-Traffic congestion during busy hour (BH) deteriorates the overall performance of cellular network and may become unmanageable unless effective and efficient methods of congestion control are developed through real live traffic data measurement and analysis. In this work, real live traffic data from integrated GSM/GPRS network is used for traffic congestion analysis. The analysis was carried out on 10 congesting cells using network management system (NMS) statistics data that span three years period. Correlation test was used to show that TCH congestion depend only on call setup success rate (CSSR) and BH traffic at cell level. An average correlation coefficient value of 0.9 was observed between TCH congestion and CSSR while 0.6 was observed between TCH congestion and BH traffic. The correlation test is important when selecting input for congestion prediction modeling.
INTRODUCTION
All over the world, cellular network operators are faced with the challenges of improving the quality of service (QoS) while increasing capacity and rolling out new services. This has resulted in many congested networks and consequently degradation of QoS due to inadequate provision of the needed resources or underutilisation of the available resources.
To cope with subscriber demands and meet Regulator standards, cellular network providers dimension network elements on continuous based using network management system (NMS) statistics, drive test trailing and customer feedbacks.
However, Nigerian Communication Commission (NCC) quarterly audit reports of GSM network performance had consistently shown that the operators have not been able to meet the set standards [1] due to network congestion [2]. If there is no hardware fault, network congestion usually occurs due to insufficient number of radio channels in access network elements [3].
This work used busy hour traffic data of access network from a live network to analyse traffic congestion in some macrocells of GSM/GPRS network. The busy hour of a network is the time when the network processes the highest traffic in a day and it is used to measure network performance, determine the robustness of a network and its dimension [4]. To measure the network performance, the pattern of busy hour traffic is considered for congestion evaluation [5] using key performance indicators (KPIs).
These KPIs are defined for different interfaces and network elements of GSM/GPRS. To cater for subscriber demand, radio frequency (RF) optimization teams use the KPIs to generate quality of service (QoS) reports to ensure minimal congestion over all the interfaces and network elements in order to avoid QoS degradation by maintaining the KPIs under pre-defined threshold [6].
Network congestion leads to poor QoS which affect grade of service (GoS) of the network, particularly during the busy hour of the day [7]. In a loss system, the GoS describes that proportion of calls that are lost due to congestion in the busy hour and can be measured using equation (1): Number _ of _ offered _ calls (1)
While a whole range of services of GSM technology are in use in Nigeria, it is obvious that the network performance in terms of QoS are degrading which proved that GSM network is either over utilized or under dimensioned. Hence the need for this analysis to identify the cells that are responsible for congestion during busy hour by statistically analyzing traffic data of a live network in order to establish the presence or absence of congestion.
II.
LITERATURE REVIEW
Traffic analysis is important for evaluating the
performance of existing networks and network optimization.
The events that occur in BTS trigger different counters in the
BSC and MSC memory that are used for performance
monitoring and network evaluation. Various KPIs that are
used to measure network performances are derived with the
help of these counters using different formulations [7]. In
practice, the performance can be monitored at different
sections of the network [
Some of the early works on GSM network elements
performance were done mostly on access part of the network
particularly at BTS level. For example, [9] proposed a traffic
model for mobile network, using Markov chain to determine
call blocking and handoff failure probabilities when no
queuing of new or handover calls is performed while [
The work of [
Reference [
In another work, [3] analyzed traffic data from a trunked
radio network operated by Ecomm in UK using OPNET
model. The findings indicated that traditional Erlang models
for voice traffic may not be suitable for evaluating the
performance of trunked radio networks. In a related work,
[
III.
METHODOLOGY
The setup for data collection in GSM/GPRS network comprises of base station subsystem (BSS) and network subsystem (NSS) connected to standalone system called NMS as shown in Figure 1.
NMS is the functional entity from which the service provider monitors and controls the entire network. The data used in this work was extracted from the NMS with the help of Ericsson Business intelligent (BI) tools installed on the standalone computer and exported to Microsoft Excel environment.
The network is composed of 742 cells, 262 BTSs. Measurements were taken from November, 2012 to September, 2014. Correlation test was used to choose only KPIs that have significant effect on TCH congestion during busy hour.
Correlation coefficient is defined as a number or function indicating the degree of correlation between two variables like X and Y. In this work, the variables are busy hour traffic and CSSR, HOSR, DCR, SDCCH and TCH Congestions as KPIs to measure the network performance. Equation (2) defines correlation coefficient as:
Correl( X ,Y )
x x y y x x 2 y y 2 (2)
Microsoft excel statistical tools was used for the correlation analysis.
IV.
BTS DAILY BUSY HOUR TRAFFIC ANALYSIS
The busy hour traffic data showed that 154 cells
experienced congestion out of 742 cells during the period
under investigation. In the cell KPIs analysis, 10 worst
congesting cells were chosen for BH TCH congestion
analysis [
Table 3 shows that BH TCH congestion has strong correlation with CSSR and busy hour traffic for the four cells.
Given the result of the correlation test, the maximum daily BH traffic carried by the ten cells and it impact on CSSR and TCH congestion is shown in Figure 2 and Figure 3.
From Figure 2 and 3, most of the cells CSSR and TCH CONG degraded when they carried maximum traffic while Cell 396B, 393C and 301C suffered worst KPIs degradation.
The behaviour of the ten worst cells when they experienced worst TCH CONG and the effect on other KPIs during the period is shown in Figure 4 to 6.
All the cells CSSR degraded and their ability to carry traffic is limited when they experienced worst TCH CONG while Cell 496B, 430C, 396B and 038B recorded worst KPIs degradation. This shows that traffic channel of these cells are not properly dimensioned. VI. CONCLUSION
The analysis of traffic channels in an network existing showed that TCH congestion beyond the acceptable 2% threshold for traffic channel (TCH) occurred in 154 cells out of 742 cells investigated.
The busy hour TCH congestion analysis showed that the congestion depends on CSSR and BH traffic. The CSSR and BH traffic has an average correlation coefficient of 0.9 and 0.6 respectively for the cells. The strong correlation showed that the knowledge of CSSR and busy traffic can be used to predict TCH congestion which is crucial for cellular network optimization and resource management.
DCR FC0496B FC0362C 2 FC0494B
FC0301C 98 2
FC0430C FC0038B 98
0.20 DCR FC0725C FC0496B FC0494B FC0430C FC0396B FC0393C FC0385B FC0362C FC0301C FC0038B NCC