Internet technology is advancing with increasing internet applications demands. However, most developing countries have poor internet experience. Thus, the Quality of Service, QoS of an internet technology, Third Generation, 3G in a developing country, Nigeria was studied. Voice over internet protocol, VoIP and Video Streaming were the internet applications while CODECs and node density were the choice of the performance parameters. Thus, the performance of the 3G network (simulated in Qualnet 6.2) was evaluated based on QoS performance metrics (Average End-to-End Delay (s), Average Jitter (s), Throughput (bps) and Mean Opinion Score). The results showed 3G network can deliver reasonable QoS for VoIP depending on CODEC choice and that node density has adverse effect on QoS. The problem identified for Nigeria was total reliance on mobile broadband and high node density. Upgrading of 3G network, migration to 4G network and revival of fixed networks were recommended to improve QoS.
There seems to still be a gap between the QoS experienced in
developed country and the developing ones despite the global
improvement. For instance, a case of the United Kingdom (a
developed country) against Nigeria (a developing nation). The
average broadband speeds reported by OOKla [
The aim of the research was to evaluate the performance of 3G broadband internet access for households in terms of service experienced by users and make recommendation on issues learnt during the study for better performance and improved end users' experience using Nigeria as the case study.
The following objectives were set in order to achieve the aim;
Present an overview of 3G network for broadband internet access;
Review past papers related to 3G broadband internet access to
residential users; Design the 3G broadband network.; Implement
the design in simulation using Qualnet 6.2 simulator [
This section presents the overview of 3G broadband internet access, 3G architecture and its Quality of Service, QoS.
Universal Mobile Telecom System, UMTS is a Third Generation,
3G telecommunication system which offers advanced data
services (broadband internet) in addition to the primary telephony
service. 3G internet access is classified as broadband [
According to Third Generation Partnership Project, 3GPP
specifications, [
The air interface between the UE and the Node B is called Uu interface while the Iub interface connects Node B to RNC. Iur is used for RNC to RNC connection while IuPS and IuCS are used for connecting the UTRAN to the PS and CS core networks respectively.
The architecture was expected to make 3G deliver good CS and PS service quality. This was based on the 3G Quality of Service, QoS requirements.
UMTS traffic has been classified based on quality of service
requirement by 3GPP into four (4) classes which are
conversational class, streaming class, interactive class and
background class [
Conversational and Streaming classes are mainly intended to be
used to carry real-time traffic flows. The main difference between
the two classes is the direction of the traffic, Conversational
traffic are bi-directional while streaming class are unidirectional.
Examples of Conversational class include VoIP and Video Chat
while audio and video streaming are common examples of
Streaming class. Interactive class and Background are mainly
representing traditional Internet applications like World Wide
Web (WWW), Email, Telnet, and FTP. The main difference
between Interactive and Background class is that Interactive class
is mostly bi-directional, while Background class is unidirectional.
3. LITERATURE REVIEW
[
The papers reviewed characterized UMTS with high delay which
is poor QoS but [
The set up for the research required 3G user(s), 3G network and internet service(s). 4.2
Sequel to the requirement, setting up a live 3G network or using
an existing one for the purpose of the research would have been
too expensive and could have negative consequence on the
network. Thus, the design was simulated using QualNet 6.2 by
scalable technologies [
The UMTS components consist of the User's equipment, UE, Node B, RLC, SGSN, GGSN and HLR. The UEs served different functions in the research depending on the application. UEs were used to make and receive phone calls (in phone call application), make VoIP call (in VoIP application), stream video (in Streaming Application) and access web page (in HTTP application). The parameters configuration discussed in this section was an extract from the entire configuration based on the role each parameter played at different layers. The physical and MAC layer configuration were very essential for the UE, Node B and the connecting air interface between them, Uu. The uplink and downlink channel were configured as 1950 MHz and 2150 MHz respectively. The uplink channel was used for communication from the UE to the Node B while the downlink represents the signal transmission from Node B to UE. The frequencies chosen were based on 3GPP spectrum allocation to Europe and Africa as specified. The antenna for the UE and Node B had similar configuration except for the transmission power and height. The antenna model used for the research were omnidirectional using two-way path loss model. Omnidirectional antenna has the same antenna gain irrespective of direction of the transmitted and received signals. This was chosen to allow even distribution of UEs around Node B for the node density scenario. The simulation area used was 1500 square meters made up of 100 m by 100m square boxes. The coverage area of the Node B was set to maximum distance of 1000m. This was because the research was set to mimic a typical urban settlement. According to NCC, most of the internet access in Nigeria are majorly in the urban settlement. The H323 model was used for the multimedia signaling protocol because it was standardized by ITU-T and has been especially focused on smooth interworking with the PSTN. The GGSN also requires dual configuration for its interface. The UMTS cellular configuration was used for GGSN interface to other UMTS core network components (SGSN and HLR) while the IP configuration was used for GGSN interface to IP Network. This was necessary for routing packets to and from 3G network. The entire wired link, (Iub, Iur, IuPS, IuCS and IuG) were configured to have abstract MAC layer with size of 10Mbps. The other core network components, HLR and SGSN were also configured using cellular model to allow full functionality of the 3G network.
The IP devices also serve different functions like UEs depending on the internet application. IP devices were used receive VoIP call (in VoIP application), video source (in Streaming Application) and web server (in HTTP application). The IP devices where configured to route packet to and from 3G network through the GGSN. 4.3.2
The Internet applications used were VoIP, Streaming (represented by Constant Bit Rate, CBR on simulator), HTTP and Background update representing each of the four QoS classes.
VOIP simulates IP telephony in H323 network. The initiator and receiver generate real time traffic with an exponential distribution function. That simulates a real life telephone conversation. Examples of VoIP application include skype and viber applications.
The Constant Bit Rate (CBR) traffic generator generates traffic at
a constant rate by transmitting packets (also called “items”) of a
fixed size at a fixed rate. It can be used to simulate applications
for which the end-systems require predictable response time and a
static amount of bandwidth is continuously available for the
lifetime of the connection. These applications include services such
as video-conferencing and telephony (voice services). [
This mimic the conventional CS call on 3G network. 4.4
The performance parameters used for the research were CODECS for the VoIP application and the node density for the streaming application. 4.4.1
CODEC encodes audio signals into digitized form at the source.
This digitized signal is sent to the receiver by using packet based
internet. The receiving terminal receives packets and decodes
them to transform into audio signal again using CODEC. Based
on [
The research is also interested in investigating the effect of node density on QoS performance of 3G network. This was particularly important as Nigeria has many users using the mobile network compared to the UK as analyzed in Section 5.5. 4.5
These were the metrics value used in measuring the user's end experience to judge the performance of the network in delivering the internet service to the user. The metrics used are, Average End-to-End Delay (s), Average Jitter (s), Throughput (bps) and ean Opinion Score 4.6 The network design was based on the components identification and configuration discussed.
The design has three different groups of scenarios.
The first group was the scenario used to verify and validate the simulation process. This was the foundation scenario which had all the applications used in the design. This scenario consisted of six users’ equipment, UEs in the UMTS network and three IP devices connected to the gateway. Two of the UEs were making UMTS call conversation, while the other four UEs were making VoIP call to an IP device (VoIP receiver), browsing HTTP website (from another IP device configured as web server), streaming video (from the and forth UE for background traffic both from last IP device configured as the CBR client. This implementation in QualNet 6.2 is as shown in Figure 4.1.
The design implemented in Figure 4.1 was inspired by the 3GPP description of internet access over 3G network which is as shown in Figure 4.3. The Node B (node 5 in the picture) was in charge of communication with the UEs while RNC (node 4) assigned the resources needed by UEs at the Node B. The core network components (HLR (node 1), GGSN (node 2) and SGSN (node 3)) were responsible for authentication of the UEs as well as routing of internet packet within the 3G network.
The second set of scenarios were set to investigate the QoS of 3G
network using VoIP application as an example of the
conversational class which required the highest QoS. This was
important to understand how well 3G performs in delivering
internet access. Unlike [
This was set up was to mimic full implementation of both Packetswitched (PS) and Circuit-switched (CS) applications as described by 3GPP as discussed earlier in Section 4.6.
UMTS call was chosen to represent CS application while VoIP call, Streaming, web browsing and background file download were chosen for PS application using each to represent the four UMTS QoS category as explained in Section 2.3. The simulation period was chosen to be 600 seconds. This simulation period has been maintained throughout the subsequent scenarios. Some of the results behaviours that verifies and validates the simulation are discussed here.
UMTS Call VoIP Streaming HTTP Background In the streaming application, item size was chosen to be 60 bytes with an interval of 0.01s to give an estimated received throughput of 48 000 bps as obtained in Calculation 5.1.
Item size = 60 bytes = (60 * 8) bits = 4800 bits Estimated received throughput = item size / interval = 4800 bits / 0.01 seconds = 48 000 bps
The streaming rate is similar to the rate of MPEG 4 which is a
video type in 3G UMTS device [
Using Calculation 5.1 for the backgroud application gave estimated received throughput of 400 000 bps.
The result obtained after running the configured design in
QualNet 6.2 verified that there was phone conversation between
the two UEs as there was exchange of CS packets between them.
The UMTS call application had no value for PS packets since it
was not a PS application.
There was VoIP communication as well. It was exchange on PS
packets as againgt the CS call and the conversation was
biredirectional. The performance metrics showed a high delay and
a moderate jitter. This was similar to the values obtained by [
The video streaming application also showed a one transfer of PS packets as backed by theory. The estimated throughout obtained from estimated throughput and the obtained throughput are both similar as well. The performance metrics showed a very low delay and a low jitter. This showed a good QoS. Also, the number of messages was equal to the number of fragments meaning there was no fragmentation. This was because the message size, fragment size and the packet size were approximately 60B (which can be obtained in using corresponding values in Calculation 5.3.). The message being smaller than the VoIP traffic is also an indication to while VoIP has more delay than the CBR traffic suggesting that a better compression could give a better QoS. The calculated size of message 60 byte is equal to the item size 60 byte configured for the streaming application.
HTTP and background application result details also followed the internet application pattern. The only notable differe is the fragmentation that occurred in HTTP application as the message size greater than 544 bytes. The framents size is 489.4 bytes as derived in Calculation 5.4 Total data received = 3.67527e+06 bytes Total message received = 383, Total fragment receive = 7503, Size of message (B) = Total data received (B) / Total message received = 3.67527e + 06 / 383 = 9596 B Size of fragment (B) = Total data received (B) / Total fragment received = 3.67527e + 06B / 7503 = 489.8B
Thus, from the analysis above, the design configuration did not only worked (verified) but also work in line with existing theory and similar experiment (validated). 5.3
The approach used was based on UMTS Quality of Service Category. As discussed is Section 2.3, UMTS QoS classification was based on the application requirement. The conversational class required the best performance because it involved real time application and also bidirectional which means the same quality is expected in both directions. The application chosen for the research was VoIP because of its increased usage and also because it has been fore-tipped to compete with conventional UMTS call. This was particularly relevant in Nigeria's context because most people find VoIP calls cheaper compared to normal CS call. This was principally significant when making international calls as the VoIP charging is based on internet usage and not location based unlike CS calls which usually have a higher tariff for international calls compared to local calls. The main QoS target of VoIP was to have a similar or even better performance when compared to the UMTS call. This research set to achieve this using different CODECs. The CODECs used were G.711 and G.729. The choice of G. 711 was because it is the most generally used CODEC and one of the earliest CODEC developed for VoIP while G. 729 was one of the latest developed CODEC for VoIP application.
The ITU describe G.711 CODEC [
The VoIP application only scenario is similar to VoIP implemetation descibed in Section 5.1.2 except that the G729 was in addition to G711. The talking time was later changed from 10s to 20s and 30s to have another set of randomly generated values. The summary of the VoIP application only scenario is summarised in Table 5.4. The speech coding for both G.711 and G.729 were calculated to be approximately 64Kbps and 8Kbps respectively for every scenario. This was compared to the 64Kbps and 8Kbps specified by ITU. This was done for the purpose of validation.
Total Data sent = 2.10224e + 06B = 1.681792+7b Talking time = 263.962s Speech coding rate = 1.681792 + 7b / 263.962s The result obtained from the VoIP application only implementation as extracted in Table 5.5 showed a better performance using G.729 compared to G.711. The range of average delay of G.729 (0.024 to 0.0287s) was low (similar to CS voice call with average delay of 0.0424s in Table 5.2) compared to the G.711 (1.1967 to 2.5190s). The same imporvement was noted for Average Jitter and MOS.
The reason for its better performance was because it has better compression for the messages. The size of a message in G.729 is 20B compared to 160 B in G. 711. Summarily, G.729 offered better QoS performance in terms of average end to end delay, jitter and MOS as it has better compression technique than G.711. The only drawback is the throughput, though not a problem from the user's view as their experience was better but on the part of the network provider, this was significant because the same average resources were applied at Node B to process both traffic. The importance of better utilisation of resources is to accommodate more users with the available resources. The result of the simulation conformed to [20] argument. As a result, it was believed that a 3G network can deliver a good QoS provided the right CODEC choice has been made. The investigation of node density effect on 3G performance will be discussed in next section having concluded on the VoIP single user investigation.
Studying the effect of node density of 3G internet access perfomance was particularly important comparing the approximate node density of the research reference countries, Nigeria and UK. The node density is the number of users per cellular base station. Nigeria has approximately 4403 users to a base station while the UK has approximately 1143 users to base station as derived in Calculation 5.8.
Total number of cellular subscriber in Nigeria = 131 910 228 [
Another indication that the QoS problem in Nigeria could be node density is that Nigeria despite having high number of users to base station, Nigeria has more internet users than UK and most of the internet users depend on the mobile network unlike UK who has both fixed and mobile network.
The node density implementation for the research was done using streaming application only. The choice of streaming application was because it has a predictive performance metrics, throughput (bps). The streaming application has an expected guaranteed rate for the download. This will help identify the point network fails to deliver to expection. Unlike the VoIP whose throughput is always randomly generated. Also streaming class requires good QoS like VoIP. Another reason for choice of streaming application over VoIP was that it is unidirectional. This was particularly important as the UMTS implementation on QualNet 6.2 was based on Release 99 which has upload limitation. The problem of upload limitation has been taken care of in the 3G later release, HSUPA. The implementation of streaming application only scenariosis similar to the implementation in Section 5.1.3. In addition to 60B item used in Section 5.1.3, items of sizes 80B and 100B were used to generate an estimated data rates of 64,000bps and 80,000 bps. The summary of the streaming application only scenarios implementation is presented in Table 5.7. The scenarios for each item size had users in increment of 10.
The result analysis for all the scenarios were based on the throughput because the throughput can be estimated for CBR applications. The result analysis was built on single UEs situation for each item size. 48063.1bps, 64084bps and 80128.9bps were obtained throughput for single UE using item sizes of 60B, 80B and 100B respectively.
The amount of resources allocated by RNC at Node B for single UE were obtained from simulation as 86796bps, 101147bps and 116873bps for 60B, 80B and 100B item size respectively. The difference between the allocated resources and the received unicast throughput caters for the control packets and forward error correction, FEC. The number of data packet for all item sizes were obtained as 59000 packets. This was because the packet transfer was assigned to send an item in 0.1s for 590s (600s (simulation time) - 10s (start time)). This will give a total of 59000 packets (590 / 0.01) as obtained. Also, the number of control packets applied by the RNC was obtained as 30003 packets. This implied that each UE will have resources assigned to cater for the data packets, control packets and the FEC. This analysis of the allocated resources was very important in order to account for the Iub size which was set as 10Mbps.
The node density investigation for the research started with 10 UEs with increment of 10 for the subsequent scenarios. This was done for the item sizes chosen.
It was observed from the result obtained that all the UEs had the estimated throughput of approximately 48000bps for the 60 B scenario up to 80 UEs scenario.
The total received throughput by UE for this scenario was calcluated as 3.84Mbps (80 * 48 000bps) while the total allocated resources by RNC was of 7.4 Mbps of 10Mpbs assigned. However, scenario with 90 UEs for 60B item size showed some UEs were unable to get up to the estimated throughput of 48000bps. The effect of the node density was clearly being noticed at this stage of the simulation. The situation of 90 UEs were similar to the 100 UEs as the number of UEs with reduced throughput increases. The scenario series for 60B item ended with 110 UEs as some of the UEs had negligible throughput. Thus, the analysis of the 60B item set of scenario was done using 10 UEs to 100UEs. The allocated resources for the 100 UEs scenario was 9.2 Mbps of 10Mbps assigned.
Similarly, it was observed from the result obtained that all the UEs had the estimated throughput of approximately 64000bps for the 80 B scenario up to 70 UEs. The total received throughput by UE for this scenario was calcluated as 4.4Mbps (70 * 64000bps) while the total allocated resources by RNC was of 7.2 Mbps of 10Mpbs assigned. However, scenario with 80 UEs for 80B item size showed some UEs were unable to get up to the estimated throughput of 64000bps. The effect of the node density was clearly being noticed at this stage of the simulation. The situation of 80 UEs were similar to the 90 UEs as the number of UEs with reduced throughput increases. The scenario series for 80B item ended with 100 UEs as some of the UEs had negligible throughput.
Thus, the analysis of the 80B item set of scenario was done using 10 UEs to 90UEs. The allocated resources for the 90 UEs scenario was 9.4 Mbps of 10Mbps assigned.
Also, It was observed from the result obtained that all the UEs had the estimated throughput of approximately 80000bps for the 100 B scenario up to 60 UEs. The total received throughput by UE for this scenario was calcluated as 4.8Mbps (60 * 80000bps) while the total allocated resources by RNC was 7.002 Mbps of 10Mpbs assigned. However, scenario with 70 UEs for 100B item size showed some UEs were unable to get up to the estimated throughput of 80000b. The effect of the node density was clearly being noticed at this stage of the simulation. The situation of 70 UEs were similar to the 80 UEs as the number of UEs with reduced throughput increases. The scenario series for 100B item ended with 90 UEs as some of the UEs had negligible throughput. Thus, the analysis of the 100B item set of scenario was done using 10 UEs to 80UEs. The allocated resources for the 80 UEs scenario was 9.6 Mbps of 10Mbps assigned.
Summarily, the number of UEs was increased to maximum of 100, 90 and 80 for the 60B, 80B and 100B item sizes respectively. Having understood the pattern of the scenario implemented for node density, the analysis of the QoS performance metrics is as follows; The values obtained were used to plot graph of average unicast received throughput per UE as shown in Figure 5.3.
The graph of the throughput (ploted in MATLAB) gave a constant throughput for some number of UEs. This is the normal behaviour expected because the traffics have been generated using CBR which normally generates at constant rate. However, at some point where the node density (Nd), was approaching the network limit in terms of available resources, the throughput dropped gradually with increase in number of UE.
The values for average end-to-end delay, Dee per UE were obtained and plotted as shown in Figure 5.4. It was observed that the Dee was directly proportional to the item size, Si. This was because the total size of items processed increases with item size. For instance, the scenario for 10 UEs sees the 3G network processing 600B size (60B * 10UEs) of item for 60B item size scenario but this get higher for 80B item size scenario with 800B size of item (80B * 10UEs) while 100B size item scenario has the highest of the three as 1000B (100B * 10UEs). It was also observed from the set of results obtained that the average end -to-end delay, Dee was increasing at constant rate with increase in node density, Nd (Dee = K*Nd) until the normal point where all scenarios experienced sharp increase in the delay. This is because as the number of UEs increases, the number and size of packets to the processed increase as well.
It was also observed that the average jitter, J was increasing at constant rate with increase in node density, Nd (J = K*Nd) until the normal point where all scenarios experienced sharp increase in the jitter. This is as plotted in Figure 5.5.
Summarily, the node density have a negative effect on the 3G QoS in terms of the average unicast received throughput, average end-to-end delay and average jitter.
The throughputs generated were within the range specified by
3GPP as 34Kbps to 384 Kbps for one way video streaming. The
work of [
The scenarios established in this section showed that 3G network can deliver good QoS for it users provided all the conditions are right. The instance used in this research was VoIP application which requires a very small delay and jitter. The research achieved good QoS for 3G network using G.729 CODEC instead of G.711. Also the effect of node density was investigated on QoS as this was one of the major problems facing Nigeria’s internet.
This Section discuss the possible ways to improve the situation as highlighted in the research aim and objectives.
The research believed that if all or most existing 3G networks in Nigeria can be upgraded, a higher capacity and better quality will be delivered to the users. Depending on 3G networks only might not be sufficient for a country like Nigeria. Other possible solutions which can be complementary to the one discussed above will be discussed in Section 6.2 and 6.3.
Advantages of migrating to 4G include but not limited to: presenting users more QoS as well as better capacity and coverage. Hence it is an important consideration for Nigeria to have a proper plan for 4G mobile broadband in their vision for better internet access for her residents.
As shown in Calculation 5.8, UK has better node density on cellular network and also have viable fixed network. It will also help both the QoS and node density if Nigeria can work on improving their fixed network.
The three recommendations will improve internet QoS and will also share the users thereby reducing the effect of node density.
The research evaluated the performance on 3G internet access in order to suggest ways to improve the internet access in developing countries like Nigeria as highlighted in the research aims and objectives. 3G network was simulated using QualNet 6.2 as described in Section 4. The QoS was investigated using VoIP application and the result showed a better performance for this using G.729 CODEC compared to G.711 as discussed in Sections 5.3 and 5.4. Also, the analysis in Section Sections 5.5 and 5.6 also showed that node density has negative effect on 3G QoS in terms of average end-to-end delay, average jitter and throughput. Three recommendations (which are improving the 3G network, migration to 4G network and having a fixed broadband as a complementary service) were proposed to improving internet accessing Nigeria in Section 6.
There were some technical constraints (related to systems and simulator) during the research. This limited the coverage of the research in terms of its absoluteness. Also, it was difficult getting real life data for implementation by telecommunication industries as they attached commercial values to this.
Comprehensive assessment of the state of infrastructure for different broadband technologies in Nigeria and the economic impact of implementation of each is strongly recommended in order to make a more informed decision on moving forward towards achieving a good QoS for her residents.
All praises and adorations are due to Almighty God. We appreciate the support of Professor Nigel Linge, all the lecturers and non-teaching staff of the University of Salford where the research was conducted. Our profound gratitude goes to our families and friends for their unquantifiable support.