Mobility Performance of LTE Co-Channel Deployment of ... - IEEE Xplore

2012 IEEE Wireless Communications and Networking Conference: Mobile and Wireless Networks

Mobility Performance of LTE Co-Channel Deployment of Macro and Pico Cells Simone Barbera

Per Henrik Michaelsen, Mikko Säily*, Klaus Pedersen

Aalborg University Aalborg, Denmark Email: [email protected]

Nokia Siemens Networks Aalborg, Denmark *Espoo, Finland Email: [email protected]

Abstract—This paper aims at analyzing the mobility performance in heterogeneous 3GPP (3rd Generation Partnership Project) Long Term Evolution (LTE) networks. The main objective is to analyze the behavior of LTE macro/pico co-channel networks with different mobility parameters, such as the “Time-ToTrigger” (TTT). For this purpose, a system simulator has been utilized to assess the performance. The system must have different settings depending on the user velocity and on which cell-layer user is being serviced. Additionally, we have considered scenarios with a mixture of users moving freely or constrained to a hotspot, and different pico cell deployments. Keywords-LTE, Mobility, Pico Cells, Heterogeneous Networks

I.

INTRODUCTION

LTE was first introduced in 3GPP Release 8 and has later evolved towards Release 9 and LTE-Advanced in Release 10, offering higher peak data rates, better average throughput and coverage [1][2]. Further improvements in terms of increased performance per area are possible by using a combination of macro cells and small cells – also known as heterogeneous networks (HetNet). Traditional cellular networks consist of base-stations with similar features, such as antennas and power levels. The macro base-stations are deployed in a regular grid to maximize the coverage and minimize the interference. Heterogeneous networks aim at overlaying the macro network with small cells; e.g. being deployed to improve coverage of areas that are problematic due to particular coverage conditions or in areas with higher traffic densities (also known as hotspots) [3]. Small cells come in several forms, ranging from micro cells and pico cells to user deployment femto cells with restricted access conditions. In this study we focus on the mobility performance in a LTE network of macro cells in a regular grid with pico cells deployed for outdoor coverage of irregularly located hotspots. The transmit powers for macro and pico cells are 46 dBm and 30 dBm respectively, corresponding to EIRP (Equivalent Isotropic Radiated Powers) of 60 dBm and 35 dBm respectively. All cells are deployed on the same 10 MHz bandwidth carrier, thus causing interference between macro and pico cells. The interference between the macro and the pico cells presents additional challenges to the timely execution of the handover (HO) process, when users are moving through the smaller pico cell coverage areas. LTE mobility performance for such co-channel HetNet cases has previously

978-1-4673-0437-5/12/$31.00 ©2012 Crown

been studied in [4], where it was found that especially the TTT setting is of great importance, and adapting the TTT value to the user velocity improved the mobility performance in terms of radio link failure and handover ping-pong (defined in Section II). In this study we further analyze how to optimize the TTT parameter for co-channel HetNet cases (i.e. we also consider layer-specific TTT settings). Based on basic mobility performance results for various TTT settings, we identify the main challenges regarding mobility, and suggest how to improve the performance. The analysis considers (1) the impact on mobility performance of the velocity of moving terminals and of the pico cells number, and (2) the mobility of users either moving freely or confined to hotspots. It is demonstrated that there is no single optimal setting of mobility parameters, hence more customized settings of parameters are required – calling for potential improvements compared to what the current LTE specifications allow. The study closely follows the ongoing 3GPP Release-11 study item concerning mobility for heterogeneous networks [5][6]. Thus, we apply the 3GPP agreed models and follow the general assumptions for mobility evaluation in LTE Release-11. The rest of the paper is organized as follows. Section II introduces the used simulation model, by defining the system parameters and explaining the deployment of heterogeneous networks with and without the hotspot model. Section III illustrates the network performance for free moving users and users concentrated on the hotspots. Section IV discusses the future work, and finally the conclusions are summarized in Section V. II.

SIMULATION MODEL

A. The Simulation Settings Dynamic system level simulations are conducted to assess the mobility performance. The general simulation methodology is according to the 3GPP guidelines [5][6], using the parameters summarized in Table I. The basic network layout consists of a regular hexagonal grid of three sector macro eNBs (Evolved universal terrestrial radio access network Node B), complemented by a number of pico cells having omni directional antennas. The pico cells are placed randomly within each macro-cell area according to the definitions in [5], fulfilling minimum distance requirements between pico and macro, as well as between picos.

2863

Handover is triggered by the A3 event, see [7]. This happens when a neighboring cell RSRP (Reference Signal Received Power) becomes better by an offset (A3 offset) than the serving cell, and this condition is fulfilled for the duration of specified TTT. The RSRP measurement imperfections and the effect of Layer-3 filtering on those measurements are modeled accurately in the simulations, in line with the 3GPP specifications [7][8]. This study does not consider DRX (Discontinuous Reception and Transmission) [2]; this would further limit the UE (User Equipment) measurement opportunities. When an A3 event is reported by the UE, the handover process is initiated. This includes handover request (signaling from source to target eNB) and command (signaling from source eNB to the UE) which are modeled by additional delays, from the reception of the A3 event. The default handover parameters are listed in Table I; to be precise, we use the mobility set #1, defined by [9] as the default setting. TABLE I. DEFAULT SIMULATION SETTINGS

Shadowing Correlation Distance TX Power Path-Loss Macro Cells Path-Loss Pico Cells HO A3 Offset Users per Macro Cell RSRP Measurement Error L3 Filtering Factor HO Preparation + Execution Time

21 2, 4 or 10 500 Seconds 10 MHz 2 GHz 8 dB Macro Cells 10 dB Pico Cells 50 m Macro Cells 13 m Pico Cells 46 dBm Macro Cells 30 dBm Pico Cells 128.1 + 37.6 log10 (R) 140.7 + 36.7 log10 (R) 3 dB 30 1 dB 4 0.15 s

Hotspot user Fig. 1. Simulated UE movement models: free moving and hotspot

•

Ping-Pong (PP) events, defined as the cases in which the UE experiences handover out of cell and then back to the same cell within Ping-Pong-Time (1 sec);

•

Radio Link Failure (RLF) event, declared when SINR has been below -8dB (Qout) and stays below -6dB (Qin) for the duration of 1s (T310) [10];

•

HO Failure (HOF) event, declared if the RLF occurs after the TTT expires, and before the handover is completed, which means during the handover preparation or execution time.

The KPIs are collected separately per HO type, followed by calculation of the empirically experienced probability. Thus, the RLF probability for macro-2-pico handovers is calculated as the number of RLFs counted on macro-2-pico transitions, divided by the total number of macro-2-pico handovers. III.

MOBILITY PERFORMANCE

A. Free Moving Users only

The study considers two mobility models, free movement and hotspot movement. Each user moves according to one of these models, and is referred to as a free user or a hotspot user respectively. The users are initially placed randomly within the full simulation or circular hotspot areas respectively. All users move along straight lines with constant velocity. The free user velocity is varied, while hotspot users always move at 3 km/h. Wrap around is applied to all users, while hotspot users additionally bounce off the hotspot perimeter in a random direction back into the hotspot area; see Fig. 1. Since the pico cells are deployed for hotspot coverage, the pico location and hotspot centers are chosen equally. The pico coverage area is defined solely by the handover process, i.e. by path loss, and is not directly influenced by the hotspot area. The full buffer data traffic model is assumed for all users, which means that data is assumed available for whatever capacity is allocated to the user.

Free user

The following primary mobility KPIs (Key Performance Indicators) are used in order to quantify the performance:

When considering the case with solely free users, we observe trends in the rate of handover as a function of the pico density, i.e. HOs/user/hour for varying number of picos per macro cell, which is independent of the user velocity. Fig. 2 presents the results for 3 km/h. This shows the rate of macro to macro handover, and handovers involving pico. As expected, the fraction of handovers involving pico cells increases significantly as more picos are deployed. This indicates an increasing importance of robust pico related handover as the density of pico cells is increased. With more than 3 picos per macro cell, we observe a higher rate of pico related handover than macro to macro handover. 40 HOs/User/Hour

Number of Macro Cells Pico Cells per Macro Simulation Time Bandwidth Macro and Pico Frequency Shadowing Standard Deviation

B. Key Performance Indicators

Macro HOs Pico HOs

30 20 10 0

2

4 10 Pico Cells per Macro Fig. 2. ‘HOs’ vs ‘Picos per Macro’ at 3 km/h. ‘Pico HOs’ are defined as pico-related HOs (macro to pico, pico to macro, pico to pico)

In order to distinguish the performance of pico hand-in (macro to pico) and pico hand-out (pico to macro) we further divide handover events accordingly. Observe that pico to pico handovers are so rare that they have been excluded from the study. The considered types of handover are then macro to macro (MM), macro to pico (MP), and pico to macro (PM). Handover ping-pong is similarly divided into MMM, MPM, and PMP.

2864

Figs. 3-5 show the mobility KPIs divided per handover types for three simulations where all users are moving at 30, 60 and 120 km/h, and there are two picos per macro cell. Fig. 3 and Fig. 5 show a drastic increase in the failure rates (HOF and RLF) as the user velocity increases, which is due to the conservative setting of TTT to 480 ms that performs well in a macro only network. We observe that the outbound pico handover (PM) is the most problematic handover type.

moving close to a pico, we only observe MP RLF in cases where the distance to the serving macro is large, see Fig. 6. The situation is more severe for pico outbound handover, since the user is moving directly out of pico and into macro coverage, which implies a steadily declining signal to interference ratio. The situation is worst when the user is moving towards a macro, see Fig. 7.

TTT: 480ms Macro, 480ms Pico

HOF Percentage per HO

50 40

30 kmph 60 kmph 120 kmph

30 20 10 0

MM

MP

Fig. 6. Macro-to-Pico RLF; the user is connected to the macro number 8, experiences RLF and then connects to the pico number 33

PM

Fig. 3. HOF probability per HO type, assuming 2 picos per macro TTT: 480ms Macro, 480ms Pico

PP Percentage per HO

6 5


4 3 2

Fig. 7. Pico-to-Macro RLF; the user is connected to the pico number 56, experiences RLF and then connects to the macro number 18

1 0

MMM

PMP

The mobility set #1, providing acceptable results in a macro only network, is clearly too conservative when deploying small cells, in which case mobility performance is poor in all cases, except when user velocity is very low.

MPM

Fig. 4. PP probability per HO type, assuming 2 picos per macro

We now consider how the mobility KPIs behave when introducing a more aggressive handover by lowering TTT, possibly introducing different values for macro and pico cells. A lower TTT causes faster handover, hence lowering the likelihood of RLF, but a too aggressive handover process may cause a drastic increase to the rate of handover ping-pong. Among the possible TTT values (listed in [4]), 480 ms, 256 ms and 128 ms are considered. Figs. 8-10 show the RLF and PP performances by using dedicated TTT values (RLF counts all failures, hence includes HOF, which shows the same trend). The following learnings are extracted:


RLF Percentage per HO

60 50


40 30 20 10 0

MM

MP

-

Fig. 8 shows that failures in handover originated from macro cells may be lowered by a lower TTT. This also slightly lowers the failures of pico outbound handover, since the time of stay in the pico is often so short that any time save by a fast hand-in gives more time for the fast triggering of the following hand-out;

-

Even at 30 km/h, a macro TTT of 480 ms is too conservative; PM RLF is 23.5% at 30 km/h (Fig. 8);

PM

Fig. 5. RLF probability per HO type, assuming 2 picos per macro

A late handover due to long TTT relative to user velocity is less problematic for the pico inbound handover, since most users are passing only part of the cell, which means that interference to the macro signal is limited. When a user is

2865

The failure rate generally decreases by decreasing the TTT, and the best case considered is to use 128 ms for both macro and pico. However, the more aggressive handover will cause an increase into the handover ping-pong. Fig. 4 and Fig. 9 show that at 120 km/h the MPM ping-pong increases from 5,5% to 21,3%;

-

PP Percentage per HO RLF Percentage per HO

-

Considering an acceptance threshold of, say, 2% as total RLF percentage, we find the required TTT to achieve a link failure rate lower than the chosen threshold. This is the optimum value fulfilling an RLF requirement while minimizing the ping-pong, since ping-pong is increasing with decreasing TTT. Table II shows the results for 30 km/h and 120 km/h. We observe that the optimum TTT increases with cell size, and decreases with user velocity. The results suggest that an even lower value of TTT, say 64 ms, is even more optimal for the pico cells.

Macro TTT ms 480 128

Pico TTT ms 256 128



40 30 20 10 0

MM

MP

PM

PP Percentage per HO


Fig. 8. RLF probability per HO type, assuming 2 picos per macro TTT: 128ms Macro, 128ms Pico

6 4


MM

MP

PM


10 MMM

MP

PM


10

5

0


MMM

PMP

MPM

p

30

0

MM

40


20

0

This section presents the mobility performance in a network with each pico cell covering a hotspot of specified size, and a number of users confined to each hotspot. There are two pico cells per macro. To see the effects of the hotspots more clearly, we first consider (Fig. 11) statistics exclusively for hotspot users (15 users per hotspot), showing the rate of handover for the four possible handovers between macro and pico cells. The hotspot radius varies from 40 m to 200 m, and the case with 30 free users per macro is shown for comparison. The number of bounces of a user inside a hotspot, i.e. changes in the direction of movement, is inversely proportional to the hotspot radius, which accounts for the steep decrease in pico related handovers for increasing radius. For the case of free users only we record the distance to the serving cell when performing handover, and produce the distribution dependent on cell type, as shown in Fig. 12. This shows that the pico coverage range is increasing with speed, and less than 100 m in all cases. This accounts for observing very similar handover statistics as with free users only, when the hotspot radius is higher than 100 m.

2 0

10

B. Hotspot Users

Rate of handover [ event / user / hour ]


50

20


Fig. 10. RLF/PP probability per HO type, assuming 2 picos per macro

TABLE II. REQUIRED TTT FOR RLF < 2% User Speed (km/h) 30 120


30

PMP

MPM

40 m 60 m 80 m 100 m 150 m 200 m no hs

35 30 25 20 15 10 5 0

Macro-Macro

Macro-Pico

Pico-Macro

Pico-Pico

Fig. 11. Rate of handover dependency on hotspot size – 3 Km/h

Fig. 9. RLF/PP probability per HO type, assuming 2 picos per macro

2866

12

1

40 m 60 m 80 m 100 m 150 m 200 m

0.9

Ping-pong per handover [%]

0.8 0.7

CDF

0.6 0.5 3 km/h, macro 30 km/h, macro 60 km/h, macro 120 km/h, macro 3 km/h, pico 30 km/h, pico 60 km/h, pico 120 km/h, pico

0.4 0.3 0.2 0.1 0

0

50

100

150

200

250

300

10

8

6

4

2

0

350

3

Distance to serving BS [m]

30 60 Speed of free moving users [km/h]

120

Fig. 12. Distributions of the distance to serving eNB, sampled per TTI, for macro and pico and user moving freely at different velocities

Fig. 14. PP per handover dependency on user velocity and hotspot radius

We now consider the case of 2/3 of users being hotspot users (10 users/hotspot), and the rest being free users. Fig. 13 shows the rate of handover for varying free user velocity and hotspot radius. As derived above, the rate is increasing with user velocity, and decreasing with hotspot radius, which accounts for the trend between groups and within groups respectively.

We have included the 120 km/h for free users for information, and are not suggesting that this is realistic. Also the random path that a free user takes through a hotspot should be seen as an average over multiple cases where a road passes along a particular path.

Fig. 14 shows the ping-pong percentage, depending on UE velocity and hotspot radius; this is a combination of several trends. A user moving at low speed stays on the cell boundary for a long time and performs ping-pong due to fading variations, where at higher velocity the rate of ping-pong is increasing with the velocity. The inverse proportional relation between hotspot radius and the rate of bouncing inside the hotspot accounts for the steep decrease in rate of ping-pong for increasing hotspot radius. When the free users are moving at 3 km/h, the handover ping-pong is mainly caused by hotspot users. Here we observe a ping-pong per handover rate up to approximately 12%, which is likely to be acceptable. We are assuming outdoor coverage of an outdoor user, which is similar to a pico covering an outdoor fair, where people are moving in a confined space around the pico, while some users are passing nearby, possibly through the area by car. p

p

Handover [events / user / hour]

300 40 m 60 m 80 m 100 m 150 m 200 m

250

200

150

100

50

0

3

30

60

120

Speed of free moving users [km/h] Fig. 13. Handover rate dependency on speed and hotspot radius

From Figs. 2 and 13 we see that a hotspot user experiences up to twice the rate of handover than that of a free user. We observe that the area of coverage by an outdoor pico cell is very small, and that there are users connecting to the macro even when very close to the pico. This clearly depends on the assumed shadowing by the environment. For instance, in case of an indoor pico covering, say, a mall, the building penetration loss makes it unlikely that an indoor user connects to the outdoor macro. IV.

POSSIBLE FUTURE IMPROVEMENTS

Further studies of the dependency of the optimal setting of handover parameters separately for macro and pico cells, and dependent on user mobility are of interest to heterogeneous networks. This includes studies of the performance achieved by applying the standardized method of estimating a mobility state based on the rate of handover and the scaling TTT dependent on the mobility state [7]. It is of particular interest to find ways of lowering the amount of outbound handover from small cells that prove to be the most problematic, e.g. by keeping fast moving UEs from making handover towards small cells. The dependency of optimal setting on cell size suggests an improvement of the mobility state estimation to take the cell size and type into account. Also the increased rate of handover due to constrained movement in hotspots suggests a study of how to distinguish this type of movement from moving at higher velocity that also increases the rate of handover, and whether it is actually necessary to make this distinction. In general we suggest further studies on adapting per UE mobility settings in line with the principles of Self Optimized Network (SON) features for Mobility Robustness Optimization (MRO). Additional mobility enhancements are being studied for LTE Release-11, e.g. as mechanisms for avoiding fast moving UEs on small cells whenever this is desirable.

2867

V.

CONCLUSIONS

This article aimed at evaluating the mobility performance in heterogeneous networks with macro and pico cells on the same carrier frequency. The study is based on different deployment scenarios and mobility parameters. The first handover parameters that are found promising for macro-only scenarios are used for the considered case (co-channel deployment of both macro and pico). Applying the former approach is found to result in unacceptable performance in terms of link failures, especially for handovers from pico to macro. Considerably improved performance can be achieved by making the handovers faster for small cells, e.g. by lowering the value of the time to trigger. This lowers the link failure rate, while increasing the rate of handover ping-pong (short stay). The optimal value of TTT achieving the lowest failure rate while minimizing the handover ping-pong proves to be different for macro and pico, hence dependent on the cell size. When separating handover in macro to macro, macro to pico, etc. we observe that the most critical handover in terms of failure rate is the outbound handover from a pico cell. In the latter case the user moves into the macro cell, as opposed to the inbound case where the user is often moving through the small target cell, keeping its connection to the macro cell if the handover is not performed in time. This study assumes deployment of pico cells for outdoor coverage, so we also consider the impact from having user movement confined to a hotspot area around the pico. This shows an increase in the rates of both handover and ping-pong that declines towards the rates observed for free moving users, such that at hotspot radius above approximately 100 m there is a little difference. The changes to mobility KPIs when

confining users to move and bounce around in a small area are similar to what is observed for users moving at higher velocity in heterogeneous networks. REFERENCES [1]

A. Damnjanovic, J. Montojo, Yongbin Wei, Tingfang Ji, Tao Luo, M. Vajapeyam, Taesang Yoo, Osok Song, and D. Malladi, “A survey on 3GPP heterogeneous networks”, in IEEE Wireless Communications, Vol.18 , Issue 3, Pages: 10 – 21, 2011. [2] H. Holma, and A. Toskala, “LTE for UMTS, OFDMA and SC-FDMA Based Radio Access”, Wiley, 2009. [3] A. Khandekar, N. Bhushan, Ji Tingfang, and V. Vanghi, “LTEAdvanced: Heterogeneous Networks”, in European Wireless Conference (EW 2010), Pages: 978 – 982, San Diego (CA), USA, 12-15 April 2010. [4] 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 2010, Pages: 492 – 496, 2010. [5] 3GPP TS 36.814, “Evolved Universal Terrestrial Radio Access (EUTRA); further advancements for E-UTRA physical layer aspects”, V9.0.0 (2010-03). [6] 3GPP TS 36.839, “Evolved Universal Terrestrial Radio Access (EUTRA); mobility enhancements in heterogeneous networks”, V0.2.0 (2011-09). [7] 3GPP TS 36.331, “Evolved Universal Terrestrial Radio Access (EUTRA); radio resource control (RRC); protocol specification”, V10.0.0 (2010-12). [8] 3GPP TS 36.133, “Evolved Universal Terrestrial Radio Access (EUTRA); requirements for support of radio resource management”, V10.3.0 (2011-06). [9] 3GPP TR 36.839, “Evolved Universal Terrestrial Radio Access (EUTRA); mobility enhancements in heterogeneous networks”, V10.4.0 (2011-11). [10] P. Legg, G. Hui, and J. Johansson, “A Simulation Study of LTE IntraFrequency Handover Performance”, in VTC Fall 2010, 6-9 Sept. 2010.

2868