Radio Network Planning Simulator Virtej

Radio network-planning simulator for telecommunication systems Iuliana Virtej Control Engineering Laboratory Helsinki University of Technology [email protected]

1 Introduction Telecommunication– central to our daily lives has changed dramatically. These changes are the result of the technological advances, the broadcast of the broadband services, the popularity of Internet and wireless communications. In such dynamic technological and economical environment, competition is increasing among the service providers and among the equipment manufacturers. Optimization of the planning network is becoming essential. Many fundamental problems are still open. The new Wideband Code Division Multiple Access (WCDMA) system is intended to be the third generation of mobile communication. WCDMA is a complementary system compared with the actual system. The main feature of the next generation mobile communication system is that data services are considered. Based on this feature, mobiles will support all Internet and multimedia services. This brings a lot of new services, but also the problems to be solved are more complex. The performance of the network should be analyzed from two points one is ‘hardware’ part and the other one is from ‘software’ part. First the reliability issues are addressed in this paper. This part will cover the ‘hardware’ performance of the network. The second part covers the availability issue. The performance of the network when the radio resources are allocated to the users is discussed. Radio network simulator what is presented in the paper will measure the performance of the implemented algorithms for power control based on the number of failures and also the quality of the transmission. The quality of the transmission is very important. Bad quality of the transmission means that the subscribers don’t have satisfactory services, so as a consequence the reduction of the subscribers to an operator is noticed. The paper tries to emphasize the importance of good services offered to the subscribers in a communication network. Also the paper is discussing the performance measurement from hardware point of view. The practical aspects are pointed out. The practical constraints are considered as well for the discussed algorithms.

2 Reliability issue in telecommunication network The design and analysis of the reliability and availability issues are considered for interconnected networks such as communication and telephony networks. In telephony networks each subscriber is considered that is able to place a call to any other subscriber in the network. Generally, the manner in which communication links are provided depends on a number of factors. Reliability is primarily concerned with ensuring that some routes in the network connecting intended source and destination are operationally. Availability is concerned primarily with ensuring that some routes are not only operational but also unblocked or idle (avoiding congestion in the network or applied to have good radio resources allocation).

2.1 Telecommunication network Given a set of geographically distributed sites, information about the traffic between them and information about the means by which they can be connected (links) is given. Network design problem is to select some candidate connections. The primer goal is to keep the cost within a specified budget and to provide sufficient connections to support the traffic offered at a specified capacity, speed or throughput. How can be improved the performance of the network? The network design has to improve the performance for example maximizing the fraction of the messages delivered within a time limit specified by the designer or client. The network design is a challenging problem. The network planner must therefore address reliability issues: How does one anticipate and deal with faulty behavior of the network? The same question can be considered for the availability issue as well. Satisfactory techniques to anticipate and to accommodate component failures are in many ways few well understood than the design and performance in a failure free system. One of the hardest tasks is to anticipate the failures in the system and how much one can reduce the performance of the network. Reliability together with availability becomes the central problem for network design. In this case the performability term is defined as a measure of performance and reliability requirements.

2.2 Reliability and performability Network reliability concerns the capability of the network to provide connections to support required network functionality. Connectivity plays an important role. A dual effort is done to extend reliability measures to treat performance measure more complex than connectivity, and to adapt performance analysis techniques to handle failures in the system. Basic knowledge of network planning and design are introduced. For example the binary situation where each link is either operating as intended, or failed completely. The probability for the connection is known. A network design is a selection of E links connecting V nodes.

A state of the network is a subset of S E containing precisely the links that are operational. A network is considered functional if there is a set of terminal nodes in the network that must all be reachable from a specified source node. The reliability can be measured in this case as a probability of connectedeness : s,Tconnectedeness is the probability that all the terminal nodes in T are reachable form a source node s in the network state , when each link operates independently with known probability. For a network with reliable nodes V and unreliable two links E , each state S E has an associate probability Pr[S] of arising and a value (S) of some performance measure . Then the expected performance is Perf(V,E)= Pr[S] (S); This performance is perhaps the most commonly studied class of performability measures. Few classes of algorithms are used to measure the performability. These methods are exact algorithms, Monte-Carlo methods, efficiently computable bounds, most probable states, most relevant states and hybrid methods.

2.3 Exact algorithms This type of algorithm is applied to analyze the coherent performability of measures. Coherence states that for any two states S and S’ if S S’ then (S) (S’). When the statistical independence holds then the performance measure is coherent. The performance, in this case, is an indicator factor for the network. The inconvenience of the algorithm is that as the number of links is getting larger the complexity is becoming higher.

2.4 Monte Carlo methods One can argue that the performance should not be exactly measured. Estimation of the value can be enough. Some assumptions about performance and/or the link failures are made. The estimation possibility is the weakness point and the strongest one of this method. The strength consists of typically one sampling plan serves for a wide variety of performability measures .The weakness part comes because that is not possible to incorporate some features about a specific network. A trade off between time and accuracy is done using these methods. The more time is used the better results of the estimation are obtained.

2.5 Efficiently computable bounds If the time to get the estimation is limited then another technique is used to get the estimation of the performance of the network. One can compute upper and lower bounds of performability efficiently. If the performability measure of the interest is connectivity-based, efficiently computable bounds methods are sufficient accurate. In a contrary situation, much faster methods can be used such as Monte Carlo methods. The bounds that are the most accurate for reliability measures appear to be those that are most dependent on the measure being connectivity-based.

2.6 Most probable states and most relevant state These methods are used in case when the performability is measured based on the most probable state or most relevant state, which can influence the performance of the network. The accuracy of the most probable state technique is difficult to compare, as one delivers absolute bounds while the other yields point estimates. Most probable state method makes some restriction on the performability measure examined. The most relevant state method is used in such cases when some states with lower probability could influence seriously the performance of the network. Specifically Pr[S] might be relatively small, but (S) can be very large. All the presented methods are used to estimate the performance of the network from the hard point of you. Application to the multi-interconnecting networks is hot research area. Flexibility and reliability of the network are main points for designing and planing network behavior. As a conclusion of this part, which can be pointed out, is that in practical situation the failures of the hardware in the interconnected networks are very difficult to be estimated. The performance of the network can be predicted using different methods. The main goal of the operator is to design a network where the number of failures in the system (hardware failures) is low. The flexibility of the system assures the reconfigured possibility of network and has also economical benefits. Another analysis of the network from the availability issues can be discussed. The addressed issue represents to avoid the congestion in the network. If an analysis of the network from radio resources allocation point of view and admission control is done then the performance of the network is considered the capacity of the system and the quality of the service. Next paragraph presents the performance of network in case that the considered radio resource used for allocation is power, which is one of the critical resources in the third generation mobile communication. First the radio network simulator is briefly described and after that power control algorithms are presented. Simulation results and performance of the network are presented. The practical constraints are taken into account. Conclusions and future research area are considered in the final part of the paper.

3 Radio Network Simulator "NetSim" 3.1 Introduction NetSim is a simulator developed to study Cellular Radio Network Control Algorithms and planning methods by Telecommunication Laboratory at Helsinki University of Technology [ 2,3, 4,5] . It is able to provide detailed information about system capacity, coverage and network control algorithms. NetSim output files consist of information about call dropping, blocking time and spatial references for these events, statistics can be easily collected and translated into visual form with help of MATLAB or other tools. NetSim is written in C language and can be updated for different radio interfaces (GSM, WCDMA), various statistical and deterministic channel models and different types of Radio Network Control Algorithms (handover, power control, admission control). NetSim is a time driven simulator, the purpose of which is to fill the gap between simulators developed to study link level signal processing algorithms like COSSAP and higher-level network simulators like OPNET and BONES. Considering C language, NetSim can be easily extended to include link level simulation or fixed network simulation and can be combined with COSSAP or OPNET tools. NetSim can simulate users behavior, various types of teletraffic, interference, power and admission control algorithms, adaptive antennas beamforming, soft and hard handover. Current version uses a deterministic propagation model for microcellular urban environment based on well-known ray tracing methods. Deterministic model gives opportunity to study different radio network planning methodologies. Existing propagation model provides information about spatial properties of radio channel for simulations radio networks with adaptive antennas at BS. The main blocks of NetSim are shown in the Figure 1.1. NetSim blocks can be logically divided in to four main models related to reference scenarios, radio propagation, and signal processing algorithms and radio network control issues. These are further discussed.

3.2 Reference scenario module The output data of the reference scenario module are MS location, velocity and activated service type data. It consists of users and teletraffic models [ 2,3, 4,5] .

3.3 Users model Users model generate information about users’location and velocities. Users in the current version are assumed to be pedestrians randomly distributed along predefined path. Current version can also model spatially non-uniform user distributions. Model for domestic users and car passengers using mobile phones is under development.

3.4Teletraffic model

Teletraffic model is for voice service simulation control and traffic channels behavior taking in to account voice activity detection. NetSim has also a model of data transmission relating to Internet services. Models for multi-rate data services are included in the current plans.

3.5 Propagation module The propagation module calculates the received signal power for each MS location. It uses propagation data of the ray tracing program. Because the distance between two calculated impulse responses is a quarter of wavelength, the cubic convolution interpolation method is used to predict the received signal strength.

3.6 Signal Processing Module Signal processing module simulates antenna and receiver algorithms. The output information of this module is post -processing SIR for each link.

3.7 Antenna model Antenna model can simulate different types of antennas at BS such as : single omnidirectional and directional antenna, distributed antennas, switched beam system, adaptive antennas beamforming based on maximum received signal power or CIR. In the current NetSim version antenna array is modeled as a single element antenna with reconfigurable antenna diagram. Antenna can change its pattern during call initialization. (for example: omnidirectional diagram
directional beams).

3.8 RAKE receiver and interference model The RAKE receiver and interference model calculates the SIR after despreading and combining all radio links. The total SIR is obtained by summing the signal-tointerference ratios of different paths corresponding to maximal ratio combining. Another technique that can also be used is the selection combining. In this technique the receiver selects the path with the best SIR. The interference consists of the thermal noise of the receiver, self-interference, and interference from the users in the same radio channel. Self-interference means here interference that is caused by the multipath propagation. All interference is modeled as additive Gaussian white noise. Combination of Adaptive Antennas technology with RAKE receiver - so called 2DRAKE receiver can be an interesting extension of this model.

3.9 Radio Network Control Module Radio Network Control Module simulates power control, admission and handover control algorithms. One of the most important functions of radio network control algorithms is radio resource management. Transmitted power is the only resource of proposed single carrier DS-CDMA systems. It can be controlled at different systems

levels and also jointly in combination with other network control methods like BS assignment and beamforming. Network Control model is also responsible for constant monitoring of SIR of all active radio links. The output data of Network Control Module consist of number of call droppings and blocking events and other system control failures with corresponding spatial and temporal references.

3.10 Admission Control Model Admission Control Model in the current NetSim version uses information about BS total received power, signal quality measurements and its variation in time for making decision about admission. More complex admission control algorithms can be further studied.

3.11 Power Control Model Power Control Model for up-link and downlink use SIR based distributed power control algorithm. Program includes return channel error and loop delay models. Open loop power control is used during call initialization period. Multiservice version of NetSim includes SIR target to service adjustment and adaptively adjusted power control step size. With this model different types of power control algorithms during soft handover can be studied.

3.12 Handover Model Handover Model includes soft and hard handover algorithms.

3.13 System requirements Operation system in NetSim is UNIX or MS-DOS (Windows version not available yet). There are no special memory requirements in running NetSim program. Practically requirements for memory are only restricted by size of files containing propagation data. UNIX or Pentium PC (for DOS and Windows version) are the current platforms.

Main Blocks and models of the “NetSim”

Propagation data

Reference Scenario

Current version

Users model: location, velocities Teletraffic model / Services Offered traffic /service spatial distribution

Propagation data processing

Radio Network Control/Resource

Signal Processing allocation methods Algorithms

Impulse responses for MS locations

COSSAP

Antenna Model RAKE receiver/Interference Model

Pedestrians on predefined route

Next versions Models for domestic users and car passengers

Control and Traffic Channel Voice service with VAD

Multiservices model, packet and circuit switched

Deterministic raytracing model

Statistical propagation model, model based on measurements

Single element antenna model wit h reconfigurable diagram

Multiple elements ant. model, 2D-RAKE

SIR

Power Control Model Admission Control Model Handover

System performance data

OPNET, BONES BS assignment

SIR based distributed Joint optimum BS assignment power control Soft /hard MAHO handover, optimum combining at the uplink

power control and beamforming. Multicarrier, multisystem network simulation

Figure 1.1. Main blocks and models of the NetSim and directions of further developments. The main blocks in Figure 1.1 [ 2,3, 4,5] shows the links between NetSim and OPNET or COSSAP. It also shows the flow diagram inside the simulator. The simulator was used to simulate and planning the telecommunication network at the system level, where all the resources are allocated for the user. The performance of the network was analyzed based on the number of quality failures. The quality of the algorithm is compared based on the quality of the transmission measure. If different power control algorithms are considered the performance of the network in terms of capacity is analyzed.

4 System model First some notations are introduced. The radio cellular system considered here has a finite cell size. The number of active users is denoted by N ms , total number of calls N calls , total number of failures N fails , number of failures due to bad quality N q _ fails , number of synchronous failures N no _ ch . The transmitted power for ith mobile is denoted by Pt i , power received from each BS (base station) through each path is Pr jL where j represents the number of base stations, and L the number of paths, mean

[

]

received power Pr mean = Prmean1 ,..., PrmeanN _ MS _ MAX , SIR (signal-to-interference ratio), SIR target (signal-to-interference ratio target). SIR is calculated by maximal ratio combining processing gain of every MS, SIR ij represents SIR calculated between the transmitter MS i and BS j , SIR at the previous time instant SIR pre and SIR pre

ij

represents SIR pre calculated between the MS i and BS j , signal quality counter quality i , where i=1,..,N_MS_MAX, j=1,..,N_MAX_BS. The number of base stations is one in this study limited to one, which implies SIR ij = SIR i and SIR pre ij = SIR pre

i

.

The signal-to-interference ratio in the kth path of the radio link between the ith mobile and jth base station, SIRijk , is expressed in [ 5] and [ 7] as:

SIRijk =

W P R i , j ,k N0 + α

L l =1,l ≠ im

Pi , j ,l +

N _ MS _ MAX

L

m =1 m≠ l

l =1

, Pm, j ,l

where W is the chip rate, R is the data bit rate, N 0 is the thermal noise power, L is the number of paths, N_MS_MAX is the maximum number of mobiles, N_MAX_BS is the maximum number of base stations, α is the orthogonality factor, P i , j , k is the received power at the jth base station from ith mobile through the kth path. The autocorrelation and cross-correlation properties of the direct sequence are assumed similar. The maximum ratio combining technique is used for simulation. The receiver combines the signals propagated through different paths by weighting the strengths of each path and chooses the strongest path of the radio link. Therefore the total SIR in N _ MS _ MAX

the link between ith and jth base station is given by SIR ij =

l =1

SIRijk . The fading

in the channel is generated using the ray tracing method. The uplink transmission is considered.

5 Power control method The uplink closed loop power control adjusts the mobile station power control in order to keep the received uplink SIR at a given SIR target. The transmitted power control command is codified on 1/2/3/4 bits. In [1] and [ 5] the idea of implementing power control for 1 bit is presented. The number of power control levels will be dependent of the number of bits. Power control steps coded on 2 bits, 3bits, 4 bits will determine 4, 8 and 16 transmitted power control commands respectively. The minimum power step is 0.25 dB and maximum is 1.5 dB. The decision is taken based on the variation of the difference (SIR pre i − SIR target ). The difference (SIR pre i − SIR target ) is the error in the closed loop power control and is denoted by ∆ SIR i for ith mobile. The value of the difference is compared with a fixed value, which represents part of SIR target . The power control is decreased or increased with

maximum step when the variation of ∆ SIR i is large compared with the fixed value. The minimum step power control is applied in case that the variation is small. The algorithm is distributed. The power control is calculated for each mobile every simulation time. The power control signal is realistic calculated based on power control error. 5.1 Power control of 2 bits For power control of 2 bits four levels of power are defined, equally distributed as number of steps for increasing and decreasing. In the range of 0.25 - 1.5 dB discrete steps of power Pc step1 and Pc step2 for each mobile are considered. Pc step1 represents the minimum step, 0.25 dB, and Pc step2 is maximum step, 1.5 dB, used for power control command. Each mobile has two power control steps defined and this gives the possibility of increasing or decreasing the power for each one with one value out of two defined. The error probability of the return channel is defined for the ith mobile as Pc error _ pr i . The proposed algorithm is divided into phases. In the first phase the sign of the power control correction is decided. In the second part the new transmitted power is calculated. The sign of the power control is derived based on the following algorithm at the k th moment of simulation: If ( ∆ SIR i < 0) then C = 1 and the power of MS (mobile) is increased with a step chosen in the algorithm. Else C = -1 and the power of MS is decreased with a step chosen in the algorithm, for i = 1 ,…, N_MS_MAX. The power control error is given in the same form as the sign for increasing and decreasing the power value at the kth moment of simulation as: If (q is less than Pc error _ pr i ) then E = -1 and the power control step is chosen incorrectly Else

E = 1 and the power control step is correct ,where q represents a variable for power control error. Now with the information that is known about increasing or decreasing the power and the information that the chosen power is either well defined or not, the power control command can be formulated. The algorithm updates the transmitted power based on the decided power control level and the Pc error _ pr i . The updating function is nonlinear. The constraints taken into account are the total transmitted power and maximum power control level which is between 0.25 dB and 1.5 dB. These are practical constraints specified in UTRA [1]. The non-linear function is calculated based on the transmitted power Pt multiplied by the decided power control level to the power E and C. The power is decreased or increased with maximum step when ∆ SIR i is large compared with fixed value and with minimum step when the difference ∆ SIR i is not large. The increasing or decreasing command is given by the variable C which is calculated for every k iteration and for each mobile. The return channel error is calculated in case that the correct power control step is not available.

5.2 Power control algorithms of 3 and 4 bits The philosophy of power control algorithms of 3 and 4 bits is similar as above. The power control commands now are codified on 3, respectively 4 bits. The number of power control levels is 8 in case of power control of 3 bits, and 16 levels in the case of 4 bits.

6 Simulation results The results of the simulations are obtained using NetSim. NetSim is a simulator developed to study Cellular Radio Network Control Algorithms and Planning methods, by Telecommunication Laboratory at Helsinki University of Technology [ 2], [ 3], [ 4], [5] . It provides a simulation environment to study system capacity, coverage and network control algorithms of WCDMA systems. NetSim output files consist information about call dropping, blocking time and spatial references for these events. Statistics can easily be collected and translated into visual form with the help of MATLAB or other tools. The following assumptions have been made in all simulations: the number of base stations is one, number of transmission paths L = 5, the maximum number of mobiles N_MS_MAX was 50. Initial value for Pc step1 i is 1.0593, and 1.4125 for Pc step2 i for each mobile, where i varies between 1 and N_MS_MAX. Value 1.0593 corresponds to 0.25 dB and 1.4125 to 1.5 dB. The decision in power control algorithm is taken based on the variation of ∆ SIR i . The return channel probability is considered 0.01. This section contains simulation results when different decisions of power control commands, of 2 bits, 3 bits and 4 bits, were used. These results were for one base station obtained using NetSim, in uplink transmission (mobiles (MS) to base station (BS)). The algorithms are compared based on the number of the total fails. One power control algorithm is considered better than other if the total number of fails at the system level is lower. The number of fails has to be as low as possible. In the ideal case the number of bits for coding the power control is one and the number of fails is 1. The results are shown in Table 0. First power control of 2 bits is simulated. The results of the simulation are presented in Table I: Total number of mobiles 50 Total number of calls 64 Total number of fails 1 Total number of quality fails 1 Tabel 0. The total number of mobiles, initiated calls, fails,and quality fails for power control on one bit over the simulation time Total number of mobiles Total number of calls Total number of fails Total number of quality fails

40 46 3 3

Tabel I. The total number of mobiles, initiated calls, fails,and quality fails for power control on two bits over the simulation time Figure 1 shows the evolution of the total number of initiated calls and the total number of fails when power control algorithm of 2 bits was used.

50 n u m b e r o f m o b ile s 45

to tal no o f m o b ile s

n u m b e r o f c a lls n u m b e r o f f a ils

40

35

to tal no o f c a lls

30

25

20

15

10

to tal no o f fa ils

5

0 10

15

20

25

30

35

40

A d m is s io n t im e

Figure 1 Simulation of power control of 2 bits. For power control of 3 and 4 bits the algorithms have the same structure as for 2 bits. The number of power levels is now 4 and 8 for increasing and the same number for decreasing. For power control of 3 bits, 4 power levels for the range 0.25- 1.5 dB are defined, and 4 for decreasing. The power control command is defined as a nonlinear function based on power control step and the calculated error. The number of bits is different because more accurate power control steps are given for each mobile. The simulation results are presented in the next table: Total number of mobiles Total number of calls Total number of fails Total number of quality fails

49 59 6 6

Tabel II. The total number of mobiles, initiated calls, fails,and quality fails for power control on 3 bits over the simulation time In Figure 2 the evolution of the total number of initiated calls and the total number of fails is presented when the power control algorithm of 3 bits is used. Comparing the number of fails with the previous algorithm, an increase of them is observed, but this is not significant. 60 number of mobiles number of calls number of fails

50

total no of mobiles

total no of calls

40

30

20

10 total no of fails

0 10

15

20 Admission time

25

30

35

40

Figure 2. Simulation power control of 3 bits A situation of more power steps for decreasing the power than for increasing the power is considered next. Five steps of power levels for decreasing the power and 3 for increasing are chosen. The results of the simulation given in Table III are better in this case when the total simulation time is 50 s. Total number of mobiles Total number of calls Total number of fails Total number of quality fails

40 48 0 0

Tabel III. The total number of mobiles, initiated calls, fails,and quality fails for power control on 3 bits over the simulation time The same algorithm can be defined for 4 bits power control. The number of the levels of power control steps is 16. First 8 steps for increasing the power and 8 steps for decreasing, are considered. The results of the simulation are similar to those in Tabel III. They are summarized in Tabel IV.

Total number of mobiles Total number of calls Total number of fails Total number of quality fails

50 45 3 3

Tabel IV. The total number of mobiles, initiated calls, fails,and quality fails for power control on 4 bits over the simulation time The evolution of the total number of initiated calls and the total number of fails is presented in Figure 3 when power control algorithm for 4 bits is used. c 60 n u m b e r o f m o b ile s n u m b e r o f c a lls n u m b e r o f f a ils

50

to t a l n o o f m o b ile s

40

30

to tal no o f c a lls 20

10

to tal no o f fa ils 0 10

15

20

25

30

35

40

A d m is s io n t im e

Figure 3. Simulation power control of 4 bits

7 Conclusions The paper is composed of two major parts. One part is describing the reliability issue in case of multi interconnecting network (communication and telecommunication networks). The other part is analysis the performance of the network in terms of capacity. The reliability issue is a design problem. Some algorithms to estimate the performance are presented here. The second part presents the algorithms for power control command in case that the number of bits is varying from 1 to 4. Total number of bits for coding the power control command is considered as a practical constraint. The new approach considered is that power control command can be transmitted on 2/3/4 bits and nonlinear heuristic power control scheme is applied based on this assumption. This coding for power control command gives a possibility to have different power levels. The power control levels are for increasing or for decreasing the power, the dynamic range of power in uplink is 80 dB. The power control algorithms for a command on 2/3/4 bits were tested in the simulator. The uplink transmission is considered here. The behavior of the algorithms is good in this case. The criteria for comparing the algorithms are the total number of failures, which should be low at the system level. The practical constrains which were taken into account are according with the proposed standard ETSI UMTS, 1998.

8 References [1] ETSI UMTS Terrestrial Radio Access (UTRA) ITU-R RTT Candidate Submission, 1998 [ 2] Mauri Honkanen, Technical Report 31.12.1995. Technology Development Center Finland - Project ,”Simulation and Signal Processing in Radio Systems”, Subproject 2.3. “Development of a DS-CDMA Radio Network Simulator” The Ray Tracing Program. [ 3] Petri Bergholm, Mauri Honkanen, Sven-Gustav Häggman, “Simulation of a Microcellular DS-CDMA Radio Network”, IEEE VTC 1995, pp. 838 -842. [ 4] Petri Bergholm, Technical Report 31.12.1995. Technology Development Centre Finland - Project, ”Simulation and Signal Processing in Radio Systems”, Subproject 2.3. “Development of a DS-CDMA Radio Network Simulator” [ 5] Antti Pietilä, “Development of the Netsim program”, Proceedings of the IRC Workshop’97, pp. 90 -91. [ 6] A. J. Viterbi, “Principles of Spread Spectrum Communication”, Addison-Wesley, Reading, Mass., USA, 1995. [ 7 ] P.Soriano, B.Sanso`, “Telecommunication Network Planning”, Kluwer, USA, 1999.