage of the cost savings of packet networks whilepreserving the functionality and investments in their existing PBXs. To achievethis, gateway routers need to support common channel signaling methods such asISDN Primary Rate signaling and Q.SIG. The former is the signaling standard forconnections between digital PBXs and the public telephone network. Q.SIG is aPBX-to-PBX peer
signaling standard that allows interoperability between value-added PBXfunctions such as conference
calling and call forwarding.

Lack of support for these protocols means that the telephonenetwork will lose functionality–clearly not a desirable outcome of moving topacket voice. Implementers must select packet voice gateway devices that arecapable of passing through all necessary signaling information to ensure amigration that is transparent to the end users.

For VoIP, the ITU has defined the H.245 standard (part ofH.323). It defines the co-ordination of multiple session elements–voice, data,and video–as well as supporting functions such as directory services.

Voice signaling is used to set up a call and monitor itsend-to-end status. Voice signaling requirements vary depending on the type ofinterface and network locations that need to be reached. For example, FXO andFXS are relatively easy to support, because they change little or not at allfrom country to country or PBX to PBX. The off-hook type is used in applicationsthat require an audio path but have no requirements for signaling. E&M onthe other hand, is more difficult to support because there are dozens of E&Msignaling protocols in use. Transparent Voice Signaling for E&M ports is apreferable way to pass the signaling states transparently from near-end userport to far-end user port.

Conclusion

Needless to say that the time has come when we cannot ignoretechnology to transmit voice over data networks. There are challenges but thensolutions do exist and with improved infrastructure, it is certainly moreproductive in many ways to implement voice in data networks. VoIP has manyadvantages and some issues and so does VoFR. The ideal way will be to implementthe solution with technology that can support both VoIP and VoFR together alongwith support of PSTN. Users within organization should not know if calls arerouted via VoIP or VoFR or PSTN and should be able to talk to each othertransparently with maintained voice quality. If the corporate network isdesigned properly it can make use of all these advantages and cut down corporatecosts and increase productivity in many ways.

Anil Gupta is director,
Convergent Communications