Joint Access Point Placement and Channel Assignment for Wireless Local Area Networks

Transcription

1 Ttle Jont Access ont lacement and Channel Assgnment for 802. Wreless Local Area Networks Author(s) Lng, X; Yeung, LK Ctaton The 2005 IEEE Wreless Communcatons and Networkng Conference (WCNC), New Orleans, LA., 3-7 March In Conference roceedngs, 2005, v. 3, p Issued Date 2005 URL Rghts Ths work s lcensed under a Creatve Commons Atbuton- NonCommercal-NoDervatves 4.0 Internatonal Lcense.

2 Jont Access ont lacement and Channel Assgnment for 802. Wreless LANs Xang Lng School of Communcaton and Informaton Engneerng Unversty of Eleconc Scence and Technology of Chna Chengdu, Chna Kwan Lawrence Yeung Dept of Eleccal and Eleconc Engneerng Unversty of Hong Kong Hong Kong, Chna Absact To deploy a mult-cell 802. wreless local area network (WLAN), access pont (A) placement and channel assgnment are two prmary desgn ssues. For a gven pattern of affc demands, we am at maxmzng not only the overall system throughput, but also the farness n resource sharng among moble termnals. A novel method for estmatng the system throughput of mult-cell WLAN s proposed. An mportant feature of ths method s that co-channel overlappng s allowed. Unlke conventonal approaches that decouple A placement and channel assgnment nto two phases, we propose to ontly solve the two problems for better performance. Due to the hgh computatonal complexty nvolved n exhaustve searchng, an effcent local searchng algorthm, called patchng algorthm, s also desgned. Numercal results show that for a typcal ndoor envronment, patchng algorthm can provde a close-to-optmal performance wth much lower tme complexty. Keywords-wreless local area networks; 802.; access pont placement; channel assgnment I. INTRODUCTION The wreless local area network (WLAN) has been an astoundng success snce IEEE 802. standards have been publshed to rule the development of WLAN. As more and more mult-cell 802. WLANs are beng deployed to provde seamless coverage, a systematc approach for determnng where an access pont (A) should be placed, and whch channel an A should be assgned become ncreasngly mportant. Efforts dedcated to WLANs desgn have produced some useful results. Rodrgues [] reported a real experence of A placement n an ndoor envronment whch ams at maxmzng the total recevng sgnal sength. Lee [2] consdered load balancng among multple As by mnmzng the load carred by the heavest-loaded A. However, these two approaches can gve no soluton when co-channel nterference between cells exsts. Kamenetsky [3] combned prunng and other refnng algorthms to solve A placement n an ad-hoc network. Leung [4] dscussed the frequency assgnment for mult-cell WLAN whle assumng an deal hexagonal coverage for each A. ark [5] noduced the throughput measurements n an offce and a conference room. Hlls [6] descrbed the general procedure n a large-scale WLAN desgn, n whch A placement and channel assgnment are carred out n two separated phases. In ths paper, we want to ontly solve the two problems, A placement and channel assgnment, by a sngle algorthm, wth a new obectve of smultaneously maxmzng the total system throughput and mantanng far resource sharng. To evaluate and compare dfferent desgn solutons, an orgnal method for throughput estmaton s frst proposed. Unlke exstng approaches, our estmaton method allows co-channel overlappng between cells. In order not to rely on the brutal force for fndng the global optmal soluton, a heurstc called patchng algorthm s also proposed. II SECIFICATION AND DESSIGN ROCESS Among the IEEE 802. standard seres, 802.b s the most-wdely deployed verson. It uses drect sequence spread specum (DSSS) as physcal layer and adopts carrer sense multple access wth son avodance (CSMA/CA) n MAC layer. The data rates supported are, 2, 5.5 and Mbps dependng on the receved sgnal senstvty thresholds (RXThresh). If the receved sgnal sength s below RXThresh but above carrer sense threshold (CSThresh), carrer busy can be detected. In an 802.b network, there are 3 channels regulated by ETSI or channels regulated by FCC. Adacent channels have specum overlappng; so generally only 3 fully solated frequency bands can be assgned for elmnatng adacent channel specal nterference. To avod son caused by hdden termnals, Request-to-Send (RTS) / Clear-to- Send (CTS) mechansm s usually adopted. The target of WLAN desgn s to satsfy the affc demands from the moble termnals whle keepng the deployment cost low. A generc desgn process for deployng a mult-cell WLAN system conssts of four phases [,2]: ) arttonng the servce area nto grds. Each grd s the basc spot for termnal countng and sgnal sength measurng. 2) Choosng canddate locatons for As. 3) Sgnal sength measurement. The measurement should be taken at every covered grd for every A canddate. The sgnal sength could also be predcted by analyss [7]. 4) Decdng A placement and channel assgnment. Our research n ths paper focuses on ths phase. III. THROUGHUT ESTIMATION To ontly carry out A placement and channel assgnment, we must have a smple and effcent way to evaluate and IEEE Communcatons Socety / WCNC /05/$ IEEE

3 compare dfferent solutons. In ths secton we propose an effcent method for throughput estmaton of a mult-cell WLAN system. Reference [8] analyzed the system performance of one A s servce regon (.e. a sngle cell), from whch the probablty ( ) that at least one moble termnal (MT) ansmts the packet n the consdered slot tme, and the probablty ( s ) that a ansmsson sezes the channel successfully are derved as follows: ) s n = ( τ () n = n τ ( τ ) (2) where n s the total number of moble termnals n a cell; τ s the probablty that a moble termnal s backoff tmer decreases to zero (and a packet wll be sent). lease refer to [8] for more detals. There are three scenaros of channel condtons shown n Fg. : (a) channel dle; (b) channel held by a successful ansmsson; and (c) channel son,.e. more than one MT ansmt at the same tme. The correspondng channel dle probablty dle, successful ansmsson probablty succ, and the channel son probablty can be obtaned as follows. = dle = succ s (3) (4) where T DIFS, T SIFS are the correspondng nterframe spaces; T RTS, T CTS, T ACK are tme for each handshakng message; T preamble s the tme consumed by preamble; T MDU- s the total ansmsson tme for MT s data packet; L LCheader, L MACheader, L FCS are the szes of the correspondng packet header and aler felds; L data- s the sze of the payload; DateRate s, 5.5, 2 or Mbps dependng on MT s receved sgnal sength. When channel son occurs, the sender wll not receve CTS frame and the son can be quckly detected; see Fg.c. So the son tme T s: T = T + T RTS If constant MAC packet sze s assumed, low data rate termnals wll hold the channel longer than hgh data rate termnals [9]. For example, f CBR servce s carred by 500- byte UD packet, the channel holdng tme for sendng a UD packet usng /2/5.5Mbps data rate are 5.80/3.6/.48 tmes of that usng Mbps. To cover the whole buldng or campus by mult-cell WLAN, some cells maybe be assgned to use the same channel. The relatonshp between such co-channel cells falls nto three types: ) Overlapped cells. In Fg. 2a, shaded areas denote the recevng range (where sgnal sength > RXThresh), and the dashed crcles denote the nterference range (where sgnal sength > CSThresh). MT can sense the carrer from both A and A 2. A MT MT 2 A 2 DIFS (8) = ) ( s (5) (a) Overlapped cells A MT MT 2 A 2 Slot RTS DIFS (b) Fully-separated cells (a) Channel dle (c) Channel son A MT MT 2 A 2 RTS CTS data ACK (b) Channel held by a successful ansmsson Fgure. Three scenaros of channel condtons Based on 802.b specfcaton, let the channel dle duraton tme T dle n Fg. a be 20 µs. In Fg. b, f RTS/CTS mechansm s adopted, the total channel holdng tme for moble termnal MT s packet ansmsson, T, s gven by: T 3 = TRTS + TCTS + Tpreamble+ TMDU + TACK + TDIFS + TSIFS (6) T SIFS L reamble SIFS LCheader MACheader + L LCheader MACheader data MDU = + Mbps DataRate L SIFS FCS + L FCS DIFS (7) (c) Incompletely-separated cells Fgure 2. Relatonshp of A cells. (The regular shaped cell s for llusaton purpose only.) 2) Fully-separated cells. In Fg. 2b, the dstance between two cells s far enough, such that the assocated MTs n one cell cannot nterfere the data ansmsson n the adacent cell. 3) Incompletely-separated cells. In Fg. 2c, two cells are separated but the dstance between them s not far enough: MT located at the boundary of cell can receve the co-channel sgnals from MT 2 who s assocated wth A 2. In ths case, MT cannot send (to A ) smultaneously as MT 2. They have mutual nterference. It should be noted that ths case of ncompletelyseparated cells has been gnored n [,2,4]. IEEE Communcatons Socety / WCNC /05/$ IEEE

4 Both contenton wthn a cell and the co-channel nterference from adacent cells can resan a MT from data ansmsson. Assume MT s assocated to A, the four possble coverage scenaros are shown n Fg. 3. For each scenaro, the area that a potental resaner of MT may resde s shaded. (The sold and dashed crcles delneate the boundares of A s recevng and nterference ranges respectvely.) A MT A 2 A MT where r s the probablty that MT holds the channel. Rsn equals to f MT s a resaner of MT ; otherwse t equals to 0. Assume each MT always has sustaned affc to send (.e. every MT wll request channel f channel dle s detected). Note that more resaners make t harder for MT to seze the channel. Ths can be seen from (0). Effcency n () s the porton of tme used by MT for ansmttng ts payload. The total system throughput THR total can be estmated by addng the respectve throughput of ndvdual MTs. THR total = THR (2) Fgure 3. Resan crtera From Fg. 3, the potental resaners of MT can be found: ) Moble termnals n MT s radatng range. The shaded area n Fg. 3a llusates where the moble termnals can be heard by MT drectly. 2) Moble termnals located n A s nterference range. Refer to Fg. 3b. 3) If MT s n A 2 s nterference range, moble termnals assocated to A 2 wll resan MT. In Fg. 3c, when A 2 communcates wth ts subscrbers, MT detects the carrer busy sgnal. MT wll keep slence untl ansmsson n A 2 s servce area ends. 4) If A and A 2 are so close that they can nterfere wth each other drectly, moble termnals assocated to A 2 are MT s resaners. Ths s shown n Fg. 3d, where A detects ansmsson n A 2 s servce area and then keeps slence. When MT requests ansmsson to A, A wll not respond. Based on the potental resaners dentfed above, MT s throughput THR can be found: THR = DataRate r Effcency (9) r = succ [ T + + rsn (a) Effcency + T succ rsn (0) ( rsn * T )] + L dle (b) A MT A 2 A 3 MT A A 2 A 3 (c) (d) * T / DataRate slot + * T data = () T IV. NEW OBJECTIVE FUNCTION Several obectve functons (OF) for A placement and channel assgnment were proposed n the lteratures. athloss mnmzaton was proposed n [3]. Although A placement can be determned by pathloss, t cannot be used for solvng the channel assgnment problem. In ths case, A placement and channel assgnment must be splt nto two separated phases. Load balance was chosen as the obectve functon n [2], and the resultng algorthm cannot solve the channel assgnment problem effectvely. In fact, prohbtng co-channel overlappng s the only consant for channel assgnment n [2]. If more and more As are placed for achevng hgher system throughput, co-channel overlappng occurs nevtably. If ths happens, the algorthm n [2] cannot return feasble soluton. In ths secton, we desgn a new obectve functon for ont optmzaton of A placement and channel assgnment. Ths functon, shown n (3), ams at optmzng both the system throughput and the farness among moble termnals. It s a product between THR total from (2) and the farness ndex β from (4). (Note that A placement and channel assgnment are carred out n the system desgn phase; a detaled throughput analyss based on the actual affc generaton s not needed. And the sustaned affc demand represents the worst stuaton that system wll encounter.) In (4), β s defned as a measure of the devatons of the throughputs acqured by ndvdual MTs. OF β = THRtotal ( N = = N N = β THR ) 2 THR 2 (3) (4) N s the total number of MTs n the system. The farness ndex β approaches when all MTs have exactly the same throughput. When the throughputs are heavly unbalanced, β converges to /N. V. ATCHING ALGORITHM Based on the obectve functon n (3), an optmal soluton for ont A placement and channel assgnment can be found by exhaustve searchng. If the number of A canddates s IEEE Communcatons Socety / WCNC /05/$ IEEE

5 large, the exhaustve search becomes exemely complex. Assume M As should be chosen from L canddates and three solated channels are avalable for channel assgnment. The number of possble solutons s on the order of M M C (3 L + ) / 2. Brutal force searchng s thus not feasble. In ths secton, we propose a tme-effcent local searchng heurstc, called patchng algorthm. atchng algorthm places As one by one to cover the affc demands untl a pre-defned number of As are placed. At each step, patchng algorthm attempts to select one A from the remanng canddate pool, whch can provde the largest OF value together wth those already placed As. The pseudo-code for patchng algorthm s outlned n Table I. TABLE I. ATCHING ALGORITHM SEDO-CODE k=; //Now we y to place the k-th A whle (k the number of As planned to place){ =the st A canddate n the canddate set; whle ( the last A canddate n the canddate set){ place A temporarly; =; //If J channels avalable, =~J whle ( J){ assgn channel to A temporarly; calculate the OF for the current k placed As, record t as OF ; ++; //y to assgn another channel nstead of prevous channel } =next; //y to place another canddate nstead of prevous A } pck up the largest OF ; A s placed and channel s assgned to t permanently; update canddate set; k++; //ready to place the next A } converge on the local optmum. The patchng algorthm starts wth an ntal canddate set that contans all canddate As,.e. {A,A 2, A L }. Then the canddate that gves the largest OF value s selected for frst A placement. The A canddate set s updated by deletng the placed A. To select and place the 2 nd A, we consder the remanng L- canddate As together wth three possble channel assgnments each. The canddate A wth a sutable channel that gves the largest OF value among all possble combnatons s then selected for the second A placement and channel assgnment. Ths process s repeated untl a predetermned number of As are placed. In each teraton, newly placed A may cause some moble termnals to re-assocate wth t, and the resan range of some moble termnals wll also be affected. If ths happens, the throughput of those affected moble termnals should be re-estmated usng (9). VI. ERFORMANCE EVALUATIONS A. Throughput Estmaton Verfcaton Frst a smple scenaro of two overlapped cells (or As) s smulated by NS-2 to verfy the throughput estmaton method we proposed n Secton III. The network topology ncludes two As; MTs are unformly dsbuted and communcate wth the fxed network va these two As. UD connectons are establshed from the MTs to the fxed network for carryng 500-byte CBR affc. Let the wreless nterface be the bottleneck n the ansmsson. The ansmsson power s set to 20dBm. The RXThreshs are -75/-79/-8/-84dBm for dfferent data rate recevers, and CSThresh s set to -94dBm. If a same channel s assgned to these two As, the estmated throughputs and those obtaned by smulatons are compared n Fg. 4. The x-axs s the dstance between the two As. When Dstance=0m (.e. these two As are fully superposed), there are 3/8/4/2 moble termnals workng at /5.5/2/Mbps data rates respectvely, so the system throughput s gven by THR total = 3 Mbps Mbps Mbps Mbps =. 73Mbps From the equaton above, we can also see that the channel holdng probablty for each MT wth Mbps data rate s ; and ther throughput effcency s Note that for Dstance=0m, all MTs have the same resan range. It s evdent from Fg. 4 that the system throughput s poor when the dstance between As s less than the nterference radus R CS ( 062m). In ths case, the two co-channel As cannot be fully utlzed. If the dstance s longer than R CS, both As work wth full utlzaton, so the system throughput s doubled to gve 3.47 Mbps. System throughput (Mbps) co-channel estmaton co-channel smulaton dff-channel estmaton dff-channel smulaton Dstance (m) Fgure 4. Total co-channel and dff-channel throughput accordng to the dstance between two As Fg. 4 also compares the scenaro that dfferent channels are assgned to the two As, denoted by dff-channel. We can see that the dff-channel overlappng between cells can slghtly mprove the system throughput. Ths s because moble termnals resdng at the boundary of A can only work at low data rate f they assocate to A ; when overlaps wth another A 2 of a dfferent channel, some low data rate termnals can now re-assocate to A 2 ; then those termnals can get a hgher data rate. Meanwhle, the overlappng between dff-channel cells wll not cause nterference. To conclude, the above study shows that our throughput estmaton method can provde a satsfactory predcton wth or wthout co-channel overlappng. The results n the subsequent subsectons are thus based on ths throughput estmaton. IEEE Communcatons Socety / WCNC /05/$ IEEE

6 B. Envronmental Assumpton Followng the desgn process descrbed n Secton II, the servce area parttonng for a typcal buldng floor s shown n Fg. 5, wth a grd granularty of 5m. A non-unform dsbuton of 0 to 6 MTs per grd s llusated by dfferent colors (or grey levels). The hotspots, where the MT densty s substantally hgher can be found. The 6 A canddate locatons are gven and numbered from 0~5. The sgnal sength s predcted by Two-Ray-Ground model [2]. The pathloss values of sgnal avelng across concrete wall and turnng rght angle at the corner are 5dB and 0dB. The predcted recevng sgnal sengths n dbm for canddate A 2 are shown n Fg. 5 as an example System throughput (Mbps) Farness ndex OF OF2 OF3 0 (a) System throughput OF OF2 OF3 0.5 (b) Farness ndex Fgure 6. System throughput and farness ndex of optmums found by exhaustve search based on three obectve functons Fgure 5. MT dsbuton and canddate A locatons n a buldng C. Obectve Functons and Global Optmums For convenence, let the obectve functon defned n (3) be OF, the obectve functon proposed n [2] be OF 2, and the obectve functon adopted n [] be OF 3. OF 2 attempts to balance the loads among As. OF 3 es to maxmze the sum of receved sgnal sengths. Fg. 6 compares the optmal solutons obtaned by exhaustve searchng based on OF, OF 2 and OF 3. We can see that our ont A placement and channel assgnment based on OF provdes the hghest system throughput and close-to-best farness ndex. Note that more than one optmal soluton can be found based on OF 2 and OF 3. Ther dsbutons are llusated and ther average values are connected by dashed lnes. From Fg. 6, we can see that there s no soluton for OF 2 when less than 6 As are placed. Ths s because optmzaton method n [2] has a consant that every MT must be served. If less than 6 As are placed, no soluton can be found to cover all termnals. Besdes, when more than 8 As are placed, there s no soluton for both OF 2 and OF 3. Ths s because those placement algorthms do not allow co-channel overlappng. If more than 8 As are placed, co-channel nterference between As cannot be avoded. D. atchng Algorthm and Local Optmums The patchng algorthm descrbed n Secton V can sgnfcantly reduce the computatonal complexty. For solvng the same problem as that n the prevous subsecton, Table II summarzes the number of solutons that are vsted by the exhaustve searchng and the patchng algorthm respectvely. TABLE II. Vsted As solutons COMUTATIONAL COMLEXITY Exhaustve searchng atchng algorthm The OF values found by the exhaustve searchng and the patchng algorthm are compared n Fg. 7. We can see that the performance of patchng algorthm s very close to the optmal. We next compare the patchng algorthm wth another algorthm run+rcc n Fg. 8. run+rcc spts A placement and channel assgnment nto two separated phases. The prunng algorthm (run) s adopted n A placement phase for mnmzng pathloss [3], and a random channel convergence (RCC) algorthm s used n the channel assgnment phase [4]. RCC adusts the channel assgnment of the bottleneck A s IEEE Communcatons Socety / WCNC /05/$ IEEE

7 OF value Global optmum patchng algorthm Fgure 7. OF values got by patchng algorthm and exhaustve searchng System throughput (Mbps) Farness ndex (a) System throughput patchng prun+rcc patchng prun+rcc 0.5 (b) Farness ndex Fgure 8. System throughput and farness ndex of local optmums neghbors to mnmze ts affc load. From Fg. 8, we can see that more than one solutons (> 20) can be found for run+rcc algorthm. Ths s because the RCC algorthm only adusts the channel assgnment n bottleneck A s neghbors. For a heavly unbalanced moble termnal dsbuton, the bottleneck A usually locates at the hotspot area. Adustng the channels n ts neghborng As cannot reduce the load of the bottleneck A effectvely. If the bottleneck cannot be removed from the hotspot to other cooler spots, ths algorthm cannot optmze the channel assgnment n lght affc demand areas. As such, channels can be assgned to the As n lght affc demand areas freely; ths generates many local optmal solutons. Overall speakng, Fg. 8 shows that splttng A placement and channel assgnment nto two phases leads to poor performance than patchng algorthm. VII. CONCLUSIONS In ths paper, we frst proposed an orgnal method to estmate the throughput of a mult-cell 802.b WLAN system. The co-channel overlappng between cells s allowed. Then an optmzaton method for ont A placement and channel assgnment was proposed, whch ams at maxmzng both the system throughput and the farness ndex. Its optmal solutons can be found by exhaustve searchng. To reduce the computatonal complexty, a smple local searchng heurstc, called patchng algorthm, was desgned and analyzed. We showed that n a typcal ndoor envronment, patchng algorthm provdes close-to-optmal system throughput and farness ndex. REFERENCES [] R.C.Rodrgues, G.R.Mateus, and A.A.F.Lourero, On the desgn and capacty plannng of a wreless local area network, IEEE/IFI Network Operatons and Management Symposum [2] Y.Lee, K.Km, and Y.Cho, Optmzaton of A placement and channel assgnment n wreless LANs, roceedngs of the 27 th Annual IEEE Conference on Local Computer Networks, [3] M.Kamenetsky and M.Unbehaun, Coverage plannng for outdoor wreless LAN systems, Internatonal Zurch Semnar on Broadband Communcatons, [4] K.K.Leung and B-J. J Km, Frequency assgnment for mult-cell IEEE 802. wreless networks, VTC 03 Fall. [5] J.A.ark, S.K.ark,.D.Cho, and K.R.Cho, Analyss of specum channel assgnment for IEEE 802.b wreless LAN, 5 th Internatonal Symposum on Wreless ersonal Multmeda Communcatons, [6] A.Hlls, Large-scale wreless LAN desgn, IEEE Communcatons Magazne, vol. 39, no., pp , Nov 200. [7] S.Armour, A.Doufex, B-S.Lee, A.Nx, and D.Bull. The mpact of power lmtatons and adacent resdence nterference on the performance of WLANs for home networkng applcatons, IEEE Trans on Consumer Eleconcs. vol. 47, no.3, pp. 502~5, August 200. [8] G.Banch, erformance analyss of IEEE 802. dsbuted coordnaton functon, IEEE Journal on Selected Areas n Communcatons, vol. 8, no. 3, pp. 535~547, March [9] M.Heusse, F.Rousseau, G.Berger-Sabbatel, and A.Duda, erformance anomaly of 802.b, Infocom2003. [0] F.Cal, M.Cont, and E.Gregor. IEEE 802. wreless LAN: capacty analyss and protocol enhancement, Infocom 98, 998. [] Y.Xao and J.Rosdahl. Throughput and delay lmts of IEEE 802., IEEE Communcatons Letters. vol. 6, no. 8, August [2] IEEE Communcatons Socety / WCNC /05/$ IEEE