Hybrid Multiplexing of Voice and Data over IP on GSM/GPRS Cellular

David E Vannucci & Peter J Chitamu Centre for Telecommunications Access and Services School of Electrical and Information Engineering University of the Witwatersrand, Johannesburg, South Africa {d.vannucci, p.chitamu}@ee.wits.ac.za

Abstract This paper investigates the benets of using voice over IP on GSM/GPRS cellular. The work is motivated by rst the complexity of having both circuit switched and packet switched connectivity on GSM/GPRS cellular and secondly by the fact that an exclusively packet based access on GSM cellular has the potential to handle more voice calls and data trafc by deploying suitably chosen hybrid multiplexing schemes that exploit the characteristics of both voice and data trafc. Index Termsvoice, VoIP, GPRS, GSM.

I. I NTRODUCTION HIS paper reports work in progress on the statistical multiplexing gains obtained by transporting voice over IP over GPRS as opposed to transporting voice in GSM circuit switched channels. Current second generation mobile systems such as GSM do not provide an effective means of offering high data rate multimedia services, with current data services being provided by a circuit switched connection. GSM networks are being upgraded to provide packet mode connection using General Packet Radio Service (GPRS) which provides data rates of up to 117.2Kbps [1]. GPRS is a packet switching service that was developed to facilitate access to IP based services by dynamically allocating radio bandwidth to users [1], [2]. Packet switching allows efcient sharing of network resources among users to reduce costs and increase the exibility of services through statistical multiplexing. GPRS is intended to support bursty data transactions such as Internet and WAP services [3], using GSM equipment, but is poorly suited to constant bit rate services such as video streaming [2]. Thus GPRS is better suited to frequent, small data packets up to 500 bytes and irregular packets up to a few kilobytes [2]. GPRS provides a migration path to the third generation network, providing a packet data mechanism using the existing equipment in the GSM architecture, while adding new elements interfacing to an IP based network [3], [4]. We propose that the utilisation of the air interface can be improved, if voice and data are transmitted using packet switching. The potential increase in usage can provide the
The Centre is supported by Telkom SA Limited, Siemens Telecommunications and the THRIP Programme of the Department of Trade and Industry.

cellular operator with benets such as: increased number of simultaneous conversations, or the same number of cellular conversations while providing a greater data bandwidth. The introduction of a packet only infrastructure can provide opportunities for equipment reuse within the GPRS architecture for applications such as xed wireless. We will introduce the concepts of General Packet Radio Service and then discuss problems associated with transmitting voice over GPRS in general. With the sharing of radio resources between voice and data services in the GSM/GPRS system, the air interface is identied to be the performance bottleneck. We propose a strategy with regard to the gains achievable by multiplexing of voice and data packets over the GSM/GPRS air interface. II. BACKGROUND TO GPRS This section discusses the GSM architecture and how the transition to GPRS is achieved. The main intention of integrating GPRS into GSM is to increase the number of connections per bearer by improving the usage of the available physical channels [2]. Thus if voice is transmitted over data channels, the utilisation of the channels can be improved further. GPRS has been standardised by the ETSI as part of the GSM phase 2+ development. GPRS usually shares the channel with GSM circuit-switched voice with one or two channels dedicated to GPRS and others being used when free channels are available [5]. Several codecs which can be used for Voice over IP (VoIP) exist, some of which support silence suppression such as G.723 and G.729 [6]. These codecs produce packets of thirty to a hundred bytes depending on the codec. With header compression and the fact that silence occurs for roughly 60 percent of the conversation the statistical multiplexing gains of the GPRS channel can be high, [7]. GPRS uses an IP backbone network with standard packet network nodes such as routers, DNS servers and rewalls, and introduced two new elements to the GSM infrastructure: the Serving GPRS Support Node (SGSN) and the Gateway GPRS Support Node (GGSN) [2], [3], [5]. The architecture of a GSM/GPRS network is shown in gure 1. Connection to the integrated services digital network (ISDN) and public switched telephone network (PSTN) is through the Gateway Mobile-services Switching Centre (GMSC).

ISDN/PSTN

GMSC

MSC

Um Gs
BTS BSC

General
Registers

Gr
IP Back bone

Gb Gn
SGSN

Gn
GGSN

BSC BTS Firewall PLMN Backbone

Gp
Border Gateway

Gi
Firewall

Internet

Fig. 1
GSM / GPRS ARCHITECTURE SHOWING CONNECTION TO PSTN/I NTERNET /ISDN AND OTHER GPRS NETWORKS

The GGSN serves as the interface between the GPRS core and the external public data network or other mobile networks. The GGSN is responsible for routing of mobile subscribers via the SGSN. Accounting of resources usage by user ows and mobility management is also handled by the GGSN. The SGSN handles the terminal mobility and authentication functions and represents the GPRS switching centre by analogy to the GSM MSC. The SGSN maps packet data addresses to the International Mobile Subscriber Identity and is responsible for routing within the IP network. Resource management as well as encryption is handled by the SGSN. Border gateways are used to interconnect GPRS IP backbones and rewalls prevent unauthorised access. GPRS employs Gaussian Minimum Shift Keying modulation with four encoding schemes, CS-1 to CS-4. Coding Scheme 1 offers the highest level of data protection. CS-1, CS-2, and CS-3 use convolutional codes and block-check sequences of varying complexity [4]. CS-4 offers no error protection of user content and provides only error detection functionality [4], [8]. See table I for encoding schemes and their associated data rates.
TABLE I
GPRS ENCODING S CHEMES AND ASSOCIATED DATA RATES [2], [8]

Check Sequence (BCS) trailer. The coded bits are the size of the frame after convolutional encoding. Puncturing is used to shorten the convolutional code in coding scheme 2 and 3. One TDMA frame, as shown in gure 2, represents eight physical channels to be shared by the GSM and GPRS. Sharing strategies are implemented by the network operator or equipment vendor to allocate the GPRS resources according to set policies. The physical channels used in GPRS for packet data trafc are known as Packet Data Channels (PDCs). Logical PDCs are mapped onto physical channels using a cyclically recurring multiframe structure [2]. Six radio blocks, each containing 8 PDCs, are mapped into one 26-multiframe, and two 26-multiframes are assembled onto one GPRS 52-multiframe. Thus a 52-multiframe represents one physical GPRS channel consisting of twelve radio blocks and one idle block, each block comprising four radio bursts distributed on timeslots with the same timeslot number in consecutive TDMA frames, as shown in gure 3. III. C HALLENGES OF VO IP OVER GPRS Transmitting voice over IP by using GPRS introduces many challenges that need to be carefully considered. Issues such as header efciency and quality of the air interface affect the successful end-to-end transmission of voice. An adequate Quality of Service (QoS) is required to transmit voice over wireless links, the main requirements being delay and bit error rate [9]. GPRS QoS proles are currently limited to only one QoS prole per Packet Data Protocol address, with GPRS phase one not allowing QoS re-negotiation to be initiated by the MS or GGSN [3], [10]. GPRS was designed for best effort trafc and as such the specication of parameters such as delay constraints are often un-implementable on an end-to-end basis. Other issues such as optimisation for real time handovers and xed bandwidth reservations are also not well supported

Scheme

Code Rate

Radio Block1

Coded Bits

Punctured Bits 0 132 220 0

CS-1 1/2 181 456 CS-2 2/3 268 588 CS-3 3/4 312 676 CS-4 1 428 456 1 Excludes USF and BCS bits

Data Rate (kbps) 9.05 13.4 15.6 21.4

The radio block presented is without the multiplexing control Uplink State Flag (USF) header and error detection Block

Radio block

52-Multiframe
0 5 10 20 40 52

Fig. 2
S TRUCTURE OF A 52- MULTIFRAME

APP TCP / UDP IP SNDCP LLC RLC MAC GSM RF

N-PDU

M S

SNDCP/LLC BSS BSSGP

SGSN
Fig. 4

GTP

GGSN

SN-DATA PDU LLC Frames RLC Blocks RLC/MAC Blocks TDMA Bursts

T UNNELING WITHIN THE GPRS S UPPORT N ODES

requirements [6]. Optimising the TBF duration is required to efciently utilise the medium. IV. E ND - TO - END S OLUTION This section illustrates why the air interface is the bottleneck. Each base station subsystem (BSS) is virtually connected to a SGSN by means of a connection-less link on the Gb interface. The BSS does not perform ow control in the uplink direction, that is from the BSS to the SGSN [2].Thus the SGSN buffer and link capacity is expected to be dimensioned such that uplink data is not discarded. All GPRS Support nodes (GSNs) are connected to the IP backbone. Encapsulation is used to transfer messages between GSNs using the GPRS Tunneling Protocol (GTP) [10]. Information is transferred between the BSS and SGSN using the Base Station Subsystem GPRS Protocol (BSSGP) as shown in gure 4. IP routing mechanisms are used to transfer the packets to appropriate GSNs. Since any transport mechanism may be used below IP the QoS and bandwidth constraints are assumed to be in the Um and Gb interfaces - illustrated in gures 1 and 4. V. P ROPOSED W ORK In order to evaluate the statistical multiplexing gains obtained by transporting voice over IP using GPRS, research will concentrate on the following aspects: The bottleneck of the system is seen as the Um and Gb interfaces, thus the focus of the research is on the air interface and SGSN. Control signalling such as that used by H.323 or SIP before and after calls will not be considered, instead focus will lie on the voice coded RTP data. Different call scenarios will be considered such as from the MS to another MS, MS to PSTN, MS to Internet phone. Initial modeling will consist of a basic model such as that shown in gure 5, with all sources being queued.

Fig. 3
GPRS P ROTOCOL S TACK AND P ROTOCOL DATA U NITS ASSOCIATED WITH EACH LAYER [8]

by GPRS. Thus a revision of the access control of the GPRS can improve the partitioning strategies by introducing variable bandwidth requirements between voice and data communication. Partitioning strategies such as complete sharing and bit rate sharing provide a high link efciency, with both strategies being able to use all available cell capacity [1]. The two competing voice over IP control protocols the ITU-T H.323 and the IETF Session Initiation Protocol (SIP) both use the Real-time Transport Protocol (RTP) to transport the voice codec stream [11]. The voice packets generated by the speech coder are encapsulated in RTP, UDP and IP before being passed to the GPRS network. A speech frame has a RTP / UDP / IP header size of 40 bytes and a payload of 15 30 bytes depending on the voice codec [12]. If IPv6 is used the header size is increased to 60 bytes [12]. Due to the high overhead, header compression is required with schemes such as compressed RTP and the favoured Robust Checksum based header Compression [12]. The GPRS Medium Access Control Layer is responsible for providing efcient multiplexing of data and control signalling on the uplink and downlink. MAC procedures include the provision of Temporary Block Flow (TBF) which allow pointto-point transfer of data in one cell between the base station subsystem (BSS) and the mobile station (MS) [2]. The BSS consists of both the base transceiver station (BTS) and the base station controller (BSC). TBF releases and re-establishments during the transmission of voice packets impact on the delay of voice packets, and can be adjusted to meet end-to-end delay

a1,m1-1 aN,mN-1 Sources

Queue
Fig. 5

Output

C ONTINUOUS TIME MODEL WITH MULTIPLE SOURCES BEING QUEUED

The voice trafc will have an average arrival rate of a1 and holding time of m1 while data will have an average arrival rate of a2 and holding time of m2 . Voice data will be considered time delay sensitive and packet loss insensitive while data would be considered packet loss sensitive and time delay insensitive. An appropriate queueing model for voice and data using GPRS will be developed. Trafc simulation over an area will be initiated according to a uniformly distributed Poisson process, a mean holding time of 120s [11] and mean talk-spurt length of 1.2s [13] will be used for the voice. A possible voice model that can be used is the 3 state Markov model which ignores double talk but models silence better [13]. Results such as the required bit rate per conversation as well as the maximum number of conversations can be deduced. The optimal delay and packet size can be calculated. The capacity at a constant level of quality can be identied in a manner similar to [11]. The encoding scheme can be varied to determine the most suitable level of protection for data, voice, and both data and voice. The effect of various voice codecs can be simulated with a data rate particular to that codec to determine what basic throughput would be generally required. VI. C ONCLUSION This paper has discussed our work in-progress on statistical multiplexing gains possible by transmitting both voice and data in GPRS packet data channels. The fact that GPRS is suited to frequent small data packets and that voice packets are of the order of a hundred bytes suggest that GPRS will be well suited to carrying voice over IP. Considering as much as 60% of a conversation is silence, the multiplexing gains would provide an operator with increased cell capacity, providing a better utilisation of the limited radio resources on the GSM radio interface. We have highlighted issues to consider in the transport of voice over IP using GPRS such as encoding schemes and header compression, and a method of research was proposed to simulate the multiplexing gains. R EFERENCES
[1] Andrew Kennedy, John Mellor, and Khalid Al-Begain, Investigation of partitioning strategies for GPRS over GSM networks, PostGraduate Network, 2001, http://www.cms.livjm.ac.uk/pgnet2001/ papers/akennedy.doc.

[2] Bernhard H Walke, Mobile Radio Networks Networking, Protocols and Trafc Performance, John Wiley and Sons, New York, USA, second edition, 2002. [3] Rajeev Koodli, Supporting packet-data QoS in next-generation cellular networks, IEEE Communications Magazine, pp. 180188, February 2001, http://www.mprg.org/people/buehrer/ research/docs/Supporting%20Packet-Data%20QoS% 20in%20Next-Generation%20Cellular%20Networks.pdf. [4] Simon N Fabri, Stewart Worrall, Abdul Sadka, and Ahmet Kondoz, Real-time video communications over GPRS, London, March 2000, IEE 3G Mobile Communication Technologies, pp. 426430, http://www.ee.surrey.ac.uk/Personal/S. Worrall/Publications/3g2000_fabri.pdf. [5] H akan Granbohm and Joakim Wiklund, GPRSgeneral packet radio service, Ericsson Review, vol. 2, pp. 8288, 1999, http://www.ericsson.com/about/publications/ review/1999_02/files/1999024.pdf. [6] Nachiappan Nachiappan and Fredrik Sjoqvist, Survey of voice over IP (VoIP), http://www.stanford.edu/nachi/VoIP_ Survey.doc, May 2001. [7] Jin Wang, Peter J McCann, Patvardhana B Gorrepati, and Chung-Zin Liu, Wireless voice-over-IP and implications for third-generation network design, Bell Labs Technical Journal, vol. 3, no. 3, pp. 7997, July-September 1998, http://www.lucent.com/minds/ techjournal/pdf/jul-sep1998/paper07.pdf. [8] Jaana Laiho, Achim Wacker, and Tom` as Novosad, Radio Network Planning and Optimisation for UMTS, John Wiley and Sons, rst edition, January 2002. [9] Hemant M. Chaskar and Upamanyu Madhow, Statistical multiplexing and QoS provisioning for real-time trafc on wireless downlinks, IEEE Journal on Selected Areas in Communications, vol. 19, no. 2, pp. 347, February 2001, http://citeseer.nj.nec.com/465875. html. [10] Christian Bettstetter, Hans-J org V ogel, and J org Ebersp acher, GSM phase 2+ general packet radio service GPRS: Architecture, protocols, and air interface, IEEE Communications Surveys, vol. 2, no. 3, pp. 214, 1999, http://www.comsoc.org/livepubs/surveys/ public/3q99issue/bettstetter.html. [11] G oran AP Eriksson, Birgitta Olin, Krister Svanbro, and Dalibor Turina, The challenges of voice-over-IP-over-wireless, Erricson Review, vol. 1, pp. 2030, 2000, http://www.ericsson.com/about/ publications/review/2000_01/article96.shtml. [12] A. Cellatoglu, S. Fabri, S. Worrall, A.H. Sadka, and A.M. Kondoz, Robust header compression for real-time services in cellular networks, 2001, pp. 124128, IEE 3G Mobile Communication Technologies, http://www.ee.surrey.ac.uk/Personal/S. Worrall/Publications/3g2001_header.pdf. [13] C Moorhead, V J Hardman, and P M Lane, Resource allocation schemes to provide QoS for VoIP over GPRS, in 3G Mobile Communication Technologies, Conference Publication No. 471. 2000, pp. 138142, IEE, ISSN 05379989.

David E. Vannucci holds a BSc (Eng.) degree in Electrical Engineering from the University of the Witwatersrand. He is presently pursuing his MSc(Eng.) degree at the University of the Witwatersrand.

Peter J. Chitamu received a BSc(Eng.)(Hons.) in Electrical Engineering from the University of Dar es Salaam and the MSc(Telematics) and Ph.D. from the University of Surrey and Southampton respectively. He is presently Senior Lecturer in the School of Electrical and Information Engineering at the University of the Witwatersrand.