Antenna and Propagaton Parameters Modelng Lve Networks Arne Smonsson, Martn Johansson and Magnus Lundevall Ercsson Research, Sweden [arne.smonsson, martn.n.johansson, magnus.lundevall]@ercsson.com Abstract In rado network performance evaluatons t s of nterest to use smulaton envronments that represent realstc networks. In ths paper two wdely used smulaton envronments are benchmarked aganst lve network path gan measurements. Some parameter adjustments are done to ft the smulaton envronments to the measured path gan and cell solaton statstcs. The need to modfy base staton antenna model parameters to match the ntra-ste cell solaton n real networks s dentfed and elaborated. In partcular an ncrease of maxmum attenuaton to 35dB, ntroducton of mechancal tlt and reducton of vertcal beam wdth are needed. In addton some adjustments are done to better represent the assumed traffc dstrbuton wth % ndoor users and expected propagaton losses resultng n a path gan level on par wth the measured network. Keywords- rado network; measurements; antenna model I. INTRODUCTION The statstcal characterstcs of rado network smulaton envronments have a decsve mpact on the performance results. Ths s true for performance assessments of new technologes as well as for feature evaluatons. For path loss dstrbuton the used propagaton and antenna models are mportant. But the actual spatal dstrbuton of users also has an mpact. If cells are planned to capture the majorty of traffc n the cell center the path loss s statstcally mproved. Also the fracton of ndoor traffc together wth buldng types has mpact. Furthermore, at hgh load the cell solaton defnes the capacty n nterference lmted networks such as WCDMA and LTE wth frequency reuse one. Also the ntra-ste and nter-ste cell solaton n smulaton envronments can be of mportance for feature evaluatons. Coordnaton technques between cells are n many cases much easer to ntroduce wthn a ste than between stes. Between stes X2 communcaton s used n LTE whch may requre standardzaton efforts and n some cases also ncreased backhaul capacty. Examples of features dependent on the ntra-ste versus nter-ste cell solaton are Inter-Cell Interference Coordnaton (ICIC) and Co-ordnated Mult Pont (CoMP) [7]. The 3GPP smulaton envronments [1] have been used for LTE evaluatons, both regardng feature evaluatons and general performance assessments. The ITU smulaton envronments [2] have been used n the IMT Advanced evaluaton process. In ths paper the envronment and the ITU Urban Macro (UMa) envronment have been compared to path gan and geometry factor dstrbutons from a commercal urban WCDMA network. Measurement reports have been logged from all mobles collectng 2 mllon samples. Ths captures the true spatal dstrbuton of traffc ncludng ndoor usage. Also two new envronments based on the 3GPP and the ITU envronments have been created by adjustng some model parameters to ft the measured network better. These new envronments models other measured networks better as well, such as the one studed n [5]. Ths paper s organzed as follows: n Secton II the used geometry factor measures are defned and the ntra-ste cell solaton measure ntroduced. The lve traffc measurement method s descrbed n secton III. Some key characterstcs of the measured network and the compared smulaton envronments are descrbed n sectons IV and V, respectvely. The results are presented n secton VI and fnally conclusons are drawn n secton VII. II. GEOMETRY FACTOR AND INTRA SITE CELL ISOLATION Cell solaton s studed through the geometry factor measure as used and descrbed n [5]. Ths measure of cell solaton s sutable for comparson between measurements of real lve traffc and data for smulaton envronments. In ths paper the per neghbor geometry factor s further separated nto nter-ste and ntra-ste geometry as llustrated n Fg. 1. As wll be shown, ntra-ste cell solaton ncludng only the co-sted neghbor cells s useful when the base staton antenna model s nvestgated. Inter-ste Fgure 1. Intra-ste cell solaton. Intra-ste 978-1-4244-8327-3/11/$26. 211 IEEE
The geometry factor for a measured sample s based on the path gan from the best cell g best_cell (n db) relatve to the path gan from neghbor cells g. A per neghbor geometry factor s defned as G = log g best _ g cell [db] (1) wth a lower bound of db. The ntra-ste geometry factor s acheved by selectng neghbor cells as the co-sted neghbor cells. Wth three-sector stes, as n these measurements, t s possble to defne one ntra-ste geometry factors to the strongest neghbor and one to the second strongest neghbor. The best cell s selected as the one havng strongest path gan per sample. Also a total geometry factor s studed usng the sum of the path gans for all neghbors G = log gbest best _ _ cell [db]. (2) g cell III. LIVE TRAFFIC MEASUREMENTS The Geo-Observablty tool [6] has been used to collect moble measurements from commercal traffc. All connected mobles are confgured to send a measurement report every 4 th second to the network. Ths captures the true spatal traffc dstrbuton ncludng ndoor usage. Every report ncludes several measures of whch the Receved Sgnal Code Power (RSCP) and scramblng code are used n ths study. RSCP for connected and up to neghbor cells are reported enablng both path loss and geometry factor analyss. For each measurement report the path gan per reported neghbor s calculated as g = RSCP P [db] (3) CPICH where P CPICH s the CPICH (Common PIlot CHannel) power. For each measurement report the path gan values are sorted and the co-sted cells are dentfed by the reported correspondng scramblng code. Also the strongest cell s used nstead of the connected cell when calculatng geometry factors n order to elmnate handover algorthm mpact. The path gan values g are then nserted nto (1) and (2) for {1,2} and {1,,} respectvely. The moble does not always detect neghbors or the two co-sted neghbors. Ths case s treated n two ways. : RSCP s set to nfnty resultng n an nfnte correspondng per neghbor geometry factor G and also an nfnte total geometry factor G f no neghbor s detected. detecton level: RSCP s set to an assumed detecton level n the moble termnal of -124dBm These two alternatves gve an ndcaton of a measurement nterval n whch undetected neghbors may exst. The neghbor detecton performance of the mobles s unknown and vares wth moble vendor and models. Also ths fxed absolute detecton level s a smplfcaton and the detecton level most probably depends on RSCP from the connected cell and the dynamc range of the recever. IV. MEASURED NETWORK The measurements are collected n a commercal WCDMA network n an urban cty envronment. The network s mature and well planned. Moble broadband s deployed wth HSPA. All connected mobles from 66 three-sector stes s logged for one and a half hour. There s a number of dfferent base staton antennas used. Most detals are unknown, however both 6 and 9 horzontal half power beamwdth (HPBW) antennas exst. A summary of some known and approxmated key parameter values are lsted n Table I. TABLE I. MEASURED NETWORK OVERVIEW Parameter Value Frequency band 2.1GHz Access WCDMA wth HSPA Number of stes/cells 66/198 Sectors per ste 3 Inter ste dstance -12m Horzontal HPBW 6 and 9 Average antenna heght Around m Average total tlt (mechancal + electrcal) Around 4 Plot power P CPICH Around 33dBm (2W) Recordng tme Around 1.5 hour Number of logged samples 2 448 V. SIMULATION ENVIRONMENT Two smulaton envronments are used: the [1] and (Urban Macro) [2]. These envronments are both used as s as well as modfed to ft measurement data. The frequency band s set to 2.1GHz to match the measured network. Key parameters for the baselne reference models are lsted n Table II. 5 mobles are randomly dropped n the system area. For each moble, the path gan g s calculated accordng to the propagaton, shadow fadng and antenna models for all cells {1,,57}. The path gan values are nserted nto (1) and (2) to calculate geometry factors. In lne wth the measurements no handover hysteress s appled, g best_cell s selected as the cell wth hghest path gan. TABLE II. SIMULATOR MODELS Parameter Frequency band 2.1GHz Cell layout Hexagonal grd 57 cells, 19 3-sector stes Wrap around Inter ste dstance m Measurement samples 5 NLOS propagaton 15.7+37.6 log(d) 2.+39.1 log(d) LOS propagaton N.A. 34.4+22 log(d) NLOS shadow fadng std. 8dB 6dB LOS shadow fadng std. N.A. 4dB Indoor shadow fadng std. 8dB 7dB d n meters, LOS: Lne Of Ste, NLOS: Non LOS
VI. RESULTS In ths secton the smulaton envronments are compared to the measurements. C.D.F:s of path gan g and geometry factors G are used to show level and dstrbutons as well as mpact on dfferent percentles. Frst the parameter dfference, the path gan and the geometry factors for all four smulaton envronments are compared. Then the mpact of ndvdual adjusted smulaton parameters s elaborated on llustrated by ntra-ste geometry factor dstrbutons. Ths s done for one of the smulaton envronments, ftted. A. Adjusted Model Summary Based on and smulaton envronments, two adjusted models are created to better ft the measured statstcal dstrbutons. The parameter adjustments are summarzed n Table III. The changes are lmted to two areas, ndoor propagaton and base staton antenna model. For the adjusted parameters a common parameter set s n general used for both envronments. The excepton s the electrcal antenna tlt whch dffers because of a dfference n antenna heght (lsted for clarty but unchanged). TABLE III. ADJUSTED PARAMETERS 3GPP case 1 ITU UMa 3GPP ftted ITU ftted Propagaton Indoor fracton % % % Outer wall loss 2dB 2dB 2dB Outer wall dstance N.A. [,25]m [,25]m In buldng loss N.A..5dB/m.5dB/m Effcency loss N.A. N.A. 14dB Base staton antenna Antenna heght 32m 25m 32m 25m Mechancal tlt 5 Electrcal tlt 15 5 2 Horzontal BW 75 Vertcal BW 15 6.5 Front to back rato 25dB 2dB 25dB Sde lobe level -2dB -2dB -25dB Max. attenuaton 25dB 2dB 35dB In there are no ndoor mobles, the lsted ndoor parameters are from ITU UM n [2]. The base staton antenna parameters are n lne wth measured antenna radaton pattern n [3] as well as wth comparsons to other real networks n [4] and [5]. The n buldng loss and effcency loss are also defned for LTE CoMP evaluatons n [7]. B. Path Gan Comparson Fg. 2 shows the path gan dstrbuton for all four smulaton envronments compared to the measurement. The envronment that lacks ndoor mobles has much hgher path gan levels compared to measurements. An expected % ndoor fracton s ntroduced to compensate for ths. As n [7], the outdoor to ndoor model from the ITU Urban Mcro (UM) envronment s ntroduced featurng 2dB outer wall penetraton loss and an n buldng loss model of.5db/m wth random unform dstrbuton between and 25 m from outer wall. Also an addtonal effcency loss s ntroduced as n [7] to match the measured path gan. Ths loss may consst of recever effcency, feeder loss and other losses. The average nter-ste-dstance n the measured network s not known but may also be larger than the smulated m. Wth ths loss the path gan level s on par wth the measured network. Even though the 3GPP envronment contans ndoor loss the same n buldng loss and effcency loss s requred to get a path gan level smlar to the measurements. Note that the effcency loss also s appled on outdoor mobles. C. Geometry Factor Comparson Fg. 3 shows the dstrbuton of the total geometry factor accordng to (2). All four envronments are compared to the measurement results. For the total geometry the measurement s rather accurate as ndcated by the relatvely small dfference compared to the detecton level. The measured lne stops at 88% because 12% of the samples dd not report any neghbor cell measurement. When comparng the smulaton models to the measurements n Fg. 3 t s clear that the default tlt used n results n a sgnfcantly hgher cell solaton. The s lmted to maxmum geometry factor of 17dB. The 9 9 6 6 2 ftted ftted -1-1 -1-12 -1 - -9 - Path Gan [db] 2 ftted ftted -5 5 15 2 25 Geometry /sum(g other ) [db] Fgure 2. Comparson of path gan. Fgure 3. Comparson of total geometry factor.
9 9 2 ο electrcal + 5 ο mechancal tlt 7 ο electrcal tlt 6 6 2 ftted ftted 5 15 2 25 35 2 5 15 2 25 35 Fgure 4. Comparson of ntra-ste geometry factor. ftted envronments do no have such clear lmtaton and they show a dstrbuton more n lne wth the measurements. In the measurement results n [5] there s also a smlar fracton of geometry factor above 2dB. D. Intra-Ste Cell Isolaton and Maxmum Attenuaton Fg. 4 shows the ntra-ste geometry factor dstrbuton accordng to (1) for measurement and smulatons. For each case there are two lnes representng the geometry factor towards the stronger co-sted cell and towards the weaker costed cell, respectvely. In the measurements, the accuracy of per cell geometry factor s somewhat worse than for the total geometry factor. 54% of the reports nclude one co-sted neghbor and only 15% both. It can be noted that above 22dB there s no neghbor reported and the mobles probably do not detect neghbors 22dB weaker than the connected or strongest cell. When comparng the smulaton results wth the measurements n Fg. 4 t s seen that both ftted models agree better wth the measurements above a geometry factor of db. The lmtaton on maxmum geometry seen n Fgs. 3 and 4 orgnates from the ntra-ste cell solaton. It s 3dB lower n 9 6 2 Vertcal HPBW 6.5 ο Vertcal HPBW 15 ο 5 15 2 25 35 Fgure 6. Vertcal beam wdth mpact, ftted envronment. Fgure 5. Mechancal tlt mpact, ftted envronment. Fg. 3 because of the contrbuton from both co-sted cells. Ths s caused by the base staton antenna parameter maxmum attenuaton whch s 2dB for and 25 db for 3GPP case 1, see Table III. It can be noted that the ndoor fracton and propagaton model adjustments do not mpact the geometry factors as these losses affect all lnks equally and thus cancel n the expresson for the geometry factor. Thus the path gan propagaton fttng can be done ndependent of the ntra-ste cell solaton antenna parameter fttng. E. Impact by Mechancal Tlt Mechancal tlt s commonly used n rado network deployments, often n combnaton wth electrcal tlt, wth mechancal tlt set to a fxed preset value and electrcal tlt used to dynamcally adjust the total tlt. The mpact of mechancal tlt n modelng s llustrated n Fg. 5. The ntra-ste cell solaton for ftted envronment s shown wth two alternatve tlt settngs resultng n a total tlt of 7. When only electrcal tlt s used the co-sted cell solaton s lmted by the front to back rato (FBR). For electrcal tlt, the vertcal back lobe s modeled as pontng wth the same down-tlt angle as the front lobe n the drecton of the center of the co-sted cells. F. Impact by Vertcal Beam Wdth A wde vertcal beam wdth also reduces the achevable ntra-ste cell solaton. In Fg. 6 the ftted ITU model s shown wth two alternatve vertcal HPBW; the orgnal ITU vertcal HPBW of 15 as well as the ftted vertcal HPBW of 6.5. Even though the vertcal beam wdth s not assocated wth a hard bound on the geometry, n contrast to maxmum attenuaton and front to back rato, the beam wdth clearly reduces the hghest percentle ntra-ste solaton by around 5dB. Wth a wde vertcal beam the nterference caused to costed cells through the back lobe s hgher. Wth 5 mechancal tlt, 15 HPBW stll transmts energy n the (mechancally uptlted) back lobe. Also n [3] a narrow vertcal beam s found to model measured antenna radaton pattern well. Furthermore, n [8]
9 Horzontal HPBW 75 ο Horzontal HPBW ο 9 Sde Lobe Level -25dB Sde Lobe Level -2dB 6 6 2 2 5 15 2 25 35 5 15 2 25 35 Fgure 7. Horzonal HPBW mpact, ftted envronment. the elevaton spreads s estmated to be n the order of one degree or less ndcatng no need for ncreased vertcal beam wdth to model reflecton paths. A wde vertcal beam also s found to result n a rather unrealstcally large tlt, such as the 15 n [1]. G. Impact by Horzontal Beam Wdth In the measured network the horzontal HPBW was a mxture of 6 and 9. It was found that a HPBW of 75 modeled ths network well. The mpact of the ncreased HPBW s shown n Fg. 7 for the ftted ITU model. The ntra-ste solaton to the strongest co-sted cell s reduced by 1 to 2dB. Ths adjustment results n better ft to the measurements n relaton to the strongest co-sted cell. H. Impact by Sde Lobe Level An ncreased vertcal sde lobe level attenuaton degrades the coverage at the sector edge somewhat. Ths also reduces the ntra-ste cell solaton but n contrast to the horzontal beam wdth manly at lower percentles, see Fg. 8 where the ftted ITU model s shown wth two dfferent sde lobe levels. Ths adjustment gves a better ft to the measurements but s dffcult to logcally motvate from an antenna modelng vew, at least when comparng to the peak of the frst sde lobe below the man beam n commercal sector antennas, whch s typcally hgher than 2 db. One possble explanaton for ths can be that when deployed on roof tops the sector antennas are sometmes separated and placed at the edges of the roof resultng n hgher solaton than can be motvated for antennas mounted on a mast. In the smulatons such a dstrbuted antenna placement s not modeled. VII. CONCLUSION Ths paper compares the characterstcs of two wdely used smulaton envronments, namely and ITU UrbanMacro, aganst lve network data. The results reveal devatons from the measured network n two aspects; the path gan level and the ntra-ste cell solaton. The latter also results Fgure 8. Sde lobe level mpact, ftted envronment. n a less representatve modelng of the nterference stuaton, the geometry factor, n the measured network. The commonly used antenna model parameters gve less realstc ntra-ste cell solaton and t s motvated to consder adjustments for future evaluaton campagns. Specfcally the maxmum attenuaton should be ncreased to around 35dB and mechancal tlt modelng should be ntroduced. Ths elmnates the unrealstc upper lmt of solaton between co-sted cells and results n a more reasonable hgher percentle of ntra-ste cell solaton. Also a narrower vertcal beam wdth s strongly recommended reducng the back lobe nterference to co-sted cells to a more realstc level and resultng n a more reasonable tlt behavor. To get a path gan level n lne wth that of the measured network, ncreased loss modelng must be ntroduced. An expected % fracton of ndoor mobles, ITU n-buldng loss, and ntroducng effcency loss results n a better algnment wth the measurements. The effcency loss s a non-specfed sum of feeder loss, recever effcency, and nter-ste dstance msmatch. REFERENCES [1] 3GPP, "Further advancements for E-UTRA physcal layer aspects", TR 36.814, V9... [2] ITU, "Gudelnes for evaluaton of rado nterface technologes for IMT- Advanced", Rep. ITU-R M.2135 (29). [3] L. Thele et al., "Modelng of 3D Feld Patterns of Downtlted Antennas and Ther Impact on Cellular Systems", ITG Workshop on Smart Antennas 29. [4] F. Gunnarsson et al., "Downtlted Base Staton Antennas - Model Proposal and Impact on HSPA and LTE Performance", VTC'8-fall. [5] A. Smonsson et al., "LTE Downlnk Inter-Cell Interference Assessment n an Exstng GSM Metropoltan Deployment", VTC'-fall. [6] Ercsson, RNO feature GEO-W, www.ercsson.com. [7] 3GPP, "Coordnated Mult-Pont Operaton for LTE physcal layer aspects", TR 36.819, V1... [8] H. Asplund, A. A. Glazunov, A. F. Molsch, K. I. Pedersen and M. Stenbaueret, The COST 259 Drectonal Channel Model Part II: Macrocells n IEEE Trans. Wreless Communcatons, vol. 5, pp. 3434 34, Dec. 26.