Aalborg Unverstet Realstc Indoor W-F and deployment Study as the Offloadng Solutons to LTE etwork Hu, Lang; Colett, Claudo; guyen, Huan Cong; Kovács, Istvan; Vejlgaard, Benny ; Irmer, Ralf ; Scully, el Publshed n: Vehcular Technology Conference (VTC Fall), 2 IEEE DOI (lnk to publcaton from Publsher):.9/VTCFall.2.639942 Publcaton date: 2 Document Verson Accepted author manuscrpt, peer revewed verson Lnk to publcaton from Aalborg Unversty Ctaton for publshed verson (APA): Hu, L., Colett, C., guyen, H. C., Kovács, I., Vejlgaard, B., Irmer, R., & Scully,. (2). Realstc Indoor W-F and deployment Study as the Offloadng Solutons to LTE etwork. In Vehcular Technology Conference (VTC Fall), 2 IEEE (pp. -6). IEEE Press. I E E E V T S Vehcular Technology Conference. Proceedngs https://do.org/.9/vtcfall.2.639942 General rghts Copyrght and moral rghts for the publcatons made accessble n the publc portal are retaned by the authors and/or other copyrght owners and t s a condton of accessng publcatons that users recognse and abde by the legal requrements assocated wth these rghts.? Users may download and prnt one copy of any publcaton from the publc portal for the purpose of prvate study or research.? You may not further dstrbute the materal or use t for any proft-makng actvty or commercal gan? You may freely dstrbute the URL dentfyng the publcaton n the publc portal? Take down polcy If you beleve that ths document breaches copyrght please contact us at vbn@aub.aau.dk provdng detals, and we wll remove access to the work mmedately and nvestgate your clam.
Realstc Indoor W-F and Deployment Study as the Offloadng Soluton to LTE etworks Lang Hu*, Claudo Colett*, guyen Huan*, István Z. Kovács, Benny Vejlgaard, Ralf Irmer, el Scully * Aalborg Unversty, Aalborg, Denmark oka Semens etworks, Aalborg, Denmark Vodafone Group R&D, ewbury, Unted Kngdom Abstract Ths paper nvestgates the downlnk performance of ndoor deployed W-F and as the offloadng soluton to the LTE macro cellular networks n a realstc large-scale denseurban scenaro. Wth an assumed broadband traffc volume growth of x compared to today s levels, t s evaluated that a dual-carrer LTE macro network wll not be able to provde suffcent servce coverage wth a Mbps mnmum data rate and, ndoor coverage s dentfed as the major bottleneck.. We evaluate the performance of ndoor W-F and cell deployment to offload the congested LTE macro network. We show that, n a dual-carrer LTE macro case wth a total of MHz spectrum, W-F access pont densty of 2/km2 s requred to meet the set target of 9% coverage wth a mnmum user data rate of Mbps. For the same scenaro t was found that an outband access pont densty of /km2 s requred. Furthermore, we show that n-band cell cannot meet the set network requrement even at a very hgh access pont densty. We also show that W-F and cell can offload the same amount of traffc when they are deployed at the same access pont densty. I. ITRODUCTIO Moble data traffc s growng explosvely wth the popularty of varous moble devces that offer ubqutous moble nternet and dverse multmeda authorng and playback capabltes. For nstance, Vodafone has seen ts data traffc grow from a trckle to a pont where t almost exceeds voce traffc already n 8; AT&T has seen a data growth of over % from 8-; Csco [] predcts that overall moble data traffc s expected to grow to 6.3 exabytes per month by 5, a 26-fold ncrease over, where moble vdeo traffc accounts for 66.4% of the total traffc. Despte the tremendous data traffc growth, operators are facng the bg challenge that the revenue per user s decoupled from the data traffc generated per user, e.g. under the current mostly adopted flat-rate prcng: whereas the data traffc grows exponentally, the revenue growth s rather slow, e.g. moble data traffc ncrease by % annually, whle the revenues only ncrease by 6% annually [2]. To allevate ths challenge, operators have to consder a cost-effectve way to evolve ther moble networks to accommodate explosve traffc as well as keepng hgh revenue. W-F s recognzed by key moble operators as a promsng soluton for cost-effectvely addng moble network capacty by leveragng low-cost access ponts and free unlcensed spectrum. W-F s a mature and wdely adopted technology n most moble devces. Mllons of exstng user deployed resdental access ponts potentally already offload a lot of moble data traffc. Thus, W-F can offer tme-to-capacty advantage over other network evoluton optons e.g. by addng more /Mcro Base Statons or upgradng 3G Base Statons to 4G LTE (Long Term Evoluton) Base Statons. As another moble data offloadng soluton, cell s a knd of low-power base staton whch s usually nstalled by end-users n the resdental and enterprse places. It ams at mprovng network coverage and capacty by ste densfcaton and ncreasng the spectrum spatal reuse. Smlar to W-F, cell utlzes the end-users fxed DSL lne as the backhaul network and s free of ste acquston fee, whch also makes cell a cost-effectve soluton. There are very few quanttatve studes on realstc W-F offloadng potental and especally the comparson wth cell offloadng, n partcular n large-scale real deployment scenaros. A recent paper [3] has studed W-F offloadng by usng the measured user moblty traces as a bass to evaluate the offloadng potental of exstng resdental W-F networks. Along another lne[4][5], there are also a few studes on the performance of offloadng based on 3GPP regular network assumptons. In contrast, we provde a comprehensve quanttatve study on W-F and cell offloadng n a large-scale real dense-urban deployment scenaro. Outdoor Pco cells can also handle ncreasng traffc, however ths paper focuses on W-F and cell offloadng. II. ETWORK MODELIG FRAMEWORK A. Cellular etwork Layout and Real Buldng Database Ths study has been carred out n a dense urban scenaro- a LTE macro cellular deployment n a European cty. The sze of the nvestgated area s approxmately.27 km 2, contanng 4 three-sector macro stes wth optmzed antenna down-tlt and average nter-ste dstance of 3 m. Furthermore, nterferng cells from base statons located outsde the nvestgated area are consdered to remove border effects. Each sector s assumed to be equpped wth 2 carrers, operatng at 8 MHz and 26 MHz bands. Furthermore, the studed area s dvded nto pxels wth x m resoluton. The offered traffc load s defned as the number of smultaneously connected users wth a mnmum data rate of Mbps average durng the peak hours. In, the measured 3G data traffc load s on average.6 users over the nvestgated area and t s predcted to ncrease by a factor of x to 558 users for the purpose of ths study. 978--4673-88-5/2/$3. 2 IEEE
The offered traffc load s fxed throughout our study. For accurate ndoor modelng, real 3D buldng database of the nvestgated area s employed as shown n Fg.. There are 96 buldngs wth 5 floors on average per buldng. It accounts for 36% of the total area. The buldng heght s also consdered: ) when estmatng the path loss between macro cells and outdoor locatons; 2) floor penetratons n ndoor propagaton 3D Propagaton Model. To accurately estmate lnk budgets, a 3D ray-tracng tool s used to evaluate path loss and antenna pattern effects wth regard to the rado lnk between macro cells and outdoor users. Such a tool models the rado propagaton at street level by consderng realstc postons and heghts of the buldngs that are mported from the prevously mentoned 3D buldng map, as shown n Fg. Gven the outdoor path loss predctons from ray-tracng tool, the ndoor penetraton loss wthn the buldng s calculated through an addtonal loss (n db) equal to.6 d + L extwall, where d s the dstance (n meters) from the ndoor locaton to the external wall observng the hghest receved sgnal strength, and L extwall defnes the penetraton through the external wall that s set at db. Furthermore, for the outdoorndoor path loss calculaton, the floor heght gan s modeled such that users located at hgher floors have a receved sgnal strength gan of 3.4 db/floor. Fgure. Path Loss (db) Predcton from a specfc cell ncludng 3D- Buldng nformaton and real antenna radaton pattern. Fgure 2. Outdoor -Indoor Floor Heght Gan and 3D user dstrbuton When consderng ndoor small cells (WF/), a statstcal model based on [6] s consdered, and t s defned as follows: PL nd (db) = 38.46 + log R +.6 d + L () 2 D,ndoor where R s the dstance between the small cell and a generc user (ndoor or outdoor), d 2D,ndoor s the dstance covered nsde the buldngs and L ow s a penetraton loss of db due to each penetrated external wall. In terms of fadng effects, fast fadng effect s not modeled snce users are statc n our study; slow fadng effect s captured by ray-tracng tool for outdoor base ow, statons and s not modeled for ndoor base statons to mprove the smulaton speed. B. Spatal Traffc Modelng The network traffc load s smulated n terms of number of smultaneous actve users, randomly placed n the map accordng to a spatal user densty map. Essentally, the user densty map s defned as a probablty of placng a user n a pxel of the map. There are 4 steps to generated spatal user densty map: ) the spatal user densty map was frstly derved from cell-level packet-swtched (Release 99 + HSDPA) traffc measurements averaged for busy hour traffc condtons. By assumng each user generates the same amount of traffc, traffc densty by the number of smultaneous actve users per cell s equvalent to the traffc densty by the carred traffc per cell; 2) on top of cell-level measurement, we also dfferentate ndoor and outdoor traffc; Typcally, n each cell coverage area, we force 7% of the traffc to be generated from ndoor area and % from outdoor area; 3) To obtan even fner granularty of the traffc densty map, on top of the above, traffc hotspot s artfcally generated by overlayng a log-normal dstrbuton wth a standard devaton of 4 db and a correlaton dstance of m; 4) To model the user dstrbuton on 3D mult-floor buldngs, the traffc densty of each pxel s further dvded among varous floors of that pxel as shown n Fg 2; Here we assume the ground floor accounts for % of the total traffc densty of that pxel and the remanng % are equally dvded nto hgher floors; In practce, the ground floors are usually shops or conference rooms whch generate more traffc than hgher floors. C. LTE etwork Modelng The user assocaton has two phases: ) user s assocated to the base staton that gves the best experenced Sgnal Interference ose Rato (SIR) across all carrers; 2) When the macro sectors have multple carrers, user assocaton s balanced among multple carrers by beng preferably allocated to the carrer wth larger product of experenced user SIR and avalable rado resource of that carrer,.e., user are prortzed to hgh SIR and hgh bandwdth carrer. The carrer aggregaton s not modeled.e., users only connect to one carrer. When users connect to or cells, the peak physcal layer user data rate s a functon of the average receved SIR at the user locaton and s approxmated by usng the SIR to spectrum effcency (SE) mappng method n [7] smlar to the studes n [8]. Rado resources are shared wth the purpose of mnmzng the number of users n outage,.e., the users who are experencng a data rate lower than a requred mnmum data rate. Gven the fxed amount of rado resources avalable per cell, the resource sharng algorthm sorts the connected users n descendng order accordng to ther experenced SIR. Then, the resources are allocated to the sorted lst of users so that whenever possble each user acheves the mnmum requred data rate. Fnally, when applcable, the remanng cell resources are allocated equally to all the served users n a round robn way. D. etwork Key Performance Indcator(KPI) The selected network KPI s the network outage level at a gven mnmum data rate, defned as the probablty: P = Pr[ R < r mn ] ()
where r mn [Mbps] s the mnmum user data rate requred for achevng acceptable user experence, R [Mbps] s the user data rate experenced on average by the -th user. It means that there s a threshold r mn [Mbps] below whch the user experence becomes unacceptable. The uplnk (UL) and downlnk (DL) data rate requrements are defned UL DL as r mn r mn respectvely. In our study, the network KPI s partcularly defned as 9% network coverage (maxmum % network outage) wth mnmum data rate of Mbps n DL and.25 Mbps n UL. Our W-F and cell deployment are drven by meetng the DL KPI, as DL s often the bottleneck of the network performance. III. WI-FI PERFORMACE MODELIG In ths study, we manly look nto the performance of 82.g W-F as the offloadng soluton, as t s the most popular nstalled 82. nterface for the tme beng and avalable n most smartphones and netbooks. As 82.n s gettng more and more popular, we also plan to study 82.n n future work. A. Physcal layer performance mappng curve Peak Data Rate (Mbps) 55 45 35 25 5 Curve-AWG (extended 82.a/g standard RSSI to TP mappng) Curve2- (Ref paper, BER e-5) 6 db Curve-AWG Curve2-Ref IEEE 82.g standard IceFyre 5 5 5 25 SIR (db) Fgure 3. W-F g SIR to PHY Peak Data Rate Mappng We frstly model IEEE 82.g physcal layer performance usng SIR to physcal peak data rate mappng curves, as shown n Fg 3. In the mappng curve, we assume that the frame sze s fxed to Bytes and that the packet error rate s %. In Fg 3, there are two mappng curves: the blue one s the mappng curve from IEEE 82.g standard under Addtve Whte Gaussan ose (AWG) channel condton, whereas the green curve s from a real 82.g W-F product (IceFyre Semconductor [9]) where a fadng channel condton s assumed. In ths paper, we assume a fadng channel condton and use IceFyre curve as the physcal layer performance bass of 82.g. It means that users need to have at least 6 db SIR value to be able to connect to W-F access pont. B. W-F rado resource sharng model In contrast to centralzed schedulng n cellular wreless system, W-F uses dstrbuted CSMA/CA (Carrer Sensng Multple Access / Collson Avodance) as the MAC (Medum Access Protocol) protocol for rado resource sharng. Due to protocol overhead e.g. DIFS, exponental back-off, the rado resource usage can be qute low e.g. 55% for a sngle 82.g user operatng on 54 Mbps data rate mode. Besdes, collsons can happen from tme to tme, whch further reduces the rado resource usage. To model rado resource usage, we employ the well-known Bchan s model []. The rado meda usage effcency s defned as P, whch s a functon of the number of users n the W-F cell and ther experenced nstantaneous SIR (Sgnal to Interference and ose rato). Due to lmted space of the paper, mathematcal dervaton of P can be found n []. Havng the rado resource usage P, the long-term average user throughput can be modeled as follows: we assume that each user generates both downlnk (DL) and uplnk (UL) traffc (beng a DL and UL user at the same tme), and there are DL users and thus also UL users n the W-F cell. To model the asymmetry load of DL and UL, a full-buffered traffc model s appled n DL whereas fnte-buffered model s appled n UL,.e., access pont always has DL frames to transmt, whereas users have UL frames to transmt wth probablty β. Defne the physcal layer peak data rate of user as PHY, whch s obtaned from the mappng curve n secton III A. The DL and UL throughput for user s computed as follows: DL β (2) Throughput = P β + PHY j PHY =... =.. j PHY UL j j Throughput =.. (3) P + β PHY j PHY =... =.. j The equatons (2) (3) are derved by usng a key property of 82. W-F networks [] under full-buffered traffc model - Throughput Farness: ) set of DL users of the same W-F cell have the same average throughput n the long term, ndependent of ther SIR; 2) set of UL users of the same W- F cell have the same average throughput n the long term, ndependent of ther SIR. To make UL/DL throughput rato UL DL fulfll the rato r mn / r mn defne n secton II, β s set as: UL r β = mn (4) DL rmn Due to the lmted space of the paper, the mathematcal dervaton of the model s not presented here. Ave User Tput (Mbps) 2 8 6 4 2 One WF Cell, Ave SIR=.8 db, UL/DL Rato=.25 Ave DL Tput Est Ave UL Tput Est Ave DL Tput Sm Ave UL Tput Sm 2 3 4 5 6 7 8 9 umber of Users Fgure 4. WF Throughput Model Valdaton
The analytcal W-F throuhgput model s valdated va a dynamc system-level 82.g smulator. Asume one W-F cell and every user has SIR of.8 db, the user throughput s evaluated agant number of users n the cell. As shown n Fg.4, the model matches perfectly wth the system-level smulaton results, especally when the number of users s large. IV. SIMULATIO ASSUMPTIOS & CASES DESCRIPTIO The W-F and offloadng potental have been evaluated by a MATLAB-based network plannng and statc smulaton tool. Four smulaton cases (see ) are consdered: as a reference scenaro, the LTE 2-carrer macro network layout s based on exstng lve 3G macro network layout whch s then upgraded to LTE. Second case presents ndoor W-F deployment on top of the macro reference layer, for varous access pont denstes. The last two consder out-band and n-band cell deployment respectvely to complement the macro network. In-band means cell operates on the same carrer as cell, whereas outband operates on a dfferent carrer than the carrer of cell. Both W-F and cells transmt at dbm and they are equpped wth omn drectonal antennas. 558 actve users are generated n the full network area of.27/km2 wth 96 buldngs (accounts for 36% of the total area). The average number of floors per Buldng s 5. The target KPI s a mnmum data rate of Mbps at 9% network coverage. Ths KPI s chosen to reflect the requrement that as many users as possble have a good user experence, rather than a smaller proporton of users havng a very hgh data rate. Wth the above assumptons, from the network smulaton, the reference macro-only network s only able to reach a network outage of 46 % compared to a target value of % and average user throughput of 6 kbps. To mprove macro network performance so as to meet the KPI, ndoor offloadng solutons-w-f and cells are requred. TABLE. SIMULATIO CASES AD SPECTRUM ALLOCATIO OVERVIEW Smulaton Cases -only Reference case & W-F & Out-band & In-band 8 MHz ( st ) (FDD, MHz) (Tx. Power 43 dbm) Spectrum Allocaton 26 MHz (2 nd ) (FDD, MHz) (Tx. Power 43 dbm) (Tx Power dbm) (Tx Power dbm) MHz ( MHz) - W-F (Tx. Power dbm) V. SIMULATIO RESULTS In ths secton, we demonstrate W-F and cell offloadng gan by extensve smulaton results. In subsecton A, we provde the fxed smulaton parameters. In subsecton - - B, we study the W-F offloadng gan under varous access pont denstes and compare t aganst LTE cell offloadng. The fxed smulaton parameters of W-F are shown n Table 2. Mult-rado technologes-lte and W-F 82.g are smulated smultaneously. We assume that all user termnals are equpped wth both LTE and 82.g rado nterfaces. We assume traffc steerng polcy between W-F and LTE as follows: whenever the user detects a W-F access ponts, t wll always frstly connect to W-F on the condton that t has at least SIR of 6 db and can get the mnmum data rate ( Mbps) f connected to W-F,; Otherwse, t connects to LTE macro network. A. Smulaton Parameters Table 2. Fxed Smulaton parameters of W-F Parameter Settng Rado standard W-F 82.g Frame sze Carrer frequency W-F channel deployment Traffc Steerng Polcy Deployment opton Buldng model Mnmum ISD of APs AP to ste UE admsson mode Traffc model Bytes MHz In-band deployed, MHz band Always connect to W-F before Indoor Traffc-drven deployment 96 3D buldngs, 36% of the total area 5 floors on average m m Open Subscrber Group Full buffered Indoor/Outdoor traffc rato 7% / % Spatal traffc modelng Table 3 Fxed Smulaton parameters of LTE cell Parameter Rado standard Carrer frequency Traffc Steerng Polcy See secton II C Settng LTE 26 MHz, MHz band In-band : Best server SIR Out-band : wth Range Extenson Best server SIR wth 3 db bas towards cell We assume that all W-F APs operate on the same channel of MHz bandwdth, whch corresponds to the worst case scenaro. Assgnng dfferent non-overlapped channels to APs can further optmze the performance. Yet, n practce, ths further gan can be cancelled out by consderng external W- F APs nterferences at 2.4GHz band. We assume a fxed MAC frame sze of Bytes. W-F ndoor traffc-drven deployment s consdered, where APs are placed n ndoor traffc hotpot area. The mnmum nter-ste dstance (ISD) of access ponts s set to m whch corresponds to the ndoor W-F coverage dameter. We assume Open Subscrber Group (OSG) model where users can be admtted to any of the W-F access ponts.
The fxed smulaton parameters of cell s lsted n Table 3. In both out-band and n-band case, cell s always deployed at 26 MHz band. network s deployed at both 8 MHz and 26 MHz band for n-band case and at 8 MHz for out-band case. In terms of traffc steerng for n-band cell, a pure best server SIR detecton method s assumed.e., users always connect to the base staton (ether or macro ste) wth the hghest SIR over all carrers. o Range Extenson (RE) feature s assumed for n-band case, snce t may result n rado lnk falure durng user moblty. o cross-ter nterference mtgaton scheme s modeled. In contrast, n the out-band case, RE s appled n the form of 3dB SIR bas towards cell n the best server SIR detecton,.e., users gan 3 db SIR bas towards cell n the best server detecton among all base statons. Other parameters are the same as W-F such as ndoor traffc-drven deployment, OSG, macro ste and AP mnmum ISD, and traffc modelng. B. W-F and offloadng gan VS Access Pont Densty We frstly study the W-F and cell offloadng gan n terms of network outage at mnmum data rate of Mbps. All our network evoluton studes are drven by the requred node deployment densty to meet the KPI - % network outage at mnmum data rate of Mbps. As shown n Fg 5, the reference dual-carrer LTE macro network has the network outage of 45% at mnmum data rate Mbps. W-F access pont (AP) densty of /km2 mprove network outage dramatcally by 38 percentage ponts.e., network outage from 45% to 7%, by usng traffc-centrc ground-floor only (GF-Only) ndoor deployment. In partcular, the APs are deployed n ndoor traffc hotspot and at ground-floor of the buldng. Ths means 2 AP/km2 s more than suffcent to reach the network KPI - % network outage. From Fg 5, t s also observed that ndoor W-F deployment can effcently mprove network outages both from ndoor and outdoor network areas, e.g. wth W-F at 2 AP/km2, the ndoor outage mproves from 32% to 2%, whle the outdoor network outage mproves from 4% to 5%. Indoor W-F deployment offloads the heavy load of macro network so as to mprove overall network performance. Lastly, smulaton results show that all outage users are from macro network whle W-F cell has zero users n outage. etwork Outage % GF-Only: Indoor/Outdoor user outage 8 2 39 umber of WF APs/km2 Outdoor Indoor % network outage KPI Fgure 5. W-F deployment: Indoor / Outdoor network outage Fg 6 and 7 show the out-band and n-band LTE cell offloadng gan n terms of network outage at mnmum data rate of Mbps for both ndoor and outdoor area. In the outband case, cell s deployed at 26 MHz band wth MHz bandwdth whereas macro network s deployed at 8 MHz band wth MHz bandwdth. Range extenson of 3 db SIR bas s appled to mprove the percentage of users connected to cells. Fg 6 shows that cell APs/km2 are needed to meet the network KPI % network outage at Mbps. Compared to the W-F case, the macro network has only a sngle carrer at 8 MHz wth MHz band, thus 4 tmes hgher AP densty than W-F s needed to reach the KPI. Smulaton results also show that all network outage s from macro network whle cell has zero outage users. In the n-band case, the stuaton s even worse n Fg 7. Even cell APs/km2 s not suffcent to reach the network KPI % network outage at Mbps. Ths s manly due to the fact that APs create strong n-band nterference to macro network at 26 MHz as shown n Fg 8. cell has very good SIR wth on average 4.7 db, however macro carrer at 26 MHz only has average SIR of.7 db. In ths case, Range Extenson combned wth nterference mtgaton schemes such as eicic can be expected to mprove n-band cell offloadng performance. etwork Outage % 8 7 6 GF-Only: Indoor/Outdoor user outage Range Extenson wth 3 db SIR 8 umber of APs/km2 Fgure 6. Out-band : Indoor / Outdoor network outage etwork Outage % GF-Only: / user outage Outdoor Indoor % network outage KPI o Range Extenson % network outage KPI 8 umber of APs/km2 Outdoor Indoor Fgure 7. In-band deployment: Indoor / Outdoor network outage Secondly, the percentage of users offloaded to W-F and cell are studed. Fg 8 shows 24% to 53% moble users can be offloaded to ndoor W-F network when the AP denstes range from 8 APs/km2 to 39APs/km2. In the case of out-band, wth /km2, 5% moble users can be offloaded to cell, almost the same as W-F offloadng. However, snce there s only sngle-carrer at macro network (compared to 2-carrers macro n W-F case), 5% moble users offloaded to stll results n 24% network outage shown n Fg 6. Therefore hgher AP densty of APs/km2 s requred to offload more users (69%) so as to meet the network KPI, as shown n Fg 6. The n-band case s not shown, snce anyway t cannot meet the network KPI even at a very hgh node densty as shown n Fg.7.
Fgure 8. SIR Dstrbuton for n-band cell + network WF Splt [%] Splt [%] CDF.3 Carr: all; Ste: all; UECat: all; (avg= 7.8dB).2 Carr:26; UECat: all; (avg=.7db) Carr:8; UECat: all; (avg= 4.dB). Carr:26; n UECat: all; (avg=4.7db) 3GPP 2-D 3GPP 2-D 3 - -5 5 5 25 SIR (db) 6 GF-Only:WF Splt 8 2 39 umber of WF APs/km2 Fgure 9. W-F deployment: % offloaded users vs.ap Densty 8 7 6.9.8.7.6.5.4 The carrer at 2.6 GHz s nterfered by the APs SIR CDF GF-Only: Splt 5% users offloaded cells 52% users offloaded 8 umber of APs/km2 Range Extenson wth 3 db SIR bas case of out-band LTE cell deployment, even wth range extenson, much hgher AP densty of AP/km2 s requred to meet the KPI. The least favorable scenaro s the deployment of n-band cells, where cells share one carrer wth the macro layer. A hgh-densty and uncoordnated cell deployment creates a strong n-band nterference couplng to the Marco layer whch results n hgh overall network outage and even wth AP/km2 the target network KPI cannot be acheved. We have also shown that for the same access pont densty W-F APs and out-band cells can offload smlar amount of users. Fnally, the performance of n-band femto could potentally be mproved wth nterference mtgaton, whch s left for future study. REFERECES [] Csco VI Forecast, Csco Vsual etworkng Index: Global Moble data Traffc Forecast Update -5, February, 2. [2] Gartner Forecast, Moble Data Traffc and Revenue, Worldwde, - 5, July 4 [3] K. Lee, Moble Data Offloadng: How Much Can W-F Delver?, ACM Coext [4] J Gora, T.E Koldng,. Deployment Aspects of 3G cells, IEEE PIMRC 9, Tokyo [5] D Caln, H Claussen On femto deployment archtectures and macrocell offloadng benefts n jont macro-femto deployments, IEEE Communcaton Magzne, Volume: 48, Issue:, [6] 3GPP TR 36.84, Further Advancements for E-UTRA, Physcal Layer Aspects, verson 9.., December 9 [7] P. Mogensen, LTE Capacty compared to the ShannonBound, Proc. IEEE 65th VTC, May 7 [8] J. Ellng, Moble Broadband etwork Evoluton Towards 5 A Copenhagne Area Case Study, Telektronkk, January [9] J. Yee and H. Pezeshk-Esfahan, Understandng Wreless LA Performance Tradeoffs, http://www.commsdesgn.com, ov 2 [] G. Banch, Performance Analyss of the IEEE 82. Dstrbuted Coordnaton Functon, IEEE JSAC Communcatons, March [] A. Duda. Understandng the Performance of 82. etworks. In Proceedngs of PIMRC 8, Cannes, France, 5-8 September 8 Fgure. Out-band : Percentage of offloaded users vs. AP Densty VI. COCLUSIO We have studed the performance of ndoor W-F and cell deployment as the offloadng solutons to a LTE macro network n a realstc dense urban area under the assumpton of x growth of moble broadband traffc growth. To evaluate the offloadng potental, we have used a 3-D rado propagaton model based on Ray-tracng wth real buldng database and spatal traffc dstrbuton from lve 3G network. In the case of ndoor W-F deployment, we showed that an access pont (AP) densty of 2 AP/km2 can already meet the target network Key Performance Indcator (KPI) of 9% network coverage wth a mnmum data rate of Mbps. In the