Practical Introduction to LTE Radio PlanningJ. Salo, M. Nur-Alam, K. Chang This paper reviews basics of radio planning for 3GPP LTE. Both coverage-limited and interference-limitedscenarios are considered. For the coverage-limited scenario LTE link budget is compared to that of 3GPP Release 8HSPA+ with 2x2 MIMO. It is shown that, given the same 5MHz bandwidth, both systems have similar cell rangesbut, for a given target bit rate, there exists an optimum LTE system bandwidth that maximizes cell range in bothuplink and downlink. For the interference-limited scenario (with random uncoordinated interference) we illustratethe relationship between average network load, cell edge target throughput and cell range, as well as the notionof interference margin for cell range dimensioning. Impact of base station antenna conﬁgurations on dual-streamMultiple-Input Multiple-Output (MIMO) performance is demonstrated by means of a real-world measurementexample. The impact of advanced LTE radio resource management features are brieﬂy reviewed. Finally, the mostimportant radio parameter planning tasks are introduced.1. INTRODUCTION Abbreviations This paper introduces LTE from the perspec- BCCH Broadcast Channeltive of radio network planning. The paper is CQI Channel Quality Indicatormainly targeted for readers with earlier experi- FDD Frequency Division Multiplexingence in radio planning and mobile communica- HARQ Hybrid Automatic Repeat Requesttions. Some prior knowledge of radio engineer- HS-DSCH HSDPA Downlink Shared Channeling and LTE is assumed as principles of OFDMA LTE Long Term Evolutionand SC-FDMA, as described in 3GPP LTE spec- PCI Physical Cell Identityiﬁcations, will not be reviewed in this paper. In- PDCCH Physical Downlink Control Channelstead the reader is referred to well-known liter- PDSCH Physical Downlink Shared Channelature references [1–3]. The most important LTE PMI Precoder Matrix Indicatorradio interface parameters are summarized in PRACH Physical Random Access ChannelTable 1 for the convenience of the reader. Our PRB Physical Resource Blockfocus is on the FDD variant of LTE, although PUSCH Physical Uplink Shared Channelmost of the discussion is also applicable to TDD. RS Reference Signal SINR Signal-to-Interference-Noise-Ratio TDD Time Division Multiplexing TTI Transmission Time Interval 1
2 J. Salo et alTable 1Summary of main LTE radio interface parameters Quantity LTE DL LTE UL System bandwidth 1.4, 3, 5, 10, 15, 20MHz 1.4, 3, 5, 10, 15, 20MHz Multiple access OFDMA single carrier FDMA Cyclic preﬁx 4.7 microsec / 16.7 microsec 4.7 microsec / 16.7 microsec Modulation QPSK, 16QAM, 64 QAM QPSK, 16QAM (64 QAM for Cat5 UE) Channel coding⋆ turbo coding turbo coding HARQ 8 processes, up to 7 retransm./proc 8 processes Power control none♮ cell-speciﬁc and UE-speciﬁc Handover hard, network-triggered hard, network-triggered Num of Tx antennas 1, 2, 4 ‡ 1 ♯ Num of Rx antennas arbitrary ≥2 Transmit diversity Space-Frequency Block Code transmit antenna selection diversity Beamforming 3GPP codebook or proprietary none Spatial multiplexing open-loop, closed-loop none† PRB allocation per TTI distributed or contiguous contiguous♭⋆ For PDSCH and PUSCH. For L1/L2 control and common channels other coding schemes can be used.♮ Downlink Only static Reference Signal power boosting has been speciﬁed by 3GPP, vendor-speciﬁc implementa-tions possible.‡ With UE-speciﬁc reference signals (vendor-speciﬁc beamforming) arbitrary number of Tx antennas.♯ For simultaneous transmission. Transmit antenna selection is possible with RF switching.† Multiuser MIMO possible in UL in 3GPP Release 8.♭ Intra-TTI and inter-TTI frequency hopping is possible in UL.2. LTE COVERAGE IN NOISE-LIMITED terference power can be further decomposed as SCENARIO I = Iown + Iother , (2)2.1. Deﬁnition of average SINR Probably the most useful performance metric where Iown and Iother are the average own-cellfor LTE radio planner is the average signal-to- and other-cell interference power. In case ofinterference-and-noise ratio, SINRave , deﬁned as HSPA, Iown = (1 − α) Pown with α ∈ [0, 1] denot- ing the average channel multipath orthogonal- S ity factor and Pown denoting the own-cell trans-SINRave = , (1) mit power. In LTE, orthogonality is often as- I+N sumed unity for simplicity, even though in re-where S is the average received signal power, I ality the following nonidealities may result inis the average interference power, and N is the non-negligible own-signal interference in LTE:noise power. In measurement and simulation • Inter-symbol interference due to multipathanalysis, the sample averages should be taken power exceeding cyclic preﬁx lengthover small-scale fading of S and I and over alarge number of HARQ retransmissions (several • Inter-carrier interference due to Dopplerhundreds of TTIs, preferably). The average in- spread (large UE speed)
Practical Introduction to LTE Radio Planning 3 • Transmit signal waveform distortion due • The noise bandwidth is about 5MHz for to transmitter nonlinearities, measured in HSPA+ UL and DL, as well as LTE DL. terms of Error Vector Magnitude For LTE UL, however, the noise bandwidth depends on the number of allocated phys- In both LTE and HSPA the effective value of ical resource blocks (frequency allocation),α depends on the multipath characteristics and since symbols are detected in time-domainreceiver implementation. For HSDPA with re- and noise is accumulated only over the ac-ceiver equalizer, α > 0.9 can be assumed in most tual occupied bandwidth.cases. In this paper, α = 1 is assumed for LTEand hence Iown = 0. The impact of α < 1 can be While HSPA+ and LTE have similar perfor-seen only at high SINRave ; at low SINRave noise mance in terms of coverage, the same is not truepower dominates the denominator of (1) over in- for system-level capacity under network inter-terference power, for both HSDPA and LTE. ference, where LTE has advantage over HSPA+ In the remainder of the paper, we drop due to more advanced radio features, such asthe subscript and denote average signal-to- multiuser-MIMO, frequency-domain schedul-interference-ratio simply with SINR. ing and inter-cell interference coordination.2.2. LTE versus HSPA+ coverage in noise- 2.3. Optimizing LTE system bandwidth for limited scenario coverage In Table 2 link budgets of LTE and HSPA+ are In this section it is shown that, for a given tar-compared. Both systems have 5 MHz system get bit rate, there is an optimum system band-bandwidth, 2Tx × 2Rx MIMO antenna system, width in DL and UL in terms of maximizingand equal antenna and RF characteristics for fair coverage. Of course, typically other decisioncomparison. Note that the link budget is carrier- criteria besides coverage come into play whenfrequency independent, as it is given in terms of choosing system bandwidth for practical net-maximum path loss, not cell range. A single work deployment. The purpose of the follow-cell in isolation is assumed, i.e., no interference ing discussion is mainly to illustrate trade-off(I = 0). between bandwidth and coverage. The following differences in link budgets can Downlink (Fig. 1): Consider a ﬁxed rate ofbe seen: information transmission for a ﬁxed transmit power. As the number of allocated Physical Re- • In HSDPA, L1/L2 control and pilot chan- source Blocks (PRBs) becomes larger, code rate nel overhead is a fraction of the total decreases and channel coding gain in turn in- downlink transmit power (typically about creases. Therefore, for a ﬁxed DL system band- 20%). In LTE, L1/L2 control channels con- width it pays off to use as many PRBs as pos- sume a fraction of DL OFDM symbols, sible to minimize required SINR1 . On the other 30% overhead in time/frequency resource hand, if one is allowed to use the system band- element usage is assumed in Table 2, cor- width as a design parameter, there exists an op- responding to three symbols wide L1/L2 timum system bandwidth for a given target bit control channel region. rate. This is due to the fact that, assuming ﬁxed total transmit power, transmit power per PRB • For a given bit rate target, required SINR (power density) increases as system bandwidth is slightly different for LTE and HSPA+. is reduced. This is illustrated in Figure 1. For bit However, in the noise-limited regime the rate target of 128 kbps, the optimum bandwidth difference in target SINR is typically small, 1 OnePRB is a contiguous chunk of 12 subcarriers, or less than 2 dBs. 180kHz, in frequency domain.
4 J. Salo et alTable 2Link budget comparison in noise-limited scenario, a single user at cell edge for 2x2 LTE and 2x2HSPA+, 5 MHz LTE bandwidth. Target bit rate: 1Mbps DL, 128Mbps UL. Quantity LTE DL HSDPA LTE UL HSUPA Transmit power⋆ 46 dBm 45 dBm♭ 23 dBm† 24 dBm Antenna + feeder gain 15 dB 15 dB −3 dB −3 dB MHA gain - - 2 dB 2 dB Rx Noise Figure 7 dB 7 dB 2 dB 2 dB SNR target −4 dB‡ −5 dB‡ −12 dB‡ −13 dB♢ RX sensitivity −105 dBm♯ −105 dBm −117 dBm −118 dBm Max path loss 164 dB 163 dB 154 dB 156 dB⋆ Sum power of two transmit antennas♭ max HS-DSCH power 42 dBm per antenna† Category 3 LTE terminal‡ All bandwidth allocated for the single user in DL and UL.♢ Eb /N0 target of 2dB for 128 kbps, processing gain of 15 dB .♯ UL and DL noise bandwidth is 5MHz.is 1.4 MHz, while for 512 kbps and 1Mbps targetbit rate, it is 3 and 5MHz, respectively. Uplink (Fig. 2): There is an optimum sys- LTE DL rx sensitivity vs system bandwidth for different bit rates −94tem bandwidth in the uplink, too, as illustratedin Figure 2. The difference to downlink is −96that the noise bandwidth equals actual instanta- −98neous scheduled UL bandwidth, which may be RX downlink sensitivity [dBm] −100less than the system bandwidth. From Fig. 2, it −102can be seen that for bit rates ≤ 256kbps the op-timum number of PRBs is about 3 − 5, meaning −104that 1.4MHz system bandwidth (6 PRBs) would −106 4Mbpsbe sufﬁcient for optimal coverage. For bit rates −108 2Mbps 1Mbpsabove 2Mbps, system bandwidth of 5MHz or −110 512kbpsmore is needed in order for system bandwidth −112 256kbps 128kbpsnot to limit receive sensitivity (UL coverage). −114 In practice, control and common channel cov- 1.4 3 5 10 System bandwidth [MHz] 15 20erage should be taken into account, too. Sim-ulation results in  indicate that for low bitrate services uplink random access channel cov-erage may be the limiting factor, instead of the Figure 1. LTE downlink sensitivity versus sys-PDSCH/PUSCH considered here. tem bandwidth for different bitrates, channel unaware scheduling. Total system bandwidth is allocated to a single user.
Practical Introduction to LTE Radio Planning 5 −98 LTE UL rx sensitivity vs allocated UL bandwidth for different bit rates where Imax = ∑K=1 Imax,k is the maximum inter- k −100 2Mbps 1Mbps ference power at cell edge, and γk is the sub- −102 512kbps 256kbps carrier activity factor of the kth cell. We assume 128kbps that the average cell load is equal for all cells, in −104 other words, γk = γ, for all k.RX sensitivity [dBm] −106 Signal-to-Interference-and-Noise ratio can be −108 written as a function of SNR, SIR and cell load −110 γ as −112 −114 S −116 SINR = (5) −118 I+N S −120 = (6) 0 2 4 6 8 10 12 Allocated UL bandwidth [MHz] 14 16 18 γImax + N 1 = γ 1 , (7) SIRmin + SNRFigure 2. LTE uplink sensitivity versus allo- S S where SIRmin = Imax and SNR = N . We againcated bandwidth for different bitrates, channel emphasize that all quantities are average val-unaware scheduling. ues, where the averaging is over small-scale fad- ing of S and I. The value of SIRmin depends on network geometry and antenna conﬁgura- tion (but not on cell range) and is in practice de-3. LTE IN INTERFERENCE-LIMITED SCE- termined from system simulations or network NARIO measurements. Typical values are SIRmin = A somewhat simpliﬁed engineering view on −4 . . . − 1dB. The interference is here implic-SINR under non-negligible other cell interfer- itly modelled as Gaussian noise, as is the com-ence is presented in this section. In particu- mon practice. It always applies that SINR <lar, interference is here characterized only by its γ−1 SIRmin and SINR < SNR. Therefore, with-average power, whereas (as in HSDPA) it is in out inter-cell interference coordination or intel-fact very bursty when examined at TTI level. ligent channel-aware scheduling, throughput isTherefore, the interplay of HARQ and interfer- limited by other-cell interference.ence (trafﬁc) time correlation is completely ne- The relation of SINR and SNR is shown in Fig.glected in the sequel, for the sake of potential 3 for different values of network load, γ. It canradio planning insights. be seen that even a low network load will satu- rate SINR, and hence throughput.3.1. SINR under random (uncoordinated) inter- 3.2. Link budget with non-negligible interfer- ference ence: Interference Margin At cell edge, the interference from K neigh- The link budget in Table 2 assumed that inter-bouring cells can be written as (I = Iother ) ference power is negligible, i.e., SNR ≫ SIRminγ in Eq. (7). When this is not the case, target SNR K in the noise-limited link budget must be substi-I = ∑ γk Imax,k (3) tuted with a modiﬁed SNR target k =1 = γImax , (4) SNRintf = SINR + NR , [dB] (8)
6 J. Salo et al SNR versus SINR for different average load γ To derive formula for the noise rise, we ﬁrst 20 γ=1 deﬁne in linear scale γ=0.5 15 γ=0.25 γ=0.1 I+N γ=0.05 NR = (9) 10 N γ=0 SNRSINR [dB] 5 = . (10) SINR 0 Substituting, SNR = NR · SINR in Eq. (7) and solving for NR results in −5 −10 1 NR = . (11) −10 −5 0 5 10 15 20 1 − γ SIR SINR SNR [dB] min Here SINR depends on the cell edge through- put target and SIRmin on the network geometry. One can see that noise rise is very steep nearFigure 3. Cell edge SINR versus SNR for dif- the pole SINR = SIRmin . Therefore, it is increas- γferent network average loads (γ), SIRmin = −3 ingly difﬁcult to reach the upper limit cell edgedB. throughput. 3.3. Trade-off between cell range, networkwhere NR denotes noise power rise due to inter- load and cell edge throughputference, and SINR is determined by the through- In this section we illustrate the relationshipput target. Cell range can then be solved in the between the three basic network planning vari-usual manner from SNRintf where the received ables: cell range, average network load and cellsignal power, S, is a function of path loss (cell edge throughput. Towards this end, we writerange) and effective isotropic transmit power.What remains to be done is to obtain an ex- Spression for the noise rise. Before doing so, it is SNR = (12) Nuseful to note, again, that SINR can at cell edge EIRP = , (13)never exceed SIRmin in Eq. (7). As SINR → SIRmin , γ γ LNSNR must in turn increase in order to keep SINR where EIRP is the effective isotropic radiatedless than its upper limit. Therefore, one sees that power, and L is the path loss, given in linearthe noise rise in Eq. (8) will be huge when the scale astarget SINR approaches SIRmin , hence drastically γshrinking the cell size. L = MAd B . (14) Graphically, the same concept can be seen inFig. 3 where the noise rise is the difference be- Here A is the median path loss at one kilome-tween the noise-limited case (γ = 0) and loaded ter distance (depends on carrier frequency andcase, see Eq. (8). For example, with 100% load propagation environment), B is the path loss ex-the noise rise is 23dB at SNR = 20dB. As SINR ponent, d is the distance in kilometers and M isapproaches SIRmin = −3 dB, large increase in the shadow-fading margin.SNR is required for diminishing gains in SINR, In linear scale, SINR as a function of cell rangeand therefore cell edge throughput. d and network load γ is
Practical Introduction to LTE Radio Planning 7 network load versus cell range, for fixed DL throughput 1 1SINR = γ (15) MAd B N SIRmin + EIRP 0.9 0.8 1Mbps 2Mbps 0.7 3Mbps There is a one-to-one mapping between SINR 4Mbps network load 0.6 6Mbpsand average throughput. Therefore, (15) de- 8Mbpsscribes the tradeoff between γ, d and through- 0.5 0.4put. In the sequel, the three two-variable specialcases are considered. 0.3 In the sequel, path loss model is based on 0.23GPP system LTE system simulation speciﬁ- 0.1cation according to L[dB] = 128 + 37 log10 (d) 0 0 0.5 1 1.5 2. Shadowing standard deviation is 8dB with DL cell range [km]shadowing margin of M = 8.7dB (95% single-cell area probability) and building penetrationloss of 20 dB. Other parameters are as givenin Table 2, except that system bandwidth is 10 Figure 4. Network load versus cell range forMHz. The SINR-throughput mapping was sim- ﬁxed downlink cell edge throughput.ulated for a 2 × 2 transmit diversity system.3.3.1. Cell range vs network load, ﬁxed cell edge throughput network load versus throughput, for fixed cell range In Fig. 4, the relation of network load and cell 20 0.5kmrange is shown for various values of cell edge 0.75km 18throughput. As network load increases, the cell 1km 16 1.25kmrange decreases for ﬁxed cell edge throughput. 1.5km Cell edge throughput [Megabits/sec] 143.3.2. Network load vs cell edge throughput, 12 ﬁxed cell range 10 In Fig. 5, the relation of network load and cell 8edge throughput is shown for various values of 6cell range. For our later discussion concerninginter-cell interference coordination, it is instruc- 4tive to note that when cell range is small, cell 2edge throughput is sensitive to network load. 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1For cell range of 0.5km, if one was able to re- network loadduce effective network load from 0.4 to 0.1 celledge throughput would double from 6 Mbps to12 Mbps. Such a reduction could be possibleby applying interference coordination between Figure 5. Network load versus cell edgeneighboring cells, hence directing interference throughput for ﬁxed cell range.to other part of the frequency spectrum. Thiswill be discussed further later in this paper.
8 J. Salo et al Cell edge throughput versus cell range, for fixed network load 4. MULTIPLE-INPUT MULTIPLE-OUTPUT 24 (MIMO) CONFIGURATIONS 22 4.1. MIMO Transmission Schemes in LTE Cell edge throughput [Megabits/sec] 20 18 γ = 0, 0.1, ..., 0.9, 1 While in common engineer talk there are ap- 16 pears to be several interpretations for the term, 14 in this paper MIMO is taken to mean any radio 12 interface that has at least two antennas at both 10 ends, i.e., with this deﬁnition smallest MIMO 8 antenna conﬁguration would then be 2Tx × 2Rx. 6 Note that in LTE all UEs are required to have 4 two receive antennas, while the number of base 2 station antennas can be2 1, 2, or 4. In principle, 0 0 0.2 0.4 0.6 0.8 1 1.2 Cell range [km] 1.4 1.6 1.8 2 there are three ways to utilize MIMO antennas: • Combined transmit/receive diversity: Normal two-branch diversity reception or trans- mission has the beneﬁt of producing twoFigure 6. Cell range versus cell edge throughput copies of the same signal for reception,for ﬁxed network load. which with a suitable signal combining technique reduces fading variation. Sim- ilarly, a 2Tx × 2Rx antenna transmission results in reception of four signal repli- cas, with the corresponding additional re- duction in fading. This version of MIMO is hence simply an enhanced diversity scheme, and as such is not by many con-3.3.3. Cell range versus cell edge throughput, sidered "true" MIMO, since it does not ac- ﬁxed network load tually increase transmission data rate. In In Fig. 6, the relation of cell range and cell LTE, space-frequency block coding basededge throughput is shown for various values diversity scheme is used; this is an open-of network load. As the cell range increases, loop scheme where UE precoder feedbackthe cell edge throughput approaches the same (PMI) is not required at the base station.limit regardless of cell load. Even the low net-work load of 10% limits throughput to less than • Beamforming: This is very similar to the di-14Mbps by creating a noise ﬂoor from other- versity transmission, the main nomenclat-cell interference. It should be noted that down- ural difference being that in beamforminglink common channels and signals alone cause one typically considers a physical antennaat least > 10% network load over the ≈ 1 MHz beam being constructed towards the UE.bandwidth over which they are transmitted. On 2 With precoding using UE-speciﬁc reference signals ("beam-top of this, reference signals are transmitted over forming") arbitrary number of antennas can be used. How-the entire system bandwidth. Sites in early FDD ever, this transmission scheme is not expected to be sup-network implementations are not expected to be ported in ﬁrst vendor releases, due to hesitance from oper- ators to deploy antenna arrays. The coverage gains fromframe-synchronized, so that it is not possible to such conﬁguration in rural/road environments would beavoid inter-site interference by cleverly coordi- substantial however, e.g., 9 dB from an 8-antenna panel ar-nating transmission in time and frequency. ray.
Practical Introduction to LTE Radio Planning 9 This requires closely spaced antennas, un- In practice, the choice of transmission scheme like the diversity schemes which require depends on instantaneous radio channel condi- at least a few wavelength antenna spacing. tions and is adapted continuously. The main dif- Beamforming using 3GPP codebook also ference to non-MIMO transmission that, in ad- requires PMI feedback from UE. Mathe- dition to SINR, now also the channel rank has matically, both beamforming and transmit to be considered, since spatial multiplexing is diversity are both special cases of so-called feasible only if the instantaneous radio chan- rank-1 precoding3 . LTE Rel 8 supports nel (during 1ms TTI) supports transmission of rank-1 precoding (or, "beamforming") us- more than one information stream, or in terms ing pre-deﬁned 3GPP codebook (for 2 of matrices, the 2Tx × 2Rx matrix channel’s con- and 4 antennas), and any vendor-speciﬁc dition number and SINR are good enough. As a beamforming when using UE-speciﬁc ref- rule of thumb, in a spatially uncorrelated chan- erence signals (arbitrary number of base nel spatial multiplexing becomes useful when station antennas). SINR > 10 dB. Looking at Fig. 3 one can see that this requires very low network load for spa- • Spatial Multiplexing: This is what many tial multiplexing to work at cell edge, assuming consider to be a true MIMO transmission that other-cell interference is uncoordinated. scheme. With beamforming and diversity, base station transmits a single stream of 4.2. Beneﬁt of MIMO information but uses the multiple anten- The beneﬁts of different MIMO schemes in nas to either reduce fading (diversity) or downlink are roughly as follows: increase signal power (beamforming). On Transmit/receive diversity: Compared to 1Tx × the other hand, with 2Tx × 2Rx spatial 1Rx the gain of 2Tx × 2Rx is about 6 − 7 dB multiplexing the idea is to transmit two due to following factors: i) transmit power is parallel information streams over the same doubled by adding another ampliﬁer (3dB); ii) bandwidth, hence theoretically doubling average received signal power is doubled be- the data rate and spectral efﬁciency. Both cause of two-antenna reception (3dB); iii) di- open-loop (only channel rank and CQI versity from four signal paths brings additional reported by UE) and closed-loop spatial gain which however strongly depends on the multiplexing are supported (also precod- propagation environment (here pessimistically ing matrix information reported by UE). assumed 0 − 1dB, higher gains are possible). Delving slightly in implementation de- Beamforming (rank-1 precoding): For slow- tails, in LTE both open-loop and closed moving UEs the gain is 1 − 2 dB higher than loop spatial multiplexing have been engi- with the transmit/receive diversity. For fast- neered in such a way that symbols in in- moving mobiles this improvement over diver- formation streams are interleaved between sity diminishes, or goes negative, since the CQI the MIMO subchannels. Thus, the aver- feedback cannot track the channel fast enough. age SINR experienced by the transmitted Spatial multiplexing: This scheme does not im- symbols is the average of the two MIMO prove link budget, but increases data rate. With subchannels’ SINRs. In uplink, spatial two antennas and an ideal MIMO radio chan- multiplexing is not supported for a single nel, the data rate would be doubled. In practice, UE, but two different UEs are allowed to gains are considerably smaller due to the fact transmit at the same time; this is called that for ideal operation the SINR of the two par- multiuser-MIMO. allel subchannels should be high enough to sup-3 Inthis paper, rank-1 precoding and beamforming are con- port the same modulation and coding scheme.sidered synonymous. This is very rarely the case, and the second
10 J. Salo et alstream must be transmitted at lower informa- (||); ii) two cross-polarized transmit antennastion bit rate. This point is further illustrated in at 0 deg and 90 deg angles (+). The dual-branchthe following subsection. UE antenna is a realistic terminal microstrip an- tenna integrated in the terminal chassis, whose4.3. Example of Measured MIMO Radio Chan- branch polarizations are not well-deﬁned, as is nel typically the case for small integrated antennas. Spatial multiplexing transmission is feasible The ﬁgure shows the powers of the two MIMOonly at those time instants when the radio chan- subchannels as a function of travelled distance.nel conditions are favorable. There exist a Roughly in the midpoint of the measurementwealth of literature on the theory of MIMO com- route the UE enters line-of-sight and the sig-munication systems, and the interested reader nal power of the stronger stream (λ1 ) increases.is referred to the references. From radio plan- The second stream power behaves differentlyning point of view, the 2Tx × 2Rx MIMO chan- for the two antenna setups. With the cross-nel power response λ at any given time instant polarized antenna base station setup the sec-can be written, on a subcarrier, as4 λ = λ1 + λ2 , ond stream (λ2 ) is about 15 dB weaker than thewhere λ1 and λ2 are the power responses of stronger one. In contrast with the vertically po-the ﬁrst and second MIMO subchannels, respec- larized antennas the second stream is about 30tively. Each of the subchannels carries one in- dB weaker. Thus, if the channel SINR is 20 dB,formation stream. As with a normal non-MIMO with the cross-polarized conﬁguration the ﬁrstchannel, the powers of the subchannels experi- stream SINR would be almost 20dB while theence signal fading. If the power of the weaker second stream would experience about SINR ≈subchannel, say λ2 , is very weak, the second in- 5 dB which would still allow fairly high bit rateformation stream should not be transmitted at transmission over the second MIMO subchan-that time instant, since it will either be buried nel. However, with the vertically polarized an-in noise, or alternatively, from link adaptation tennas the second stream would be SINR ≈ −10point of view, it is more spectrally efﬁcient to dB, which would not support high-rate trans-transmit one stream with higher-order modu- mission of the second stream.lation and/or less channel coding. In either Fig. 8 illustrates the same measurementcase, usage of spatial multiplexing depends on routes using two different ﬁgures of merit,the instantaneous channel conditions which are namely the total MIMO channel power re-reported to, and tracked, by the base station. sponse and ratio of MIMO subchannel pow-When the second subchannel is weak, the chan- ers, λ1 . From the upper subplot it can be seen λ2nel is said to have "rank one", and spatial multi- that the total power responses of the two an-plexing is not used5 . tenna setups are slightly different. Vertically Fig. 7 presents an example of a narrow- polarized conﬁguration has almost negligiblyband measurement of a 2Tx × 2Rx channel in higher received power in the line-of-sight, whilean urban environment at 2.1 GHz carrier fre- the cross-polarized conﬁguration, in turn, hasqurency. Two base station antenna conﬁgu- slightly better power response in non-line-of-rations are shown: i) two vertically polarized sight. From the lower subplot we note that thetransmit antennas at 3 wavelength separation cross-polarized antenna setup has advantage in terms of the second subchannel power in line-of-4 The symbol λ is used here for historical reasons, since the sight conditions; the second subchannel is aboutMIMO subchannel power is related to the eigenvalues of theMIMO radio channel. 10 dB weaker than the ﬁrst, compared to the 305 Mathematically, the rank of the 2Tx × 2Rx channel is prac- or so dBs for the vertically polarized case.tically always two, so this terminology is strictly speaking a Summarizing the main radio planning mes-misnomer.
Practical Introduction to LTE Radio Planning 11sage from the two ﬁgures: two vertically polarized BTS antennas(||), three wavelength separation 0 λ1 • Dual-stream transmission is only feasible λ2 subchannel power, dB −10 when the SINR of the second subchannel −20 is high enough. This depends on the total −30 channel SINR and the ratio of the subchan- −40 nel powers. −50 0 2 4 6 8 10 12 14 16 18 20 distance, meters • Base station antenna conﬁguration has im- 0 two cross−polarized BTS antennas (+) λ1 pact on the spatial multiplexing perfor- λ2 subchannel power, dB −10 mance in two respects: total received sig- −20 nal and spatial multiplexing gain. −30 −40 • In terms of total received signal power, the −50 0 2 4 6 8 10 12 14 16 18 20 two compared antenna conﬁgurations are distance, meters roughly equal with slight advantage for v- pol in line-of-sight, and vice versa in non- line-of-sight. Figure 7. Example of measured MIMO radio • In terms of spatial multiplexing gain, x-pol channel, power response of the MIMO subchan- has large advantage in line-of-sight. There nels for two base station antenna conﬁgurations. is no noticeable difference in non-line-of- sight. channel power response, v−pol versus x−pol MIMO channel power response, dB The results indicate that cross-polarized base 0 v−pol (||) −5 x−pol (+)station antenna conﬁguration should be favored −10when usage of spatial multiplexing is desired, as −15it offers more robust performance in all channel −20conditions. While these conclusions are in this −25paper illustrated by means of a single measure- −30 5 10 distance, meters 15 20ment example, similar conclusions have been ratio of MIMO subchannel powers, v−pol versus x−poldrawn from large channel measurement cam- 50paigns reported in literature . 40 λ1 / λ2, dB 30 205. RADIO RESOURCE MANAGEMENT FEA- 10 TURES 0 5 10 15 20 distance, meters So far, we have discussed a simple "baseline"LTE radio conﬁguration, assuming fully ran-dom (uncoordinated) interference and no spe-cial radio resource management features. TheLTE radio interface speciﬁcation Release 8 sup- Figure 8. Example of measured MIMO radioports several advanced air interface functional- channel, total channel power response and sub-ities that potentially improve network perfor- channel power ratio for two base station antennamance in certain environments. These are ex- conﬁgurations.plained in the following.
12 J. Salo et al5.1. Frequency-Aware UL/DL Scheduling from inter-cell interference coordination, to be Baseline scheduler assumes no knowledge of discussed next.the instantaneous frequency response of the ra-dio channel. For slow-moving UEs such knowl- 5.2. Inter-Cell Interference Coordinationedge is available at the BTS in the form of sub- There are two basic methods to coordinateband CQI feedback from UE. In case there is frequency usage between cells in a network:only a single user to be scheduled in downlink, static and dynamic. Static coordination meansit is optimal to allocate whole bandwidth, since ﬁxed allocation of frequency resources per cellthis offers highest coding gain. In UL, the opti- over extended periods of time while dynamicmal bandwidth for a single simultaneous user frequency allocation means fast coordinationmay be less than the system bandwidth, see within a time frame of seconds or even less,Fig. 2 for an example. Focusing on the down- without need for manual operator intervention.link, in the case there are multiple simultane- In LTE, dynamic inter-cell interference coordina-ous UEs to be scheduled during a TTI, the base tion is inherently supported by 3GPP-speciﬁedstation can schedule each UE in the frequency signalling between base stations.subband with the strongest channel power re- The network-level throughput gain from fre-sponse, hence avoiding fading dips in frequency quency use coordination is most visible withdomain. This results in frequency-domain mul- low-to-medium network load; when the net-tiuser scheduling gain, which can be quantiﬁed work load is high no coordination will helpin terms of network-level throughput gain and since all frequency resources of all cells tendcell edge throughput gain. to be used most of the time. Hence, inter- Practical (including system imperfections) ference coordination is most useful for non-achievable downlink multiuser scheduling congested networks. Looking at Fig. 5, con-network-level throughput gains have been sim- siderable cell edge throughput gains are possi-ulated to be about < 30 − 40% at network level, ble if one can reduce the effective network loadfor 3 − 7 users . For the cell edge, throughput seen by the UE by allocating neighbor cell fre-can be almost doubled with a suitable schedul- quency resource on a different part of the sys-ing algorithm (always vendor-dependent). Here tem bandwidth. From Fig. 5, if the cell ra-the gain should be interpreted relative to the dius is small the throughput gain can be con-throughput of a frequency-unaware scheduler siderable. At network-level, throughput gainswith the same instantaneous number of users similar to channel-aware scheduling have beenper TTI. Of course, even if the network-level reported in literature for low-to-medium net-throughput increases, actual per-user through- work loads (γ < 50%). Summarizing, inter-put will decrease when the number of simulta- cell interference coordination is mostly useful inneous users increases. small-range cells during off-peak hours or dur- Another beneﬁt of the frequency-aware ing early phases of network life cycle.scheduling emerges in case the time correlationof other-cell interference power is long enough 5.3. Uplink Fractional Power Controlso that the scheduler has time to learn the in- LTE supports both cell-speciﬁc and UE-terference spectrum and divert transmission to speciﬁc uplink power control. In LTE down-a cleaner part of the system bandwidth. This link, there is no power control in the conven-is only possible if UE is moving slowly and tional sense. With LTE power control schemethe interference is persistent in time over, say, it is possible to compensate only a fraction ofa few dozens of TTIs. This interference rejec- the path loss, reducing transmit power of celltion gain can be seen similar to what is obtained edge UEs that generate most of the interference to other cells. This is in contrast to conventional
Practical Introduction to LTE Radio Planning 13power control schemes that attempt to compen- UE cannot measure and report two cells withsate path loss to the full extent allowed by the the same PCI. This part of the PCI allocationUE transmit power. The beneﬁt of the fractional should not pose a problem since there are 504scheme is that, due to reduction in UE trans- PCIs deﬁned.mit power, network-level interference is reduced The second, perhaps less obvious, purpose ofand overall UL throughput is increased. This the PCI is to serve as a resource allocator param-improvement is at the expense of reduction in eter for downlink and uplink Reference Signalscell edge UL throughput. For each optimization (RS).goal - maximizing cell edge throughput or net- DL Reference Signal: Downlink reference sym-work throughput - there exist an optimum op- bols ("LTE pilot signal") are allocated in a time-timum power control parameter set. For early frequency grid as shown in Fig. 9. The ﬁguresystem simulation examples, see . illustrates three frame-synchronized cells with PCIs 9, 10, and 11. For example, these could6. INTRODUCTION TO RADIO PARAME- represent different cells of the same eNB site6 . TER PLANNING In time domain the RS are always transmitted in the same OFDM symbol. However, in fre- The three most important tasks in LTE pre- quency domain each cell has a different shiftlaunch radio parameter planning are: i) Phys- determined by modulo-3 of the PCI, denotedical Cell Identity (PCI) allocation, ii) Physical PCI mod3. In Fig. 9, cell #1 has PCI mod3 =Random Access Channel (PRACH) parameter 0, cell #2 has PCI mod3 = 1, while cell #3planning, and iii) uplink reference signal se- has PCI mod3 = 2. With this PCI allocation,quence planning. Obviously, there are numer- the RS of different cells do not overlap in fre-ous other types of air interface parameters that quency, which results in less interference oncan be tuned in LTE, such as UL power control UE channel estimation. If neighbouring sitesand handover thresholds. However, the three are frame-synchronized (e.g., using GPS), or thetasks picked out above are the ones that must frame timing offset between sites is known, thebe planned before network launch, while the re- PCI allocation should also be coordinated be-maining ones can arguably be optimized also af- tween neighbouring sites, which brings addi-ter launch, provided that reasonable default val- tional complexity in the allocation process. Onues are available. In the following, we brieﬂy re- the other hand, if the neighbouring sites are notview the basics of the three main pre-launch pa- frame-synchronized or the frame timing offset isrameter design tasks. More detailed treatment not known (i.e., random), one cannot coordinateof parameter optimization and planning will be PCI allocation between sites.a topic of a separate exposition. It should be noted that assigning PCIs as shown in Fig. 9 results in RS overlapping with6.1. PCI planning the control and data Resource Elements of the In LTE, Physical Cell Identity (PCI) allocation neighbouring cells. Therefore, a design choiceis a task somewhat similar to scrambling code must be made between RS↔RS interference andallocation in WCDMA. PCI is encoded in the RS↔PDSCH/PDCCH intererence. The currentphysical layer synchronization signal transmis- engineering consensus seems to be that the lat-sion and is used by the UE for neighbour cell ter choice is favoured, due to the fact that ifhandover measurement reports. Hence, as in two frame-synchronized cells have the sameWCDMA, the PCI should uniquely identify the 6 Note that Fig. 9 shows transmission of both transmit anten-neighbouring cell to the serving eNB, within a nas in the same grid, to save space. Frame synchronizationcertain geographical area. Consequently, PCI in this context means that the DL transmission of a radioreuse distance should be large enough so that frame starts at the same time instant in all three cells.
14 J. Salo et alFigure 9. Location of Reference Symbols within one PRB for different PCIs. Frame-synchronized cellsshown.PCImod3, the additional drawback is that their tentially interfering cells is different. This is be-Primary Synchronization Signals will interfere cause in the simplest scheme the DM RS base se-with each other, causing problems in cell search quence index is equal to u =PCI mod30, whereand handover measurements. u = 0 . . . 29 is the base sequence index. An ex- UL Reference Signal: LTE uplink shared data ample of this simple PCI-based DM RS sequencechannel (PUSCH) carries Demodulation Refer- allocation scheme is shown in Fig. 10. In prac-ence Signal (DM RS). Optionally also Sound- tical network deployments this simple planninging Reference Signal is transmitted in the up- criterion cannot be always fulﬁlled. As a remedylink. The uplink DM RS are constructed from to such a case, there are additional sequence al-Zadoff-Chu sequences which are divided into location schemes built-in in the 3GPP speciﬁca-30 groups. Roughly, this means that for a given tion, most notably a static base sequence indexnumber of PRBs allocated in the uplink there offset parameter, DM RS cyclic shift planning,are 30 different base sequences that can be used and pseudo-random base sequence hopping ("u-as the reference signal7 . The cross-correlation hopping"). These will be discussed next.between the base sequences is on average low,which is beneﬁcial from inter-cell interferencepoint of view. It follows that the planning re- 6.2. UL Reference Signal sequence planningquirement is that the neighbouring cells should The simple PCI-based based sequence uplinkbe allocated different base sequences. The sim- DM RS allocation scheme from the previous sec-plest method is to ensure that PCI mod 30 of po- tion may be difﬁcult to plan since the same base7 There sequence is reused in every 30th cell, which may are actually two groups of 30 sequences deﬁned lead to insufﬁcient cell separation.for PRB allocations of 6 or more. These can be pseudo-randomly alternated. However, this option is not considered There are three options to reduce uplink inter-in this paper. cell interference in such cases:
Practical Introduction to LTE Radio Planning 15Figure 10. Simple PCI-based uplink DM RS allocation scheme. Different base sequence u allocated toevery cell based on PCImod30, every cell has ∆ss = 0.Figure 11. Example of how the parameter ∆ss can a base sequence collision between two cells. Theindoor cell has been allocated PCI=30 which would result in u = 0 if ∆ss = 0, hence creating ULinterference with cell PCI=0. Setting ∆ss = 29 gives u = (30 + 29) mod 30 = 29 instead.
16 J. Salo et alFigure 12. Example of u-hopping. Cluster #1 has hopping-pattern σ = 0 and cluster #2 has hopping-pattern σ = 1. Sequence offset ∆ss is different in all cells within a cluster to prevent systematic basesequence collisions. Random base sequence collisions are possible in the cluster border.Figure 13. Example of cyclic shift planning. Cells of the same site are allocated the same base sequenceu using the offset parameter ∆ss . Inter-cell interference between cells of the same site is mitigated bysetting different cell-speciﬁc cyclic shift (cs) for cells of a site.
Practical Introduction to LTE Radio Planning 17 • Bypass the simple PCI-based base alloca- patterns deﬁned since there are 504 PCIs de- tion scheme by explicitly deﬁning the base ﬁned, i.e., σ = 0 . . . 16. A foreseen method of sequence used in the cell. This brings ad- PCI planning, where near-by cells are assigned ditional ﬂexibility to base sequence allo- near-by PCI values, results in grouping of cells cation, and effectively decouples the PCI into "clusters-of-30", where within each cell clus- planning from uplink DM RS base se- ter the same hopping-pattern is used. The off- quence planning. set ∆ss must be set differently for cells within a cluster as otherwise continuous base sequence • The base sequence u can change pseudo- collisions take place. At the border of two cells randomly for every 0.5ms time slot. having different σ, inter-cell interference is av- This planning option randomizes base se- eraged since two different hopping patterns are quence collisions and averages inter-cell utilized. An example of this planning scheme is interference. shown in Fig. 12. • Different cyclic shifts of a ZC sequence are Cyclic shift planning: So far we have consid- orthogonal. This can be utilized by assign- ered only how to reduce inter-cell interference ing a different cyclic shift on two cells that by assigning different base sequences to every use the same base sequence u. The cell- neighbouring cell. Another interference reduc- speciﬁc static cyclic shift is broadcasted on tion method follows from the fact that ZC se- BCCH. quences have the useful property that two dif- ferent cyclic shifts of the same base sequence are Deﬁning u independently from PCI: The simplest orthogonal9 . Therefore, if a pair of cells havescheme assigns the base sequence index u to a been allocated the same base sequence u, inter-cell as modulo-30 of PCI. Optionally, the base cell interference can be reduced by assigning asequence can be assigned to a cell as different cyclic shift to the cells. This scheme can be applied to cells of one site as shown in Fig. u = ( PCI + ∆ss ) mod 30 , 13. Other applications, tangential to our present discussion, are uplink multi-user MIMO (wherewhere ∆ss = 0 . . . 29 is an offset parameter each UE uses a different cyclic shift) and, in LTE-signalled on BCCH. In the simple PCI-based A, uplink single-user MIMO (where each UEscheme ∆ss = 0. With ∆ss it is possible to avoid transmit antenna uses a different cyclic shift), orcollisions in cells that would otherwise use the a combination of the two. It should be notedsame u due to PCI allocation. An example of ∆ss that cyclic shift planning can also be combinedplanning is shown in Fig. 11. with the other two schemes discussed in this Pseudo-random u-hopping: If u-hopping is acti- section.vated, the base sequence used in the cell changesat every time slot in a pseudo-random fashion. 6.3. PRACH parameter planningThe index of base sequence in time slot n 8 The random access procedure in LTE uplink begins when UE transmits a preamble to the un = (νn + PCI + ∆ss ) mod 30 , eNB. The speciﬁc preamble is selected randomly by UE from a pre-deﬁned set of 64 Zadoff-Chuwhere νn = 0 . . . 29 is pseudo-random integer sequences. To avoid call setup anomalies, eachdeﬁned by the hopping-pattern. The hopping cell (within a reuse distance) has its own uniquepattern deﬁned used in a cell is deﬁned by the set of 64 preambles, and the information of theindex σ = ⌊ PCI ⌋. Thus, there are 17 u-hopping 30 speciﬁc set to use in the cell is broadcasted on8 There are 20 time slots in a radio frame. The hopping pat- 9 Cyclicshifts of extended ZC sequences used in uplink DMtern is re-initialized at the beginning of every radio frame. RS are not orthogonal, however.
18 J. Salo et alFigure 14. Example of root sequence allocation for different cells. Four sequences per cell are allocated,PRACH conﬁguration index = 8, see Table 3.BCCH. The preamble sequence allocation de- dius. One can see that a cell requires ﬁve ZCpends on cell range: for example, if one (un- sequences per cell for up to 7.3km radius, whichrealistically) requires that all cells must be ac- is typically sufﬁcient for urban and suburbancessible from 100km away then there would be macro cells. This results in sequence reuse fac-only a total of 839 preamble sequences available. tor of at least 839/5 ≈ 167 cells, hence allowsAs each cell must have exactly 64 preambles, for easy planning process. Example of root se-the total number of cells within a reuse distance quence allocation is shown in Fig. 14. Thewould then be only 839 ≈ 13, and in this case 64 point to notice here is that root sequence indiceseach cell consumes 64 ZC sequences (one full of cells must not overlap within the reuse dis-ZC sequence per preamble) out of the total of tance10 . As for typical scenarios there are plenty839 ZC sequences available for PRACH. More of ZC sequences available, for safety margin andsequences can be generated if the "random ac- ease of planning one can overdimension the ran-cess radius" of cells is dimensioned more real- dom access radius of the cells in a given plan-istically; continuing the example, if maximum ning region, to simplify allocation. In the exam-random access cell range is chosen as 1km, one ple of Fig. 14 each cell consumes four sequences,ZC sequence generates 64 preambles, instead corresponding to maximum radius of 5.4km forof the one preamble in the 100km case. sum- every cell.marizing, the PRACH parameter planning task If root sequences used in neighbouring cellsconsists of deciding i) the maximum cell range overlap, the transmitted preamble may be de-for random access, and ii) allocating the ZC se- tected in multiple cells. The drawback are thequences to cells. Table 3 lists the number of ZC sequences re- 10 Toreduce signalling load, only the index of the ﬁrst rootquired per cell, for a given random access ra- sequence used in the cell and the PRACH conﬁguration in- dex is transmitted on BCCH.
Practical Introduction to LTE Radio Planning 19"ghost" preambles which consequently result radio resource management features and de-in unnecessary PDCCH and PUSCH resource tailed radio parameter planning tasks were in-reservation in those cells that whose random ac- troduced.cess responses the UE chooses to neglect11 . REFERENCESTable 3 1. H. Holma et al (eds.), LTE for UMTS, Wiley,PRACH ZC sequence parameters from the 3GPP 2009.speciﬁcation. 2. F. Khan, LTE for 4G Mobile Broadband, PRACH con- Number Random ac- Cambridge University Press, 2009. ﬁguration in- of root se- cess radius 3. S. Sesia et al (eds.), LTE, The UMTS Long dex quences per [km]⋆ Term Evolution: From Theory to Practice, cell Wiley, 2009. 1 1 0.7 4. 3GPP TS 25.814, v7.1.0, 2006. 2 2 1 5. P. Suvikunnas et al, Comparison of MIMO 3 2 1.4 4 2 2 Antenna Conﬁgurations: Methods and Ex- 5 2 2.5 perimental Results, IEEE Trans VT, 2008 6 3 3.4 6. A. Pokhariyal et al, Frequency Domain 7 3 4.3 Packet Scheduling Under Fractional Load 8 4 5.4 for the UTRAN LTE Downlink, IEEE VTC, 9 5 7.3 2007 10 6 9.7 7. C. U. Castellanos et al, Performance of Up- 11 8 12.1 link Fractional Power Control in UTRAN 12 10 15.8 LTE, IEEE VTC, 2008. 13 13 22.7 14 22 38.7 15 32 58.7 Jari Salo received the degrees of Master of 0 64 118.8 Science in Technology and Doctor of Science in⋆ Includes 1.2µsec implementation margin for delay Technology from Helsinki University of Tech-estimation nology (TKK) in 2000 and 2006, respectively. Since 2006, he has been with European Com- munications Engineering Ltd, where he works as a consultant for mobile wireless industry and operators. His present interest is in radio and7. CONCLUSION IP transmission planning and optimization for wireless networks. He has written or co-written In this paper topics related to radio planning 14 IEE/IEEE journal papers, 35 conference pa-in LTE were discussed. Besides the link budget pers, and is the recipient of the Neal Shep-for coverage and interference-limited cases, the herd Memorial Best Propagation Paper Awardtradeoff between cell range, cell edge through- from the IEEE Vehicular Technology Society,put and network load was discussed under un- for the best radio propagation paper publishedcoordinated interference. Impact of MIMO and in IEEE Transactions on Vehicular Technologyantenna conﬁgurations was illustrated using a during 2007. He is also a co-creator of themeasurement example. Finally, some advanced wide-band spatial channel model adopted by11 Similar problem exists in GSM networks where ghost ran- 3GPP for LTE system-level simulations. E-mail:dom accesses result in unnecessary SDCCH reservation. email@example.com
20 J. Salo et al Mohammad Nur-A-Alam obtained his Mas-ter of Science degree from Tampere Univer-sity of Technology (TUT) in 2009 and is cur-rently working towards Ph.D in the same de-partment. After attaining his BSc. in Electricaland Electronic Engineering from Islamic Univer-sity of Technology in 2003, he worked in Tele-com Malaysia International as a Network Plan-ning and Optimization Engineer. From 2008 to2010, he was with ECE Ltd as LTE Consultant.He is currently working for Nokia Siemens Net-works as system solution manager and his inter-est are in LTE/HSPA+ radio network planning. Kwangrok Chang received the degrees ofMaster and Ph.D in Electronic and Electrical En-gineering from POSTECH (Pohang Universityof Science and Technology) in 1995 and 1998,respectively. In July of 1998 he joined LG Elec-tronics Ltd for two years, after which he hasbeen with Nokia Siemens Networks Ltd., wherehe is currently heading the Network Planningand Optimization Group of Japan and Korea.His present interests are network capacity man-agement and enhancement of end user perfor-mances in high speed data networks, such asHSPA, and network planning/dimensioning ofLTE system.