Chapter 11: System Architecture Michel Mouly 1 11.1 Scope The decisions that were taken in GSM3/SMG3 were centred on the specification of signalling protocols. Signalling protocols can be understood as languages for exchanging control infor- mation between distant nodes, such as network nodes or mobile stations. The exchanged information is quite variegated. It includes data such as identities, called numbers, nature of call, description of allocated resources, result of measures, and so on. 11.2 Architecture Before the specification of a protocol can be started, it is necessary first to determine the two entities that are exchanging the information. This in turn requires a description of the overall system in terms of entities, such as mobile stations, terminals attached to mobile stations, radio base stations, switches, … The analysis and description of a system in terms of entities, of interfaces between such entities and of dialogs flowing over these interfaces, in short of the system architecture, was originally a derived task of GSM3; this task was eventually given to a sub-group of SMG3, and to a spawn of SMG3, SMG12. The choices driving the specification of the architecture of a telecommunications system such as GSM are determined by a variety of factors. Some splits are derived naturally from fundamental requirements, such as the split between the user equipment and the infrastruc- ture, with radio as the means for communication between the two sides. Roaming, interfacing with non-GSM telecommunications networks, interfacing with non-GSM user terminals, are other requirements leading to natural splits. Some other splits have been introduced to answer particular requirements from the opera- tors, for example to centralise some functions and thus to reduce costs. For instance, the separation between switches and radio site equipment allows the procurement of high capa- city switches (i.e. able to handle the traffic from a vast area, covered by many radio sites). Another kind of operator requirement that has been a leading factor in some architectural decisions is the wish to increase the possibility for competition between equipment manu- facturers. An example in this category is the split between the radio resource controlling node (the BSC) and the switches. 1 The views expressed in this chapter are those of the author and do not necessarily reflect the views of his affiliation entity. GSM and UMTS: The Creation of Global Mobile Communication Edited by Friedhelm Hillebrand Copyright q 2001 John Wiley & Sons Ltd ISBNs: 0-470-84322-5 (Hardback); 0-470-845546 (Electronic) Finally, many architecture decisions arose from the progressive enhancement of GSM past the first phase. Adding new functions was in some cases done by introducing new nodes, to ease manufacturing and/or deployment. On the other hand, splitting a system, which has a unique goal (allowing end users to communicate), in distinct and often distant entities leads to needs for co-ordination of the actions of these entities so that the common goal can be achieved. The additional require- ment to allow different manufacturers to compete on the equipment market (both for infrastructure and for user equipment), as well as giving some level of independence between operators, leads to the need for common, publicly available, specifications of the key protocols ruling the communications between distant entities determined by the archi- tectural choices. This sums up the main activities of SMG3/SMG12: to choose and to describe the general system architecture, to decide for which of the interfaces so determined a set of standard protocols should be produced, and to specify among those protocols the signalling ones. 11.3 General Trends The a posteriori analysis of the decisions taken by GSM3 distinguishes between two main approaches: on one side the adoption and, if needed, adaptation of techniques and protocols existing in the non-mobile telecommunications world, and on the other side the creation of original solutions, usually to answer problems specific to mobile telecommunications. If the latter approach illustrates the pioneering work and creativity of SMG3, the former has shaped more fundamentally the design of GSM. GSM was not designed independently from the rest of telecommunications. On the contrary, the services provided by GSM are primarily an extension to mobile users of the services available to users accessing networks through wirelines. For this reason, the tech- nical choices made in wireline telecommunications have had a deep influence on most key technical choices in GSM. Two periods can be distinguished when looked at from the wireline point of view. The initial, foundation building, technical decisions for GSM were taken around 1985–1988, at a time when the telecommunications ‘‘ paradigm’’ was ISDN, characterised by integration of voice and data, digital transmission of voice, and packet transmission for signalling. The second period started around 1992–1993 and was full-blown around 1995, and corresponded to the shift to all-packet transmission, with a focus on the Internet. The areas where SMG3 has been the most creative were, not astonishingly, those where no on-the-shelf solution was available from wireline techniques. From a signalling and architectural point of view, the essential difference between cellular access such as in GSM and wireline access is that a given user is not tied to a single access point, but on the contrary, is allowed to access from many points, spread geographically at the scale of the world, and from an administrative point of view, between many operators. This ‘‘ roam- ing’’ capability has a very important impact on how to route calls towards users, since their location is known only imperfectly. Mobility during calls leads to dynamic routing modi- fications, the handovers. For all these points, and many others, SMG3 built on the experience gained in NMT and other first generation cellular networks, using original ideas from its members to set down principles for digital cellular networks that remain valid through the next generation. GSM and UMTS: The Creation of Global Mobile Communication302 11.4 Historical Perspective 11.4.1 Before 1985 The examination of the reports and participant lists of the GSM meetings in those first years shows that the topics that were to be those of GSM3 were rather marginal. The main advocate of those topics, and of their importance, was Jan Audestad, who was to become the first chairman of GSM3. Important guidelines were established, if not officially, during this period, in particular the idea that GSM was to be designed mainly to interwork with ISDN. 11.4.2 1985–1988 GSM3 was created officially in 1985, in the first batch of creation of technical sub-commit- tees of the GSM standardisation group. Activities related to network architecture and proto- cols had started some months ago, as part of the GSM meetings, and the newly acquired independence allowed the working pace to increase. At that time, the radio transmission was not defined. The main tasks of GSM3 were then the general architecture, with as a main guideline the adaptation of ISDN switching and routing, as well as the protocols involved in call establishment and management. The main difference with ISDN was mobility, and in particular roaming. This led to the development of the Mobile Application Part (MAP), a protocol suite to be fitted in the larger protocol suite that is Signalling System No. 7 (SS7, the packet network used between ISDN switches for routing and call management). The other area of work was the adaptation of the ISDN protocols between the switch and the user equipment. From the start there was a distinction in GSM3 between ‘‘ network’’ related activities, meaning aspects related to switches and routing, and then roaming, and ‘‘radio’’ related activities, meaning what happened more locally, between the switch and the user equipment. A third activity was identified, namely the adaptation of supplementary services. These activities were allocated to sub-sub-groups, named WPC, WPA and WPB (WP for working party), in the order of presentation. This working organisation was essentially unchanged up to 1995. WPC, dealing with switching network architecture, set down very early the main concepts, in particular the very important notion of Home Location Register (HLR) and the technical principles for roaming, with such notions as location registration (allowing users to register with the local GSM switch) and two-branch routing of calls towards GSM users (so that a single number can be used to call a GSM user, wherever he/she is). This was a formalisation and clarification of ideas already existing in the analogue system NMT. The result has been one of the key factors for worldwide adoption of GSM, since with this technical basis (plus the financial aspects, treated by the MoU) roaming could be offered with not too much effort by any new operator. In WPA, dealing in particular with the radio interface, work started on the protocol architecture, but was somewhat limited due to several factors, in particular the fact that the main aspects of radio transmission were set down only during 1987. Several key decisions were taken during this period, decisions which are still acting for instance in the third generation standards. They deal with the architectural separation between what relates to the transmission means over the radio interface, and what transits over that interface, but is Chapter 11: System Architecture 303 essentially independent of it. The two key decisions concern first the system architecture, and second the radio interface protocol architecture. It was clear from the start that the cellular concepts implied that there would be much more emission/reception geographical points (base stations) than switches, if scale economies were to be obtained by using high capacity switches. This led naturally to the notion of switches and base stations as distinct nodes. The more in-depth studies of the architecture showed quickly that there were tasks that fit well in neither node, in particular those related to handing over connections from one base station to another when users moved too far away during calls. Some centralisation over several base stations seemed needed, and it appeared best not to overload switches with such computing hungry radio-related tasks as those required to determine when a handover is needed and to which base station. Thus came the three node architecture, the base station, the switch and the base station controller in between. This decision led to another, more difficult one, about whether protocol suites for the newly created interfaces (between base station and controller, and between controller and switch) were to be standardised, or left to node manufacturers. Such decisions are not technical, but the result of strategic equilibrium between on one hand operators, which favour open inter- faces, i.e. interfaces for which one standardised protocol suite exists, so to allow them independent choices for sources of the different nodes, and on the other hand manufacturers, which prefer to avoid the competition thus created. In the case of the GSM radio sub-system, the operators, no surprise, favoured the stan- dardisation of protocol suites for both interfaces, whereas manufacturers tried to limit the standardisation to a single one, with some more discussions on which one. To make the story short, the position that prevailed was that of the operators (no surprise once again, the GSM group being dominated by operators at that time), with a qualification that the effort was to be put mainly on the protocol suite for the switch to controller interface (the A interface). The other key decision taken by SMG3/WPA was the split of the so-called ‘‘ layer 3’’ protocol over the radio interface in three parts, one protocol for call management, one for mobility management and the other for radio management. Though this was not that obvious at the time, this decision is consistent with the system architecture, since it distinguishes the protocol that terminates in the base station controller (for radio management) and those terminating in the switch. In the short-term, the main advantage was to separate the part that could be done by a simple adaptation of ISDN protocols (the call management protocol) from those that were specific to cellular access (radio management and mobility manage- ment) and thus required a specific focus. With hindsight, those two decisions were excellent, because they set as a foundation the division between the radio access part (base stations and their controllers) and the network part (switches, location registers). Though the protocol suite developed by GSM3 was not fully universal, there are clear indications now that the A interface is the right point for which a universal protocol suite could be developed. Other cellular networks followed this route, as well for instance systems offering mobility with transmission through satellites. The Iu protocol suite developed for third generation is fully in the same spirit, and one step further in the direction of a universal protocol suite. WPA had many other tasks. Within this period the main lines for such aspects as handover support, access to the system from idle mode, network and cell selection, and paging over the radio interface, were established, among others. GSM and UMTS: The Creation of Global Mobile Communication304 11.4.3 1988–1992 In 1988, a complete first draft of the main protocol suites (MAP, A interface and radio interface) was available. Thanks to the decision to split the work between phase 1 and phase 2, the following years were devoted mainly to transforming these drafts into stable documents. Curiously, there is little of interest to describe in this period, though it was one of very intensive work. The work load was very high, and amount of documents dealt with and generated during this period was enormous. However, they concerned myriads of decisions concerning details; none of the founding decisions taken beforehand were challenged, and no new requirements led to decisions important to note here. 11.4.4 1992–1995 The years 1992–1995 correspond to the development of the phase 2 program. The main work was simply to modify the protocols to fit a number of new functionalities, all fitting rather naturally into the framework. Many of them were in the area of supplementary services, and the corresponding group, WPB had a lot to do. The introduction of some of those new features revealed some weaknesses in the capacity of the protocols to incorporate them with ‘‘ upward compatibility’’ . By this term is meant the possibility to interwork a node which has been upgraded with the new features and a node which has not. This is not really a problem when both nodes are under the control of the same operators, and then not of major importance in telecommunications so far. On the other hand, a consequence of roaming and of procurement from multiple, competitive, manufacturers, is that a network has to accept users with any equipment, old or new, and has to interwork with many other networks worldwide. Since it is difficult (!) to coordinate upgrading of all user equipment and of all networks in the roaming web, upward compatibility is a strong require- ment for the radio interface and for the interswitch protocol suite, the MAP. As a consequence, it was accepted that the transition between phase 1 equipment and phase 2 equipment would be done the hard way, but that everything should be done so that further enhancements can be introduced smoothly. This led to a number of modifications, very technical in detail, of both the radio interface protocols and the MAP. 11.4.5 1995–1999 Up to 1995, the work of GSM3 (SMG3 at that time) had been well delimited and consistent. With the advent of the phase 21 program, the work spread in many directions and the group expanded correspondingly, with unfortunately a loss of focus. A number of new topics arose that were too big to be dealt with together with the maintenance of the existing standardised features. Moreover, many were by nature weakly linked with these existing features. Another factor was the devolution of full work items to T1P1. The result was that the work in SMG3 was split into many independent tasks, minor and major. Some examples of interest are dealt with below. Chapter 11: System Architecture 305 11.5 Supplementary services Many phase 21 work items were supplementary services in the ISDN meaning of the term, and thus fell to SMG3. The adaptation to GSM of services offered with wireline ISDN was more and more difficult, because the easy one were done first. Some services such as Call Completion on Busy Subscribers (CCBS) proved to be tricky to combine with mobility. 11.6 GPRS The GSM architecture designed in the 1980s was heavily influenced by ISDN. In the ISDN perspective, packet switching is to be used only for signalling. Transmission of user data is limited to circuit. As a consequence of this limited view, ISDN switches, and hence GSM switches, since they are usually derived from them, have a high capacity for circuit switching, and a small one for packet switching. The latter was sufficient, however, so that the very low rate packet service called Short Message Service (SMS) was acceptable, but widely insuffi- cient for the emerging packet services, in particular for wireless access to the Internet. The consequence of this was that those services could be offered only using a circuit to cross the switch, and, unfortunately, the A interface. The topic of one major phase 21 work item was to provide some improvement of this state of affairs, and that was GPRS. It fell to SMG3 to study the system architecture for GPRS. However, because the members of WPC, the group in which switching and routing are dealt with, were mainly ISDN experts, and because the experts sent for dealing with GPRS architecture were mainly Internet experts, WPC was considered by both parties as unsuited to study the item, and a special technical sub-group was created. Other factors led in the same direction to estrangement between the original ISDN core of GSM and the GPRS work item. One was the weight of in-the-field equipment. Both operators and manufacturers felt that it was important to minimize the impact of the introduction of any new feature on the existing infrastructure, and on this basis they were reluctant to accept, or even to study in depth, an integrated architecture. Another factor was the preference of Internet people for something totally separate from the ISDN and PSTN. That was what happened for wireline telecommunications, and there was seemingly no reason why this should not be suitable to cellular. And a third factor was a unexamined trust in packet multiplexing as a source of major improvement. All this resulted, after discussions that lasted over years, in an almost totally parallel architecture, with the existing series of nodes essentially unmodified, and a new series of nodes (in particular a switch) and of protocol suites entirely devoted to GPRS. Moreover, the new part was designed on the simple, and seemingly natural idea that the Internet transmis- sion techniques have ‘‘ just’’ to be extended down to the cellular user equipment. In practice, the feature was technically very ambitious, and a long time was necessary to cover all the issues. The difficulties came from two directions: the contradiction between the lack of integration between the GPRS infrastructure and the original infrastructure and the need for such an integration in the user equipment; and the impact of mobility. The double architecture led in particular to two mobility managers, one in the circuit switch and the other in the GPRS switch (plus the idea of a third mobility manager, with mobile IP). Also, the protocol suite developed for the switch to radio controller interface (in correspondence with the A inter- GSM and UMTS: The Creation of Global Mobile Communication306 face), influenced by Internet approaches to transmission, took a narrow view to the handover issue. It is interesting to note that in the third generation, the development of the Iu protocol suite did not follow the GPRS approach but a more integrated view. 11.7 UMTS One consequence of the explosion of SMG3 in many parallel working activities was that the lack of a group dealing with the system architecture per se was more and more visible. In the early years, SMG3 plenary meetings were the main place where architectural consistency was checked. With the increase in the number of topics in phase 21, time was insufficient for that, and it was difficult to attract the required experts in meetings dealing mainly with preparing a synthetic report of what all the sub-groups did. More or less at the same moment, the need was expressed to integrate third generation activities into the main SMG groups. A sub-group of SMG3 was then created to answer both problems, and was nominally in charge of decisions related to architecture, both for GSM and the ongoing phase 21 features, and for the UMTS. In 1998 this group became a technical sub- group of SMG, under the name of SMG12. In some way, this group happened too late for GSM, and too soon for UMTS. On the GSM side, it was too late to come back on past architectural decisions, in particular regarding GPRS. On the UMTS side, the group discussed many interesting concepts such as the enhancements of roaming known as ‘‘ Virtual Home Environment’’ or a universal switch to radio access system protocol suite. However, recent events showed that such big steps forward were not compatible with the short-term objectives of the 3GPP group. 11.8 Onward With the advent of 3GPP, the organisation of the work dealing with the subject matter of SMG3 and SMG12 was drastically changed. The architectural aspects are dealt with 3GPP/ SA, the next avatar of SMG12, and have been set nearer to the group dealing with services. All technical aspects involving the switches have been grouped together and separated from the radio-subsystem aspects. The protocol suite for the radio interface, which was henceforth dealt with in a single group, has been split so that the protocols terminating in the switch are dealt with together with other switch-related topics, and the protocols terminating in the radio sub-system are treated in the group dealing with radio transmission. Despite these reorganisations, there is a true continuity from GSM to UMTS in the areas of architecture and protocols. On the technical side most of the key developments and ideas of GSM architecture and protocols are still living in UMTS. In the switching and routing area, the standards are common to GSM and UMTS. The continuity is also clearly visible in the working methods, and, not least, the membership. 11.9 Key people The output of the SMG3 and SMG12 amounts to tens of thousands of pages of specifications. Over the years, several hundred people participated in the different meetings, representing the work of possibly thousands of contributors spread over dozens of companies. Some contri- Chapter 11: System Architecture 307 butors will be named, though it should be kept in mind that their contributions represent only a very small part of the overall work. As in any project, the influence of individuals is more marked at the beginning. This is particularly true for the switching aspects, which brings to mind the names of Jan Audestadt and Christian Vernhes who participated in the founding technical work. On the radio sub-system side, a small group of people has animated the work for quite a long time, mainly Roland Bodin (whose influence is deeply missed since his premature death), Michel Mouly, Chris Pudney, and also Niels Andersen and Franc¸ois Courau. However, SMG3 and SMG12 key people are the first of hundreds of remote contributors, who provided the substance on which the chairpersons, the meeting secretaries and the participants worked. GSM and UMTS: The Creation of Global Mobile Communication308 . in UMTS. In the switching and routing area, the standards are common to GSM and UMTS. The continuity is also clearly visible in the working methods, and, . continuity from GSM to UMTS in the areas of architecture and protocols. On the technical side most of the key developments and ideas of GSM architecture and protocols