Spatial Analysis of GSM Subscriber Call Data Records

By Ireti Ajala

The most valuable asset of many companies is not their products or services, but their data. This is particularly true in the communications industry. Trapped inside the customer billing systems is a gold mine of data that holds the key to customer retention, reduced expenses, customer self-service and overall competitive advantage.

In the mobile communications networks, the core business is selling airtime to subscribers. All the information about that airtime is tracked through call data records, or CDRs. CDRs are used to bill customers. Because of the huge amount of data that would need to be processed, they are usually not analyzed to add business value. However, with shrinking margins and increasing pressure to improve revenues, operators now are looking closely at ways to use these data to their advantage.

GIS is one powerful tool that can be used to analyze CDRs, allowing the operators to see a precise, up-to-date picture of the entire network and to better understand the calling patterns of their subscribers, with a view to knowing their networks better and offering subscribers customized services, and hence increasing revenue.

This article describes how MTN Nigeria has been using GIS to investigate the relationship between the geographic spread of subscribers, using CDRs and network stipulated key performance indicators (KPIs) like traffic.

A CDR is generated at the switching center of a GSM network each time a successful call goes through. The switching center generates huge volumes of these records, sometimes running into terabytes. It is therefore a serious challenge to analyze these raw data to enhance the making of informed decisions.

CDRs can be analyzed based on the location of subscribers, cells, market share, handset usage, etc. Unlike other network monitoring statistical analyses that attempt to analyze the network purely from radio or engineering points of view, CDRs are analyzed principally from the subscribers’ points of view, and we use GIS to do this. In telecommunications, “where” is everything, which is why GIS is a very useful tool in telecommunication - more than half of all decisions made in telecommunications are geographic in nature.

GIS and CDR Analysis
Using a map to analyze CDR data can provide a dramatic improvement over traditional row-column methods, allowing operators to see a precise, up-to-date picture of the entire network, and visually identify the location of subscribers, or the most profitable cell in the network in terms of the amount of time spent originating calls.

Some time ago at MTN Nigeria, we undertook to understand the geographic spread of our subscribers based on their most used cell during working hours over a period of one month. We knew that the engineering network-monitoring tool could achieve this, but we were interested in looking at this issue strictly from a subscriber’s point of view, and hence we decided to use CDRs for this analysis. By looking at CDRs, we can understand activity per customer.

We were particularly interested in understanding the geographic areas within our network where there exists a correlation between subscribers and traffic generated. This became important to us because it was noted that some places have large numbers of subscribers but generate only relatively smaller amounts of traffic compared to other places. From our previous studies of our network, we have come to understand the fact that the number of subscribers in most places is not necessarily proportional to the amount of traffic generated; however some areas within our network are exceptional. This new study will help us to continue to locate these exceptional places, as they portend huge sources of revenue for us.

To achieve this goal of this study, we decided to produce two maps to do a comparative analysis – one map showed the geographic spread of our subscribers and the other showed the geographic spread of traffic generated across the network for the same area. We believed that overlaying these two maps would give us a tool that would help us better size our network.

The data requirements for this analysis are as follows. Each is described below.
  1. Summary CDR data
  2. Equal Power Boundary Areas (EQBAs) or predicted coverage arrays
  3. Vector map layers
Summary CDR
The CDR server architecture differs from one operator to the other depending on the CDR analyzing application used (see Fig. 1). At MTN, an Oracle 9i database is used for this purpose. It provides a very flexible architecture for data storage and access.

The database can be accessed in real-time using an application server, or queried by customer applications such as a billing system, or accessed by a reporting tool such as FACTS Reports. A listener application (a set of tools that allow the operator to pull data directly from the database) is co-hosted with the database server running on the data layer. It collects the data posted by the probes and then updates the database using the collected data

Figure 1: A typical CDR Database and listener configuration. (Click for larger image)


The raw CDRs generated at the switch are called the Toll Ticket File, or TT File for short, and are processed into summarized CDR tables. These summary tables make it possible to aggregate all activities on the basis of the serving cell. It is therefore possible to give a figure for the total number of subscribers and the hours spent on each cell. We accumulated the summarized CDRs generated over a one month period based on subscribers’ most used cells during the working hours and their most used cells during non-working hours.

These data formed the basis for the attribute table required for the GIS application. Table 1 (below) is a sample of the result of a query to find the number of subscribers making calls using each cell as their most used cell over a period of one month and the amount of time spent originating calls.

_

EQBAs or predicted coverage arrays
EQBAs are geo-referenced polygons that represent the radial distance of the signal from the serving cell, based on radio’s interaction with the environment. The planning tool combines a number of coverage predictions into a raster surface, which contains the best coverage values for each location, as well as other information regarding the serving cell. The EQBAs formed the spatial layer used for this particular application.

Figure 2: Equal Power Boundary Areas of Lagos. (Click for larger image)


Vector map layers

Vector layers like city street maps and point location layers were used to further track subscriber behavioral patterns down to the micro level. These helped us to associate geographic areas to the aggregated CDR tables and further helped in understanding what was happening where (see Fig. 3).

Figure 3: Map of Lagos showing land use and road network. (Click for larger image)


Since this analysis used data from a period of one month, it was not necessary to create dynamic maps that showed subscribers’ movement patterns every hour.

The key to joining the EQBAs with the CDR Summary table is the “cell ID.” The cell ID uniquely identifies each EQBA polygon, and is used in the CDR table as well. It was used as the primary key to join the two databases using a simple SQL script. Once this joining had been accomplished, a thematic map, based on realistic range categories, was created that gave insight into the geographic spread of subscribers (see Fig. 3)

The map below (Fig. 4) was produced in September 2005 to show the geographic spread of our subscribers across the Lagos metropolis during working hours. The darker the patch, the more concentrated the subscribers around that particular area.

Figure 4: Showing subscribers’ distribution across Lagos in September 2005 during working hours. (Click for larger image)


A similar map (see Fig. 5) was produced for the same area based on call traffic generated. When this traffic was compared with the subscriber thematic map, it was interesting to note that in some cases, the number of subscribers was directly proportional to the amount of traffic generated; examples of such places are Alaba International Market and Victoria Island. This makes sense, because Victoria Island is the economic capital of corporate Lagos, while Alaba is one of the most important commercial centers in West Africa, where people come from different parts of Africa to buy both household and industrial electrical appliances. We also noticed a common patch at Ajah, which is becoming one of the fastest growing suburbs in Nigeria. It is estimated that a new house is occupied in Ajah every four hours, and this has been going on for the past three years. We were really excited by this result as it showed that for certain areas in Lagos, a high concentration of subscribes did indeed yield high traffic. However, we wondered if we could assume this condition applied everywhere in Lagos. From the study, we further discovered that in most places, this rule of thumb was not applicable. We thus learned that we couldn’t assume this general rule – many subscribers equals high traffic – applies to the entire network; we would have to take each area on an individual basis.

Figure 5: Showing traffic distribution across Lagos Metropolis in September 2005. (Click for larger image)

We were really excited by the result of this analysis and thought it would be interesting to compare the map of calling congestion with the geographic spread of subscribers to see if a large concentration of subscribers had lead to congestion. The result of this second analysis was quite interesting because a large chunk of signaling network resource is usually wasted on a phenomenon called “flash.” A flash, as commonly used in Nigeria, describes a situation whereby a subscriber initiates a call and quickly cuts it before the called party can actually pick it up. People use "flashing" to tell the person they’re calling, “Hey, I am trying to call you but I do not have enough credit to call, so could you please call me?" Most times, the called party will usually respond by calling. So flashing does not last for more than a few seconds but it does rob subscribers who want to make legitimate use of network resources. We discovered that incidents of congestion usually happened where there was a large subscriber base. However, we also discovered that, as simple as this analysis was, we still could not assume it applied everywhere in the network – like the first analysis, we learned we need to take each geographic location on an individual basis.

The above are just two examples in which we have been using GIS to analyze CDRs to our advantage. We have also used the combination of GIS and CDRs as a marketing tool (gathering business intelligence about competitors), network resource dimensioning tool, revenue assurance tool, handset performance and penetration analysis tool, a site selection tool (siting customer care centers), and a routing tool, to mention just a few applications. Comparing the analysis from CDR with other Engineering network monitoring statistics can provide a dramatic assistance to the operator in understanding the network better, with a view to providing unprecedented quality service to the subscribers. This can constitute the leading edge in a highly competitive industry such as the telecommunication industry.

Caveat
The views contained in this article are not those of MTN Nigeria Communications Limited or those of its Management or Board of Directors.

Reference
Ernest C.A. Ndukwe (Nigeria Communication Commission’s Executive Vice Chairman -2004) Welcome Address at the commissioning of the Digital Bridge Institute, Utako.

Sue Marek (2004) Plugging Revenue Leaks.

A Communication Data Storage (Intellectual Property Digital Library)

Surveyor Product Overview (CDR Analyzer –from GL Communication Inc.

CDRLive Product Overview (A CDR Analysis and Warehousing application from LGR Telecoms)

P.A. Burrough (1993) Principles of Geographical Information Systems for Land Resources Assessment

Ireti Ajala (2005) GIS and GSM Network Quality Monitoring: A Nigerian Case Study

Ireti Ajala (2003) The role of GIS in business decision making process of locating Customer Care Centers using GSM subscriber distribution analysis

Ireti Ajala (2003) GIS – A decision making tool – The Experience of MTN Nigeria

Published Wednesday, March 8th, 2006

Written by Ireti Ajala



If you liked this article subscribe to our newsletter...stay informed on the latest geospatial technology

© 2016 Directions Media. All Rights Reserved.