You are on page 1of 10

3G/4G/WLAN/WMAN P L A N N I N G A N D O P T I M I Z AT I O N

AUTOMATED OPTIMIZATION OF SERVICE COVERAGE AND BASE STATION ANTENNA CONFIGURATION IN UMTS NETWORKS
IANA SIOMINA, PETER VRBRAND, AND DI YUAN, LINKPING UNIVERSITY

ABSTRACT
Deployment and maintenance of UMTS networks involve optimizing a number of network configuration parameters in order to meet various service and performance requirements. In this article we address automated optimization of service coverage and radio base station antenna configuration. We consider three key configuration parameters: transmit power of the common pilot channel (CPICH), antenna tilt, and antenna azimuth. CPICH power greatly influences coverage. From a resource management point of view, satisfying the coverage requirement using minimum CPICH power offers several performance advantages. In particular, less CPICH power leads to less interference and higher system capacity. Optimal CPICH power, in its turn, is highly dependent on how the other two parameters, tilt and azimuth, are configured at radio base station antennas. Optimizing antenna tilt and azimuth network-wise, with the objective of minimizing the CPICH power consumption, is a challenging task. The solution approach in this article adopts automated optimization. Our optimization engine is a simulated annealing algorithm. Staring from an initial configuration, the algorithm searches effectively in the solution space of possible configurations in order to find improvements. The algorithm is computationally efficient; thus, we can optimize large networks without using excessive computing resources. We present a case study for a UMTS planning scenario in Lisbon. For this network, automated optimization saves up to 70 percent of the CPICH power used in the reference network configuration. In addition, the optimized network configuration offers significant performance improvement in terms of fewer overloaded cells and lower downlink load factor.

487.0

488.0 X, [km]

Deployment and maintenance of UMTS networks involve optimizing a number of network configuration parameters. The authors address automated optimization of service coverage and radio base station antenna configuration.

INTRODUCTION
Automated radio network planning and optimization in Universal Mobile Telecommunications System (UMTS) networks have attracted

increasing interest in the last several years [13]. Network planning refers to the process of designing network structure and determining network elements subject to various design requirements. Network optimization amounts to finding a network configuration to achieve the best possible performance. Automated network planning and optimization allow operators to better deal with the complexity of UMTS network design, a task that is often beyond the reach of a manual approach. In addition to making the network design process time-efficient, planning tools incorporating automated optimization can significantly reduce network deployment and maintenance costs. UMTS network optimization involves a tradeoff between many factors, such as service coverage, network capacity, quality of service (QoS), equipment costs, and expected revenues from network operation (Fig. 1). From a long-term perspective, the primary objective of an operator is to maximize revenue. This objective plays a major role in the network definition phase. Issues involved in this phase include the choice of technology and its expected evolution, deployment strategy, service specification, as well as coverage and capacity requirements. Each phase of a network life cycle also involves short-term objectives and goals. For example, minimizing equipment cost is very important in network dimensioning when major equipment investments are required. Equipment cost is also of particular importance in network expansion and upgrade. In detailed planning as well as network operation and maintenance, the type and amount of equipment are typically given, and the focus is on optimizing network configuration and parameter setting to achieve the best possible network performance. Optimization problems arising in various phases of a network life cycle differ not only in their objectives, but also in the set of design parameters with which they deal and the level of detail. Average site-to-site distance, site location, sectorization, antenna type, and average antenna height are usually addressed in the dimensioning phase [4, 5], whereas detailed planning focuses

16

1536-1284/06/$20.00 2006 IEEE

IEEE Wireless Communications December 2006

Network life phases Definition Dimensioning Detailed planning Operation and maintenance Expansion/ upgrade

Optimization goals and constraints Equipment and other costs Service coverage Revenues from network operation

Design parameters Re-configuration cost

Trade-off

Site-to-site distance Site location Number of sectors Antenna type Antenna height Antenna azimuth Antenna tilt CPICH power Handover parameters

Network capacity

Quality of service (QoS)

Automated optimization

Satisfying service coverage requirements using a minimum amount of CPICH power offers several performance advantages. Since the maximum transmit power available at RBS is constant, less power consumption by CPICH makes more power available to traffic channels.

Flexibility

I Figure 1. Network design and optimization. on many radio base station (RBS) configuration parameters [1]. In this article we address automated optimization in detailed planning after network rollout. In detailed planning, making major changes in network topology and layout is typically not an acceptable option for an operator. Instead, the goal is to optimize some key configuration parameters, including antenna azimuth, antenna tilt, pilot power, and soft handover parameters [2, 6, 7]. These parameters are at different levels of flexibility. For example, changing antenna azimuth requires higher effort and cost than changing electrical tilt. We focus on coverage planning and capacity optimization in the downlink (DL) direction. In particular, we aim for full service coverage and improved cell isolation in order to decrease interference and allow more users to be served. We consider three key RBS configuration parameters: common pilot channel (CPICH) power, antenna tilt, and antenna azimuth. CPICH power determines service coverage. To utilize power resources efficiently, CPICH should not be set to more than what is necessary to guarantee coverage. The optimal level of CPICH power, in turn, depends on the other two parameters (i.e., tilt and azimuth of RBS antennas). We present a system model and a solution approach for the above optimization problem. The system model captures the relation between CPICH power and the other two configuration parameters. The core of our solution approach is a simulated annealing algorithm. Being effective as well as computationally efficient, this algorithm does not demand excessive computing resources to optimize large UMTS networks. We organize the remainder of the article as follows. In the next section we discuss the aforementioned three RBS parameters in more detail. We present the system model and formalize the optimization problem. The solution approach is presented. A case study for a UMTS planning scenario in Lisbon is presented. Finally, we draw some conclusions and discuss extensions.

RADIO BASE STATION CONFIGURATION PARAMETERS


CPICH TRANSMIT POWER
In a UMTS network a cell announces its presence through the CPICH, a fixed-rate downlink physical channel carrying a predefined bit/symbol sequence. Typically, each cell has one CPICH. CPICH signals, or pilot signals, are used by mobile stations for channel quality estimation, cell selection/reselection, and handover evaluation. From a resource management standpoint, satisfying service coverage requirements using a minimum amount of CPICH power offers several performance advantages. Lower CPICH power means less power consumption by common channels and more power available to traffic channels, as well as reduction of cell overlap and DL interference. Since the maximum transmit power available at RBS is constant, less power consumption by CPICH makes more power available to traffic channels. This benefit becomes particularly significant if the common practice of setting the power levels of some common channels, such as synchronization channel (SCH), paging indicator channel (PICH), common control physical channel (CCPCH), and acquisition indicator channel (AICH), relative to that of CPICH, is adopted in the network [8]. Moreover, excessive pilot power adds to DL interfer-

IEEE Wireless Communications December 2006

17

overlap, in turn, improves inter- and intracell interference, power consumption, and capacity. There are various tools that simplify the task of steering azimuth. In most situations, however, adjusting azimuth has to be done manually on site.
Y

PREVIOUS WORK
In recent literature on UMTS network optimization, some or all of the aforementioned three configuration parameters are used in order to improve network performance. Garcia-Lozano et al. [11] optimize CPICH power and electrical tilt with the objective of maximizing capacity and balancing cell load. A similar optimization problem was addressed by Gerdentsch et al. in [6]. Eisenbltter et al. [1] presented comprehensive optimization models for UMTS network optimization, and proposed heuristic algorithms based on integer programming. The models use CPICH power, antenna tilt, antenna azimuth, as well as some other configuration parameters as decision variables.

(a)

(b)

I Figure 2. 3D interpolation of antenna diagrams: a) 6 mechanical downtilt; b) 6 electrical downtilt. ence, and increases cell overlap and potential pilot pollution area. On the other hand, coverage problems will arise if the CPICH power becomes too low. Therefore, to address the CPICH power parameter, a reasonable approach is to minimize CPICH power subject to coverage guarantee. Research on adjusting CPICH power for the purpose of controlling cell size and balancing cell load has been presented in [9, 10].

SYSTEM MODEL
Let I = {1, , I} and J = {1, , J} denote the set of cells and the set of bins in a UMTS network, respectively. A bin is a small square area. We use a grid of bins to model the entire service area. We assume the same signal propagation conditions in a bin. Thus, the bin size determines the resolution and amount of signal propagation data. Assuming that each cell uses a single directional antenna, we define set K i to represent all possible antenna configurations of mechanical tilt, electrical tilt, and azimuth of cell i, i I. A network-wise configuration is denoted by a vector k = (k1, k2, , kI), where ki Ki; i I. Obviously, k K1 K2 KI. Mobile stations (MSs) continuously scan for CPICH signals. For each detected CPICH signal, a MS measures the carrier-to-interference ratio (E c /I 0 ). This is the ratio between the received chip energy of CPICH and the total received power spectral density. To provide service coverage to a MS, there must exist at least one cell, for which the measured value of CPICH E c /I 0 meets a threshold 0 defined by the MS. (We assume that the threshold is the same for all MSs in a bin.) For a given network configuration k = (k1, k2, , kI), the Ec/I0 requirement of an MS in bin j with respect to cell i is defined in Eq. 1:
k Pi giji l I k PlTot gljl + j

ANTENNA TILT
Antenna tilt is the angle of the main beam of the antenna below the horizontal plane. Positive and negative angles are also referred to as downtilt and uptilt, respectively. Antenna downtilt results in stronger signals from the home cell and less interference to neighboring cells. However, coverage may suffer if antennas are downtilted excessively. Antenna downtilt can be adjusted mechanically and/or electrically. Mechanical tilting means to adjust the physical angle of the brackets in which an antenna is mounted. Electrical tilt does not change the physical angle of an antenna, but adjusts the radiating currents in the antenna elements to lower the beam in all horizontal directions. Thus, electrical tilt implies changing the antenna radiation pattern, which is not affected if mechanical tilt is used. Figure 2 shows signal propagation (power gain) in 3D space for an antenna downtilted by 6 using mechanical and electrical downtilt. In performance, mechanical downtilt is more useful in a coverage-limited environment, whereas electrical downtilt is a better option in a capacity-limited environment [3]. Recently, electrical downtilting has been significantly simplified by using remote electrical tilt (RET) controllers, which eliminates costly site visits. Electrical and mechanical tilt can be combined. The resulting flexibility is one of the advantages of improving network performance via antenna downtilt. Due to electro-magnetic compatibility (EMC) regulations in many countries, however, the range of possible tilt values can be quite limited in practice, especially in city environments.

0 Pij (k ) = 0

l I

k PlTot gljl + j k glji

. (1)

ANTENNA AZIMUTH
Antenna azimuth is the horizontal angle between the north and the antennas main lobe direction. Antenna azimuth is another configuration parameter having great influence on service coverage as well as cell overlap. Sometimes, adjusting antenna azimuth can significantly reduce cell overlap without sacrificing coverage. Less cell

In Eq. 1, Pi is the CPICH transmit power of k cell i, gij i [0, 1] is the power gain between bin j and the antenna of cell i under configuration ki, PTot is the total DL transmit power in cell l, and l j is the thermal noise power in bin j. From the inequality to the left, we derive parameter Pij(k). This is the minimum CPICH power required if we let bin j be under coverage of cell i in configuration k. We aim at optimizing network performance under heavy traffic load consuming all power

18

IEEE Wireless Communications December 2006

(a) Reference (2 W, 0, 0, 120)

(b) Lower CPICH power (1 W, 0, 0, 120)

(c) Mechanical downtilt (2 W, 6, 0, 120)

(d) Electrical downtilt (2 W, 0, 6, 120)

(e) Changed azimuth (2 W, 0, 0, 140)

I Figure 3. The effect of changing design parameters on the received CPICH signal strength. available at RBSs. Thus, in our model PTot equals i the maximum power Pimax available in cell i (i.e., we assume the worst-case interference scenario). From a planning point of view, it is reasonable to consider this scenario when dealing with service coverage. As an implication of the assumption of worst-case interference, P ij (k) is a function in k only. The coverage area of a cell, on the other hand, is determined by both k and the cell CPICH power we choose to set. In Fig. 3 we illustrate how CPICH power and antenna configuration of a cell affect the received CPICH signal strength. In the figure, the CPICH signal strength is represented by color. The values in parentheses denote CPICH power, mechanical downtilt, electrical downtilt, and azimuth, respectively. The optimization problem we address can be formalized as: find a configuration vector and a CPICH power vector (one power value per cell), such that every bin is in the coverage area of at least one cell, and the total CPICH power is minimized. We have chosen CPICH power as the objective function because CPICH power influences power consumption of common channels, cell size and load, and reflects the total interference in the network. This optimization problem is very complex and extremely difficult to solve. We tackle the problem in two steps. In the first step we restrict CPICH power to be uniform in all cells. Thus, in this step we consider finding a configuration vector k* to minimize the uniform CPICH power. One particular reason for treating uniform CPICH power, in addition to problem decomposition, is that it is a common practice in currently deployed UMTS networks. The output of the first step is a configuration vector. In the second step we apply the optimization algorithm in [9] to optimize nonuniform CPICH power under this configuration vector. For configuration vector k, the minimum uniform power to provide coverage in all bins is given by Eq. 2 derived as follows. For bin j, it is optimal to let it be covered by the cell having minimum P ij(k). This gives the min-operation. The max-operator is then applied because CPICH power is restricted to be uniform. P (k ) = max min Pij (k ).
j J i I

term used to denote a class of difficult problems in the theory of problem complexity). For such problems, guaranteeing optimum requires, in the worst case, an enumeration of all possible solutions (i.e., configurations). The number of possible antenna configurations is enormous in a network of 100 cells and five candidate antenna configurations per cell, the total number of possible configuration vectors is 5100. Expecting to find the optimum using reasonable computing resources is therefore unrealistic. In the next section we present an algorithm aimed at quickly finding a near-optimal solution.

SOLUTION APPROACH
To solve the problem of minimizing uniform CPICH power, we use simulated annealing (e.g., [12]), a probabilistic meta-heuristic algorithm for searching in the solution space of hard optimization problems. This iterative algorithm simulates the physical process of annealing, in which a substance is cooled gradually to reach a minimum-energy state. In simulated annealing, a sequence of solutions to an optimization problem is generated, and the best among them becomes the output. A new solution is generated by modifying part of the current one. The new solution is accepted if it has a better objective function value (corresponding to lower energy in physical annealing). The algorithm also allows occasional non-improving moves with some probability that decreases over time. A nonimproving move means to go from one solution to another with a worse objective function value. This type of move helps to avoid getting stuck in a local optimum. The probability of accepting a non-improving move is determined by two factors. The first is the amount of worsening value in objective function. A non-improving solution with small (positive) has a higher probability of being accepted than one with larger . The second factor is the temperature parameter T. Higher T means higher probability. Algorithmically, the probability p is calculated as p = e(/T). Parameter T is decreased gradually. Eventually, the process converges to a frozen state in which the probability of accepting inferior solutions is almost zero. In a practical implementation simulated annealing terminates if a maximum allowed number of iterations has been reached, or no improving move has been found in a number of consecutive iterations. In our implementation is the relative differ-

(2)

Finding the optimal configuration k* for uniform CPICH power is a non-convex optimization problem with numerous local optima and attraction areas varying greatly in size. Moreover, it can be shown that this problem is NP-hard (a

IEEE Wireless Communications December 2006

19

Start

Select an initial configuration k compute P(k)

k*=k, P(k*)=P(k)

Initialize algorithm parameters

Generate a new configuration k compute P(k)

P(k)<P(k)? Yes k=k, P(k)=P(k)

No

With probability p, let k=k and P(k)=P(k)

P(k)<P(k*)? Yes k*=k, P(k*)=P(k)

No

Update temperature

No

Termination? Yes Return k* and P(k*)

to cover this bin. Critical bins vary by configuration. A move in the algorithm is as follows. The algorithm selects a critical bin, say j. Cells are then sorted, in descending order of their influence in bin j (e.g., by their power gain with j). One of the first m cells in the sorted sequence is chosen randomly with some exponential probability, that is, cells with higher influence are more likely to be chosen. (In our case study m = 30.) Assume that cell i is chosen. The algorithm selects a new antenna configuration in cell i, denoted ki, such that for bin j, the result of the min-operation in Eq. 2 decreases or, if reduction is not found in a number of attempts, increases as little as possible. The new configuration is k = (k 1 , k 2 , , k i, , k I ). Note that the sets of critical bins are typically different in k and k, and P(k) may be higher than P(k). For an antenna configuration, we can apply the optimization algorithm in [9] to find a solution of nonuniform CPICH power. The algorithm minimizes the total CPICH power in the network. Embedding the algorithm in [9] into simulated annealing would require excessive computing time. Therefore, in our case study we optimize nonuniform CPICH power after we find a set of antenna configurations in the network that minimize uniform CPICH power.

A CASE STUDY
THE NETWORK
The test network originates from a planning scenario for the downtown area of Lisbon [13]. Network statistics and some parameter settings are summarized in Table 1. For the sake of simplicity, all antennas are assumed to be of the same type. Locations and heights of the antennas are given. We use a model in [1] to obtain path loss predictions for specific antenna downtilt and azimuth from path loss predictions for an isotropic antenna and antenna diagrams. For this network there is a reference scenario, in which antenna azimuth values are given and no downtilt is present. The uniform CPICH power is set to 10 percent of the maximum RBS DL transmit power in this scenario. Being designed to deal with full service coverage under worst-case interference, our system model does not assume a given traffic or user distribution. However, for performance evaluation, it is interesting to examine how a solution in antenna configuration and CPICH power behaves under a realistic traffic scenario. The snapshot we use has 2618 active users distributed among the eight services shown in Table 1. Each service has its average bit rate and Eb/N0 threshold. Those services together having most users are speech telephony (67.86 percent of users), data streaming (20.38 percent of users), video telephony (6.68 percent of users), and file downloading (4.75 percent of users).

End

I Figure 4. A flow chart of the simulated annealing algorithm.

ence in uniform CPICH power, that is, = (P(k) P(k))/P(k), where k and k are the current and new configurations, respectively. Let N denote the maximum allowed number of iterations. In addition to N, algorithm parameter specification consists of two tuples: (p0, 0) and (pN, N). Here, p0 is the probability of accepting a solution with relative difference 0 in the first iteration, and pN and N are the corresponding entities after N iterations. From (p0, 0) and (pN, N ), we derive two temperature values: T 0 = 0 /ln p 0 , and T N = 0 /ln p N . The algorithm uses T0 as the initial temperature, which is then reduced by a scaling factor in every iteration such that the temperature becomes T N after N iterations. For our case study in the next section, we use 0 = 0.05, p0 = 0.5, N = 0.01, pN = 0.25, and N = 2500. A flow chart of the algorithm is given in Fig. 4, where k* denotes the best configuration found by the algorithm. Assuming configuration k = (k1, k2, , ki, , k I ) is the current solution, an iteration in the algorithm involves generating a new antenna configuration vector k from k. The objective function value P(k) is computed by Eq. 2. Among all bins, a bin defining the maximum in the max-operator (we call it a critical bin) is of particular interest, simply because a necessary condition to obtain improved uniform CPICH power is to reduce the minimum power required

PERFORMANCE METRICS
Transmit Power of CPICH and Other Common Channels CPICH power is the objective used in the optimization problem. In addition to CPICH, a cell uses a number of other common channels

20

IEEE Wireless Communications December 2006

Network statistics Number of sites Number of cells (I) Number of bins (J) Bin size Area size 60 164 52,500 20 m 20 m 4200 m 5000 m Antenna characteristics Antenna type Frequency Antenna gain Polarization Half-power beam width Adjustable electrical downtilt range Kathrein 742265 2110 MHz 18.5 dBi +45 +65 [0, 6] Parameter setting Max. DL transmit power (Pimax, for all cells) Thermal noise power (j, for all bins) Total DL common channels power, excluding CPICH, relative to CPICH power CPICH Ec/I0 target (0) Electrical downtilt range Mechanical downtilt range Possible azimuth adjustment Services Conversational Streaming Interactive Speech telephony (12.2 kb/s, 5.5 dB) Video telephony (64 kb/s, 4.7 dB) Streaming multimedia (64 kb/s, 4.2 dB) Web browsing (32 kb/s, 1.9 dB) Location-based service (32 kb/s, 3.35 dB) MMS (32 kb/s, 3.35 dB) Email (32 kb/s, 3.35 dB) File download (64 kb/s, 1.9 dB) 20 W 1.55e-14 W 80% 0.01 [0, 6] [0, 6] {0, 5, 10}

Designed to deal with full service coverage under worst-case interference, our system model does not assume a given traffic or user distribution. However, for performance evaluation, it is interesting to examine how a solution in antenna configuration and CPICH power behaves under a realistic traffic scenario.

Background

I Table 1. Network statistics and parameter setting.

(primary and secondary SCH, primary CCPCH, PICH, AICH, and secondary CCPCH). The total amount of power of these common channels is typically set in proportion to that of CPICH [8]. We assume that the total transmit power of the other common channels amounts to 80 percent

of the CPICH power. Thus, the total power consumed by all DL common channels in cell i is 1.8 times the CPICH power Pi. CPICH Coverage Overlap Cells overlap in a bin if their CPICHs all satisfy the inequality in Eq. 1.

IEEE Wireless Communications December 2006

21

When all three parameters are subject to optimization, the power reduction is 64.5 percent, the number of highly loaded cells goes down to three, and among them only one is overloaded. On average, DL load is improved by a factor of three.

We compute the percentage of bins in which at least two cells overlap and the percentage of bins where four or more cells overlap. Cell overlap is a necessary condition for soft handover (SHO). However, large overlap areas may result in high SHO overhead and high pilot pollution. Usually, it is desired to have no more than 30 percent of users in SHO. DL Load Factor We define DL load factor of a cell as the ratio between the total allocated transmit power and the maximum power available in the cell; that is,

RESULTS
The results of performance evaluation are summarized in Table 2. Following our problem decomposition approach, performance evaluation consists of two parts. In the first part we evaluate solutions obtained by the simulated annealing algorithm minimizing uniform CPICH power. To study the effect of mechanical tilt, electrical tilt, and antenna azimuth, we run the algorithm for each of the configuration parameters, as well as combinations of them. The second part of performance evaluation addresses solutions of nonuniform CPICH power obtained by the algorithm in [9]. In each of the two parts we also evaluate the solution of optimizing CPICH power for the antenna configuration used in the reference scenario (reference antenna configuration). Note that optimizing uniform CPICH power for the reference antenna configuration amounts to computing the CPICH power by Eq. 2. Along with solutions of uniform and nonuniform CPICH power, Table 2 presents performance metrics for the reference scenario. The set of possible antenna downtilt degrees is {0, 1, , 6} for both mechanical and electrical tilting. Antenna azimuth can be adjusted, in 5 steps, within the range [10, 10] relative to the initial direction. When mechanical and electrical tilt are combined, preference is given to the latter, that is, mechanical downtilt is used only if electrical downtilt has reached its 6 maximum. This is because electrical tilting, in addition to being less costly than mechanical tilting, offers better performance in a capacity-limited urban environment. The simulated annealing algorithm discussed earlier and the algorithm in [9] are designed to handle large-scale planning scenarios. The average computing time for the two algorithms does not exceed 15 and 5 min, respectively, on a moderate-speed computer (Pentium 4-M notebook with a 1.8 GHz CPU and 1 Gbyte RAM). We make several observations from the results of uniform CPICH power in Table 2. Adjusting azimuth only gives 34.5 percent power saving in comparison to the reference value. Although this is a very impressive amount, downtilting, in particular electrical downtilting, has an even larger effect. Note that downtilting leads to not only more power reduction, but also significantly smaller numbers of highly loaded and overloaded cells, as well as considerably lower average DL load. When all three parameters are subject to optimization, the power reduction is 64.5 percent, the number of highly loaded cells goes down to three, and among them only one is overloaded. On average, DL load is improved by a factor of three. So far, we have not discussed results on cell overlap, which deserves a couple of remarks. First, optimizing all three configuration parameters reduces areas covered by four or more cells by 46.8 percent a significant improvement over the reference scenario. Second, the reduction is moderate for an overlap threshold of two cells, indicating that the negative impact on potential soft handover is small. Let us examine solutions for nonuniform CPICH power. For the reference antenna con-

iDL =

PiTot Pimax

(3)

Note that earlier, PTot was set equal to Pimax. i Here, P iTot is the total power allocated for the traffic snapshot used in performance evaluation. Power PTot equals the sum of transmit power of i common channels and dedicated traffic channels needed to support all users served by cell i, that is, PiTot =
j J (i ) s S s d s s pij + 1.8 Pi . j

(4)

In Eq. 4, J(i) is the set of bins for which i is the serving cell, djs is the number of users in bin j requesting service s, S is the set of services, s is s the activity factor of service s, and pij is the minimum transmit power required to provide one user in bin j with service s. We assume that a user is always served by the cell with the highest CPICH Ec/I0. Then for antenna configuration k s = (k1, k2, , ki, , kI), pij can be calculated by the following equation for signal-to-noise ratio:
s k pij gij k k s (1 j )giji ( PiTot s pij ) + gljl PlTot + j l i

= s.

(5)

In Eq. 5, j is the orthogonality factor in bin j, and s is the threshold value for service s. Equations 4 and 5 together define a linear equation system, which can be solved by a standard solver. However, the solution may not be feasible due to power limit, that is, PTot may exceed i Pimax in some cell i, and service requests of some users have to be denied. To account for such infeasibility, we extend the equation system to an optimization model. The model uses the equation system as the constraints and has the objective function of minimizing the number of cells in which some users are denied service. Solving the model leads to a vector of DL power levels P iTot , i I. Other approaches for deriving DL powers can be found, for example, in [14]. Highly Loaded Cells and Overloaded Cells It has been suggested that in a stable network the DL cell load should not exceed 70 percent [13]. Thus, we define the set of highly loaded cells as those for which iDL > 0.7. Moreover, we count the number of overloaded cells. Cell i is overloaded if the total DL transmit power required to serve all users exceeds P imax . Because some users are denied service, however, in these cells we have iDL = 1.0.

22

IEEE Wireless Communications December 2006

Optimization scenarios

Common channels transmit power (W) CPICH Total 3.60

Coverage overlap (%) 2 cells 0.93 4 cells 0.48

Number of highly loaded cells 24

Number of overloaded cells 17

Average DL load factor

Reference scenario

2.00

0.41

Uniform CPICH power Reference antenna configuration Azimuth Mechanical tilt Electrical tilt Mechanical tilt + azimuth Electrical tilt + azimuth Mech. tilt + el. tilt Mech. tilt + el. tilt + azimuth 1.89 1.31 0.93 0.82 0.84 0.74 0.75 0.71 3.40 2.36 1.67 1.48 1.51 1.33 1.35 1.27 0.93 0.91 0.89 0.87 0.88 0.85 0.78 0.79 0.48 0.44 0.38 0.34 0.36 0.31 0.24 0.26 24 19 8 5 6 4 4 3 16 12 7 2 4 3 4 1 0.40 0.32 0.20 0.17 0.18 0.15 0.17 0.14

Nonuniform CPICH power Reference antenna configuration Azimuth Mechanical tilt Electrical tilt Mechanical tilt + azimuth Electrical tilt + azimuth Mech. tilt + el. tilt Mech. tilt + el. tilt + azimuth 0.88 0.88 0.73 0.68 0.71 0.65 0.64 0.62 1.58 1.58 1.32 1.22 1.28 1.17 1.16 1.12 0.89 0.89 0.89 0.86 0.87 0.85 0.78 0.78 0.40 0.41 0.37 0.34 0.36 0.31 0.23 0.24 16 18 7 5 5 4 4 3 9 11 6 1 4 3 3 1 0.26 0.26 0.18 0.15 0.16 0.13 0.16 0.13

I Table 2. Performance evaluation.

figuration, optimization reduces the power of common channels by 53 percent in comparison to the optimal uniform CPICH power solution for the same antenna configuration. However, the power saving reduces to 13 percent when azimuth and both tilts are used. Even more, the improvement in the other performance metrics is small when antenna configuration is optimized. Thus, the main benefit of adopting nonuniform power is less power consumption. In the last row of the table, power consumption of common channels is reduced to less than one third of that in the reference scenario. Another observation, which is probably more interesting than power reduction, is that combining electrical tilting and nonuniform CPICH power performs close to the best results in the table. Thus, most of the potential performance gain can be achieved at a very low cost! Figure 5 visualizes coverage and load statistics in some network configurations. In the figure we use the level of darkness to represent

the number of overlapping cells. (In the darkest areas in Fig. 5a 10 cells overlap.) White areas are covered by one CPICH signal. The figure also shows RBS locations and antenna azimuth. Each antenna is represented by a line. The line width reflects cell load, and the length shows the antenna tilt. The longest lines represent antennas with zero tilt. Moreover, lines in red show those cells where some users are denied service, (overloaded cells). We observe that adopting nonuniform CPICH power, by itself, reduces overlap and cell load to some extent. However, the effect on reducing excessive cell overlap and improving cell load is clearly more dramatic if we optimize antenna configurations (Fig. 5b) or combine the two options (Fig. 5d). In comparison to Fig. 5d, antenna tilt and/or azimuth are changed in many cells in Figs. 5b and 5d, suggesting that a manual approach to tuning antenna configuration can hardly achieve the same performance as automated optimization.

IEEE Wireless Communications December 2006

23

CONCLUSIONS
Approaching service and performance goals in a UMTS network necessitates optimization of network design and configuration. In this article we have addressed optimization of three configuration parameters at RBS in UMTS networks: CPICH power, antenna tilt, and antenna azimuth. How these parameters are configured heavily influences service coverage, power consumption, and cell load. For this reason, the parameters are frequently used for tuning network performance. The approach presented in this article adopts automated optimization. Two cornerstones of the approach are a system model

that captures the interplay between the configuration parameters, and an algorithm that can deal with the complexity of the optimization problem. Our case study demonstrates the benefits of the approach. Optimized CPICH power and antenna configuration offer significant performance gain. In comparison to the reference scenario, automated optimization reduces power consumption of common channels and average cell load by approximately 70 percent, and the number of overloaded cells from 17 to only one. Moreover, a significant reduction in CPICH power is a strong indication of reduced interference in the network.

4289.0 4288.5 4288.0 4287.5 4287.0 Y, [km] Y, [km] 4286.5 4286.0 4285.5 4285.0 4284.5 4284.0

4289.0 4288.5 4288.0

10 9 8

4287.5 7 4287.0 6 4286.5 4286.0 4285.5 4285.0 4284.5 4284.0 5 4 3 2 486.0 487.0 488.0 X, [km] (b) 4289.0 4288.5 4288.0 8 4287.5 7 4287.0 Y, [km] 6 4286.5 4286.0 4285.5 4285.0 4284.5 4284.0 5 4 3 2 486.0 487.0 488.0 X, [km] (d) 489.0 490.0 1 10 9 489.0 490.0 1

486.0

487.0

488.0 X, [km] (a)

489.0

490.0

4289.0 4288.5 4288.0 4287.5 4287.0 Y, [km] 4286.5 4286.0 4285.5 4285.0 4284.5 4284.0

486.0

487.0

488.0 X, [km] (c)

489.0

490.0

I Figure 5. Coverage and load statistics for selected solutions: h) reference scenario; i) uniform CPICH + mechanical tilt + azimuth; j) nonuniform CPICH; k) nonuniform CPICH + mechanical tilt + electrical tilt + azimuth.

24

IEEE Wireless Communications December 2006

Most of the performance gain can be conveniently implemented in the network through electrical tilting and adjusting CPICH power. Moreover, the algorithm presented in the article is computationally efficient. It can be used as a tool to evaluate many potential solutions of network design (RBS location, sectorization, and antenna height) in an early stage of a network planning process. There are a number of directions for extending the work in this article. In particular, uplink coverage, which is an important performance factor, is not addressed in the current work. A very interesting research topic is thus to extend the system model to take both downlink and uplink into consideration. A second topic of research interest is algorithms that optimize both CPICH power and antenna configuration for balancing cell load. Such algorithms can be used in a semionline fashion to effectively adapt network configuration to changes in user and traffic patterns.

[7] I. Siomina, P-CPICH Power and Antenna Tilt Optimization in UMTS Networks, Proc. Advanced Industrial Conf. Telecommun., July 2005, pp. 26873. [8] J. Laiho, A. Wacker, and T. Novosad, Eds., Radio Network Planning and Optimization for UMTS, Wiley, 2002. [9] I. Siomina and D. Yuan, Pilot Power Management in WCDMA Networks: Coverage Control with Respect to Traffic Distribution, Proc. 7th ACM Symp. Modeling, Analysis and Simulation of Wireless and Mobile Sys., Oct 2004, pp. 27682. [10] K. Valkealahti et al., WCDMA Common Pilot Power Control with Cost Function Minimization, Proc. IEEE VTC Fall 02, Sept. 2002, pp. 224447. [11] M. Garcia-Lozano, S. Ruiz, and J. J. Olmos, UMTS Optimum Cell Load Balancing for Inhomogeneous Traffic Patterns, Proc. 60th IEEE VTC, vol. 2, Sept. 2004, pp. 90913. [12] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, Optimization by Simulated Annealing, Science, vol. 220, no. 4598, May 1983, pp. 67180. [13] MOMENTUM Project, IST-2000-28088, http://momentum.zib.de. [14] M. J. Nawrocki, M. Dohler, and A. H. Aghvami, Eds., Understanding UMTS Radio Network Modeling, Planning and Automated Optimization: Theory and Practice, Wiley, 2006.

ACKNOWLEDGMENTS
The authors wish to thank the group of the MOMENTUM project for providing network data, and Fredrik Gunnarsson at Ericsson Research, Linkping, Sweden, for technical discussions. This work is funded by the Swedish Research Council and Center for Industrial Information Technology (CENIIT), Linkping Institute of Technology, Sweden. The work has been carried out within the framework of COST Action 293, Graphs and Algorithms in Communication Networks.

BIOGRAPHIES
I ANA S IOMINA (iansi@itn.liu.se) received her M.Sc. degree in scientific computing from the Royal Institute of Technology, Sweden, and her M.Sc. in computer science from Stockholm University in 2002 and 2004, respectively. Currently, she is working toward her Ph.D. degree in infrainformatics with emphasis on applied system optimization for wireless networks at the Department of Science and Technology, Linkping University, Sweden. Her main research interests include wireless network modeling and simulation, and radio network planning and optimization for UMTS, wireless LANs, and ad hoc networks. PETER VRBRAND (petva@itn.liu.se) received a B.Sc. degree in mathematics, and M.Sc. and Ph.D. degrees in operations research at the Department of Mathematics, Linkping University, in 1982, 1985, and 1988, respectively. Since 2001 he has held a chair in operations research at the Department of Science and Technology, Linkping University. His research interests span integer and combinatorial optimization, with applications in telecommunications, logistics, and transportation planning. D I Y UAN (diyua@itn.liu.se) received his M.Sc. degree in computer science and engineering, and Ph.D. degree in operations research at Linkping Institute of Technology in 1996 and 2001, respectively. At present he is associate professor in telecommunications at the Department of Science and Technology, Linkping University. His research interests span design, analysis, and resource optimization of telecommunication systems. He has authored or co-authored over 30 refereed articles in international journals and conference proceedings. His current research addresses network design and bandwidth allocation in UMTS systems, and resource management in ad hoc networks.

There are a number of directions of extending our work. In particular, uplink coverage is not addressed in the current work. A very interesting research topic is thus to extend the system model to take both downlink and uplink into consideration.

REFERENCES
[1] A. Eisenbltter et al., Mathematical Methods for Automatic Optimization of UMTS Radio Networks, Project rep. D4.3, IST-2000-28088, MOMENTUM, 2003. [2] A. Hglund and K. Valkealahti, Automated Optimization of Key WCDMA Parameters, Wireless Commun. and Mobile Comp., vol. 5, no. 3, May 2005, pp. 25771. [3] J. Lempiinen and M. Manninen, Eds., UMTS Radio Network Planning, Optimization and QoS Management for Practical Engineering Tasks, Springer, 2004. [4] E. Amaldi, A. Capone, and F. Malucelli, Planning UMTS Base Station Location: Optimization Models with Power Control and Algorithms, IEEE Trans. Wireless Commun., vol. 2, no. 5, Sept. 2003, pp. 93952. [5] J. Zhang et al., Mathematical Modeling and Comparisons of Four Heuristic Optimization Algorithms for WCDMA Radio Network Planning, Proc. 8th IEEE Intl. Conf. Transparent Opt. Networks), June 2006, pp. 25357. [6] A. Gerdenitsch et al., A Rule-Based Algorithm for Common Pilot Channel and Antenna Tilt Optimization in UMTS FDD Networks, ETRI J., vol. 26, no. 5, Oct. 2004, pp. 43742.

IEEE Wireless Communications December 2006

25

You might also like