NEXT GENERATION WIRELESS LAN SYSTEM DESIGN 1. Chutima Prommak, Joseph Kabara, David Tipper, Chalermpol Charnsripinyo

Size: px

Start display at page:

Download "NEXT GENERATION WIRELESS LAN SYSTEM DESIGN 1. Chutima Prommak, Joseph Kabara, David Tipper, Chalermpol Charnsripinyo"

Gregory Roberts
6 years ago
Views:

1 NEXT GENERATION WIRELESS LAN SYSTEM DESIGN Chutima Prommak, Joseph Kabara, David Tipper, Chalermpol Charsripiyo Departmet of Iformatio Sciece & Telecommuicatios 35 N. Bellefield ave., Uiversity of Pittsburgh, PA 5260 ABSTRACT A importat issue i the widespread deploymet of ifrastructure based wireless local area etworks (WLANs) is the etwork desig. I this paper, we propose a ew WLAN desig approach that focuses o assurig sufficiet data rate capacity to meet expected user demad i the coverage area, while still satisfyig sigal coverage ad iterferece level requiremets. Notig the low cost of WLAN access poits, we formulate a ovel mathematical etwork desig model withi the framework of costrait satisfactio problems. Our model is termed the capacity based WLAN costrait satisfactio problem (Cap-WLAN CSP). The solutio of the Cap-WLAN CSP model yields a etwork desig based o data rate demad by providig the access poit locatios, the frequecy chael allocatio, ad power levels required for the WLAN to meet expected user demads. Our umerical results illustrate that the capacity based approach is more appropriate for the desig of WLAN systems tha those of traditioal coverage based desigs. I. INTRODUCTION Wireless local area etworks (WLANs such as those built o the IEEE 802.b stadard, are experiecig tremedous growth, providig cosumers ad busiesses mobile data etworkig capabilities that complemet the mobile voice capabilities of cellular phoes. The deploymet of WLANs ad the growth i the umber of subscribers has bee pheomeal ad is expected to cotiue or icrease at its fast pace [-3, 3]. The fact that laptop computers ad other mobile devices (e.g., PDAs) are becomig cheaper, smaller ad more powerful has drive the demad for WLAN services. A importat issue i such wide spread deploymet of WLANs is the etwork desig. The etwork desig determies the umber, locatio ad cofiguratio (e.g., frequecy, power level, etc.) of WLAN access poits (APs) ad the etwork capacity (aggregate bit rate) provided to a specific geographic area. I curret practice, WLANs are largely desiged i the basis of a trial ad error, measuremet based approach. Specifically, oe places APs i buildigs at opportuistic locatios, measures the received sigal stregth i the desired coverage areas of the buildig ad adjusts the AP locatios, power levels, frequecy chael etc., based o the observed coverage [5]. All of this is doe maually, ad is labor itesive. Such a approach is expesive ad time cosumig whe deployig large umbers of WLAN APs. I the research literature, work has appeared o developig algorithms for the desig of WLANs i a idoor eviromet [4-6]. This work seeks to create a coverage based WLAN desig; that is esurig that a adequate received sigal stregth ad sigal-to-iterferece ratio (SIR) are maitaied i the iteded service area. These papers are similar i that they formulate optimal access poit/ base statio placemet problems with very similar objective fuctios ad vary oly slightly i the assumptios ad the approaches to solve the optimizatio problem. The curret coverage based WLAN desig approach is sufficiet for small etworks of a few APs where user desity is low ad traffic load is light. However, as the umber of WLAN users ad applicatios icreases, etwork capacity becomes a issue ad a fudametally differet approach to etwork desig is required. A etwork desig solutio must accout for wireless user desity, expected user subscriber profiles, traffic models for various applicatios, ad support for QoS classes. Support for these requiremets i tur requires desigig the WLAN system based o a data rate desity criterio because the first step towards providig ay kid of QoS is to esure availability of the ecessary badwidth (data rate). However, curret work i WLAN desig igores this issue ad cocetrates o providig sigal coverage ad acceptable iterferece levels i cells. While these factors play a crucial role i the overall desig, they are ot sufficiet for guarateeig a particular aggregate data rate capacity i a specific geographic area. I the recet literature Kabara[7] ad Hills [8] discuss the eed for cosiderig of capacity requiremets i WLAN desig ad Tutschku [, 4] presets similar argumets for the cosideratio of traffic load i the desig of circuit switched cellular voice etworks. I this paper, we preset a ew desig methodology for ifrastructure based WLANs capable of supportig a data rate demad (data rate desity) i a give area. The desig methodology will determie the umber of access poits (APs frequecy chaels, power level ad the placemet of the APs, that will satisfy a set of costraits that iclude the data rate desity requiremet, radio propagatio coditios ad physical limitatios like receiver sesitivity. Ulike curret optimizatio approaches we formulate the desig problem as a costrait satisfactio problem (CSP) [2] that we will refer to as Capacity based WLAN costrait satisfactio problem (Cap-WLAN CSP). The remaider of the paper is orgaized as follows. Sectio II presets the formulatio of the Cap-WLAN CSP ad discusses a solutio techique. Sectio III illustrates etwork Fuded i part a NSF grat ANIR ad a NIST Critical Ifrastructure Protectio grat

2 desig results ad discusses computatioal complexity. Fially, sectio IV cocludes the paper. II. CONSTRAINT SATISFACTION PROBLEM FORMULATION The optimizatio based WLAN desig approaches [4-6] aim to miimize the umber of ad optimize the locatio of the access poits. This objective results i a very large ad complex solutio space ad such desig formulatios are NP hard, thus heuristic sub-optimal solutio techiques to the optimizatio models have bee proposed. However, it is uecessary i the desig of WLAN systems to miimize the umber of the APs due to the low cost of the AP compared to the wireless devices with which they are commuicatig. However, overprovisioig the service areas leads to serious system performace degradatio due to co-chael iterferece. Thus, we propose that it is more appropriate ad effective to formulate the desig problem as a costrait satisfactio problem rather tha a optimizatio problem. Our algorithm represets a service regio as discrete space of grid size m m. The grid poits represet cadidate locatios to istall APs ad specify the locatios that require radio sigal coverage. I our desig experimets we cosider two cases, the first allows APs to be located at ay grid itersectio. The secod restricts AP locatios to a more arrowly defied feasible space (e.g., located oly i hallways). The geeral Cap-WLAN CSP desig approach is show i figure. The etwork desig algorithm is structured ito two mai parts. The first part ivolves determiig the miimum umber of APs ecessary for a give service requiremet ad iitializig APs cofiguratio. The secod stage implemets a solutio algorithm for the CSP ad determies the AP parameters, icludig locatios, power levels, ad frequecy chael, such that the desiged WLAN system satisfies the service requiremets. Two sets of iput compoets are provided to the Cap-WLAN CSP algorithm. The first set of iputs defies the service eviromet icludig the spatial traffic demad distributio ad the physical structure of the service area. The secod set of iputs icorporates the path loss models that approximate radio propagatio i the give physical service eviromet. The Cap-WLAN CSP formulatio is defied by the triple (V, D, C where V = the set of variables, D = the set of fiite domais associated with the variables, ad C = the set of costraits. We represet ay two-dimesioal space with (x, y) coordiates. Let A deote the set of N access poits {ap, ap 2,, ap }, U deote the set of wireless users or demad odes i the service area {u, u 2,, u m }, GC deote the set of grid poits {g, g 2,, g c } represetig the area that requires radio coverage. We defie a set of variables V = {p j, f j, u ij, g hj, (x j, y j )} where p j is the power level of access poit j, j A; f j is the frequecy chael of access poit j, j A; u ij is a biary variable that idicates whether user i associates with access poit j or ot, i U, j A; g hj is a biary variable that idicates whether grid poit h ca receive sigal from access poit j or ot, h GC, j A; (x j, y j ) idicates the locatio of access poits, j A. We also defie the domais of the variables, D be the set of domais of the form {D p, D f, D uij, D ghj,d (xj,yj) } where: D p = the domai of p j variable for j A, D f = the domai of f j variable for j A, D uij = the domai of u ij variable = {0, } for i U, j A, D ghj = the domai of g hj variable = {0, } for h GC, j A, D (xj,yj) = the domai of (x j, y j ) variable = { x mi < x j < x max ad y mi < y j < y max } for j A. Iput : -User locatios - traffic demad -Structure of service area Feasibility check Access poit iitializatio Part : determiig #Access poits Add access poit Path loss models i = Try other frequecy chael i D f Move AP to other locatio i D (x,y) No solutio foud No solutio foud i = i+ Check costrait i pass fail Try other power level i D p No solutio foud Move operator: Brute-Force search i = N c No Yes Part 2 : CSP module Output : - # Access poits - Access poits' parameters: - locatio - Frequecy chael - Power level Figure Cap-WLAN CSP Algorithm 2

3 We defie a set of costraits C = {C, C2, C3, C4, C5, C6}. Each costrait put restrictios ad requiremets to the WLAN desig as follows: C states that each wireless termial is associated to oe access poit. C2 states that the sigal received at each wireless termial must be greater tha the receiver threshold sesitivity. C3 assures that the traffic demad of wireless termials assiged to a particular AP does ot exceed the data rate capacity of the AP. Here, we icorporate the effective capacity coefficiet (β) to capture the effects of capacity reductio due to the umber of wireless termials ad wireless termials traffic characteristics associated to the AP. C4 specifies the iterferece threshold of the wireless termial. C5 states that a portio of mea data rate from all wireless users i a service area is served by available APs. α specifies a portio of all traffic demad that we cosider servig. C6 states that the radio sigal will be available across the specified coverage space. This coditio allows the grid poit to be able to receive radio sigal from more tha oe access poits, i.e. it allows over-lappig of the access poits coverage areas. g hj = if the received sigal stregth at the grid poit h from the access poit j is greater tha P R ad the associated iterferece level is below P I ; g hj = 0 otherwise. Note that costrait C, C2, C4 ad C6 iclude the path loss model, which may vary for differet service area eviromets. To solve the Cap-WLAN CSP, we employ a brute force search techique, where variable values are tested i sequece. I the ext sectio we demostrate that a brute-force search is i some case efficiet eough. III. NUMERICAL RESULTS A. 802.b desig example Numerical etwork desig experimets were coducted for a small ad large sigle-floor service area with low ad high data rate requiremet scearios. The small service area is the fourth floor (2m 35m) of the Iformatio Sciece (IS) buildig at the Uiversity of Pittsburgh. The large service area is the first floor (60m 65m) of the Hillma library at the Uiversity of Pittsburgh. We classify wireless users ito three categories based o typical demads i this eviromet. Those with hadheld computers (smaller devices with lower computatioal power, smaller memory like a HP Jorada) access the etwork at 50 Kbps. The secod class of wireless devices eed a medium data rate of 250 Kbps as they employ laptops to mostly read ad access some Iteret sites with little multimedia cotet. The third category cosists of high data rate wireless termials who use their laptops at the fullest etwork speed possible (curret measuremets idicate that typical laptops trasmit or receive data at about 2 Mbps). These wireless termials utilize remote file systems ad streamig audio/video. The domais for the variables were tailored to curret 802.b practices; D p = {5, 20, 24} i dbm, ad D f = {2.42, 2.437, 2.462} i GHz. These experimets employed the log distace path loss model [0] to estimate radio propagatio characteristics. Iput parameters were selected with respect to the service eviromet, the buildig structure ad the 802.b specificatio. Here, 0 = 3.02 [0], K σ = 0 [0], P R = -80 dbm, P I = -90 dbm, ad C ap = Mbps [9]. The desig aims to serve Figure 2 Costrait Satisfactio Problem Formulatio: 2-D capacity-based WLAN desig Path loss model: L d d ij ( f j,( xi ( x j, y j ) = L( d0 ) + 00 log + Kσ L( d0) = 2 4πd 0 f j 0log Iput parameters: d 0 the referece distace d ij the distace betwee user i ad access poit j 0 the path loss expoet K σ the shadow fadig margi P R the received sigal stregth threshold P I the sigal iterferece power threshold C ap access poit capacity d i traffic demad from user I β access poit effective capacity coefficiet α portio of traffic demad guarateed to be served Costraits: C: uij =, i U j= C2: uij ( p j L( f j, ( xi ( x j, y j ) PR, i U j = m C3: diuij β Cap, j A i= C4: 0 log ( uik ).log pk L( f k, ( xi ( xk, yk ) k = f k f = j, i U, uij = m m C5: diuij α di j= i= i= C6: ghj, h GC j= 0 ( ) PI all traffic demad, i.e., α = o a m x m grid. Due to the cotetio based MAC protocol, it is assumed that the access poits throughput reduces to 90% of the access poit full capacity, i.e. β = 0.9. Figure 4 illustrates the case of the small service area with a coverage-based desig. I this example the WLAN system utilizes two access poits to cover the etire service areaat the specified sigal stregth threshold, give the propagatio eviromet. However, this desig does ot provide adequate capacity to all wireless users. I this etwork there are a total of 33 users, 23 users require 250 Kbps each ad 0 users require 2 Mbps each, this results i total traffic demad of Mbps. Hece, the use of two access poits ca ot support all traffic load. For this same service sceario, however, the Cap- WLAN CSP desig approach estimates the umber of APs from the aggregate traffic demad, which i this case requires a miimum of three APs. The a brute-force search algorithm determies access poits parameters, icludig locatios, power levels, ad frequecy chaels as show i figure 5. 3

4 Aother example of the small service area illustrates the more complex case where we cosider a higher user desity ad traffic demad as depicted i figure wireless users are classified as 6 of 50 Kbps wireless termials, 38 of 250 Kbps wireless termials, ad 4 of 2 Mbps wireless termials. I this case a miimum umber of four APs are required. The solutio for this case assigs differet power levels to access poits resultig i differet coverage size as depicted i figure 6. The majority of low bit rate users are covered by access poit AP. I the left ad top right corers of the service area, small groups of users demadig much higher data rate are served by smaller coverage of,, ad. The effects of desity ad distributio of wireless users o the WLAN system ifrastructure are more obvious i large service areas. Cosider the desig of the large service area with the light ad heavy load scearios. Figure 7 ad 8 show differet system cofiguratios for both scearios. Agai for the light load sceario, the desig is comparable to the coverage-based desig where sufficiet sigal stregth is provided across the service regio. We ca observe that the coverage-based desig is ot suitable for heavy load scearios, whereas the capacity-based desig approach takig ito accout user desity ad traffic load yields a effective system cofiguratio for the high traffic load eviromet. B. Computatioal Requiremets The brute-force search algorithm is simple to implemet but i the worst case, the whole solutio space must be searched exhaustively. Cosider the complexity of the CSP module for the problem with umber of variables, e umber of costraits, ad each variable cosists of a cadidate values. There are altogether a possible combiatios (cadidate solutios) of - tuples. Thus, the complexity of the exhaustive-search algorithm is O(ea ). We ca see that the complexity of the CSP module icreases expoetially with the umber of variables. Table shows computatioal times ruig o a 350MHz Su Ultrasparc II processor to solve the WLAN desig for the small service area with various user desity ad traffic scearios by the brute-force approach. Figure 3 depicts the risig of the computatioal time as the umber of wireless users ad the traffic load icrease. We see that the computatio time rises to more tha 5000 secods for eve a moderate, 90, umber of users. However, as discussed earlier, we also cosidered the case of a restricted set of locatios for the AP. I this case we see that whe we restrict AP locatios to lay withi the hallway area, which are ofte desirable locatios because of proximity to power ad wired etwork ifrastructure. We see that by restrictig the space eve the brute force search time, while growig is oly secod for a 90 user space. Table Computatioal time of brute-force search # Demad Aggregate # Access CPU time (sec) odes i service area traffic demad (Mbps) poits required Search the etire area Search alog the wall IV. CONCLUSIONS I this paper we propose a ew WLAN desig strategy called capacity based WLAN desig. The problem is formulated as a costrait satisfactio problem, which ca guaratee ot oly radio coverage to the target service area but also provide a specified data rate capacity to carry the traffic demad from each user i the service area. We coducted several desig experimets, which illustrate the beefits of the capacity-based WLAN desig approach over the traditioal coverage-based desig. Curretly, by limitig the search space to the most desirable locatios eve a brute-force search techique succeeds i reasoable time. ACKNOWLEDGMENT We gratefully ackowledge Prashat Krishamurthy for his thoughtful ad costructive commets. REFERENCES [] US News ad World Report. pp Dec3, 999. [2] J. T. Geier, Wireless LANs, 2 ed. Idiaapolis : Pretice Hall, 200. [3] A. Satamaria, ad F. J., Lopez-Heradez, Wireless LAN stadards ad applicatios Bosto: Artech House, 200. [4] M. Ubehau, "Coverage plaig for idoor wireless LAN systems," Staford/MatthiasUbehau.Abstract.pdf., Jue200 [5] D. Stamatelos ad A. Ephremides, "Spectral efficiecy ad optimal base placemet for idoor wireless etworks," IEEE Joural o selected areas i commuicatios, vol. 4, o. 4, pp , May996. [6] R. C. Rodrigues, G.R. Mateus, ad A.A. F. Loureiro, "O the desig ad capacity plaig of a wireless local area etwork," IEEE Coferece o Network Operatios ad Maagemet Symposium, pp , [7] J. Kabara, P. Krishamurthy, ad D. Tipper, "Capacity based etwork plaig for wireless data etworks," Proceedigs of IST Mobile Commuicatios Summit, Sept. [8] A. Hills, "Large-scale wireless LAN desig," IEEE Commuicatio Magazie, vol. 39, o., pp , Nov.200. [9] IEEE-SA Stadards Board. Part : Wireless LAN Medium Access Cotrol (MAC) ad Physical Layer (PHY) specificatios: Higher-speed physical layer extesio i the 2.4 GHz bad [0] T. S. Rappaport, Mobile radio propagatio: Large-scale path loss, Wireless commuicatios: Priciples & Practice, 996, pp [] K. Tutschku ad P. Tra-Gia, ``Spatial Traffic Estimatio ad Characterizatio for Mobile Commuicatio Network Desig, IEEE Joural o Selected Areas i Commuicatios, Vol. 6, No. 5., pp , Jue,998 [2] Z. Ruttkay, "Costrait Satisfactio - a Survey," CWI Quarterly, vol., o. 2-3, pp , 998. [3] J.Kabara ad D.Tipper, ``A Desig Tool for Wireless Local Area Networks, Pittsburgh Digital Greehouse, April, 2002 [4] K. Tutschku, ``Demad-based radio etwork plaig of cellular mobile commuicatio systems, INFOCOM 98, vol. 3 pp , 998 [5] B.J. Beigto ad C.R. Bartel, Wireless Adrew: Buildig a high speed campus-wide wireless data etwork, Mobile Networks ad Applicatios, vol. 6 pp.9-22,

CPU time (secods) 0000 000 00 0 0. 0.0 Search the etire area Search alog the wall Chael, 33 Users Total traffic = 25.75 Mbps AP capacity = Mbps AP Chael 0.

5 CPU time (secods) Search the etire area Search alog the wall Chael, 33 Users Total traffic = Mbps AP capacity = Mbps AP Chael # users Figure 3 Expoetially computatioal time of bruce-force search approach Figure 4 Small service area usig coverage-based desig Chael 6 33 Users Total traffic = Mbps AP capacity = Mbps Chael, 68 Users Total traffic = Mbps AP capacity = Mbps Chael Chael AP Chael Chael AP Chael 6, Figure 5 Small service area usig capacity-based desig Figure 6 Small service area with heavy load usig capacity-based desig Chael: Chael: AP4 Chael: AP4 Chael: AP5 Chael: 6 AP6 Chael: Chael: 6 AP Chael: Chael: Chael: AP Chael: AP 6 Figure 7 Large service area, light load usig capacity-based desig (Comparable to coverage-based desig) 5 Figure 8 Large service area, heavy load usig capacitybased desig

Lecture 4: Frequency Reuse Concepts

Lecture 4: Frequency Reuse Concepts EE 499: Wireless & Mobile Commuicatios (8) Lecture 4: Frequecy euse Cocepts Distace betwee Co-Chael Cell Ceters Kowig the relatio betwee,, ad, we ca easily fid distace betwee the ceter poits of two co