ARTICLE A QoS Estimation Algorithm from Caller Ringtone Analysis in GSM

estimation. ABSTRACT – Call Setup Time (CST) is one of the key performance indicators (KPIs) that Mobile Network Providers (MNPs) are mostly appertain. It has been established that long CST usually severely affects the user experience. Owing to the limitations associated with gleaning the CST data from MNPs, this paper provides the development of QoS estimation algorithm from various CST parameters. The algorithm involves the determination of CST; Inter-Burst time; Intra-Burst time and Call duration in time domain. The caller Frequency content was also determined by the application of fast Fourier Transform before computing the Mean Square Error (MSE). The eventual QoS rating is done after the computation of the MSE from various individual parameters. Four hourly data consisting of 10 sets each were collected three times in a week for four weeks for each MNP’s for creating Caller Ringtone dataset and testing the developed algorithm. Performance analysis of the system in accurately determining: CST; Intraburst time; Interburst time and Call durations were carried out. Results obtained shows that the proposed technique accurately computes these parameters and maximum error obtained was to the value of 10%. Furthermore, the QoS obtained shows an error margin of less than 5 % was observed when the developed technique was compared to the ground truth. Thus, the proposed algorithm was able to compute the QoS using Caller Ringtone only, thus independent of


RELATED WORK
As demands on mobile communication increases, the ability of MNPs to plan and manage capacity, and to deliver superior service to customers, is being severely challenged. Owing to increasing number of users and several smart devices and applications, MNPs have been experiencing some challenges in the areas of improved coverage, cost, channel capacity and QoS [9][10][11]. A problem in hardware, transmission, coverage, or interference may result in an increase in the call setup time. A faulty Transceiver or combiner or a wrongly connected RF cable makes seizing of the SDCCH or TCH complex, and thus resulted in call setup time been increased. Call Setup Time (CST) refers to the time for an end-to-end call to be set up by MS through radio network equipment.
It has been established that long CST usually severely affects the user experience. Hence, CST is one of the key performance indicator that MNP's are most concerned about as it provides a measure to regulate and guarantee the QoS requirements [7,12]. Author in [13], described and evaluates real call set up success rate in GSM and also states the possibility of its implementation using the current technologies in GSM and difference between the real and implemented CSSR. The "real" as the Kollár explains means that CSSR is calculated as ratio of the assigned traffic channel (TCHs) to the channel requests. It was concluded that more complex formulation which utilized the Immediate Assignment Success Rate (IASR), TCH Assignment Success Rate and SDCCH Success rate must be used for measuring CSSR. Kollár further stated that this formulation was the best approach despite a higher effort on the processor part of the equipment where the CSSR is to be calculated. It was noted that the formulation did not cover the case when the Direct TCH Assignment feature is enabled.
Author [1], gave insight into the causes of call setup failures in a GSM service test area was studied and the use of RF optimization process to increase call success rate was presented. The adopted methodology is drive testing, post processing and data analysis. The authors conclusion were that most of the network problems are caused by increasing subscribers and the changing environment and also that RF optimization should be carried out frequently in order to improve the network performance with the existing resources [14][15]. CST can be estimated by drive test through the calculation from L3 messages. For GSM network, CST is the period from Requesting a Channel until Alerting, it is usually 7-8 s for Mobile-to-Mobile Calls. The CST measurement can be achieved through traffic measurement and Drive Test (DT). However, based on the need for self-determining measurement, we examine the DT approach, which guarantees independence in the next section [5,16].

Drive test
The procedure used in Drive Test involves using a car containing mobile radio network air interface measurement equipment that can detect wide range of the physical and virtual parameters of mobile cellular service in a given test region. Coverage, capacity and QoS of a mobile radio network can be evaluated with Drive Testing method. Drive test equipment typically collects information and services running on the network such as voice or data, radio frequency scanner and GPS information to provide location logging [2,17]. Drive tests is aimed at tracing the signalling on the Um interface and on the Aibs interface in different scenarios hence, the signalling process can be analysed comprehensively and the problem can be located easily [9,[18][19]. There are diverse types of tools to carry out a DT among which are: JDSU E6474A v15.2, TEMS Investigation, Nemo Outdoor. However, in this work, an algorithm was developed using MATLAB and LabView to estimate QoS from the analysis of caller ringtone in GSM network. The result gotten will then be used in estimating QoS as explain in the next section.

METHODOLOGY
The methodology used in this research is divided into 4 blocks each block set to achieve each objective. A test area was selected; Data were collected every 4 hours i.e., 2am, 6am, 10am, 2pm, 6pm and 10pm three times in a week i.e., Mondays' Wednesdays' and Saturdays' for 4weeks and for each hour 10 data (caller ring tones) was collected. The caller tone dataset was then Pre-processed and analysed using the developed algorithm and bench mark against results obtained from manual determination of the caller tone parameters. A standard caller tune file for Globacom was downloaded from the internet as ground truth caller tune. The file was converted with the format converter to a wav file and was fed into the developed algorithm to extract the call parameters for computation of necessary QoS. These parameters were then use as a standard against our dataset to compute the MSE as well as QoS estimation of MNPs base on the difference between the ground truth (downloaded tone) and our recorded tone for selected tone 1-10. In estimating the Qos of a caller tune, the Euclidean Distance Accuracy Measure has been used. The Euclidean distance measures the closeness between the determined caller tune parameters and the ground truth.

Direct listening
The ring tones were listened to by few people one after the other to choose the best of each 10 set. The chosen tone was fed into the design to extract the relevant call parameters such as (silent mode time (SMT), InterBurst Time, IntraBurst time, Call Duration, and Number of Burst). Also, some other parameter like mean squared error, Amplitude of the burst, frequency, Signal-to-noise ratio (SNR) etc. were used for further analysis. journal.ump.edu.my/ijsecs ◄

Globacom standard
These MNP has their SMT to be 7 seconds for calls between their customers, it has a standard pattern and sound which completely differs from other network hence the ring tone that best suit this standard was choose from the set. The chosen tone was then used as an Observe Value (y) in QoS estimation.

Test area
Since the aim of the research was to assess the QoS provided by mobile operators as perceived by the user, areas where these services are mostly used were selected, i.e., Suburban Districts of Abuja; these districts are not within the Federal Capital City (FCC) but because of their proximity to the FCC, they have attracted some development and many people who work in the FCC live in these suburban districts. For the purpose of this research Lugbe district is chosen. Lugbe; under Abuja Municipal Area Council (AMAC) is one of the fast-growing suburban settlements in Abuja. It is largely residential and densely populated. Lugbe is divided into five districts namely Lugbe south, Lugbe north, Lugbe central, Lugbe west and Lugbe east. Lugbe has been brought into lime light because of its proximity to the city centre and to the Abuja airport. This made significant development to be attracted to the area.
Heavy traffic is experienced in the early hours of the morning between 6.00am to 10.00am and 4.00pm to 8.30pm on the lanes from Lugbe to city centre and city centre to Lugbe respectively. Journey to or from Lugbe to city centre could take well over one hour during these periods

Proposed QoS estimator algorithm to estimate QoS from caller ringtone in GSM network
QoS estimator from caller ring tone in GSM network is being proposed as part of ongoing research using a citizen sensing approach. This section only focuses on the discussion of the proposed algorithm for QoS estimation which consists of four distinct stages as shown in Figure 1.  The technique also allows the caller ring tone parameters to be analysed and their mean square error (MSE) was used with respect to SMT to estimate the QoS of the MNPs under observation. Figure 3 shows the flow chart for the caller ringing tone acquisition.  (1) Where is the silence period end time and denotes the start time of the silence period. Also, Intra Burst Time (IntraBT) can be estimated by finding the difference in the time the first peak is detected from the time the end of a burst is detected. ( Where and is the start and end of burst respectively. Value for Inter Burst Time (InterBT) can be gotten from subtracting the end time of the detection first burst from the time the next burst started.
InterBT t t =− journal.ump.edu.my/ijsecs ◄ The process involved in analysing sound quality includes Signal in Noise and Distortion (SINAD) measurement, Amplitude measurement and Frequency measurement. This is carried out by following four distinct stages as shown in Figure 5. It was observed that the input signals have sudden and infrequent zero values occurring due to high sampling frequency of 8000Hz per samples and this affect the computational procedure. However, averaging method was used to annul the presence of these zero values in the signal i.e., we were able to effectively compress 8000 samples per second to 8 samples per second without distorting the signal. It was also observed that values that are less than 0.01 are noise; thresholding method was then used to reduce them to zero. The resulting signal is then differentiated to facilitate the effective tracking of gradients. A positive gradient indicates the start of a burst and the return of the gradient to zero denotes the end of burst and the start of an inter burst time. The algorithm used for sound quality analysis for calls originating from different mobile network providers to other mobile network providers in a fairly quiet environment is shown in Figure 6 while Figure 7 is the labview design of the sound quality analyser. journal.ump.edu.my/ijsecs ◄

Preliminary results and discussions
In this section, the caller tone dataset was analysed using the developed algorithm and bench mark against results obtained from manual determination of the caller tone parameters. A standard caller tune file for Globacom was downloaded from the internet as ground truth caller tune. The file was converted with the format converter to a wav file and was fed into the developed algorithm to extract the call parameters for computation of necessary QoS. These parameters were then use as a standard against our dataset to compute the MSE as well as QoS estimation of MNPs base on the difference between the ground truth (downloaded tone) and our recorded tone for selected tone 1-10. journal.ump.edu.my/ijsecs ◄ The normalized absolute error used in the measure of the performance of the system is given as: (4) Where a1m is the measure parameter and a1g is the graound truth. Analysis was carried out on the different hour ranging from less busy hour to the busiest hour in the dataset. Table 1 shows the caller tone parameters for Airtel dataset for day1. It is worthy of note that are all time used throughout this paper are measured in seconds.

Performance Analysis of the technique in detecting Silent Mode Time (SMT)
Monday is regarded as the busiest day in a week in Nigeria. From observation of Globacom network, with 2am being regarded as the time with less traffic as most mobile users are sleeping. Performance analysis was carried out on the dataset obtained on Airtel network at 2am dataset. Performance analysis of the developed algorithm in detecting SMT from dataset is as shown in Table 1a and Table 1b.     Table 1a and Table 1b shows some selected results obtained from the determination of SMT on a less-busy and mostbusy hour on Monday while Table 1c and Table 1d shows the performance analysis of the developed algorithm in detecting SMT in less busy hour and busy hour respevtively. The result shows that the maximum error recorded from the developed algorithm as compared with the manually computed results is 4.21%. In some instances, the computed error was as low as 0.0% and the average percentage error was 1.2%. Also, it is evident in Table1a and Table that the error during the off peak i.e., 2am is far lesser than that of when the system is busy i.e., 6pm. This shows that as the system becomes busy, the QoS deteriorate. Similarly comparing, results of similar hours on Monday with that of Wednesday shows that more calls happened during the early hours of Wednesday than on Monday, thus the reason for increase in the error recorded on Wednesday. Generally, the results obtained from Table 1a-Table 1d shows that the developed algorithm is highly accurate in the determination of SMT and is a good measure of QoS parameters.

Performance analysis of the technique in detecting number of burst
The number of bursts in a caller tone plays a significant role in the determination of GSM QoS; hence the performance of the developed algorithm in accurate determination of the number of bursts is presented in this subsection. The performance has been done on 2am and 6pm dataset that represent both less busy and most-busy period on a busy day. Results obtained are discussed herewith.     Tone-1  10  10  0  0   Tone-2  11  11  0  0   Tone-3 Table 2a and Table 2b shows the results obtained in accurately detecting number of bursts in different caller tone on a Monday while Table 2c and Table 2d shows the results obtained in accurately detecting number of bursts in different caller tone on a Wednesday. Similar to the analysis in detecting SM, two scenarios have been reported in this subsection also. It can be observed that the error in the determination of number of bursts shows significant difference between 2am calls and 6pm calls. This also justify the earlier assertion that MNP recorded more calls during 6pm peak duration. Furthermore, comparing Monday data to that of Wednesdays shows significant increase in error; does an indication of reduced QoS on those days. In conclusion, the developed algorithm has the capability to accurately determine the number of bursts in different scenarios.

Performance analysis: QoS estimation
In estimating the Qos of a caller tune, the Euclidean Distance Accuracy Measure have been used. The Euclidean distance measures the closeness between the determined caller tune parameters and the ground truth. Parameters used to compute the final MSE include: SMT; InterBurst Time; IntraBurst Time; No of Burst; Call Duration and call Frequency. The MSE is computed as follows: The Euclidean distance accuracy measure is given as   Table 4 shows the analysis on the dataset for different MNPS at 2am. It can be observed from the Table 4 that mobile network connections between Airtel network and GLO network at 2am is good as the average MSE is 0.78 which is rated high with respect.

CONCLUSION
This study was undertaken to determine call setup time and ringtone quality in GSM network independence of the mobile network provider. For the purpose of this study GLOBACOM Nigeria was used as the source. We gathered ringtone every four hours for a month in one of the sub-urban regions of Abuja using a SAMSUNG GALAXY S4 and SAMSUNG GALAXY A3 although data were collected from other networks too to have a general knowledge of the ring pattern. Our data revealed that ringtone parameters for the networks differs even when measurement is taken in the same region and also, we discover that SMT has a great impact on the quality of service of the mobile network, the longer the SMT the higher the possibilities of call drop hence the lower the QoS of the mobile provider. Mean square error (MSE) is used as a performance index that indicates how well the algorithm can adapt to a given solution. A small minimum MSE is an indication that the QoS estimator design is accurately modelled and can predict, adapt and converge to a given solution. However, a very large MSE usually indicates that the system is not accurately modelled.