4.4 Results analysis
4.4.1 Simulation tests
The simulation tests are executed using the dynamic LTE system level simulator presented in Chapter 3. This simulator represents the LTE radio access network and allows to estimate the most important KPIs for each cell and adjacency. The Cell Outage Model proposed in Section 4.2 is integrated in this simulator in order to assess the different COD algorithms.
The simulation scenario is a part of a real network consisting of 75 cells. The simulator includes a user mobility model that allows to simulated users with different speeds. The simulator also includes an HO algorithm to generate realistic HO statistics. The HO algorithm considered in this work is based on the A3 event. When the signal level received by a user from a neighboring cell exceeds the signal level received from the serving cell by a margin, HOM, the HO is executed. Table 4.1 summarizes the main configuration parameters of the simulations.
Table 4.1: Simulation parameters Parameter Configuration
Cellular layout 75 cells (25 eNBs) Transmission direction Downlink
Carrier frequency 2.0 GHz System bandwidth 1.4 MHz Propagation model Okumura-Hata
Log-normal slow fading, σsf = 8 dB
and correlation distance=50m Channel model Multipath fading, ETU model Mobility model Random direction, 3, 10, 50 km/h Base station model Tri-sectorized antenna, SISO,
PT Xmax =43 dBm
Handover Triggering event = A3 Measurement type = RSRP HOM = 3 dB
Time resolution 100 TTI (100 ms)
In total, 307 outage situations are generated in the simulation: 49 are outages that do not affect the eNB so that KPIs from the cell are available and the related availability KPI indicates the outage situation; and 258 are outages that affect the eNB so that KPIs from the cell are not available in the OSS. In addition, 141 eNB-OSS failed connections have been simulated. The different COD algorithms have been activated during the simulation to assess their effectiveness. Table 4.2 presents the results in terms of the false positive rate and the false negative rate for each COD method. The false positive rate is the percentage of cases wrongly detected as positive among the total outage cases and the false negative rate is the percentage of undetected cases among the total of normal cases simulated.
Table 4.2: Simulation results
Result Availability KPIs Lack of KPIs inHO statistics
False_Negative_Rate 84% 16% 5.9%
False_Positive_Rate 0% 0.9% 0%
The Availability KPIs algorithm obtains 0% of false positives but 84% of false negatives. This value coincides with the frequency of occurrence of the outage cases with the eNB affected because this method cannot detect an outage problem when KPIs are not being reported.
The Lack of KPIs algorithm has nonzero false positive and false negative rates. The value of the false negative rate (16%) coincides with the frequency of outages with available KPIs. Moreover, another important disadvantage of this method is that it produces 0.9% of false positives, equal to the frequency of OSS failed connections, since the algorithm identifies every eNB-OSS failed connection as outage.
It is possible to define an algorithm that combines the Availability KPIs and the Lack of KPIs methods. With this algorithm, the false negatives of both methods can be eliminated. However, this algorithm would have an important number of false positives since the algorithm would classify all the failed eNB-OSS connection situations as cell outages.
Finally, the proposed inHO statistics algorithm is able to detect most simulated outages, leading to a low percentage false negatives (5.9%). This situation is related to cells with low traffic so that they should have a small impact on the overall network performance. When a cell with low traffic is in outage, the number of inHO in the current measurement period would be equal to zero. The algorithm is able to detect the outage only if the number of inHO in the immediately preceding measurement period is nonzero. However, it is possible that the cell does not manage any inHO in the earlier measurement period due to the low traffic. In this case, the algorithm cannot detect the outage problem, resulting in a false negative.
The proposed algorithm produces a 0% of false positives, considering the simulated situa- tions, since the availability KPIs allow to detect the potential false positives cases (i.e., cells with very low traffic that have not inHO in a certain measurement period although no problems are affecting it). However, in a real network, new situations may occur that produce a false positive, e.g., a cell with no inHO during a certain hour and no KPI available due to an OSS connection failure. Nevertheless, since those situations are not very common, the real percentage of false positives will be very low.
After analyzing the proposed COD algorithm, it is possible to make a comparison with state- of-the-art approaches. Specifically, most COD algorithms found in the literature (e.g., [27, 28]) are based on detecting cells in outage by monitoring KPIs from the cell in outage. All these methods can be represented by the Availability KPIs approach in the simulations. The main drawback of these solutions is that they can detect outages only when there are available KPIs from the cell in outage. This situation occurs when the outage does not affect the eNB.
but user traces are needed to perform these approaches. For this reason, this method is not appropriated for being implemented in real networks since the feature to collect traces is normally disabled.
In [29], a detection algorithm is presented. In this case, detection is done based on neigh- boring cell measurements. Specifically, the algorithm applies a complex method to conclude if a cell is in outage based on the degradation generated in the cell edge of neighboring cells. The effectiveness of this approach depends on the level of degradation caused by the outage. How- ever, in many situations, it is difficult to appreciate the degradation caused by the outage in the signal quality experienced by the users in the neighboring cells, especially when there is high cell overlapping. To illustrate this, a figure is included below. Fig. 4.4 shows the value of the 50th percentile of the SINR experienced by the users in the neighboring cells of a site in outage. It can be seen that some neighboring cells experience degradation, but others experienced an im- provement. In fact, the average value obtained shows a slight increase. Specifically, the average value of the 50th percentile of the SINR in the normal situation is 2.5012 dB and the obtained value in the outage situation is 2.6491 dB. Consequently, although the algorithm proposed in [29] can obtain good results in many situations, this approach has an important limitation to adapt to different scenarios (e.g., scenarios with a high level of overlapping).
1 2 3 4 5 6 7 8 9 10 −3 −2 −1 0 1 2 3 4 5 6 7 Neighbor cells 50th percentile of SINR Normal situation Outage situation
Figure 4.4: 50th percentile of SINR of the neighboring cells of a site in outage.