Chapter 2 System Model
2.5 Source Model
The conversational class traffic is modeled as the ON-OFF model [25] shown in Fig. 2.3. During ON period, voice packets are generated with rate Dv bps. During OFF period, there is no packet generated. This model has a transition rate with value y in the ON state and a transition rate with value z in the OFF state.
Fig. 2.3 : Packet trace of a typical on-off voice model
Fig. 2.4 depicts the packet trace of one video streaming session model, which is composed of a sequence of video frames generated regularly with a constant interval Tf [22]. Each video frame consists of a fixed number of slices Ns, where each slice corresponds to a single packet. The size of packet is denoted by Ps, and the inter-arrival time between each packet is Tp.
Fig. 2.4 : Packet trace of one video streaming session model
Fig. 2.5 shows the Packet trace of one HTTP session model. The interactive class traffic can be modeled as a sequence of packet calls (pages), and each packet call consists of a sequence of packet arrivals, which is composed of a main object and several embedded objects [22]. Four parameters, including the inter-arrival time
Treading (reading time), main object size Sm, embedded object size Se, the number of
embedded objects per packet call Ne, and the packet inter-arrival time Tp are used in this model.
Fig. 2.5 : Packet trace of one HTTP session model
The background class traffic is modeled as a sequence of file downloads [22]
and is shown in Fig. 2.6. Denote the size of each file by Sf, and the inter-arrival time between each file by Tf.
Fig. 2.6 : Packet trace of a typical FTP session model
Chapter 3
Preference Value-Based Cell Selection (PVCS) Scheme
3.1 Introduction
The proposed preference value-based cell selection (PVCS) scheme is composed of three stages, candidate cells selection, preference value calculation, and target cell determination. As shown in Fig. 3.1, for a MS’s call request, the candidate cells selection is used to filter out unsuitable cells. All suitable cells form candidate cell set C, where C={C1 , C2 , ... , CK }. The preference value calculation calculates each preference value Pi of candidate cell Ci , i =1, 2, ... , K, in the candidate cell set. These preference values are evaluated by considering the factors of loading, QoS, and mobility to maximize overall system utilization, maintain the call request’s QoS requirements, and minimize handoff occurrence frequency. Ultimately, the target cell determination selects a best-fit cell for the call request of a MS. In the following, each stage is described in details.
Fig. 3.1 : The block diagram of PVCS scheme
3.2 Candidate Cells Selection Stage
This is the first stage to sift out the candidate cells for the call request from all cells according to received signal strength constraint, cell loading constraint and dwell time constraint. A feasible cell has to meet these three constraints so as to be included in the candidate cell set C for a call request. If there is no cell in the candidate cell set C, the call request is blocked.
3.2.1 Received Signal Strength Constraint
The MS measures the received signal strength of pilot or beacon signal from the neighboring BSs or APs in order to pick out the cells which offer sufficient connection quality. If the received signal strength of pilot/beacon from cell i, denoted by RSSi, exceeds a given received signal strength threshold RSSth, the cell i will be categorized as an available cell which offer sufficient connection quality. In addition, the different power characteristics of different wireless radio interfaces make the predefined received signal strength threshold different.
3.2.2 Cell Loading Constraint
This constraint is used to guarantee that the admittance of a call request will not influence the QoS requirements of existing connections. The admission controller gathers loading information from all available cells. Assume that a call request is required to notify the BSs of its QoS requirements when it asks to access one cell to transmit data. The QoS requirements contain maximum bit rate, packet error ratio, transfer delay, etc. [24] Consequently, the loading intensity of cell i before admitting a new call request, denoted by ρi( 0≤ρi ≤ ), can be calculated as the sum of the 1 loading intensity of existing call requests in the cell i,
1 number of existing call requests in the cell i [18].
Moreover, if infinite buffer size is assumed, the equivalent capacity can be derived as mean bit rate [26]. Thus we define the mean bit rate of a call request as its equivalent capacity to estimate the cell loading intensity increment Δ when the ρ call request enters the cell. If the call request is admitted by cell i, the cell loading intensity should not exceed a predefined threshold, ρth, which is set by the radio network planning. The cell loading constraint can be expressed by
i th.
ρ + Δ ≤ρ ρ (3.1)
Otherwise, the cell will be excluded from the candidate cell set.
In the WCDMA system, the loading incrementΔ can be estimated as [18] ρ
0
where f is the other cell to own cell interference ratio, W is the chip rate of the WCDMA system, R is the bit rate of the new call request, v is the activity factor of the new call request, and Eb/N0 is the bit-energy divided by noise spectral density for
In the OFDMA-based WMAN system, the mean equivalent capacity of WMAN can be estimated as 4 q L N T× × × / (bps). Then the loading incrementΔ can be ρ estimated as
/(4Δ = ×ρ R v × × ×q L N T/ ), (3.3)
In the WLAN system, the measurement-based cell loading estimation is used.
Let Ts be the total busy occupation transmission time consisting of successful transmission time and collision time in the latest observation duration Td. The loading of cell i is defined as ρi =T Ts/ d. The available cell i will be included in the candidate cell set if the following condition is satisfied,
i th.
ρ ≤ρ (3.4)
Note that the values of ρth in WCDMA, WMAN, and WLAN systems could be different.
3.2.3 Dwell Time Constraint
Generally speaking, if an MS can stay in the cell as long as possible, the MS’s call request has high probability to finish data transmission in this cell. This also implies that the handoff occurrence frequency and signaling overhead for handoffs could be reduced. Assume that the maximal dwell time of an MS, which is acquired from the diameter of the cell i divided by the estimated velocity of the MS, is Tmaximum,i. The dwell time Tdwell,i for an MS in cell i can be calculated according to the estimated velocity, position, and direction of motion of the MS and cell coverage by using (2.7) and (2.8). Let xi =Tdwell,i/Tmaximum,i. xi = 1 means that the MS travels along the diameter of the cell i and has larger chance to reduce handoff occurrence frequency. On the contrary, if xi is smaller than a predefined threshold xth, this means that the MS would enter and leave the cell i quickly and has high probability to
initiate handoff. Note that the values of xth in WCDMA, WMAN, and WLAN systems could be different.
Accordingly, the constraint is designed to check if the dwell time for an MS in cell i is too temporal to be suitable for the candidate cell. That is
i th.
x ≥x (3.5)
Otherwise, the cell will be excluded from the candidate cell set.
Fig. 3.2 : The flow chart of candidate cells selection
3.3 Preference Value Calculation Stage
After obtaining candidate cells for the call request, all candidate cells in C compete with each other to serve the call request according to their preference values.
The meaning of preference value is to judge the degree of suitability for the target cell, which considers maximizing overall system utilization, maintaining the call request’s QoS requirements, and decreasing the number of handoff. To achieve these goals, the preference value of candidate cell Ci, denoted by Pi, is defined as a linear combination of three factors given by
+ ,
i i i
P =αL βQ +γMi (3.6)
where Li is the loading factor of candidate cell i, Qi is the QoS factor of candidate cell i, Mi is the mobility factor of candidate cell i, and α ,β,γ are the weights of different factors whose relation is α β γ+ + = . The meaning of the three factors and 1 how to design these factors are described as follows.
3.3.1 Loading Factor
The main concept of loading factor is to strike a balance between system utilization and cell loading. When a call request arrives, the PVCS scheme prefers to arrange it to the cell which can minimize overall cell loading differences. Through load balancing, it can achieve the objective of maximizing overall system utilization.
Accordingly, the loading factor Li can be defined as
K K K ρk are the loading intensity of cell j and k before accepting new call request. In the
denominator of equation (3.8), the first summation represents the differences in loading intensity between the cell i if the new call request is accepted and other cells.
The last dual summation represents all combinations of loading intensity differences except the cell i. Obviously, if the denominator of Li is smaller, it implies that the new call request would cause less loading difference if it is admitted by Ci, which is more likely to be a target cell.
3.3.2 QoS Factor
Utility is a measurement of call request’s satisfaction level with the perceived QoS. The different forms of utility function characterize different elastic traffic class.
In this thesis, voice services are assumed to be constant bit rate (CBR) traffic. Its QoS requirements on data rate, packet delay, packet dropping rate are so stringent that its utility function can be represented by a steep concave function. Video and HTTP services are assumed to be variable bit rate (VBR) traffic. Their QoS requirements are less stringent than voice services. Thus the slope of utility function is gentler than voice services. FTP services are assumed to be available bit rate (ABR) and greedy traffic, which means more data rate leads higher satisfaction level. An increasing concave function can describe its utility. Based on these utility concepts, a QoS factor Qi for each candidate cell i is expressed as a product of three QoS-related utility functions, which is given by
(3.8)
( ) ( ) ( ),
i i i
Q =U B ×U D ×U Ri
where U(Bi), U(Di), and U(Ri) are the normalized utility functions of data rate, packet delay, packet dropping rate for candidate cell i, respectively. Note that the last two utility functions are only used for real time services. As shown in Fig. 3.2, Fig. 3.3, Fig. 3.4, if a candidate cell has greater QoS measures to fulfill the QoS requirements
from utility functions would increase exponentially. On the contrary, if a candidate cell does not have sufficient QoS measures to satisfy the QoS requirements of the call request, the utility values would be zero.
Therefore, the utility functions of data rate for four traffic classes described in Section 2.4 are defined as
{ }
( ) max 1-i a b Bi i i,0 , i 0, 0,
U B = e − × a ≥ bi ≥ (3.9)
where BBi is the allowed data rate measured in the candidate cell i, ai is the requirement parameter, bi is the elasticity parameter, and By,req, y=voice, video, HTTP, and FTP denote the data rate requirement of the call request for four traffic classes, respectively.
Note that the allowed data rate measured in WCDMA can be obtained by (2.1), the achievable modulation order in WMAN can be estimated by (2.2), (2.3), and the allowed data rate measured in WLAN is obtained by measurement-based cell loading intensity estimation. Besides, the larger value of requirement parameter ai means higher data rate requirement and the larger value of elasticity parameter bi means steeper curve, that is, less elasticity in data rate requirement. To ensure , we define
( ) [0,1]i U B ∈
, /
y req i i
B =a b .
Fig. 3.3 : Utility functions of data rate for four traffic classes
Moreover, the utility functions of packet delay are defined as
(3.10) ( ) max{1-i c Di i di,0}, i 0, 0,
U D = e × − c ≥ di ≥
where Di is the average packet delay measured in the candidate cell i, ci is the elasticity parameter, di is the requirement parameter, and Dy,req, y=voice, video, is the maximum delay tolerance of the call request, respectively. Note that the values of average packet delay for real time services are computed separately. Different traffic class has different average packet delay in the same candidate cell. In addition, the larger value of requirement parameter di means higher maximum delay tolerance and the larger value of elasticity parameter ci means steeper curve, that is, less elasticity in delay tolerance. To ensure U D( ) [0,1]i ∈ , we define Dy req, =d ci/ i.
Fig. 3.4 : Utility functions of packet delay for real time services Similarly, the utility functions of packet dropping rate are defined as
(3.11) ( ) max{1-i e Ri i fi,0}, i 0, 0,
U R = e × − e ≥ fi ≥
where Ri is the average packet dropping rate measured in the candidate cell i, ei is the elasticity parameter, fi is the requirement parameter, and Ry,req, y=voice, video, is the maximum allowable dropping rate of the call request for real time services,
respectively. Similar to the above case, the values of average packet dropping rate for real time services are computed separately. Furthermore, the larger value of requirement parameter fi means higher maximum allowable packet dropping rate and the larger value of elasticity parameter ci means steeper curve, that is, less elasticity in packet dropping rate. To ensure U R( ) [0,1]i ∈ , we define Ry req, = f ei/ i.
Fig. 3.5 : Utility functions of packet dropping rate for real time services
3.3.3 Mobility Factor
The mobility factor considers two aspects of an MS, inclusive of the dwell time and the relative position. In this thesis, the proposed PVCS scheme favors to arrange a high mobility MS to large cell for the purpose of avoiding frequent handoffs. A small cell that would incur more handoffs has a less chance of becoming the target cell.
For an MS, it is assumed that the average dwell time of total candidate cells is Taverage. Let ri =Tdwell i, /Taverage. If ri is larger than one, the MS has larger chance to finish data transmission in the large candidate cell i. However, if ri is smaller than one, this means that the MS has high probability to initiate handoff when it enters the candidate cell i. Consequently, the mobility factor Mi for the aspect of dwell time can be defined as
,
On the other hand, the candidate cell i will be more suitable if the MS is nearer to the BS of candidate celli. The mobility factor will be larger. In order to decrease the number of handoff, the mobility factor will get smaller if the MS is farther from the BS of candidate celli. Therefore, the mobility factor Mi for the aspect of relative position can be defined as
, radius of candidate celli, and cri,th is a predefined threshold. Note that the values of cri,th in WCDMA, WMAN, and WLAN systems could be different. Finally, the mobility factor Mi can be calculated from the average of two aspects of an MS.
3.4 Target Cell Determination Stage
The preference value contains the preference from the viewpoints of the call request and the service provider. If the preference value of the candidate cell is maximal, it has the largest opportunity to satisfy the QoS requirement of call request and achieve the objective of service provider to maximize overall system utilization.
Consequently, the target cell determination is formulated as an optimization problem given by
(3.14)
* arg max{ },i
i = i P
where i is the ith candidate cell in the candidate cell set C, and i* is the index of selected candidate cell for the call request.
Chapter 4
Simulation Results and Discussions
4.1 Simulation Environment
The simulation environment contains 7 subnetworks overlapping with each other, which subnetwork is as shown in Fig. 2.1. The system parameters in the heterogeneous wireless access environment are listed in Table 4.1. The channel model and pedestrian (3 km/hr) and normal mobility (30 km/hr) and high mobility (80 km/hr) models of MS have been introduced in Chapter 2.
Table 4.1: System parameters for WCDMA, WMAN, and WLAN
Parameters WCDMA WMAN WLAN
Cell radius 2 Km 2 Km 0.1 Km
Frame (slot) duration 10 ms 5 ms 9 us
Carrier frequency 2 GHz 2.5 GHz 2.4 GHz
Number of cells 7 7 28
Loading intensity threshold (ρth) 0.85 1 0.85
Dwell time threshold (xth) 0.05 0.05 0.2
the other cell to own cell interference ratio (f)
0.55
Number of subchannels (L) 4
Number of data subcarriers per subchannel (q)
48
Number of slots per frame (N) 16
4.2 Traffic Model Parameters and QoS Requirements
As described in chapter 2, there are four classes of traffic considered. The traffic model parameters for conversational, streaming, interactive, and background traffic classes are shown in Table 4.2, 4.3, 4.4, and 4.5, respectively.
Table 4.2: Conversational Class Traffic Model Parameters
Component Distribution Parameters
ON time Exponential Mean = 1 sec
OFF time Exponential Mean = 1.35 sec
Packets per second Deterministic 50
Packet size Deterministic 28 bytes
Call holding time Normal Mean = 90 sec, Variance = 20 sec
Data rate during active period 11.2 Kbps
Activity factor 0.426
Mean data rate 4.77 Kbps
Table 4.3: Streaming Class Traffic Model Parameters
Component Distribution Parameters Inter-arrival time between each
video frame (Tf)
Deterministic 100 ms
Number of packets (slices) in a video frame (Ns)
Deterministic 8
Packet size (Ps) Truncated
Pareto
Data rate during active period 80 Kbps
Activity factor 0.8
Mean data rate 64 Kbps
Table 4.4: Interactive Class Traffic Model Parameters
Component Distribution Parameters Main object size (Sm) Truncated
Lognormal
Min. = 100 bytes, Max. = 2 Mbytes Mean = 10710 bytes, Std. dev. = 25032bytes Embedded object size (Se) Truncated
Lognormal
Min. = 50 bytes, Max. = 2 Mbytes Mean = 7758 bytes, Std. dev. = 126168 bytes Number of embedded objects
per page (Ne)
Truncated Pareto
Min. = 2, Max. = 53 Mean = 5.64, α = 1.1 Inter-arrival time between each
page (Treading)
Exponential Mean = 30 sec
Packet size Deterministic Chop from objects with size 1500 bytes
Packet inter-arrival time (Tp) Exponential Mean = 0.13 sec
Call holding time Normal Mean = 120 sec, Variance = 30 sec
Data rate during active period 92.3 Kbps
Activity factor 0.136
Mean data rate 12.55 Kbps
Table 4.5: Background Class Traffic Model Parameters
Component Distribution Parameters Inter-arrival time between each
file (Tf)
Exponential Mean = 180 sec
Packet size Deterministic 3000 bytes
Call holding time Normal Mean = 180 sec, Variance = 40 sec
Data rate during active period 88.9 Kbps
Activity factor 1
Mean data rate 88.9 Kbps
Different traffic classes have different QoS requirements. The QoS requirements of each traffic class are listed in Table 4.6.
Table 4.6: The QoS Requirements of each traffic class Traffic class Requirement Value
Required BER 10-3
Required Eb/No 4 dB
Max. delay tolerance 40 ms Conversational
(Voice)
Max. allowable packet dropping rate 1%
Required BER 10-4
Required Eb/No 3 dB
Max. delay tolerance 100 ms Streaming
(Video)
Max. allowable packet dropping rate 1%
Required BER 10-6
4.3 UGT based network selection scheme
The proposed PVCS scheme is compared with the UGT based network selection scheme [12]. When a new call or a handoff call arrives, UGT will find which networks are suitable for the call request first. After obtaining the candidate networks, UGT will compute the utility value from the satisfaction of QoS requirements of the call request and the network preference from predefined cooperative game for each candidate network. Finally, by choosing the maximum linear combination of utility values and network preference from all candidate networks, the most suitable network for the call request can be obtained.
4.4 Simulation Results and Discussions
It is assumed that a call request could only connect to one cell at the same time in this thesis. For each cell, it is assumed that the new call arrival rate of conversational, streaming, interactive, and background traffic class calls in the heterogeneous wireless access environment are AR, AR×1/3, AR×1/3, and AR×1/6 (calls/second), respectively, where AR is the unit arrival rate. In the simulation, AR is chosen from 0.1 to 0.9.
Fig. 4.1 shows the new call blocking rate. It can be found that PVCS has lower new call blocking rate in the high arrival rate. The reason is that PVCS chooses the cell which can minimize overall cell loading differences by using loading factor in order to achieve load balancing. However, UGT selects the cell which has the most resource. It may cause non-real-time calls to be blocked when the cell loading is close to the predefined threshold. The result comparison also implies that the system with PVCS can accommodate more calls than the system with UGT when the arrival rate becomes high. Number of accommodated calls is shown in Fig. 4.2.
Fig. 4.1 shows the new call blocking rate. It can be found that PVCS has lower new call blocking rate in the high arrival rate. The reason is that PVCS chooses the cell which can minimize overall cell loading differences by using loading factor in order to achieve load balancing. However, UGT selects the cell which has the most resource. It may cause non-real-time calls to be blocked when the cell loading is close to the predefined threshold. The result comparison also implies that the system with PVCS can accommodate more calls than the system with UGT when the arrival rate becomes high. Number of accommodated calls is shown in Fig. 4.2.