the challenges of voice-over-ip-over-wireless - ericsson · pdf file20 ericsson review no. 1,...

IntroductionToday, the accumulated volume of data traf-fic is on the verge of surpassing the accu-mulated volume of voice traffic in all pub-lic networks. Given the growth in the areasof wireless voice and data, we see that thecombination of mobile and Internet com-munication constitutes the driving force be-hind third-generation wireless systems,which promise to support at least 144 kbit/s(384 kbit/s) in all radio environments, andup to 2 Mbit/s in low-mobility and indoorenvironments.

The standardization of third-generationwireless systems is rapidly progressing in allmajor regions of the world. These systems—which go under the names of IMT-2000(ITU), UMTS, and EDGE (ETSI/3GPP)—will extend the services provided by currentsecond-generation systems (GSM, PDC, IS-136, and IS-95) with high data-rate capa-bilities. The main application for these ser-vices will be wireless packet transfer; for in-stance, for wireless access to the Internet.However, support will also be provided for

high data-rate circuit-switched services,such as real-time video.1

UMTSThe universal mobile telecommunicationssystem (UMTS) is being standardized in theThird Generation Partnership Project(3GPP), which is a joint effort between theEuropean Telecommunications StandardsInstitute (ETSI) and the Association ofRadio Industries and Broadcasting (ARIB,Japan). The basic radio-access technologyfor UMTS/IMT-2000 in all major areas ofthe world is wideband code-division multi-ple access (WCDMA). The 1999 release ofthe UMTS standard was the first to be im-plemented in commercial products.

The radio-access part—the universal ter-restrial radio access (UTRA)—includes afrequency-division duplex (FDD) mode anda time-division duplex (TDD) mode. TheFDD mode is based on pure WCDMA,whereas the TDD mode includes an addi-tional time-division multiple access(TDMA) component.

The WCDMA system, which uses wide-band direct-sequence technology (DS-CDMA), fully supports the UMTS andIMT-2000 requirements for 384 kbit/swide-area coverage and 2 Mbit/s local cov-erage. Particularly noteworthy features ofWCDMA are • support for interfrequency handover,

which is necessary for high-capacity hier-archical cell structures (HCS);

• support for capacity-improving technolo-gies, such as adaptive antennas and multi-user detection.

• built-in service flexibility, which pro-vides spectrum-efficient access for currentas well as future applications; and

• efficient handling of bursty applicationsvia an advanced packet-access mode.

WCDMA also provides efficient support formultimedia services; that is, for transferringmultiple services on one connection.1,2

EDGEThe GSM and TDMA/136 technologiesmake up the foundation on which the com-mon radio access for data services will be of-fered. The enhanced data rates for GSM andTDMA/136 evolution (EDGE) concept,which ETSI and the Universal WirelessCommunications Consortium (UWCC)have adopted as the migration path fromGSM and TDMA/136, fulfills the require-ments for third-generation wireless systemsaccording to IMT-2000. EDGE is capable

20 Ericsson Review No. 1, 2000

The challenges of voice-over-IP-over-wirelessGöran AP Eriksson, Birgitta Olin, Krister Svanbro and Dalibor Turina

Much as second-generation radio access brought mobile telephony capa-bilities to the mass market, third-generation radio-access technology willintroduce value that extends beyond basic telephony. The widespreadgrowth of the Internet has created a mass market for multimedia andinformation services. The challenge of providing this access is twofold:from the market perspective, the challenge is to merge the installed baseof users in cellular and Internet environments; in terms of technology, thechallenge is to find common denominators for cellular solutions on oneside, and efficient Internet access on the other. To successfully meetthese challenges, the third-generation wireless systems must be designedto offer a multitude of services, provide considerable flexibility, structuredQoS handling, and cost-effective access, while ensuring coverage withhigh radio spectrum efficiency.

These design challenges need to be resolved in the future evolution ofUMTS and EDGE standards, beginning with the standardization of release2000 in ETSI/3GPP, both regarding the radio-access networks and thecommon core network. A common denominator for this evolution is theInternet protocol suite, which provides end-to-end solutions for transportand network communication, including cellular (radio) access links overUMTS/EDGE.

The authors describe the requirements and current trends in the evolu-tion of the third-generation wireless systems toward supporting moregeneric multimedia applications on a common platform. More especially,they discuss the challenges and solutions involved in designing a packet-switched radio-access network that efficiently supports VoIP service,highlighting session control for VoIP service, quality-of-cellular service forthe cellular link, and the challenges of VoIP over wireless.

Ericsson Review No. 1, 2000 21

of offering data services of up to 384 kbit/sand is thus a global complement to theUMTS radio-access network.

The roadmap for EDGE standardizationhas been divided into two phases. Initial em-phasis was placed on enhanced general pack-et radio service (EGPRS) and enhanced circuit-switched data (ECSD). According tothe ETSI time plan, these standards werepart of the 1999 release. The second phaseof EDGE standardization, which is target-ed for release in 2000, will define improve-ments for multimedia and real-time ser-vices. Other objectives will include thealignment of services and interfaces withUMTS, to allow EDGE and UMTS to sharea common core network.

Real-time IP applications over wireless Second-generation radio-access technologybrought mobile telephony to the market.Third-generation radio-access technologywill extend beyond basic telephony: a com-mon, IP-based transport and service plat-form will offer mobile users a multitude ofreal-time and interactive services.2

Typical services with real-time require-ments are voice and video, as well as delay-sensitive applications, such as traffic-sig-

3GPP Third-generation Partnership Program

AMR Adaptive multirateAPI Application program interfaceARIB Association of Radio Industries

and BroadcastingBER Bit error rateCDMA Code-division multiple accessCN Core networkCRC Cyclic redundancy codeCRTP Compression for real-time

protocolDS-CDMA Direct-sequence CDMADTMF Dual-tone multiple frequencyDTX Discontinuous transmissionECSD Enhanced circuit-switched dataEDGE Enhanced data rates for GSM and

TDMA/136 evolutionEED Equal error detectionEEP Equal error protectionEGPRS Enchances GPRSETSI European Telecommunications

Standards InstituteFDD Frequency-division duplexFER Frame error rateGGSN Gateway GPRS support nodeGSM Global system for mobile

communicationGW GatewayGPRS General packet radio service

H.323 ITU-T standard for multimediaapplications

HCS Hierarchical cell structureIETF Internet Engineering Task ForceIMT International mobile

telecommunicationIP Internet protocolIPv4 IP version 4IPv6 IP version 6IS Intermediate standardISDN Integrated services digital networkITU-T International Telecommunication

Union – TelecommunicationsStandardization Sector

LAN Local area networkMS Mobile station (terminal)MT Mobile terminalNAS Non-access stratumPDC Personal digital communicationPDCP Packet data convergence protocolPDP Packet data protocolPSTN Public switched telephone networkPT Payload typeQoS Quality of serviceRAB Radio-access bearerRAN Radio-access networkRLC Radio link controlRNS Radio network serverROCCO Robust checksum-based header

compression

RRC Radio resource control RSVP Resource reservation protocolRTCP RTP control protocolRTP Real-time transport protocol SAP Service access pointSDP Session description protocolSDU Service data unitSGSN Serving GPRS support nodeSIP Session initiation protocolSMS Short message serviceTCP Transmission control protocolTDD Time-division duplexTDMA Time-division multiple accessTE Terminal equipmentUDP User datagram protocolUE User equipmentUED Unequal error detectionUEP Unequal error protectionUMTS Universal mobile

telecommunications systemUTRA UMTS radio accessUTRAN UMTS terrestrial radio-access

networkUWCC Universal Wireless

Communications ConsortiumVoIP Voice over IPVoIPoW Voice over IP over wirelessWCDMA Wideband CDMAWWW World Wide Web

BOX A, ABBREVIATIONS

Mobile

Mobile Internet

Internet

1000

800

600

400

200

01996 1997 1998 1999 2000 2001 2002 2003 2004

Year

Millions of subscribers

Figure 1The strong growth of mobile communication is expected to continue. Experts predict thatby the year 2003/2004 there will be close to one billion subscribers of cellular systemsworldwide. Similarly, the Internet will continue to grow. By 2004, the number of sub-scribers to the Internet is also expected to reach one billion. Of this group, more than 350million persons will subscribe to the mobile Internet.

naling systems, remote sensoring, and sys-tems that provide interactive access toWWW servers. The focus of this article,however, is on voice service. The voice ser-vice of third-generation wireless systemsmust, at the very least, offer the same highlevel of voice quality, and be as spectrum-efficient, as present-day second-generationrealizations. The challenge is to implementend-to-end service on IP-based transport.

The main advantage of running IP all theway over the air interface is service flexibil-ity. To date, cellular-access networks havebeen optimized for voice quality and spec-trum efficiency. The demand for service flex-ibility adds a new parameter, as illustratedby Figure 2. Since there are no dependen-cies between an application and the accessnetwork, almost anyone can develop new ap-plications. But for services like voice over IP over wireless (VoIPoW), the main chal-lenge is to achieve quality and spectrum efficiency.

To date, all cellular systems that providevoice service have been optimized in a two-dimensional space whose X-axis and Y-axisare voice quality and spectrum efficiency, re-spectively. Now, a third dimension is beingadded in the form of IP service flexibility.By bridging the radio interface with IPpackets, we suffer a lot of protocol overhead,

which runs counter to the goal of spectrumefficiency.

Network architectureoverviewTo facilitate our ensuing discussion, let usbriefly describe the VoIP service. The basiccomponents of the voice service are two userterminals with IP-based voice applicationsand a network that provides end-to-endtransport between the terminals (Figure 3).The terminals exchange voice samples usingthe real-time transport protocol (RTP),which has been standardized by the IETF.

In some situations, terminals can estab-lish and maintain communication withoutthe involvement of a third-party entity. Atother times, however, the two user end-points cannot establish end-to-end commu-nication without outside intervention; forexample, when they do not know one an-other’s IP address or do not use the samevoice codec. In these cases, a control planeframework is used to route incoming trafficand to negotiate terminal capabilities (codecsupport, multiparty conferencing, and soon)—in traditional telecommunicationsthis functionality, which is referred to as callcontrol, is provided by, say, a GSM mobileswitching center. In the IP world, there are


IP s

erv

ice

flexib

ility

Ideal solution

Pure IP

Circuit-switched voiceVoice

quality

Spectrum efficiency

Figure 2The voice-over-IP-over-wireless challengecube.


two main methods of providing call-controlfunctionality: ITU-T RecommendationH.323, and the IETF session initiation pro-tocol (SIP).

Originally intended for LAN environ-ments, H.323 is an ITU standard for multimedia applications. Today, however,the standard is being adopted for broaderusage. H.323 encompasses a complete architecture and a set of protocols, such asH.225 for call control and H.245 for bear-er control. H.323 uses IETF protocols, suchas the real-time protocol and the resourcereservation protocol (RSVP).

Besides end-user terminals, the H.323 ar-chitecture encompasses gatekeepers, gate-ways, and multiparty units. In this context,emphasis is put on the gatekeepers and gate-ways—which constitute the VoIP server.The gatekeeper part is the controlling unitthat provides call-control functionality; thegateway part contains the user plane func-tions. H.323 call control is based on Q.931,which is also used in GSM and ISDN.

The session initiation protocol, which isan IETF standard draft, is only one compo-nent in the IETF alternative to the H.323paradigm for a complete multimedia archi-tecture. Other necessary protocols and com-ponents include the session description pro-tocol (SDP), the service access point (SAP),and the real-time control protocol (RTCP).

The session initiation and session de-scription protocols (SIP/SDP) do not makeup an architecture; they were designed for

session initiation. In contrast to H.323 andGSM/ISDN, SIP/SDP does not provide acomplete call-control mechanism—an SIPproxy primarily provides routing and ad-dressing services; device management is notincluded. However, the SIP proxy (or VoIPserver) can be enhanced to include func-tionality for offering other services such astranscoding. The session initiation protocolis associated with a paradigm in which callcontrol is distributed over several entities,and in which the user terminal plays a cen-tral role in coordinating these entities.

In summary, the two IP-based terminalsexchange voice samples that have been en-capsulated in RTP over the IP network. Theterminals exchange control signaling be-tween themselves or, with assistance fromnetwork entities such as a VoIP server, es-tablish and maintain communication ses-sions through the network according to ei-ther the H.323 or SIP paradigm.

Both the session initiation protocol andH.323 support end-to-end solutions inwhich the network solely functions as a bear-er. In this case we assume that an SIP orH.323 network call agent can, if so re-quested, support the end point (the termi-nal) with transcoding services.

The mobile terminal supports cellular ac-cess (UMTS/WCDMA or EDGE) and acomplete VoIP application that is based oneither SIP or H.323. We assume that anadaptive multirate (AMR) codec will besupported by future VoIP clients.

Backbone router

VoIP serverCall/session routing Transcoding Multiparty conference units

Edge router

SGSN GGSN

Codec VoIP ctrl

Cellular IP terminal Fixed IP terminal

IP

Radio

Codec VoIP ctrl

IP

Ethernet

RNS

Figure 3Basic VoIP components.

In addition to basic UMTS packet-switched access, the network contains func-tions for adapting media, routing calls, andfor authenticating users and services.3

Overview of the QoSconcept for the cellularlinkThe user’s overall perception of quality ofservice (QoS) is a blanket judgement of theaggregate contributions from essential com-ponents of the communication system. Intypical communication involving UMTS orGSM/EDGE, the transmission phasethrough the radio-access network (RAN) isonly one part of the overall end-to-end com-munication. Thus, even when the radio-access network provides excellent quality ofservice, there is no guarantee that the end-user’s perception of the service or applica-tion is good.4

Bearer services within UMTS andGSM/EDGEGiven that frequency spectrum is a sparseresource in wireless communication sys-tems, we can greatly benefit from applyinga tailored-to-fit quality-of-service conceptto the radio-access network. This way, usingas few radio resources as possible, we canmatch each connection with the stipulatedquality requirement.

To achieve a given network quality of ser-vice, a bearer service—with clearly definedcharacteristics and functionality—is set upfrom the source to destination. Each bearerservice on a specific layer offers its individ-ual characteristics through services that areprovided by subordinate layers. The shadedsections in Figure 4 indicate bearer servicesthat depend on services provided over theair interface.

Once again, because frequency spectrumis a sparse resource, we readily see the ben-efit of being able to classify traffic, in orderto guarantee system capacity and quality ofservice. By being able to differentiate traf-fic flows in the network, we can define fourapplication-related service classes withinUMTS and GSM/EDGE: • The conversational service class is used for

real-time services, such as ordinary tele-phony voice; for instance, VoIP and video-conferencing. The vital characteristics ofthis class are low transmission delay andpreserved time relationships, or low-delayvariation, in the traffic flow.

• The streaming service class applies to real-time audio and video-streaming applica-tions. In contrast to the conversationalclass, this category comprises one-waytransport.

• Typical applications associated with theinteractive service class are WWW brows-ing and Telnet. The fundamental charac-teristic of the interactive class is a request-response pattern, which makes round-tripdelay an important characteristic. In ad-dition, all data transfer must have a lowerror rate.

• The background service class is used for best-effort traffic. Examples of services in thisclass are electronic mail (e-mail), shortmessage service (SMS) and file transfer.Here, too, all data transfer must have alow error rate, although the requirementsfor transfer delay are less stringent.

The transport for each service class can beconfigured in a way that optimizes the effi-ciency of the radio network and fulfills thequality-of-service requirement.

Different radio-access bearers (RAB)transport the services through the radio-access network. Each RAB is associated witha set of attributes that specify the requiredquality (bit rate, delay and error rate) andsupply information on the characteristics ofthe traffic flow. This information is essen-tial for • providing a connection with good quali-

ty through the radio-access network; and• using spectrum efficiently. Examples of RAB attributes are service class,guaranteed bit rate, transfer delay, servicedata unit (SDU) loss rate, residual BER, andtraffic handling priority.

Service requirements for voice

Unequal error detection

Ordinarily, the bits in a frame from a cellu-lar voice codec are divided into three class-es: 1a, 1b and 2. Bit-error sensitivity variesbetween these classes—Class 1a includes themost sensitive bits; Class 2, the least sensi-tive bits.

In a typical second-generation system, thebits in Class 1a are covered by a cyclic re-dundancy code (CRC) that checks for errorsin the frame. Thus we say that the voiceframe uses an unequal error detection (UED)scheme.

If information on the different classes ofbit-error sensitivity cannot be transferredfrom the codec to the radio-access network,or if the bits in the voice frame are not or-



ganized into classes, then the UED schemecannot be used. Instead, an equal error de-tection (EED) scheme—a CRC that coversthe entire voice frame—is introduced. Tohave the quality in these two cases, eachmust receive the same number of frameswith a bad CRC.

Where circuit-switched traffic is con-cerned, only frames with a bad CRC con-tribute to the frame error rate. But in an IP-based radio network, frames with a badCRC, frames that are lost due to jitter, andfatal errors in the IP header contribute to theframe error rate. By fatal errors in the IPheader we mean user datagram protocol(UDP) checksum errors, errors in the linklayer, and header decompression errors.

Unequal error protection

The calculation of the bit error rate (BER)only includes those errors that occur in bitsnot protected by the CRC. The residual er-rors in bits protected by the CRC must beas close to zero as possible. If any residualbit errors exist in the Class 1a bits, the voicedecoder might produce noticeable artifacts.

If UEP is not available (but UED is), thelowest FER requirement of Class 1a and theBER requirement of Class 1b constitute thechannel requirements.

End-to-end delay

The ITU-T recommends the following lim-its on one-way delay:• 0-150 ms—acceptable for most applica-

tions.• 150-400 ms—acceptable provided the

administrator knows what impact thetransmission time will have on the trans-mission quality of other user applications.

• Greater than 400 ms—unacceptable forgeneral network planning; however, insome exceptional cases (such as satellitehops) this limit will be exceeded.

A typical second-generation cellular systemhas a one-way delay of less than 100 ms.Thus, to achieve the same high end-to-endquality, the one-way delay requirement forthe conversational service (terminal-to-PSTN) of a third-generation wireless systemmust also be less than 100 ms.

Optimizations for VoIPoW

ChallengesThe main objective of the VoIPoW conceptis to port voice service to the new packet-data-based platform while retaining the

quality of service that users associate withpresent-day circuit-switched realizations(for instance, GSM circuit-switched voiceservice). To meet this objective, severalquality-of-service issues have to be resolvedon the core network and radio-access net-work sides. One objective is to offer voice asan integral part of emerging public and pri-vate conversational multimedia services,thereby converging the service with archi-tecture solutions developed in the IP do-main. A long-term migration solution tothis realization implies that equivalent so-lutions to most telephony services must beoffered over the wireless IP solutions.

It should be noted that the voice-only ap-plication is considered a special case—aspart of a multimedia session, voice has morecomplex requirements; that is, in terms ofBER and delay, different streams in a multimedia session may have vastly differ-ent QoS requirements. The Internet worldon the fixed side is currently experiencing atremendous growth in number of users andapplications. At the same time, the goal ofbridging the wireless air interface intro-duces additional design challenges. Themain constraint is the capacity of the radiolink, which is a sparse resource that must beused with care. Consequently, one addi-tional requirement for the radio-access net-

TE MT RAN CN

End-to-end service

UMTS/EDGE bearer serviceLocal bearer

service

External bearer service

Edge node toward Iu

UMTS/EDGE

Gateway toward external networks

CN bearer service

Radio-access bearer service

Iu bearer service

Radio bearer service

Physical radio bearer

service

Backbone bearer service

Physical bearer service

CN TE

Figure 4Hierarchical structure of bearer services within UMTS and GSM/EGDE.

work side is to obtain the same or compara-ble spectrum efficiency for voice services asfrom present-day circuit-switched systems.

To achieve spectrum efficiency, we canbenefit from characterizing different pack-et data streams in terms of bandwidth anddelay requirements. Characterizations ofthis kind are useful when implementing ad-mission control algorithms that accommo-date multiple user data streams in availablespectrum. We can also benefit from apply-ing different methods of minimizing theamount of data (such as RTP/UDP/IP head-er compression and session signaling com-pression) to obtain adequate spectrum effi-ciency for voice. By minimizing the amountof overhead, we obtain nearly the same levelof spectrum efficiency as for the referencecase, in which only voice frames are trans-ferred across the radio through a circuit-switched connection.2

Traffic classification in IP networksWhen a bearer is requested from the UTRAor GSM/EDGE radio-access network, theservice it will bear is described by a numberof parameters (radio-access bearer attribut-es), such as guaranteed bit rate, residualBER, packet loss rate, and delay. Therefore,to request an optimum VoIP bearer from theradio-access network, the requesting entitymust know the attribute settings of theradio-access bearer. The procedure for de-termining and assigning values to the at-tributes of the radio-access bearer is called

traffic classification. Obtaining the informa-tion needed to set these attributes is a fun-damental problem of transferring IP servicesover a radio network. In the IP world, theapplication and transport are separate andindependent, whereas in the cellular worldthe two are usually integrated. Consequent-ly, service data must be obtained for settingup an appropriate radio-access bearer. Apartfrom the RTP stream, the VoIP session alsocontains control signaling traffic flows withdifferent characteristics. Moreover, a VoIPsession might easily be extended to includevideo or a background FTP transfer.

Explicit method

The introduction of an application programinterface (API) between the application andthe radio link enables the explicit orderingof proper radio-access bearers. The API canbe client-based or server-based. By client-based we mean that the API is between theapplication part and the radio part in themobile terminal. By server-based, we meanthat the API is defined on the fixed side ofthe network between the call agent (anH.323 server, for example) and the cellularradio network.

Implicit method

A more transparent method of obtaining ser-vice data is to develop a flow-classificationalgorithm that examines the headers ofpackets and extracts information on flowsfor identifying and characterizing them.Figure 5 shows the principle that applies inreal-time situations. The RTP header con-tains a payload-type (PT) field that identi-fies the source codec. However, the PT canalso be dynamic, in which case informationon the codec is unavailable. In this case, thealgorithm measures parameters, such aspacket size and inter-arrival time. Fromthese parameters, it might be possible toidentify the codec or to give direct input tothe radio-bearer parameters. How much in-formation the algorithm can provide de-pends on the depth of the tree. For instance,the algorithm could be extended to capturespecific signaling messages. Obviously, thecomplexity of the tree and the time it takesto go through it are limiting factors.

Header compression for real-time IP The large headers of the protocols used whenvoice data is sent over the Internet consti-tute a major problem with voice over IP overwireless. An IP packet with voice data willhave an IP header (20 octets), a UDP head-


TCPUDP

Non- RTP

Measurement phase

RTP

codec

Figure 5Flow classification tree.


er (8 octets), and an RTP header (12 octets)for a total of 40 octets. In IPv6, the IP head-er is 40 octets, which raises the total to 60octets. The size of the voice data depends onthe codec and can be as small as 15 to 30octets. These numbers make a strong argu-ment for terminating the IP protocols be-fore the air interface: the IP/UDP/RTPheaders consume too much bandwidth andmake inefficient use of valuable radio spec-trum. However, it is possible to overcomethis problem with header compression.

No header information in a voice packetcan be discarded, but because there is a highdegree of redundancy in the fields of head-ers of consecutive packets that belong to thesame packet stream, this information can becompressed by means of header-compressionalgorithms. These algorithms maintain acontext, which is essentially the uncom-pressed version of the last transmitted head-er at each end of the channel over whichheader compression occurs. The compressedheaders solely carry changes to the context.Static header fields need not be transmitted

at all, and fields with only minor changescan be updated with only a few bits. How-ever, when frames are lost or damaged overthe channel, as can occur on cellular links,the context on the downstream side cannotbe updated properly and the decompressionof subsequent headers produces incorrectheaders. Thus, the header-compressionschemes must have mechanisms for in-stalling context, for detecting when the con-text is out of date, and for repairing thedownstream context when it is incorrect.

Several header-compression algorithmsare being standardized by the IETF. ForVoIP, the compressed RTP (CRTP) cancompress 40-octet IPv4/UDP/RTP headersto a minimum of 2 octets.5 To repair con-text, the CRTP relies on an upstream linkover which the decompressor sends requestsfor updating the headers. All packets re-ceived by the decompressor are lost whilethe context is out of date, since the headerscannot be decompressed. The round-triptime over the link will thus limit the effi-ciency of the context-repair mechanism.

VoIP session

Voice samples

RTP RTP TCPUDP

RTCP Basic SIP DTMF Charg. Suppl. service

UDP UDP

UDP

PDP_RTP

PDP_BE

PDCP_BE

RAB0

PDCP_Voice PDCP_Sig

RLC_BE RLC_Voice RLC_Sig

PDP_Sig

UDP RSVP

PDCP

RLC

MAC

PHY

IP IP IP IP IP IP IP

Figure 6Conceivable realization of a VoIPoW usersession.

Link-level simulations show that the pack-et-loss rate for CRTP is about four timesgreater than for an ideal scheme in a VoIP-over-WCDMA scenario. To be viable, theheader-compression scheme for VoIP-over-wireless cannot be less efficient but must beless fragile than CRTP.

Ericsson’s response to these requirementsis a header-compression scheme that pro-vides a high degree of compression and issuitably robust for use in cellular applica-tions (Figure 7). The robust checksum-based header-compression (ROCCO)scheme is heavily geared toward local con-text repair.6 A checksum that covers theoriginal (uncompressed) header is includedin the compressed header to introduce a re-liable way of detecting when • the context is out of date; and • local attempts to repair the context have

succeeded.Moreover, to achieve really good compres-sion and robust performance without losinggenerality, Ericsson has also introduced acompression profiles concept. To achieveoptimum performance, different compres-sion profiles handle different RTP streamsand channel conditions. Various profiles arebeing developed for voice and video streams.A general, all-purpose profile might also bedeveloped. Figures 8 and 9 compare ro-

bustness and compression performanceagainst channel quality for VoIPoW servicewhen tested with different types of com-pression schemes and channel types.

In addition to the ROCCO header check-sum, a code that is included in the com-pressed header provides the decompressorwith information on how the header fieldshave changed; for example, due to loss overthe cellular link. For the ROCCO VoIP pro-file, this code contains enough informationon previous headers to enable local repair ofthe context after several (up to 26) consecu-tive packets have been lost between the com-pressor and the decompressor. The profilewith the maximum compression ratio has aminimal header size of one octet. The abil-ity to repair context locally virtually elimi-nates the negative effect of long round-triptimes on header-compression performance.

Radio-access bearers for VoIPThe main challenge in designing radio-access bearers is to find suitable implemen-tation points (Figure 2). These are domi-nated either by requirements for IP serviceflexibility or by requirements for spectrumefficiency. Different points satisfy these re-quirements to different degrees. For exam-ple, the amount of protocol overhead (thatis, the IP header) cannot be minimized if IPservice flexibility must be combined withend-to-end encryption. On the contrary, ifwe solely want to implement voice service,we can use a radio-access bearer to obtain asolution, which in terms of spectrum effi-ciency, is comparable to that of present-daysolutions.

In summary, the radio-access network ofa third-generation all-IP wireless systemshould support VoIP applications by meansof a• voice-optimized radio-access bearer

whose service characteristics and spec-trum efficiency match that of circuit-switched voice—for instance, by means ofthe AMR codec.4 All IP-related overheadis terminated before bridging the wirelessair interface. A trusted proxy is needed onthe network side and no IP service flexi-bility is offered to the mobile user.

• voice-optimized radio-access bearerwhose service characteristics and spec-trum efficiency are similar to that of circuit-switched voice. Voice frames aretransferred across the air interface togeth-er with the compressed IP overhead—forexample, using header-compression algo-rithms.1 Additional optimization might


Reconstruct header

Compressor Decompressor

Original header

CRC

Payload

Yes

No

Give up

Try again with modfiedreconstruction

Reconstruction correct?

Forward packet Request update

Figure 7ROCCO—the robust checksum-based header-compression scheme.


be obtained from unequal error protection(UEP) or from unequal error detection(UED) that matches the voice codec inuse. This solution allows for migration toan all-IP wireless system that efficientlysupports telephony service.

• generic radio-access bearer for real-timeuser-to-user IP connectivity; that is, bysupporting data streams that belong tothe multimedia session and usingRTP/UDP/IP. This solution providestotal IP flexibility to new IP multimediaapplications and in cases where securitymechanisms (IPsec) do not allow for head-er compression.

Overview of VoIPoW realizations inUTRA and GSM/EDGE networksTo provide efficient and cost-effective solu-tions for deploying end-to-end IP-basedmultimedia services that satisfy the elevat-ed standard of QoS for UMTS (the radiopart), the UTRA network must select theradio-access bearers that best suit the car-ried service. By optimizing how the radioresource is provided, we can also maximizesystem capacity. Figure 8 depicts a globaloverview of the data streams that relate to aVoIPoW call.

Data streams with similar characteristicsand approximately the same QoS require-ments can be logically grouped and mappedonto the same radio bearer. Within aVoIPoW session, a few relevant classes havebeen identified:• RTP voice stream (RAB2)—the transfer

of voice data requires the lowest possibledelay, no jitter, and BER of 10-4 (less if arobust header-compression mechanismcannot be used). Since low delay require-ments do not permit retransmission, thepreferred choice for the voice stream is atransparent RLC service—with guaran-teed peak bit rate—that does not intro-duce any additional protocol overhead.

• Application signaling (RAB1)—RTCP,H.323 or SIP, RSVP. This category hasless stringent delay requirements but de-mands better data integrity than voice.The requirement for data integrity can bemet by means of retransmission at theRLC level. In some cases, the quality re-quested by this kind of signaling cannotbe matched by a pure best-effort radio-access bearer. Therefore a guaranteedminimum bit rate should be consideredwithin the acknowledged RLC servicemode. To avoid over-providing orthogo-nal codes—provided more bandwidth is

Packet loss rate [%]

CRTP-WCDMA

CRTP-EDGE

ROCCO-WCDMA, 2oct

ROCCO-EDGE, 2oct

ROCCO-WCDMA, 1oct

ROCCO-EDGE, 1oct

0

1

2

3

4

5

6

7

8

9

10

10-4

10-3

10-2

Bit error rate

Figure 8Global overview of data streams that relate to a VoIPoW call.

Average header size (octets/packet)

CRTP-WCDMA

CRTP-EDGE

ROCCO-WCDMA, 2oct

ROCCO-EDGE, 2oct

ROCCO-WCDMA, 1oct

ROCCO-EDGE, 1oct

10-4

10-3

10-2

Bit error rate

1

1.5

2

2.5

3

3.5

4

Figure 9Simulated results of CRTP header compression.

required for occasional bursty signalingduring voice—the use of common orshared channels should be considered.

• Radio resource control (RRC) and non-access stratum (NAS) signaling(RAB0)—UTRA network control signal-ing is used to exchange messages betweenthe UMTS network and the user equip-ment. Rapid and reliable delivery of mes-sages can have a major impact on the per-formance of the entire system. Therefore,the transmission of UTRA network mes-sages must be guaranteed, reliable, andhave low delay and top priority.

Note: not all the data streams of theVoIPoW session depicted in Figure 8 are ac-tive at the same time. In fact, most of themcan be decoupled.

Evaluating the capacity of VoIPoWSystem simulations have been made to de-termine how the introduction of IP serviceflexibility in the WCDMA-based UMTSsystem affects capacity. The focus of the sim-ulations was on the radio link, which is con-sidered to be the bottleneck of the system.Four different cases, each representing a dif-ferent technical solution, were studied:

1.Circuit-switched voice service with equalerror protection. This case is included asa benchmark. The capacity of this solu-tion is used to normalize the other results.In all likelihood, the circuit-switchedvoice service in WCDMA will use un-equal error protection. Thus, in terms ofcapacity, the studied results are slightlypessimistic. Nonetheless, the results ofthis solution suffice for comparison withvoice-over-IP cases, to investigate the ef-fects of the upper protocol layers on sys-tem capacity. For the air interface, this so-lution is the equivalent of one that ter-minates IP-related overhead in the net-work.

2.IP-based voice service with fullRTP/UDP/IP header. This IP servicesends the complete RTP/UDP/IP headerover the air interface.

3. IP-based voice service with CRTP header compression. The same IP-basedservice as in no. 2, above, apart from theintroduction of the CRTP header-compression scheme.

4.IP-based voice service with ROCCOheader compression. The same IP-basedservice as in no. 2, above, apart from theintroduction of the ROCCO header-compression scheme.

The capacity figures for each of the differ-ent cases were derived in a WCDMA systemsimulator. The simulator models an areawith hexagonal cells covered by three sectorsites. During the simulation, mobile callswere initiated according to a Poisson processand distributed uniformly over the area. Thecall duration figures were distributed expo-nentially, with a mean holding time of 120seconds. During calls, the mobile terminalgenerated or received voice frames from amodel of the AMR 12.2 codec operating indiscontinuous transmission (DTX) mode.Before and after calls, some IP-related sig-naling (H.323 and RSVP) was transferredover the radio link. The model did not in-clude RTCP and RSVP signaling duringcalls.

The objective was to estimate capacity ata constant level of quality. However, becauseit is difficult to measure perceived voicequality in a simulator, the quality measureof a single connection was based on frameerror rate (FER). Listening tests indicatethat there is a correlation between the per-ceived voice quality of a connection andFER.

The measure of system quality—to de-termine system capacity—was not based on


System quality, the 95th percentile of FER

Relative load0.40.3 0.5 0.6 0.7 0.8 0.9 1 1.1

0.009

0.010

0.011

0.012

0.013

0.017

0.016

0.015

0.014

Circuit- switched

ROCCO HC

CRTP HC

Full RTP/UDP/IP header

Figure 10Simulation results of UMTS voice services.


an average connection, but on the numberof connections with acceptable quality. Inthis investigation, adequate system qualitywas defined as less than 1% FER for at least95% of all connections. For a given level ofsystem quality, capacity was defined as themaximum load for which this level can beachieved. Capacity was measured relative tothe reference case (1) whose capacity was 1.0.

The simulation results (Figure 10) showthat for low loads, • the interference level of the system is mod-

erate; and • the power control is able to set the indi-

vidual power levels in such a way that95% or more of the users have connectionswith the desired level of quality(FER² 1%).

When the load was increased, the interfer-ence level also increased, and at some pointthe system became overloaded. When thesystem becomes overloaded, interference isexcessive and too many connections are bad;that is we reach the capacity limit of the sys-tem. When the complete RTP/UDP/IPheader was sent over the air interface, ca-pacity dropped to approximately 50% thatof the circuit-switched reference case. WhenCRTP header compression was applied (Fig-ure 10), capacity dropped to approximately80%; when ROCCO header compressionwas used, capacity dropped only 10% (to ap-proximately 90%).

The impact on capacity of RSVP andRTCP signaling, which was not included inthe simulations, depends on the extent towhich it can be reduced and compressed. Ifthe signaling is not reduced, the loss in ca-pacity due to IP-related signaling will equalthat caused by compressed headers.

ConclusionThe widespread growth of the Internet hascreated a mass market for multimedia andinformation services. The challenge of pro-viding these services via third-generationwireless systems is twofold: from the mar-ket perspective, the challenge is to mergethe installed base of users in cellular and In-ternet environments; and in terms of tech-nology, the challenge is to find common de-nominators for cellular solutions and effi-cient Internet access. To succeed in meetingthese challenges, third-generation wirelesssystems must be designed to provide a mul-titude of services, offering considerable flex-ibility and cost-effective access with struc-tured quality-of-service handling and en-

suring high radio-spectrum efficiency. TheUMTS and GSM/EDGE radio-network ar-chitecture and quality-of-service concept aredesigned to support the needs of present-dayand future applications. The concept of bear-er services at different network levels makesup the basis for providing end-to-end quality-of-service—transport through theradio-access network is provided via radio-access bearers.

The main objective of the VoIPoW con-cept is to port voice service to the new pack-et-data-based platform while maintainingthe perceived quality-of-service and spec-trum efficiency associated with present-daycircuit-switched wireless systems. Giventhis objective, we see that we cannot choosea single point of implementation. Instead,the main challenge is to find suitable pointsof implementation that satisfy the voice-service requirements for IP service flexibil-ity or spectrum efficiency. By introducingtraffic classification and header compressionwe can offer a spectrum-efficient VoIPoWservice with high voice quality and IP ser-vice flexibility.

The aim of designing the third-generation all-IP wireless network is to sep-arate core and radio-access network compo-nents, thereby allowing a common packet-switched core network (based on GPRS) tobe used for UMTS and GSM/EDGE radio-access networks.

A key objective of third-generation all-IPnetworks is to provide a capable service plat-form for IP-based applications. The solu-tions we have described for audio streamsand associated control protocols will ad-vance the UMTS network another step to-ward becoming a full-fledged service plat-form that can support demanding services,such as IP-based conversational multi-media.

1 Nilsson, M.: Third Generation Radio AccessStandards. Ericsson Review Vol. 76(1999):3, pp. 110-121.

2 Nilsson, T.: Toward third-generation mobilemultimedia communication. EricssonReview Vol. 76 (1999):3, pp. 122-131.

3 Dahlin, S. and Örnulf, E.: Network Evolutionthe Ericsson Way. Ericsson Review 76(1999):4, pp.174-181.

4 Madfors, M.: Radio Access Network andQuality of Service in Future Wireless Net-works. Future Telecom Forum 99, Beijing,China, December 1999.

5 Casner, S. and Jacobson, V.: CompressingIP/UDP/RTP Headers for Low-Speed SerialLinks, RFC 2508, February 1999.

6 Jonsson, L.-E., Degermark, M., Hannu, H.and Svanbro, K.: Robust checksum-basedheader compression (ROCCO). Internet-Draft (work in progress) draft-jonsson-robust-hc-03.txt. Ericsson Research, Janu-ary 2000.

REFERENCES

the challenges of voice-over-ip-over-wireless - ericsson · pdf file20 ericsson review no. 1,...

Documents