This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808292, IEEE
Communications Letters
1

Swapped Sectors Detection Based on

Mobility Statistics
Omar Kaddoura, Raquel Barco, Inmaculada Serrano, Juan J. Sánchez-Sánchez

Abstract—Among the multiple faults that might happen in a

cellular network, operators spend long time trying to identify
swapped sectors, that is, during the network rollout phase,
feeders from baseband units to antenna units may result crossed,
thus leading to swaps in service areas for different sectors
managed by the same radio equipment. In this paper, two
different methods are presented in order to automatically detect
antenna misconfigurations due to crossed feeders. Results show
that the proposed methods are valid for troubleshooting of
swapped sectors, thus saving operators time. Moreover, both (a) Pair case (b) Trio case
methods have been successfully validated on real networks.
Fig. 1: Swapped cases
Index Terms—Swapped sectors, mobility statistics, handover,
troubleshooting, self-healing, Self-Organising Networks
swaps are not detected. The only alternative to find those
undetected swapped sectors in a network is a costly process
I. I NTRODUCTION that implies to carry out drive testing and involves site visits.
WAPPED sector term is used to describe the problem that This paper proposes a novel methodology to detect swapped
S arises when sector radio jumpers are terminated to the
wrong antenna sector, or trunk fibers or jumpers are terminated
sectors in an automatic manner. Among the advantages of
using this new methodology are cost reduction, efficiency
to the incorrect sector radio position. For instance, when the improvement and the removal of human factor error.
feeder for sector A is connected to sector C and vice versa. In this paper, firstly, the problem will be formulated. Then,
When this happens, swapped sectors typically continue to two novel methods for detecting swapped sectors will be
provide good coverage. However, this might mask several proposed. Finally, results will be presented and conclusions
subjacent issues derived from the reversion of the parameters outlined.
associated to those misconfigured sectors. Some of the issues II. P ROBLEM FORMULATION
found in networks with swapped sectors are: greater degree of
interference, poor uplink signal strength, poor performance in Radio Frequency (RF) drive and walk testing [1] [2] are
handovers and poor performance in terms of capacity. widely used by operators in order to detect swapped sectors.
Fig. 1 shows how the coverage areas of some sectors are Those methods consist of the utilization of user stations in
swapped. The coverage areas for sectors A, B and C are order to obtain sectors’ RF fingerprint. Thus, once sectors’
represented as A’, B’ and C’, respectively. Fig. 1a corresponds serving areas are determined, it is possible to detect swapped
to a scenario where two sectors of the same site (B and C) are sectors according to Fig. 1. However, the main drawbacks
swapped (swapped pair) and, thus, their expected azimuths are of drive/walk testing are time of execution and associated
shifted. Fig. 1b corresponds to a scenario where three sectors costs since it involves a lot of equipment (UEs, Scanner) and
in the same site are rotated (rotated trio). operational expenses for processing logs and reporting [3]. On
In order to avoid those issues, operators define and enforce top of that, when drive/walk test is used to detect swapped
swapped sector prevention processes in order to identify and sectors, there is a need for multiple site visits during swapped
correct misconfigurations, if any, prior to unlock and set the sectors corrections. Usually, a three visit plan is performed:
sectors on-air. Those processes consist of a method that pre- drive test for detection, feeders correction and final drive test
checks the radio access system to decide whether the site has for verification.
been correctly built in relation to any probability of swap To overcome the necessity of performing drive/walk tests,
sectors. The final objective is to facilitate the reduction of a new method was published in [4]. The system described
the time period between the moment the site construction in this patent presents a new methodology to detect swapped
is complete and the moment there is an acceptance by the sectors based on interference measurements reported by UEs
operator, allowing for successful and timely site launches. in measurement report messages. This information can be
Although adherence to swapped sector prevention processes obtained from logs recorded by network elements. However,
minimizes return site visits and delays in declaring a site this method has several drawbacks. Firstly, due to the stochas-
construction complete, the processes are not faultless since tic property of the RF channel, RF measurements cannot
they depend on engineers expertise and, in some occasions, be considered as reliable enough. Secondly, processing L3
messages like measurement reports requires a high processing
capacity.

1089-7798 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808292, IEEE
Communications Letters
2

More recently, a newer method was described in [5]. This

patent presents a system capable of detecting swapped sectors
using network mobility statistics. The main advantage of this
solution is the low processing capacity required, since network
statistics are provided by the network as standardized in [6].
This letter proposes two novel methods to detect swapped
sectors based on the same principles than [5]. However, these
two methods enhance the detection since mobility statistics
are not used to discard neighbors but to weight them. Besides,
neighbors’ distance is now used as a relevant source of data.
(a) Known beamwidth (b) Unknown beamwidth
III. P ROPOSED METHOD Fig. 2: Solid angle calculation
This letter presents two methods for detecting cases where,
due to an error in the site connections, the coverage areas of
two or more sectors in that site are swapped. These methods
are based on the analysis of network mobility statistics. Among
others, these statistics consist of the number of handover at-
tempts performed between two sectors, being identified which
one is the source sector and which one the target sector.
Additionally, in order to apply these two methods, the
topology information of the network should be known. This
is, the spatial location (latitude and longitude coordinates)
of every sector and its neighbors as well as the expected (a) Mobility statistics for sector (b) Mobility statistics for sector
azimuth of the antennas and/or the antenna beamwidth. Both α β
methods are valid for any mobile communications network Fig. 3: Example of handover for two sectors in the same site
whose sectors are sectorized, e.g. GSM, UMTS or LTE. This
implies that the antenna patterns are not omnidirectional but f can be any function of distance. xloc , yloc and iloc are the
they are directive having a certain beamwidth. location coordinates of sectors x, y and i, respectively. #HO
is the number of handover attempts between source and target
A. Method I
sectors.
This method is based on the fact that handover attempts Due to the directionality of antenna patterns in cellular
from a given sector are expected to be performed towards sec- networks, the probability of performing handover to distant
tors located in the direction of its antenna azimuth. Knowing neighbors increases when they are located in the direction
the number of sectors in a site and their expected azimuths, it of the source sector antenna’s main lobe. In order to take
is possible to calculate the width of the angle where the main advantage of this property, a function of distance has been
neighbors of every sector are expected to fall into. For each included into (1). Finally, it is expected that the majority
sector, that width will be defined as sector’s solid angle. of handovers attempted from the same source sector fall
Different criteria can be used in order to calculate this solid inside its solid angle (i.e.: W HOxx is the highest among
angle for every sector. As an example, as shown in Fig. 2a, a all for every sector x). Therefore, in case the majority of
site is divided into three sectors with corresponding azimuths handovers are attempted towards sectors located in the solid
Azα = 0o , Azβ = 100o and Azγ = 260o , and the sectors angle of a co-sited sector y, sector x will be candidate to
beamwidth is 60o for all of them. Then, the solid angles would have its actual antenna azimuth pointing in the direction of the
be SAα = [330o , 30o ], SAβ = [70o , 130o ] and SAγ = [230o , expected azimuth of sector y. Two similar reciprocity criteria
290o ]. A different criterion can be used in case the antennas’ are established in order to determine swapped pairs or rotated
beamwidth is unknown. Here, as shown in Fig. 2b, the solid trios. Thus, two sectors α, β are detected as swapped when the
angle bounds can be calculated as the bisector of the azimuths value of the cost function W HOαβ is higher than W HOαα
of the sectors in the site. In this case, the solid angles for the and W HOβα is higher than W HOββ . Whereas for rotated
previous azimuths would be SAα = [310o , 50o ], SAβ = [50o , trios α, β, γ the reciprocity criterion is W HOαβ is higher than
180o ] and SAγ = [180o , 310o ]. W HOαα and W HOβγ is higher than W HOββ and W HOγα
Once the solid angle is obtained for every sector, a cost is higher than W HOγγ .
function for every of the 2-tuples of sites sectors is calculated Fig. 3 shows an example of an scenario with swapped
depending on the location of the neighbor sectors and the sectors. In this figure, dark gray triangles represent sectors
number of outgoing handovers attempts, as in (1). under study, black dots are target sectors and arrows indicate
X the direction of handover attempts between sectors.
W HOxy = #HOxi ×f (dxi ); i ∈ SAy ; xloc = yloc 6= iloc (1) d
i
For the case depicted in Fig. 3, if f (d) = min(d x)
(where
where x is the source sector, y is the sector whose solid angle min(dx ) is the distance from sector x to its closest neighbor,
will be used, i is the target sector located inside y solid angle used to normalize distances) the resulting values for the cost
SAy , dxi is the distance between source and target sectors, function would be those in Table I.

1089-7798 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808292, IEEE
Communications Letters
3

Algorithm 1: Swapped sectors detection Method I TABLE I: METHOD I - EXAMPLE RESULTS

Input : Mobility statistics, topology information Pair case Equation Cost function result
Output: Swapped sectors detected W HOαα 10 × 3 ÷ 2 + 70 × 2 ÷ 2 85
1 for each site do W HOαβ 80 × 2 ÷ 2 + 60 × 2 ÷ 2 140
2 Calculate solid angles for site’s sectors (SA) W HOαγ 10 × 4 ÷ 2 20
3 Calculate cost function for the 2-tuples of site’s sectors W HOβα 50 × 3 ÷ 2 + 70 × 4 ÷ 2 215
permutations with repetitions (W HO) W HOββ 35 × 2 ÷ 2 + 30 × 2 ÷ 2 65
W HOβγ 15 × 4 ÷ 2 30
Apply reciprocity rules
4 if W HOαβ > W HOαα and W HOβα > W HOββ then
Sectors α and β are swapped; TABLE II: METHOD II - EXAMPLE RESULTS
5 else if W HOαβ > W HOαα and W HOβγ > W HOββ
and W HOγα > W HOγγ then Computed
Neighbors Computed Computed
Sectors α, β and γ are rotated; Sector Iteration distance
to combine direction handovers
(km)
6 else There are no sectors swapped;
α 1 1-2 47.73o 2 150
7 end α 2 1,2-3 78.96o 4.18 210
α 3 1,2,3-5 75.69o 4.14 220
α 4 1,2,3,5-4 82.6o 4.13 230
Therefore, for the case in Fig. 3 it may be concluded that α 5 1,2,3,5,4-6 84.11o 4.11 235
sectors α and β are swapped since W HOαβ = 140 is higher β 1 1-5 14.33o 3.65 120
than W HOαα = 85 and W HOβα = 215 is higher than β 2 1,5-2 19.94o 3.2 155
β 3 1,5,2-3 28.14o 2.91 185
W HOββ = 65. β 4 1,5,2,3-4 37.28o 3.09 200
Method I is detailed in Algorithm 1. β 5 1,5,2,3,4-6 40.64o 3.08 210
B. Method II d
f (d) = min(d x)
. In this table, angles used for anglexi
Method II uses handover statistics to compute an actual component are [28o , 65o , 105o , 240o , 340o , 180o ] for sector
azimuth of a sector. Thus, this computed azimuth can be α and [20o , 65o , 110o , 240o , 358o , 180o ] for sector β.
compared with the expected one to determine whether the As in the previous method, two similar reciprocity criteria
sector antenna is pointing to the expected direction. are established in order to determine swapped pairs or rotated
As during the second phase in Method I, this phase utilizes trio based on the fact that computed azimuths are expected to
the location of the neighbor sectors and the number of outgo- be closer to actual azimuths than to azimuths of other sectors
ing handover attempts. In this case, an azimuth is computed in the same site.
for a given sector x by iterating over its neighbors. In every Thus, for the case in Fig. 3 it is concluded that sectors α
iteration, the two neighbors having the highest number of and β are swapped since Azcomputedα is closer to Azβ than
handover attempts are combined into a single sector whose to Azα and Azcomputedβ is closer to Azα than to Azβ .
direction, distance and number of handover attempts are cal- Method II is detailed in Algorithm 2.
culated according to (2). The algorithm will finish when there
is only one neighbor remaining, being the computed azimuth IV. R ESULTS
the computed direction of this. With the aim of measuring the performance of the proposed
Ax methods, a dynamic system-level LTE simulator [7] has been
Azcomputedx = used. A real network consisting of 219 nodes divided into 657
Bx
Cx sectors has been replicated in the simulator. The antennas of all
DISTcomputedx =
Bx sectors are correctly connected which means that there are no
N
X swapped sectors in the original network. Table III summarizes
#HOcomputedx = #HOxi the main configuration parameters of the simulations. Once
i
N
the initial configuration has been set, different simulations have
been carried out. These simulations consist of random swapped
X
Ax = #HOxi × f (dxi ) × anglexi (2)
i sectors (both swapped pairs and rotated trios). All combina-
N
X tions of 0 to 50 random swapped pairs and 0 to 50 random
Bx = #HOxi × f (dxi ) rotated trios have been done, making a total of 2,601 different
i
scenarios. Finally, for every one of the previous scenarios, 100
N
X simulations have been performed to add statistical significance
Cx = #HOxi × f (dxi ) × dxi
i
to the results. Thus, 260,100 cases have been studied in overall.
For every one of the previous simulated cases, the two methods
where N is the number of neighbors of sector x used for the proposed in this letter and the one presented in [5] (henceforth
calculation (N = 2 per iteration), anglexi is the direction referenced as Method 0) have been implemented and their
in azimuthal coordinates (starting from North and clockwise) results evaluated to determine the effectiveness of these.
from source sector x to target sector i, dxi is the distance Solid angles in Method I have been calculated using bisec-
between source and target sector and #HOxi is the number tors of sectors’ azimuth as explained in III-A. Moreover, 4
of outgoing handover attempts from sector x to sector i. different functions of distance have been utilized to evaluate
For the case depicted in Fig. 3, Table II summarizes the both Method I and Method II. Regarding Method 0, the
computed azimuth values obtained after every iteration when reference values specified in its description have been used.

1089-7798 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808292, IEEE
Communications Letters
4

Algorithm 2: Swapped sectors detection Method II

Input : Mobility statistics, topology information
Output: Swapped sectors detected
1 for each site do
2 for each sector in site do
3 while number of neighbors > 1 do
4 Look for the two neighbors having the highest number of
handover attempts and merge their computing direction,
distance and number of handover attempts
5 end
Apply reciprocity rules
6 if abs(Azcomputedα −Azβ ) < abs(Azcomputedβ −Azβ )
and abs(Azcomputedβ −Azα ) < abs(Azcomputedα −Azα )
then Sectors α and β are swapped;
7 else if
abs(Azcomputedα −Azβ ) < abs(Azcomputedβ −Azβ ) and
abs(Azcomputedβ −Azγ ) < abs(Azcomputedγ −Azγ ) and Fig. 4: Simulation Results
abs(Azcomputedγ −Azα ) < abs(Azcomputedα −Azα )
then Sectors α, β and γ are rotated; 83.83% averaged true positives ratio, respectively. In terms
8 else There are no sectors swapped; of false positives, linear function of distance has resulted the
9 end best performing in Method I and so has quadratic function
10 end
of distance in Method II; being the averaged ratio of false
positives 0.63% and 5.06%, respectively.
TABLE III: SIMULATION PARAMETERS Both methods have also been tested in three real LTE
mobile communication networks. The first network consisted
Parameter Configuration
Cellular layout 657 sectors (219 eNBs) of 5267 sectors where Method I detected 5 cases and Method II
Transmission direction Downlink detected 8 cases. The second network consisted of 4582 sectors
Carrier frequency 1.8 GHz where Method I detected 4 cases and Method II detected
System bandwith 1.4MHz
Propagation model Okumura-Hata 8 cases. The third network consisted of 7969 sectors where
Log-normal slow fading σsf =8dB Method I detected 11 cases and Method II detected 25 cases.
and correlation distance = 50m Moreover, drive tests were performed confirming detected
Channel model Multipath fading, ETU model
Mobility model Random direction, 50km/h cases as actual cases.
Base station model Tri-sectorized antenna, SISO, V. C ONCLUSIONS
PT XM AX =43 dBm
Handover Triggering event= A3 This letter presents two methods for automatically detecting
Measurement type = RSRP swapped sectors in mobile communication network by using
HO off = 3dB
Time resolution 100 TTI (100 ms) mobility statistics and avoiding drive test activities.
The proposed methods have been compared with prior art in
Once the simulations have been performed, two statistical terms of effectiveness. After a statistical significant number of
indicators have been obtained. These are, the ratio of true simulations, Method I has shown the lowest number of false
positives (i.e.: detected swapped cases divided by the total positives whereas Method II has the best detection rate.
number of swapped cases) and false positives (i.e.: non- Additionally, both methods have been tested in real LTE
swapped cases detected as swapped divided by the total mobile communication networks. They were useful for detect-
number of nodes in the network). As described before, a total ing swap cases and saving operation costs. Besides, performed
of 260,100 simulations have been performed. Among these, drive tests confirmed detected cases as actual swaps.
true positives have been calculated for 260,000 simulations
R EFERENCES
corresponding to scenarios with at least one swapped case
and false positives have been calculated for 100 simulations [1] J. Zhang, J. Sun, and D. Yang, “Application of drive test for QoS
evaluation in 3G wireless networks,” in Proc. IEEE ICCT, vol. 2, 2003,
corresponding to the scenario where there are no swapped pp. 1206–1209.
cases. The average of these indicators has been calculated per [2] N. Shaikh, “Sector swap identification guide.” [Online]. Available:
method configuration. https://www.scribd.com/doc/36864048
[3] W. Jiao, H. Yiling, W. Zaixue, and Y. Dacheng, “A novel model of drive
Finally, Fig. 4 depicts the results of the tests carried out. test data processing in wireless network optimization,” in Proc. IEEE Int.
From the results, it can be concluded that both Method I and Symp. on Personal Indoor and Mobile Radio Communi. (PIMRC), Sep.
Method II achieve better results in terms of true positives than 2006, pp. 1–4.
[4] K. Guo, R. Chang, S. Qian, and X. Chen, “Method and apparatus for
Method 0 while keeping a low figure of false positives. Among diagnosing feeder misconnection,” Patent US 13/732.255, 5 9, 2013.
these, Method I is the best option to minimize false positives [5] R. Menon and E. Gormley, “Method and system for crossed antenna feed
and Method II is the best one to maximize true positives. detection,” Patent PCT/US2016/023 800, 9 29, 2016.
[6] Performance measurements Evolved Universal Terrestrial Radio Access
With respect to function of distance, it has been evidenced Network, 12 2016, 3GPP TS 32.425 v14.1.
that its use may improve the results of both methods. Quadratic [7] P. Muñoz et al., “Computationally-efficient design of a dynamic system-
level LTE simulator,” Int. Journal Electron. Telecommun., vol. 57, no. 3,
function of distance is the best performing one in terms of true pp. 347–358, Sep. 2011.
positives. Method I and Method II have provided 71.58% and

1089-7798 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.