A Network Controlled QoS Model over the 3GPP Evolved Packet Core

*Marius Iulian Corici, *+Fabricio Carvalho de Gouveia, *+Thomas Magedanz * Fraunhofer Institute FOKUS, Kaiserin-Augusta-Alee 31, D-10589, Berlin, Germany *+ Technical University of Berlin, Franklinstr. 28-29, D-10587, Berlin, Germany {marius-iulian.corici@fokus.fraunhofer.de, fabricio.gouveia@fokus.fraunhofer.de, thomas.magedanz@fokus.fraunhofer.de}

1. Introduction
Present network devices are able to access services over various technologies e.g. WLAN, UMTS, WiMAX, LTE etc. and to exchange data simultaneously over them, as depicted in Figure 1. In order to optimize the service continuity in this environment with multiple heterogeneous access networks, various handover mechanisms are currently studied (e.g. IEEE 802.21 Media Independent Handover [1]). They concentrate especially on the seamless quality of the sessions, considering that the network where the user is relocated after the handover has the necessary resources to sustain the service continuity.

Figure 1: Typical Mobile Network Connectivity

However the user end-points decide to which network to connect, this leading to a general tendency of over-saturating the access networks where the cost is minimal. In this case the network provider can not offer the required quality of service for the connected users, even though it could have provided services over other access networks. Being agnostic to the momentary load of the networks, a multi-card user end-point will choose its network selection based on the user preferences, on the knowledge gained from previous

connections and on the static policies introduced by the network operator in the devices. Therefore the network binding and session parameters decided without taking into consideration the momentary parameters of the network. In a worse case scenario, the user end-point connects to one network only to find that its requirements are not supported. Thus, for receiving the required service, it has to connect to another network, where this problem could repeat. At this moment, the provisioning of resources is done by the user end-point. When a service is started, the user end-point has to send some messages in order to negotiate the parameters of the session and to reserve resources according to them in order to ensure the quality of the data transmission. If the service can be provided at different levels of resource requirements, the reservation has to be repeated until one of the levels is satisfied, which can be extremely time consuming over some of the access technologies. To remedy this shortcomings of the user end-point oriented QoS architecture, a new model is proposed. It considers that the resource reservation is triggered not by the client directly, but by a network entity to which the service was signaled. The decision on the level of resources to be reserved is done by passing through a set of filters, in a policy oriented manner and with respect to the momentary resources available on the access networks. The main focus is on the scenarios of a single operator able to offer services over multiple access technologies. First a state-of-the-art on QoS mechanisms in 3GPP is done. The existing architectural designs are presented from a Quality of Service perspective, than the improvements are asserted. The conclusions are drawn from the comparison of the existing infrastructures and the one here proposed.

2. QoS and Architecture State-of-the-Art

The IP Multimedia System (IMS) [10] represents today the global service delivery platform. The IMS is a complete signaling framework, able to integrate different types of services in a unified manner as seen from the users perspective, using as signaling protocol the Session Initiation Protocol (SIP) [5]. The IMS structure also enables the connectivity of devices using different access networks in a unified manner [12], reducing the management cost of the operators that deploy multiple types of access technologies. By its ability to integrate multiple services as

application servers, the IMS enables various services to be defined and deployed in a fast and flexible manner. The Open IMS Core developed at FOKUS [11] serves as the signaling platform for the architecture here presented. It contains the structures necessary for user registration and user reachability. The user end-point (or User Equipment UE) access to the network is done in a network controlled manner. Also access to various services, both local to the home domain of the operator and remote to other parties is done on a network technology agnostic level, offering an interface easy to integrate in various user end-point devices. The 3GPP IMS has adopted a Policy based approach for QoS provisioning [4]. Policy based networking allows a dynamic and automated control of network resources by the operator, where resource allocation decisions are done based on session information and local policies, which define the expected behavior of the network. High level policies are specified without interfering with IP-CAN specific management. A critical component in this architecture is a logical entity that has a northbound interface to the signaling plane and a southbound interface to the bearer plane to provide linkage and synchronization between the two planes. This system was chosen to manage resources on both the session and bearer levels in a tight but flexible manner decoupling the core network components and procedures from the subtleties of the access technologies. Such an intermediate node has been defined under several standards and given different functional names including Policy Decision Point [6], Resource and Admission Control Function (RACF) [7], or in the 3GPP release 7 and IMS architecture such an element is called a Policy and Charging Rules Function (PCRF) [3]. The Evolved Packet Core (EPC) is a new network architecture defined by 3GPP for the System Architecture Evolution (SAE) ([1], [14]), able to interconnect the signaling with the QoS on the access networks, thus connecting the offered signaled services to the data transmission path in a standardized manner. IMS is integrated at the network layer with the Evolved Packet Core (EPC) [13], [14], which enables the actual resource reservations and the policy control on the different access technologies. As depicted in Figure 2 it comprises of the following entities: Mobility Management Entity (MME) It is responsible for managing and storing UE context and for generating temporary identities to be allocated to the UEs. It also is responsible for checking the authorization whether the UE may connect to the PLMN (Public Land Mobile Network).

Serving Gateway - a router situated in the local domain of the user end-point which has to function transparently for the agnostic internet, and is able to offer local provisioning, like access control, translation from circuit switched to packet switched networks and resource provisioning.

Packet Data Network Gateway - the PDN Gateway has a similar functionality with the Serving Gateway, but it relates only to packet switched networks and it is located in the home domain of the user end-point.

Policy and Charging Rules Function (PCRF) is the entity that interconnects the EPC with the IMS control network. It receives the policies that have to be enforced on the data path, keeps the state of the session and announces the IMS control of the exception cases. One of the goals of EPC is to be able to offer not only home domain, but also visited domain break-through for data. For this reason the PCRF functionality was separated into home domain functionality and visited domain one (hPCRF and vPCRF).

For untrusted network access an evolved packet data gateway was considered (ePDG). It authenticates the users and controls the traffic.

The EPC appeared from the necessity to converge different types of networks, thus having as main goal the transparency of access technology features to the core network of the service provider. At this moment, the EPC is in the early stages of specification and has not defined the mechanisms necessary for an end-to-end communication. However the central component of the architecture is the PCRF, which ensures a backwards compatibility and gives a perspective on the evolution of the 3GPP standards. The Evolved Packet Core is introduced in order to transparently unify the parameters of different technologies, like the UMTS, the 3GPP WLAN, non-3GPP access technologies (e.g. WiMAX) and a future Evolved Radio Access Network called Long Term Evolution (LTE) or Evolved - UMTS Terrestrial Radio Access Network (E-UTRAN). Each of these technologies comes with its own specific access functions. The core itself manages and stores the user end-point context and the user end-point services and network information in a Mobility Management Entity (MME).

Figure 2. Evolved Packet System

An evolved Policy and Charging Resource Function (PCRF) [3] is connected to the core in order to provide the necessary information for controlling the data information received from the user end-points. The overall QoS concept of EPC, and of the new technology it is introducing, the E-UTRAN, is based on the control of the EPC Bearer QoS parameters QoS Class Identifier (QCI) and Allocation and Retention Priority (ARP) [13]. QCI is a scalar (also called Label) that represents the QoS characteristics that the EPC is expected to provide for the Service Data Flow (SDF). This Label is used by routers to access node-specific parameters that control bearer level packet forwarding treatment (e.g. admission thresholds, queue management thresholds), which are specified by the operator. This Label Characteristic is standardized and comprises the following elements: Bearer Type (GBR or Non-GBR) L2 Packet Delay Budget L2 Packet Loss Rate.

The Bearer Type parameter is for checking if the EPC bearer is permanently allocated. The L2 Packet Delay Budget denotes the time that a link layer Service Data Unit (SDU) (e.g., an IP packet) may reside within the link layer between an access node and a UE. The L2 Packet Delay Bucket is meant to support the configuration of scheduling and link layer functions. Last, The L2 Packet Loss

Rate determines the rate of SDUs of non congestion related packet losses. This is for allowing appropriate link layer protocol configurations [13]. The ARP is the parameter used to decide whether a bearer establishment / modification request can be accepted or rejected (in case of resource limitations like available radio capacity for GBR bearers). It is also used to decide which bearers to drop during exceptional resource limitations (e.g. at handover). Each GBR bearer is associated with a Maximum Bit Rate (MBR) QoS parameter. The GBR corresponds to the minimum bit rate to be provided to a GBR bearer and the MBR limits this bit rate in order to avoid packets discarding a rate shaping function. The MBR may be greater than or equal to GBR for a particular GBR bearer. Table 1 depicts the QCI/Label characteristics [13].
Table 1 - Standardized QCI/Label Characteristics (Source [13])

QCI Characteristic 1 (GBR) 2 (GBR) 3 (GBR) 4 (non-GBR) 5 (non-GBR) 6 (non-GBR) 7 (non-GBR)

L2 Packet Delay Budget < 50 ms 50 ms (80 ms) 250 ms Low (~50 ms) Low (~50ms) Medium(~250ms) Medium(~250ms)

L2 Packet Loss Rate High (e.g.10-1) Medium (e.g.10-2) Low (e.g.10-3) e.g. 10-6 e.g. 10-3 e.g. 10-4 e.g. 10-6

Example Services Realtime Gaming VoIMS Streaming IMS signalling Interactive Gaming TCP interactive Preferred TCP bulk data

8 (non-GBR)

High (~500ms)

n.a.

Best effort TCP bulk data

The IEEE 802.11 standard defines two types of services: the distributed coordination function (DCF) which supports delay-insensitive data and the point coordination function which supports delay sensitive transmissions (PCF). The first mechanism is based on the CSMA/CA where the stations compete for the transmission environment. For collisions, it considers a random idle time between some specific boundaries. The second function is a centralized polling-based approach which avoids collisions by polling the mobile nodes individually for transmission. This function offers a better primitive for service differentiation between stations. However, the mechanism leads to a longer time usage of the channel for the mobile nodes with a lower rate transmission. Also stations, being the only ones that know the type of traffic transmitted, do not take part in the decision of the polling, making it a static one in regard to the quality of service required. In order to solve these problems two new functions were introduced in a Quality of Service standard [16]. First one, the Enhanced Distributed Channel Access (EDCA), is similar to DCF, but it considers a dynamic value for the idle time, depending on the type of traffic the terminal has. For this, four types of priority are defined. The packets are classified by the mobile node into one of these categories and queued for transmission, based on the priority level. The second one, the Hybrid Coordination Controlled Access (HCCA) extends the rules of the EDCA by introducing a polling mechanism for the stations, depending on their resource reservation and the time prior reserved. By using the HCCA, the stations are able to pre-reserve a specific access time on a regular period. The HCCA is similar to the QoS solution for UMTS [8]. It also considers a set of priority classes and a channel reservation for each specific terminal. The reservation is done considering two main parameters: the maximum bandwidth and the guaranteed bandwidth for each session flow. This enables to discriminate between different flows and their priorities, not considering the whole mobile node traffic as to have a specific priority. In WiMAX ([15], [16]) each flow has a pre-defined set of parameters and a set of parameters that is communicated when the resources are reserved. This enables a more detailed resource reservation which can be dynamically changed if the user preferences change and the network conditions allow it. The QoS support mechanisms presented here for some specific access technologies do not presume a method of provisioning. Each one of them introduces some control mechanisms and

communication protocols. The scope of this paper is not to evaluate and compare these mechanisms, but to introduce a new perspective on how the resources may be reserved using the already existing mechanisms. The new concept and the complimentary architecture are described in the next sections.

3. Network controlled QoS Concept Description

Each of the forward presented technologies offers a method for provisioning the required resource reservations, user end-point based. This mechanism enables no signaled services to have the requirements fulfilled based on the client expectations and considering an agnostic network. In this section the client controlled resource reservation and the new proposed network mechanism are described and assessed in a single operator scenario. As depicted in Figure 3, a user end-point (UE) is communicating with a service provisioning infrastructure, in order to obtain the service required by the user. The UE communicates using one access technology and is able to transmit data to the internet through an Access Network Gateway. In order to control the resources available in the network and also for charging proposes, the operator of the access technology uses some access network control, here depicted for simplicity as one entity.

Figure 3: Service and QoS Provisioning Architecture

The client oriented resource reservation presumes that the service provisioning infrastructure is not able to communicate with the resource reservation path. Thus after the service is provisioned, by this understanding that the communicating parties negotiate a common session profile, the UE has to use a separate mechanism to enforce the resource reservation on the data path. This may lead to multiple exception cases when the network is not able to offer the required resources as the two signaling flows are separated. The network controlled QoS presumes that during the session negotiation for a specific service, the service provisioning infrastructure communicates with the access network control of the network where the user is located, which at its turn enforces the policies both on the access network and on the access network gateway. On a first empiric evaluation the network controlled QoS is introducing a larger delay in the session provisioning. But considering that the client oriented resource reservation is done after the session profile negotiation and using a different signaling, this may lead not only to a larger delay, but also to multiple possible error cases, e.g. a session profile is negotiated, but only one of the parties involved in the session is able to reserve resources in its access network. In the architecture presented in the previous section, the communication between the service infrastructure (i.e. IMS) and the data path is done using the Evolved Packet Core (EPC) standardized by 3GPP. By integrating the two infrastructures, a new possibility of network control QoS appears. It presumes a trade-off by reducing the terminal knowledge and introducing it as part of the network control. The present IMS scenarios use the client controlled resource reservation for the access networks and the network based one for the core network QoS. With the increase number of terminals that can use multiple access networks simultaneously (e.g. UMTS and WiFi or WiMAX and WiFi), the resource reservation on the access networks depends more on the knowledge of the terminal. Even though the Evolved Packet Core offers a convergent model for accessing the required services, it does not consider user end-points with the capacity of parallel attachment to multiple access systems. Thus the reservation of resources and its connection to mobility still remain as open issues. A multiple access could be resource consuming if no prior information is exchanged with the Packet Core. The decision of connecting to a specific access point is taken by the user end-point,

based on some criteria like cost, signal strength and prior knowledge [8]. For a possible second connection, even though the first one has been established, the user end-point continues to consider only those local policies. Thus, it remains agnostic of the load information that is kept by the core, decreasing the possibility of finding a network able to offer the required surplus of resources. Some of the access technologies, like WLAN [9], use frequencies from the public spectrum. Thus the user is left to choose from multiple networks, from which some pertain to the operator and can offer the required services and some pertain to other independent parties. In order to eliminate this shortcoming, knowledge of the candidate networks is required. An alternative is to store all the information about all the networks of the operator in the user end-point. This will lead to memory consumption and to a possible duplication from malicious thirdparties. The same problem would appear in the case a static criterion of determining the possible good networks is inserted in the terminal. Therefore the only acceptable alternative is to receive information about usable networks, using another communication channel, if established. This information should be restricted to the geographical position of the user end-point and the user profile. By having multiple interfaces, on a service provisioning, a malicious user might send multiple resource reservation requests on all the interfaces available. Using the actual QoS model, in which the user has to reserve resources on the data path, various types of attacks could affect the network. Also in the public frequency networks the users can not be completely identified. Other devices could take over session provisioning of a user end-point and duplicate it, creating a denial of service attack. In order to limit this possibility a network provisioned QoS model was considered. Presently the services could be offered on multiple levels of quality. This is due to the coalescence of multiple services in a new service e.g. voice and whiteboard and to the appearance of different transmission mechanisms that could adapt in real-time to different bandwidth changes e.g. the video codecs H.261, H.263. Therefore the QoS provisioning mechanism should adapt to this evolution in the service. In the actual architectures, the reservation of a tiered service is done by multiple requests sent by the user end-point until the constraints of one class of reservation is satisfied. For example for an IPTV service multiple transmission rates could be considered. If the best one considered by the

user end-point cannot be satisfied, a request for the subsequent one is required. This process can repeat until the resources for one class can be reserved. This could lead to overly-extended session setup delays, which affect the user perception of the service as a whole. The network controlled QoS model here proposed is able to solve the resource allocation using only one step request.

4. Network controlled QoS Architecture

Figure 4 depicts the enhancements to the 3GPP EPC. A new function was introduced: the QoS Information Function (QIF) which keeps track of the resources that are available on the access networks and pre-reserves them on requests coming from the PCRF. The information is either gathered from the MME which communicates directly to the technology specific enforcement points or by other administrative mechanisms. Triggered by the service signaling, the PCRF filters the requested profile of the user according to the information received from the QIF. After the policy class is decided for the required service, it is enforced both to the access network using the specific mechanism and to the Serving and PDN Gateways, which act as access network gateways, having the possibility of exchanging packets with the core network. This way the resources are reserved on the full path from the UE to the Access Gateway and in the core network. The new interfaces introduced, have both the function of enforcement of different policies and of transmitting information about the momentary load of the resource on which the policies are enforced. Through interface I1, information about network resources availability is sent from the QIF to the PCRF. Using this information, the user profile and the cost information the PCRF is able to decide which the most appropriate access technology for the user is. The QIF receives information using interface I2 from the access networks. When a specific QoS is enforced on the access network, it also sends a message to the QIF on an administrative path. The enforcement of the QoS resources is done by the PCRF both on the access networks using interface I3 and on the Access Network Gateways using interface I4. It is necessary for the resources to be enforced on the access network gateways as to be considered for the connection between the operator network and the other networks. The QoS request has also to be enforced

on the access networks in order to be sent to all the technology specific entities in the path between the user end-point and the access gateway. I5 is a the typical IMS interface (Rx+) on which information about the IP filters necessary to identify the session and the QoS requirements of the user end-point are sent. Using this interface messages about policy control and charging are transferred from the user end-point to the Packet Core, in a protected way, by passing through the IMS signaling infrastructure. If one of the connections of the user end-point is done using a large coverage network, like the 3GPP UMTS, a Tracking Area is defined for the user. Having this information, the PCRF is able to determine the networks using other access technologies in the vicinity of the user end-point. By correlating this information with the information about the momentary load of the networks in the QIF and with the user specific policy and charging policies, the PCRF can announces the user endpoint to which networks could connect in order to enhance the quality of the service.

Figure 4. Enhanced System Architecture Evolution

By using the mechanisms of IMS, the user is geographically localized in a network secure manner, also offering the possibility to enhance the resources used accordingly to the specific service requests.

Two typical scenarios illustrating the usage of the architecture are further analyzed: the registration scenario and the QoS service provisioning scenario.

3.1. Registration Scenario

This scenario, as depicted in Figure 5, describes how a default signaling channel is allocated, when a new terminal registers with the network, in a technology transparent manner. When a new user end-point registers to a technology specific access network, information is sent to the technology specific access control point (1), which can determine the identity of the user for specific network types e.g. UMTS or cannot for other networks e.g. WLAN. In both situations the information has to be passed to the PCRF (2), in order to reserve a user specific or a general minimal default resource as to make the signaling possible. First the resources have to be reserved on the anchor point of the access networks (3, 4), then confirmed as to be reserved and enforced through the network to the user (5, 6). The user specific policy might ensure more resources for the default resources than the anonymous default reservation. Therefore after the user registers with the IMS infrastructure (7), a reallocation of the default resources is considered, if the momentary network capacities permit it. The PCRF receives the user information from the IMS structure (8) and taking into account the information from the QIF it enforces it on the anchor point and on the access network to the user (9, 10).

UE

Access Network Controller

Access Network Gateway

QIF PCRF

IMS

Network Registration(1) Network Registration(2) Default Reservation(3) Default Reservation(4) Network Registration(5) Network Registration(6) IMS Registration(7) User Default Reservation(8)

User Default Reservation(9) User Default Reservation(10) User Default Reservation(11) User Default Reservation(12)

Figure 5. Registration Scenario with the Enhanced EPC

By separating the anonymous registration from the user registration a better allocation of the resources is obtained. Also due to the fact that all the traffic passes through the access gateways a filter could be added. This can help in restricting the access of anonymous users to other domains than the one controlled by the operator. Therefore the service provider could secure the network from being used by unregistered parties. If in the future the default resource allocation is decided to be used as a bearer of data for third party services, the allocation of the resources can be done dynamically, using a profile inserted in the registration messages and evaluated by the PCRF. For example if the operator decides that for a specific connected user, a specific bandwidth should be available for non-signaled services, after the user registers with the IMS infrastructure, this service can be offered, by using this reallocation mechanism. Also this resource can be dynamically adjusted by subsequent registration requests.

3.2. QoS Provisioning Scenario

A typical resource reservation scenario using the same network oriented QoS provisioning in the access network is depicted in Figure 6. When a resource allocation request arrives from one of the user end-points (e.g. a SIP INVITE request) (1), the IMS signaling infrastructure makes a request to the PCRF (2). The PCRF, after combining the user information, with the set of policies and with the momentary load of the network received from the QIF, decides for a specific resource class and enforces it into the access gateway (3). Also this enforcement request is sent to the access network controller (4), signaling that the resources that are reserved are less than the user required, if necessary. The access network controller allocates the resources to the terminal on the access network; also if the user has the possibility to use other interfaces and they are in an inactive state, based on the Tracking Area of the end-point it creates a list of the interfaces which can be found by the terminal (5). The usable networks are than passed through a filter of the QIF (6), in order to keep only the ones that are highly probable of sustaining the resources required and send to the IMS infrastructure (7). The confirmation for the low level QoS that was reserved and the information about other networks that could enhance the service quality are sent back to the user end-point (8). At this moment the service could be started with a low resources allocation.

UE

Access Network Controller

Access Network Gateway

QoS Request (1)

PCRF

IMS

QoS Request (2) User Specific Reservation (3) User Specific Reservation Request (4) Allocating resources & Localization (5) User Specific Reservation Response with other usable networks (6) User Specific QoS Response with other usable networks (7) QoS Response with other usable networks (8) Connecting to another network (9) QoS Request for re-reservation (10) QoS Request (11) User Specific Reservation (12) User Specific Reservation (13) User Specific QoS Response (14) QoS Response for re-reservation (15)

Figure 6. QoS Provisioning Scenario

In the meantime, using the information received in the QoS request, about possible free networks in the vicinity, the end-point connects to another network, authenticates and receives a default reservation as it was previously described (9). This user end-point decision should not consider only the information received, but also the signal strength of the network it wants to connect to and the mobility of the user end-point. When the connection to the secondary network is completed the UE sends a new QoS Request for the same service session (10). The request is then processed by the PCRF and QIF and enforced on the complete data path from the user endpoint to the access gateway.

3.3. Policy Provisioning

A policy description, as received from the signaling core, can be a set of codecs and their usage as in the case of present scenarios. From this set of codecs the bandwidth and the service class are deduced and sent from the IMS Core to the Evolved Packet Core. Information containing possible networks to which the user could connect is received from the location management which may be present for some access technologies (e.g. UMTS). This enables the PCRF to select a set of policies that could be enforced if the resources permit it. The PCRF receives the degree of occupation of the networks from the QIF. By combining this information with the policies from the PCRF a reply is sent to the user. It contains a set of networks and which resources could be used if the user connects to that network. This enables the user end-point to receive information about the neighbor networks and their possibility to sustain its services. Also information about networks of other operators can be sent by using this mechanism, if roaming service is required and desired. The information of network locations can be introduced statically in the location mechanism. It does not have to match exactly the networks that are seen by the user at one specific moment, but it has to cover that set in order to give a larger set of possibilities to the user end-point. Thus, the networks that are considered by the PCRF can cover a larger area than the exact location of the user. This enables the user to select the network to connect based on its mobility and on other local criteria. This information is particularly useful for users that are already connected to a network and the network cannot sustain anymore the service requirements e.g. the signal of the network is rapidly decreasing due to an increase of the distance between the access point and the user end-point. This way the handover decision does not contain only the signal strength and the mobility parameters of the user, but also the degree of QoS availability in the network to which the user end-point is going to connect. User end-point requests and network resource availability is sent using the service signaling. This enables the user end-point to send QoS requirements, which, by this mechanism, can be resolved in a network secure manner. It also allows information to be sent from the network to the user end-point containing the reserved resources and the possible other resources available. The

network is completely controlled by the operator, thus reducing the risk of possible attacks and the error recovery delay. By using this complex mechanism, services have at the beginning at least a minimal resource reservation. Afterwards, if the network momentary capacities permit it, supplementary resources could be allocated according to the request of the parties involved. This leads to a provisional client satisfaction and to a multiple class behavior of the QoS provisioning. Multiple-card end-points can effectively use their network capacities in order to get access to more resources in a centralized coordinated manner controlled by the network operator. Thus the operator can control tighter the data traffic through the core. This increases the degree of usability of the network and decreases the delay due to exception mechanisms in the QoS provisioning. Also the multi-card terminals do not have to bind the second interface to some random network, expecting some user-space service to use the provided resources. By using the double reservation protocol here presented, the second card can be connected only after some resources are required, thus reducing the power consumption of the user end-point.

4. Conclusions
In this paper we presented an enhancement to the 3GPP EPC able, to optimize the QoS provisioning in a single operator scenario. It is improving the service user perception, especially for multi-card end-points and for tiered services. Transparent inter-operator scenarios have to be studied, in order to ensure roaming possibilities. A simple model was proposed for describing the user policies and the mechanism of processing these policies both on the IMS and on the Evolved Packet Core. A future study may present how the policies are provided by the user end-point to the Evolved Packet Core, the entities in the IMS Core that should modify and transmit the messages and how the selection is done by the PCRF, QIF and access technology controller inside the core. We have then proposed some additions to the 3GPP specification, in order to optimize the service provisioning. From both the operator and the user end-point, the services are better provided using a network QoS provisioning, which ensures a better QoS management from the operator and a better mechanism for exception handling as in the former 3GPP architecture.

A QoS Information Function (QIF) in the Evolved Packet Core able to register the load of the access networks was considered. This function is able to receive information about the various networks of the operator and to provide filters for selecting the ones that have enough resources for the services required by the user. For some access technology the EPC has knowledge of the user location by using a Tracking Area. An extension to this Tracking Area mechanism was introduced. It contains not only the networks of the tracked type e.g. UMTS, but also the networks that can be reachable by the user, by means of other technologies e.g. WLAN. A correlation between the QIF information and the information on the available networks provided by the extended Tracking Area provides a better network selection for multi-card user end-points. In order to obtain this correlation and the enforcement of the resources, two interfaces were considered. The first one, between the PCRF and the access network controller, is able to enforce policies on the mobility access and to inform about these enforcements. The second one, between the PCRF and the access gateways, both System Gateway and PDN Gateway, is able to enforce the policies in a transparent manner as seen from other parties involved in the service. The concepts here presented apply to the access networks of an operator. It remains also as an open issue the optimizations that have to be considered for the roaming user end-points in order to benefit from these core enhancements.

5. References
[1] IEEE 802.21Working Group. Ieee p802.21/D7..1: Draft ieee standard for local and metropolitan area networks: Media independent handover services (work in progress, 2007. [2] 3GPP TR 23.228, 3GPP System Architecture Evolution Report on Technical Options and Conclusions, November 2006. [3] 3GPP TS 23.203, Policy and Charging Control Architecture, December 2006. [4] TS 23.107 Quality of Service (QoS) Concept and Architecture, www.3gpp.org; 2006. [5] J Rosenberg, et al., SIP: Session Initiated Protocol, RFC 3261, June 2002. [6] RFC 2753 (A Framework for Policy-Based Admission Control, January 2000). [7] ES 282 003 Resource and Admission Control Sub-system (RACS); Functional Architecture. [8] 3GPP TS 23.107, Quality of Service (QoS) Concept and Architecture, June 2005.

[9] 3GPP TR 23.836, Quality of Service (QoS) and Policy Aspects of 3GPP Wireless Local Area Network (WLAN) Interworking, December 2005. [10] 3GPP TS 23.228 IP Multimedia Subsystem (IMS), December 2006. [11] D. Vingarzan, P. Weik, T. Magedanz, Introducing the FOKUS Open Source IMS Core System for Multi Access Multimedia Service Environments, IIRs Telecoms Signalling World Forum 2006, London, October 2006. [12] K. Knuettel, D. Witaszek, T. Magedanz, The IMS playground @ FOKUS an open testbed for generation network multimedia services, Tridentcom 2005, February 2005. [13] 3GPP TS 23.401 GPRS Enhancements for E-UTRAN access, October 2007. [14] 3GPP TS 23.402 3GPP Architecture Enhancements for non-3GPP accesses, May 2007. [15] IEEE 802.16 group. IEEE Standard for Local and metropolitan area networks Part 16: Air Interface for Fixed and Mobile BroadbandWireless Access Systems Amendment for Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands., 2004. [16] IEEE 802.16 group. IEEE Standard for Local and metropolitan area networks Part 16: Air Interface for Fixed and Mobile BroadbandWireless Access Systems Amendment for Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands., 2005.