Mitigating Mobility Signaling Congestion in LTE by Overlapping Tracking Area Lists

Sara Modarres Razavi

Department of Science and Technology Linkping Institute of Technology SE-601 74, Norrkping, Sweden

Di Yuan
Department of Science and Technology Linkping Institute of Technology SE-601 74, Norrkping, Sweden

sarmo@itn.liu.se ABSTRACT
Avoiding signaling congestion in location management of cellular networks is becoming increasingly important as the population of user equipments (UEs) rapidly grows. Congestion can occur due to massive mobility of UEs behaving in a similar manner, such as the train movement scenario. In this paper, we explore the use of overlapping tracking area lists (TALs) in Long Term Evolution (LTE) networks for congestion mitigation that is not possible with the conventional tracking area concept. Each cell can use multiple and overlapping TALs, to be allocated to UEs requesting their TALs from the cell. We show that nding the optimal proportional use of TALs can be formulated as a linear program. Solving the linear program minimizes the maximum tracking area updates occurring between the cells. We present numerical results to illustrate the performance of the approach. The experiments demonstrate the eectiveness of overlapping TALs for mitigating mobility signaling congestion.

diyua@itn.liu.se
Term Evolution (LTE) networks, the focus of location management is on LTE-idle mode UEs. The network knows the location of LTE-idle UEs within the granularity of a few cells forming a tracking area (TA). The UEs are frequently in the idle mode and have no active radio connection with the network. The excessive signaling required for tracing the LTE-idle UEs in some scenarios may signicantly decrease the quality of service in some parts of the network, and could create signaling resource congestion. The conventional TA scheme is almost the same concept as the Location Area (LA) in the circuit-switched (CS) domain and the Routing Area (RA) in the packet-switched (PS) domain in GSM, GPRS and UMTS networks [1]. A UE updates its location whenever it changes a TA. In order to place a call to an idle UE, the Mobility Management Entity (MME) pages all cells of the TA in which the UE is currently registered. Hence, the tracking area updates (TAU) are uplink signaling messages from UEs to the MME, and pagings are downlink signaling messages from the MME to UEs. One of the well-known performance consideration is the signaling overhead of TAU versus that for paging. In densely-populated cities, there is a drastic dierence in the mobility patterns between day and night. There are typically many UEs moving into a central area in the morning by jam-packed public transportation, causing signaling congestion around the center of the city. In this case, the UEs at the border of TAs may not have a sucient quality of service during peak hours [3]. This is an undesirable situation from both the network and UEs standpoint [11, 13]. Reconguring TAs in order to adapt to the UEs movement behavior during pick hours typically requires temporally tearing down the cells, which is a costly process from the service perspective [14]. The current paper aims to mitigate the signaling congestion resulted from mass mobility of idle UEs without performing a TA revision. In comparison to earlier generations of cellular networks, LTE networks allow for a more exible conguration of TA design by the list concept introduced in 3GPP Release 8 [2]. In this scheme, instead of assigning one TA to each UE, one UE can have a list of TAs. The UE receives a TA list from a cell, and keeps the list, until it moves to a cell that is not included in its list. In LTE standards, a cell is also able to give dierent lists to dierent UEs. The UE location is known in the MME to at least the accuracy of the list allocated to that UE. In our study, we mitigate the excessive TAU signaling at the TA boundaries by assigning UEs with dierent and overlapping tracking area lists (TAL). There are still many open problems unexplored in the

Categories and Subject Descriptors

C.2.1 [Computer-communication Networks]: Network Architecture and DesignWireless communication ; G.1.6 [Numerical Analysis]: OptimizationLinear programming ; I.6.5 [Simulation and Modeling]: Model Development

General Terms
Design, Management, Performance

Keywords
location management, mass mobility congestion, tracking area list

1.

INTRODUCTION

By the dramatic increase in the population of user equipments (UEs) in cellular networks, the study of avoiding signaling congestion scenarios becomes important. In Long

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for prot or commercial advantage and that copies bear this notice and the full citation on the rst page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specic permission and/or a fee. MSWiM11, October 31November 4, 2011, Miami, Florida, USA. Copyright 2011 ACM 978-1-4503-0898-4/11/10 ...$10.00.

285

Figure 1: Example of TAU storm at the border of two TAs. study of TAL. One main problem is how to estimate the overall signaling overhead of a fully deployed TAL scheme in a large scale network. The UEs are allowed to move in the cells contained in their TAL without any TAU. Thus, a cell contains UEs with dierent TALs depending on the cells from which the TALs are assigned. This diversity makes the estimation of signaling overhead dicult. The only way to calculate the exact overall signaling overhead of the network based on TAL design is to have the accurate data traces of UEs movements. Unfortunately, the real data of UE traces is not available or too expensive to obtain. In the previous studies [15, 16], to overcome this problem, the authors generated a UE-trace scenario based on cell load and handover statistics of the network. This paper explores a new scheme in which the signaling overhead calculation is independent of UE traces. The overlapping TAL scheme in this study allows the cells to have a group of lists and to assign UEs dierent TALs. This feature could not be used in the framework of the TAL scheme suggested in [15] and [16]. The contributions of the paper can be summarized as follows. First, we develop the new overlapping TAL scheme. Second, we theoretically present how the scheme works with small instructive examples. Third, we present a linear programming (LP) model which minimizes the number of TAUs while keeping the overall paging overhead lower than some specied limit. The numerical results demonstrate that the proposed model is able to balance the load for representative scenarios of massive mobility signaling congestion.

crossing a TA boundary (Figure 1) [13]. In this study, we propose a practical method to design an overlapping TAL scheme in order to overcome the second problem. Introducing the concept of list gives more exibility to the operators in their TA management. Two of the so far discussed candidate schemes are as follows: The Overlapping TA scheme in which one cell holds a list of overlapping TAs and a UE is assigned only to one TA in the cells list. The other candidate is the Multiple TAs scheme in which a cell belongs to only one TA, but a UE can be assigned with more than one TA using a list. The objective of these two schemes is to reduce the TAU overhead, while assuming that the problem of paging is a less critical issue. The Multiple TAs scheme is considered as the more preferred scheme among the two in [20]. We have previously proposed the TAL scheme in which both UEs and cells are assigned a list of non-overlapping TAs [15, 16]. The assumption was that a cell gives only one common TAL to all UEs getting updated in that cell. Inevitably, this imposes restriction on the performance of TAL. It has been shown that the TAL scheme is potentially able to reduce the overall signaling overhead of the network compared to the conventional TA scheme [17]. However the challenge in applying this scheme is that UEs in one cell have dierent lists. It is already mentioned in the previous section that the perspective dierence makes the estimation of signaling overhead dicult. In the current study, we are interested in a framework where there is no need of having the UE traces for calculating the exact value of signaling overhead. The congestion mitigation problem in this paper has been previously considered in [11]. The authors of [11] use randomized time distribution of cell reselections to avoid the mass mobility signaling congestion. However, their mechanism requires additional conguration on the network side in order to provide the algorithm parameters, while our approach uses the existing feature in LTE systems.

3. TAU AND PAGING CALCULATION

Tracking area is dened as an area in which a UE may move without updating the MME. The initial design task is the formation of TAs inside a network, and the natural objective in TA planning is to reach an optimal balance between TAU and paging. In real-life networks, splitting a site into dierent TAs is not a common practice. Therefore, although TA assignment is applicable to the cell level, we keep the discussion at the site level. Let us denote the set of sites in a network by N = {1, . . . , N } and the TA of site i by ti . A TA design can be alternatively represented by an N N symmetric and binary matrix S; in which element sij represents whether or not two sites are in the same TA, i.e., sij = 1 if ti = tj , 0 otherwise. (1)

2.

RELATED WORKS

All the problems related to LA and RA planning and optimization can be generalized to the study of TA. Hence, there is an extensive amount of literature on the conventional TA concept [5, 17, 23, 24]. In this section we only review the limited amount of work that has been so far published on the concept of TAL, and the problem of massive mobility congestion. Since 2007, a great attention has been given to the development of the list concept in the 3GPP standardization procedure and the telecommunication industries [13, 19, 20, 21, 25]. The TA list has been suggested to solve a couple of specic problems existing in the conventional TA scheme. One is that the TA list can prevent the frequent updates when a UE keeps hopping between two or more adjacent cells in dierent TAs (the so called ping-pong eect) [19]. Second, TA list can solve the problem of high uplink trafc due to simultaneous updates of a large number of UEs

The S matrix has the transitive property: If sites i and j are in the same TA, and sites j and k are in the same TA, then i and k are also in the same TA. Let ui be the total number of UEs in site i scaled by the time proportion that each UE spends in site i. For the same time period, hij is the number of UEs moving from site i to site j . The values of ui and hij can be assessed by site load

286

and handover statistics of active UEs [14]. The amount of overhead of one paging and one update are denoted by cp and cu , respectively. The exact relationship between cu and cp depends on their radio resource consumption [7]. Moreover, parameter is the call intensity factor (i.e., probability that a UE has to be paged). The calculated CT AU in Equation (2) below accounts for the total TAU overhead of the network. The TAU overhead of site i is proportional to the number of UEs moving from site i to other sites such as site j , when i and j are not in the same TA. CT AU =
iN j N :j =i

Figure 2: Three sites along a train path. Table 1: Possible TA congurations of the ABC network with the corresponding CT AU and CP AG . CP AG No. Conguration CT AU 0 1 (A),(B),(C) cu (hAB + hBA +hBC + hCB ) 2 (A,B),(C) cu (hBC + hCB ) cp (uA + uB ) 3 (A),(B,C) cu (hAB + hBA ) cp (uB + uC ) 4 (A,B,C) 0 2cp (uA + uB + uC ) Table 2: The list of TALs for each site in ABC network. A B C {B } {C } {A} {A, B } {A, B } {B, C } {A, B, C } {B, C } {A, B, C } {A, B, C } Let us assume that each of the sites A, B and C are a separate TA (conguration 1) and they can assign UEs with dierent TALs. The TALs assigned by each site should contain that site. Table 2 presents the set of all possible TALs for each site. Site A and C are not giving TAL {A, C } to their UEs as there is no direct handover between these two sites; this list only increases the paging overhead without any improvement in TAU. From the table, it is observable that the UEs in the ABC network have the possibility to hold one of the following six TALs: ({A}, {B }, {C }, {A, B }, {B, C }, {A, B, C }). ta Let ta denote any TA of the ABC network; we dene Xtal as the percentage of the UEs inside each ta having tal as their list by that ta. The following three constraints are valid. Equation (6) assures that all the UEs receiving their lists from A use one of the three candidate TALs of A. The same logic for TAs B and C is represented by Equations (7) and (8), respectively.
A A A XA + XAB + XABC =1 B XB

cu hij (1 sij )

(2)

The computed CP AG in Equation (3) below is the overall paging overhead of the network. The paging overhead of site i is proportional to the number of incoming calls of the UEs in i and the number of other sites in the same TA. CP AG =
iN j N :j =i

cp ui sij

(3)

4.

AN INSTRUCTIVE EXAMPLE

Let us consider an example of three sites, A, B, and C, which are along a train path in Figure 2. The optimum conventional TA conguration of this part of the network is one of the four possible combinations presented in Table 1. The parenthesis in the second column of the table specify dierent groupings of these three sites. According to Equations (2) and (3), the overall TAU and paging overheads are calculated for each of the four congurations. The conguration {(A, C ), (B )} is practically not a feasible combination, as it has been assumed that there is no direct handover from A to C in the example. If we relax the transitivity condition of the S matrix, then the S matrix in (4) could become a possible conguration for the example in Figure 2. 1 1 S = 1 1 0 1 0 1 1

(4)

u p CT AU + CP AG = c (hAC + hCA ) + c (uA + 2uB + uC ) (5)

The overall signaling overhead of S is given in (5). In the example, hAC = hCA = 0, and CT AU = 0, hence the S conguration has the lowest TAU overhead compared to the congurations in Table 1. Assume that uA uB and uC uB (for example considering that there are train stations in A and C with large number of UEs standing in the platform), then the paging overhead CP AG is almost half of the paging overhead of the fourth conguration in the table. This example shows that the transitive property of the S matrix limits the performance of the conventional TA scheme. Here, we are interested to show that the TA list concept has the potential to reduce the overhead by relaxing this property, achieving an eect similar to that of S . However, the challenge in using the list concept is that the UEs in one site might have dierent lists depending on the sites from which the lists are assigned.

(6) (7) (8)

B XAB C XC

+ +

B XBC C XBC

+ +

B XABC C XABC

=1 =1

According to the denition of TAL, the UEs holding the list {A, B, C } can move in the ABC network without any TAU, and they could have also received this TAL from any of the sites. To make the calculation of signaling overhead tractable, we impose the restriction that the percentage of assigning a common TAL from dierent TAs are equal. This A B B C means that we require XAB = XAB , XBC = XBC and A B C XABC = XABC = XABC . With the mentioned assumption, there is no need to continue with the superscripts of elements (Xtal ).

287

5.

OVERLAPPING TRACKING AREA LIST

In the overlapping TAL scheme, the sites inside each TA hold a list of overlapping TALs. The cells update the UEs with dierent TALs in their lists. We consider the conventional TA design as the backbone structure, and create the lists by dierent combinations of TAs. The restriction explained in the previous section is applied here. It means that in the overlapping TAL scheme, the percentages of assigning a common list from dierent sites to the UEs in the network are equal. The biggest diculty in calculating the signaling overhead resulted from the TAL scheme was the challenge of having UEs with dierent TALs in one cell. However now with the denition of Xtal and the overlapping TAL scheme, the percentage of UEs holding each TAL is known. Therefore, unlike previous studies where the signaling overhead was calculated by following each UE movement pattern and call arrivals, here we can directly use the TAU and paging overhead formulas given in (2) and (3). To overcome the signaling congestion in the train scenario, the cells (sites) along the railway path can give dierent TALs to the UEs inside the train. Because UEs in the train are holding dierent TALs, not all of them will perform TAU at the same time.

Figure 3: Denition of Xij in Model (10).

6.

SOLUTION APPROACH

We aim to formulate a solution approach which nds the optimal proportional use of TALs by the sites along a railway path. The objective is to minimize the maximum TAU signaling overhead (M T AU ), while the total paging overhead growth (T P AG) is bounded by a limit value in comparison to the overall paging overhead of the TA design. For simplicity, we rst consider the small example of the ABC network. In this case, we assume that A, B and C are the sites along a railway path at a peak hour. Now that it is possible to calculate the TAU and paging overheads based on Equations (2) and (3), we can minimize the maximum TAU signaling by applying the LP model in (9) and obtaining the percentages of UEs holding each TAL. The rst three constraints assure that the use of candidate TALs for each TA sum up to 100%. The next two constraints limit the TAU overhead of each site to be less or equal to M T AU . The last two constraints calculate the total paging overhead addition of the three sites, and keeps it lower than the Limit. min M T AU subject to: XA + XAB + XABC XB + XAB + XBC + XABC XC + XBC + XABC cu hAB XA + cu hBA (XB + XBC ) cu hBC (XB + XAB ) + cu hCB XC cp uA (XAB + 2XABC ) + p c uB (XAB + XBC + 2XABC ) + cp uC (XBC + 2XABC ) = = = 1 1 1 M T AU M T AU (9)

Without loss of generality the terms site and TA are used interchangeably here. In general, the sites considered in a train scenario have the unique property that they only have handover with their two adjacent sites. This fact forces the TALs to contain consecutive sites. We dene Xij as the percentage of the list containing i, j and all the sites between them. For an illustration of an example of Xij see Figure 3. We generalize the idea in Model (9) to N sites in Model (10). Apart from the non-negativity constraints, the number of constraints of Model (10) is of O(N ), and the number of variables is of O(N 2 ). This shows that the model is polynomial in size and it can be solved in polynomial time. The low computational complexity of the model is a denite advantage of our solution approach. The rst set of N constraints ensures that each site assigns all its UEs with feasible TALs. The second set of N 1 constraints computes the TAU signaling overhead of each site. The rst sum considers UEs moving from i to j without having j in their TAL, and the second sum is for UEs moving from j to i without having i in their TAL. The remaining constraints calculate the total paging overhead addition of all sites {1, .., N } and bound T P AG to a certain value. The paging overhead of each called UE depends on the length of its TAL, which in this case is j k. min M T AU subject to:
i N

(10)

Xkj = 1, i = 1, .., N
k=1 j =i i N

cu hij
k=1

Xki + cu hji
k =j

Xjk M T AU

i = 1, .., N 1, j = i + 1
N i N

cp
i=1

ui (
k=1 j =i

(j k)Xkj ) = T P AG T P AG Limit

Xij 0, i = 1, .., N, j = 1, .., N, i j

7. PERFORMANCE EVALUATION
To evaluate the performance of the overlapping TAL scheme proposed in this paper, we consider two synthesized scenarios, which are constructed to resemble the reality. In both scenarios, it is assumed that 5% of the UEs are paged in every site. The overhead of a single update cu is set ten times as much as cp [10]. The load and handover statistics of the UEs are given for one hour period of time. The LP

T P AG Limit 0

T P AG XA , XB , XC , XAB , XBC , XABC

288

Maximum TAU overhead at TA boundaries

25 20

Conventional TA Scheme Overlapping TAL Scheme

Figure 5: Congestion Scenario 1.

15
Maximum TAU overhead at TA boundaries 700 600 500 400 300 200 100 0 0 0.2 0.4 0.6 0.8 Limit based on paging overhead ratio 1

10 5 0 0

Conventional TA Scheme Overlapping TAL Scheme

0.2 0.4 0.6 0.8 Limit based on paging overhead ratio

Figure 4: Comparing the performance of the two schemes for the baseline scenario. model dened in (10) has been implemented in the CPLEX optimizer [8]. Both scenarios represent UE movement by train. We illustrate that the overlapping TAL scheme can mitigate the congestion problem in both scenarios of the network. As the LP model is polynomial in size, applying it to largescale scenarios of this kind is scalable. The basic element of this study is TA rather than site. To compare the performance of the overlapping TAL scheme with the conventional TA scheme, we consider the TALs to be the combinations of overlapping TAs. However, it is obvious that the TALs at site level can improve the possible potentials of TALs. Baseline Scenario Both scenarios involve a railway path with ve TAs {A, B, C, D, E } of 150 km in length (see Figures 5 and 7). Each TA covers a large area and contains ve sites. The boundaries of each TA have the lowest handovers between UEs in that region. This conguration is the optimum conventional TA design by model [22] based on the overall signaling overhead of the specied part of the network. The overall signaling overhead is 841.7 (CT AU = 122, CP AG = 719.7). The data excludes the mass mobility of UEs in train. We apply Model (10) to the baseline scenario and run it for all Limit values from 0 to maximum. It only takes several seconds for the solver to calculate the optimum proportions of TALs for each Limit value. Next, we design the conventional TA congurations by minimizing the maximum M T AU , the same objective in Model (10), and all the Limit values. Note that while the Limit value is 0, all the ve TAs are separate in the conventional TA conguration, and the equivalent solution in the overlapping TAL scheme adapts all the UEs in each TA to contain only that TA in their lists. On the other extreme, when the Limit value is maximum, all the ve TAs are merged into one in the conventional TA conguration. The overlapping TAL, in this case, gives a list containing all the ve TAs to every UE. Hence, for the two extremes of the Limit the optimal conventional TA scheme and the overlapping TAL coincide. The comparison of the conventional TA scheme and the overlapping TAL for the baseline is illustrated in Figure 4. As it is explained above, the graphs coincide at the two

Figure 6: Comparing the performance of the two schemes for Scenario 1.

ends. It is observable that the conventional TA scheme is changed in four discrete steps. By increasing the limit on the paging overhead, the TAs tend to merge. The behavior of the overlapping TAL scheme is smooth as the proportions of each TAL can be continuously changed. By allowing 10% addition of the paging overhead, we can reduce the maximum TAU overhead at the TA boundaries (M T AU ) by more than 50% with the overlapping TAL scheme, while the conventional TA scheme can only reduce it by 20%. Note that even in this uncongested scenario where mass mobility is not considered, the overlapping TAL scheme can signicantly perform better than the conventional TA scheme by overcoming the transitive restriction explained in Section 4. Congestion Scenario 1 In this scenario (Figure 5), a train moves from A to E having 600 passengers onboard. One example can be TGV Eurostar train which can move at a speed up to 300 km/h and may carry up to 784 passengers [13]. After a main station in C , the number of onboard passengers is reduced to 400. We assume that all passengers inside the train hold cell phones. It takes about half an hour for the train to pass the whole path (ignoring the time spent at the main station). We aggregate the load and handover data resulted from the train and add it to the baseline scenario. All the UEs in the train passing each TA boundary will be added to the corresponding handover. The load is increased according to the proportion of time the train is spent in each TA. For example, the train spends about 6 min in A, therefore the 1 load of A is increased by 600 10 = 60, due to the fact that the handover and load data are given for one hour period of time. For various Limit values, Figure 6 illustrates the behavior of the two schemes in minimizing the M T AU . Here the Limit refers to the additional paging overhead due to the

289

Table 3: Results of Scenario 1 for a certain Limit. Limit TPAG MTAU Conventional TA 215.91 173.55 424.00 Overlapping TAL 215.91 215.91 163.28 Table 4: Results of Scenario 2 for a certain Limit. Limit TPAG MTAU Conventional TA 215.91 175.49 471.00 Overlapping TAL 215.91 215.91 176.59

Figure 7: Congestion Scenario 2.

Maximum TAU overhead at TA boundaries 700 600 500 400 300 200 100 0 0 0.2 0.4 0.6 0.8 Limit on paging overhead addition 1

Conventional TA Scheme Scenario 1 Overlapping TAL Scheme Scenario 1 Conventional TA Scheme Scenario 2 Overlapping TAL Scheme Scneario 2

Figure 8: Comparing the performance of the two schemes for Scenarios 1 and 2. conventional TA congurations and the proportions of the lists in the overlapping TALs compared to the original ones. By comparing the values on the y-axis in Figures 4 and 6, we can conclude that Scenario 1 is highly congested. When the Limit is equal to 20% of the overall paging overhead, equivalent to 143.94 in Scenario 1 (shown by the vertical line in Figure 6), the conventional TA scheme can reduce the maximum TAU signaling at the TAU boundaries by 0.01%. In the same situation, the overlapping TAL scheme reduces the M T AU by 60.9%, equivalent to 242.01. This case study demonstrates the benet of the overlapping TAL scheme in mitigating the congestion problem of the train scenario. Congestion Scenario 2 Now, we add another train in the opposite direction to Scenario 1 (Figure 7). This train, which can be a regional train with speed reaching 160 km/h, has 450 passengers in the beginning and it stops in every TA on its way. It takes one hour for the train to pass the specied railway path. The same data aggregation is applied for this scenario. Figure 8 illustrates the performance of the two schemes for Scenario 2. The results of Scenario 1 in Figure 6 is replicated to Figure 8 to ease comparison. Couple of observations follow from Figure 8. First by adding the second train to the scenario, the gap between the conventional TA schemes of Scenario 1 and 2 in some parts is more than ten times bigger than the corresponding gap for the overlapping TAL scheme. The proportions of TAL in reducing the maximum TAU overhead are mostly based on the largest amount of simultaneous movements of the UEs. Second, the overlapping TAL scheme performs much better than the conventional TA scheme while mass mobility in both directions is present. Let us consider that the Limit is 0.3 of the overall paging overhead, equivalent to 215.91 (shown by the vertical

line in Figure 8). The optimal conventional TA conguration based on Model (10) is {(A, B, C ), (D), (E )}, which can be represented by XABC = 1, XD = 1 and XE = 1 for both scenarios. However, the optimal TAL proportions for Scenario 1 are: XA = 0.27, XE = 0.18, XAB = 0.26, XCD = 0.26, XDE = 0.60, XABC = 0.39, XBCD = 0.05, and XABCD = 0.08, while the optimal TAL proportions for Scenario 2 are: XA = 0.29, XE = 0.41, XAB = 0.24, XCD = 0.11, XDE = 0.37, XABC = 0.42, XBCD = 0.25, XCDE = 0.17 and XABCD = 0.05. The overhead results of these congurations are given in Tables 3 and 4 for the two scenarios, respectively. The numerical results show that the overlapping TAL scheme is able to mitigate the rst congestion scenario by 61.49% and the second congestion scenario by 62.51% more than the conventional TA scheme for the Limit under consideration. The fundamental exibility of the overlapping TAL scheme in assigning dierent proportions of each TAL signicantly improves the performance compared to the conventional TA scheme.

8. CONCLUSIONS
We have presented a low-complexity solution procedure, based on linear programming, in order to nd the optimal proportional use of the lists of TAs for the proposed overlapping TAL scheme. The congestion mitigation eect is achieved by minimizing the maximum TAU signaling overhead in the network. The study shows that by assigning dierent TALs to the UEs involved in massive mobility, one can avoid the excessive TAU signaling messages at the TA boundaries. The transitive property of the conventional TA conguration is relaxed in the overlapping TAL scheme, which considerably improves the load balancing capability. There are several topics deserving further research. One is to apply the overlapping TAL scheme in a large scale network to see its potentials compared to the conventional TA scheme and the previously suggested TAL algorithms. Another interesting subject worth studying is to extend the scheme by considering more data in hand, for example assuming that the destination of each UE in a train is known.

9. ACKNOWLEDGMENTS
The work has been supported by CENIIT, Link oping University, Sweden, and the Swedish Research Council.

10. REFERENCES
[1] 3GPP. TS 23.002 v9.1.0, Network Architecture, 3G Release 9, 2009. http://www.3gpp.org/ftp/specs/archive/23 series/23.002/.

290

[2] 3GPP. TS 23.401 v8.6.0, General Packet Radio Services (GPRS) enhancements for Evolved Universal Terrestrial Radio Access Network (E-UTRAN) access, 3G Release 8, 2009. http://www.3gpp.org/ftp/specs/archive/23 series/23.401/. [3] 3GPP. TR 23.880 v0.1.0, Stage2 for registration in densely-populated area (RED), 3G Release 8, 2007. http://www.3gpp.org/ftp/specs/archive/23 series/23.880/. [4] 3G Americas, The Benets of SON in LTE, White Paper, 2009. [5] I.F. Akyildiz, J. McNair, J.S.M. Ho, H. Uzunalioglu, and W. Wang, Mobility management in next generation wireless systems, Proc. of IEEE, vol. 87, pp. 1347-1384, 1999. [6] S.K. Das, S.K. Sen, R. Jayaram, and P. Agrawal, A distributed load balancing algorithm for the hot cell problem in cellular mobile networks, Proc. of the 6th IEEE International Symposium on High Performance Distributed Computing, pp. 254-263, 1997. [7] P. Garcia, V. Casares, and J. Mataix, Reducing location update and paging costs in a PCS network, IEEE Transactions on Wireless Communications, vol. 1, pp. 200-209, 2002. [8] ILOG CPLEX 12.2, Users Manual, 2010. [9] S.S. Kim, A.E. Smith, and S.J. Hong, Dynamic load balancing using an ant colony approach in micro-cellular mobile communication systems, Advances in Metaheuristics for Hard Optimization, Natrual Computing Series, vol. 10, pp. 137-152, 2008. [10] K. Kyamakya and K. Jobmann, Location management in cellular networks: classication of the most important paradigms, realistic simulation framework, and relative performance analysis, IEEE Transactions on Vehicular Technology, vol. 54, pp. 687-708, 2005. [11] T. Mach and R. Tafazolli, Mass mobility signaling congestion avoidance mechanism using randomized time distribution of cell reselections, in Proc. of International Conference on Telecommunications (ICT 09), pp. 238-242, 2009. [12] S. Mitra and S. DasBit, On location tracking and load balancing in cellular mobile environment: a probabilistic approach, in Proc. of the 5th International Conference on Electrical and Computer Engineering (ICECE 08), pp. 121-126, 2008. [13] Mitsubishi Electric, Tracking areas sizes and tracking

[14]

[15]

[16]

[17]

[18] [19] [20]

[21]

[22]

[23]

[24]

[25]

area list optimization, Technical Report in 3GPP TSG RAN WG3 Meeting, R3-071931, 2007. S. Modarres Razavi and D. Yuan, Performance improvement of LTE tracking area design: a re-optimization approach, in Proc. of the 6th ACM International Workshop on Mobility Management and Wireless Access (MobiWac 08), pp. 77-84, 2008. S. Modarres Razavi, D. Yuan, F. Gunnarsson, and J. Moe, Exploiting tracking area list for improving signaling overhead in LTE, in Proc. of IEEE Vehicular Technology Conference (VTC 10), 2010. S. Modarres Razavi, D. Yuan, F. Gunnarsson, and J. Moe, Dynamic tracking area list conguration and performance evaluation in LTE, in Proc. of IEEE Global Communications Conference Workshop (GLOBECOM Workshop 10), 2010. S. Modarres Razavi, Tracking area planning in cellular networks, Thesis, Link oping University, 2011. http://liu.diva-portal.org/smash/get/diva2:402919. IST-2000-28088 Momentum project. http://momentum.zib.de. Motorola, Long Term Evolution (LTE): A technical overview, Technical White Paper, 2007. NEC, Tracking area concepts and inuence on paging load, Technical Report in 3GPP TSG RAN2 Meeting, R2-070655, 2007. QUALCOMM Europe and T-Mobile, New SON user case: tracking area optimization, Technical Report in 3GPP TSG RAN WG3 Meeting, R3-071594, 2007. D.W. Tcha, T.J. Choi, and Y.S. Myung, Location-area partition in a cellular radio network, Journal of the Operational Research Society, vol. 48, pp. 1076-1081, 1997. T. Winter, Mobility management and network design for UMTS, in Proc. of IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 04), 2004. V.W.S. Wong and V.C.M. Leung, Location management for next generation personal communication networks, IEEE Network, vol. 14, pp. 18-24, 2000. ZTE, Self optimization for tracking areas, Technical Report in 3GPP TSG RAN WG3 Meeting, R3-071415, 2007.

291