Measurement-Based Evaluation of Link Budget Gain from 3GPP LTE Uplink Transmit Antenna Selection

Transcription

1 Measurement-Based Evaluaton of Lnk Budget Gan from 3GPP LTE Uplnk Transmt Antenna Selecton J. Salo ECE Ltd Teknkante 2 FI-25 Espoo, Fnland Emal: jar.salo@eceltd.com K. Haneda, V.-M. Kolmonen A.A.H. Arem, P. Vankanen Aalto Unverst School of Electrcal Engneerng Department of Rado Scence and Engneerng, SMARAD P.O. Bo 3, FI-76 Aalto, Fnland Emal: {katsuuk.haneda, velmatt.kolmonen, abdullah.arem, pertt.vankanen}@aalto.f Abstract Ths paper presents a measurement-based evaluaton of lnk budget gan from uplnk transmt antenna selecton n 3GPP LTE. In contrast to earler studes, the results are based on channel measurements and realstc non-omndrectonal moble termnal antennas. Further, n addton to the mpact of fast selecton based on small-scale fadng power, also the gan from slow selecton based on average sgnal power s evaluated. In the evaluated measurement cases, fast and slow fadng based transmt antenna selecton result n up to 3.7 and 3.2 db average power gan, respectvel, wth respect to the fed uplnk transmt antenna case. Ths mples that wth practcal nonomndrectonal termnal antennas a major porton of the gan can be realed based on slow selecton, hence justfng a smple open-loop transmt antenna selecton scheme whch s also applcable to FDD sstems. I. INTRODUCTION Cellular network performance studes and lnk budget calculatons are tpcall based on the assumptons of omndrectonal antenna pattern and unform angular power dstrbuton at the moble termnal end. From numerous measurement campagns t s however evdent that - whle useful for smplfng analses - nether of the assumptons hold n realstc envronments [], [2]. Frst, ntegrated handheld termnal antennas do not have omndrectonal antenna pattern (n an planar cut) when held near hand and/or head. Second, channel angle-ofarrval power spectum s rarel unform n realstc cellular outdoor propagaton scenaros, at least when eamned over a dstance of a few doens of meters or so. The dea of 3GPP LTE uplnk transmt antenna selecton s that the User Equpment (UE) uses the optmum (based on some crteron) transmt antenna for uplnk transmsson. The selecton of the UE transmt antenna can be made b the Base Staton (BTS) or the UE. In case of closed-loop selecton, antenna nformaton s sgnalled to the UE n the uplnk schedulng grant as specfed n [3]. In the open-loop transmt antenna selecton UE selects the transmt antenna If channel angular power spectrum was measured and averaged over, sa, man klometers, t s lkel that the same amount of power arrves from all amuth drectons wth equal probablt. autonomousl wthout feedback from BTS; the actual selecton method s not defned b 3GPP Release 8. The support of transmt antenna selecton s optonal for the UE and, n addton to the closed-loop and open-loop modes, the network ma decde to confgure a fed UE transmt antenna. The antenna selecton mode s set sem-statcall b hgher-laer sgnallng. From network plannng and performance evaluaton pont of vew, the queston arses of the mean and outage lnk budget gan from the uplnk transmt antenna selecton. Related work: The 3GPP Rel8 LTE uplnk transmt antenna selecton s a specal case of the general MIMO antenna selecton problem [4]. 3GPP research reports on uplnk transmt antenna selecton nclude [5], [6], [7], where some ntal performance studes were made. In a more complete contrbuton [8], the gan of open and closed loop transmt antenna selecton, ncludng frequenc-selectve schedulng and Hbrd Automatc Repeat Request (HARQ) retransmssons, was consdered. For the baselne case wthout HARQ and usng frequenc-unaware schedulng, the gan at % SNR outage was obtaned at about 3dB for closed-loop and 2.5dB for open-loop antenna selecton. These results were based on smulatons and on fast selecton of nstantaneous channel. Open-loop transmt antenna selecton utlng a non-3gpp ntra-subframe antenna swtchng scheme was studed n [9]. In [], t was suggested that downlnk cclc dela dverst mght mprove uplnk transmt selecton gan. The contrbuton closest to our paper was presented n [], where the mpact of antenna drectvt and uplnk tranmst antenna selecton were evaluated based on ndoor measurements for a mult-antenna setup n a non-3gpp settng. Interestngl, UE orentaton statstcs collected from actual smartphone users behavour was ncluded n the evaluaton. Medan gan of transmt antenna selecton was evaluated as 3 db, for the specfc antenna selecton scheme emploed n [], see theren for detals. Some references can also be found n patent lterature. Contrbuton of ths work: Whle uplnk transmt antenna selecton s well-known to practoners n the feld, evaluaton of the resultng lnk budget gan n practcal use cases has

2 receved less attenton. Ths paper presents a case stud evaluaton of the gan from UE transmt antenna selecton wth realstc drectonal termnal antennas and measured outdoor propagaton channel. Here, b realstc termnal antenna we mean that the antenna structure used n evaluaton has been specfcall desgned for a dual-antenna handheld devce. Compared to earler papers, the added value of ths paper nclude: ) evaluaton s based on outdoor channel measurements; ) non-omndrectonal termnal antennas are used, ncludng the mpact of hand grp; ) transmt antenna selecton gan based on averaged sgnal level s also evaluated (applcable also to FDD sstems). Organaton of the paper: In Secton II the problem s motvated b means of a measurement eample. In Secton III the channel measurements, antennas under test and measurement processng s eplaned. Results are gven n Secton IV and Secton V concludes the paper. Note on termnolog: Throughout the paper we assume that the nstantaneous path loss between an two pars of receve and transmt antennas s recprocal on the gven carrer frequenc. Thus, reversng the roles of transmtter and recever, the nstantaneous gan from transmt antenna selecton can be evaluated based on recever measurements rotaton # (deg) rotaton #4 (9deg) rotaton #6 (5deg) r Ant r Ant dstance, meters Fg.. Eample of average receved power seen for dfferent UE orentatons, measurement route. Base staton Moble route Moble route 2 Moble route 3 II. MOTIVATION: A MEASUREMENT EXAMPLE Fg. shows a measurement eample of the mpact of antenna orentaton on the receved average power of a dualantenna termnal (average s over small-scale fadng). The measurement and post-processng detals are gven n Sec. III. In the eample, the same measurement route of about 85 meters length s effectvel traversed three tmes, n each case the UE havng dfferent orentaton. Fg. 2 shows a map of the measurement ste together wth the coordnate sstem of the moble antennas. The plots n Fg. are from the route 2. The BTS antenna heght 2 s 3 m. It can be seen that n the mddle of the measurement route, UE enters lne-of-sght whch can be seen n Fg. as ncrease n receved sgnal power. Fg. 3(a) llustrates the structure of the consdered moble handset antenna together wth the hand model provded n the SEMCAD [2]. The three rotatons of the moble handset model n Fg. correspond to the -as of the antenna pontng, 9, and 5 of the coordnate sstem of the measurement envronment shown on the map. From Fg., dependng on the amuth rotaton angle, the average receved sgnal level of the antennas dffers b up to 8 db n non-lne-of-sght and up to db n lne-ofsght. It s noteworth that the average power s consstentl dfferent between UE branches over several meters dstance. Therefore, t s concevable that even (slow) average-sgnalbased selecton of transmt antenna would result n power gan at the BTS recever. 2 The average rooftop heght s about 25 meters. Fg. 2. Map of the MIMO channel soundng n one of the measurement stes n the Helsnk ct center. Element Element # #2 4 mm (a) mm m Fg. 3. (a) Two-element moble handset antenna structure and (b) the hand grp model consdered n our analss [2]. III. MEASUREMENTS AND POSTPROCESSING A. Used antennas The handset antenna model used n ths paper s shown n Fg. 3(a). Fg. 4 shows the amuthal radaton patterns of (b)

3 5 2 2 [db] (a) 33, #, # [db] (b) 33, #, # Fg. 4. Radaton patterns of two-element antennas embedded n the chasss of the moble termnal [2] (a) n free space and (b) wth the nfluence of the hand grp. The plots are on the - plane defned n Fg. 3(a). the two antenna elements mplemented n a moble handset chasss, descrbed n more detal n [2], one n a free space and another under the nfluence of the hand grp. The hand grp affects the radaton pattern notabl n reduced radaton effcenc and ncreased cross-polaraton levels. Mamum gan of the antenna was.5 and 5.8 db n the antenna under the absence and presence of the hand grp. Effcenc of the two antennas was 4. db n the free space, whle t decreased to 6.6 and 8.6 db wth the hand grp. It s mportant to note that the hand grp made the effcenc of the two antenna elements dfferent b 2 db, producng a statc power mbalance between the branches. B. Channel measurements Here onl a bref summar s gven, for further detals the reader s referred to [3]. Channel measurements were performed n outdoor mcrocellular envronment n Helsnk ct center at 2.54 GH. The base staton was equpped wth a two-element horontall polared lnear patch antenna arra and was located n a crane. The moble staton was wth a sphercal antenna arra consstng of 32-element dualpolared patch antennas. The moble was put on a trolle on.5 m heght above the ground to measure MIMO rado channel responses on 7 measurement routes stretchng over 59 m. Snce the channel response was acqured ever quarterwavelength, we have obtaned a total of about 65 MIMO channel responses. C. Embeddng the antenna patterns on the measured rado channel From the measured MIMO channel responses, multpath angle-of-arrval and polarmetrc comple ampltude was estmated at the moble end usng a spato-temporal beamformer. The beamformer compensates for the characterstcs of the sphercal arra b the knowledge of comple mult-polared antenna patterns. Therefore t s possble to embed our moble handset antenna characterstcs nto the measured MIMO channels b combnng the radaton patterns of the antennas wth the estmated multpaths [3]. The resultant MIMO channel responses shows the effect of the consdered moble antennas n the measurement envronment. The moble antenna was rotated n 3 steps on the - plane n Fg. 3(a) to smulate random orentaton of UE, leadng to 2 realatons of MIMO channels at sngle moble locaton. The effectve channel matr, after weghtng the (appromatel) sotropcall measured channel response wth the rotated antenna patterns, s denoted wth [ ] h, h H eff, = 2,, () h 2, h 22, where s the tme nde. We remark that the elements of H eff, are narrow-band channel mpulse responses. D. Calculaton of transmt antenna selecton gan The receved power of antenna branch k (wthout precodng) at tme nstant s P k = h k, 2 + h k2, 2. (2) As the channel s assumed recprocal on the gven frequenc, ths s also the receved power at the base staton after mamum rato dverst combnng when UE antenna k s transmttng (assumng the same transmt power). The nstantaneous transmt antenna selecton gan for the th receve tme sample s G = P k opt P ref, (3) where P ref depends on the reference antenna used to calculate the gan. In ths paper we evaluate the gan for P ref = P and P ref = P 2. The optmal transmt antenna k opt for the th measurement sample s formall defned as k opt = arg ma F k [{P k }]. (4) k It should be noted that the optmal antenna can be the same as the reference antenna n whch case the gan s db for that measurement sample. The generc antenna selecton cost functon s denoted as F k [ ] and t s some functon of the power samples {P k }. In the sequel, transmt antenna selecton gan s measured for two specal cases of the antenna selecton cost functon F k [{P k}]: antenna selecton based on nstantaneous sgnal level (fast fadng) and selecton based on averaged sgnal level (slow fadng). ) Selecton based on fast fadng: In ths case the antenna selecton cost functon s based on the nstantaneous sgnal: F k [{P k}] = P k. In terms of mplementaton ths mples that the UE has nformaton of the nstantaneous uplnk receved sgnal ether b means of assumed channel recproct (openloop transmt antenna selecton) or transmt antenna selecton commands transmtted from BTS to UE (closed-loop transmt antenna selecton).

4 2) Selecton based on slow fadng: In ths case the antenna selecton cost functon s based on the movng average of receved sgnal power: F k [{P k }] = L L+ where L s the length of the averagng wndow. In terms of mplementaton ths mples that the UE averages the downlnk sgnal and uses open-loop transmt antenna selecton to select more favourable transmt antenna. Assumng that uplnkdownlnk path loss s recprocal wthn a small frequenc scalng constant defned b the duple gap, ths scheme s sutable also for FDD sstems. The slow fadng correlaton has been shown to be n the range between.9 GH,.8 GH and 2. GH bands n [4], hence the slow fadng correlaton for duple dstance wthn one band s epected to be hgher 3. The drawback of the slow selecton s the unavodable averagng dela whch occasonall results n negatve gan. 3) Average branch power mbalance: To gan nsght, mbalance n average receved power of UE antenna branches s used as an llustratve metrc. When reference antenna k = s used t s defned as L+ IB = P L+ P, 2 and the nverse of ths when reference antenna k = 2. IV. RESULTS In ths secton we present measurement results of transmt antenna selecton gan. The results are stated for two antenna confguratons: UE n free space UE held n the poston shown n Fg. 3(b) Averagng length L = (25 wavelengths, or 3.5 meters@2. GH) s used n all cases wth slow fadng - based transmt antenna selecton. Tunng L mpacts results slghtl but s left for further stud. All presented emprcal dstrbuton functons of the nstantaneous gan G are based on aggregaton of all seven measurement routes (wth 2 amuth orentatons each) nto a one long measurement route. In the case of slow selecton, the frst L = measurement samples of each ndvdual route are dscarded from gan calculatons to avod edge effects. In the legends, FF gan ndcates the average value of G when fast selecton s used, and SF gan ndcates the average value of G when slow selecton s used. G has been defned n (3). 3 Ths s also underlng assumpton n 3GPP LTE uplnk open-loop power control whch s based on the assumpton of recprocal uplnk-downlnk path loss [3]. P k A. UE n free space Fg. 5 shows the result for UE n free space, antenna pattern shown n Fg. 4(a). The cumulatve dstrbuton functon (cdf) of the nstantaneous gan G s shown for both antennas used as reference. Addtonall, power mbalance cdf s also shown. Average transmt antenna selecton gan s 2.6 db and 2.4 db for the two references when transmt antenna selecton s based on the fast fadng sgnal. For the slow fadng based selecton the values are.8 db and.7 db, respectvel. The result llustrates that when the UE n free space s rotated over the entre amuth angle, the selecton of reference antenna does not make a great dfference, snce the branch antenna patterns are rotated replcas of each other (see Sec. III-A). For the (deal dela-less) selecton based on nstantaneous sgnal the gan cdf obvousl shows no negatve gans. For about 5% of samples there s no gan snce the optmal transmt antenna s the reference antenna tself; ths eplans the step functon n the cdf. It s noteworth that for 2% of samples the gan eceeds 5 db. Ths s snce for some UE rotatons the reference antenna s strongl non-optmal compared to the stronger branch a result of the drectve channel and non-omndrectonal antenna patterns. For selecton based on the averaged sgnal the gan cdf shows some negatve gans for about % of samples. Ths s due to: ) when the average antenna branch powers are roughl equal small-scale fadng results n man negatve gan samples due to non-optmum antenna selecton; ) averagng dela n transmt antenna selecton and hence t s not possble to track the average sgnal power when the phscal channel undergoes a quck transton. On the other hand, the gan eceeds 5 db for 8% of samples, whch s almost the same as gan from fast transmt antenna selecton. From Fg. 5, slow selecton results n loss of.7.8 db n average gan wth respect to fast selecton, manl because of the tal showng negatve gan values durng fast channel changes. From the two-sded power mbalance cdf t can be seen that for about 25% of samples the average power mbalance eceeds 5 db llustratng the gan potental n slow antenna selecton. Up to db branch power mbalance has also been reported n [5]. B. UE held n user s hand In ths case the UE s held b the user as shown n Fg. 3(b). Some observatons: Antenna suffers from the 2dB effcenc loss caused b the user hand. As the user hand poston s dffcult to predct n practce, ths demonstrates the addtonal robustness effect brought b the transmt antenna selecton. The average gan from fast selecton s 3.7 db when antenna s used as reference. The average gan from slow selecton s 3.2 db when antenna s used as reference. The gan s 7dB or more for 2% of samples when antenna s used as reference.

5 gan over ref antenna FF gan 2.6dB SF gan.8 db gan over ref antenna 2 FF gan 2.4dB SF gan.7 db fast fad slow fad mbalance Fg. 5. of nstantaneous gan G for UE n free space. FF gan = average gan for selecton based on fast fadng sgnal, SF gan = average gan for selecton based on averaged sgnal gan over ref antenna FF gan 3.7dB SF gan 3.2 db gan over ref antenna 2 FF gan.7db SF gan.2 db fast fad slow fad mbalance Fg. 6. of nstantaneous gan G for UE held n user s hand. FF gan = average gan for selecton based on fast fadng sgnal, SF gan = average gan for selecton based on averaged sgnal. C. Dscusson The most nterestng observaton of the presented results s that, for handheld UEs, a large porton of the gan of transmt antenna selecton can be harvested b usng the slow selecton crteron; ths ndcates that the usage of average-sgnal based open-loop selecton can mprove end user throughput, reduce uplnk nterference as well as ncrease UE batter lfe. For 2% of samples there s no dfference between slow and fast selecton, a result of the large power mbalance n UE antenna branches n realstc operatng envronments. The evaluaton n ths paper s based on pont-frequenc channel mpulse responses and hence most applcable to evaluate the antenna selecton gan for narrow-band frequencunaware scheduled uplnk transmsson. In the case where the channel coherence bandwdth s larger than the scheduled bandwdth, frequenc-aware uplnk channel schedulng can result n addtonal gan when average branch power mbalance at UE s large. However, assumng that the power mbalance s small, fast antenna selecton actuall consumes part of the frequenc-aware schedulng gan [8]. An mportant mplementaton ssue not dscussed so far s that of the power loss ncurred b the RF swtch requred to mplement the transmt antenna selecton wth mnmal duplcaton of RF crcutr. Tpcal assumpton s that the RF swtch results n.5 db mplementaton loss [8]. V. CONCLUSION We have presented an evaluaton of the gan from uplnk transmt antenna selecton b usng outdoor channel measurements and realstc handheld termnal antenna model. The average gan from fast and slow selecton was 3.7 and 3.2 db, respectvel, when the termnal s held n hand grp poston. For 2% of measurement samples the gan eceeded 7 db for both antenna selecton modes. The advantage of slow selecton s that t can be used n open-loop mode and s also usable n FDD sstems wthout uplnk soundng reference sgnal. REFERENCES [] R. Verdone, A. Zanella (Eds.), Pervasve Moble and Ambent Wreless Communcatons: COST Acton 2, Sprnger, 22 [2] L. M. Correa (Ed.), Moble Broadband Multmeda Networks Technques, Models and Tools for 4G, Elsever, 26 [3] 3GPP TS 36.23, Evolved Unversal Terrestral Rado Access (E-UTRA); Phscal laer procedures, [4] A.F. Molsch, M. Wn, MIMO sstems wth Antenna Selecton, IEEE Mcrowave Magane, Vol. 5, No., pp.46-56, March 24 [5] 3GPP R-5398, Transmt Antenna Selecton Technques for Uplnk E- UTRA, Insttute for Infocomm Research (I2R), Mtsubsh Electrc, NTT DoCoMo [6] 3GPP R-6829, Random Access Transmt Dverst, ETRI [7] 3GPP R-6334, Closed Loop-Based Antenna Swtchng Transmt Dverst, NTT DoCoMo, Insttute for Infocomm Research, Mtsubsh Electrc, NEC, Panasonc, Sharp [8] H. Kawa, H. Taoka, K. Hguch, and M. Sawahash, Performance Evaluaton of Closed-Loop Based Antenna Swtchng Transmt Dverst Assocated wth Frequenc Doman Channel-Dependent Schedulng n E-UTRA Uplnk, IEEE PIMRC, 27 [9] J. Lu, L. Wang, M. You and P. Wen, A Specal Open-Loop Transmt Antenna Selecton Mode n the SC-FDMA Sstems for 3G LTE, IEEE MILCOM, 27 [] L.M.A. Jalloul, N. Cnk, B. M. Hochwald, A. Paulraj, Wh Downlnk Cclc Dela Dverst Helps Uplnk Transmt Dverst, IEEE VTC Sprng 29, Barcelona [] A. A. San, L. Zhong, and A. Sabharwal, Drectonal Antenna Dverst for Moble Devces: Characteratons and Solutons, ACM MobCom, 2 [2] A.A.H. Arem, K. Haneda, and P. Vankanen, Ste-specfc evaluaton of a MIMO channel capact for multantenna moble termnals promt to a human hand, 5th European Conference on Antennas and Propagaton (EuCAP 2), pp , Rome, Ital, Aprl 2. [3] P. Suvkunnas, J. Vllanen, K. Sulonen, C. Icheln, J. Ollkanen, and P. Vankanen, Evaluaton of the Performance of Multantenna Termnals usng a New Approach, IEEE Trans. Instrum. Meas. Vol. 55, No. 5, pp , Oct. 26. [4] B. Van Laethem, F. Qutn, F. Bellens, C. Oestges, Ph. De Doncker, Correlaton for Mult-frequenc Propagaton n Urban Envronments, Progress n Electromagnetcs Research Letters, vol. 29, pp. 5 56, 22 [5] W. A. T. Kotterman, G. F. Pedersen, K. Olesen, Dverst Propertes of Multantenna Small Handheld Termnals, EURASIP Journal on Appled Sgnal Processng, vol. 24, pp , Jan. 24