M A S T E R ' S T H E S I S
Contention-based uplink transmission in unscheduled access
Lilian Sunna
Luleå University of Technology MSc Programmes in Engineering Computer Science and Engineering
Department of Computer Science and Electrical Engineering Division of Computer Communication
2010:009 CIV - ISSN: 1402-1617 - ISRN: LTU-EX--10/009--SE
Contention-based uplink
transmission in unscheduled access
Lilian K. S. Sunna
LTU
Dept. of Computer Science and Electrical Engineering
Ericsson AB
Dept. of Wireless Access Networks Div. of Wireless IP Optimizations
January 15, 2010
A BSTRACT
Voice communication has traditionally been carried out on circuit switched networks only.
This is not true today since some voice communication is done over IP-networks, such as the Internet. A general term for this type of voice communication transmission is Voice over Internet Protocol, or simply VoIP. The main benefit of VoIP is its cost-efficiency concerning both system maintenance and international calls. VoIP-services will probably have an increased popularity with the switch to Long Term Evolution (LTE), due to its high capacity and bit-rates.
Voice communication is highly sensitive to end-to-end delay. Obtaining transmission re- sources in LTE using the conventional contention-free transmission method contributes to the end-to-end latency. This is due to the scheduling signalling which can only be done in specified intervals on a scheduling channel, so called D-SR intervals. The D-SR interval can be long if the load is high.
This work investigates the possible capacity gain for VoIP-users by using an alternative contention-based transmission method. This method uses a pre-set broadcast grant to be used by any user. However, as the load increases, the contention from several users will become more common since the grant is shared by everyone. Hence, restrictions to the alternative transmission method in order to decrease the load are also analysed.
Two different grant methods are evaluated: the conventional contention-free and the contention-based. A system simulator is used to investigate the two grant methods. The simulator models a complete LTE system in a multi-cell environment. The simulations show it is necessary to limit the contention-based access to only VoIP packets of a cer- tain size. Unlimited use of contention-based access is clearly shown to be useless: the capacity is zero VoIP-users per cell. Further on, when comparing contention-free access to contention-based limited to VoIP packets for different D-SR intervals, it is found that the two methods have similar VoIP-user capacities for D-SR interval of 10 ms and lower.
VoIP packet limited contention-based transmissions have a significantly higher VoIP-user capacity for D-SR intervals over 10 ms.
iii
P REFACE
The difficulty level of this thesis project turned out to be significantly higher than I initially expected, but it pushed me to gain invaluable knowledge. It has truly been a rewarding experience. Realization of the project would not have been possible without the support and guidance from several people and thanking them is truly a pleasure.
First, I would like to thank M˚ arten Ericson, my supervisor at Ericsson. His positive attitude, encouragement and knowledge is highly appreciated and has been of great as- sistance.
I would like to thank Mats Nordberg for giving me the opportunity to realize this thesis.
Kristofer Sandlund deserves a special thanks for his invaluable aid with practically ev- erything regarding the simulator, the protocol standards and always taking the time to listen. There is no doubt that this thesis project never could have been finished without his support!
I am also indebted to thank all the employees at Ericsson Research for their help and guidance with problems of all magnitudes. Also, thanks to Ulf Bodin, my supervisor at Lule˚ a University of Technology, for all the input and tips.
Last but not least, I would like to thank my family and friends for all the support and love they have given me. The joy and happiness their presence brings, makes even the greatest challenge seem possible.
v
C ONTENTS
Chapter 1 – Scope 1
1.1 Purpose . . . . 1
1.2 Problem statement . . . . 1
1.3 Objective . . . . 2
1.4 Delimitations . . . . 2
1.5 Outline . . . . 3
Chapter 2 – Introduction 5 2.1 VoIP . . . . 5
2.2 LTE . . . . 6
Chapter 3 – LTE implementation in the simulator 9 3.1 Chapter introduction . . . . 9
3.2 Simulation system environment . . . . 9
3.3 Conventional resource scheduling . . . . 10
3.4 Channel mapping for the uplink . . . . 11
3.5 Obtaining uplink transmission resources . . . . 12
3.6 Key parameters . . . . 16
3.7 Protocol layers of the LTE architecture . . . . 17
Chapter 4 – Implementation 21 4.1 Granting uplink resources . . . . 21
4.2 Transport format . . . . 21
4.3 Failed transmissions . . . . 22
4.4 Simplification of the simulation models . . . . 24
Chapter 5 – Simulations 27 5.1 The simulation system design . . . . 27
5.2 Definitions . . . . 28
5.3 System description and parameter settings . . . . 30
5.4 Iteration parameters . . . . 31
5.5 Limitation . . . . 32
Chapter 6 – Results 35
6.1 Pure VoIP, no limitation . . . . 35
Chapter 7 – Discussion 45
7.1 Analysis of results . . . . 45
7.2 Reliability of results . . . . 47
7.3 Conclusions . . . . 49
7.4 Further studies . . . . 49
Appendix A – Abbreviations 53 Appendix B – Simulation results: CDF of latencies 55 B.1 Pure VoIP . . . . 55
B.2 Mixed traffic: VoIP and web browsing . . . . 70
Appendix C – Simulation results: user satisfaction, different D-SR intervals 77 C.1 Pure VoIP . . . . 77
C.2 Mixed traffic . . . . 81
viii
C HAPTER 1 Scope
1.1 Purpose
Voice communication has traditionally been carried out on circuit switched networks.
This is not true today since some voice communication is done over IP-networks, such as the Internet[1]. Voice over Internet Protocol, or simply VoIP, is a general term for this type of voice communication transmission. VoIP-services will probably have an increased popularity with the switch to Long Term Evolution (LTE), due to its high capacity and bit-rates. LTE is only a packet switched network: the circuit switched technology previ- ously used for voice communication will not work. LTE networks are the last step toward the next generation mobile telecommunication networks: 4G.
Whenever a mobile, or any type of wireless user equipment, has data to send via a wireless base-station, it needs to be granted uplink transmission resources. Without granted resources, the user equipment is unable to communicate through the base-station.
In the present LTE standard, the access to system resources is initiated by transmitting a scheduling request from the mobile to the base station, see section 3.2.
The scheduling signalling between the user equipment and the base-station adds to the latency. For most services, latency is not a problem and can easily be handled, but voice communication demands short and constant latencies. There is a possibility of decreasing this delay is by switching to another access method: the contention-based transmissions.
With this alternative method, the signalling to request uplink transmission grants is not needed. The purpose of this thesis project is to examine this possibility.
1.2 Problem statement
Contention-based transmission can be achieved with a pre-set broadcast grant, see section 3.3. By merely broadcasting the grant, the scheduling request signalling described in
1
section 3.2 becomes unnecessary. For small loads this may work well, but the collision probability increases with the load. This is due to the risk of several users transmitting simultaneously using the pre-set broadcast grant, which increases with the load. The idea is to examine different use of contention-based grants and find under which properties and restrictions the average satisfaction ratio and capacity for VoIP-users will be maximised.
This will also be compared to the conventional contention-free access method.
1.3 Objective
Higher satisfaction ratio and capacity for VoIP-users requires decreasing the uplink trans- mission resource access delay. Since contention-based grants have a high contention probability for high load, suitable access limitation parameters will be evaluated. Four types of restrictions to the contention-based grants will be compared with contention-free transmissions. The comparison will also be done to contention-based grants available for all users. Evaluation will take consideration to the initial delay due to gaining access to the signalling channel.
The evaluation is done using simulations of a radio access network. This Java-based simulator is developed at Ericsson AB.
1.4 Delimitations
• Only LTE wireless network system is considered.
• This study will only focus on the end-to-end latency for VoIP. Latency for other services such as FTP and web browsing will not be considered.
• There are numerous sources that contribute to VoIP latency, such as access latency, routing, switching and processing at the receiver and the transmitter. Only the latency due to the air interface delay between a user equipment and its base-station will be considered here.
• Mixed traffic analysis is limited to two simultaneous traffic types: VoIP and web browsing.
• Four types of limitations to the contention-based transmissions are done: QoS-class, RLC-buffer size, voice frames and a combination of QoS-class and RLC-buffer size.
• The analysis will be based on simulations of model descriptions together with a
random component. However even the most complete simulator can never imitate
the infinite possibilities and chaos of reality.
1.5. Outline 3
1.5 Outline
This report describes part of the current LTE standards and architecture as it is imple-
mented in the simulator (see chapter 3). The new complementary uplink transmission
method is described in section 3.5.2. See section 4 for the necessary changes to the com-
plementary transmission method and the simulator environment for this evaluation. The
most important results can be found in section 6, the rest can be found in appendices B
and C. The analysis and conclusions can be read in section 7.
C HAPTER 2 Introduction
2.1 VoIP
2.1.1 Introduction
The idea of VoIP was born in 1974 as Vinton G. Cerf and Robert E. Kahn presented the paper ”A protocol for Packet Network Intercommunication”[2]. One of the best known VoIP-services today is Skype, but numerous other services are available on the market.
The main benefit of VoIP is its cost-efficiency for both the provider and consumer. The cost for international calls can be greatly decreased by switching to VoIP, because deliv- ering voice packets on IP-network is as cheap as other data traffic transmissions. Mainte- nance costs can be greatly reduced by switching to VoIP, since the voice communication is done through existent IP-traffic system instead of a separate network.
2.1.2 Delay causes and their effects
Even though VoIP was presented in 1974, it has not been widely used until recently.
The main problem has been the poor sound quality and long and varying delay, mostly due to low bandwidth[3]. Voice services are extremely sensitive to transmission delay, which increases as the bandwidth decreases. Even a constant latency of merely a few hundred milliseconds has a significant impact on the experienced sound quality[18]. As the latency increases, the conversation becomes hesitant and unnatural.
This thesis is limited to reducing the delay between the caller and the base-station through its access method. There are however other aspects which adds to the com- plete latency, such as routing/switching, sample resolution and frequency, jitter and buffering[4]. A slightly longer and constant end-to-end delay does not affect the speech quality as much as jitter influence.
5
2.2 LTE
2.2.1 A short history
The mobile telecommunication networks of today consists of 1G, 2G and 3G. The older systems are being phased out as the bit-rate and capacity demands increase. The his- tory can be read in e.g. ”Multiple Access Protocols for Mobile Communications: GPRS, UMTS and beyond”[5].
1G It started in the late 1970s and early 1980s with the first generation (1G), a low-speed fully-automatic analogue telecommunication system. There are several standards, such as NMT, AMPS and TACS, but all are voice-only and can only offer low system capacity. The 1G standards suffered from clicking noises interrupting the speech, as network control messages were transmitted on the same channels as voice.
2G The second generation (2G), also known as GSM (Global System for Mobile com- munications), was unlike the first generation digital and came in the early 1990’s.
The capacity was significantly increased with the use of low bit-rate speech codecs and dedicated channels for network control information. However, GSM still has traffic channels contaminated by handover signalling.
Unlike 1G standards, the technology in 2G did not only support voice calls but also was able to offer short message service (SMS). This was unexpectedly a huge success and changed the way people communicated. Other features were added as well, such as fax and voice mail.
CDMA is another 2G access method[6], but not a 3GPP specification[7].
3G Third generation services includes voice calls, video calls and wireless data. Unlike 2G systems, the voice and data transmissions can be done simultaneously in 3G, which opened up for a whole set of new services such as mobile TV, video confer- encing and location based services. A higher spectral efficiency also increased the capacity.
LTE As our usage of services with high data-rates and quality of service demands in-
crease, 3G is not sufficient and a new standard is necessary. Long Term Evolution,
or LTE, is the last step from 3G to the next generation of mobile telecommunication
networks, 4G. In 2014, 2.7 billion people are estimated to use mobile broadband
and most of them will be served by Long Term Evolution or High Speed Packet
Access (HSPA) networks[8]. HSPA[10] is another type of bridge between the 3G
and 4G standards and offers up to 14 Mbit/s for the downlink and 5.8 Mbit/s
for the uplink. LTE is more advanced than HSPA and offers IP-transparency (see
section 2.2.2), hence a closer step towards 4G.
2.2. LTE 7
Release 8 of the LTE standardisation has been presented and has been frozen since December 2008[9]. The current LTE equipments are based on this release. Further improvement of the standardisation is possible in future releases.
2.2.2 Current LTE general features
LTE release 8 feature[9]:
• Uplink access scheme is DFTS-OFDM and downlink is OFDMA.
• Compatible with earlier 3GPP-standards.
• Variable bandwidth: 1.4, 3, 5, 10, 15 and 20 MHz.
• 15 kHz sub-carrier spacing.
• Modulations QPSK, 16QAM and 64QAM. More useful bits can be squeezed into a packet with a higher complexity of modulation and good channel quality. See section 3.3.4.
• Low latencies. Short set-up time and short transfer delay. Minimum transmission time per interval (TTI) as low as 1 millisecond. Ideally a packet can be transmitted each millisecond.
The idea of LTE is not merely to increase the bit-rate. The future usage of mobile telecommunication moves clearly towards All-IP Networks, AIPN[12], as the trend of IP traffic usage in mobile systems increases. Mobile data traffic is estimated to double every year through 2013 and increase 66 times from 2008 to 2013[11]. AIPN offers gains such as decreased cost for the operators, flexibility of deployment and improved user experience[13].
With the use of IP transparency that AIPN brings, the access can also be done with WiFi, WiMAX or regular wired systems. Hence, AIPN is not limited to UTRAN[14] and GERAN[15] based access networks[12], i.e. 3G and GSM/EDGE.
LTE supports handover between its own systems and existing mobile networks, which means it can be partially built while regular mobile broadband is offered where LTE lacks coverage.
More information can be found on the 3GPP website: www.3gpp.org/LTE. LTE does
not sufficiently meet the standards of 4G due to one area: bit-rate. The peak bit-rate
required for 4G is 1Gbit/s for stationary and 100 Mbit/s for mobile usage. With LTE
Advanced the 4G ITU requirements should be fulfilled.
2.2.3 Transmission resources: frequency-time domain
LTE supports both frequency division duplexing (FDD) and time division duplexing (TDD) within the same single radio access technology. Base-stations in an LTE network can schedule users on several different spectrum bands, more commonly known as sub- bands, every transmission time interval. This works the same way for the transmission time intervals, or TTIs.
A user can be scheduled for one or more subbands one specific TTI and scheduled for more subbands in a future TTI. The following scheduling does not have to use same sub- bands as the previous one. Although there are other scheduling restrictions, the scheduler has higher freedom to chose appropriate resources for users and still the availability to try to fully utilize its set of subbands.
In June 2009, fifteen paired and eight unpaired spectrum bands were specified for LTE
and this will be increased[8].
C HAPTER 3 LTE implementation in the simulator
3.1 Chapter introduction
The following chapter describes the LTE standard and architecture as it is used in the simulator. The conventional contention-free uplink transmission method (see section 3.5.1) and the lower protocol architecture (see section 3.7) are examples of the LTE standard. Parameter settings are also described, such as the scheduling algorithm (see section 3.3.2).
Last section, i.e. section 3.7, describes the LTE lower and higher protocol architecture.
This is necessary for the understanding of chapter 4.
3.2 Simulation system environment
The simulator is a Java-based object-oriented event-driven radio access simulator devel- oped at Ericsson AB.
The simulator implements the standardisation as well as other functions of an LTE system over a time-period in pseudo-reality. Several models are used for the simulations: system descriptions, physical environments (here: urban topology), users and their characteris- tics (location, movement, activity etcetera), traffic models, radio channels, interference, antenna models to name a few. Some aspects are directly affected by the pseudo random component, such as user location and channel disturbance.
9
3.3 Conventional resource scheduling
3.3.1 General
If a user equipment wants to transmit data to the base-station, it needs to be granted up- link transmission resources by the base-station. The base-station has to use a scheduling algorithm (see section 3.3.2) to be as fair as possible to all users when deciding when a user equipment can transmit and how many subbands it can use. Otherwise, some users might not be scheduled at all.
Next step is for the base-station to decide the modulation method (see section 3.3.3) and the transport format (see section 3.3.4) for the uplink transmission. These two aspects specifies how the uplink transmission will be done. All uplink transmission information will be packaged into a grant, which will be transmitted from the base-station to the user equipment.
3.3.2 Scheduling algorithm
The default setting in the simulator is round-robin as grant scheduling algorithm. This type of scheduling dedicates equal resources per user and distributes them in a circular fashion. For example, if there are two resources and three users, then for the first trans- mission time interval (TTI) users 1 and 2 will be scheduled. The second TTI users 2 and 3 will be scheduled. Users 3 and 1 will be scheduled the third TTI, repeating as long as the users have data to transmit.
In reality most radio access networks use some sort of quality of service weight for their scheduling. This is necessary in order to avoid unfairness, since the latency tolerance differ depending on service. VoIP is highly sensitive to latency, but when transferring large data files between users, latency is not as significant.
Interference is a large problem for transmissions. The interference will vary quickly: the interference from other users can be high one TTI and insignificant the next TTI. In or- der to avoid inter-cell interference, the scheduler will pick randomly a subband from the set of available subbands. If the user to be scheduled should be granted more subbands, it will be a subband next to the previous one.
3.3.3 Transport format
The transport format specifies how many subbands to use each transmission time interval together with a specified modulation method. This simulator branch uses a scheduler with a greedy selection algorithm for its contention-free transmissions. The transport format is decided upon several user aspects: the amount of data in the transmission buffer, its channel quality indicator, settings whether or not to minimise the transport format, etcetera. The channel quality is measured every 40 milliseconds in the simulator.
The base-station estimates the channel quality indicator for a specific user on its uplink
3.4. Channel mapping for the uplink 11
transmission on the physical uplink control channel (PUCCH, see section 3.4). Since this is something a user never can measure by itself, it cannot decide its own transport format.
3.3.4 Modulation method
LTE offers three different modulation types: QPSK, 16-QAM and 64-QAM[16]. Higher modulation complexity enables the transport block to carry more data.
QPSK Quadriphase-shift keying carries the information in the phase of the transmission wave. The phase can take the value of four equally spaced values: π/4, 3π/4, 5π/4 and 7π/4. These phases corresponds to a set of bits each: 10, 00, 01, 11. Hence, QPSK carries two bits of information per interval.
16-QAM 16-Quadrature Amplitude Modulation carries information in the phase and in the amplitude. This enables it to carry a larger amount of information than the QPSK modulation. M-ary QAM-modulations are able to carry L information bits, where M = L
2. This enables 16-QAM to carry 4 bits of information.
64-QAM 64-Quadrature Amplitude Modulation works the same way as 16-QAM. Since √ 64 = 8, this modulation method can carry 8 bits of information.
3.4 Channel mapping for the uplink
There are three different types of channels: physical, transport and logical[17].
The physical channels in the uplink are used for carrying user information and data.
Physical Random Access Channel (PRACH) is only used for initial access and synchro- nization.
Physical Uplink Control Channel (PUCCH) carries the Uplink Control Information (UCI).
When transmitting a scheduling request, it is done on PUCCH. However the load on PUCCH increases with the number of users. To decrease the load, the scheduling re- quest interval has to increase. Hence a user has to wait longer for access to PUCCH if the system expects a larger amount of users. An alternative method is to increase the number of PUCCH channels: this will decrease the load without increasing the access interval. There is a cost - more PUCCH means less PUSCH.
The Physical Uplink Shared Channel (PUSCH) is used for transmitting data. Unlike
PUCCH and PRACH, scheduling controls access to the PUSCH. The scheduling grant
given by the base-station will specify which resources the user equipment can use on
PUSCH.
The transport channels carry the characteristics of transmission information on the radio interface. These channels are a service from the physical layer to the MAC-layer and higher layers.
Uplink Shared Channel (UL-SCH) is the transport channel that handles uplink data transmission.
Random Access Channel (RACH) is used for synchronization with the base-station and initialization. The physical counterpart PRACH carry information to this transport channel. Even though the RACH never transmits any transport blocks, it is regarded as a transport channel.
The logical channels are defined by the type of information they carry. Control channels carry control and configuration information needed for transmissions. Traffic channels contain data from the users.
Common Control Channel (CCCH) handles control information.
Dedicated Traffic Channel (DTCH) transmits the data traffic from the user equipment.
Dedicated Control Channel (DCCH) transmits the control information from the user equipment. This channel handles the individual configuration information such as han- dovers.
The downlink has several more channels than these three.
In Figure 3.1, the relations between the physical channels, transport channels and logical channels for the uplink can be seen.
3.5 Obtaining uplink transmission resources
3.5.1 Conventional method: contention-free transmissions
The conventional way of obtaining transmission resources in radio access networks is through contention-free access. Each specific transmission is individually scheduled by the base-station and the grant is then transmitted to the user equipment. This grant will inform the user equipment which subbands the base-station has pre-allocated for it together with information about when and how to use them. The signalling of the resource request required for the transmission initialization adds to the overall latency.
Contention-free resource access can be done with the following steps:
1. The mobile has to wait for the physical uplink control channel (PUCCH, see section 3.1) to become available. This dedicated scheduling request (D-SR) interval is determined by the base-station. The D-SR interval is in this example 5 ms, hence the average delay is 2.5 ms.
With numerous VoIP users, the interval should be larger in order to minimize the
3.5. Obtaining uplink transmission resources 13
Figure 3.1: Channel mapping for logical channels, transport channels and physical channels[17].
load on the PUCCH. This can lead to average delays as large as 20 ms (for D-SR intervals of 40 ms).
2. The user equipment transmits a D-SR on the PUCCH along with a buffer status report (BSR).
Time: 1 ms.
3. The base station decodes the received D-SR and generates a scheduling grant (SG).
Time: 3 ms.
4. Transmission of the SG.
Time: 1 ms.
5. The user equipment processes the received SG and encodes the uplink data.
Time: 3 ms.
6. Transmission of the uplink data on the physical uplink shared channel (PUSCH).
Time: 1 ms
7. Processing delay for the base station.
Time: 3 ms.
These steps are shown in Figure 3.2. Obtaining resources with the access scheme described above, does not include the necessary signalling for synchronisation and ini- tialisation at the start-up. The simulator implements the delays for each step as an approximation to reality.
This description is also a generalization. The signalling of scheduling request is merely a bit transmitted on PUCCH: it does not contain a buffer status report. The buffer status report is essential for the scheduling at the base-station: for a fair scheduling for user equipments, the base-station needs to be aware of the amount of data each user equip- ment wants to transmit. In order for the base-station to be aware of the actual buffer size at the user equipment, it will first transmit a smaller grant, but large enough to fit a buffer status report. At this point the user equipment will have triggered its buffer status report. Next time the user equipment will try to transmit using the dedicated resources, its buffer status report will be appended to the payload. When the base-station receives this, it will grant for the now known buffer size at the user equipment.
As mentioned previously, the D-SR interval in Figure 3.2 is 5 milliseconds. In reality, this scheduling interval is selected by the base-station and dependent on its load. To decrease the load on PUCCH, the number of PUCCH resources can be increased. However it would be at the cost of PUSCH resources.
3.5.2 Alternative method: contention-based transmissions
This alternative method is not a part of the LTE standard yet, but has been discussed at 3GPP meetings and was partly already implemented in the radio network simulator for evaluation purposes.
The contention-free access method described in section 3.5.1 has a significant latency whenever there is a long scheduling request interval. This could be avoided with a contention-based grant. With this method, the base-station broadcasts a grant any user equipment can use for its transmission. With this shared grant, there is no need to wait for the PUCCH to become available. The contention-based uplink transmission can be done with the following steps:
1. The base-station waits for the beginning of the transmission time interval (TTI).
Average time: 0.5 ms.
2. Transmission of the contention-based resource grant. This will be read by the user equipment on the physical downlink control channel (PDCCH). The grant is transmitted by the base-station every TTI when it feels it has available subbands.
Time: 1 ms.
3.5. Obtaining uplink transmission resources 15
0 ms
10.5 ms 7.5 ms 2.5 ms
6.5 ms
11.5 ms 3.5 ms
Time
SR
SG
Data
14.5 ms
Contention free
UE eNB
Figure 3.2: Contention-free access between a user equipment (UE) and a base-station (eNB).
3. The user equipment processes the received scheduling grant and encodes the uplink data.
Time: 3 ms.
4. Transmission of the uplink data. This data is transmitted using the contention- based resources previously granted by the base-station.
Time: 1 ms.
5. Processing delay for the base-station.
Time: 3 ms.
This is also shown in Figure 3.3. Nevertheless, it means using the same uplink data
transmission resource for all user equipments (UEs). If two or more UEs are trying to
access the same resource simultaneously, there is a risk of contention of packets. If this
happens often, it will significantly affect the throughput: if retransmissions are enabled,
retransmitting the unsuccessful data will increase the delay. With no retransmissions, the
unsuccessful data will merely be discarded. Clearly the use of contention-based grants needs to be restricted such that only the users that has most to gain shall use it.
0.5 ms 0 ms
5.5 ms 4.5 ms
1.5 ms
Time
CB Grant
Data
8.5 ms
UE
Contention based
eNB
Figure 3.3: Contention-based access between a user equipment (UE) and a base-station (eNB).
3.6 Key parameters
The purpose of this thesis is to evaluate the VoIP-user capacity and satisfaction ratio for two uplink transmission methods: contention-free (see section 3.5.1) and contention- based (see section 3.5.2). Two key parameters are necessary for estimating the satisfaction ratio: the amount of transmitted packets from one end and the amount of received packets and their delays at the other end.
If two or more users are simultaneously transmitting using a contention-based grant, their
transmissions will be on the same subtends. This means that the base-station will receive
3.7. Protocol layers of the LTE architecture 17
a signal which is a combination of all the simultaneous contention-based transmissions.
It is highly important that the simulator will take this into consideration. Since the contention-based grant does not take into consideration a user’s channel quality, it is important that the simulator has a varying channel quality.
3.7 Protocol layers of the LTE architecture
3.7.1 General
The transport protocol consists of several different layers. As the transport block ascends in the protocol hierarchy, the header of the current layer is removed and checked. Like peeling an onion, so are the transport blocks handled in the protocol hierarchy. This can be seen in Figure 3.4.
Figure 3.4: Lower protocol layer structure for LTE[17].
The description of the lower layers in the LTE protocol architecture[17], starting at the lowest and ascending:
1. Physical layers 2. MAC
3. RLC 4. PDCP
3.7.2 Physical layers
The physical layer is the first and lowest layer and handles[17]:
• Mapping of the signal to the correct time-frequency resource.
• Mapping of the physical channels to the transport channels.
• Coding/decoding.
• Modulation/demodulation.
• For the uplink: the DFT.
Cyclic Redundancy Check (CRC) is used as an early error detection. It is calculated for each transport block and attached after it. When a transport block is received, the same CRC is calculated at the receiver and then compared with the received appended CRC. If an inconsistency is detected, a retransmission can be triggered by signalling a HARQ-error at the MAC-level.
3.7.3 MAC
Medium Access Control (MAC) handles the uplink and downlink scheduling and HARQ retransmissions[17]. HARQ (Hybrid Automatic Repeat reQuest) is a method used for error-control at the MAC-level. The uplink and downlink scheduling is only performed at the base-station since no user equipment performs any type of scheduling.
MAC also uses logical channels as a service to carry information to the RLC-level. Hence,
this layer handles the mapping of logical channels to the appropriate transport channels
and the multiplexing of different logical channels. This can be seen in Figure 3.1.
3.7. Protocol layers of the LTE architecture 19
3.7.4 RLC
RLC is an abbreviation for Radio Link Control and handles segmentation, concatenation and ARQ (Automatic Repeat reQuest)[17]. The packets received from PDCP may be too large to fit inside the available size left at the RLC-level. Hence it may need to be segmented into smaller pieces. RLC handles the segmentation and keeps track of whether or not the following and/or previous service data unit (SDU) is concatenated with the current one.
Channel condition fluctuations can cause one segment of a larger PDCP packet to be lost. It would be a waste of resources retransmitting successful segments from the same PDCP packet. A retransmission operates at the RLC-level between the transmitter and the receiver, in order to only retransmit the lost segment. Therefore transmission errors due to noise, an unpredicted drop of channel quality, etcetera, can be handled at this level. Status reports will be transmitted from the RLC entity at the receiver regarding the incoming segments. The RLC entity at the transmitter receives this status report and is able to act upon it. A retransmission will be performed in the acknowledged mode, but ignored in the unacknowledged mode. VoIP traditionally uses the unacknowledged mode in order to minimise the mouth-to-ear delay.
3.7.5 PDCP
The Packet Data Convergence Protocol (PDCP) layer handles header compression and ciphering[17]. Without no specified header compression, the data unit will be unchanged.
Robust header compression (ROHC) is recommended, but not a part of the standard.
3.7.6 Above
The PDCP is followed by IP for data traffic. The IP (Internet Protocol) layer handles the addressing and routing of packets on a packet-switched interwork. The addressing is distinguished such that packets can be delivered from its source to its host without any other information.
Layers on top of the IP-layer differs depending on service. Figure 3.5 shows the up-
per layers of two different services: VoIP to the left and web browsing to the right. The
IP layer is at the bottom and the higher the layer, the further up in the protocol hierarchy.
Lower layers
IP IP
UDP TCP
RTP Web
VoIP
Figure 3.5: Higher protocol layer structure for VoIP (left) and web browsing (right).
C HAPTER 4 Implementation
4.1 Granting uplink resources
Whenever the base-station has not fully scheduled the uplink, i.e. has available resources, it will transmit a contention-based grant. In reality this is a broadcast: it will be trans- mitted once to all users and any user in range will receive it. This is not the case in the Java-based simulator used for the study. A pseudo broadcast is used instead.
The base-station will start with scheduling the regular contention-free users - the ones who have transmitted a scheduling request and have been granted resources. Second step is to schedule all users which have their channel quality indicator triggered. If there still are available resources, then the remaining users will be scheduled for the same contention-based resources. The grant will be duplicated and transmitted on the physi- cal downlink control channel (PDCCH) to all the remaining users.
The base-station in the radio access simulator is unaware of users that have not had their initial signalling on physical uplink control channel (PUCCH). In order to be scheduled for contention-based resources, the user needs to first be scheduled for the conventional contention-free resources. In reality this initial contention-free scheduling will not be re- quired, although the users still needs to do all the initial signalling with the base-station to get synchronized.
4.2 Transport format
The transport format in contention-free access differs depending on the user to be sched- uled: its buffer size, channel quality, etcetera (see section 3.3.3). For contention-based scheduling this brings a significant problem: when the base-station is transmitting its contention-based grant, it is unaware of which user will actually use it. Even if the base- station would know all of the users’ current channel qualities, it would not know which one to base the transport format on. In order to avoid this, the pre-set contention-based
21
transport format needed to be robust enough to handle most channel qualities, but still be able to carry enough information to be useful. Since the object of this paper is to minimise the latency for VoIP-users, it is important that one grant is able to fit one voice frame (see section 5.3). The voice frames in the simulator at the RLC-level (see section 3.7.4) are all 296 bits each and the silence insertion descriptor (SID) frames are 96 bits.
The size of the grant was set to 328 bits, to be able to append headers.
Since the channel quality is unknown, it was decided that the modulation method should be one with high robustness. Hence the choice fell on QPSK (see section 3.3.4). The transport format tables used for this simulator implementation, could use 1, 2, 3, 4, 5, 6, 8 or 9 subbands and still fulfil the requirements of 328 bits and QPSK modulation.
Using a large amount of subbands would be affected by the background load: if there are already several users scheduled for regular resources, the contention-based grant might not be able to be scheduled at all. Given the above, the choice fell on two subbands in order to often schedule contention-based grants.
4.3 Failed transmissions
4.3.1 Failure detection
Using contention-based resources leads to a high probability of collision. This is due to enabling several users transmissions in the same cell on the same subbands and transmis- sion time interval. In reality, one user can successfully be decoded at the base-station if the others’ transmissions were unsuccessful due to path loss or other disturbances. The base-station could by analysing the signal-to-noise ratio of the successful transmission, be able to detect the failed transmission attempts of the other users. For contention-free transmissions this is not an issue: the base-station is aware of which users are enabled for transmission since the transmissions are already scheduled. The base-station would also not schedule several users for the same resource, but it is the core of contention-based access.
The contention-based grant is broadcast to the users in the cell, meaning the base- station has no clue which of the user is actually trying to transmit. If the transmission attempt fails, then the packet cannot be decoded and the source cannot be detected. If an acknowledgement (ACK) is sent to the user of the successful transmission, then the users that failed might interpret this as acknowledgements of their transmissions.
The simulator uses a simplified collision model when it comes to contention of several
transmissions. If several users try to transmit simultaneously on the same resource, the
regular channel model in the simulator will treat them as if they are separate transmis-
sions. In other words, it does not take into consideration the intra-cell interference of
the multiple transmissions and will decode each transmission as if it was alone. This
4.3. Failed transmissions 23
is not suitable for simulations with higher load on contention-based resources, therefore changes were required. Apart from the regular actions of the collision model, it will also look at the amount of users trying to simultaneously access the same subband. If only one user attempts use the grant, then everything proceeds as normally. For several users, then all will be considered as failures and negative acknowledgements (NACK) will be transmitted. Thus ignoring the case when one user might succeed despite contention from other simultaneous transmissions using the same resources.
4.3.2 Error handling
There are two types of error detection: either a received HARQ NACK for a certain pro- cess or a process time-out at the user equipment, triggered by no received ACK/NACK before its deadline. The error detection by time-out is simple, due to the synchronous nature of the simulator: there are no variations. The time-out interval in the simulator is set to 8 milliseconds. The reality is unfortunately not synchronous. In order to cope with smaller variations, extra intervals are necessary and the time-out is thus in the magnitude of 13-14 milliseconds.
As mentioned in section 3.7.4, there are two types of transmission modes: acknowl- edged mode (AM) and unacknowledged mode (UM). AM has by definition its own error handling and retransmission. The normal error procedure for UM is to merely discard the data and proceed as normal. If several user equipments in UM try to repeatedly transmit their packets simultaneously using contention-based resources, nothing will get through and the packets will be lost. Due to the high collision probability, the services using contention-based resources needs some sort of retransmission. VoIP-services can- not use the retransmission from AM, since it leads to longer latencies. Hence another retransmission procedure needed to be implemented for UM. The error handling for UM is as follows:
• A transmission error is detected at the user equipment, either by a received NACK or triggered by a time-out.
• If the failed transmission was contention-free, proceed as normal UM: discard the data and continue as previously. If contention-based, put the payload back into the buffer and switch to contention-free access with a scheduling request. The new buffer status estimator will do its buffer size calculation after the negatively ac- knowledged data has returned to the buffer.
There is however an exception: if the payload is older than the maximum tolerable delay of 140 milliseconds, then it will not be returned to the buffer at a NACK.
The switch to contention-free transmission will still occur together with a current
buffer status estimator. Retransmitting an out-dated packet is a waste of resources, seeing that the data unit will be regarded as useless at the receiver.
• After the scheduling request has been sent, the user equipment will disregard any contention-based grant while waiting for its contention-free grant. At the reception of its requested contention-free grant, continue as normally.
In order to return the payload to the transmission buffer after a failure, the transmission process for UM needed to be changed slightly as well. After transmitting a packet, the user equipment stores the sent data while waiting for an ACK or NACK. The buffer with stored sent data will be updated at every new transmission and out-dated data will be removed. The expiration time of the stored data is important: otherwise an unnecessary high amount of data would be stored. If an ACK is received when in UM, nothing will happen. The corresponding packet with the same identity number will merely be deleted from the buffer after 140 milliseconds.
The error handling above also depicts the switch from contention-based to contention-free transmissions. A user equipment is able to switch back to contention-based access if the contention-free transmission was successful. At the next triggering of scheduling request transmission, the user equipment will instead halt and wait for next contention-based grant.
4.4 Simplification of the simulation models
Several simplifications of the simulation models were made. The reason was simply the time-constraints of the project and the complexity of the simulator.
• Header sizes have been dramatically cut. Handling the retransmission of failed contention-based packets in unacknowledged mode turned out to be cumbersome since in many cases it demanded resegmentation. There was clearly not enough time to fix this properly. In order to ensure that a retransmitted packet would not get segmented during the retransmission, header sizes had to be minimised.
Dynamic robust header compressions on PDCP-level is often used, but cutting down the header sizes such as done here is more like a simplistic ROHC. Instead, UDP and IP headers are both set to zero. The RTP header size is 12B as default setting, but is here set to 3B.
• As mentioned previously, the retransmission model could not handle segmentation.
As a result, only users with an unsegmented data unit for transmission that can completely fit inside the transport format can use the contention-based grant.
• Collision model is simplified. Whenever two users transmit on the same subband
simultaneously, both packets will be regarded as corrupted and NACKs will be
transmitted. In reality, different channel qualities and signal strengths might lead to
4.4. Simplification of the simulation models 25
one packet being successfully decoded and the other one corrupted. The simulated
case is thus a worst-case scenario. It would have been too cumbersome for this
project to implement a proper collision model.
C HAPTER 5 Simulations
5.1 The simulation system design
The simulations are done for the transmission between a user equipment (for example a cellphone) located in a cell within the system and a non-wireless device at the core of the system, see Figure 5.1. For the uplink the following delays will occur:
1. Processing delay at the front-end of communication. Front end in Figure 5.1: device B. Estimated time: 5 milliseconds.
2. Air interface delay. Delay due to transmission between front-end and its base- station. If the base-station detects a failure, a retransmission may be done (de- pending on mode). This is not a fixed delay and is dependent on load in the cell and channel quality. However the largest tolerable delay for this simpler system is 140 milliseconds. After subtracting the other fixed delays, 80 milliseconds is left for the air interface delay.
3. Delay due to switching and routing between the base-station at the front-end to the back-end (device A). Estimated time: 20 milliseconds.
4. Processing delay at the other end of communication (device A). Estimated time:
35 milliseconds.
This is a simpler scenario: the worst case scenario for access transmission time is be- tween two wireless devices, see Figure 5.2. The maximum tolerable end-to-end delay for two-way voice communication was decided to be 240 milliseconds. With this mouth-to- ear delay, 85% of all conversational speech users are very satisfied or satisfied[18].
Delay due to switching and routing is estimated in total to 40 milliseconds. An addi- tional 5 milliseconds at the front-end and 35 milliseconds at the other end is required for
27
Figure 5.1: The simpler end-to-end transmission model used in the simulator.
processing. This leaves us with 160 milliseconds for two air interference delays, hence a maximum of 80 milliseconds latency per wireless link at each end. Using the easier case implemented in the radio access simulator and still making sure to handle the worst case scenario, leaves us with a system maximum end-to-end delay of 140 milliseconds.
5.2 Definitions
Packet failure There is a difference between completely lost packets and late packets
flagged as lost. If a voice packet does not reach the destination at all, it is completely
lost. If a packet arrives at the destination after its deadline of 140 milliseconds, it
is flagged as lost, since it is useless. See section 4.3.2 for discarded and late data
5.2. Definitions 29
Figure 5.2: The worst case end-to-end transmission model.
due to failure detection and its handling.
Satisfaction A user is regarded as satisfied if it receives 99% of the voice frames on time. Hence, up to 1% of the packets can be late or lost.
Capacity The system capacity is dependent on the satisfaction ratio. The capacity
criteria in this work is the amount of users when 95% of them are satisfied.
Figure 5.3: Simulation model of seven base-stations with three hexagonal cells each.
5.3 System description and parameter settings
System The simulated system consists of seven base-stations with three cells each, see Figure 5.3. Every cell is in a hexagonal form with the radius of 166.7 m and a typical urban environment model. The attenuation model includes path loss.
Radio interface The uplink and downlink has 25 subbands each and a bandwidth of 4.5 GHz.
Users The number of initial users is specified. The users will be randomly distributed over the system and will move in a straight line with the velocity of 3 km/h. See section 5.4.2 for amount of users in the simulations.
The users can use different services. In these simulations, the users are predefined as one of two different services:
• VoIP
A voice frame is generated every 20 milliseconds and it is crucial it arrives before its deadline of 140 milliseconds. It would be a waste of resources trans- mitting voice frames of silence every 20 milliseconds. When one user is de- tected as silent, the voice frame is replaced with a silent insertion descriptor (SID). First SID is generated after 20 milliseconds, same interval as with voice frames. But if the silence continues, the SID interval will double its interval at every new reception of a SID frame, until the interval is 160 milliseconds, assuming no new speech. The SID interval will remain at 160 milliseconds until the speech continues.
The speech activity ratio is set to 0.5: each end of the conversation is active
approximately half of the time. When one user stops talking, the other one
begins. The mean talk spurt duration is set to 5.0 seconds. The call length
5.4. Iteration parameters 31
mean value is set to 30.0 seconds with a 0.4 seconds graceful termination pe- riod. The simulations are 30.5 seconds each, in order for the last transmitted packets to be received before the simulation terminates.
A VoIP-user in these simulations is considered as satisfied if 99% of the packets arrives and has a mouth-to-ear delay up to 140 milliseconds.
• Web browsing
The web browsing service behaves differently to VoIP services. VoIP has a small amount of data transmitted often while web transmits a larger amount of data more seldom. The uplink traffic for web services comes from requesting web pages or transmission of data. The simulation model imitates persons surfing the Internet: there can be seconds between uplink data transmissions.
The simulator uses a web traffic model according to what is used in ”IEEE 802.16m Evaluation Methodology Document (EMD)”[19].
Random component Five different seeds are used for every iteration. The seeds are the input to the pseudo-random generator. The random component affects the initial position of each user in the system, the voice activity of each user and channel condition fluctuations. The post-processing adds the result of all seeds together and calculates an average, to get a result as accurate as possible.
5.4 Iteration parameters
5.4.1 Dedicated scheduling request interval
As mentioned in sections 3.5.2 and 3.4, the load on the physical uplink control channel (PUCCH) increases with the number of user equipments. One method to decrease this load is to increase the dedicated scheduling request (D-SR) interval. The D-SR interval to be used in real implementations is unknown. Hence it is of interest to see the effect on the contention-based and contention-free access methods for different D-SR intervals.
The D-SR intervals simulations will be done for four values: 5, 10, 20 and 40 milliseconds.
The actual interval to use in reality should be between the extreme values of 5 and 40 milliseconds.
5.4.2 Amount of users
It is of high interest to evaluate the possible gain of contention-based resources in re-
lations to the load. Simulations over different D-SR intervals can show the effect of
different background loads, but it does not take consideration to contention from other
users. Since all simultaneous transmissions on same subbands (see section 4.3.1) are un-
successful, retransmissions will probably be common when the load is high.
For low load simulations, the amount of users will be between 1 and 21 VoIP users in the whole system. Since there are 21 cells in the system, low load will be simulated with up to one user per cell on average. For higher load simulations, the simulations will be done for 1, 10, 20, 30, 40 and 50 VoIP users per cell. For the complete system of 21 cells, this means simulating between 21 and 1050 VoIP users. When simulating mixed traffic, there will be a background load of 50 web browsing users per cell.
5.5 Limitation
5.5.1 General
In order to minimize the load on contention-based resources, limiting the usage may increase the gain. This limitation will be done at the user-end. The base-station will transmit its grant independently, and it is the user that decides whether or not it shall use the grant. This decision is done by comparison with the pre-set limitation. The general idea is to limit such that only the users that has most to gain by a possible delay decrease shall use the contention-based grant.
5.5.2 Quality of service limitation
Since web browsing is not as delay sensitive as VoIP, the gain of a latency decrease would be more significant for VoIP. Some simulations will be done where only VoIP users can use the contention-based grant. As described in section 4.3.2, whenever a contention-based transmission is unsuccessful, the retransmission will be done with the contention-free method. Hence, a VoIP user can use the contention-free method, but a web browsing user can never use a contention-based grant with this quality of service limitation.
The quality of service class limitation is of course useless without mixed traffic scenario, since all VoIP users is of the same quality of service class.
5.5.3 Buffer limitation
When there is a larger amount of data in the transmission buffer, the overhead given by the signalling of scheduling request and its grant is not as significant. There is also the possibility of utilizing a better channel quality with higher modulation, which enables carrying more bits in the same transport block. Hence, it is of interest to evaluate the possible gain of limiting the contention-based grant to the RLC buffer size.
The objective is to minimise delay for VoIP-users, which has small amount of data arriving
often in their RLC buffer. For this reasons, the buffer limitation is set to 312 bits: large
enough to fit up to one voice frame.
5.5. Limitation 33
5.5.4 Voice frame limitation
When considering VoIP as the main service for contention-based grant, it may be im- portant to differ between the main two types of data frames it uses, see section 5.3. The SID frames are smaller than the voice frames, but not as useful since they do not carry important voice information. Limit the access to only voice frames may be of interest.
This also means quality of service limitation, since web browsing does not have any voice frames.
5.5.5 Combined limitation
The buffer limitation does not take into consideration the traffic type. The services differ
when it comes to quality of service classes. There might be a gain by combining the two
limitations and evaluate the result.
C HAPTER 6 Results
6.1 Pure VoIP, no limitation
6.1.1 Low load
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 10 20 30 40 50 60 70 80 90 100
VoIP users per cell
satisfied users [%]
CB DSR=5ms CB DSR=40ms CF DSR=5ms CF DSR=40ms
Figure 6.1: Satisfaction ratio versus VoIP-user load per cell in average. Contention-based (CB) and contention-free (CF) simulations for two D-SR interval: 5 and 40 ms.
35
For pure VoIP and low load, conventional contention-free access is clearly better as long as the D-SR interval is short (5 ms). There is a huge drop of satisfaction ratio when the D-SR interval is increased to 40 ms: most of the VoIP-users in contention-free access will be unsatisfied. With an extremely low load of VoIP-users and long D-SR interval, the satisfaction ratio for contention-free transmissions is merely in the magnitude of 20%.
See Figure 6.1.
Unlimited contention-based transmissions for low load and pure VoIP-traffic scenario does not vary much with the D-SR interval. The largest gain can be found for D- SR interval of 40 ms: contention-based transmissions are almost as good as the best contention-free transmissions (5 ms D-SR interval).
See list below for parameter settings.
Access method Contention-free, contention-based
Number of cells 21
VoIP users in system [1, 3, 6, 9, 12, 15, 18, 21]
Web browsing users in system 0
Dedicated scheduling request interval (D-SR) [5, 40] ms
Limitation None
6.1. Pure VoIP, no limitation 37
6.1.2 High load
5 10 15 20 25 30 35 40 45 50
0 10 20 30 40 50 60 70 80 90 100
VoIP users per cell
satisfied users [%]
CB DSR=5ms CB DSR=10ms CB DSR=20ms CB DSR=40ms CF DSR=5ms CF DSR=10ms CF DSR=20ms CF DSR=40ms
Figure 6.2: Satisfaction ratio versus VoIP-user load per cell in average. Contention-based (CB) and contention-free (CF) simulations.
Unlimited contention-based access has higher satisfaction ratio than contention-free access as long the D-SR interval is long (40 ms) and the total amount of VoIP-users is below 10 VoIP-users/cell. This is shown in Figure 6.2. With D-SR intervals of 5, 10 or 20 ms alternatively a load over 10 VoIP-users/cell, contention-free access will have a signif- icantly higher ratio of satisfied users than the unlimited contention-based transmissions.
See the list below for parameter settings.
Access method Contention-free, contention-based
Number of cells 21
VoIP users in system [21, 210, 420, 630, 840, 1050]
Web browsing users in system 0
Dedicated scheduling request interval (D-SR) [5, 10, 20, 40] ms
Limitation None
6.2 Pure VoIP, contention-based limitations
5 10 15 20 25 30 35 40 45 50
0 10 20 30 40 50 60 70 80 90 100
VoIP users per cell
satisfied users [%]
CB: voice frames CB: buffer size CB: unlimited CF
Figure 6.3: Satisfaction ratio versus VoIP-user load per cell in average. Two types of contention-based limitation: RLC-buffer size and voice frames. Contention-free and unlimited contention-based transmissions for same D-SR interval.
Figure 6.3 depicts the different types of contention-based transmissions and the contention- free transmission method, with a D-SR interval of 40 ms. For low load (below 10 VoIP- users per cell), all types of contention-based transmissions has a higher satisfaction ratio than contention-free transmissions. However as the load increases, the limited contention- based transmissions are a significant improvement when compared to contention-free transmissions. The highest satisfaction ratio can be reached by limiting on voice frames.
The buffer frame limitation allows also transmission for smaller frames such as SIDs.
See the list below for parameter settings.
6.2. Pure VoIP, contention-based limitations 39
Access method Contention-based, contention-free
Number of cells 21
VoIP users in system [21, 210, 420, 630, 840, 1050]
Web browsing users in system 0
Dedicated scheduling request interval (D-SR) 40 ms
Limitation RLC buffer size (312 bits)
Voice frame
6.2.1 Capacity evaluation
Figure 6.4 depicts the cell capacity in average for three types of transmission method:
contention-free, unlimited contention-based and voice frame limited contention-based.
Contention-based transmissions limited to voice frames clearly has the best capacity.
Only contetion-free transmissions when the D-SR interval is 10 ms has higher capacity:
the difference is 2 users. However for longer D-SR interval, contention-free has zero ca- pacity, while voice frame limited contention-based has a capacity over 30 users.
The capacity is defined as the load when the satisfaction ratio is at 95%. See Figures C.1, C.2 and C.3 in appendix C for the relevant graphs.
5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40 45 50
DïSR interval [ms]
Capacity [users]
CF
CB (voice frames) CB (unlimited)
Figure 6.4: Average cell capacity. Contention-free (CF), unlimited contention-based (CB (un-
limited) and voice frame limited contention-based transmissions (CB (voice frame)).
6.3. Mixed traffic 41
6.3 Mixed traffic
6.3.1 Contention-based limitations
5 10 15 20 25 30 35 40 45 50
0 10 20 30 40 50 60 70 80 90 100
VoIP users per cell
satisfied users [%]
CB: QoS CB: Buffer size CB: Voice frame CB: QoS + Buffer size CB: unlimited CF
Figure 6.5: Satisfaction ratio versus VoIP-user load per cell in average. Contention-based limitation of QoS, buffer size, voice frames and combined limitation of QoS and buffer size.
Contention-free and unlimited contention-based transmissions.
Unlimited contention-based transmissions is useless with this high amount of traffic: the satisfaction is practically zero even with a low VoIP-user load. The web browsing users as background load is clearly more than the unlimited contention-based method can handle.
Contention-based transmissions limited to quality of service-class (VoIP) has higher satisfaction ratio than contention-free transmissions as long as the load is low. As the load increases, this limitation is clearly not sufficient.
Combining limitation of RLC-buffer size with quality of service-class shows little dif-
ference to only RLC-buffer size limitation.
Clearly, the highest satisfaction ratio can be reached by limiting the contention-based transmissions to voice frames. For high load (50 VoIP-users per cell), these transmissions will have a satisfaction ratio of 10% while contention-free and unlimited contention-based transmissions have 0%.
See the list below for parameter settings.
Access method Contention-based, contention-free
Number of cells 21
VoIP users in system [21, 210, 420, 630, 840, 1050]
Web browsing users in system 1050 Dedicated scheduling request interval (D-SR) 40 ms
Limitation QoS-class (VoIP)
RLC buffer size (312 bits) Voice frames
Combination of QoS and buffer size
6.3. Mixed traffic 43
6.3.2 Capacity evaluation
Figure 6.4 depicts the cell capacity in average for three types of transmission method:
contention-free, unlimited contention-based and voice frame limited contention-based.
The voice frame limited contention-based transmissions clearly has the best capacity:
from 41 users at low load to 22 users at high load.
The capacity is defined as the load when the satisfaction ratio is at 95%. See Figures C.4, C.5 and C.6 in appendix C for the relevant graphs.
5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40 45
DïSR interval [ms]
Capacity [users]
CF
CB (voice frames) CB (unlimited)
