Improved Mobility Performance in LTE Co-Channel ... - IEEE Xplore

GC'12 Workshop: The 7th IEEE International Workshop on Heterogeneous, Multi-Hop, Wireless and Mobile Networks

Improved Mobility Performance in LTE Co-Channel HetNets Through Speed Differentiated Enhancements Simone Barbera

Per Henrik Michaelsen, Mikko Säily*, Klaus Pedersen

Aalborg University Aalborg, Denmark E-mail: [email protected]

Nokia Siemens Networks Aalborg, Denmark and *Espoo, Finland E-mail: [email protected] in small cells. Among others, it has been concluded that radio link failures are relatively high for co-channel pico outbound handovers for high speed UEs [11], which tend also to have undesirable short-time-of-stay in pico cells, generating large amount of signaling overhead from the many experienced pico related handovers. This paper is focused on developing optimized solutions covering both low and high speed users in LTE HetNet cochannel scenarios with macro and pico cells. As a detailed mathematical formulation and derivation of such solutions is rather challenging, so we are mainly taking a heuristic approach in this study, followed by evaluation of the proposed solutions by means of extensive system level simulations. First, an enhanced autonomous UE mobility state estimator is proposed. Depending on the UE mobility state, solutions are derived for minimizing high speed UEs on pico layer, while trying to maximize the offload of low speed UEs. The proposed solutions are mainly UE autonomous, meaning that they require minimum network assistance and related signaling overhead. Performance results confirm that the proposed solutions offer attractive benefits.

Abstract—This paper analyzes the mobility performance of LTE (Long Term Evolution) co-channel heterogeneous networks (HetNet) with macro and pico cells. Improved methods for differentiating offload and mobility robustness as a function of the UE (User Equipment) mobility are proposed. The suggested solution comprises two key elements, namely enhanced UE MSE (Mobility State Estimation), as well as optimized methods such that high speed users are primarily kept at the macro layer, while the offload to pico cells for low speed users is maximized. The proposed methods are designed as UE autonomous solutions, requiring minimum assistance and signaling from the network. Extensive system level simulations are used to quantify the benefits. Results confirm that the proposed solutions offer improvements in several mobility key performance indicators such as radio link failure, number of handovers, offload to pico layer. Keywords-LTE, Small Cells, Heterogeneous Networks, Mobility

I.

INTRODUCTION

Numerous studies show increasing demands for higher broadband capacity of cellular networks, which calls for addition of small cells to supplement traditional macro cells installations [1][2]. Such network topologies are commonly referred to as heterogeneous networks or multi-layer networks. LTE HetNet scenarios have recently attracted lots of attention in both industry and academia with an exhaustive number of public studies, in particular concerning interference management challenges [3]. Another important aspect of LTE HetNet is the mobility performance, as addressed in this study. The mobility framework for LTE was originally developed and analyzed by 3GPP (3rd Generation Partnership Project) for homogeneous macro-only networks, and was therefore not explicitly optimized for HetNet scenarios. The focus is therefore to analyze the performance of standard LTE mobility mechanisms for HetNet scenarios, and to propose corresponding improvements where needed. The open literature already counts several LTE mobility studies. As an example, LTE mobility performance for co-channel HetNet cases was studied in [4], where it was shown that adapting the handover timing to the user speed improved the mobility performance. Studies on mobility state estimation were presented in [5] and [6] for macro-only networks. Improvements for mobility state estimation were proposed in [7] by explicitly taking into the account the cell types. An ongoing study on LTE HetNet mobility performance and related improvements in 3GPP, namely the Release 11 HetNet Mobility Study Item [8][9][10], has concluded that one of the main identified mobility challenges relates to fast moving UEs

978-1-4673-4941-3/12/$31.00 ©2012 IEEE

The rest of the paper is organized as follows: Section II describes the system model and outlines the addressed problems. Section III contains the proposed solutions, while Section IV includes corresponding performance. Finally, concluding remarks are provided in Section V. II.

SYSTEM MODEL AND PROBLEM FORMULATION

A. Baseline Assumptions The network topology consists of a regular hexagonal grid of three-sector macro cells with pico cells deployed either to provide higher hotspot capacity or better coverage at locations with modest macro-layer performance. Co-channel deployment is assumed, meaning that the macro and pico eNBs (EvolvedUTRAN Node B) share the same bandwidth. Handovers are triggered by the A3 event [12], when the RSRP (Reference Signal Received Power) of a neighboring cell is an offset better than the serving for a certain time known as the TTT (Time To Trigger). The analysis includes the measurements Layer-3 filtering effect. Similarly, handover delays (inter-eNB and eNB-to-UE signaling) are taken into account [13]. Current specifications provide a mechanism for UEs to estimate their mobility, called Mobility State Estimation (MSE). MSE is defined in [12] for Connected Mode UEs and in [14] for Idle Mode UEs. This study focuses on Connected Mode cases only. Three mobility states are defined: normal, medium and high. The UE is by default in the normal state.

426

The mobility state is updated if one of the following conditions is fulfilled:

1HO > NHO_H during THOmax

NHO_M1HO < NHO_H during THOmax

where NHO is the number of past experienced handovers (HOs) by the UE. The values of parameters NHO_M and NHO_H define the thresholds for medium and high mobility states. Parameter THOmax expresses the time window in which the counts are evaluated. The UE does not count ping-pong handovers between the same two cells. The UEs fulfilling condition (1) update their status to high mobility; the UEs fulfilling condition (2) update their status to medium mobility. It is possible for the network to configure the UE such that the TTT is scaled depending on its mobility state. Here the principle is to down-scale the TTT for high mobility state to ensure fast handovers for high-speed users.

Figure 1. A possible set of weights to be used in the handover count

B. Gray-listing Solution The Gray-listing solution aims at limiting fast moving users access to pico cells unless needed for interference reasons. A list of conditionally restricted cells is signaled to the user by e.g. informing the UEs of the pico cells Physical Cell Identity (PCI). All UEs perform Radio Resource Management (RRM) measurements on Gray-listed cells for mobility purposes, according to the current LTE specifications. A UE in high mobility state does generally not report measurement events, e.g. A3, for a Gray-listed cell, so handover is not triggered. The cell is still measured for the evaluation of the escape mechanism. Handover is triggered by poor radio conditions towards the serving cell, which is detected by RSRQ (Reference Signal Received Quality) being below a given threshold, while the RSRQ on the Gray-listed cell is acceptable for access, i.e. greater than another given threshold. The solution is on Fig. 2, where a UE in high mobility state moving along the indicated path reports Pico#1 only when the serving cell RSRQ goes below Threshold1, and the Pico#1 RSRQ is above threshold2. Later Pico#2 is not reported, since the RSRQ on the serving cell stays high. Assuming that Pico#3 is deployed for coverage reasons, thus bridging the two macro cells, then it is simply not included in the Gray-list, and it is accessed based on ordinary measurement reporting. The benefit of this solution is that it is mainly UE autonomous. Threshold1 should be set so it corresponds to the RSRQ value at which the lowest acceptable throughput for the UE can be achieved.

B. Problem Formulation and Objectives The main problem addressed in this study is how to develop a mechanism where high speed UEs are primarily kept at the macro layer while still trying to maximize the offload of other UEs to pico cells. By doing this, the identified mobility challenges for high speed UEs and small cells are eliminated. However, as this study considers co-channel scenarios, not all high speed UEs can avoid HOs to pico cells. UEs that experience too severe interference from a pico cell will eventually have no other option than being handed over to the pico to avoid radio link failures and dropped calls. Thus, there is need for an “escape” mechanism to still allow high speed UEs to perform handover to pico cells if absolutely necessary. The current LTE mobility specifications fall short of offering such mechanisms, since not taking the characteristics of different cell types into account, e.g. into the MSE algorithm, meaning low correlation between the actual UE speed and its mobility state for HetNet scenarios. One of the objectives is therefore to propose a more suitable and generic MSE method working also for HetNet scenarios. Secondly, while the current LTE specifications offer mechanisms for accelerating the HOs for UEs in high mobility state by down-scaling the TTT, we seek also a mechanism for avoiding high UEs to access pico cells. The overall objective is to develop UE autonomous solutions addressing the aforementioned problems with only minimum network assistance and signaling support. III.

1 for a macro-to-macro, 0.45 for a macro-to-pico, 0.25 for a pico-to-macro and 0.1 for a pico-to-pico. The suggested MSE weights relate the cells type coverage, and therefore depend on the effective radiated power for the considered types. In a real network, the MSE weights could be subject to further fine tuning by using Self Optimization Network (SON) techniques.

PROPOSED MOBILITY ENHANCEMENTS

A. Enhanced Mobility State Estimation The proposed improvement to the UE MSE algorithm is based on incrementing the count NHO with different values depending on the serving and target cell characteristics. Due to the smaller coverage area of the pico cells, it makes sense to give less weight to the pico-related events, such that a reasonable correlation between the MSE and the user speed is obtained. As the UE has no explicit knowledge of the cell type, it is simply proposed that the network signals the MSE count increase as part of the handover command, which is anyway signaled to UEs after the network receives the A3 event from the terminal. Weights producing good MSE results are (Fig. 1):

Figure 2. Sketch of the Gray-listing principle

427

TABLE II.

C. Cell Dependent TTT Scaling According to the current 3GPP standard, the network can specify a TTT for measurement events. TTT is independent on the target cell towards which the trigger is evaluated [12]. It is now considered to extend this, such that multiple values of TTT may be specified, each applying to a specified set of target cells. It is also assumed that a UE entering a cell is configured by a measurement setup with specific TTT values to this cell. As a special case it is considered to specify TTT dependent on the types of cells. In a given cell the UE is configured by a pair of TTT values applying to a list of macro and pico cells, which achieves dependency on the target cell type. Two pairs of such TTT values is in use, one in macro cells and another in pico cells, which achieves dependency on the serving cell. It is thus possible to specify values of TTT that depend on the types of both the serving and target cell. It is also assumed that the MSE based scaling of TTT is extended, such that a scaling factor greater than one may be specified, thus up-scaling of TTT. Current specifications only allow for down-scaling of TTT [14]. An example of the total number of TTT values that may thus be specified is shown in Table I. Here the general value is 256 ms applied in normal MSE in any case. For handover towards macro, down-scaling is applied to expedite handover at medium and high MSE. For macro-to-pico handover no scaling is applied at medium MSE, and up-scaling is applied at high MSE, to lower the likelihood of a fast moving UE to trigger towards a pico cell. This thus achieves an implicit way for avoiding fast moving UEs access to pico cells. This is not beneficial in the pico-to-pico case, so here down-scaling is applied, same as for handover towards macro. TABLE I.

•

Value 10 MHz, 2 GHz

Shadowing Standard Deviation

8 dB Macro, 10 dB Pico

Shadowing Correlation Distance

50 m Macro, 13 m Pico

Base Station TX Power

46 dBm Macro, 30 dBm Pico

Macro Cells Path-Loss

128.1 + 37.6 log10 (R)

Pico Cells Path-Loss

140.7 + 36.7 log10 (R)

Number of Users / Sims Time

30 per Macro-cell (21 Macros) / 200 s

RSRP Zero Mean Gaussian Measurement Error

1 dB St. Dev.

L3 Filtering Factor

4

Handover Delay (Prep + Exec)

0.15 s

MSE THOmax, NHO_M, NHO_H

30 s, 2, 4

A3 offset

3 dB

Default TTT

256 ms

B. Performance In order to quantize the impacts of the pico cell density on the MSE, distributions of the the MSE handover count within the evaluation window are collected, for different pico densities and UE speeds. The distributions obtained for the current MSE (no weights applied [12][14]) are shown in Fig.3.

Macro to Macro

Macro to Pico

Pico to Macro

Pico to Pico

Normal MSE Medium MSE High MSE

256 128 128

256 256 480

256 128 128

256 128 128

PERFORMANCE ANALYSIS

A. Simulation methodology Dynamic system level simulations are conducted to assess the mobility performance. The simulation methodology is according to the 3GPP guidelines, with default parameters as summarized in Table II. The simulated network consists of a regular grid of three-sector macro base stations, with a number of pico cells at random locations within each macro cell. The users start at random locations, and move in random directions at constant speed along straight lines. Only Connected Mode is simulated, by applying a simple full buffer traffic model. Cases with different number of pico cells per macro cell area, and different UE velocities are considered. The following main KPIs are extracted from the simulations: • •

Parameter Bandwidth and Frequency

SERVING TARGET TTT

TTT [ms]

IV.

SIMULATION SETTINGS

Figure 3. CDF of the MSE counts within THOmax. The numeric labels on each curve indicate the UE speed in km/h

The stepwise curves are due to the integer MSE counts. The counter values increase approximately proportional with the UE speed, except when limited to zero counts at low speed. In a network with homogeneous cell density the higher rate of handover at higher cell density is addressed by applying a correspondingly higher MSE threshold. This way one can achieve approximately the same distribution of users on mobility state, independent on the cell density. This is not feasible in the heterogeneous network due to the varied local density of pico cells, and the MSE threshold must be the same across cells, due to the MSE count being collected across cells.

Handovers (HOs) per user per hour; Off-Loading to Pico, measured as the percentage of users served by pico cells; Radio Link Failure (RLF) events, declared when the downlink UE SINR has been below -8dB (Qout) and stayed below -6dB (Qin) for the duration of the T310 time-window, configured to one second.

428

The local environment has a varying number of pico cells, but it appears that by applying the proper weights one can achieve MSE counter distributions which are almost independent on the pico cell density. The Fig. 4 shows the same distributions as on Fig. 3, only with a weighted MSE counter. The curves are smoother due to the fractional increments for pico-related HOs, and the curves are now aligned. For comparison Fig. 4 shows the distributions without pico cells as well, which is not dependent on the weighting. It is observed from other simulations that the sensitivity on the weights is not very high.

network is providing the increment by which the UE updates the MSE count. The alternative would be a complex UE procedure. The proposed solution requires minimum of signaling, has low UE complexity, and allows for future optimization without changes to the 3GPP standard.

Figure 6. User trajectory through (a) the center of the macro hexagons and (b) the macro cell borders

Figure 4. CDF of the MSE counts within THOmax for the chosen weights [1 0.45 0.25 0.1]

Figs. 7-9 compare different cases according to Table III. For Gray-listing the RSRQ thresholds are set to -20 dB and -12 dB, corresponding to SINR of -8 dB (RLF threshold) and 0 dB.

With the MSE thresholds at 2 and 4, Fig. 4 shows the proportion of times that a given mobility state is selected dependent on the UE speed. This is summarized on Fig. 5. km/h 3 30 60 120

Normal 100% 80% 30% -

Medium 20% 65% 40%

TABLE III. MSE Count Weighting

High 5% 60%

ReferenceCase enhanced-MSE GraylistScaling TTTservingTarget

Figure 5. Mobility state distribution dependent on UE speed for MSE thresholds 2 and 4, and weighting as in Fig. 4

No Yes Yes Yes

SIMULATED CASES TTT=[256,256,128]ms for normal, medium and high mobility states, respectively No Yes Yes No

Gray List

TTT as in Table I

No No Yes No

No No No Yes

Fig. 7 shows the number of handovers per user per hour at different velocities. The rate of handover is approximately proportional to the speed, as expected. With Gray-listing, the rate of handover increases less than other cases at high speed, since less UEs access a pico. This is not observed with celldependent TTT. Fig. 8 shows the off-loading in terms of the time average of the percentage of connections to pico cells out of all connections. The impact is clear with Gray-listing, but marginal with cell-dependent TTT. On Fig. 9, we first observe an expected reduction in RLF by applying MSE-based scaling of TTT. Gray-listing provides an added gain, whereas with cell-dependent TTT the RLF at high speed gets worse. The number of RLF per handover is less than ~2% for all UE speeds for cases with Gray-listing. Figs. 10-11 show comparison of the Gray-listing option with the current and enhanced MSE in terms of offload to pico cells and RLFs. Here the Gray-listing with enhanced MSE is resulting in the best performance, as the option with the current MSE results in too low offload to the picos for the low to medium UEs speed. This is a consequence of having too many UEs in the high mobility state when using the current MSE.

The enhanced MSE captures a combination of UE speed and the rate of handover. The above classification is such that “Normal” includes both UEs moving very slow, 3 km/h, and UEs that move at some speed (30-60 km/h) but happen to pass very few pico cells, hence achieving a very low MSE count. Similarly UEs will be classified higher when they happen to pass a lot of pico cells. The distinction is on the average time between mobility events. MSE is clearly affected by the UE trajectory. In the examples in Fig. 6, the UE is moving at 120 km/h, but in the first case is estimated in high mobility 77.6% of the simulation time, while in the second case is estimated in high mobility 81.5% of the simulation time. This is because in the second case the UE is moving through the macro cell borders. The simulations considered only two cell sizes, but shapes and sizes may vary much more in real networks. All of this indicates that the network should adapt the MSE weights to the topology of the network, e.g. city center versus sub-urban area, and the optimum values will differ from region to region. Such adaptation and optimization may be achieved when the

429

HOs per UE per Hour

1000

3 kmph 30 kmph 60 kmph 120 kmph

800 600 400 200 0

ReferenceCase

enhanced-MSE

GraylistScaling TTTservingTarget

Figure 7. HOs/UE/Hour – 4 Picos per Macro

Pico Users Percentage

20


15

10

V.

The main objective of this study was to propose additional UE autonomous enhancements to existing LTE handover procedures to improve co-channel HetNet performance for a wide range of UE speeds. Previous studies show that access to small cells should ideally be conditionally restricted to allow only low to medium mobility users. To achieve this goal, a stable enhanced mobility state estimator was developed, as well as corresponding mechanisms for the different mobility states. The developed Gray-listing solution achieves the goal of keeping fast moving users out of small cells, unless strictly required, and thereby reduces the amount of radio link failures and signaling overhead, while maintaining the desired offload to pico cells for slow moving users. An alternative approach with up-scaling of TTT for high mobility users into pico cells was also evaluated. Such solutions provide similar effect as the Gray-listing option, but with slightly lower performance for the considered parameter settings used in this study.

5

REFERENCES 0

ReferenceCase

enhanced-MSE


[1]

Figure 8. Pico Users Percentages – 4 Picos per Macro [2]

RLFs per UE per Hour

70


60 50

30

[4]

20 10 ReferenceCase

enhanced-MSE

[5]


Figure 9. RLFs/UE/Hour – 4 Picos per Macro 35 Pico Users Percentage

[3]

40

0

[6] 3 kmph 30 kmph 60 kmph 120 kmph

30 25

[7]

20 15

[8]

10 5 0

[9] ReferenceCase

RefCase

GLenhanced-MSE

GrayListEnhancedMSE

GLcurrent-MSE

GrayListCurrentMSE

[10]

Figure 10. Pico Users Percentages – 10 Picos per Macro 250 RLFs per UE per Hour

CONCLUSIONS


200 150

[11]

[12]

100

[13]

50 0

[14] ReferenceCase

RefCase

GLenhanced-MSE

GrayListEnhancedMSE

GLcurrent-MSE

GrayListCurrentMSE

Figure 11. RLFs/UE/Hour – 10 Picos per Macro

430

A. Damnjanovic, et.al., “A Survey on 3GPP Heterogeneous Networks”, in IEEE Wireless Communications, Pages 10-21, Vol. 18, Issue 3, June 2011. A. Damnjanovic, et. al., “UE’s role in LTE Advanced Heterogeneous Networks”, in IEEE Wireless Communications, Pages 164-176, Vol. 50, Issue 2, February 2012. D. Lopez-Perez, I. Guvenc, G. de la Roche, M. Kountouris, T.Q.S. Quek, and Jie Zhang, “Enhanced Intercell Interference Coordination Challenges in Heterogeneous Networks”, in IEEE Wireless Communications, Pages 22-30, Vol. 18, Issue 3, June 2011. Y. Lee, B. Shin, J. Lim, and D. Hong, “Effects of Time-to-Trigger Parameter on Handover Performance in SON-Based LTE Systems”, in APCC, Pages 492 – 496, 2010. J. Puttonen, N. Kolehmainen, T. Henttonen, and J. Kaikkonen, “On Idle Mode Mobility State Detection in Evolved UTRAN”, in ITNG, Pages 1195 – 1200, 2009. N. Kolehmainen, J. Puttonen, T. Henttonen, and J. Kaikkonen, “Performance of Idle Mode Mobility State Detection Schemes in Evolved UTRAN”, in ISWPC, Pages 584 – 588, 2010. Lei Yixue, and Zhang Yongsheng, “Enhanced Mobility State Detection based Mobility Optimization for Femto Cells in LTE and LTE-advanced Networks”, in IET ICCTA, Pages 341 – 345, Beijing, China, October 14-16, 2011. 3GPP RP-110709, “Revised WID on Study on for Hetnet Mobility Enhancements for LTE”, 3GPP TSG-RAN Meeting #52, May 2011. 3GPP Technical Report 36.814, “Evolved Universal Terrestrial Radio Access (EUTRA); further advancements for E-UTRA physical layer aspects”, V9.0.0 (2010-03). Available at www.3gpp.org. 3GPP Technical Report 36.839, “Evolved Universal Terrestrial Radio Access (EUTRA); mobility enhancements in heterogeneous networks”, V0.5.0 (2012-03). Available at www.3gpp.org. S. Barbera, P.H. Michaelsen, M. Säily, and K. Pedersen, "Mobility Performance of LTE Co-Channel Deployment of Macro and Pico Cells", in IEEE WCNC, Pages 2890 – 2895, Paris, France, April 01-04, 2012. 3GPP Technical Specification 36.331, “Evolved Universal Terrestrial Radio Access (E-UTRA); radio resource control (RRC); protocol specification”, V10.5.0 (2012-03). Available at www.3gpp.org. H. Holma, and A. Toskala, “LTE for UMTS, OFDMA and SC-FDMA Based Radio Access”, Wiley, 2009. 3GPP Technical Specification 36.304, “Evolved Universal Terrestrial Radio Access (E-UTRA); User Equipment (UE) procedures in idle mode”, V10.5.0 (2012-03). Available at www.3gpp.org.