Cell Outage Compensation in LTE Networks

Cell Outage Compensation in LTE Networks: Algorithms and Performance Assessment

M. Amirijoo1, L. Jorguseski2, R. Litjens2 and L. C. Schmelz3

1 Ericsson, Link?ping, Sweden,

2 TNO ICT, Delft, The Netherlands,

3 Nokia Siemens Networks, Munich, Germany

mehdi.amirijoo@https://www.360docs.net/doc/021989415.html,, {ljupco.jorguseski,remco.litjens}@tno.nl,

christoph.schmelz@https://www.360docs.net/doc/021989415.html,

Abstract — Cell outage compensation is a self-healing function and as such part of the Self-Organising Networks concept for mobile wireless networks. It aims at mitigating the degradation of coverage, capacity and service quality caused by a cell or site level outage. Upon detection of such an outage, cell outage compensation tunes a variety of control parameters, e.g., electrical antenna tilt and uplink target received power level, in cells surrounding the affected site, in order to minimise the outage-induced performance effects in accordance with the operator policy. In this paper we outline the cell outage management framework, propose concrete compensation algorithms and assess the achieved performance effects in various scenarios. The reported simulation results show that the proposed compensation algorithm is able to recover a significant percentage of the users that would otherwise be dropped, while still providing sufficient service quality in the compensating cells.

I.I NTRODUCTION

Outages in mobile cellular networks come in different degrees of significance and may have different reasons. The failure of a site, a sector, or a physical channel or signal can be caused by, e.g., hard- or software failures (radio board failure, channel processing implementation error, etc.), external failures of power supply or network connectivity, or erroneous configuration. The largely manual analysis to detect outages in today’s networks is rather suboptimal in terms of the involved response time, the incurred performance degradation and, potentially, revenue loss.

Cell Outage Management (COM) comprises mechanisms for both automated Cell Outage Detection (COD) and Cell Outage Compensation (COC), and is an integral part of the self-organising network concept in E-UTRAN [1][2], with the objective to enhance the network robustness and resilience without requiring manual intervention. Figure 1 depicts the different elements and workflow of COM in cellular networks. The depicted example is characterised by a site outage, whose pre-outage service area is indicated in red. A variety of measurements from different sources (user terminals, base stations, Operation & Maintenance centre) are fed to the COD algorithm, which determines whether, where and what type of outage has occurred, and triggers the COC algorithm as well as the operator’s maintenance department for possible manual repair. The COC algorithm translates its measurement input to compensation measures in terms of an adaptation of one or more control parameters in surrounding cells, in line with the operator-formulated policy regarding the trade-off of local and regional performance effects. The COC function is characterised by an iterative process of radio parameter adjustment and evaluation of the performance impact until convergence is reached. Refer to [11] for a more elaborate discussion of cell outage management.

Figure 1: Overview of cell outage management.

The issue of optimising coverage and capacity in wireless cellular networks has already been addressed with off-line optimisation approaches (e.g., [3], [4]). For COC, however, it is important to develop methods that adjust multiple involved parameters on-line and in real-time. A recent study [5] addresses the real-time automatic reconfiguration of the base stations neighbouring the outage area via rule-based and genetic algorithms. In another study [6], reinforcement learning and fuzzy logic are combined to optimize tilt settings. Using genetic algorithms and reinforcement learning result, however, in trials with potentially negative impact, which may not be acceptable in real networks. Further, the approach in [6] requires fuzzification of input and definition of fuzzy rule sets, which may be difficult to realize. In this paper we present a novel approach for COC in LTE (Long Term Evolution) networks that meet the shortcomings of previous work.

The paper is structured as follows. Section II provides the key elements of COC. In Section III the COC algorithm is described. The assessment approach is outlined in Section IV, followed by the simulation results in Section V. Section VI ends this paper with some concluding remarks.

II.C ELL O UTAGE C OMPENSATION C ONCEPT

In this section we describe the different components related to COC in greater detail, considering control parameters, measurements and the operator policy.

In general all radio parameters that somehow influence coverage, capacity and spatial properties of service quality are candidate control parameters for COC. Based on [14] where

Measurements

Control Parameters, e.g., tilt and P

e.g. Q

978-1-4244-8331-0/11/$26.00 ?2011 IEEE

the COC potential of several control parameters was assessed in different scenarios, the target received power density P0 of the Physical Uplink Shared Channel (PUSCH) and the antenna tilt are selected as most effective COC parameters. The uplink power level for PUSCH transmission is controlled by the uplink power control. The open loop part of the power control derives the uplink power level from, among others, a target received power density P0 (per Physical Resource Block, PRB). Decreasing P0 generally results in lower power levels for PUSCH transmissions, and consequently lower UL inter-cell interference levels. Hence we have on one hand extended UL coverage due to reduced inter-cell interference per PRB, while on the other hand the per-user throughputs are decreased due to reduced per-PRB channel rates, i.e., lower Modulation and Coding Scheme (MCS) and, consequently, a higher PRB utilisation.

Modern antennas are capable of electrical adaptation of both the orientation of the main antenna lobe and the antenna pattern, e.g. via remote electrical tilt or beam forming techniques. Adjusting the antenna tilt shapes the cell footprint. For example, decreasing the antenna downtilt (up tilting) extends the footprint further away from the base station location, which can be used for providing coverage in the area affected by a neighbour cell outage.

The measurements used for COC may be obtained from the eNodeBs (e.g., cell loads, radio link/handover failure statistics, interference levels, etc.), UEs (e.g., reference signal received power, link failure reports, etc.) and O&M system providing quality-based KPIs, e.g. the 10th UL (DL) throughput percentile Q UL(Q DL). Note that if UE measurements can be combined with positioning information different types of coverage maps can be generated and used for more advanced COC algorithms. This is left for further study and not addressed in the remaining sections.

The operator policy is the key differentiation factor among the network operators. These policies define the trade-off an operator is making between e.g. investment and operational costs versus network performance, as well as between the different network performance aspects (coverage, capacity, quality). An example of an operator policy that is used in this study for COC is to maximise coverage given minimum constraints Q T,UL and Q T,DL on the 10th UL and DL throughput percentiles, respectively (i.e. Q UL≥ Q T,UL and Q DL≥ Q T,DL). An operator policy may depend on location, time as well as on the operational state of a cell, e.g. whether it is in ‘normal’ state or involved in the compensation of an outage in its vicinity.

III.A LGORITHMS

The extensive controllability study presented in [14] revealed that by altering P0 and tilt we can effectively trade service quality for enhanced coverage. This potential is utilised in the following compensation algorithms, which are executed in a stand-alone manner in each compensating cell.

Considering a P0-based COC algorithm, the UL and DL quality in each cell are continuously measured, whereby P0 is decreased under the constraints that

min

,0

,

,

,

,P

P

Q

Q

Q

Q

UL

T

UL

DL

T

DL

≥

≥

≥

where P0,min is the minimum tolerable P0 (see below). Clearly, there are several ways of setting P0 such that the above-specified constraint is satisfied. Key considerations for computing appropriate P0 include (i) the observed UL and DL quality over a time windo w, i.e., Q DL and Q UL respectively, (ii) outage quality targets Q T,DL and Q T,UL, and (iii) current and historical P0 values.

The particular solution studied in this paper is as follows. The basic principle is that if the experienced service quality exceeds the required level, this slack is exploited to improve coverage, in line with the P0impact noted in Section II. More specifically, when Q DL and Q UL exceed their respective outage quality targets Q T,DL and Q T,DL, then P0 is reduced in order to increase coverage. Otherwise, P0 is increased. This is summarised in Figure 2, where ΔP0 is the step size and P0,max denotes the maximum allowable P0.

Figure 2: Proposed P0-based COC algorithm.

In general, the parameter P0,min can be determined such that, e.g., the UL SINR is higher or equal to a threshold which supports the lowest MCS. Let SINR min = -6.5 dB be the minimum tolerable UL SINR, based on the lowest MCS. The received signal power S at the base station must thus satisfy)

(

max

min

I

N

SINR

S+

×

≥in linear scale, where I max is the maximum UL inter-cell interference that can be observed. Consequently, this gives that )

(

max

min

I

N

SINR

P+

×

≥. The parameter I max is based on the maximum measured UL inter-cell interference during the last X minutes/hours and includes a margin for unforeseen increases in inter-cell interference.

An algorithm similar to the P0-based compensation algorithm, has also been developed for antenna tilt. This algorithm reacts to DL and UL quality in the own cells and UL inter-cell interference and tunes the antenna tilt such that the DL and UL qualities satisfy requirements and significant UL inter-cell interference is avoided. If Q DL≥Q T,DL and Q UL≥Q T,UL then the antenna is uptilted under the condition that the UL inter-cell interference does not increase by a certain threshold. Although the simulation results of the tilt-tuning algorithm are promising and sometimes better than the P0-tuning algorithm, more care needs to be taken when deploying such functionality compared to the P0-tuning algorithm. This is due to the intricacy of real networks when it comes to, e.g., propagation, location of base stations, and direction of antennas, which brings additional complexity that needs to be handled when, e.g., uptilting an antenna may result in a loss of users close to the base station, which needs to be balanced

with users saved in the outage area. We have developed an

initial solution for automatic tilt tuning and present the preliminary results in Section V.

IV. A SSESSMENT A PPROACH The performance enhancement achieved by the proposed COC algorithms is assessed for a few data-only scenarios, which are characterised by distinct inter-site distance and load settings. The choice is based on the following arguments. In a sparse, coverage-driven network layout, little potential is likely to exist for compensating outage-induced performance

loss, while in a dense, capacity-driven network layout, this potential is much higher. The latter is particularly true when traffic loads are relatively low. Detailed specifications of the considered coverage- and capacity-driven layouts and the associated traffic loads, are given in Section V.

Although the COC algorithms operate independently in

each compensating cell, in the assessment approach a

distinction is made between two distinct compensation cell groups (see Figure 3). Cell group 1 (green arrows) comprises the three adjacent cells directed towards the outage area, while cell group 2 (blue arrows) comprises the six cells directed sideways. Separate assessments address the case where only group 1 contributes to the compensation, or both groups 1 & 2.

In practice, the set of compensating cells can be determined

either on- or off-line, based on e.g. predictions, simulations or measurements (handover rates, RSRP/RSRQ 1 reports, etc).

N loss

# served users

Figure 3: Compensating cells (l) and coverage metrics (r).

The objective of the assessment is to determine the degree of compensation that is attainable by control parameter (P 0 and tilt) adaptation in different scenarios, as well as its sensitivity w.r.t. the number of compensating cells. We hereby concentrate on the coverage gains and the quality sacrifice made in the compensating cells to achieve these gains.

The achieved coverage is expressed by either the relative (RCF ) or absolute compensation factor (ACF ):

,,nom

save loss save L N

ACF N N RCF ≡≡

where N loss is the total number of users that loose coverage due to the outage, N save is the total number of users that are recovered due to the COC actions, and L nom is the nominal (or average) number of users served by a cell (see also Figure 32).

1

Reference Signal Received Power/Quality.

2 Since the vertical axis (‘# served users’) in Figure

3 covers the entire affected area, while L nom is a cell-level metric, L nom is not illustrated in the figure.

As such, ACF gives the number of saved users relative to the

average number of served users in each cell. RFC and AFC are computed for the whole simulation area and are formed

during the steady-state (SS), where the results have converged

(during the final iterations). Herein, a user is ‘covered’ if three conditions are satisfied: (i) the Reference Signal (RS) SINR exceeds -6 dB; (ii) the RSRP exceeds -127 dBm; and (iii) the experienced SINR on the DL and UL is high enough to support the lowest MCS, i.e. SINR ≥ -6.5 dB.

The service quality is expressed by the 10th user throughput percentile in up- and downlink, i.e., Q UL and Q DL . A distinction is made between the post-outage/pre-compensation quality Q pre-comp and the post-compensation service quality Q comp . Service qualities Q pre-comp and Q comp are always

computed over the first tier of neighbours (group 1 and 2). V. S IMULATION R ESULTS The sensitivity analysis is performed using a Monte Carlo-based LTE network simulator. We consider a hexagonal layout of 19×3 cells. Key system parameters are listed in Table 1 (largely based on [12]). Table 1: Key system parameters. Capacity-driven layout Coverage-driven layout Inter-site distance 500 m 2200 m

Antenna downtilt 15o

5o

System bandwidth

10 MHz Maximum base station output power 46 dBm Average RS power 33 dBm Maximum UE output power 25 dBm

Path loss 128.1 + 37.6 log 10 r , with r in km Shadowing σ = 8 dB, inter-site correlation of ?, decorrel.

distance = inter-site distance / 15

Antenna model 3GPP 3D model

Noise level

-199 dBW/Hz in DL, -195 dBW/Hz in UL Service

Generic elastic data service with a requested throughput of 1 Mb/s (DL) & 250 kb/s (UL)

The considered scenarios are specified by the inter-site distance and the traffic load. The coverage-driven network layout is characterised by an inter-site distance of 2200m and a default (pre-outage) antenna downtilt of 5o . For this layout we consider only a low traffic load of (on average) 1 UE per cell (with a traffic model as given in Table 1). The coverage-driven layout represents a noise-limited scenario. The capacity-driven network layout is characterised by an inter-site distance of 500m, a default antenna downtilt of 15o , and low, medium or high traffic loads of (on average) 1, 23 and 47 UEs per cell. We assume outage-based performance targets of Q T,DL = 128 kbps and Q T,UL = 64 kbps.

A. Performance Summary

In this section we present performance results for the P 0-tuning algorithm. Specifically, we assess the compensation gains under different traffic loads, and for cases whether the compensation group comprises either only cell group 1 or both cell groups 1 and 2.

Figure 4 shows RCF and ACF for different traffic loads and compensation groups. We see that the compensation potential in terms of coverage improves as the load decreases. One interesting observation is that RCF > 1 during medium and low load when both group 1 and 2 are included in the compensation, i.e. we have that N save > N loss, which may seem contradictory at first sight. This is due to the dimensioning and parameter tuning of the network in a pre-outage setting, where coverage and quality are balanced. Since the compensation algorithms are not aware of changes to coverage (the standards lack appropriate measurements enabling accurate coverage observation), they simply reduce quality in exchange for improvements in coverage. This can as such imply that coverage targets used in the network planning are even exceeded during an outage. Further, we see that RCF and ACF are higher when including cell groups 1 and 2 in the compensation procedure, compared to the case where only cell group 1 performs compensation. Including both groups 1 and 2 results in an even less inter-cell interference compared to when compensation is done only by cell group 1, which enables the cells in group 1 to ‘penetrate’ even more into the outage area.

Figure 4: RCF (a) and ACF (b) for capacity-driven layout for different loads and compensation groups.

Figure 5 shows the quality degradation when outage compensation is applied. In general, both UL and DL quality is reduced due to more users being served, which share a given amount of resources. In addition, the UL quality is reduced more than the DL quality, since reducing P0 results in lower SINRs and hence a lower spectral efficiency. Further, the quality degradation is most visible for high and medium load and can be as low as 50% of the quality without any compensation (Q pre-comp). The quality reduction does, however, not decrease during low load, since there are enough resources to meet the UEs’ bit rate requirements (see Table 1).

Figure 5: DL quality (a) and UL quality (b) for capacity-driven layout for different loads and compensation groups.

As can be seen in Figure 5, the quality degradation is mostly visible when including both group 1 and 2 in the compensation (recall that Q pre-comp and Q comp are computed over group 1 and 2). When including both group 1 and 2 in the compensation, more cells suffer from quality degradation and, as such, we observe a greater quality reduction. This is an additional trade-off that needs to be considered, i.e., how many neighbouring cells that suffer from quality degradation and on the other hand contribute to increasing the coverage. We have also analysed the results of the coverage-driven layout, which show that the compensation potential is less for a coverage-driven layout, i.e., the ACF is lower compared to the capacity-driven layout. Further, we have no quality degradation in DL and a very small degradation in the UL. This is due to the low load and the fact that the network is underutilised, which is typically the case for networks dimensioned for coverage.

The results from tilt tuning show that similar or better performance can be expected, depending on the load situation.

In particular, higher compensation potential exists during

medium and low load when employing tilt-tuning. B. Temporal Behaviour

In this section we study the temporal behaviour of the P 0-tuning algorithm in a capacity-driven layout during high load.

Figure 6: (a) P

0 of the compensating cells, where max, mean and min is computed over the compensating cells (b) Mean of the DL and UL quality metrics computed over the compensating cells.

Figure 7: Snapshot of outage situation with largely converged compensation (iteration > 200).

An outage occurs at iteration 50, resulting in a reduction of number of served UEs and the creation of coverage holes, as shown in Figure 7. Compensation starts at iteration 100, whereby P 0 is gradually decreased, resulting in a reduction of mainly UL quality and in an increase in coverage and number of served users.

VI. C ONCLUDING R EMARKS In conclusion, an overview of cell outage management has been presented as a self-organising functionality in LTE

networks, followed by specific algorithmic solutions and evaluations to cell outage compensation. In the presented

algorithm, the parameters of cells neighbouring a failed cell/site can be automatically adjusted such that the coverage is maximised given constraints on quality defined in terms of

cell-edge user throughout. Depending on the considered scenario in terms of e.g., traffic loading, simulation results show the majority of the users can be recovered at the cost of an acceptably reduced quality in the compensating cells.

Key challenges that remain are, for example, accurate real-time coverage estimations, which enable a richer and more powerful operator policy where requirements on coverage can

be explicitly expressed and monitored. More specifically, if

appropriate measurements that enable accurate observation of coverage can be developed then this would facilitate the

inclusion of a coverage component in the operator policy and

in the algorithms. This would then result in a better match between coverage and quality balancing in network

dimensioning and planning, and during compensation.

A CKNOWLEDGMENT

The presented work was carried out within the EU-sponsored FP7 SOCRATES project [13].

R EFERENCES

[1] NGMN, ‘Use cases related to self organising networks. Overall

Description’, 2007.

[2] 3GPP TS 32.541, ‘Self-Healing OAM; Concepts and Requirements’,

v1.2.0, 2010.

[3] I. Siomina, P. Varbrand and D. Yuan, ‘Automated optimisation of

service coverage and base station antenna configuration in UMTS networks ’, Wireless Communications Magazine , vol. 13, no. 6, 2006. [4] D. Fagen, P. Vicharelli and J. Weitzen, ‘Automated coverage

optimisation in wireless networks ’, VTC ‘06, Montreal, Canada, 2006. [5] E3, ‘Simulation based recommendations for DSA and self-management’, FP7 E3 project ICT-2007-216248, 2009.

[6] R. Razavi, S. Klein, H. Claussen, ‘Self-Optimization of Capacity and

Coverage in LTE Networks Using a Fuzzy Reinforcement Learning Approach ’, PIMRC ’10, Istanbul, Turkey, 2010

[7] 3GPP TS 36.211, ‘Physical Channels and Modulation’, v8.9.0, 2009. [8] A. Simonsson and A. Furuskar, ‘Uplink power control in LTE –

Overview and performance ’, VTC ’08, Calgary, Canada, 2008. [9] 3GPP TS 36.213, ‘Physical layer procedures’, v8.8.0, 2009.

[10] J. Niemel? and J. Lempi?inen, ‘Impact of mechanical antenna downtilt

on performance of W-CDMA cellular networks ’, VTC ‘04, Milan, Italy, 2004.

[11] M. Amirijoo, L. Jorguseski, T. Kürner, R. Litjens, M. Neuland, L.C.

Schmelz and U. Türke, ‘Cell Outage Management in LTE Networks ’, ISWCS ’09, Siena, Italy, 2009.

[12] 3GPP TR 36.814, ‘Further advancements for E-UTRA – Physical layer

aspects’, v1.0.1, 2009.

[13] SOCRATES project, www.fp7-socrates.eu , 2010.

[14] M. Amirijoo, L. Jorguseski, R. Litjens and R. Nascimento,

‘Effectiveness of Cell Outage Compensation in LTE Networks ’, CCNC ’11, Las Vegas, USA, 2011.

Failed Cell Compensating Cell Operational Cell Covered Area Coverage Hole

Compensated Coverage Hole