Vous êtes sur la page 1sur 4

Steered Optimization Strategy for Automatic Cell Planning of UMTS Networks

Sana Ben Jamaa, Zwi Altman, Jean-Marc Picard and Arturo Ortega
France Telecom R&D, RESA/NET 38-40 rue du Gnral Leclerc 92794 Issy les Moulineaux, France zwi.altman@francetelecom.com
Abstract This paper aims at answering the question of how local an optimization problem of a UMTS network should be defined to guarantee an effective optimization by an automatic cell planner (ACP). Based on the interference matrix approach, sets of base stations, denoted as influence sets, which are the most coupled with the problematic ones, are identified and introduced to the ACP for optimization. The size of an influence set is a key issue: under-sizing the set will reduce the effectiveness of the optimization whereas over-sizing it will increase the cost of the solution implementation. Comparison of the steered optimization with a global one illustrates the robustness of the proposed approach. Keywords: steered optimizationg; automatic cell planning; interference matrices; UMTS networks

following questions: how local should an optimization problem of a UMTS network be defined and which stations should be considered for possible modifications by the ACP to guarantee a robust and efficient optimization. To this end, one should identify the base stations, denoted herein as the influence set of stations, which are coupled with the problematic one. The size of an influence set should be chosen with great care: undersizing the set will reduce the effectiveness of the optimization process whereas over-sizing it will increase the cost of the solution implementation on the field. The methodology utilized to construct the influence set is based on the interference matrix approach. The robustness of the procedure is demonstrated by assessing the performance according to QoS criteria which are distinct from the ones used by the ACP during the optimization. The paper is organized as follows: Section II presents briefly the interference matrix formulation in uplink (UL) and downlink (DL). Section III describes the construction of influence sets using the interference matrices. Numerical results for the steered optimization are described in Section IV, followed by conclusion remarks in Section V. II. INTERFERENCE MATRIX APPROACH

I.

INTRODUCTION

High quality network planning is one of the challenging tasks for UMTS mobile operators, and requires deep understanding of WCDMA mechanisms. Base stations in the network are often coupled and the modification of parameters (e.g. antenna tilt, common channel power or admission control threshold) in one station can impact performance of other stations. This feature of WCDMA networks is due to the utilization of the same frequency bandwidth by all the users, and renders planning and quality design complex [1]. The performance and profitability directly depend on the capability of the network planner to optimally parameterize the network. The complex nature of the 3G network planning has stimulated the development of Automatic Cell Planners (ACP) that automatically handle complex design tasks. The ACP optimizes the network performance in terms of capacity, coverage, quality of service (QoS) and mobility by adjusting antenna parameters (antenna type, tilt and azimuth angle) and common channel transmitted powers. Different strategies for automatic cell planning have been studied such as capacity and QoS optimization for a given coverage [2], and multiobjective optimization for capacity and coverage [3]. Steered optimization deals with a local type of optimization, i.e. adding sites in a densification process, modifying parameters of base station with poor performance or with dysfunctions. However, in WCDMA systems one cannot speak of local behavior in the strict sense due to interference based coupling effects. This paper aims at answering the

The influence sets utilized in the steered optimization process are derived using the interference matrices and are briefly described presently. The network area is divided into meshes m, over which basic WCDMA quantities such as interference can be computed. A Monte-Carlo based network simulator is utilised to compute the average state of the network in equilibrium. The interference matrices in DL, KDL, and in UL, KUL, are calculated using the following formulas [4]:
DL K ij = m i

I ij (m)
i

(1)

UL = I tot ( j ) K ij

mobile k j

PL( k , j ) PL( k , i )

(2)

where cell i,

stands for the set of all the meshes belonging to

I ij

(m)

is the DL interference generated by

communications within the j-th cell on a probe mobile on

0-7803-8887-9/05/$20.00 (c)2005 IEEE

mesh m within the i-th cell. Itot(j) denotes the total UL interference received by the j-th base station, k - the signal to noise target ratio of the k-th mobile, and PL(k,j) - the path loss between the mobile k and the station j. A perfect power control hypothesis is assumed in (2). Eq. (2) is derived as follows: The signal from mobile k is received by its serving station j with the power I tot ( j ) k , and

A.

Computation of equivalent mutual coupling

To account for both received and generated interference, the interference matrix is rendered symmetric by averaging the Kij and Kji elements. The resulting modified interference matrix, ~ K , is defined as: ~ K ij = K ij + K ji / 2 (3)

the corresponding transmitted power is I tot ( j ) k PL(k , j ) . By dividing the last expression by PL(k,i) one obtains the interference produced by the mobile k at the station i.

III.

INFLUENCE SET DERIVATION

To gain a deeper insight on the interference based coupling between the sectors1, it is of interest to see the interference histogram produced and received by a given sector, on a couple of sector basis. Figure 1 shows two histograms of the average received (white) and produced (gray) UL interferences of a sector within a dense urban environment. The average interferences comprising the histograms are taken from the i-th column and i-th row of the UL interference matrix, respectively. It can be seen that more than 10 sectors are within 10 dB from the intra-cellular interference (which is
UL the diagonal term K ii and corresponds to the highest value in the histogram).

~ Since K is symmetric, the IS can be computed using only its ~ rows (or columns). Next the matrix K is normalized with respect to its diagonal terms, to give an equivalent F-factor ~ matrix, F, which can be obtained from K by dividing each row by the corresponding diagonal element (intracellular interference term): ~ K ij (4) Fij = ~ K ii

The normalization in (4) has two advantages: first the elements of F are dimensionless and second, the dependency of the matrix elements to the network environment is reduced. The term Fij can be considered as a mutual coupling term. The IS of the station i, Si, can be computed by selecting all the stations with the corresponding i-th row (or column) elements above a given threshold (i.e. -15 dB which represents about 3% of mutual coupling) Si = j

ji

Fij > threshold

(5)

Eq. (5) is applied for both UL and DL, so that two sets are obtained: SiUL and SiDL . The influence set is then given as the union of the two sets: Si = SiUL SiDL (6)

In the subsequent optimization process only the base stations of the influence set have been set for possible modification by the ACP. IV.
Figure 1. Histogram of the average UL interference generated (gray) and received (white) by a sector in a dense urban environment.

CASE STUDY

The Influence Set (IS) of a station is the set of stations that are mostly coupled with it and have the strongest influence on its performance. Interference is particularly well suited to represent coupling between stations since it directly impacts basic quantities related to network performance, such as UL and DL loads, QoS and capacity. For a couple of stations i and j, the interference matrix provides the information of how much the station i is interfered by - and interferes with the station j, via the elements Kij and Kji respectively. In the following, the computation of ISs is derived. It is noted that other derivation of ISs are possible.
1

A heterogeneous network is considered with 132 sectors located in an urban and a dense urban environment. In a first step, a dynamic simulator emulating a real network is utilized to identify the sectors with the worst performance in terms of average blocking and dropping rates. Three problematic stations have been identified, and selected for the steered optimization, colored red in Figure 2: Station A (upper one in the Figure) on the border of the dense region, station B, in the dense region, and station C (lower part in the Figure) in the peripheral region. To compute the corresponding influence sets, a threshold of -15 dB in (5) is used for both SiUL and SiDL . Three influence sets for the sectors A, B and C are constructed (colored blue in Figure 2): - SA: 7 sectors - SB: 8 sectors with one sector in common with SA

The terms base station, station and sector have the same meaning and can be interchanged in the present paper.

- SC: 4 sectors When adding the influence sets to the three problematic sectors, a total number of 21 sectors are set for possible modifications by the ACP while the other 111 are used for evaluation only (i.e. they cannot be modified but their impact on the network is taken into account). It is noted that overlapping influence sets pose no problems since the performance is evaluated globally. Hence, coupling effects between influence sets are inherently accounted for by the optimization process. To assess the effectiveness of the steered optimization, it is compared with a global one in which all the network sectors can be modified.

0,8 0,7 0,6

Initial network Global optim Steered optim

Blocking rate

0,5 0,4 0,3 0,2 0,1 0 0 10 20 30 40 50

Station number
Figure 3. Average blocking rate for the initial (diamonds), steered (triangles) and global (squares) optimized networks in decreasing order.
0,4 0,35 0,3

Initail network Global optim Steered optim

Dropping rate
Figure 2. Three influence sets (in blue) are computed for three problematic sectors (in red).

0,25 0,2 0,15 0,1 0,05 0 0 10 20 30 40 50

Figures 3 and 4 show the average blocking and dropping rates respectively for the initial (diamonds), the steered (triangles) and global optimization (squares), in descending order. It is noted that the order of the stations in the three curves of each figure may not be preserved. One can see that the steered optimization (in which up to 21 sectors can be modified) improves a little the global network performance, as could be expected. The global optimization that allows all the 132 sectors to be modified by the ACP, considerably improves the overall performance of the network. Furthermore, one can see that the ACP particularly improves the blocking rate, which is directly related to capacity increase. Improvement in dropping rate is smaller, and can be enhanced by a different type of optimization related to Radio Resource Management (RRM) parameters [6]. Next, the results for the (filtered) blocking rate as a function of simulated time for the three problematic sectors A, B and C are depicted in Figures 5, 6 and 7, respectively. One can see that the average levels of the blocking rates for the steered and global optimized networks are comparable. For these sectors the two optimization approaches considerably improve the blocking performance. It is also important to note that for sector C, an influence set of only 4 sectors has been derived, i.e. only 5 sectors need to be introduced to the ACP for possible modifications to achieve high quality results.

Station number
Figure 4. Average dropping rate for the initial (diamonds), the steered (triangles) and global (squares) optimized network.
1

0,8

Blocking rate

0,6

0,4 Initial network Global optim 0 50 Steered optim 100 150 200 250 300

0,2

Simulated time (4xsec)


Figure 5. Filtered blocking rate as a function of simulated time for the initial (continuous line), steered (red squares) and global optimized networks for sector A.

0,6 0,5

V.

CONCLUSIONS

0,4 0,3 0,2 0,1 0 50 100 150 200 250 300

Initial network Global optim Steered optim

This paper has presented a methodology for steered optimization in the context of automatic cell planning. Steered optimization allows to determine the influence sets, namely the set of sectors that should be introduced to the ACP for possible modifications to achieve high quality results while minimizing the number of modified sectors. It can be utilized to improve performance of sectors that suffer from poor quality of service (QoS), and to efficiently densify the network. The following results are highlighted: (i) A considerable improvement in blocking rate of the problematic sectors has been obtained by the steered optimization, comparable to that achieved by the global optimization procedure. (ii) The number of sectors defined by the influence set may be small, between 4 and 8 in the examples studied in this paper. (iii) Considerable improvement brought about by the steered optimization are obtained for QoS indicators that are computed dynamically (e.g. blocking rate), whereas the ACP utilizes static based indicators to guide the optimization. This result indicates the robustness of the optimization procedure. REFERENCES
Initial network Global optim Steered optim

Blocking rate

Simulated time (4xsec)


Figure 6. Filtered blocking rate as a function of simulated time for the initial (continuous line), steered (red squares) and global optimized networks for sector B.
0,9 0,8 0,7

Blocking rate

0,6 0,5 0,4 0,3 0,2 0,1 0 50 100 150 200 250 300

Simulated time (4xsec)


Figure 7. Filtered blocking rate as a function of simulated time for the initial (continuous line), steered (red squares) and global optimized networks for sector C.

The average (filtered) blocking and dropping rates for the initial, steered and global optimized networks are summarized in Tables 1 and 2.
TABLE I. AVERAGE BLOCKING RATES FOR THE STATIONS A, B AND C FOR THE INITIAL, STEERED- AND GLOBAL OPTIMIZED NETWORKS.

Blocking rate Station A Station B Station C

Initial 0.69 0.37 0.56

Steered 0.36 0.09 0.43

Global 0.46 0.09 0.37

TABLE II. AVERAGE DROPPING RATES FOR THE STATIONS A, B AND C FOR THE INITIAL, STEERED AND GLOBAL OPTIMIZED NETWORKS.

[1] H. Holma and A. Toskala, WCDMA for UMTS, Radio Access for Third Generation Mobile Communications, John Wiley & Sons Ltd, England, 2000. [2] S. Ben Jamaa, Z. Altman, J.M. Picard and B. Fouresti, "Optimisation des rseaux mobiles utilisant les Algorithmes Gntiques", chapter 8 in Mtaheuristiques pour l'optimisation difficile, (in French) Editor Eyrolles, 2003. [3] S. Ben Jamaa, Z. Altman, J.M. Picard and B. Fouresti, "Multiobjective strategies for automatic cell planning of UMTS networks", IEEE International Symposium VTC 2004, Genoa, Italy, May. 11-14, 2004. [4] S. Ben Jamaa, Z. Altman, A. Ortega and B. Fouresti "UMTS design strategies based on indicator matrix approach", IEEE Int. Conf. on Communications, ICC 2004, Paris, France, June 20-24, 2004. [5] D. E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning, Addison-Wesley Publishing Comp. Inc., New York, 1989. [6] H. Dubreil, Z. Altman, V. Diascorn J.M. Picard and M. Clerc, "Particle Swarm optimization of fuzzy logic controller for high quality RRM auto-tuning of UMTS networks", IEEE International Symposium VTC 2005, Stockholm, Sweden, 29 May-1 June 2005.

Dropping rate Station A Station B Station C

Initial 0.35 0.085 0.09

Steered 0.15 0.02 0.07

Global 0.135 0.035 0.09

Vous aimerez peut-être aussi