Spatial traffic estimation and characterization for mobile communication network design by Tutschku_ K.; Tran-Gia_ P by farhatmasood

VIEWS: 11 PAGES: 8

									804                                                        IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 16, NO. 5, JUNE 1998




       Spatial Traffic Estimation and Characterization
        for Mobile Communication Network Design
                        Kurt Tutschku, Student Member, IEEE, and Phuoc Tran-Gia, Member, IEEE



   Abstract— This paper presents a new method for the estima-          use in mobile system engineering. Hence, the demand-based
tion and characterization of the expected teletraffic in mobile         design of mobile communication systems requires an efficient
communication networks. The method considers the teletraffic            traffic estimation and characterization procedure which is at
from the network viewpoint. The traffic estimation is based
on the geographic traffic model, which obeys the geographical           the same time both accurate and simple to use. Such a method
and demographical factors for the demand for mobile com-               will be introduced in the following presentation.
munication services. For the spatial teletraffic characterization,         The paper is organized as follows. In Section II we in-
a novel representation technique is introduced which uses the          troduce a new demand-based and integrated mobile network
notion of discrete demand nodes. We show how the information           planning approach. In Section III we provide first an overview
in geographical information systems can be used to estimate
the teletraffic demand in an early phase of the network de-             on traffic source models which are used so far in mobile
sign process. Additionally, we outline how the discrete demand         network design. In Section III-A we define a spatial traffic es-
node representation facilitates the application of demand-based        timation model which takes into account the geographical and
automatic mobile network design algorithms.                            demographical factors for the expected teletraffic in a service
  Index Terms—Clustering methods, communication system per-            region. Subsequently, we introduce the demand node concept,
formance, facility locations, land mobile radio cellular system,       which is a novel and efficient technique for the representation
network planning, traffic estimation.                                   of the spatial distribution of the teletraffic using discrete
                                                                       points. Section IV outlines a traffic characterization procedure
                        I. INTRODUCTION                                which is capable of deriving a demand node distribution from
                                                                       publicly available geographical data. To generate the demand

T     HE design of future generation wireless communication
      networks is facing three major challenges. First, there is
a tremendous increase in the demand for mobile communi-
                                                                       nodes, we present a recursive partitional clustering algorithm.
                                                                       Section V demonstrates how the demand node concept can be
                                                                       applied for locating base stations. Section VI summarizes the
cation services. Second, due to deregulation acts, competition         presentation.
between mobile service providers is increased and customers
can switch almost instantaneously to the most economical                              II. DEMAND-BASED SYSTEMATIC
provider. And third, new multiplexing and access technologies                           MOBILE NETWORK PLANNING
like space division multiple access (SDMA), code division
multiple access (CDMA), or wireless local loop (WLL) require              The major drawback of commonly used conventional mo-
new network planning methods in order to obtain an efficient,           bile network planning methods is their focus on radio fre-
economic, and optimal wireless network configuration.                   quency (RF) aspects. Their main objective is to provide
   The primary task of mobile system planning is to locate and         a sufficient radio signal coverage throughout the complete
configure transmission facilities, i.e., base stations or switching     service area, cf. [6]. However, economical aspects of system
centers, in the service area of the network and to interconnect        deployment and operation are either not addressed effectively
these nodes in an optimal way. To achieve an efficient con-             or are considered only in a very late stage of the design
figuration of these spatial extended systems, new teletraffic            process. The handling of this issue is usually left to the
models are required to evaluate their spatial performance, cf.         expertise of the network design engineer.
[24]. In particular, the design of mobile networks has to be              The demand-based and integrated approach overcomes this
based on the analysis of the distribution of the expected spatial      disadvantage by regarding the spatial teletraffic demand dis-
teletraffic demand in the complete service area. However,               tribution in the service area as a major input factor among
most of the traffic models applied so far for the demand                the design constraints. The new approach is depicted in
estimation characterize the traffic only in a single cell, e.g.,        Fig. 1. The main cellular design constraints are organized into
[10]. Other traffic models, like the highway Poisson-arrival-           four equally and parallel considered basic modules, cf. [22]:
location model (PALM) proposed by Leung et al. [13], give              radio transmission, mobile subscriber, system architecture, and
deep theoretical insights but they are too complex for practical       resource management. This structured set of input parameters
                                                                       is used by the integrated concept for the synthesis of a cellular
   Manuscript received December 1997; revised February 1998.           configuration. The system configuration is generated by the
                                        u
   The authors are with Lehrstuhl f¨ r Informatik III, University of   automatic network design sequence.
   u                         u
W¨ rzburg, Am Hubland, W¨ rzburg, D-97074 Germany (e-mail: tutschku@
informatik.uni-wuerzburg.de; trangia@informatik.uni-wuerzburg.de).        In contrast to the conventional design method, the new
   Publisher Item Identifier S 0733-8716(98)04108-0.                    approach starts with the analysis of the expected teletraffic
                                                     0733–8716/98$10.00 © 1998 IEEE
TUTSCHKU AND TRAN-GIA: SPATIAL TRAFFIC ESTIMATION AND CHARACTERIZATION                                                          805



                                                                    in order to increase system efficiency. In Section III-B we
                                                                    examine these types of models in greater detail.

                                                                    A. Traffic Source Models
                                                                       Due to their capability to describe user behavior in detail,
                                                                    traffic source models are usually applied for the characteriza-
                                                                    tion of the traffic in an individual cell of a mobile network.
                                                                    Using these models, local performance measures like new call
                                                                    blocking probability or handover blocking probability can be
                                                                    derived from the mobility pattern. Additionally, these models
                                                                    can be used to calculate subjective quality-of-service values
                                                                    for individual users.
                                                                       Overview on Traffic Source Models: A widely used single
                                                                    cell model was first introduced by Hong and Rappaport [10].
                                                                    Their model assumes a uniformly distributed mobile user
                                                                    density and a nondirected uniform velocity distribution of the
Fig. 1. Integrated planning approach.                               mobiles. Under this premise, performance values like the mean
                                                                    channel holding time and the average call origination rate in
                                                                    a cell can be computed.
demand within the considered coverage area, cf. Section V.             A more accurate modeling of the calling behavior of users in
Due to the equal and parallel contribution of all basic mod-        a single cell was proposed by Tran-Gia and Mandjes [19]. The
ules, the new concept is capable of obeying the interactions        model considers a base station with a finite customer popula-
and dependencies between the design objectives. Hence, the          tion and repeated attempts. The appealing characteristic of the
capacity and teletraffic engineering objectives can be addressed     model is the assumption of a small, finite user population. This
early and in an appropriate way. Moreover, the new ap-              is the typical case in real networks, cf. Section IV. However,
proach is able to find tradeoffs between contrary objectives         the model is limited to a single cell and does not consider the
like high user coverage and equipment minimization and              spatial variation of the teletraffic within the service area.
is capable of achieving comprehensive optimized wireless               El-Dolil et al. [4] characterized the mobile phone traffic on
network configurations. Additionally, the approach constitutes       vehicular highways by assuming a one-dimensional mobility
a forward engineering technique and facilitates the application     pattern. They derive the performance values by applying
of automatic network design algorithms.                             a stationary flow model for the vehicular traffic. An ex-
                                                                    tended one-dimensional highway model with a nonuniform
                     III. TRAFFIC ESTIMATION                        density distribution, denoted as the highway PALM model,
                                                                    was investigated by Leung et al. [13]. For traffic character-
   In mobile communication networks, the teletraffic originat-       ization, fluid flow models with time-nonhomogeneous and
ing from the service area of the system can be described by two     time-homogeneous traffic have been used, as well as an
traffic models which differ in their view of the network. 1) The     approximative stochastic traffic model.
traffic source model, which is often referred to as the mobility        A limited directed two-dimensional mobility model was
model, describes the system as seen by the mobile unit. The         investigated by Foschini et al. [5]. The model assumes a
traffic scenario is represented as a population of individual        spatially homogeneous distribution of demand and an isotropic
traffic sources performing a random walk through the service         mobility structure. Chlebus [2] investigated a mobility model
area and randomly generating demand for resources, i.e., the        with a homogeneous demand distribution but which assumes
radio channels. An overview of these models is provided in          a nonuniform velocity distribution. The traffic orientation is
Section III-A. 2) In contrast, the network traffic model of a        nondirected and uniformly distributed.
mobile communication system describes the traffic as observed           The application of these traffic source models in real net-
from the nonmoving network elements, e.g., base stations            work planning cases is strongly limited. Some models, like the
or switches. This model characterizes the spatial and time-         highway PALM model, give a deep insight to the impact of ter-
dependent distribution of the teletraffic. The traffic intensity      minal mobility on cellular system performance; however, they
   is in general measured in call attempts per time unit and
                                                                    are rather complex to be applied in real network design. Other
space unit [calls/(sec km )]. Taking into account the mean
                                                                    models, like the one suggested by Hong and Rappaport [10],
call duration        , the offered traffic is                  (in
                                                                    can only be applied for the determination of the parameters in
[Erlang/km ]). This measure represents the amount of offered
                                                                    an isolated cell, due to their simplification assumptions.
traffic in a defined area.
   Both traffic models are used in mobile communication
system design. The latter model is of principal interest when       B. Traffic Intensity
determining the location of the main facilities in a mobile           Since the cellular network planning process requires a
network, i.e., the base stations and the switching centers. These   comprehensive view of the expected load and since the traffic
components should be located close to the expected traffic           source models only focus on a single cell, a network tele-
806                                                               IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 16, NO. 5, JUNE 1998




                                          (a)                                                                (b)




                                          (c)                                                                (d)
Fig. 2. Demand node concept: (a) geographical and demographical data, (b) traffic matrix, (c) service area tessellation, and (d) demand node distribution.


traffic model has to be specified. Therefore, we define the                                The location        is a coordinate in , and
traffic intensity function            This function describes the               is the size of the unit area element.
number of call requests seen by the fixed network elements                         Using the assumption that every mobile unit has the same
in a unit area element at location          during time interval               call attempt rate      at time the traffic intensity
              The coordinates          of the area element are                 can be readily obtained
integer numbers. Due to the definition given above, the traffic
intensity function is a matrix of traffic values representing the                                                                                       (2)
demand from area elements in the service region, cf. Fig. 2(b).
The traffic intensity           can be derived from the location                   Since in real world planning cases it is almost impossible
probability and the call attempt rate of the mobile units.                     to directly calculate the location probability             from
   Under the premise that this probability            is known,                the mobility model, the traffic intensity has to be derived from
                                                                               indirect statistical measures.
the average number of mobile units                            in a certain
area element at time is
                                                                               C. The Geographic Network Traffic Model
                                                                                  The offered traffic in a region can be estimated by the
                                                                        (1)    geographical and demographical characteristics of the service
                                                                               area. Such a demand model relates factors like land use, pop-
                                                                               ulation density, vehicular traffic, and income per capita with
  Here,           is the probability that, if the system is                    the calling behavior of the mobile units. The model applies
viewed from the outside, there is a mobile unit at location                    statistical assumptions on the relation of traffic and clutter type
TUTSCHKU AND TRAN-GIA: SPATIAL TRAFFIC ESTIMATION AND CHARACTERIZATION                                                                 807



with the estimation of the demand. In the geographic network              usually dropped and the traffic models are reduced to stationary
traffic model, the offered traffic            is the aggregation            models describing the peak traffic. The maximum load is the
of the traffic originating from these various factors                      value of the traffic during the busy hour, cf. [15].
                                                                             There remains a pitfall for the network designer: the busy
                                                                   (3)    hour varies over time within the service area. In downtown
                                                                          areas the highest traffic usually occurs during business hours,
                                                                          whereas in suburban regions the busy hour is expected to be
where                  is the traffic generated by factor in an            in the evening. Therefore, the network engineer has to decide
arbitrary area element of unit size measured in Erlangs per area          how to weight the different traffic factors, i.e., how to obey
unit,     the number of call attempts per time unit and space             the different market shares of various user groups in the traffic
unit initiated by factor           is the mean call duration of           model of the network.
calls of type and               is the assertion operator

                     traffic factor    is not true at location             D. Traffic Discretization and Demand Nodes
                     traffic factor    is true at location                    The core technique of the traffic characterization proposed
                                                             (4)          in this paper is the representation of the spatial distribution
   So far, the planning of public communication systems uses              of the demand for teletraffic by discrete points, denoted as
geographic traffic models which have a large granularity. In               demand nodes. Demand nodes are widely used in economics
these cases a typical unit area size is in the order of square            for solving facility location problems, cf. [7].
kilometers, i.e., in public cellular mobile systems this is the              Definition: A demand node represents the center of an area
size of location areas, cf. [8]. For the determination of the             that contains a quantum of demand from teletraffic viewpoint,
location of transmission facilities a much smaller value is               accounted in a fixed number of call requests per time unit.
required. Their locations have to be determined within a spatial             The notion of demand nodes introduces a discretization
resolution of 100 m. Thus, a unit area element size in the order          of the demand in both space and demand. Consequently the
of 100 m       100 m is indicated here.                                   demand nodes are dense in areas of high traffic intensity and
   Traffic Parameters: The values for , which are the traffic               sparse in areas of low traffic intensity. Together with the
values originating from factor per area element, can be                   time-independent geographic traffic model, the demand node
derived from measurements in an existing mobile network and               concept constitutes, in the context of cellular network design,
by taking advantage of the known causal connection between                a static population model for the description of the subscriber
the traffic and its origin. A first approach is to assume a                 distribution.
highly nonlinear relationship. A general structure to model                  An illustration of the demand node concept, given in
this behavior is to use a parametric exponential function. In             Fig. 2(a), shows publicly available map data with land
our proposed geographic model, the traffic-factor relationship             use information for the area around the city of W¨ rzburg,
                                                                                                                                 u
is defined to be                                                           Germany. The information was extracted from ATKIS, the
                                                                          official topographical cartographical data base of the Bavarian
                                                                   (5)    land survey office [1]. The depicted region has an extension
                                                                          of 15 km        15 km. Fig. 2(b) sketches the traffic intensity
where     is constant and   is the base of the exponential                distribution in this area, characterized by the traffic matrix:
function.                                                                 dark squares represent an expected high demand for mobile
  To reduce the complexity of the parameter determination                 service and bright values correspond to a low teletraffic
we introduce the normalization constraint                                 intensity. Fig. 2(d) depicts a simplified result of the demand
                                                                          discretization. The demand nodes are dense in the city
                                                                   (6)    center and on highways, whereas they are sparse in rural
                                                                          areas.
                                                                             In principle, the two-dimensional teletraffic density ma-
where                is the size of the service area,                is   trix, cf. Fig. 2(b), is sufficient to characterize the teletraffic
the size of a unit area element, and           is the total teletraffic    distribution in the service area. However, the application of
in this region. The value of               can be measured in an          the demand node representation significantly decreases the
operating cellular mobile network.                                        computational requirements in network design. Due to the
   The structure of the geographical traffic model given in (3)            use of discrete point representation in mobile system design,
and (5) appears to be simple. However, due to its structure the           it is no longer necessary to calculate the field strength at
model can be adapted to the proper traffic parameters. This                every point in the area. It is adequate to compute the field
capability enables its application for mobile system planning.            strength values only at the location of the demand nodes, cf.
   Stationary Geographic Traffic Model: The above proposed                 Section V. Moreover, the discrete presentation can be used
model                also includes the temporal variation of the          to characterize the clustering effect of users in the service
traffic intensity in the service area. Since communication                 area. The demand node concept enables the evaluation of the
systems must be configured in such a way that they can                     impact of this user clumping on network performance, cf.
accommodate the highest expected load, the time index is                  [18].
808                                                              IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 16, NO. 5, JUNE 1998




                                    (a)




                                                                              Fig. 4. Algorithm 1: Demand node generation.


                                                                              not usually collected with respect to mobile network planning.
                (b)                                     (c)
                                                                              For example, ATKIS’s main objective is to maintain map
                                                                              information. It uses a vector format for storing its drawing
Fig. 3. Original map information data and data preprocessing: (a) unordered
lines, (b) adjacent open and closed polygons, and (c) closing lines.          objects.
                                                                                 To determine the clutter type of a certain location, one has
                                                                              to identify the land type of the area surrounding this point.
                IV. TRAFFIC CHARACTERIZATION                                  This requires the detection of the closed polygon describing
                                                                              the shape of this area. Since most maps are printed on paper,
A. Traffic Characterization Procedure                                          the order in which the lines of a closed shape are drawn does
   Based on the estimation method introduced in the previous                  not matter [see Fig. 3(a)]. To identify closed polygons, one has
section, the traffic characterization has to compute the spatial               to check if every ending point of a line is the starting point of
traffic intensity and its discrete demand node representation                  another one. If a closed polygon has been detected, the open
from realistic data taken from available data bases. In order to              lines are removed from the original base and replaced by its
handle this type of data, the complete characterization process               closed representation. In addition, due to the map nature of
comprises four sequential steps.                                              the data, two adjacent area objects can be stored by a closed
                                                                              and an open polygon [see Fig. 3(b)]. Some data might also
   Step 1) Traffic model definition: Identification of traffic
                                                                              be missing [see Fig. 3(c)]. In this case line closing algorithms
            factors and determination of the traffic parameters
                                                                              have to applied. After the preprocessing step only closed area
            in the geographical traffic model.
                                                                              objects remain in the data base and the traffic characterization
   Step 2) Data preprocessing: Preprocessing of the informa-
                                                                              can proceed with the demand estimation.
            tion in the geographical and demographical data
                                                                                 Traffic Estimation: Step 3) of the traffic characterization
            base.
                                                                              process uses the geographical traffic model defined in Step
   Step 3) Traffic estimation: Calculation of the spatial traffic
                                                                              1) for the estimation of the teletraffic demand per unit area
            intensity matrix of the service region.
                                                                              element. The computed traffic values are stored in the traffic
   Step 4) Demand node generation: Generation of the dis-
                                                                              matrix. To obtain the traffic value of a certain unit area
            crete demand node distribution by the application
                                                                              element, the procedure first determines the traffic factors which
            of clustering methods.
                                                                              are valid for this element and then computes the matrix entry
   Traffic Model Definition: The definition of geographical                      by applying (3).
traffic model in Step 1) of the characterization procedure
is based on the arguments given in Section III-C. A simple
but accurate spatial geographic traffic model is the basis for                 B. Demand Node Generation
system optimization in the subsequent network design steps.                     The generation of the demand nodes in Step 4) of the
   Data Preprocessing: The data preprocessing in Step 2) is                   characterization process is performed by a clustering method.
required since the data in geographical information systems are               Clustering algorithms are distinguished into two classes, cf.
TUTSCHKU AND TRAN-GIA: SPATIAL TRAFFIC ESTIMATION AND CHARACTERIZATION                                                          809




Fig. 5. ICEPT’s network design sequence.


[11]: 1) the partitional clustering methods, which try to con-     right-angled bisections, the shape of the tessellation pieces is
struct taxonomies between the properties of the data points,       always rectangular. To overcome these drawbacks, we also
and 2) the hierarchical clustering methods which derive the        investigate hierarchical agglomerative clustering algorithms.
cluster centers by the agglomeration of input values.              These methods are able to obtain tessellation pieces of an
   The algorithm proposed for the demand generation is a           arbitrary shape and of a predefined traffic value.
recursive partitional clustering method. It is based on the
idea of dividing the service area until the teletraffic of every           V. AUTOMATIC CELLULAR NETWORK DESIGN
tessellation piece is below a threshold Thus, the algorithm
constructs a sequence of bisections of the service region. The     A. The ICEPT Planning Tool
demand node location is the center of gravity of the traffic
weight of the tessellation pieces.                                    To prove the capability of the demand estimation and to
                                                                   show the feasibility of the integrated and systematic design
   The demand node generation algorithm is shown in Al-
                                                                   concept ICEPT (Integrated Cellular network Planning Tool), a
gorithm 1, found in Fig. 4. The function left area()
                                                                   prototype of a planning tool for cellular mobile networks, was
divides the area into two rectangles with the same teletraf-
                                                                                                          u
                                                                   implemented at the University of W¨ rzburg [23]. The tool’s
fic and returns the left part of the bisection. The function
                                                                   core components are the automatic network design algorithm
right area() returns the right piece. In every recursion
                                                                   SCBPA (set cover base station positioning algorithm) [20],
step the orientation of the partitioning line is turned by 90 .
                                                                   and the traffic characterization procedure as described in
The recursion stops if every rectangle represents a traffic
                                                                   Section IV.
amount less than the minimal quantization value The func-             The SCBPA algorithm is a Greedy heuristic which selects
tion traffic() evaluates the amount of expected teletraffic         the optimal set of base stations that maximizes the proportion
demand in the area.                                                of covered traffic, i.e., the ratio of the demand nodes which
   An example for the bisection sequence of the algorithm          measure a path loss on the forward/reverse link above the
is shown in Fig. 2(c). The numbers next to the partitioning        threshold of the link budget.
lines indicate the recursion depth, and to make the example           In the ICEPT prototype, the design constraints and ob-
more vivid, not every partition line is depicted. The upper left   jectives are implemented as exchangeable modules. These
quadrant of Fig. 2(c) only shows lines until the recursion depth   modules are indicated in Fig. 5 by the names of the design
3, the lower left only the lines until depth 4, the lower right    aspects, e.g., radio wave propagation. ICEPT uses as radio
only the lines until depth 5, and the upper right quadrant lines   wave propagation models the common Hata model [9] and
until depth 6.                                                     the COST231 model [17].
   The partitional clustering algorithm of Algorithm 1 is a fast      Network Design Sequence: The network design sequence
but simple clustering method. However, its accuracy strongly       of ICEPT is depicted in Fig. 5. In contrast to the conven-
depends on the quantization value         which only gives an      tional cellular design method, the demand-based approach of
upper bound for the traffic represented by a single demand          ICEPT starts with the traffic estimation. Therefore, the tool
node. Moreover, since the algorithm constructs a sequence of       first generates the demand node distribution of the planning
810                                                      IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 16, NO. 5, JUNE 1998



                                                                    cation systems. The proposed method considers the teletraffic
                                                                    from the viewpoint of the network. Its traffic estimation
                                                                    is based on the geographic traffic model, which obeys the
                                                                    geographical and demographical factors in the service area for
                                                                    the teletraffic demand estimation. The characterization of the
                                                                    spatial distribution is facilitated by the application of discrete
                                                                    points, denoted as demand nodes. We demonstrated how the
                                                                    demand node pattern can be derived from the information
                                                                    in geographical data bases. In addition, we outlined how
                                                                    the demand node representation enables the application of
                                                                    demand-based automatic mobile network design algorithms.
                                                                       However, since rather insufficient geographical and demo-
                                                                    graphical information often is available in practical cellular
                                                                    system design, it would be of great interest to generate demand
                                                                    node patterns artificially by the application of spatial point pro-
                                                                    cesses, such as those described by Latouche and Ramaswami
                                                                    [12], Stoyan and Stoyan [16], and Cressie [3]. Furthermore,
                                                                    these reference processes could be used to evaluate different
                                                                    planning scenarios.
                                                                       Future applications of the demand node concept are not
                                                                    limited solely to mobile communication systems. The concept
Fig. 6. ICEPT planning result: Base station locations.              can also be used in other network planning scenarios where
                                                                    transmission facilities have to be deployed in a service area,
region. Afterwards it computes the coverage areas for all           such as in wireless access networks, wireline data networks,
potential transmitter locations and configurations. In the next      or ADSL (asymmetric digital subscriber line) systems [14].
step, ICEPT checks whether or not the traffic and hardware           The demand node concept facilitates a revenue-based system
constraints are obeyed at these potential sites. Invalid configu-    design.
rations are removed and are not considered in the optimization         The traffic characterization procedure introduced in
step. After completing the verification, the SCBPA algorithm         Section IV will soon be commercially available as CUTE
generates the cellular configuration by selecting a subset of        (CUstomer Traffic Estimation tool). More information can
transmitter configurations from the potential set that maxi-         be found on the World Wide Web at http://www.infosim-
mizes the coverage. During the selection of the transmitters        usa.com/CUTE/.
the SCBPA algorithm not only considers the demand coverage,
but also minimizes the carrier-to-interference (C/I) ratios, cf.                              ACKNOWLEDGMENT
[21]. Subsequently, the tool computes the carrier separation
constraints and constructs a frequency allocation plan. If the        The authors would like to thank T. Leskien, P. Liebler,
frequency allocation plan is valid, ICEPT again verifies C/I         and M. Heuler for their valuable programming support and
values of the configuration. In the event that the C/I constraints   M. Wolfrath, K. Leibnitz, and O. Rose for fruitful discussions
are not obeyed, the separation constraints have to be increased.    during the course of this work.
If the C/I specifications are met, the network design stops with
the output of the cellular radio network configuration.                                            REFERENCES
                                                                     [1] ATKIS (Amtliches Topographisches Kartographisches Informations Sys-
B. Planning Result                                                       tem) Bavarian Land Survey Office, Munich, Germany, Tech. Rep. 1991.
                                                                     [2] E. Chlebus, “Analytical grade of service evaluation in cellular mobile
   ICEPT was tested on the topography around the city center             systems with respect to subscribers’ velocity distribution,” in Proc. 8th
      u
of W¨ rzburg. The task was to find the optimal locations of               Australian Teletraffic Research Seminar, 1993, pp. 90–101.
nine transmitters in this terrain. The result of the algorithm is    [3] N. Cressie, Statistic for Spatial Data. New York: Wiley, 1991.
                                                                     [4] S. A. El-Dolil, W.-C. Wong, and R. Steele, “Teletraffic performance
depicted in Fig. 6. The base station locations are marked by             of highway microcells with overlay macrocell,” IEEE J. Select. Areas
a diamond-shaped symbol            The lines indicate the convex         Commun., vol. 7, pp. 71–78, Jan. 1989.
hull around the set of demand nodes which are supplied by            [5] G. J. Foschini, B. Gopinath, and Z. Miljanic, “Channel cost of mobility,”
                                                                         IEEE Trans. Veh. Technol., vol. 42, pp. 414–424, Nov. 1993.
the base station. The SCBPA algorithm was able to obtain a           [6] A. Gamst, E.-G. Zinn, R. Beck, and R. Simon, “Cellular radio network
75% coverage of the teletraffic of the investigated area. The             planning,” IEEE Aerosp. Electro. Syst. Mag., vol. 1, pp. 8–11, Feb. 1986.
                                                                     [7] A. Ghosh and S. L. McLafferty, Location Strategies for Retail and
total computing time for the configuration, including the traffic          Service Firms. Lexington, MA: Heath, 1987.
characterization, was about 4 min on a SUN Ultra 1/170.              [8] S. Grasso, F. Mercuri, G. Roso, and D. Tacchino, “DEMON: A
                                                                         forecasting tool for demand evaluation of mobile network resources,”
                                                                         in Proc. Networks ’96, Sydney, Australia, 1996, pp. 145–150.
                         VI. CONCLUSIONS                             [9] M. Hata, “Empirical formula for propagation loss in land mobile radio
                                                                         services,” IEEE Trans. Veh. Technol., vol. 29, pp. 317–325, Aug. 1980.
  This paper presented a new model for the characterization of      [10] D. Hong and S. S. Rappaport, “Traffic model and performance analysis
the expected spatial demand distribution in mobile communi-              for cellular mobile radio telephone systems with prioritized and non-
TUTSCHKU AND TRAN-GIA: SPATIAL TRAFFIC ESTIMATION AND CHARACTERIZATION                                                                                     811



       prioritized handoff procedures,” IEEE Trans. Veh. Technol., vol. VT-35,                                                                    u
                                                                                                          Kurt Tutschku (S’96) was born in W¨ rzburg, Ger-
       pp. 77–92, Aug. 1986.                                                                              many, in 1966. He received the Diploma degree
[11]   A. K. Jain and R. C. Dubes, Algorithms for Clustering Data. Engle-                                 in computer science in 1994 from University of
       wood Cliffs, NJ: Prentice Hall, 1988.                                                                 u
                                                                                                          W¨ rzburg, where he is currently working towards
[12]   G. Latouche and V. Ramaswami, “Spatial point patterns of phase type,”                              the Ph.D. degree.
       in Proc. 15th Int. Teletraffic Congr., Washington, DC, 1997.                                           From 1991 to 1992 he was a Visiting Student
[13]   K. K. Leung, W. A. Massey, and W. Whitt, “Traffic models for wireless                               at the University of Texas at Austin, TX. In 1994
       communication networks,” IEEE J. Select. Areas Commun., vol. 12, pp.                               he was a Visiting Scientist at the Stuttgart Research
       1353–1364, Oct. 1994.                                                                              Lab of the Daimler-Benz Corporation. In November
[14]   K. Maxwell, “Asymmetric digital subscriber line: Interim technology                                1994 he became a Member of the Scientific Staff of
       for the next forty years,” IEEE Commun. Mag., vol. 34, no. 10, pp.                                 the Department of Distributed Systems at University
       100–106, 1996.                                                                 u
                                                                                 of W¨ rzburg. His main fields of research are planning methods for mobile
[15]   M. Mouly and M.-B. Pautet, The GSM System for Mobile Communica-           communication systems and high-speed networks and the application of neural
       tion. Palaiseu, France: self-published, 1992.                             networks for control of nonlinear dynamic processes.
[16]   D. Stoyan and H. Stoyan, Fraktale—Formen—Punktfelder: Methoden
       der Geometrie-Statistik. Berlin, Germany: Akademie-Verlag, 1992.
[17]   G. L. St¨ ber, Principles of Mobile Communication. Boston: Kluwer
                 u
       Academic, 1996.
[18]   P. Tran-Gia and N. Gerlich, “Impact of customer clustering on mobile
       network performance.” Inst. Comput. Sci., Univ. W¨ rzburg, Germany,
                                                             u                                           Phuoc Tran-Gia (M’80) received the Dipl.-Ing.
       Res. Rep. 143, 1996.                                                                              degree from Stuttgart University and the Dr.-Ing.
[19]   P. Tran-Gia and M. Mandjes, “Modeling of customer retrial phenomenon                              degree from University of Siegen, both in electrical
       in cellular mobile networks,” IEEE J. Select. Areas Commun., vol. 15,                             engineering.
       pp. 1406–1414, Oct. 1997.                                                                            In 1977 he joined Standard Elektrik Lorenz (now
[20]   K. Tutschku, “Demand-based radio network planning of cellular com-                                Alcatel), Stuttgart, where he worked on the software
       munication systems,” in Proc. IEEE Infocom ’98, San Francisco, 1998.                              development of digital switching systems. From
[21]           , “Interference minimization using automatic design of cellular                           1979 to 1982 he worked in the Department of
       communication networks,” in Proc. IEEE/VTS 48th Veh. Technol. Conf.,                              Communications, Siegen, and from 1983 to 1986 he
       Ottawa, Canada, 1998.                                                                             was with the Institute of Communications Switch-
[22]   K. Tutschku, N. Gerlich, and P. Tran-Gia, “An integrated approach to                              ing and Data Techniques, Stuttgart University. In
       cellular network planning,” in Proc. 7th Int. Network Planning Symp.      1986 he joined IBM Zurich Research Laboratory where he worked on
       (Networks ’96), 1996.                                                     the architecture and performance evaluation of computer communication
[23]   K. Tutschku, K. Leibnitz, and P. Tran-Gia, “ICEPT—An integrated           systems. Since 1988 he has been a Professor in the Faculty of Mathematics
       cellular network planning tool,” in Proc. IEEE/VTS 47th Veh. Technol.                                                  u
                                                                                 and Computer Sciences, University of W¨ rzburg, Germany. His current
       Conf., Phoenix, AZ, 1997.                                                 research areas include performance modeling of communication systems,
[24]   P. E. Wirth, “The role of teletraffic modeling in the new communication    planning and optimization of mobile communication networks, and analysis
       paradigms,” IEEE Commun. Mag., vol. 35, no. 8, pp. 86–93, 1997.           of manufacturing systems.

								
To top