LTE網(wǎng)絡(luò)優(yōu)化方案：上下行鏈路不均衡的優(yōu)化分析

1、Tdoc-R2-0727213GPP TSG RAN WG2 #58bisththOrla ndo, U.S A, 25- 29 June 2007Agenda item:4.5.1Source:NTT DoCoMo, Telecom Italia, T-MobileTitle:Use of cell specific offsets and readi ng n eighbour BCHDocume nt for:Discussi on1. In troducti onIn RAN2#58 in Kobe, RAN2 has decided that, to allow for suffic

2、ie nt mobility con trol without NCL, an offset value shall be included in BCH, and that the UE shall read the neighbour cell BCH to obtain this offset value both in ACTIVE and IDLE modes 1. The offset value biases the measured quantity of the corresponding cell for mobility control. It was expressed

3、 by operators that this offset is necessary primarily to control the cell boudaries considering the DL andUL coverage imbalanee, caused by DL/UL feeder cable loss differenee (due to TMA) and eNBs having different tran smissi on powers adjoi ning in the n etwork 2. However, i n RAN Ple nary #36 in Bu

4、sa n, the decisi on was take n back after some ven dors expressed concerns on the han dover/cell reselecti on delays and UE battery con sumpti on 3. Revisit ing this issue, this docume nt expla ins why cell specific offsets are thought n ecessary, summarises con cer ns of readi ng n eighbour BCH, an

5、d prese nts our positi on on the issue.Note that the support for opti onal NCL for in tra-freque ncy cells has already bee n agreed in RAN2, and this has not bee n reopened. The optional NCL should serve purposes such as to set serving-neighbour pairwise specific offsets or to blacklist certain cell

6、s. It can also be used to speed up cell detection, although relevance of this is pending RAN4 respon se. Hen ce, the only ope n questi on that n eeds to be addressed is whether UE reads n eighbour BCH and obta ins the offset value included therewith, ” and this is the exact focus of this paper.2. Us

7、e of cell specific offsets2.1 DL/UL imbala nce problemAs mentioned in 2, the need for a cell specific offset is mainly motivated by the fact that eNBs of different power classes can be adjoining in many places throughout the n etwork, and that each cell has differe nt DL and UL feeder cable losses (

8、i.e., DL/UL feeder loss differe nce due to TMA). By sett ing approprite offset values, the DL/UL imbala nce can be mitigated. Before going into how offsetting works, the DL/UL imbalance problem has to be understood.Figure 1 shows the principle of DL/UL imbalance caused by cable loss difference. Assu

9、ming two base stations, having the same antennas and propagati on coefficie nts, the cell bou ndary will be at the cen tre (equidista nt) based on path loss (UL orie nted). However, if the two base stati ons have differe nt cable losses (or differe nt tra nsmissi on powers), the cell boundary will d

10、eviate from the centre based on Ec/N0 (DL oriented), hence causing DL/UL imbalance.Tx power = 43 dHm_ =. _-Tx power = 43dl/ul imbaianca-Fig. 1 DL/UL imbalance principle.2.2 Mitigat ing DL/UL imbala neeThe DL/UL imbala nee problem can be mitigated/tolerated in a n umber of ways:Alt.1:Mobility control

11、 based on Ec/NO (do noth in g)Alt.2:Adjust DL total tra nsmissi on power based on the UL coverage (DL/UL bala ncing)Alt.3:Adjust DL pilot tran smissi on power based on the UL coverageAlt.4:Use cell specific offsetsEach solution is described in detail in the sequel. Note that these solutions are not

12、exclusive, and can be combined if so desired.To simplify the discussion, the traffic distribution aspect is omitted for the qualitative assessment below. However in practice, cell pla nning has to take into accou nt real traffic distributi ons, which can be far from ideal uni form. It should be no t

13、ed that this adds ano ther dime nsion to n etwork pla nning, which can be quite complicated.2.2.1 Alt.1: Ec/N0 based mobility con trol (do nothi ng)The first solution is to do nothing special, and just rely on Ec/N0 to control mobility (Fig.2). This can be optimum for the DL, however, the UL will de

14、teriorate especially at cell edge. If the UE has sufficient transmission power (typ. small cells), it will tra nsmit at a larger power to satisfy its QoS (i.e., the required SIR for a desired rate). If theUE didnothave eno ugh power (typ. large cells), the likely con seque nce in a scheduler based s

15、ystem as in LTEisthatit tra nsmits(is scheduled) more freque ntly so that the desired rate can be met. I n either case, this will create larger in terfere nce at the n eighbour cell, and hence decreases system capacity in the UL.Cable loss= 3dBeNBULdegradaflionFig. 2 Ec/N0 based mobility control (do

16、 nothing).TMA2#2.2.2 Alt.2: DL total Tx power adjusti ng (bala ncing)#The second solution is to adjust the total DL transmission power of the base station such that the DL boundary matches the UL boundary, while the power ratio of the pilot symbols used for mobility measurements is maintained. An ex

17、ample case is show n in Fig. 3. The total power of the base stati on hav ing a smaller cable loss is reduced, such that the emitted power from the antenna is equal to that of the neighbour having a larger cable loss. This will balanee the DL/UL bou ndaries, hence resolv ing imbala nee, and will be t

18、he optimum in terms of the UL. This has a ben efit in that it limits DL in terfere nee at the n eighbour, and gen erally improves lor/(loc+N0) at cell bou ndaries. However, it has some drawbacks as the cell with the reduced power now has limited capacity due to reduced power. It may also reduce chan

19、nel estimation and cell detection performances due to reduced pilot power. This will be more evident in noise limited deployeme nts (e.g., large cells).3#Fig. 3 DL total Tx power adjusting (balancing).2.2.3Alt.3: DL pilot Tx power adjusti ngIn stead of adjusti ng the total tran smissi on power, the

20、tra nsmissi on power of the pilot symbols can be reduced,whilemaintaining the total power (Fig. 4). This would also resolve DL/UL imbalanee. Since the total power is maintained and data tran smissi ons can be allocated larger powers, the cell can provide larger capacity compared to Alt.2. However, t

21、his will create larger in terfere nce at the n eighbour cell, and will reduce cha nnel estimati on and cell detecti on performa nce due to reduced pilot power. This alter native will also require some further adjustme nts e.g.,The tran smissi on powers of other DL com mon cha nn els (such as BCH) al

22、so n eed to be adjusted con sideri ng the cha nnel estimati on quality.The pilot/data symbol power ratio must be adjusted and signalled to the UE so that it can correctly demodulate 16QAM or 64QAM sig nals.These adjustme nts will have to be performed for each cell, which can in cur exte nsive effort

23、s on operators.eNBTMACable ktss I#DUUL balancetTx power = dBmPlot power = S3Tx power = 43 cSmPiol power = dBm f 28 dBmFig. 4 DL pilot Tx power adjusting.2.2.4 Alt.4: Use cell specific offsetsAno ther alter native is to use cell specific offsets (Fig. 5). With this alternative, the total and pilot po

24、wers do n ot have to be modified, and the DL/UL imbala nce can be mitigated by in stead, setti ng a cell specific offset at each cell. The offset value can be set such that it reflects the cable loss (DL/UL difference). By having the UE take into account the offset value in mak ing mobility decisi o

25、ns, the imbala nce can be resolved.The offset can be used to optimise cell bou ndary for the UL, DL, or any where in betwee n, by sett ing the appropriate offset value. If the offset is used to optimise boundary for the UL, for the cell transmitting at a higher power (left in Fig.5) the DL quality/c

26、apacity improves (1), however, the neighbour cell (right in Fig.5) will suffer degraded DL quality/capacity (2) and degraded DL com mon cha nn el (CCH) quality (3). Note that if the cell bou ndary is optimised for the DL (offset = 0), the performa nce will be the same as for Alt.1.Since the pilot po

27、wer is not reduced as in Alt.3, the cha nnel estimati on quality does not degrade with this alter native. The pilot power is usually adjusted considering the optimum (tradeoff) allocation between the pilot and data, and in that sen se, this alter native allows to maintain the optimum power allocati

28、on sett ings (i.e., it should not require power adjustme nts per cell). Although sig nalli ng of the offsets in cur some additi onal overhead, this overhead can be trivial considering the entire system bandwidth. As such, it offers a considerably simpler solution to resolving DL/UL imbala nce, espec

29、ially if the offset is read from the n eighbour BCH.Cable loss0GGH UTMhJdBIMAThe cdl bouncteiy can be set liembty by the uffseL 上4、戸-一" 二X.麗畝Mi論噥fipsc茜BCCH.£iirs =SdBTMA.Cable loss'=a<ra4lx power = 43 d&ncNBFig. 5 DL/UL imbalance mitigation by use of cell specific offsets.2.2.5Q

30、ualitative comparis onTx powEf = 43 cEmTable 1 summarises qualitative comparis on of the four alter natives.Table 1 Qualitative comparison.AlternativeAlt.1: Ec/N0 based mobility control (do nothing)Alt.2: Adjust DL total Tx power (balancing)Alt.3: Adjust DL pilot Tx powerAlt.4: Use cell specific off

31、setsIDLEUL (RACH)ConProProProDL (CCH)Pro(Con)(in noise limited deployments)ConConACTIVEULCon(reduced capacity due to larger interference)ProProProDL (interior)ProCon(reduced capacity due to reduced total power)(Con)(reduced capacity due to reduced channel estimation quality)ProDL (cell edge)ProProCo

32、n(reduced capacity due to larger interference)Con(reduced capacity due to larger interference)UE complexityProProProCon(UE needs to consider offsets)OAMPro(easy to operate)Con(needs DL power adjustment per cell)Con(needs pilot power and data/pilot ratio adjustments per cell)Con(need to set BCH info

33、per cell, but can be easy esp. if not by NCL)2.2.6 Capacity comparis onHere, two extreme sce narios, i.e., Alt.2 (bala ncing, hereafter) and Alt.4 (offsett ing, hereafter), both optimised for UL, #are compared. Note that Alt.3 is expected to perform somewhat similarly as Alt.4, although any differen

34、ee cannot be assessed without detailed analysis. (It can be expected that Alt.3 performs worse due to poorer channel estimation quality.) The deployme nt Cases 1 (ISD = 500 m) and 3 (ISD = 1,732 m) in TR 25.814 4 are assumed, with vertical antenna patter ns and tilt ing also take n into accou nt. Th

35、e feeder cable loss was modelled as log-no rmal with -5 dB, +5 dB limits, with the sectors of the same eNB having the same cable loss. The log-normal std. deviation was set in a range 1-4 dB, and the media n was set such that the result ing gain on average is equal to that assumed in Cases 1 and 3.

36、Snapshot system level simulations were performed to obtain the Ior/(Ioc+N0) distribution, and from this, the normalised capacity was derived considering two factors: CCH overhead according to the lor/(loc+N0) at 2% coverage, and DL-SCH capacity scaling according to the average Ior/(Ioc+N0). For the

37、CCH factor, it was assumed that a 2 dB degradati on in Ior/(Ioc+N0) in curs a 3 dB larger overhead for CCH (based on our in ternal an alysis).Figures 6 and 7 compare the Ior/(Ioc+N0) distribution of offsetting (Alt.4) and balancing (Alt.2). It can be observed that offsett ing produces worse Ior/(Ioc

38、+N0) in Case 1 (Fig. 6(a), 7(a). This is due to in creased in terfere nee at cell edge. However, in Case 3, offsetting produces better Ior/(Ioc+N0) even at around cell edge (Fig. 6(b), 7(b). This is because bala ncing (Alt.2) reduces the total tra nsmissi on power and in creases the impact of therma

39、l no ise throughout the en tire cell.0.80.40.21015Ior/(Ioc+N0) dB0 -100.60.80.60.40.2No imbala nce Offset 1dB Offset 2dB Offset 3dB Offset 4dB Bala nce 1dB Bala nce 2dB Balance 3dB Balance 4dBIor/(Ioc+N0) dB(b)Fig. 6 lor/(loc+N0) distribution for (a) Case 1 and (b) Case 3.0.020.080.060.04 No imbala

40、nceOffset 1dBOffset 2dBOffset 3dBOffset 4dBBala nce 1dB Bala nce 2dBBala nce 3dB Bala nce 4dBIor/(Ioc+N0) dB0.10.080.060.040.02-12-10 -8-6Ior/(Ioc+N0) dB-4-145-14#(b)Fig. 7 Ior/(Ioc+N0) distribution (cell edge) for (a) Case 1 and (b) Case 3.-14#Offset-Balance0 IILI01234Log-no rmal std. dev. of DL/UL

41、 imbala nce dBFigure 8 compares the system capacity. As shown in Fig. 8(a), offsetting provides less capacity in Case 1 when some CCH overhead is considered. This is mainly due to the larger CCH overhead caused by lor/(loc+N0) degradation at around cell edge. However, as shown in Fig. 8(b), offsetti

42、ng provides larger capacity than balancing in Case 3. The capacity is about 20% larger at 2 dB imbala nce and 10% CCH overhead. This is because bala ncing reduces the total power and hence in creases the impact of thermal no ise. With offsett ing, the impact of thermal no ise is kept at mini mum. Th

43、erefore, offsetting is beneficial in providing larger capacity in moderate/large cell scenarios (more evident in thermal noise limited scenarios).0% CCH e b 10% CCHv t 20% CCH “30% CCH18 n-6 n-24 n-18 n-6 4 a a2 aT 90123Log-no rmal std. dev. of DL/UL imbala nce dB0(a)(b)Fig. 8 Normalised capacity fo

44、r (a) Case 1 and (b) Case 3.2.2.7Ben efits of cell specific offsetsSummaris ing above discussi ons and an alysis, the ben efits of cell specific offsets are:Offsetti ng provides larger system capacity whe n cell size is moderate/large (i.e., more evide nt in thermal no ise limited sce narios);Offset

45、ti ng allows the operator to flexibly choose betwee n UL optimised and DL optimised;Easy to operate as the offset is just a parameter on BCCH (or DCCH), and does not involve complicated power adjustme nts.Note that bala ncing by DL tran smissi on power adjusti ng is any way possible, regardless of s

46、upport for an offset mechanism. The mechanism to utilise cell specific offsets allows the operator to cope with DL/UL imbalance in various ways, provid ing flexible cou ntermeasures depe nding on the deployme nt sce nario. Without the offsett ing mecha ni sm, the operator is restricted to either tol

47、erati ng UL losses, or toen gage a complicated process of adjust ing DL tran smissi onpowers. Therefore, the offsett ing mecha nism should be supported in LTE.3. Readi ng neighbour BCH3.1 Ways of sig nail ing cell specific offsetsIn section 2, the need for cell specific offsets has been justified. T

48、he question then becomes how the offset values are sig nalled to the UE. As discussed in 2, there are two ways:Alt.1:UE reads offset in eluded in neighbour BCH- The offset is set for its own cell in BCH (no need to care about neighbours).- The same offset applies to all the neighbouring cells (1-to-

49、all). The UE has to read the n eighbour BCH to avoid any pin g-p ongs 5.This was decided as ma ndatory in RAN2#58 in Kobe, but reope ned in RAN Ple nary #36.Alt.2:Use NCL The offsets applicable to the releva nt n eighbours are broadcast by NCL.- The offset can be specific to certa in serv ing-n eigh

50、bour relati on ship (1-to-1).- The use of NCL causes larger overhead compared to Alt.1.- NCL needs to be planned and set by OAM, which can be complicated.- This is already supported opti on ally in LTE (RAN2 decisi on).The only questi on that n eeds to be addressed is whether LTE shall support Alt.1

51、 or not.3.2 Concerns of read ing n eighbour BCHAs discussed in detail in 2 and above in section 2, the offset can be used primarily to mitigate DL/UL imbalanee. As such it is prevale nt that each cell requires a differe nt offset value, and that it is sufficie nt for most cases if the offset is 1-to

52、-all. That is, the 1-to-1 granularity is not necessary to cope with DL/UL imbalanee, but only necessary in irregular cases like tunn els. Hen ce, it would be preferable, from OAM aspects, that Alt.1 is adopted. However, as expressed by 2 and in particular 3, there are several con cer ns of man dat i

53、ng the UE to read n eighbour BCH:UE will have to read BCH for each cell detected;一 It can be battery con sum ing if the UE cannot decode BCH and repeatedly tries decodi ng for a detected cell;This problem is more evident if the BCH error rate does not match (is worse than) the cell detection perform

54、a nee;Note: This may be prevented to some extent by specifying UE behaviours, e.g., UE only reads BCH ifRSRP > threshold, and/or UE considers offset = 0 if it fails reading BCH after some attempts.UE impleme ntati on may be more complex.- UE may have to decide whether to read BCH or not upon dete

55、cting a cell, e.g., UE does not have to read for cells in dicated in NCL or if RSRP < threshold, etc.Larger BCH overhead to make it robust.- E.g., by reducing the coding rate, increasing Tx power, repetition, soft combining, etc.These aspects n eed to be discussed in RAN WGs, and comme nts from U

56、E ven dors are especially valuable in this respect. If these concerns are serious, LTE should certainly not support reading of neighbour BCH.3.3 Other aspects of read ing n eighbour BCHIn discussing the pros/cons of reading neighbour BCH, another aspect that might be worth considering is the CSG cel

57、l (home-eNB) sce nario. The requireme nts of CSG cells 6 state no tably:7. It shall be possible to minimise the quantity of measurements which UEs perform on CSG Cells, if the UE does not bel ong to the User Group of a specificCSG Cell.8. The mobility procedures shall allow a large n umber of (small) CSG Cells to be deployed within the coverage of e-UTRAN, UTRAN and GERAN macro-layer cells. Deployme nt of (additio nal) CSG Cells shall n ot require rec on figurati on of other eNodeB (E-UTRAN) or RNC (UTRAN) or BSS (GERAN).This could be easily achieved if the UE