In this article, we would discuss on ML framework which can
be used to identify one of the most difficult problem for a telecom operator i.e
Sleeping Cell.
What is sleeping cell???
The Sleeping Cell problem is a particular type of cell
degradation in Long-Term Evolution (LTE) networks. In practice such cell outage
leads to the lack of network service and sometimes it can be revealed only
after multiple user complains by an operator. a cell can become sleeping cell due
to SW / HW bug, unfortunately this problem of sleeping cell don’t have any alarm
in network Elements, thus it makes almost impossible to find sleeping cell in
network, unless customer complaint is received.
Though there are different type of sleeping cell few can be
identified with the help of KPI analysis, where as still there are few specific sleeping cell scenario like RACH sleeping which is not easy to get identified via KPI
analysis, where Base station doesn’t permit new
RACH attempt , but ongoing calls will be going on without any problem
and slowly when all calls will go down BS will seize its services, because of
this scenario operator can identify the issue much later.
Here we will discuss about the sleeping cell scenario which
we can identify with the help of KPI, for eg. a fault in antenna gain, which result
in degradation of cell services.
To identify this kind of sleeping cell we are proposing a
K-Nearest neighbor based ML model to identify the sleeping cell , which will
also help customer to reduce drive test requirement and in -turn can save cost.
SLEEPING CELL DETECTION FRAMEWORK:
To identify the sleeping cell in our frame work , its
basically a 3 steps process.
a)
Measurement
b)
Sleeping cell detection
c)
Sleeping cell localization.
Measurement for Sleeping Cell
detection:
In SON there is a feature Minimize
drive test (MDT ) , the idea behind this feature is that instead of performing
drive test via drive test engineer , network can use UE to get all the information
from the UE which are there in this location.
Overview about this feature can be
find out in below link:
https://www.sharetechnote.com/html/Handbook_LTE_MDT.html
The MDT measurement functionality allow
operator to collect measurements either periodically or event based
Sample of MDT report
After receiving measurement data from
MDT, the data is further processed by cleaning and scaling it to lower dimension
by using techniques like PCA , MDS etc. , as we know the measurement data are of
very high dimension in nature and put them in a single vector.
V = {RSRPS, RSRPN1, RSRPN2, ...RSRPN3,
...RSRQS, RSRQN1...RSRQN3, CQI}
Here S stands for serving cell and N stands
for neighbouring cell.
The dimensionality reduction is a crucial
step as a high-dimensional KPI database poses challenges for network engineers
as well for experts. The real network is complex and dynamic in nature, and it
is often not possible to identify few KPIs that really capture the behaviour of
the system. On the other hand, projecting the data onto fewer dimensions of
maximum variance uncovers the true structure which ultimately aids the cell
profiling process. Moreover, less computational effort is required which
consequently leads to low detection delays.
Then this KPI data can be used with state
of art anomaly detector model in ML i.e k-Nearest
Neighbor based Anomaly Detector (k-NNAD)
, Local Outlier Factor based Anomaly Detector (LOFAD) etc.. , to
identify the abnormality of the KPI in the network.
Sleeping cell localization.
Once the deployed ML based anomaly
detection based mechanism , done the classification of abnormal and Normal cell
, the UE reported information via MDT feature can be used to identify the sleeping
cell. The MDT measurement reports also contains
time and location information, which are not used in cell detection,However, based
on the coordinate information, the classified measurements can be further
mapped to network topology. As a result, the cell which corresponds to the
highest number of abnormal measurements can be easily identified as shown in
Figure 2.
CONCLUSION
AND FUTURE WORK:
In this Article, we suggested a un
supervised machine learning framework
for automating the sleeping cell detection process in an LTE network. Our
proposed approach first acquire key performance measurements from the
fault-free operating network. The data is further embedded into a
lower-dimensional space. The embedded measurements are used to build a normal
profile of the network by training the k-NNAD and LOFAD detection models. The
models are later used to automatically detect abnormal measurements from the
test scenario. Finally the UE reported coordinate information is employed to
localize the position of sleeping cell.
1)
S. Ham¨ al¨ ainen, H. Sanneck, C. Sartori ¨
et al., LTE Self-Organising Networks (SON): Network Management Automation for
Operational Efficiency. John Wiley & Sons, 2012.
2)
M. M. Breunig, H.-P. Kriegel, R. T. Ng, and
J. Sander, “Lof: identifying density-based local outliers,” in ACM Sigmod
Record, vol. 29, no. 2. ACM, 2000, pp. 93–104.
3)
https://www.sharetechnote.com/html/Handbook_LTE_MDT.html
4)
O. Manzanilla-Salazar, F. Malandra, and B.
Sansò, “eNodeB failure detection from aggregated performance KPIs in smart-city
LTE infrastructures,” in 15th International Conference on the Design of
Reliable Communication Networks, DRCN 2019, Coimbra, Portugal, March 19-21,
2019, 2019, pp. 51–58.
5)
E. J. Khatib, R. Barco, A. Gómez-Andrades, P.
Muñoz, and I. Serrano, “Data mining for fuzzy diagnosis systems in LTE
networks,” Expert Systems with Applications, vol. 42, no. 21, pp. 7549–7559,
2015.
6)
“Smart cities M2M traffic characterization and
performance analysis,” https://www.trafficm2modelling.com/home, accessed:
2019-02-01.
7) S. Chernov, M. Cochez and T. Ristaniemi, "Anomaly Detection Algorithms for the Sleeping Cell Detection in LTE Networks," 2015 IEEE 81st Vehicular Technology Conference (VTC Spring), Glasgow, 2015, pp. 1-5, doi: 10.1109/VTCSpring.2015.7145707.
