15
FOCUS A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis M. Theologou Management of the power control operation in HIPERLAN/2 networks Published online: jjj Ó Springer-Verlag 2003 Abstract BRAN/WLAN systems, e.g., HIPERLAN/2, IEEE 802.11a, etc., are seen as a promising solution for covering residential, business, transport, tourism, etc., environments, and generally areas of high demand, characterized as ‘‘hot spots’’. This paper presents man- agement functionality for augmenting the potential exploitation of one of these systems, HIPERLAN/2, by making feasible their (loose or tight) integration in a composite radio (CR) context. The approach will be the following. The first point will be to revisit the main features of a HIPERLAN/2 system. Next, the architec- ture of a general Service and Network Management System (SNMS), which has been developed for assisting wireless systems in their operation in a CR context, will be briefly presented. The next main point will be the presentation of the functionality of the SNMS compo- nent that is tailored to the managed HIPERLAN/2 technology and specifically to the configuration of the Power Control (PC) functionality. An algorithm for configuring the PC operation, based on a greedy algo- rithm and a neural network, will be presented. A rele- vant resource management problem, which should be efficiently solved for exploiting HIPERLAN/2 networks, will be addressed. Numerical results will be presented. Keywords HIPERLAN/2 IEEE 802.11a IEEE 802.11h Power control 1 Introduction Present day wireless telecommunications networks, which are primarily narrowband, are mostly used for circuit-switched voice services. The migration of second generation (2G) towards the 2.5G era [1] and the development of third generation (3G) mobile wireless systems aim to enable networks to provide users with instantaneous bit rates of up to 2 Mbit/s, significantly improving packet-data transmission and mobile multi- media applications. This will be materialized through the gradual introduction of the Universal Mobile Tele- communications System (UMTS) [2, 3, 4]. In addition, even higher data rates can be obtained for local area networks using novel short-range wire- less technologies. Bandwidth demanding, real-time and interactive multimedia services, such as high-quality vi- deo distribution, client-server multimedia applications, and data-bank access, are typical applications for this technology. Therefore, new wireless networks with broadband capabilities are being sought to provide high- speed integrated services (data, voice, and video) with cost-effective support for Quality of Service (QoS). This leads to the introduction of systems collectively called Broadband Radio Access Networks (BRANs) and/or Wireless Local Area Networks (WLANs). A particular class of such systems, operating at the 5 GHz band, promises to offer high data rates at ade- quate capacity volumes, for short-range communica- tions with limited mobility. Therefore, they are seen as a promising solution for covering residential (home), corporate (business, office, etc.), transport (e.g., airport, train, etc.), and other environments, which are often characterized as hot-spot areas. The class includes the IEEE 802.11a and 802.11h systems, the High Perfor- mance Radio LAN Type 2 (HIPERLAN/2) [5–11], specified by the European Telecommunications Stan- dards Institute (ETSI), and Japan’s High Speed Wireless Access Network (HiSWAN). The spectrum allocation in the three systems is presented in Fig. 1 Moreover, a recent trend, often called ‘‘wireless be- yond 3G’’, assumes that cellular, BRAN/WLAN and DVB (Digital Video Broadcast) systems can be co- operating systems in a composite radio (CR) infra- structure [12–14]. According to the CR concept, a Soft Computing j (2004) 1 – 15 DOI 10.1007/s00500-003-0354-3 A. Oikonomou(&) National Technical University of Athens, Electrical and Computer Engineering Department, Telecommunications Laboratory, 9 Heroon Polytechneiou Street, Zographou 15773, Athens, Greece E-mail: [email protected] 5 0 0 0 0 3 5 4 Journal number Manuscript number B Dispatch: 12.1.2004 Journal : Soft Computing No. of pages: 15 Author’s disk received 4 Used 4 Corrupted Mismatch Keyed

A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

FOCUS

A. Oikonomou � P. Demestichas � K. TsagkarisG. Koundourakis � M. Theologou

Management of the power control operation in HIPERLAN/2 networks

Published online: jjj� Springer-Verlag 2003

Abstract BRAN/WLAN systems, e.g., HIPERLAN/2,IEEE 802.11a, etc., are seen as a promising solution forcovering residential, business, transport, tourism, etc.,environments, and generally areas of high demand,characterized as ‘‘hot spots’’. This paper presents man-agement functionality for augmenting the potentialexploitation of one of these systems, HIPERLAN/2, bymaking feasible their (loose or tight) integration in acomposite radio (CR) context. The approach will be thefollowing. The first point will be to revisit the mainfeatures of a HIPERLAN/2 system. Next, the architec-ture of a general Service and Network ManagementSystem (SNMS), which has been developed for assistingwireless systems in their operation in a CR context, willbe briefly presented. The next main point will be thepresentation of the functionality of the SNMS compo-nent that is tailored to the managed HIPERLAN/2technology and specifically to the configuration of thePower Control (PC) functionality. An algorithm forconfiguring the PC operation, based on a greedy algo-rithm and a neural network, will be presented. A rele-vant resource management problem, which should beefficiently solved for exploiting HIPERLAN/2 networks,will be addressed. Numerical results will be presented.

Keywords HIPERLAN/2 � IEEE 802.11a �IEEE 802.11h � Power control

1 Introduction

Present day wireless telecommunications networks,which are primarily narrowband, are mostly used for

circuit-switched voice services. The migration of secondgeneration (2G) towards the 2.5G era [1] and thedevelopment of third generation (3G) mobile wirelesssystems aim to enable networks to provide users withinstantaneous bit rates of up to 2 Mbit/s, significantlyimproving packet-data transmission and mobile multi-media applications. This will be materialized throughthe gradual introduction of the Universal Mobile Tele-communications System (UMTS) [2, 3, 4].In addition, even higher data rates can be obtained

for local area networks using novel short-range wire-less technologies. Bandwidth demanding, real-time andinteractive multimedia services, such as high-quality vi-deo distribution, client-server multimedia applications,and data-bank access, are typical applications for thistechnology. Therefore, new wireless networks withbroadband capabilities are being sought to provide high-speed integrated services (data, voice, and video) withcost-effective support for Quality of Service (QoS). Thisleads to the introduction of systems collectively calledBroadband Radio Access Networks (BRANs) and/orWireless Local Area Networks (WLANs).A particular class of such systems, operating at the

5 GHz band, promises to offer high data rates at ade-quate capacity volumes, for short-range communica-tions with limited mobility. Therefore, they are seen as apromising solution for covering residential (home),corporate (business, office, etc.), transport (e.g., airport,train, etc.), and other environments, which are oftencharacterized as hot-spot areas. The class includes theIEEE 802.11a and 802.11h systems, the High Perfor-mance Radio LAN Type 2 (HIPERLAN/2) [5–11],specified by the European Telecommunications Stan-dards Institute (ETSI), and Japan’s High Speed WirelessAccess Network (HiSWAN). The spectrum allocation inthe three systems is presented in Fig. 1Moreover, a recent trend, often called ‘‘wireless be-

yond 3G’’, assumes that cellular, BRAN/WLAN andDVB (Digital Video Broadcast) systems can be co-operating systems in a composite radio (CR) infra-structure [12–14]. According to the CR concept, a

Soft Computing j (2004) 1 – 15DOI 10.1007/s00500-003-0354-3

A. Oikonomou(&)National Technical University of Athens,Electrical and Computer Engineering Department,Telecommunications Laboratory,9 Heroon Polytechneiou Street,Zographou 15773, Athens, GreeceE-mail: [email protected]

5 0 0 0 0 3 5 4Journal number Manuscript number B Dispatch: 12.1.2004 Journal : Soft Computing No. of pages: 15

Author’s disk received 4 Used 4 Corrupted Mismatch Keyed

Page 2: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

network provider (NP) can rely on diverse radio tech-nologies for efficiently covering service area regions.This may mean that the NP either possesses licenses fordeploying and operating diverse radio systems (tightintegration between the radio technologies), or cooper-ates with other NPs that own alternate radio networks(loose integration between the radio technologies). Effi-cient coverage means offering as high as possible (re-quired) Quality of Service (QoS) levels, at adequatecapacity volumes, in a cost-effective manner.A typical CR scenario will include 2.5G/3G mobile

networks, BRAN/WLAN and DVB systems. In thiscontext, BRAN/WLAN systems should properly man-age their resources, so as to have capacity available,which can be used for cooperating with other networksof the infrastructure. The cooperation is materializedthrough the agreement of absorbing traffic from othernetworks of the CR infrastructure, in order to assistthem to the handling of new service area conditions (e.g.,hot-spot situations, traffic demand alterations, etc.), orservice management requests. Achieving this operation,however, requires upgraded service and network man-agement systems (SNMSs). This paper will present(essential parts of ) such an SNMS. It will be assumedthat a BRAN/WLAN network, operated by an arbitraryNP, is covering a given area. The proposed managementfunctionality extends the exploitation possibilities (andtherefore, the chances of success) of the BRAN/WLANnetwork, by enabling its (loose) integration in an overallCR infrastructure, which comprises also other NPs thatoperate various types of networks.It should be noted that BRAN/WLAN systems pos-

sess a ‘‘self-sufficient’’ mode of operating, in the sense,that they are adequately dynamic (autonomous) foradapting to the environment conditions. This is mainlymotivated by the fact that these systems will operate in alicense-exempt spectrum band. Therefore, under theseconditions, the introduction of management function-ality can be essential. The reason is the provision ofstatistical guarantees regarding performance and QoS,towards the CR infrastructure. The CR infrastructure

will pose various requirements. The presented manage-ment functionality aims at the fast adaptation of thesystem to the new requirements.The control actions in a HIPERLAN/2 network are

mainly Link Adaptation, Dynamic Frequency Selection(DFS), and Power Control (PC). The quality of theradio link, which is dependent on the radio environment,changes over time and in accordance with traffic insurrounding radio cells. To cope with variations, a LinkAdaptation scheme is applied. In addition, the DFSoperation allows several operators to share the availablespectrum and avoids the use of interfered frequencies.Frequency selection is based on interference measure-ments performed by the access point and associatedmobile terminals. The functionality of this paper is in thedirection of managing the PC operation, in order toensure that the traffic handled by the network is servedin a most efficient way, reducing, nevertheless, the gen-erated interference as much as possible.The reference system for this paper will be

HIPERLAN/2, however, our practices are applicable tothe IEEE 802.11a, IEEE 802.11h and HiSWAN systemsthat have similar specifications. Our approach in thispaper is the following. Section 2 revisits the basic fea-tures of the HIPERLAN/2 system. Section 3 brieflypresents a management system that enables HIPER-LAN/2 networks to act as parts of a CR environment.Section 4 presents the management functionality thatwill configure the PC operation [15, 16]. It will be basedon a greedy algorithm (Sect. 4.4) and a neural network(Sect. 4.5). Sections 5 and 6 include sample numericalresults and concluding remarks.

2 HIPERLAN/2 Overview

This section revisits the basic features of the HIPER-LAN/2 system [5–11].Figure 2a depicts the reference architecture ofHIPERLAN/2 networks. Each Access Point (AP) con-trols a cell. It offers wireless connectivity to the mobile

Fig. 1 Current spectrumallocation at the 5 GHzband

2

Page 3: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

terminals (MTs) of the cell, and is, therefore, the inter-face between the radio and the fixed network. Typically,a HIPERLAN/2 system will comprise a fixed networksegment that enables the interworking with core net-works, as well as the communication between APs.

Figure 2b presents, in a high level manner, the func-tionality (protocol stacks) of an AP. The convergencelayer realizes the mapping between the core networkprotocols and the lower layers of the HIPERLAN/2system. The physical layer uses Orthogonal FrequencyDivision Multiplexing (OFDM). The carrier spacing is20 MHz, which means that in Europe there can be 19carriers available (most likely, 12 will be for outdoor/indoor use, and 7 only for indoor use). Each carrier is

Fig. 2 a Reference architecture of HIPERLAN/2 networks. bFunctionality (protocol stacks) of an AP. c Structure of the mediumaccess control (MAC) frame

3

Page 4: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

split into 52 sub-carriers (48 are used for data and 4 arepilots). Table 1 lists the seven physical layer modesprovided regarding the sub-carrier modulation andcode-rate schemes, and the resulting bit-rates. Themodulation schemes supported are binary phase shiftkeying (BPSK), quaternary PSK (QPSK), 16-quadratureamplitude modulation (16-QAM), and 64-QAM. Thecode-rates are 1/2, 3/4, and 9/16.The medium access control (MAC) is based on a time

division duplex (TDD) and time division multiple access(TDMA) scheme, controlled by the AP. Figure 2(c)depicts the structure of the MAC frame. It has a fixedduration of 2 ms and consists of several phases. Thebroadcast phase contains control information. Theframe channel phase describes the (allocation of re-sources in the) current MAC frame. The access feedbackchannel phase contains information on previous randomaccess attempts. The uplink and downlink phases con-tain data from/to MTs. The direct link phase containsinformation exchanged between MTs, without the directinvolvement of the AP, according to the ad-hoc networkparadigm. The random access phase is used by MTs forestablishing associations with APs, requesting resources,once associated with an AP, and finally, conductinghandovers.The data link control (DLC) layer is divided into a

control and a user plane part. The control part supportsa number of procedures: (i) association control, i.e.,association establishment, authentication, encryptionkey exchange, disassociation, etc.; (ii) DLC control, e.g.,connection set-up/release/modify, multicast join/leave,etc.; (iii) radio resource control, i.e., link adaptation,power control, dynamic frequency selection (DFS) andhandover.The DFS operation, which enables an AP to auto-

matically select its frequency, based on interferencemeasurements, will not be further analyzed in this paper.The link adaptation operation enables the AP to selectthe highest possible physical layer mode (Table 1), forboth the uplink and downlink, based on radio linkquality measurements, and mainly from the Carrier toInterference Ratio (CIR), conducted by both the MTand the AP. The power control operation decreases theinterference caused in other cells, or other systems, in thesame band.The CIR is one of the important resources of the

system. The higher the CIR in a cell, the higher the

physical layer mode that can be selected, and conse-quently, the throughput that can be achieved. Theappropriate physical layer mode is selected through thelink adaptation operation. The transmitter power con-trol operation is a means for improving the CIR levels.Section 4 presents management functionality for man-aging (properly configuring) the power control proce-dure, and therefore improving the CIR.

3 Service and network management system

This section briefly presents the SNMS that enables aHIPERLAN/2 network to be loosely integrated in a CRinfrastructure (Fig. 3). The detailed description of theplatform can be found in [14].The system is composed by three entities: (i) Moni-

toring, Service management interworking and ResourceBrokerage (MSRB); (ii) Resource Management Strate-gies (RMS); Network and Environment Simulator(NES).Firstly, the MSRB entity [17], identifies new service

area conditions (e.g., new traffic demand patterns), andaccepts and responds to service management requests.Through these capabilities the SNMS is capable of act-ing in a reactive (to new service area conditions) andproactive (accepting and satisfying service managementrequests) mode. The resource broker capability is im-posed by the CR concept. It enables the generation, andnegotiation on, a set of offers. References [18–22] aresome samples of background material, on which thiscomponent can be based.Whenever one of the triggers above emerges (new

service area condition, service management request, orresource brokerage request) the MSRB initiates theRMS operation. The RMS component finds cost-effec-tive reconfigurations of the managed network thatachieve the required capacity figures. Using the inputfrom the MSRB component certain management actionsare taken, to ensure that the status of the network re-mains adequate. These actions, as already stated, can bethe proper configuration of the operations in the control

Table 1 Physical layer modes for HIPERLAN/2

Physical layermode

Modulation Code rate Physical layerbit rate (Mbps)

1 BPSK 1/2 62 BPSK 3/4 93 QPSK 1/2 124 QPSK 3/4 185 16 QAM 9/16 276 16 QAM 3/4 367 64 QAM 3/4 54 Fig. 3 Deployment of the management system for the HIPERLAN/2

network

4

Page 5: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

domain. The current paper investigates the configura-tion of the power control operation.The Network Environment Simulator (NES) enables

validation of some management decisions prior to theapplication in the network, off line testing and demon-stration. It is not further analyzed in this paper.

4 Management of the power control operation

This section presents the ETSI standard requirementsfor the PC procedure, and our approach to the man-agement of the PC operation.

4.1 Constraints and requirements for the powercontrol procedure

The HIPERLAN/2 standard ([9]) defines the acceptablelevels of transmission, between which the APs and MTsare allowed to operate. For both, the transmissionpower must be greater than )15 dBm and lower than30 dBm (or even lower in accordance to regulatoryrequirements).The AP transmission power (AP Tx Level) ranges

from )15 dBm to the maximum, with an increase step of3 dB. However, the standard rather than defining atransmission power for the MTs, defines a desired re-ceived power level at the AP, from the MT. This powerlevel (AP Rx UL Level), ranges from )71 to )43 dBm.The MT transmission power must be therefore such thatthe received power at the AP reaches its desired value,without however exceeding the transmission power of itsassociated AP nor of course its maximum transmissioncapabilities. To accomplish that, the MT makes mea-surements on the link to estimate the current pathloss,and defines its power at such a level so that the receivedpower at the AP is the desired one.Therefore, the power control procedure defines two

values for each AP, the transmission power and the re-quired reception power (namely AP Tx Level andAP Rx UL Level). In the following, the algorithm forconfiguring these two values will be presented.

4.2 Basic assumptions

The proper configuration of the target, uplink anddownlink, transmission powers allowed, constitutes themanagement of the power control operation. The man-agement functionality should ensure fast adaptations tosevere traffic alterations, involving large segments of thenetwork. The new traffic assigned can be the outcome ofthe handling of new service area conditions by themanagement infrastructure.The throughput (capacity) of HIPERLAN/2 cells

is influenced by their CIR, which depends on the allo-cation of carriers to APs, and on the (uplink anddownlink) transmission powers used. Therefore, the

component managing the power control procedure canassist to the achievement of the target CIR levels byproperly configuring the power control related opera-tions and parameters.The algorithm for configuring these parameters is

presented below. The algorithm evolves in two phases.The first phase intends to find a solution fulfilling aminimum CIR requirement for all cells. This is accom-plished using a neural network approach. The secondphase of the algorithm performs computations, whichresult in the increase of certain cells’ requirements. Thefirst phase algorithm is then invoked again to generatenew results.

4.3 General problem formulation

The input for the algorithm is the set of cells (V ) andtheir requirements in terms of total mean bandwidth forall services required for each cell (bwUL vð Þ,bwDL vð Þ8v 2 V ) expressed for the uplink and downlink.Another required piece of information is the carrierallocation to cells, as well as information on the servicearea layout and cell coverage, the propagation condi-tions and the equipment capabilities.It should be noted at this point, that the exact posi-

tion of the users requiring the services is not taken underconsideration both because the users are not consideredto be static and because we intend to minimize thenecessary input data for the algorithm. Therefore, weconsider the cell to be divided in a certain number ofareas, and users to be uniformly distributed in each areaaccording to the size of the area. If we assume thedivision of each cell in n areas, each area will be definedbetween the homocentric circles with radius m � ðR=nÞ,and ðmþ 1Þ � ðR=nÞ, with 0 � m � n� 1. If we mark thetotal demand in each cell as bw vð Þ, the demand in eacharea is then bw vð Þ pR2 mþ1ð Þ2�pR2m2

pR2n2 ¼ bw vð Þ mþ1ð Þ2�m2

n2 . Thus,each cell is divided in areas, marked from 0 to n� 1.The total number of available frequencies is F , and

the exact allocation of carriers to cells is provided by avector ACAP. The service area layout is described throughthe positions of the APs and the radius of the cells.These are given as posAP vð Þ, belonging to cell v (v 2 V )and rAP vð Þ correspondingly. From these given informa-tion, if we mark P as the total area covered by the net-work, for each pixel p (p 2 P ) belonging to the area twofunctions can be identified. (i) A function returning theset of pixels, pAP vð Þ, belonging to cell v (v 2 V ); (ii) afunction, cAP pð Þ, returning the cell to which pixel p be-longs.The propagation conditions are described through an

attenuation model typical for HIPERLAN/2 systems([23]), resulting in a function set of pixel-level attenua-tion values, where each value aP p1; p2; kð Þ p1; p2ð Þ 2 P 2

�,

0 � k � Fj jÞ provides the attenuation of a transmissionthat originates from pixel p1 and terminates at pixel p2,when the distance of the carriers used in cAP p1ð Þ andcAP p2ð Þ is k. The equipment capabilities specify the

5

Page 6: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

maximum transmission powers of MTs and APs, �pMT

and �pAP, respectively.

4.4 First phase of the algorithm

After gathering the above-mentioned information, thefirst phase of the algorithm intends to find a solutionfulfilling the minimum requirements for all cells in thecoverage area. Using results in bibliography ([24]) it canbe found that 5 dB is the minimum CIR for the per-mission of services in a HIPERLAN/2 system, therefore,the minimum requirement for each cell is reaching atleast a 5 dB Carrier to Interference (CIR) at the edges ofthe cell. The first target for the power control procedureis ensuring that the n� 1 area of each cell v (which is themost difficult area to ensure the desired CIR), reachesthe cirDL

n�1 vð Þ ¼ cirULn�1 vð Þ ¼ 5 dB 8v 2 V , where cirDL

n�1ðvÞis the CIR in the n� 1 area of v cell in the downlink (andcirUL

n�1ðvÞ in the uplink). Obviously the other areas of thecells will receive a higher CIR.The solution of the problem provides the allocations

of transmission power to APs and MTs, APAP ¼tpAP vð Þf j8v 2 V g and APMT ¼ tpMT vð Þf j8v 2 V g, respec-tively. The notation tpMT vð Þ and tpAP vð Þ corresponds tothe maximum, uplink and downlink, transmissionpowers, which should be allowed, in cell v, by the powercontrol operation. The objective function is targeted tothe minimization of the aggregate transmission powers.The assigned powers should maintain the required CIRlevels and be compatible with the equipment capabilities.Moreover, the uplink power should not exceed thedownlink power in each cell.Fig. 4a and b describes the formulations of two sub-

problems that compute the APAP and APMT allocations.

The IPMT vð Þ notation represents the aggregate interfer-ence sensed by the reference transmission (MT) of cell v.Likewise, IPAP vð Þ is the aggregate interference sensed bythe AP of cell v. It should be noted here that in bothformulations the MT is considered to be in the n� 1area of the cell v (worst case scenario). The solutionalgorithm is influenced by the scheme analyzed in [16,25, 26]. It employs a greedy algorithm proceeding in aniterative manner. In each round there are computationsof the aggregate interference in each cell. Moreover,there are assessments on the compliance with the con-straints and on the convergence of the algorithm. Thesolution algorithm is shown in Fig. 5.

4.5 Soft computing enhancement for the firstphase algorithm

Even though the greedy algorithm provides a solution, itstill needs some time to achieve it. In real time condi-tions this period could be crucial for the stability of thenetwork. A solution would be to implement a neuralnetwork, trained off-line under the supervision of thegreedy algorithm.Trying to select the most appropriate input vector for

the neural network, a mandatory observation is thatonly the cells that share the same frequency and thuscontribute to the co-channel interference have to be ta-ken into account. Thus, the input sequence consists ofthe CIR targets of the cells that interfere to the examinedcell plus the CIR target for the examined cell itself. Theoutput of the neural network will be the transmissionpower of the AP that controls the cell.The output power values are quantized into the pre-

defined, discrete power levels [9]. Therefore, the

Fig. 4 Power control config-uration. a Downlink. bUplink

6

Page 7: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

proposed neural network shall be used for classificationand more specifically will perform multiple-class identi-fication.Several types of neural networks could be used for

that purpose with the feed-forward backpropagationnetwork and the probabilistic network being the bestcandidates [27]. The latter one is selected and applied tothe solution of the downlink problem, as presented inthe previous section.

4.5.1 Probabilistic neural network

In general, Probabilistic Neural Networks (PNNs) are aclass of radial basis function networks, which combinesome of the best attributes of statistical pattern recog-nition and feed-forward neural networks [28].The PNN architecture is shown in Fig. 6. The neu-

rons in the input layer distribute the input to the pattern

units. It further consists of neurons allocated in threelayers after the input:

– pattern layer: there is one pattern node for eachtraining example. Each pattern neuron forms a dotproduct of the weight vector and the input pattern

Fig. 5 Solution algorithmfor the first phase

Fig. 6 PNN architecture

7

Page 8: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

vector. After that, the product is passed through aselected activation function. Each pattern neuroncomputes a distance measure between the input andthe training case represented by that neuron.

– summation layer: the summation layer has one neuronfor each class (possible –discrete- output). Eachsummation neuron, associated with a single class,sums the pattern layer neurons corresponding tonumbers of that summation neuron’s class. It pro-duces at its net output a vector of probabilities.

– output layer: The output neuron is a threshold dis-criminator that picks the maximum of these proba-bilities and produces one for that class and zero forthe other classes.

Working as a classifier, the main problem is to determinethe class membership of a k-sized multivariate randomvector X ¼ x1; . . . ; xk½ �, into one of N ¼ ðn1; n2 . . . nmÞpossible groups. If the probability density functions(PDF), pi Xð Þ, are known for all populations, thenaccording to the Bayes optimal decision rule [29], each Xis classified into population i if

hicipi Xð Þ > hjcjpj Xð Þ 8j 6¼ i

where hi: the prior probability of a sample being drawnfrom population i. ci: the cost from misclassifying asample from population i.The above decision rule can be also applied even if

the PDFs are not known, provided that Parzen’s tech-nique is used on the training sample to find the pi Xð Þestimates of the density function of each population i[30]. The Parzen’s PDF estimator uses a weight functionthat maximizes its values for small distances between thetraining and unknown points and decreases toward zerofor high distances, respectively (WF x�xi

r

� �). Actually, it is

an average of that weight function across the trainingset. The common density estimator, using the ‘‘Gauss-ian’’ weighting function, is given by

p xð Þ ¼ 1

2pð Þr2rrn

Xn�1t¼0e� x�xij j2=2r2

where r stands for the dimensionality of the input pat-terns, n denotes the number of training patterns and r(sigma) is the scaling parameter that controls the widthof the area of influence. The larger sample size is used,the smaller r values to be set.In the sequence, a PNN is applied to the problem of

downlink allocation of transmitted powers (seeSect. 4.4). Numerical results are also presented.

4.5.2 Simulation and results

We will assume the network topology depicted inFig. 11. The number in parenthesis next to the APnumber, corresponds to the frequency used by the Ac-cess Point. The total number of frequencies is 7. Thenetwork will be described in more detail in a later sec-tion.

The simulation below concentrates on a specific cell,cell=1, and aims at finding the transmitted power of thecorresponding AP. As stated before, the input consistsof the CIR targets of the interfering cells (i.e. cells 10, 18,20 and 28) plus the CIR target for cell 1. Therefore,the training data is a sequence that contains 3000 vec-tors of type: ink ¼ cirk1 cirk10 cirk18 cirk20 cirk28½ �,k ¼ 1,…,3000. It is important to state that, since thenumber of cells sharing the same frequency is not pri-marily known, the size of vector ink is not fixed andtherefore multiple PNNs, deferring in this size, have tobe trained.The output target i.e. the power transmitted by

AP ¼ 1, is defined by the greedy algorithm and plays therole of the network supervisor. The only factor thatneeds to be appropriately configured is the factor r,which is optimized by trying many values and select thebest one in terms of increase in network performance.The trials showed that too small values for r lead up topoor generalization capabilities, while too large r valuesconceal details.Figure 7 shows that the network performs perfectly

when dealing with a sample within the training sequence.Nevertheless, it is unfair to judge a classifier based on itsperformance in classifying its training set. Consequently,the next step is to pick up a sample of 100 unknownvectors of type ink and observe the neural network’sperformance in the general population. This procedureis followed for two different sequences of such vectors.Figure 8 and 9 show the response of the selected PNN,which results in 89% success in the first case and 83%success in the second case.

4.6 Second phase of the algorithm

After the first phase of the algorithm reaches conver-gence, the second phase of the algorithm is triggered.The results are stored and the cell noted as vMAC, isfound. Cell vMAC is selected from all cells as the cell withthe maximum usage in time of its MAC frame. Thiscalculation is feasible, as both the required bit-rate isprovided and the Physical Rate is calculated through theachieved CIR levels. Using link level simulations ([23]) afunction phyrate(cir) connecting the CIR and thePhysical Rate is derived (This function can be definedthrough the Physical Rate versus CIR graph). The cal-culation of vMAC is the following.

vMAC ¼ v 2 V : max

Xn�1m¼0

bwUL vð Þ mþ1ð Þ2�m2

n2

phyrateðcirULm ðvÞÞ

þbwDL vð Þ mþ1ð Þ2�m2

n2

phyrateðcirDLm ðvÞÞ

!!:

The requirement cirDLn�1 vMACð Þ ¼ cirUL

n�1 vMACð Þ (in Fig. 4aand b) for the cell vMAC is then raised by a reasonablestep (i.e. 1 dB) and the first phase of the algorithm is

8

Page 9: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

invoked again. A greater raise step will lead to fasterhowever more approximate results. Each time conver-gence is achieved, the new vMAC with the maximum usageof its MAC frame, is calculated and the algorithm of thefirst phase is rerun, while the last found working solutionis stored. This procedure continues until either the firstphase algorithm cannot provide a solution, or the MACusage for all cells v 2 V , is lower than a minimumthreshold, ensuring that even under unpredicted raises inthe demand, the MAC frame will not be fully coveredresulting in blockage of users demanding services. In

either case, the last stored solution is the outcome of thealgorithm. For the uplink, the transmission power ofthe MTs is converted, through the pathloss formula, tothe desired level of reception at the AP. The secondphase is presented in Fig. 10

4.7 Discussion on the management algorithm

As it is seen, the behavior of the algorithm guarantees,not only that the mobile terminals which are near the

Fig. 7 Training set classifi-cation – 100% success

Fig. 8 General set classifica-tion 1 – 89% success

9

Page 10: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

edge of the cell will receive the required CIR to operatenormally, but also that the cells with the heaviest trafficwill lower their usage of the MAC window, allowingmore mobile terminals to be served at a certain area. Thealgorithm also provides the functionality to lower asmuch as possible (without exceeding the AP and MTcapabilities) the usage of MAC frame if serious altera-tions in traffic are anticipated, or to keep the meanpower at a low level (satisfactory, nevertheless) if noserious variations are anticipated and therefore theusage of the MAC frame can be kept at a high level.

5 Results

This section includes indicative results, mainly, on thebehavior and efficiency of the HIPERLAN/2 configu-ration component, and the Power Control itself. Anindicative test case will be realized. It is assumed there isan initial condition that corresponds to a certain load,performance and configuration for the network. At anext phase there is a new condition, caused by theadditional traffic that should be absorbed, possibly inorder to assist another network of the CR infrastructure.Most likely a 2.5G or 3G mobile network will face suchproblems, and therefore, will need the assistance of theHIPERLAN/2 network. The HIPERLAN/2 configura-tion component, using the PC scheme, is applied toadapt the network to the new condition.Figure 11 depicts the service area and the network

used in our test case. There are 36 cells, organized in a6 · 6 structure. The radius of each cell is assumed100 m. The overall service area covered by the networkis 1 km2. Our study does not depend on the exact struc-

ture of the network. Other networks of the similar sizeand connectivity degree could have been used instead.Only co-channel interference is taken into account in thetest case. Our focus is restricted to co-channel interfer-ence for facilitating the presentation of our methods (re-sults or whatever). Our work can readily be expanded toinclude more general interference conditions, e.g., basedon [31]. In accordance with the standardization, themaximum transmission power of APs is assumed 30 dBmand the maximum allowed uplink interference is set to)43 dBm. It will be assumed that the APs of the net-work can access only 7 frequencies, due to the back-ground interference induced to the other carriers byother systems, using the 5 GHz band in the area. Theassignment of frequencies to the cells is depicted also inFig. 11 (the number in the parenthesis next to the APnumber).At the initial condition of the network it is assumed

that there are (approximately) 600 users (subscribers) ineach cell. The users access two services through theHIPERLAN/2 network. Half of the users access services1 and the other half access service s2 (300 users perservice). Service s1 is analogous to conversational video.It requires 64 Kbps on both directions (uplink anddownlink). Service s2 is analogous to a streaming servicerequiring an average of 128 Kbps, on the downlink, and8 Kbps for the uplink. Each user generates 0.02 Erlangs(20 mErlangs) for the service he is subscribed to. Be-cause of the uniformity of the demand in traffic as wellas the low level of the common demand, the PowerControl procedure would produce a common powerlevel for transmission.Figure 12 is focused on our test case. It presents the

distribution of the additional traffic load that should

Fig. 9 General set classifica-tion 2 – 83% success

10

Page 11: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

be accommodated, and the outcome of the handling ofthe HIPERLAN/2 configuration component. Fig-ure 12a shows that the service area can be split inthree sets. In the heavily shaded area there are 600more users for each service, therefore the demand isthree times more than the initial case. In the lessshaded area the additional users compared to the ini-tial condition are 300 more for each service. Finally inall the other cells, the demand remains as it was in theinitial condition.Figure 12b shows the outcome of the configuration

of the power control operation and parameters on thedownlink. Specifically, it depicts the maximum trans-mission power per AP that can be imposed by thepower control operation, in order to preserve the tar-get CIR levels. The values range from 6 dBm, inmoderately loaded cells in the periphery of the net-work, to 12 dBm, in more heavily loaded cells. Thesevalues are significantly lower than the maximum pos-sible values, which have been used as a reference in theconfiguration of the frequency selection operation.Figure. 12c shows the outcome of the configuration ofthe power control operation and parameters on theuplink. As it was expected, higher power values areallocated to the cells where the demand is increased.What is really interesting is to investigate the results

of the proposed power allocation when it is applied tothe HIPERLAN/2 network. In order to achieve this,two simulations are run (in the NES component of themanagement system). In both simulations, the config-uration for the network is the one corresponding tothe number of users, user activity and type of servicesmentioned above. However, in the first simulation,fixed power levels for the Power Control procedure areused (and the PC scheme proposed is not applied). Inthe second, the results of the PC scheme proposed areapplied. During the simulations, statistics on the per-formance of the network are kept, and at the end ofthe simulation the mean measured values are calcu-lated. The mean rate of transmission for the PhysicalLayer for both downlink and uplink is calculated(including mistakes and retransmissions), and the timeused out of the 2 ms of the MAC frame for the DLand UL are produced, for every AP.In Fig. 13, the mean Physical Layer Rate is de-

picted both without and with the PC for all APs.Next, in Fig. 14, the time used in the MAC frame isprovided for both cases. Conclusions on the perfor-mance of the PC procedure can be extracted. First ofall, the Physical Layer transmission rate is increasedfor the cells with higher demand, while it is slightlyreduced in some of the cells with lower requirements.This was not only anticipated, but is also one of theFig. 11 Network and service area considered in our test cases

Fig. 12 The test case. a Distribution of additional traffic load in theservice area. b Outcome of the configuration of the power controloperation on the downlink. c Outcome of the configuration of thepower control operation on the uplink

c

Fig. 10 General presentation of the second phase algorithm

11

Page 12: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

12

Page 13: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

major benefits of the PC scheme proposed. Cellsheavily loaded, achieve better link quality than what

they would get, and therefore the demand is betterserved where needed. Very important is also the factthat the utilization of the MAC frame is lowered inthe cells in which the demand is higher, and slightlyincreased in cells with lower demand. That leads to a

Fig. 13 a Mean physical bit-rate for the downlink. b Mean physicalbit-rate for the uplink

13

Page 14: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

more uniform usage of the MAC frame in all cells,and creates the possibility to accommodate more

users, if needed, even at cells which are more loaded,and in case the PC management procedure was notused, would have to block users trying to establishconnection. Last but not least, using the neuralnetwork approach, the Power Control procedure

Fig. 14 a Time consumed in each MAC frame in the downlink. bTime consumed in each MAC frame in the uplink

14

Page 15: A. Oikonomou P. Demestichas K. Tsagkaris G. Koundourakis …s3.amazonaws.com/publicationslist.org/data/kostas.tsagkaris/ref-7/J5_SC3_Management of...FOCUS A. Oikonomou P. Demestichas

accomplishes these in a fast and self-operating way,using power levels significantly lower than the maxi-mum allowed values.

6 Conclusions

This paper presented management functionality foraugmenting the potential exploitation of BRAN/WLANsystems through the use of a Network ManagementSystem using a Power Control scheme. The approachused was the following. The first point was to revisit themain features of a HIPERLAN/2 system. Next, thearchitecture of a general SNMS, which has been devel-oped for assisting wireless systems in their operation,was presented. The last main point was the functionalityof the SNMS component that is tailored to the managedHIPERLAN/2 technology and specifically the PowerControl functionality integrated to the ManagementSystem. The algorithm for configuring the PC operation,based on a greedy algorithm and a neural network, waspresented. A resource management problem, which wasefficiently solved with the use of PC for exploiting HI-PERLAN/2 networks, was addressed. Numerical resultswere presented.Issues for further study are the following. First, the

development of a combinative scheme, which willimplement both DFS and PC to lower (if possible) thepower used. The second issue is the report of furtherexperience that will be obtained from the experimenta-tion with the SNMS platform and the PC algorithm.

Acknowledgements. This work was partially funded by the Com-mission of the European Communities, under the Fifth FrameworkProgram, within the IST project MONASIDRE (IST-2000-26144:Management of Networks and Services in Diversified RadioEnvironment).

References

1. The evolution of TDMA to 3G. Special issue in IEEE PersonalCommunications 6(3): June 1999

2. Holma H, Toscala A (2000) W-CDMA for UMTS. J. Wiley &Sons

3. Ojanpera T, Prasad R (2001) Wideband CDMA for thirdgeneration mobile communications. Artech House

4. Laiho J, Wacker A, Novosad T (eds) (2002) Radio networkplanning and optimisation for UMTS. J. Wiley & Sons

5. Khun-Jush J, Schramm P, Malmgren G, Torsner J (2002)HIPERLAN/2: Broadband wireless communications at 5 GHz.IEEE Commun Mag 40(6):June

6. Khun-Jush J, Schramm P, Malmgren G, Torsner J (2000)HiperLAN type 2 for broadband wireless communication,Ericsson Review, No. 2

7. European Telecommunications Standards Institute (ETSI);Broadband Radio Access Networks (BRAN); HIPERLANType 2; Physical Layer (TS 101 475)

8. European Telecommunications Standards Institute (ETSI);Broadband Radio Access Networks (BRAN); HIPERLAN

Type 2; Data link control layer; Part 1: Basic data transportfunctions (TS 101 761-1)

9. European Telecommunications Standards Institute (ETSI);Broadband Radio Access Networks (BRAN); HIPERLANType 2; Data link control layer; Part 2: Radio link control (TS101 761-2)

10. European Telecommunications Standards Institute (ETSI);Broadband Radio Access Networks (BRAN); HIPERLANType 2; Data link control layer; Part 4: Extension for homeenvironment (TS 101 761-4)

11. European Telecommunications Standards Institute (ETSI);Broadband Radio Access Networks (BRAN); HIPERLANType 2; Network management (TS 101 762)

12. IST project MONASIDRE (Management of Networksand Services in a Diversified Radio Environment) Web sitehttp://www.monasidre.com

13. IST project CREDO (Composite Radio for EnhancedService Delivery During the Olympics) Web sitehttp://www.ist-credo.org

14. Demestichas P, Papadopoulou L, Stavroulaki V, Theologou M,Vivier G, Martinez G, Galliano F (2002) Wireless beyond 3G:managing services and network resources. IEEE Computer

15. Bambos N (1998) Toward power-sensitive network architec-tures in wireless communications: Concepts, issues, and designaspects. IEEE Personal Commun 5(3): June

16. Demestichas P, Kotsakis G, Tzifa E, Demesticha V, Anag-nostou M, Theologou M (2002) Power allocation in the contextof dimensioning the air-interface of third-generationW-CDMA-based cellular systems. Inter J Commun Sys 15:375–400

17. Demestichas P, Koutsouris N, Koundourakis G, Papadopou-lou L, Stavroulaki V, Theologou M (2002) Brokerage of wire-less systems’ resources in a composite radio context. Challengesand achievements in e-business and e-work. IOS Press

18. Dutta P (1999) Strategies and games: theory and practice. MITPress, Cambridge, Massachussets

19. Lewicki R, Saunders D, Minton J (1999) Negotiation: readingsexercises and cases. McGraw-Hill, Boston

20. Shell G (1999) Bargaining for advantage: negotiation strategiesfor reasonable people. Viking, Penguin Books, New York

21. Ghosh S (1998) Making business sense of the Internet. HarvardBusiness Review 76(2):

22. Kalakota R, Whinston A (1997) Electronic commerce: a man-ager’s guide. Addison-Wesley Publishing Company

23. Lin Z, Malgren G, Torsner J (2000) System performanceanalysis of link adaptation in HiperLAN type 2. VTC

24. Doufexi A, Armour S, Butler M, Nix A, Bull D (2001) A studyof the performance of HIPERLAN/2 and IEEE 802.11a phys-ical layers. VTC

25. Demestichas P, Igoumenidis A, Kotsakis G, Demesticha V,Tzifa E, Anagnostou M, Theologou M (2000) Transmissionpower management and control in third-generation W-CDMA-based cellular systems. In: Proc 10th IEEE MediterraneanElectrotechnical Conference (Melecon’ 2000), Cyprus

26. Kotsakis G, Papavassiliou S, Demestichas P (2000) Decentra-lised power control algorithms and their convergence for multi-service CDMA-based cellular systems. In: Proc 5th IEEESymposium on Computer and Communications (ISCC’2000).Antibes, France

27. Haykin S (1999) Neural networks. A comprehensive founda-tion. 2nd edn, Upper Saddle River, Prentice Hall, NJ

28. Specht D (1990) Probabilistic neural networks. Neural Net-works 3:109–118

29. Masters T (1993) Practical neural network recipes in C++.Academic Press, New York

30. Cacoulos T (1966) Estimation of a multivariate Density. Annalsof the Institute of Statistical Mathematics, Tokyo 18(2):179–189

31. Demestichas P, Tzifa E, Theologou M, Anagnostou M Inter-ference oriented carrier assignment in wireless communications.IEEE Commun Lett J

15