Mobile Broadband Networks Performance Modeling Based on Experimental Studies

W. Hashim,S. Dzulkifly,N.A Abd Ghafar Wireless Communication Cluster, MIMOS Berhad,

Technology Park Malaysia, Bukit Jalil, 57000 Kuala Lumpur, Malaysia.

wahidah.hashim@mimos.my

A.F. Ismail, K. Abdullah Dept. of ECE, Faculty of Engineering, International

Islamic University Malaysia, Kuala Lumpur, Malaysia. af_ismail@iium.edu.my

Abstract—Real time performancemeasurements of mobile broadband network can beutilized in the process of developing a model that capable to accurately outline the behavior of anactual network. Such knowledge of network behavior can help operators in examining the current network performance as well as predicting its future condition.The model is anticipated to be able to providereliable estimation of connectivity speed to mobile broadband users. Thepaper outlinesstochastic live network behavior deduced from performance of two mobile broadband network types namely the HSPA and IEEE 802.16e. The reported findings are an initial research outcome that is to be tested out in the next phase of the project. The real time download data rates were examined usingLinux Operating System.Identified networkparameters are the received data rates as well as the signal strength indicator (RSSI). Obtained results show the variation of the network performance over time which can provide the knowledge of the best and most reliable network to be connected.

Keywords-component; Heterogeneous wireless networks, always best connection, HSPA, IEEE 802.16e, mobile broadband, wireless communication, RSSI, data rate, probability distribution,network intelligence.

I. INTRODUCTION The penetration of wireless broadband services has

shown an exponential increase due to the flexibility of the services in providingconnectivity while being mobile [1]. According to Akamai Technologies, one of the key factors that influence the broadband penetration is good cumulative average speed offered by various service providers within thatcountry [2]. As of in the fourth quarter of 2011, the average global speed is 2.3 Mbps [2].

This highlightsthe importance of considering the connection speed as one of the key factors influencing the growth in the numbers of broadband users [2]. Combining both connection speed and the signal strength as the deterministic parameters in the examination of network condition could help the process of securing good connectivity. A complex system embedded with continuous monitoring capability of both parameters will be required. The system is expected to be operating foran extensive period of time. Therefore, to solve this issue, it is suggested that the system has a pre-defined knowledge developed

from a probability model derived within a specific period during themonitoring process.

There are several existing network monitoring softwares available in the market that can be used to observe the condition of wireless network such as Wireshark, Kismet, WiFi Sniffer, InSSIDer, and others [3, 4, 5]. However,these softwares arelimited to monitor only the WiFi/Wireless LAN and the Ethernet. As for mobile broadband subscribers, the network tool usually came as a package along with their subscribed services.Nonetheless, thesenetwork tools only provide the network availability information of that particular time. No network assessment is executed to supervise the current network state as well as anticipating future condition. Apart from that, there are also networktools that are used to administer the network activity. These tools however, are usually used in a wider network scale and are not suitable for individual analysis [6].

There are several attemptson modeling the network condition using probability theory. Bootstrap approximation method had been proposed to be used in the evaluation of services parameters’ dynamic quality [7].This is due to the characteristics’ instability of the parameters over a short period of time within the network. This method, however, comprisesa complex mathematical modeling that could overwork the whole process. In addition, this approximation method computes the sample by relying on the independent assumption which is not suitable when dealing with real time monitoring. Meanwhile, position prediction mechanism (PPM) had been suggested to be used to predict the location of mobile users within the cells[8]. The prediction modelis quite straight-forward but is not applicable for observing the network behavior.

This paper presents the report onnetwork behavior studies involving the fundamental real time network condition indicators which are the RSSI and the data rate (connection speed). In addition, the paper also elaborates the distribution model that best represents the data.The rest of the paper is organized as follows; section II presents the studiesmonitoring setup, while section III discusses themobile broadbandreal data probability modeling. Section IV elaboratesthe experimental results and finally, section V concludes the studies.

This study is part of the research funded by Ministry of Science, Technology and Innovation Malaysia (MOSTI)

II. SYSTEM MODEL This section details the specified hardware and operating

system that wereused to capture the network data that later to be converted into probability. Two subscribed mobile broadband services areHSPA mobile broadband service and IEEE 802.16e service.

A. Experimental Setup Subscribed mobile broadband services usually comes

with a widget package that would notify the signal strength and the connection speed to users as shown in Fig. 1 below.

With reference toFig. 1, dotted circle indicates the network signal strength while the full circle shows the connection speed.

Figure 1 Mobile Broadband Graphical User Interface (GUI)

Continuous automatic data retrieval that works like the normal GUI was developedusing open source platform Linux. Thenetwork conditions wereretrieved repetitively over a period of time using the bash script of the programming language.

Fig.2 illustratesthe experimental setup involvingtwo subscribed services, integrated into acomputer that emulates the platform for dataretrieval using Linux Operating System (OS). These two subscribed services (HSPA and IEEE 802.16e) are monitored through Linux terminal assisted by the specified bash script commands.As shown in Fig. 2, the HSPA and IEEE 802.16e are embedded into the USB port of the computer.

Figure 2 Laboratoryexperimental setup ©2012 MIMOS Berhad.

All rights reserved

B. Process Flowchart Fig.3 outlinesthe steps involved in developing the

network performance model.

Figure 3Steps in developing the network performance model©2012 MIMOS Berhad. All rights reserved

At the beginning of the process, data retrieved through stochastic technique wereanalyzed before converted into the probability mass function (p.m.f.). Once a model had been confirmed, it wasvalidated before it can be deployed as the input for further network decision in providing best connectivity based on each connectivity performance.

III. NETWORK PERFORMANCE MODEL The data retrieved through network monitoring is

typically in a random and discrete manner which can be modeled into a specific probability mass function. The initial observation from the retrieved data has shown existence of two graph pattern types which uniform and positively skewed. These two distribution patterns were studied inorder to verify the initial assumption.The model was verified by testing it against the chi-squareand positively skewed test.

A. Chi-Square The chi-square is being calculated by assuming that the

discrete data samplesarecollected within a specific time intervalfor the perceived uniform pattern. In this case(1) would be involving n as the number of data samples [9].

(1)

where; a = 1,2,3 …..n

The basic chi-square test calculation, χ² is as follows [9]:

(2)

where; Oi= observed data

Ei= expected data

The calculated chi-square is compared with the tabulated chi-square which in this case, the degree of freedom, ν; is shown in (3):

ν = a - 1 (3)

Null, Ho and alternative hypothesis, Hı are manipulated as initial assumption of this chi-squareprobability whereby [9]:

Ho = Not uniform (4)

Hı = Uniform (5)

Under this assumption, H�is conformed againstthe perceived uniform modelif [9]:

(6)

Otherwise, the data is assumed to fulfill the alternate hypothesis, Hıif [9]:

(7)

Through calculation, it was verified thatthe data fulfill the alternate hypothesis, Hısince the calculated chi-square value is largerthan the tabulated chi-square value. It can be consideredthat the data is uniformly distributed.

B. Positively Skewed An asymmetric graph is said to be positively skewed if it

follows the following condition [11]:

(8) The methods of measuring the positive skewness can be divided into three manners which are [11]:

i- Karl Pearson’s skewness ii- Bowley’skewness iii- Kelley’s skewness

In our experimental studies, we are using the Karl Pearson’s skewness method [11]whereby the skewness coefficient, Sk is defined in eq. (9):

(9) where; σ = standard deviation

However, in the case where mode, Mocannot be

identified clearly, the empirical relation between mean, median and mode define:

(10)

Since mode is not utilized, the skewness coefficient is then obtained by eq. (13)

(11)

From the deduced skewness coefficient value, it can be further verifiedthat the perceived pattern follows skewed positive graph if:

Sk> 0, positively skewed (12) Sk< 0, negatively skewed (13)

The positively skewed distribution usually shows the tail

of the asymmetry curve directed towards the left as depicted in Fig. 4 below [11].Fig. 4 below is the generalized illustration of the positively skewed distribution graph.

Figure 4 Positively Skewed Distributions [11]

In the study, it was able to verify that good received signal does not provide enough information of the network condition. This is shown from findings of the RSSI and data rate included in the sub-section below.

Probability, Pb

x

IV. EXPERIMENTAL RESULTS The monitoring of both IEEE 802.16e and HSPA mobile

broadband have shown variation that when verified, follows the chi-square and positively skewed distribution test.Note to be taken, the monitoring data is captured within a specific time stamp in real time environment where severe fluctuation and low data rates were caused by the saturated usage during the day time. This phenomena, however, is an ideal environment in modelling the data rates to achieve our main target of mapping the best data rates between the two services.

A. Received Signal Strength Indicator (RSSI) The RSSI indicate thenetwork signalstrength level. It

can be used to infer the conditions of the network since higher RSSI value is usually assumed to have agood connectivity.

Fig. 5 depicts the monitored average RSS level for IEEE 802.16e as well as HSPAmobile broadband service. The duration is from 8:30 amuntill 4:00pm throughout 7 days.

Figure 5 RSSI for IEEE 802.16e and HSPA Mobile Broadband

It can be observed that theRSSI for both IEEE 802.16e service and HSPA mobile broadband services are almost stable, perhaps due to the static positioning of the monitoring setup. The fluctuations experienced by HSPA mobile broadband service might have occurred due to interference that exists within the environment setup.

There are likely exists several interference sources such as the others unseen RF transceivers within the environment setup and/or due to physicalintervention to the static positioning of the setup.

Under the assumption that both IEEE 802.16e and HSPA mobile broadband RSSI arein discrete uniform, the obtained probability distribution graph is plotted in Fig. 6.

Figure 6 RSSI P.M.F. for IEEE 802.16e and HSPA Mobile Broadband

From the above result picture the RSSI distribution probability for both IEEE 802.16e and HSPA mobile broadband is seen to be the same.

The reason of theequiprobablebehavioral level might be due to theircharacteristicsthat illustrate the pattern of the random discrete data. In other words, even though the RSSI values for both IEEE 802.16e and HSPA mobile broadband is different, both services show RSSI stability throughout the period.

B. Data Rate (Connection Speed) Another additional aspectthat can be considered as one

of the key factors that gauge the network condition is the connection speed. The data rate or connection speed is usually obtained by generating traffic such as file transfer. In the study, video streaming was used as the traffic generator in the attempt to quantifycurrent network’s connection speed.

Fig.7 portrays the captureddata rate for both IEEE 802.16e and HSPA mobile broadband.

Figure 7 Data Rate for IEEE 802.16e and HSPA Mobile Broadband

The variation of the data rate proves that the RSSI alone is not able to fully describe the condition of the network.

0100200300400

8:30

9:00

9:30

10:0

010

:30

11:0

011

:30

12:0

012

:30

13:0

013

:30

14:0

014

:30

15:0

015

:30

16:0

0

Data rates(kbps)

Time of the day (a.m./p.m.)

IEEE 802.16e HSPA

100-90-80-70-60-50-40

8:30

9:00

9:30

10:0

010

:30

11:0

011

:30

12:0

012

:30

13:0

013

:30

14:0

014

:30

15:0

015

:30

16:0

0

RSSI(dBm)

Time of the day (a.m./p.m.)

IEEE 802.16e HSPA

0.030.040.050.060.070.080.09

Probability, Pb

Time of the day (a.m./p.m.)

IEEE 802.16e HSPA

From the figure, IEEE 802.16e services initially shows good data rate while HSPA mobile broadband shows lower data rate. It can be suggested that thelow data rate for HSPA mobile broadband service might have occurreddue to high population of users within the base station distribution.

The IEEE 802.16e service did show higher data rate, at the start of the day. Nonetheless, as the hours passed, it started to deteriorate to the point where its data rate is even lower than the HSPA mobile broadband services.

Figure 8 Data Rate Probability for IEEE 802.16e and HSPA

Mobile Broadband

Fig.8 demonstratesthe probability distribution for both IEEE 802.16e and HSPA mobile broadband services. The blue line indicatesHSPA mobile broadband service while green line represents the ideal curve of positively skewed distribution.

It can be assumed that HSPA mobile broadband service showsconsistent stability although it has lower data rate connectivity. On the other hand, theIEEE 802.16e was observed to provide higher data rate than HSPA mobile broadband in certain periods of time. Following suchfindings, a network management system or operators can utilizesuch results to determine whenever one network is down, and the system can be switched to the better network.

V. CONCLUSIONS

From experimental studies and probability analyses, the findingssupporttheinitial hypothesisthat connection speed, combined with the RSSI value can be exploited to derivemore accurate network condition information.This probability model can be used as a benchmark in determining the network with best connectivity at all time or certain duration of time. Theproposed probability model can be used as a reference to structure a more extensive and

accurate model that could promote a reliablemobile broadband connectivitywithin a country. Future work will include more genericand complete real data retrieval that includes a longer window frame of data monitoring.

ACKNOWLEDGMENT The studies are part of deliverables to Ministry of

Science, Technology and Innovation of Malaysia. The reported research findings are part of the main research funded under Sciencefund.It is a collaboration project under supervision of MIMOS Berhad incorporating researchers from International Islamic University (IIUM) and Universiti Teknologi Malaysia (UTM).

REFERENCES [1] E. Pigliapoco and A. Bogliolo, “Enhancing Broadband Penetration in

a Competitive Market,” 2010 2nd International Conference on Evolving Internet, no. D, pp. 159-163, Sep. 2010.

[2] I. Insight, “The State of the Internet, 4th Quarter 2011 Report,” Akamai Technologirs Inc., vol. 4, no. 4, 2011.

[3] Retrieved May 25, 2012, from www.wireshark.org/

[4] Retrieved May 12, 2012, from www.kismetwireless.net/

[5] Retrieved May 12, 2012, from www.metageek.net/products/inssider/

[6] M. Sidibé and A. Mehaoua, “QoS Monitoring for End-to-End Heterogeneous Networks Configurations Management,” 2008.

[7] E. H. Ong and J. Y. Khan, “Dynamic Access Network Selection with QoS Parameters Estimation: A Step Closer to ABC,” VTC Spring 2008 - IEEE Vehicular Technology Conference, pp. 2671-2676, May 2008.

[8] C. Gu, M. Song, Y. Zhang, and J. Song, “Access network selection strategy using position prediction in heterogeneous wireless networks,” Frontiers of Electrical and Electronic Engineering in China, vol. 5, no. 1, pp. 23-28, Sep. 2009.

[9] Cohen L., Manion L., Morrison K., Morrison R.B. K., Research Methods in Education, NY, USA: Routledge, 2007, pp. 525

[10] Frederick J. Gravetter, Larry B. Wallnau, Essentials of Statistics for the Behavioral Sciences,USA: Cengage Learning, 2011, pp. 545 – 549.

[11] Naval Bajpai, Bussiness Statistics, India: Pearson Education India, 2009, pp. 136.

[12] F. Mekuria, “Enabling Wireless Broadband Technologies and Services for the Next Billion Users,” no. September, pp. 13-15, 2011.

[13] E. V. A. G. Ustafsson, A. N. J. Onsson, and E. R. R. Esearch, “Always Best Connected,” no. February, pp. 49-55, 2003.

[14] S. A. Sharna, M. Murshed, and M. Ieee, “Adaptive Weight Factor Estimation from User Preferences for Vertical Handoff Decision Algorithms,” pp. 1143-1148, 2011.

[15] E. Pigliapoco and A. Bogliolo, “Enhancing Broadband Penetration in a Competitive Market,” 2010 2nd International Conference on Evolving Internet, no. D, pp. 159-163, Sep. 2010.

0.058

0.06

0.062

0.064

0.066

8:30 9:30 10:30 11:30 12:30 13:30 14:30 15:30

Probability, Pb

Time of the day (a.m./p.m.)

HSPAIEEE 802.16ePositively Skewed Curve (IEEE 802.16e)