On the Design and Capacity Planning of a Wireless Local Area Network

Ricardo C. Rodrigues, Geraldo R. Mateus, Antonio A. E Loureiro

f Computer Science Department o Federal University of Minas Gerais Caixa Postal 702 30123-970 Belo Horizonte, MG Brazil Email: (rick,mateus,loureiro) @dcc.ujing.br
Abstract

The design of a wireless local area network (WLAN) has two major issues: determining the best placement of base stations (BS) and assigning the frequency channels for those stations. The correct BS placement minimizes installation costs. The adequate channel assignment reduces signal interference and improve network throughput. This paper reports a real experience where we applied the concepts of two classical outdoor problems namely the optimal base placement problem and the fixed channel assignment problem to build a WLAN in an indoor environment.

Keywords
Network Design, Capacity Planning, Network Operation, WLAN, Optimization Models

1. Introduction
The great advance on wireless communications during the last years must continue in the future. Cellular phones and paging systems are good samples of wireless applications and several others are being developed. The tendency is that these applications will not be offered isolated as today but integrated. The goal is to provide cellular telephony, paging, Web access and e-mail, among other services, integrated in a unique computational device. Although wireless networks are clearly the present goal, the study of the wireless technology does not apply only to the current systems, but to a large variety of applications that will be available in the coming years.
0-7803-5864-3 0 2000 IEEE

336

Session Eight

Network Design and Planning

The first experiences in wireless local area networks (WLANs) used proprietary hardware [ 1,2]. The recent publication of IEEE 802.11 [3] specificationregulating the media access method in WLAN started an industry technology competition to adapt their products to the new standard. The first IEEE 802.11 devices were available in the market by early 1998 and some projects have just begun to migrate to these new products [4]. In this work, we report a real experience where we installed an IEEE 802.1 1 compliant WLAN into an indoor environment. We studied two phases of the installation process: choosing the best placement for the base stations (BS) and assigning the frequency channels to these stations. This work intends to share our experience on installing an IEEE 802.11 WLAN and stimulate information exchange between various research groups. This paper is organized as follows. Section 2. describes the acquired hardware we have used. Section 3. presents the installation process phases. Section 4. discusses the model adopted. The results obtained are presented in Section 5. and the conclusions in Section 6..

2. Hardware Description
We used three WavePOINT-I1 Access PointsLucent Technologies (the base stations) that work as Ethernet bridges by receiving data from the wired backbone through a UTP (or BNC) connection and bypassing the received packets through a wireless network interface card (NIC). The mobile users receive these packets through another NIC attached to its laptop or other computational device. The hardware employed in the network is IEEE 802.1 1 compliant. This standard defines the media access method and the physical layer specifications of a WLAN. It defines two modulation techniques: DSSS (Direct Sequence Spread Spectrum) and FHSS (Frequency Hopping Spread Spectrum). These modulation techniques are incompatible with each other and it should be observed when expanding the wireless network. Our wireless network operates at 2.4 GHz (microwaveband), provide a bandwidth of 2 Mbps with fallback to 1 Mbps, DSSS modulation and CSMNCA access method (Carrier Sense Multiple Access/Collision Avoidance) with ACK. The avoidance collision mechanism is needed in a wireless network as a host cannot detect a collision after it sends a packet, different from an Ethernet network.

3. The Installation Process

In this section we present the phases involved in the installation process and the developed optimization model.

Design and Capacity Planning of Wireless Local Area Network

331

3.1 Mapping The Demand Area

As a pilot experience and due to a limitation of having only three base stations available, we considered the third and fourth floors of our building as our demand area. Figure 1 shows the demand area mapping of each floor. The lighter areas correspond to aisles. They were excluded from the demand area because they are in a circulation area and their signal levels are usually better than inner rooms (fewer obstacles). Thus, the impact caused by its removal is irrelevant. The dark areas correspond to faculty rooms and labs. They are, effectively, our demand areas.
Caption: -Aisles

- Demand Area - Base Station Candidate Position

Figure 1: Demand area of 3rd e 4th floors Laboratory rooms mainly occupy the third floor while faculty rooms mainly occupy the fourth floor. The set of rooms forms a rectangular area with an inner free space. The demand area mapping was obtained by dividing the total area into small quadrangular pieces of demand points. Due to the variable room size, we defined a square size unit so that the area of each room is an integer multiple of this square area. The idea is to avoid situations where a unit belongs to two adjacent rooms. With this mapping, we got 1144 square units of 0,7O x 0,70 m2, being 592 on the 3rd floor and 552 on the 4th floor.

338

Session Eighr

Network Design and Planning

3.2 Choosing Candidate Locations

In the next stage, we had to choose candidate locations to the BS. A good candidate square must offer low cost of installation and good attendance area. Questions like physical security, available infrastructure and flexibility are also relevant. The process of choosing candidate locations with different characteristics let us understand the BS behavior and its reach, trying to discover locations that could give us a better coverage. Thus, we chose three candidate locations on laboratories at the third floor, one on a faculty room of the 4th floor and two on the aisles (3rd and 4th floors). Chosen locations are shown in Figure 1.

3.3 Signal Measurement

After choosing the candidate locations, we must calculate or measure the signal level received from each candidate BS at each demand point. On outdoor environments, this is usually calculated through signal prediction algorithms. In our work, however, we preferred to measure the signal received at each demand point. The great number of demand points (1 144) and the small area of each point (0,49 m2) would make very difficult to measure the signal level at each demand point. The problem was solved by grouping the demand points into small but larger groups. Typically, we formed groups of four or six demand points. For each group, we performed only one measurement and we assumed that the measured signal level has the same value for all group elements. With this simplification, the number of necessary measurements for each BS was reduced from 1144 to 253. The communication signal quality is measured in decibel (db). A higher value means a better signal quality. Values over 20 db indicate excellent quality. Values between 11 and 20 db indicate acceptable signal quality and values under 10 db means poor or no communication capacity. The signal measurement was done using the software WaveManagerKlient IEEE, implemented by the WavePOINT-I1 manufacturer. This software lets us register the received signal from each BS simultaneously and save the information in a log file for later treatment. During the measurement, however, we verified that the signal level at each point was very sensitive to obstacles and highly dependent of the mobile unit orientation. Given the same demand point as reference, if we turn the mobile unit to left or right by 90 degrees, we can get a completely different signal level. Figure 2 shows the signal level variation from two distinct BS that arrives at the same demand point while the mobile unit is turned around by 360 degrees. The signal level variability according to the mobile unit orientation introduced a new component to the problem. It is not sufficient to measure the signal level of the BS at each point, but it is also necessary to choose or calculate the signal value that best represents the signal quality at that point. We defined a method for measuring the signal that could provide the most representative signal level at each demand point. The data was collected while the mobile

Design and Capacity Planning of Wireless Local Area Network

339

Signal Level Variation Along Time

35
30
25
20 15

10

5i

Figure 2: Signal level variation along time

unit was turned around slowly until it completed 360 degrees. Thus, we tried to represent all possible orientations of the mobile unit for that point. With this methodology, we obtained approximately 80 signal level values from each BS for each demand point. The next step was to analyze and treat the collected information. The BS signal level at a demand point is a discrete function. This behavior is because the signal is very susceptible to environment variations. We defined, thus, a methodology for treating the signal samples. For each set of BS signal samples at a demand point, we sorted the values obtained at that point. After that, we discarded 40% of the samples, 20% referring to the lower sorted values and other 20% referring to the higher sorted values. Although this discard can look very high, it is important to minimize the influence of sporadic values of the measured signal on the point. The average of the remaining 60% signal samples was the value used to represent the signal on that point. The signal level distribution on an indoor environment can be observed through signal quality maps. These maps show the arriving signal level from each candidate BS for each demand point. Figure 3 shows the signal quality map for a base station located on an aisle at 3rd floor. The X point indicates the BS location. Dark colors represent the best signal level inside rooms and laboratories. Aisles signals were not measured. Note that the signal level inside a room is approximately constant. It occurs because the biggest signal obstacles are walls, metal materials (elevators) and the concrete between floors and rooms. This fact shows that our approach of grouping demand points into small but larger groups was valid. Another obvious factor responsible for

340

Session Eight

Network Design and Planning

Caption: -Aisles

- Demand area with signal > 2Wb E - Demand area with signal between 10 and ZWb

- Demand area with signal < 1Wb

- Base Station Candidate Position

Figure 3: Sample of a signal quality map signal level fall is the distance between the base station and the demand point.

3.4 Interference
When installing the WLAN, it is necessary to specify the channel frequency to be used by each BS. The stations used on our work allow us to choose one and only one channel between 11 available. If two adjacent base stations use the same or near channels, they can interfere and degrade the network performance. We ran some tests to verify the influence of interference on an IEEE 802.1 1. We used two laptops to download two identical files of approximately 20 Mbytes from distinct servers of our wired LAN. We ran various tests and verified that the minimal channel distance between two interfering base stations on our environment was three. Some scenarios tested are described below: Scenario 1. Each laptop downloaded the file on different times through the same base

Design and Capacity Planning of Wireless Local Area Network

34 1

station; Scenario 2. Each laptop downloaded the file simultaneously through the same BS; Scenario 3. Each laptop downloaded the file simultaneously through different BS using the same channel frequency. The distance between the BS was far away enough to avoid interference between them; Scenario 4. Each laptop downloaded the file simultaneously through different BS using the same channel frequency. The BS was close enough so that there was interference between them; Scenario 5. Each laptop downloaded the file simultaneously through differentBS with channel distance equal to 1. As the BS was very close, there was interference between them; and Scenario 6. Each laptop downloaded the file simultaneously through different BS with channel distance equal to 3. Although the base stations were very close, the distance channel of 3 avoided interference between them. The results provided in Table 1 for each scenario show that the presence of interference degrades the network performance. For only two laptops in different interfering cells, the network suffered almost 45% of degradation. An adequate channel assignment is very important for avoiding interference between the BS and thus improving the wireless network throughput.

Scenario Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6

Laptop1 151.32 86.65 151.73 94.24 96.37 151.09

Laptop2 172.92 93.19 170.77 92.25

90.92
172.39

Table 1: Transfer performance (Kbps) according to the scenario operation The solution adopted to the channel assignment problem consists in sorting the communication channels according to its frequency band and allocating them to the base stations obeying the minimal channel distance. Operating on distant frequency bands, the stations do not cause interference with each other.

4. Optimization Model
We developed an integer linear programming optimization model to determine the best placement for the base stations and the adequate channel allocation that would reduce signal interference and improve our WLAN throughput. We do not consider user density as in cellular system models due to two reasons.

342

Session Eight

Network Design and Planning

The first one is the nature of a WLAN connection. In cellular phone systems, after establishing a connection, it is assumed that the user is talking and its traffic cannot suffer considerable delays. Each frequency channel can be used by only one user during its call. Thus, when projecting a cellular phone system, the user density clearly determines the number of channels allocated to each cell. In W A N systems, however, the communication channel can be shared among various users and some delays are acceptable. The presence of various users in a single cell obviously degrades the network throughput, but it still allows network accessibility. The second reason is that the initial goal of this work was to cover the maximum demand area. As a pilot installation and having very few mobile users, the user density was not covered at this phase of our study.

4.1 Optimal BS Placement Problem

The optimal base station placement problem consists in distributing the available BS through the demand area to assist the desired covering goal using the minimal number of base stations. The covering goal varies accordingly to the design requirements. Usual goals are the total coverage of the demand area, partial coverage with maximum economical return and coverage with demand supply guaranteed. The solution is found using integer linear programming models able to inform the minimal number of base stations needed to meet the desired coverage. This approach is traditionally used in cellular phone systems. In our environment, however, we have a fixed number of available base stations: three. Our problem is to know what is the best BS placement to meet our demand area and what is the percentage of the total area covered with these base stations. Our solution was to develop an integer linear programming model that, given the number of available base stations and the signal level from each candidate BS at each demand point, provides the best station placement that maximizes the total area covered. This approach provides more flexibility than the traditional one:
0

Allows us to inform the number of available base stations. This is a typical situation where there is a limited budget; Allows us to estimate the number of needed base stations depending on the desired coverage. In pilot installations, it can be enough to cover only part of the demand area; Allows us to assess the obtained gain when installing new base stations in the system; and In case a BS fails, allows us to assess the placement changes to be applied to the working base stations so that the WLAN can work, even without its full capacity, until the BS is reinstalled.

Design and Capacity Planning of Wireless Local Area Network

343

4.2 Fixed Channel Assignment Problem

The fixed channel assignment problem consists in allocating frequency channels to base stations to attend the demand at each cell and minimize interference between near base stations. On traditional cellular phone systems, it is allocated various noninterfering channels to each BS. The number of allocated channels to each BS is proportional to the demand on that cell. In our system, however, we must assign only one frequency channel to each BS, conserving the minimal channel distance between near base stations. Despite the fact that there are various solutions proposed for this problem [ 5 ] , we decided to develop a very simple model that could give us a good solution efficiently. The main idea is to represent our system through a non-directed graph where nodes represent the candidate base stations. The existence of an edge between two nodes means that the nodes cause interference between them and therefore, if both are installed, they must obey the minimal channel distance.

4.3 The Optimization Model

The developed integer linear programming model is described as follows. Let M be the demand points set, N the candidate BS set, S the number of availablebase stations and Ak the set of mutually exclusive base stations. Given two or more mutually exclusive BS, only one can be selected (installed). Let aj be a boolean variable that assumes value 1 if the station j is selected and 0 otherwise. Also, let Nj be the set of points i the attendance priority attended by station j, sij the signal of station j at point i, w of point i, and z i the area of point i. Let also xij be a decision variable that assumes value 1 when the point i is assigned to station j and 0 otherwise. Let also K be the ordered set of available channels, Cik the decision variable that assumes value 1 when channel k is assigned to station i and 0 otherwise, and let d be the minimal distance channel for the system. The model is given by:

344

Session Eight

Network Design and Planning

subject to:

Cik

+
1E { { k

- d + 1.. . k d - 1 } 1 ( k - d + 1 2 1) A ( k + d - 1 5 IKl))

c +

cjl

1,Vk E K , V ( i , j ) E

(7)

The objective function (1) must be maximized as a function of the points that have higher signal level, attendance priority and physical area. The attendance priority stimulates the attendance of the most important demand points, such as points where it is more usual the use of portable computers, meeting rooms, etc. In our experiments, all the demand points had the same attendance priority. Constraints (2) to (5) refer to the BS placement problem. According to constraint (2),a point can be attended by only one station. Although this situation does not occur in the reality, this constraint is necessary to let the chosen stations cover the largest number of distinct possible points, increasing the total area coverage. Constraint (3) limits the number of available stations for installation. Constraint (4)states that only installed base stations can attend demand points. Constraint (5) limits in only one the number of mutually exclusive stations that can be selected for each set Ak. Constraints (6) and (7) refer to the fixed frequency channel assignment problem. Constraint (6) limits the number of channels to be assigned to each BS. Constraint (7) states that if a channel k is allocated to BS i, the interfering BS cannot use channels from range k - d 1to k + d - 1. In other words, this constraint forces base stations to obey the minimal channel distance. The computational model was implemented using the AMPL language [6] integrated to CPLEX [7]. These are well-known commercial softwares used to solve optimization problems. The model was solved in less than 10 seconds, running on a Sun SPARCstation Ultra with 128 Mbytes of RAM. The solution provided for this particular model is optimum.

Design and Capacity Planning of Wireless Local Area Network

345

5. Results
We performed two sets of tests with the base stations installed at candidate locations, thus obtaining data about six candidate base stations. In our study we simulated the use of 1 to 6 stations to observe the increase in the covered demand area despite having only three base stations available. To bypass the channel assignment constraints, we reduced the minimal channel distance from 3 to 1 during these simulations. Figure 4 shows two important characteristicsof the results. The first one is that the increase in the total covered area slows down as we increase the number of available stations. Even having high increasing factor for few base stations, the environment saturation due to the installation of various BS tends to stabilize the total covered area. The second characteristicrefers to the reduction of the area covered by each BS as we increase the number of base stations. In this way we reduce the number of users attended by each BS and, consequently,the demand for the communication bandwidth.
Covered area by each Base Station (%)

100,oo

Q Base Station 2

80,OO
60,OO
40,OO

Total Area se Station 5 Base Station 3 Base Station 1

1 2 3 4 5 6

Q Base Station 3

U Base Station 4
W Base Station 5

20,oo

Base Station 6 Total Area

0,oo
Number of Base
Stations
b

Figure 4 : Covered area by each antenna or base station Figure 5 presents the final result. The three chosen base stations were installed on 3rd floor, exactly where the rooms are bigger and have less obstacle interference. That figure also shows the signal level at each demand point. The attendance on third

346

Session Eight

Network Design and Planning

floor is excellent while on fourth floor it is only reasonable. It occurs because the base stations are installed on 3rd floor and also due to the great variety of obstacles on 4th floor. Despite these problems, we were able to cover 86% of our demand area with signal level over 10 db using only the three base stations available.

6. Conclusions
Wireless local area networks provide great flexibility for their users and introduce several new questions not treated by wired networks. The first question is the WLAN installation itself. It is necessary to measure or predict the signal at each point of the demand area, locate the base stations and then assign one frequency channel to each base station. This work presented a very simple and efficient integrated integer linear programming optimization model for solving both base station placement and fixed frequency channel assignment problems in indoor environments as described in Section 4.. Note that the model presented can be applied either when the number of base stations is fixed, as in our case, or the network designer wants to determine the best possible coverage which will be provided by a given number of base stations. We also described in Section 3. a methodology that we followed to design o u r W A N . It seems to us that this methodology can be repeated in different environments where people want to install a WLAN and there is no previous experience as in our case. This work was developed using a real installation and employed a hardware compliant with the recently published IEEE 802.1 1 specification. This specification is very important as it encourages the development of interoperable WLAN hardware. Our future work will be the analysis of signal curves and try to develop a model where it could be possible to measure the signal in only a few points and expand these measurement by interpolation to obtain the whole demand area signal levels.

References
[ 13 http:Nwww.ini.cmu.edu/WIRELESS/Wireless-Infrastructure.html. Wireless An-

drew Project at Carnegie Mellon University, USA. [2] B. J. Bennington and C. R. Bartel. Wireless Andrew: Experience Building a High Speed, Campus-Wide Wireless Data Network. Third Annual ACM/IEEE Intemational Conference on Mobile Computing and Networking (Mobicom), Budapeste, Hungary, pages 5545,1997. [3] IEEE 802.1 1. IEEE Standard for Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. 1997.

Design and Capaciry Planning of Wireless Local Area Network

347

[4] B. J. Bennington and C. R. Bartel. Wireless Andrew: Building a High Speed, Campus-Wide Wireless Data Network. Special Issue on Wireless Internet and Intranet Access of the ACMBaltzer Joumal Mobile Networks and A(1ications (MONET}},1998.

[5] R. B. A. Eisenblatter, M. Grotschel, and A. Martin. Frequency Assignment in Cellular Phone Networks. Annals o f Operations Research, 76:73-93, 1998. [6] R. Fourer, D. M. Gay, and B. W. Kernighan. AMPL: A Modeling Language for Mathematical Programming. Duxbury Press/Brooks/Cole Publishing Company, 1993. [7] CPLEX Optimization. Using CPLEX Callable Library and CPLEX Mixed Integer Library. Version 5.0. 1997.

348

Session Eight
Caption:

Network Design and Planning

- Aisles - Demand area with signal > 20db E - Demand ares with signal between 10 and 20db
@

- Demand area with signal

c 1Wb

- Base Station Candidate Position