Allocation of Multiple Serices in Multi-Access Wireless Systes Anders Furuskär Wireless@KTH, Royal Institute of Technology, Sweden and Ericsson Research anders.furuskar@era.ericsson.se Abstract This paper discusses principles for allocating ultiple serices onto different sub-systes in ulti-access wireless systes. Faorable near-optiu sub-syste serice allocations that axiize cobined ulti-serice capacity are deried through siple optiization procedures. These faorable serice allocations are either characterized by that the relatie resource cost for supporting serices is equal in all sub-systes, or they are extree points where serices are isolated in different subsystes. As a consequence of this, serices should typically be ixed in sub-systes with conex capacity regions, and isolated in sub-systes with concae capacity regions. Siple user assignent algoriths based on this are also discussed. Additionally, illustrating the ain findings of the analysis, soe syste exaples are gien, including a ixed GSM/EDGE and WCDMA case study. Keywords: Multi-Access, Multi-Serice, GSM/EDGE, WCDMA. I. INTRODUCTION Two expected characteristics of future wireless networks are support for ultiple serices and the use of ultiple radio access standards. This raises the question of how to best allocate the different serices onto the different access technologies (henceforth denoted sub-systes. In this paper, principles for how this serice allocation should be done and the resulting cobined capacity is discussed. More specifically, gien the ulti-serice capacity regions for the different sub-systes and a total traffic ix, it is analyzed how serices should be allocated onto the different sub-systes for axiu capacity. Seeral interesting ulti-access concepts ay be found in the literature. In [], a so-called always-best-connected concept coprising GSM/EDGE, WCDMA, CDMA and WLAN access technologies is discussed as a candidate for a future generation wireless network. Conergence of the aboe cellular networks with broadcast networks such as DAB and DVB is discussed in []. On a ore detailed leel, eans for exchanging traffic load and quality inforation between GSM/EDGE and WCDMA, as well as inter-working with WLAN-based systes are being standardized by the 3 rd Generation Partnership Project (3GPP. Analyses of the singleserice trunking gain enabled by the larger resource pool when cobining such systes hae been presented in e.g. [3]. No preious studies on ulti-serice allocation in ultiaccess systes hae been found howeer. This paper begins with a discussion of what differentiates user assignent in ulti-access systes fro single-access systes. This is followed by definitions of the ulti-serice, ulti-access syste odels and perforance easures used. Deriations of capacity-wise faorable, near-optiu subsyste serice allocations are then proided. Next, soe siple exaples illustrating the deried principles and the achieable capacity are presented. A perforance estiation for a cobined GSM/EDGE and WCDMA syste is also included. Finally, soe siple serice allocation algoriths are discussed. II. MULTI-ACCESS SERVICE ALLOCATION Within a single-access syste, handoer and cell-selection procedures typically assign users to an access port or a set of access ports with sufficiently good radio conditions. The ore accurate the estiates of the radio conditions, the better the access port assignent. This principle ay also be generalized for assigning users to access ports in ulti-access systes. Soe additional characteristics ay need to be taken into account in the ulti-access case howeer. These include:. The capability to handle arious serices ay differ between the sub-systes. Therefore the serice types of the users need to be taken into account.. The aount of inforation on indiidual users radio conditions that is aailable across sub-systes ay be less than that aailable internally within the sub-systes. Based on the aboe characteristics, soe interesting cases of inter-sub-syste inforation aailability ay be outlined: a The sub-systes hae no inforation of the situation in other sub-systes. Access attepts are accepted if possible, and otherwise redirected to another sub-syste. b The sub-systes can exchange load inforation per serice-type, and the serice-type of users is known. Access attepts ay be directed based on serice type. c The sae inforation is aailable across sub-systes as within sub-systes. Access attepts ay now also be directed based on expected radio resource consuption. The focus of this paper is on case b. Case a is seen as a reference case towards which the results are copared. A discussion on principles to use in case c is also included. Apart fro siplicity reasons, this scope is also otiated by the fact that current 3GPP standards include signaling eans supporting siilar solutions. III. SYSTEM MODELS AND PERFORMANCE MEASURES In this paper, a ulti-access syste consisting of two subsystes supporting two serices, oice and data, is considered. A generalization to a case with arbitrary nuber of subsystes and serices is straightforward, but for lack of space
d d ax Q Q in or Q d Q d in d d ( d ( Q Q in and Q d Q d in d ( ax d^( Figure. A capacity region exaple. reasons left out. The sub-systes are assued to hae the sae coerage area, and terinal capabilities to be such that any terinal can connect to any sub-syste. Beyond that the access ports of the sub-systes ay be arbitrarily located independently of each other. IV. SYSTEM LOAD AND SERVICE MIX The total nubers of oice and data users are denoted and d respectiely. The nuber of users allocated to sub-systes and are denoted, d, and d. A serice-ix α ay now be defined as the fraction of oice users aong the total nuber of users: α ( + d ( Siilarly within the sub-systes (, : α + d ( ( Apart fro the serice ix, the relatie serice allocation of serice n, i.e. the fraction of the nuber of users of serice n allocated to access technology is also of interest. This ay be expressed as: + ; d ( d (3 ( d + d A. User and Syste Quality The qualities of indiidual oice and data user in subsyste are denoted by the rando ariables q and q d respectiely. It is assued that q and q d are real-alued scalars. The distributions of these rando ariables depend on a coplex ariety of paraeters, including e.g. traffic, obility and syste characteristics. Typically howeer, for a fixed set of other paraeters, the quality decreases with an increase in the nuber of users in the sub-syste in question. A syste-leel quality of the oice and data serices in sub-syste ay be expressed as a function of the distribution of the corresponding indiidual user quality. For exaple, assuing that oice users are satisfied if they experience a quality exceeding q in, then Q Pr( q q (4 in is the probability that a oice user in sub-syste is satisfied, i.e. roughly easuring the fraction of satisfied users. B. Multi-Serice Capacity and Capacity Regions To easure the syste capacity, a set of iniu syste leel qualities Q in and Q d in ay be defined. Then, for a Figure. In a ulti-access syste the cobined capacity is the su of the sub-syste capacities. certain set of traffic, obility and syste assuptions etc., the sub-syste capacity for serice ix α ay be defined as: c( α ax( + d : Q Q in & Qd Qd in (5 By repeating the analysis for the full range of serice ixes a capacity region such as that depicted in Figure ay be constructed. For traffic loads inside the capacity region acceptable quality is thus sustained for both serice types. The cure deliiting the capacity region ay be expressed as a function d (. C. Multi-Access Cobined Capacity Regions The aboe perforance easure ay also be applied to ulti-access ulti-serice systes. The cobined capacity for a serice ix α is then defined as: c( α ax( + d : Q Q in & Qd Qd in (6 The corresponding cobined capacity region, deliited by a function d(, consequently deliits the axiu nuber of users that can be accoodated by both sub-systes while aintaining acceptable quality for both serice types in both sub-systes. For a gien oice load, the function d( deliiting the cobined capacity region ay be reasonably well approxiated by (see discussion below: d ˆ ( d ( + d( ; for + (7 Figure depicts a siple exaple. Noticeably the shape of the cobined capacity region depends on the sub-syste serice allocation, i.e. on and in the aboe equation. This fact is utilized in the next section when axiizing the cobined capacity regions. The approxiation in Eq. (7 disregards effects of trunking oer a larger resource pool, and thus underestiates the cobined capacity. For exaple, with 5% hard blocking, linearly suing the capacities of two sub-systes with 5 channels each underestiates the cobined capacity with soe 7%. Siilar effects ay be expected for interference liited systes, where the diersity gain steing fro reduced the risk of outage in all sub-systes as copared to a single sub- Note that this is a stronger requireent than acceptable quality aeraged oer the sub-systes, which thus ay yield soewhat higher capacity. The existence of single-ode terinals howeer otiates the stronger requireent.
syste is neglected by linear suation. For large sub-syste capacities the error should howeer be liited. V. SERVICE ALLOCATION STRATEGIES As discussed aboe, the cobined capacity region depends on how serices are allocated onto the sub-systes. Below, sub-syste serice allocations that axiize the cobined capacity region are deried. Since the objectie function d^( is an approxiation of d(, the resulting serice allocations ay not be regarded optiu. Motiated by the relatiely sall expected approxiation error, they ay howeer be considered near-optiu, and also denoted as such or siply as faorable. Proble: Gien the oice-data capacity regions of two subsystes: d ( and d (, for a total oice load, what subsyste serice allocations β and β fulfilling + axiize the sustainable data load d^ d + d? Solution: For a fixed oice load, d^ is gien by: (, d( + d( d( + d(, (8 ax(, ax in(, ax To axiize d^, its deriatie with respect to ay be taken: ˆ d(, ( d( + d ( (9 ( d ( ( d ( The zeros are thus found at β such that: ( d( ( d ( ( Whether these correspond to local axiu or iniu alues of d^ depend on whether the second deriatie of d^ is saller than or greater than zero respectiely: > (, ( d ( + ( d ( in ax ( If no zeros of the first deriatie are found, axiu alues of d^ are found at the extree alues of : ˆ If d(, ( d( ( d( > : If ax( in(, (, ( d ( ( d ( in( ax ax (, ax : ( To find global axius of d^, all local axius ust be copared including the extree alues of. Fro this global faorable solution, the full faorable, near-optiu serice allocation is gien by -, d d ( and d d (. The relatie serice allocations are gien by applying Eq. (3, e.g. β / etc. Further, to find faorable serice allocations for the whole range of serice ixes the aboe procedure ust be repeated for all total oice loads [, ax + ax ]. Fro these the desired apping between total serice ix α and faorable relatie sub-syste serice allocation β (α, β (α, β d (α and β d (α ay be calculated easily. Soe interesting characteristics of the aboe solution ay be obsered. First, inner faorable sharing points are characterized by that the capacity region slopes are equal, i.e. the cost in data users per oice user is equal in both sub-systes. This is intuitiely pleasing, since if this was not the case, a better solution could be found by oing data users fro the ore expensie subsyste to the other. It is further seen that a axiu in between the endpoints is only found if the su of the second deriaties is less than zero. Intuitiely this eans that serices should be ixed within the sub-systes if either both subsyste capacity regions are conex (here defined as d /, or one of the is ore conex than the other is concae. If this is not the case, the ost efficient serice allocation is found at the extree points, i.e. by as far as possible separating oice and data users. This further eans that one of the sub-systes will sere only oice or only data users. This is also the case when the capacity regions are linear; i.e. d / is constant. Then the following rules apply: If d / > d / : axiize (or equally β If d / d / : iniize (or equally β The intuitie eaning of these rules gets soewhat blurred by the fact that the deriaties are negatie. Howeer, the first rule ay be interpreted as that if d decreases slower with than d does with, should be increased at the expense of. The aboe procedure ay straightforwardly be generalized to arbitrary nubers of serices and sub-systes. The results are siilar. The faorable near-optiu serice allocations are characterized by that the gradients of the now ultidiensional capacity regions are equal in all sub-systes. VI. SOME SIMPLE ILLUSTRATIVE EXAMPLES This section presents soe fabricated, but illustratie, exaples of cobined capacity regions achieable through eploying the serice allocation rules of Section V. Figure 3 - Figure 5 show exaples of capacity regions; both per subsyste and cobined. The faorable serice allocations used to construct the cobined capacity region are also depicted. Starting fro a point on the cobined capacity region, ectors corresponding to the serice allocations in the two subsystes can be followed towards the origin. Figure 3 depicts a case with linear capacity regions in both sub-systes. Since in this exaple d / d / always holds, faorable serice allocations are found, for any serice ix, by allocating as any oice users as possible to subsyste, and as any data users as possible to sub-syste.
3.5.5 Equal serice ix allocation.5 d^(.5 d ( d^(.5 d ( d (.5.5.5 3 Figure 3. Cobined capacity region and faorable serice allocations for a case with linear sub-syste capacity regions Intuitiely this is correct since sub-syste is relatiely better at handling oice user than sub-syste, and ice-ersa for data users. The perforance of a schee that allocates serices to achiee equal serice ixes in both sub-systes is also depicted. Noticeably this schee results in a concae cobined capacity region. Further, the gain oer such a schee is deterined by the difference in the slope of the capacity regions. As seen in Figure 3, with linear capacity regions, the allocation principles becoe siple: serices should as far as possible be allocated to the syste relatiely best at supporting the. As a result, serices are ixed in only one of the systes, whereas the other only seres one type of users. With at least one conex capacity region, the situation becoes different, as depicted in Figure 4. Here, an inner solution with d / d / ay be found. To reach this faorable solution, serices should as far as possible be ixed in the syste with a conex capacity region. Further, the serice ix of this sub-syste should be such that d / d /. A consequence of ixing serices in the syste with a conex capacity region is that serices are ixed also in the syste with a linear capacity region. Hence, serices are here ixed in both sub-systes. Figure 5 depicts a situation with one linear and one concae capacity region. Noticeably here serice ixes fulfilling d / d / ay also be found. These howeer correspond to local inius. The faorable serice ixes are instead found at the endpoints of the concae capacity region. Consequently, serices are isolated in the syste with the concae capacity region. Noticeably, it is still possible to achiee a linear cobined capacity liit. VII. GSM/EDGE & WCDMA WITH MIXED VOICE & WWW.5 d (.5.5.5 3 3.5 Figure 4. Serices should be ixed within sub-systes with conex capacity regions. Mixed oice and WWW capacity regions for both GSM/EDGE and WCDMA exist, and are presented in e.g. [4] and [5]. In both cases rather linear regions are often achieed, which is what can be expected under interference liited conditions with interference aeraging. Although typical, this will howeer not always be the case. In systes with soe for of blocking or queuing, as well as in cases where the axiu tolerable interference depends on the serice ix, non-linear capacity regions appear. Exaples of such scenarios include sparse reuse GSM/EDGE-based systes for which oice bearers becoe blocking liited [4], and WCDMA with a ix of regular and HSDPA-based bearers [6]. Unfortunately the aboe capacity regions are deried under different syste, radio and traffic assuptions, and can thus not be directly used for ulti-access capacity ealuations. Howeer, assuing that the shapes of the capacity regions do not change draatically with the aboe assuption differences, they ay siply be re-scaled to fit the single-serice endpoints under equal assuptions. This should be a fair approxiation as long as the systes stay interference liited. The necessary single-serice oice and WWW capacity coparisons, based on equal assuptions, indeed exist and are used in the below analysis. Assuing a MHz spectru allocation, for a certain oice serice and certain syste, radio and traffic assuptions, the WCDMA oice capacity is about 5 Erlang per sector. The corresponding figure for GSM/EDGE is 5 Erlang per sector. For a perceied user throughput of 5kbps, the WWW capacity is about 7 kbps per sector for WCDMA, and about 6kbps per sector for GSM/EDGE. Beyond this, the capacity regions are approxiated to be linear. Figure 6 shows the resulting cobined capacity region and faorable sub-syste serice allocations. As far as possible, oice users are allocated to GSM/EDGE, and WWW users are allocated to WCDMA. A gain of up to 5% in ters of supported WWW users for a fixed oice load is achieed oer equal-serice-ix-allocation. For a fixed nuber of WWW users, a gain of up to % is achieable. It should be noted that the difference in slope of the capacity regions depends on The quoted figures are alid only for the set of assuption ade. Different assuption ay yield different results.
.5.5 d (.5.5 Figure 5. Serices should be isolated in sub-systes with concae capacity regions. the quality requireents of the serices, especially for WWW. With other requireents other results are achieed. VIII. d ( d^( SIMPLE USER ASSIGNMENT ALGORITHMS The principles of Section V yield sub-syste serice allocations that axiize cobined capacity. A siple user assignent algorith aking use of these results ay be realized by easuring the total serice ix α, and assigning oice and data users to sub-systes and according to the associated faorable relatie serice allocations β (α, β (α, β d (α and β d (α respectiely. It should be noted that this procedure ay be regarded nearoptiu only in an aerage sense, and proided that the only basis for the assignent is the serice type. Hence, in each single realization better assignents ay well exist. In addition to the serice-type, the relatie resource consuption of each user in each sub-syste ay also be used as a basis for user assignent. If aailable, such inforation ay be used to assign users where they are expected to consue the least radio resources, thereby increasing capacity. An assignent rule cobining both serice-type and resource consuption-based-assignent ay be realized e.g. by ultiplying the faorable relatie serice allocations β (α, β (α, β d (α and β d (α with inerse functions of the expected resource consuption in the different sub-systes. The relatie allocation rate is thus increased in the sub-syste where the expected resource consuption is the sallest. IX. CONCLUSIONS 8 6 4 GSM/EDGE WCDMA Cobined 5 5 5 3 Using a straightforward axiization procedure, faorable near-optiu sub-syste serice allocations in ulti-access systes ay be found. These faorable serice allocations are either characterized by that the relatie resource cost for supporting serices is equal in all sub-systes, or they are extrees where serices are isolated in different subsystes. Consequences of this include that serices should typically be ixed in sub-systes with conex capacity regions, and iso- Figure 6. Cobined capacity region and faorable serice allocations for a GSM/EDGE and WCDMA. lated in sub-systes with concae capacity regions. With linear capacity regions serices should as far as possible be allocated to the sub-syste relatiely best at supporting the. The gain achieable by eploying the proposed serice allocation strategy copared to a reference case of equalserice-ix allocation depends on the characteristics of the capacity regions of the sub-systes. Roughly, the ore different the sub-syste capacity regions, the higher the gain. The principles are readily applicable to ulti-serice GSM/EDGE and WCDMA ulti-access systes. It is further possible to cobine the serice-based allocation principles with assigning indiidual users in the sub-syste where their relatie resource cost is iniized. This is expected to yield further increased capacity. REFERENCES Equal serice ix allocation [] M. Frodigh, S Parkall, C. Roobol, P. Johansson and P. Larsson, Future-Generation Wireless Networks, in IEEE Personal Counications, October. [] R. Keller, T. Lohar, R. Tönjes and J. Thielecke, Conergence of Cellular and Broadcast Networks fro a Multi-Radio Perspectie, in IEEE Personal Counications, April. [3] A. Tölli, P. Hakalin and H. Hola, Perforance of Coon Radio Resource Manageent (CRRM, in proceedings of IEEE ICC. [4] A. Furuskär, Can 3G Serices be offered in Existing Spectru?, licentiate thesis, aailable at www.s3.kth.se/ radio/publication/pub/andersfuruskar_5.pdf. [5] R. De Bernardi et al., Load Control Strategies for Mixed Serices in WCDMA, in proceedings of IEEE Vehicular Technology Conference spring, p. 85-9, ol.,. [6] A. Furuskär et al., Perforance of WCDMA High Speed Packet Data, in proceedings of IEEE Vehicular Technology Conference spring.