Statistical Analysis of Handover Process Performance in a Cellular Mobile Network in the City of Quito, Ecuador ^†

Abstract

This paper presents an overview of the findings in the handover (HO) process performance within three routes in Quito, Ecuador. We used the Net-Monitor Software to gather information from one of the three national mobile operators. Then, we used the R tool to analyze the HO performance. We analyze several performance metrics, such as HO types, HO conditions, and the ping-pong process. Analysis of the results of the outdoor drive tests demonstrate that the radio frequency (RF) parameters, such as Received Signal Strength Indicator (RSSI), Reference Signal Received Quality (RSRQ), power margin, times radio frequency measurements repeats, and HO percentage to nearest BS, are extremely important during different HO types and ping-pong processes because there are statistical differences in these measured RF parameters. The main measurement results demonstrate that RSSI difference between inter HO and intra HO is 20 dB, whereas HOs are performed when the mobile device (MS) gets farther from the base station (BS), approximately 50% of total HOs. Operator achieves a high ping-pong rate of approximately 10% of total HOs.

1. Introduction

Mobile network evolution aims to enhance mobile broadband capabilities according to the higher service demands. Mobile Network Operators (MNOs) continuously monitor the quality of service (QoS) to guarantee high-network performance. HO is essential in mobile network performance since it refers to transferring connections from one serving BS to another while allowing the MS to maintain continuous communication (call or data session) while moving between different cell areas, Base Stations (BSs), and different technologies within the network. It reflects the QoS, network performance, user experience, and other crucial indicators. In any technology (2G to 5G), typically, the HO process follows these steps [1]:

Measurement: The MS acquires measures of the signal strength quality of nearby BSs and from its current serving BS.

Triggering: The HO process starts when the signal strength and (or) other parameters of the current serving cell drop below certain thresholds or when a neighboring cell’s parameters become significantly better.

HO Decision: Based on the MS shared measurements, the network operator performs an HO decision-making process to determine whether an HO is necessary and which neighboring cell the MS should connect to. The decision considers factors such as signal strength, quality, available resources, hysteresis, and HO policies. It is important to note that the network operator defines these. Due to this, the HO process does not have to be to the nearest BS.

Preparation: Once the HO decision is made, the target BS is informed of the incoming HO. Resources are allocated and prepared to accommodate the incoming MS call or data session.

Execution: The MS is instructed to switch its connection from the current serving cell to the target cell.

HO Completion: The HO process is complete after the mobile device successfully connects to the target base station. The new serving cell now serves the MS, and the communication continues.

The HO can be classified as “hard” (if it disconnects the MS from the current serving cell before connecting to the target cell) or “soft” (if there is a smooth transition. This happens when there is an overlap between the two cells’ coverage) [2]. There are other HO classifications, such as intra-cell HO (between different sectors of the same BS), inter-cell HO (between neighboring BSs within the same network), and inter-system HO (between different cellular technologies, e.g., 4G to 5G). The specific HO procedures and parameters may vary depending on the used technologies (i.e., GSM, CDMA, WCDMA, LTE, 5G) and the network architecture. The aim is to provide the best possible QoS and maintain an uninterrupted user experience as the MS moves throughout the network [3]. One key objective of the 4G and 5G networks is to improve mobile broadband capabilities with higher demands among service consumers, such as high-speed Internet connections for urban, suburban, and rural areas. Many mobile network operators (MNOs) monitor the quality of service in terms of multiple services to guarantee high-network performance [4]. In [5], the authors provide an extensive performance evaluation of five national MNOs in Malaysia using metrics such as Reference Signal Received Power (RSRP), RSRQ, signal-to-noise ratio (SNR), throughput in downlink (DL)/uplink (UL), ping, and HO.

The 5G network spectral efficiency may improve by reducing the coverage of BSs, which reduces the number of users by each BS and enhances frequency reuse; however, this increases the HO rate, i.e., the successive change in handling BS for a mobile user. Accordingly, the MNO may implement a capacity gain but at the cost of increased HO rates and higher signaling overheads caused by the HO procedure. In [6], the study found that the basic HO scenario in 5G new radio (NR) is very similar to the long-term evolution (LTE) except for the involved entities and a slight change in HO steps. For instance, the basic HO procedure in NR is completed in 12 steps, while in LTE, it goes through 18 steps.

HO between several networks or cells of the same network is a problem for mobile nodes. In [7], the authors proposed a mathematical model for Wrong Decision Probability (WDP) and Handover Probability (HP) to better understand the behaviors of QoS parameter-based HO algorithms under different network conditions and end-user requirements; they also suggest a WDP based HO algorithm to improve the HO performance. Therefore, making HO decisions based on WDP can provide better performance. Similar work has been carried out in [8], using the same data collection methodology that we describe in this paper, plus Decision Tree Algorithms, to obtain a predictive HO approach in LTE networks.

Recent research has focused on optimizing HO control parameters appropriately for efficiently address HO issues during user mobility [9,10,11]. In [9], the authors propose a fuzzy-coordinated, self-optimizing HO scheme to achieve a seamless HO while users move in multi-radio access networks. The load balancing optimization function adaptively adjusts the settings of HO control parameter to achieve balance uneven loads between adjacent cells. In [10], the authors have focused on analyzing the performance of load balancing self-optimization within 5G cellular networks. The conflict resolution technique is introduced in the self-optimization network, which is responsible for addressing contraction between mobility robustness optimization and load balancing optimization [11]. This technique performs optimization by obtaining the weight function for input parameter and then monitoring the HO types occurring during HO performance.

The rest of this work is organized as follows: Section 2 provides information about the data recollection methodology and geographic description of the 3 routes. Section 3 describes the data preprocessing for further statistics, statistical analysis of the different RF parameters at the HO process, and a brief discussion about the findings of the results. Finally, it presents conclusions.

2. Data Measurements

There are several tools for mobile cellular communications network data collection. After a careful comparison and investigation, Net Monitor was selected for the data measurements since it allows monitoring and logging of mobile network parameters without using specialized equipment (1 sample per second of 33 variables: report, sys_time, sim_state, net_op_name, net_op_code, roaming, net_type, call_state, data_state, data_act, data_rx, data_tx, gsm_neighbors, umts_neighbors, lte_neighbors, rssi_strongest, tech, mcc, mnc, lac_tac, node_id, cid, psc_pci, rssi, rsrq, rssnr, slev, gps, accuracy, lat, long, band, arfcn). It allows log exportation in .kml and .csv formats.

The data measurements were performed in Quito/Ecuador from 28 November 2022 to 18 December 2022 and were taken on three routes. The collected data were saved on several Microsoft Excel software program spreadsheets. These routes were Quitumbe (37,382 samples taken at 5.872 km/h on average), San Bartolo (15,260 samples at 10.6 km/h), and La Floresta (20,000 samples at 10.8 km/h).

It must be noted that Quito has an elevation of 2850 m asl. Its shape is very long (40 km) and very narrow (4 km), and its population is approximately 3 million people (Density = 7500/km²). Its buildings have ten floors on average, so it is suitable for the urban environment of a small or medium-sized city in the Okumura–Hata propagation model [12]. Figure 1 displays where the data measurements were taken (yellow routes) and Figure 2 shows the points where the HO events occur (heightened in red color). These three routes were defined according to their high usage of the HO process. The Quitumbe and San Bartolo routes were selected since many mobile subscribers take these routes to travel from home to work and vice versa. The route, named La Floresta, was chosen due to its high traffic of pedestrian users. In this work, we select only one MNO. In Ecuador, 5G networks have not been put into commercial operation yet, only trial sites have been deployed. No 2G data were collected since this network is poorly used and the HO process mainly occurs between 3G and 4G technologies. Thus, the tested area mostly coverage with the 4G network. We carried out a methodology through walk tests for measuring the cellular network parameters using a test terminal.

Similarly, we conducted drive-tests along the main streets and avenues around the cell sites. The goal was to gather samples of the network’s behavior, MS performance, and signal quality vs. mobility, mainly by observing different HO. At each GPS location, the Net Monitor tool simultaneously measures the parameters around the cell sites of the MNO in the three routes described earlier.

3. Sample Processing

Some additional information needs to be included in the gathered data and must be included to have a complete data set for HO analysis. This additional information was obtained with the data set applying algorithms and getting out parameters like a power margin, HO type, ping-pong effect, etc. The BS coordinates were gathered from the Cell Mapper website “https://www.cellmapper.net (accessed on 15 June 2023)“ and verified during the drive tests. We pre-processed the data collected from field tests using the output files exported from the Net Monitor tool in the .csv extension. We merged all the samples from individual files into a single file. After that, we transformed raw data into valid numerical data by modifying cell formats for numerical values, dates, times, and signs. Then, we carried out specific algorithms for each column and related the measured parameters at each location point for each route. We separated the “sys_time” variable into several columns for precise date and time pointers. We added columns for general HO types (HO yes/no, Intra-cell/Inter-cell, Inter-Radio Access Technologies (RAT)) and another column for the “power margin” value when there is a handover between cells, represented as the difference between the RSSI levels (rssi–rssi_strongest). Also, we added a column to calculate the power difference by subtracting each interval’s new RSSI level from the previous one. Using the Haversine formula, we added columns to calculate the distance between each BS and the MS within the mobile network. The following flowchart describe how to find new features obtained from the samples of the observations for the HO process.

Figure 3 shows the flowchart of the feature “HO_yes_no” to find the HO existence. It compares 2 variables: contiguous “node_id” (BS node id) and “cid” (BS cell id); when these are the same, then “no_HO”, and when these contiguous values are different, then “yes_HO”.

Figure 4 shows the flowchart of the feature “HO_intra_inter_BS”. This feature differentiates the HO types, like intra-cell or inter-cell. It compares the neighbor’s node ID during the HO process. If these are the same, there is “intra_HO”; if these are different, there is “inter_HO”.

Figure 5 shows the flowchart of the feature “HO_tech”. This feature differentiates the HO between RAT. It compares two neighboring fields in the “tech” variable when the HO process is present.

Figure 6 shows the flowchart of the feature “HO_to_nearest_BS”. This feature shows if the HO process connects the MS to the nearest BS. First, we calculate the distance between BS and each MS in the route and compare if the connection establishment during the HO process is to the nearest BS.

Figure 7 shows the flowchart of the feature “times_rssi_repeats_before_HO” and “times_rsrq_repeats_before_HO”. When a HO happens, these features show if the RSSI and RSRQ values repeat themselves during the previous 20 measurements and count the occurrences.

Figure 8 shows the flowchart of the feature “ping_pong”. This feature depicts the “ping-pong” phenomenon in the HO process. For this, we count HO in 10 s previous, determine if MS is farther from the HO BS, and count HO in the same BS.

4. Results and Discussions

We post-processed the data collected using the RStudio 2023.06.0 Build 421 version software. We implemented box plots and statistical analyses to obtain the main radio frequency parameters included in the HO scenarios and mobility in the target area. We analyzed and deduced the performance and behaviors for each radio parameter in the measurements and samples for each route. Figure 9 shows the HO type (intra and inter) vs. RSSI for the 3 routes.

If the RSSI decreases to −81 and −86 dBm in the Quitumbe and San Bartolo routes, an intra-cell HO occurs in mobile networks. If the RSSI reduces further between −99 and −106 dBm, an inter-cell HO is performed. While the La Floresta route presents different behavior, intra-cell HO occurs in MNO when RSSI level is approximately −24 dBm, and inter-cell HO occurs when RSSI level is approximately −86 dBm. The results indicate that RSSI level is a key metric for HO type. For example, the RSSI level achieved by Quitumbe is −107 and −86 at inter HO and intra HO, respectively.

Concerning the RSRQ parameter, the HO process is distributed almost 50/50 between inter and intra-cell HO (−13/−17). Inter or intra HO are triggered with similar RSRQ levels in the Floresta route and between 1 dB difference at intra and inter in the other routes.

Figure 10 displays the HO type (intra and inter) vs. times RSSI measurements repeat before HO for the 3 routes. If a RSSI value in the La Floresta route repeats itself more than 20 times, then MNO performs an intra HO and 4 times for inter HO. The Quitumbe and the San Bartolo routes present similar behavior with six and three RSSI repeats before intra and inter HO, respectively. As in the previous case, the results indicate that RSSI repeats is a key metric for HO type.

Concerning the RSRQ repeats parameter, the Floresta route presents a singular behavior, RSRQ repeats itself 20 times for intra HO and 5 times for inter HO. The Quitumbe route presents nine repetitions for intra HO and seven for inter HO, while four repetitions for the San Bartolo route perform an inter HO and intra HO.

Figure 11 shows the HO type vs. power margin for all routes. The difference between the current and previous RSSI values is the power margin. Intra or inter HO is activated when power margins are primarily between 1 and 2 dB.

The results reveal the HO frequency to the nearer or farther BS. In practice, the HOs are performed when the MS gets farther from the HO BS between 42% and 48%. In all routes, MNO achieve that HO occurs when the MS goes nearer the HO BS between 46% and 51%, and the remaining HOs are performed when the MS is stationary. HO to farther BS leads to possible HO failure due to physical distance from the server and the reduction in the quality of radio channel.

The results demonstrate the HO rate with ping pong and without ping pong. The majority of the HOs do not show the ping-pong behavior. Less than 10% of the HOs have this problem.

Figure 12 and Figure 13 illustrate the RSSI and RSRQ for HO process with ping pong and without ping pong (no_ping_pong), respectively, for all routes. The Quitumbe and San Bartolo routes present similar behavior with RSSI down to −102/−107 dBm with ping-pong effects and no ping pong when RSSI is between −82 dBm and −103 dBm. The La Floresta route presents ping pong when RSSI is −24 dBm and no ping pong when RSSI is −82 dBm. An average RSRQ difference of approximately 1 to 3 dB achieves between ping pong and no ping pong. The RSSI and RSRQ may be sufficient key metrics for mobile systems to determine or evaluate the occurrence of ping pong.

5. Conclusions

This study provided an HO performance evaluation of MNO in Quito, Ecuador. The data measurements were conducted in a city with 4G and 3G technology infrastructure. The existing network was tested and evaluated with various performance metrics in two scenarios: with intra and inter HO and with the ping-pong process. Most results shown for the three routes had similar results. The difference of average performance metrics (RSSI, RSRQ, power margin, and measurement repeats) is extremely important during different HO types and the ping-pong processes. Therefore, this finding indicates that the MNO may implement algorithms to predict the HO types and the ping-pong process in a specific tested area with the use of radio frequency measurements. In addition, approximately 50% of total HO processes does not occur to nearest BS. This problem needs to be resolved to reduce the HO failure probability. Limitations not considered in this study and can be a point to future works.