AN END-TO-END QoS FRAMEWORK FOR MULTIMEDIA STREAMING SERVICES IN 3G NETWORKS

H. Montes, G. Gmez and D. Fernndez Nokia Networks, IP Mobility Networks, NSR-Mlaga SCT, P.T.A, Mlaga (Spain) & Dpto. Ingeniera de Comunicaciones, E.T.S.I. Telecomunicacin, Universidad de Mlaga E-mail: {ext-hector.montes, ext-gerardo.gomez, ext-daniel.fernandez}@nokia.com

Abstract - In this paper, an end-to-end Quality of Service (QoS) framework for streaming services in 3G mobile networks is considered. A solution based on a Public Land Mobile Network (PLMN) hosted multimedia streaming service is studied to avoid accessing through an external IP packet data network (IP-PDN) to streaming services. Under this scenario, the interaction between UMTS and IETF protocols and mechanisms for a streaming session is analyzed. As a part of the proposed framework, a Radio Resource Management (RRM) scheme for the support of streaming services over GSM/EDGE Radio Access Network (GERAN) based third generation (3G) mobile networks is presented. Simulation results on the performance of this RRM scheme show the efficiency for the transmission of streaming services over EGPRS networks. Keywords - Multimedia Streaming, Quality of Service, 3G Mobile Networks, Radio Resource Management. I. INTRODUCTION Multimedia streaming services are receiving considerable interest in the mobile network business. Supporting reliable real time services is a decisive aspect for the increasing migration towards packet based telephony networks. For UMTS, deploying an all-IP architecture is a promising standardization trend due to the convergence between IP technologies and telephony services [1]. Streaming services are also technically supported over evolving 2G and 3G wireless networks, thus streaming clients will soon be implemented in advanced wireless communication devices. Inside this new group of services, there are a variety of applications (e.g. audio and video on demand) with different traffic source statistical characteristics [2]. For audio streaming, the generated traffic is rather non-bursty whereas video traffic has a more bursty nature. One key issue is how mobile networks can support this kind of services. In these "Pre-All-IP" service cases the used radio bearers can be chosen from either circuit switched (CS) or packet switched (PS) bearer set. PS bearers provide more trunking gain and better resources utilization while CS bearers offer better performance for those services with stringent delay requirements. All the multimedia services are mainly characterized, from network point of view, by the need to guarantee certain QoS requirements.

Providing end-to-end QoS for multimedia streaming services implies the harmonized interworking between protocols and mechanisms specified by IETF and 3GPP [3]. This harmonization also covers the QoS provisioning within the different 3G network subdomains and the external IPPDN through which the user accesses to the service. In this paper, the end-to-end QoS management of streaming services in 3G mobile networks is considered. Particularly, the possibility of employing a PLMN hosted multimedia streaming service is studied to avoid accessing through an external IP-PDN to streaming services. With this solution, the mobile operator hosts a streaming server or a proxy server within the PLMN, allowing the provision of sufficient QoS to users of wireless streaming terminals. The presented analysis of the multimedia streaming session is chronologically divided in two phases: service activation and service utilization. The transmission of multimedia services with stringent QoS requirements implies a conversion of the existing GSM/EDGE network into real 3G networks [4]. This conversion entails the enhancement of the radio interface, with the necessity of a complete RRM functionality, aware of the QoS requirements of the new multimedia services. II. OVERALL SCENARIO DESCRIPTION A. Description of the service: Multimedia Streaming A generic framework for a typical multimedia streaming service consists of content creation and retrieval system. When providing a streaming service, a media server opens a connection to the client terminal and begins to stream the media to the client at approximately the playout rate. During the media receiving, the client plays the media with small delay or no delay at all. This technique does not only free up limited terminal memory, but it also allows the media to be sent live to clients as the media event happens. The user needs a player, which is a special program that decompresses and sends video data to the display and audio data to the speakers. This client application must be able to control the streaming flows (control plane) and manage the media flows (user plane). The client also has to interface

0-7803-7589-0/02/$17.00 2002 IEEE

PIMRC 2002

with the underlying transport network technology, its specific protocols and data bearers dedicated to the service. The 3GPP PS multimedia streaming service is being standardized [5] based on control and transport IETF protocols, such as Real-Time Streaming Protocol (RTSP), Real-Time Transport Protocol (RTP) and Session Description Protocol (SDP), as fig. 1 shows. RTSP is an application level client-server protocol, used to control the delivery of real-time streaming data [6]. Both RTP and its related control protocol called Real-time Transport Control Protocol (RTCP) convey media data flows over User Datagram Protocol (UDP) [7]. RTP carries data with real time requirements while RTCP conveys information of the participants and monitors the quality of the RTP session.
Signaling Media

the GGSN is connected to a RTSP proxy, which is also connected to the Streaming Server. Therefore, no external IP-PDN is involved in providing the streaming service. III. SERVICE ACTIVATION This phase is described at three levels of abstraction. In the first place, the service activation procedure from UE viewpoint is briefly outlined. Secondly, the signaling interchanges between application entities by using RTSP in order to establish the session are presented, as well as the media codec negotiation. Finally, all the signaling messages and mechanisms in lower layers (i.e. UMTS protocols) are explained in details. A. User Equipment Operation The service activation from user viewpoint can be described as follows. At first, user initiates the streaming client application, which connects to the UMTS network by using a socket Application Program Interface (API). The application requests a primary Packet Data Protocol (PDP) context which is opened to a specific access point with interactive UMTS traffic class and other suitable UMTS QoS release 99 parameters. A socket is opened for RTSP negotiation and it is tied to the interactive PDP context. The user then selects an audio streaming content. The application activates a streaming handler to take care of the streaming content. When the RTSP negotiation reaches the SETUP phase, a secondary PDP context is activated with QoS parameters suitable for audio streaming (RTP traffic) and for transport signaling (RTCP traffic). The RTP flow will start running through the streaming PDP context. New sockets are opened for RTP and RTCP traffic and they are tied to the streaming secondary PDP context. B. Application Layer Signaling The application layer signaling interchange between the UE and the streaming server is outlined in fig. 3. A primary PDP context is activated for the RTSP signaling between the terminal and the streaming server. By means of RTSP messages, information about the encoding of the media and the corresponding UDP port number is exchanged. The SDP message [10] describes the streaming media the UE is about to receive. It should be noted that RTSP specification defined by IETF [6] does not mandate the use of the DESCRIBE method for this media initialization phase. However, in order to function properly, any RTSP-based system must receive the description of the media one way or other. The 3GPP standard, defining the protocols and codecs for the transparent end-to-end PS streaming service in 3G networks [5], mandates the use of the DESCRIBE method for the conveyance of the media description. Afterwards, a secondary PDP context for the streaming media (RTP and RTCP flows) is activated. When resources for the media are successfully reserved, the UE sends the

SDP
Application layer Transport layer Internet layer Telecom layer

Audio/video RTP RTCP UDP IP

RTSP TCP

UMTS protocol stack

Fig. 1. Protocol stack for signaling and media flows of streaming services B. Architecture of the Mobile Network A general overview of the considered UMTS network architecture is depicted in fig. 2. Detailed description of the entities, interfaces and protocols in UMTS is given in [8].
PLM N UMTS O&M BTS RAN
Iu Uu

RTSP Proxy (PCF) HLR SGSN

App. Server GGSN

SMLC

Packet CN

RAN GW BTS

UE

Fig. 2. End-to-end network architecture In addition to the User Equipment (UE), main entities involved in QoS management are: UMTS Terrestrial Radio Access Network (UTRAN) and GERAN, Serving GPRS Support Node (SGSN), Home Location Register (HLR), Gateway GPRS Support Node (GGSN), and Application Server and RTSP Proxy. End-to-end QoS in the UMTS Release 5 is based on the IP bearer service concept, which consists of the extension of the UMTS BS defined in the UMTS release 1999 [9] to take into account the QoS in the external IP-PDN. In our model,

streaming server a PLAY request in order to start to receive the stream. The server sends the stream in form of a RTP flow. Likewise, RTCP traffic is sent for QoS control of the corresponding RTP data flow.
UE GGSN RTSP server

(GTP-c) to indicate the GGSN that a new PDP context has to be created. As the primary PDP context is not intended for real-time traffic, no resource reservations are needed in the Core Network (CN). The GGSN creates the primary PDP context based on similar admission criteria to those employed by the SGSN. Thereafter, the GGSN notifies the SGSN that the primary PDP context for RTSP has been successfully created and the SGSN sends a SM message to the application in the UE. Once the streaming server accepts the RTSP connection request, the UE triggers a secondary PDP context activation procedure, used for both unidirectional RTP traffic and bidirectional RTCP traffic. The UE translates user data application requirements into UMTS QoS profile for streaming class. Thus, table I shows an example of QoS profile for both RTP and RTCP data traffic. QoS parameters requested for the PDP context take into account the full RTP/UDP/IP headers. Thus, no UDP/IP header compression is assumed in IP level when requesting QoS. Table 1 Proposed QoS Profile for RTP traffic
QoS99 Parameter name Traffic Class Maximum bitrate for uplink Maximum bitrate for downlink Maximum SDU size Delivery of erroneous SDUs SDU error ratio Transfer Delay Guaranteed bitrate for uplink Guaranteed bitrate for downlink Parameter value Streaming 0 kbps 90 kbps 1060 bytes No 10-2 2s 0 kbps 72 kbps

Primary PDP context activation (RTSP signaling)

Secondary PDP context activation (one for RTP traffic and another for RTCP)
[RTSP] PLAY [RTSP] 200 OK

RTP and RTCP flows

Fig. 3. RTSP session initiation procedure in UMTS network C. UMTS Signaling Procedures Once the application level signaling procedure is presented, further insight about the UMTS signaling is provided. In UMTS, all signaling associated with service session establishment is carried out by control plane through different QoS management functions (i.e. bearer service management, subscription and admission). Control plane functions are distributed along several protocol layers. In the first place, a primary PDP Context is activated, as aforementioned, for RTSP signaling using UMTS interactive traffic class [3]. QoS requirements of the application in the UE are mapped to UMTS QoS attributes. The UMTS QoS interactive traffic class has a priority based handling instead of guarantees based handling, being reliability requirement the target in this case. Since primary PDP context is used for RTSP signaling, a UMTS QoS profile with interactive traffic class, high priority and low error rate is appropriate. A Session Management (SM) protocol message from the UE to the SGSN initiates the PDP context activation procedure. After the SGSN has validated the service for that user by querying HLR, local admission control is performed (e.g. based on the state of buffers and CPU load). Then, the SGSN maps the UMTS QoS attributes to Radio Access Bearer (RAB) QoS attributes and triggers a RAB assignment procedure in the RAN by using the Radio Access Network Application Protocol (RANAP). In the RAN, admission control is basically based on the availability of radio resources. Once a new PDP context is accepted, RAB attributes are mapped to Radio Bearer (RB) parameters used in physical and link layers (e.g. spreading codes, transmission modes, etc.). A RB according to these parameters is established and it is reported to the SGSN, which employs GPRS Tunneling Protocol for Control Plane

For a given Service Data Unit (SDU) Error Ratio, the larger the SDU size, the smaller the radio block error rate, meaning the reliability requirements for radio link are more demanding. Since a more protective coding scheme must be used, the bitrate is lower (for the same radio blocks sent), implying larger delay. Therefore, maximum SDU size should be commonly considered with the required SDU error ratio. From network viewpoint, smaller SDUs allow easier compliance to reliability requirements by relaxing the radio link adaptation. Moreover, a trade off between reliability and delay relevancy should be found. This compromise needs to be communicated from UE application to the network or the application criteria for SDU size should be always conservative. Once the QoS profile is derived, the secondary PDP context is activated. This procedure is quite similar to the above explained for the primary PDP context. The main differences in the secondary PDP context activation procedure are located in the RAN (UTRAN and GERAN) admission control.

IV. SERVICE UTILIZATION Once the connection is established, the RTP data flow needs an appropriate QoS provisioning. In the IP transport domain, DiffServ mechanism [11] is employed (i.e. both in Core Network and between GGSN and RTSP Server, being GGSN responsible for mapping UMTS QoS parameters to DiffServ parameters). DiffServ mechanism is based on different Per Hop Behaviors (PHB). Each PHB consists of the rules used to treat packets in specific ways inside the network. More specifically, PHB denotes a combination of forwarding, classification, scheduling and drop behaviors at each hop. For streaming traffic, two groups of PHB can be applied: Expedited Forwarding (EF) or Assured Forwarding (AF). EF PHB target is to provide tools to build a low loss, low latency, low jitter, and assured bandwidth end-to-end service within the DiffServ domain, with the drawback of the complexity it introduces in the system. Due to the undemanding QoS requirements of streaming services, mainly in comparison with other real-time traffic like VoIP services, AF PHB can be used. Inside AF PHB group there is a number of PHB delay classes, each of them with a number of drop precedence levels. For streaming traffic the highest priority should be used. In the radio domain there are basically two options for conveying the data flow: CS bearer or PS bearer. The CS approach has the inherent drawback of the waste of resources, mainly in case of bursty traffic, as is the case of streaming traffic [2]. In other words, if resources are shared, trunking gain is obtained. The challenge comes from the need of guaranteeing certain bandwidth on shared channels whose radio link capacity is continuously varying, so enhanced RRM mechanisms are necessary for that purpose. The Enhanced QoS (EQoS) framework is a complete RRM system designed for the transmission of Guaranteed Bitrate (GBR) services (as streaming services) over EGPRS networks [12]. EQoS scheme consists of different functionalities: in the set up phase, an Admission Control and Channel Allocation scheme is used to accept/reject new allocation requests, according to the QoS requirements and the available radio resources; a Packet Scheduler, with the goal of providing to each allocated connection the needed air transmission time in order to guarantee its bitrate requirements; and a Quality Control functionality, in charge of monitoring whether the provided QoS to each connection is in accordance with the negotiated QoS. Main characteristics of EQoS feature are: use of acknowledged Radio Link Control (RLC) mode of operation, due to undemanding delay requirements; and multiplexing of several GBR users over the same physical channel, by means of the use of shared Medium Access Control (MAC) mode. This use of shared channels leads to a better trunking efficiency, when comparing with dedicated

channels. The reason for that trunking gain is the bursty nature of the incoming streaming traffic. V. EQoS PERFORMANCE VALIDATION A. Simulation Environment The performance of the EQoS framework has been evaluated by means of dynamic system level simulations. All the simulations have been made over a macro-cellular scenario, formed by 75 cells in a regular three-sectored hexagonal layout. Users are uniformly distributed through the system coverage area. 3 km/h user mobility has been employed in the simulations. Only one traffic TRX per cell has been used, with a frequency reuse 3/9. GSM air interface is modeled at burst level, including multipath fading, shadowing, distance attenuation, and both co-channel and adjacent channel interference. RLC/MAC protocols are implemented according to the specifications, including selective ARQ, retransmission bitmaps, polling and both transmitter and receiver sliding windows. Streaming traffic source is modeled with an autoregressive model for the packet size distribution with random variable following a Gamma distribution, and a deterministic packets interarrival time. The video codec is generating traffic at 16, 32 and 64 kbps as mean bitrate. Non-real time traffic is modeled trying to use all the remaining capacity not used by streaming traffic. B. Simulation Results The EQoS RRM is designed to guarantee the required bitrate for different GBR requirements. However, the GBR requirement determines the maximum load that the system is able to support. The system is clearly able to manage more users with lower bitrate, since the lower the guaranteed bitrate, the higher the statistical multiplexing gain, i.e. low bitrate involves higher number of connections multiplexed over the same timeslot, as shown in fig. 4. Other advantage of the use of shared channels is that nonreal time traffic can be transmitted over the same channels, making use of the remaining capacity left by streaming connections. Obviously, the higher the streaming capacity supported by the network, the lower the non-real time capacity multiplexed with it, as it is also depicted in fig. 4. As above stated, streaming traffic does not have demanding delay requirements. However, in fig. 5, transfer delay distribution for IP packets at link level is shown. It can be seen how transfer delay requirements are fulfilled. In addition, this figure provides useful information for the dimensioning of compensating buffers used in application layer for streaming services, for EGPRS mobile networks.

140

Spectral Efficiency [kbit/s/Cell/Mhz]

120 100 80 60 40 20 0 non-real Time Streaming

has also been noted. EQoS RRM scheme to provide guaranteed bitrate services is depicted. Simulation results show that the proposed scheme is behaving as expected. Both guaranteed bitrate and transfer delay requirements for multimedia streaming traffic are fulfilled. ACKNOWLEDGEMENTS This work has been performed as part of the cooperation agreement between Nokia and the University of Malaga. This agreement is partially supported by the Program to promote technical research (Programa de Fomento de la Investigacin Tcnica, PROFIT) of the Spanish Ministry of Science and Technology. REFERENCES
GBR = 16 kbps GBR = 32 kbps GBR = 64 kbps

GBR = 16 kbps

GBR = 32 kbps

GBR = 64 kbps

Fig. 4. Streaming and non-real time Spectral Efficiency

0.9

PCDF

0.8

0.7

[1] 3GPP; TSG Services and Systems Aspects, "Architecture for an all IP network", TR 23.922 v1.0.0, October 1999 [2] A. Mena and J. Heidemann, "An Empirical Study of Real Audio Traffic", in Proceedings of Conference IEEE INFOCOM 2000, vol. 1, pp. 101-110

0.6

0.5 0 0.25 0.5 0.75 1 1.25 1.5

Frame Delay [sec]

[3] 3GPP; TSG Services and Systems Aspects, "QoS concept and architecture", TS 23.107 v5.3.0, 2002 [4] T. Halonen, J. Romero, J. Melero, GSM, GPRS and EDGE Performance, Evolution Towards 3G/UMTS, Wiley & Sons, 2002 [5] 3GPP; TSG Services and Systems Aspects, "Transparent End-to-End PS Streaming Services (PSS); Protocols and Codecs", Rel4, TR 26.234 v4.2.0, 2001 [6] H. Schulzrinne, A. Rao, and R. Lanphier, "Real Time Streaming Protocol (RTSP)", IETF RFC 2326, 1998 [7] H. Schulzrinne, S. Casner, R. Frederick and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", IETF RFC 1889, 1996 [8] 3GPP; TSG Services and Systems Aspects, "General Packet Radio Service (GPRS); service description"; stage 2, TS 23.060 v4.3.0, 2002 [9] 3GPP; TSG Services and Systems Aspects, "End-to-end QoS concept and architecture", TS 23.207 v5.2.0, 2002 [10] M. Handley and V. Jacobson, "SDP: Session Description Protocol", IETF RFC 2327, 1998 [11] S. Blake, D. Black, M. Carlson, Z. Wang and W. Weiss, "Architecture for Differentiated Services", IETF RFC 2475, 1998 [12] D. Fernndez, H. Montes, "An Enhanced QoS method for guaranteed bitrate service over Shared Channels in (E)GPRS", in Proceedings of 54th IEEE Vehicular Technology Conference, Alabama, spring 2002

Fig. 5. Frame Transfer Delay Distribution In case connections with lower delay requirements are transmitted through EGPRS networks, maximum transfer delay can be controled by means of the Admission Control, restricting the number of multiplexed connections per timeslot. In fig. 5, transfer delay for a GBR requirement of 64 kbps is lower than for a GBR requirement of 16 kbps. The reason is that number of streaming connections multiplexed over the same timeslot in the 64 kbps case is much lower than in the 16 kbps case. By means of reducing the number of 16 kbps connections accepted in the system, transfer delay can be reduced to the desired requirement. VI. CONCLUSIONS Since supporting reliable real time services is a decisive aspect for packet based telephony networks, an end-to-end QoS framework for streaming services in 3G mobile networks is considered. This paper addresses a solution based on a PLMN-hosted multimedia streaming service. Service activation is described at three levels of abstraction: from UE viewpoint, RSTP signaling interchanges between application entities and UMTS signaling procedures. Provisioning of streaming services over 3G mobile networks based on GERAN radio technology has been also tackled in this paper. Results from traffic behaviour analysis have shown the convenience of using PS bearers. In shared channels, the challenge of assuring capacity for such traffic