Chapter 2 System Model
2.4 Traffic Model
Since the storage capacity of eNBs can be enhanced easily by hardware devices, we basically assume that each serving UEs of eNB has infinite buffer. Because multiple services are provided by operators, various service traffics are proposed with
13
individual characteristics and requirements [39]. We focus on the four most common traffic types, which are voice traffic of real-time (RT) service, video traffic of RT packet would be dropped if the packet violates its maximum delay tolerance.
Voice traffic is modeled as a two-state voice activity in Figure 2.4, which corresponds to active and inactive states. The state update is made at the speech encoder periodically with changing probability from active to inactive state of Pa, and that from inactive to active state of Pi. However, it provides different fixed-sized packets to UE in active and inactive states with different fixed periods.
1-Pi
Video traffic is modeled as a stable video frame flow in Figure 2.5. It provides a fixed duration and a constant number of packets in one video frame duration.
14
Moreover, the packet size and the inter-arrival time between packets in a video frame are both based on the truncated Pareto distribution.
time
... ...
frame duration frame duration
inter-arrival time
Figure 2.5: Video traffic model.
HTTP traffic is modeled as a web-browsing behavior in Figure 2.6, which includes web-page downloads, reading and parsing duration. The web-page contains a main object and several embedded objects, and the number of embedded objects is based on the truncated Pareto distribution. Besides, the sizes of main and embedded objects are based on the truncated log-normal distribution. The reading duration is the time interval between two web-page downloads, and the parsing duration is the time interval between two objects in a web-page download, while the duration of reading and parsing are both based on the exponential distribution.
web-page downloads web-page downloads
main object embedded objects reading
duration
parsing duration
time Figure 2.6: HTTP traffic model.
15
FTP traffic is modeled as a sequence of file downloads in Figure 2.7. Its file size is based on the truncated log-normal distribution, and the inter-arrival time between two successive files is based on the exponential distribution.
file download
inter-arrival time
file download file download
time Figure 2.7: FTP traffic model.
However, the detail parameters of each traffic model would be described in Chapter 4.
16
Chapter 3
Cognitive Priority-based Resource Management Scheme for Macro- Femto Networks
Since the HeNB has the ability of cognitive radio, it could sense all sub-channels and choose the most suitable sub-channels to avoid interfering with MeNB and the neighboring HeNBs. Once the HeNB determines the available sub-channels to transmit data, it allocates sub-channels, modulation order, and power to its serving UEs. Moreover, the HeNB could serve HUEs and MUEs since it adopts the hybrid access policy. Each UE has its QoS requirements, which should be guaranteed by the resource management scheme. Therefore, the problem of radio resource management for the hybrid access HeNB with the ability of cognitive radio is very complex due to its multiple degree of freedom.
In this thesis, we adopt the priority-based service discipline to design a resource management scheme, which determines the priority value to each UE and then serves UEs based on their priority values. In order to maximize the throughput of HeNB and guarantee QoS requirements of its serving UEs, we first formulate the resource management problem of hybrid access HeNB to an optimal problem in the following.
After that, we propose a cognitive priority-based resource management (CPRM) scheme to find a sub-optimal solution for this optimal problem.
17 modulation order of QPSK, 16-QAM, or 64-QAM, respectively, on sub-channel n at the l-th sub-frame. Therefore, the total throughput of UE k at the current frame can be obtained by constraints to satisfy the limitation of HeNB.
(i) QoS Fulfillment Constraint for RT service
Since the RT UE k has the QoS requirements of the maximum delay
(ii) QoS Fulfillment Constraint for NRT service
Since the NRT UE k has the QoS requirements of the minimum transmission rate, its transmission rate, Rk, should be larger than or equal to R .k* We then have
18
*, NRT
k k
R R k , (3.3)
where ΨNRT is the set of NRT UEs. Note that, since BE UE only has the QoS requirement of BER and the BER requirement can be easily fulfilled by setting its transmission power, we do not consider the QoS fulfillment constraint for BE service here.
(iii) Sub-channel Allocation Constraint
In the macro-femto system, a sub-channel can be allocated to at most one UE at each sub-frame, since HeNB is only equipped with one transmit antenna.
Thus, the sub-channels allocation constraint is given by
1 ,
The total power allocation for downlink data transmission at HeNB should have a limitation. Let P* be the maximum transmission power. We then have the power budget constraint as
The sub-channel n is regarded as available if the detected RSSn is less than or equal to the threshold of RSS, RSSth. Hence, the cognitive sub-channel availability constraint can be expressed by
, 0, if , , , .
k n
n l th
b RSS RSS k n l (3.6)
(vi) HUE QoS Satisfaction Constraint
In order to avoid violating the QoS requirements when UE is urgent, each UE should transmit some bits at each frame. Since HeNB is mainly established to serve HUEs, the QoS requirements of HUEs must be guaranteed first. Let γk
19
be the minimum transmission bits of UE k at the current frame. We then have the HUE QoS satisfaction constraint by
1 , resource management problem of HeNB can be formulated to an optimal problem as follows,
It is complicated to find an optimal solution for the optimal problem of (3.8) by exhaustive search. Therefore, we propose the cognitive priority-based resource management (CPRM) scheme to heuristically find a sub-optimal allocation solution, B*, in (3.8) to maximize the system throughput and satisfy the QoS requirements of UEs.
20
3.2 Cognitive Priority-based Resource Management Scheme
The cognitive priority-based resource management (CPRM) scheme can sense the available sub-channels from the environment, assign a suitable priority value to each UE, and then allocate the radio resource to UEs according to their priority values.
Figure 3.1 shows the HeNB with the CPRM scheme. The CPRM scheme is in charge of sub-channel allocation, modulation assignment, and power allocation for HeNB. It contains three algorithms: cognitive channel determination (CCD) algorithm, priority and minimum bits determination (PBD) algorithm, and priority-based resource allocation (PRA) algorithm.
The input of the CCD algorithm is the sensing information, which is the RSS detected on each sub-channel at the last frame. According to the sensing information, the CCD algorithm determines the number of sub-frames required for HeNB to sense sub-channels and the set of available sub-channels for HeNB. The inputs of the PBD algorithm are users’ QoS fulfillment information and queue information. Based on these information, the PBD algorithm assigns a service priority value and the minimum transmission bits to each UE. Note that, the outputs from the CCD and PBD algorithms are the input of the PRA algorithm. The inputs of the PRA algorithm are the channel information, the QoS fulfillment information, the queue information, the set of available sub-channels, UEs’ priority values, and their minimum transmission bits. According to these inputs, the PRA algorithm can allocate the suitable sub-channel, modulation order, and transmission power to UEs. Details are given in the following.
21
Figure 3.1: The HeNB with the CPRM scheme.
3.2.1 Cognitive Channel Determination Algorithm
The cognitive channel determination (CCD) algorithm determines the minimum number of sub-frames required for HeNB to sense sub-channels, and finds the available sub-channels for HeNB to avoid interfering with MeNB and its neighboring HeNBs. Let Ln be the minimum required number of sub-frames for HeNB to sense
22
deviation on sub-channel n in false alarm problem, R is the sensing deviation on n sub-channel n in mis-detection problem, and Q(a) is the complementary distribution function of the standard Gaussian and is given by
2
respectively. By adding (3.11) and (3.12), we have
1 * 1 *
On the other hand, if the sub-channel n satisfies the cognitive sub-channel availability constraint of the system constraint (iii), it would be regarded as the available sub-channel. Therefore, the set of available sub-channels for HeNB, denoted by Φ, is given by
n RSSn RSSth and Ln L, 1 n N
. (3.15)
3.2.2 Priority and Minimum Bits Determination Algorithm
The priority and minimum bits determination (PBD) algorithm assigns the priority values to UEs according to their packet delay or transmission rate. It also determines the minimum transmission bits at the current frame to UEs based on the
23
QoS fulfillment information and queue information.
First, we introduce the time-to-expiration (TTE) parameter to represent the degree of urgency of UE [42]. The TTE value for UE k is denoted by Vk in unit of violate its delay requirement and its HOL packet will be dropped after the next frames.
On the other hand, for NRT UE k, Vk is derived from its minimum transmission rates;
that is If Vk < 0, the NRT UE k is regarded as urgent, since it already violates its minimum transmission rate requirement.
We have the following two priority assignment rules:
(i) HUE has a higher service priority than MUE.
(ii) RT UE has a higher priority value than NRT and BE UEs, while urgent NRT UE has a higher priority value than un-urgent RT UE.
Accordingly, the service priority of UE k, denoted by uk is designed as follows,
,
where ∆uk is the priority value of each service type, and is given by
*.
24 because the maximum value of ∆uk is 2. For the same service type, the smaller the Vk
is, the larger the ∆uk. The ∆uk of RT UE k is set to be , which is always larger than that of un-urgent NRT UE, since the maximum value of ∆uk of un-urgent NRT UE k is 1. The ∆uk of urgent NRT UE k is set to be μ, which is between the minimum value of ∆uk for urgent RT UE k (1.5) and the maximum value of ∆uk for un-urgent RT UE k (1.3). Therefore, the priority assignment rule (ii) is satisfied. On the other hand, the ∆uk of BE UE k is set to be 0, since the BE service is the background traffic. By the design of ∆uk, the QoS fulfillment constraints (i) and (ii) can be satisfied.
The minimum number of transmission bits allocated to UE k at the current frame to avoid violating QoS requirements, γk, is given as packet should be completely transmitted at the current frame. Because Vk of NRT UE k may be less than 1, the denominator of γk is given by max{1,Vk}. Since BE HUE is
1 1
k 1
V
25
the subscribed UE of HeNB, it should also get the service. We consider that the BE HUE should obtain at least one resource unit. Thus, the γk of BE HUE k is set to be
3Ne,which is the average transmission bits at one resource unit. Moreover, the value of γk of BE MUE k is set to be 0, since it is not the subscribed UE of HeNB.
3.2.3 Priority-based Resource Allocation Algorithm
The PRA algorithm allocates sub-channel, modulation order, and transmission power to UEs with the maximum priority value.
Let Ψc be the set of backlogged UEs with the maximum priority value, where
1max and 0, 1
.c k uk k Kuk Bk k K
(3.20)
In order to achieve high system throughput and save the transmission power, the optimal pair of UE k* and sub-channel n* is selected based on the maximum SINR by
* * power budget constraint. Based on the achievable SINR between UE k* and HeNB on sub-channel n*, denoted by k**, system constraint (ii), the modulation order with **,
k
26
k the PRA algorithm will allocate the next sub-frame of sub-channel n* to UE k* until k* 0 or the sub-channel n* has no void sub-frame.
Once the resource allocated to UE k* is completed, the CPRM scheme sets its priority value to be 0 and looks for another best pair of (k*, n*). If the sub-channel n* has no void sub-frame for data transmission, it is removed from Φ. Note that, in order to satisfy the sub-channel allocation constraint of the system constraint (i), one sub-channel is allocated to one UE at each sub-frame. If there is still the residual resource after all UEs are served (uk ≤ 0, 1 ≤ k ≤ K), the PRA algorithm sets γk = Bk, 1
≤ k ≤ K, and redo (3.21) to allocate the residual resource to the backlogged UE with the maximum SINR. The procedure of the PRA algorithm is repeated until all radio resource is allocated to the selected UEs or no backlogged UEs exists.
3.2.4 Summary of The CPRM Scheme
Based on the CCD, PBD, and PRA algorithms, the CPRM scheme can find a sub-optimal solution for (3.8) to maximize the system throughput of HeNB, and guarantees the QoS requirements of its serving UEs. The CPRM scheme can be summarized as the following five steps.
Step 1. Initialize the CPRM scheme. Set Ln of all sub-channels by (3.14) in the CCD algorithm. Also, set uk and γk of all serving UEs by (3.17) and (3.19),
27 respectively, in the PBD algorithm.
Step 2. Find the set of the available sub-channels, Φ, by (3.15) in the CCD algorithm.
Only when |Φ| > 0 and l ≤ L, the CPRM scheme would go to step 3.
Step 3. Find a pair of UE k* and sub-channel n* by (3.21) in the PRA algorithm. Then, allocate suitable and by (3.22) and (2.7), respectively, to UE k* on sub-channel n* at the l-th sub-frame. If |Ψc| = 0, end.
Step 4. Pre-allocate the next sub-frame to the selected UE k*. If * 0
k and lʹ ≤ L, set untilk* 0.
Step 5. If |Φ| > 0 and * 0,
k go to step 3. If |Φ| > 0 and uk ≤ 0, 1 ≤ k ≤ K, set γk = Bk, 1 ≤ k ≤ K, and go to step 3. If |Φ| = 0, end.
The pseudo code of the CPRM scheme is given as follows.
,
3: // Find the set of available sub-channels.
4: Find by (3.15).
5: // If there is available sub-channel, perform the CPRM scheme.
6: 0 and
12: // Find an optimal pair of UE and sub-channel.
13: Find ( , ) by (3.21).
16: // Check the system constraint (ii).
17:
27: // Pre-allocate resource to UE on sub-channel .
{ }
33: // If there is residual resource and UE
k
* still has the required bits, continue to allocate the resource to the UE with the required bits.
34: 0 and 0
35: Go to line 7.
36: // If there is residual resource and all backlogged UEs
k
if then
can compete it, allocate the resource to the UE with their residual HOL packet bits.
37: 0 and == 0, 1 ,
Check whether this frame is end.
44:
28
3: // Find the set of available sub-channels.
4: Find by (3.15).
5: // If there is available sub-channel, perform the CPRM scheme.
6: 0 and
12: // Find an optimal pair of UE and sub-channel.
13: Find ( , ) by (3.21).
16: // Check the system constraint (ii).
17:
27: // Pre-allocate resource to UE on sub-channel
{ }
33: // If there is residual resource and UE
k
* still has the required bits, continue to allocate the resource to the UE with the required bits.
34: 0 and 0
35: Go to line 7.
36: // If there is residual resource and all backlogged UEs
k the resource to the UE with their residual HOL packet bits.
37: 0 and 0, 1 ,
29
Chapter 4
Simulation Results
4.1 Simulation Environment
In the simulations, parameters of the considered macro-femto networks are set to be compatible with the 3GPP E-UTRA standard [36]. The parameter values are listed in Table 4.1. The system bandwidth in each sector is 5 MHz, which is divided into 25 sub-channels. Each sub-channel has 144 allocation units for data transmission.
Therefore, the maximum transmission rate per macro sector is equal to 21.6 Mbps, which is obtained when each sub-channel delivers data by the highest modulation order of 64-QAM. Each sector has 3 femto blocks with the same number of deployed HeNBs of 16. The total traffic intensity (ρ) is defined as
total arrival rates of all traffics
maximum transmission rate per macro sector.
(4.1)
In each femto block, the numbers of voice, video, HTTP, and FTP UEs are set to be the same, and each HeNB is deployed with the same number of HUEs. The number of MUE is twice than that of HUE in each traffic type. Each HUE would stay in a fixed position, and each MUE would move in any direction from [-π, π) with constant speed of 3 km/hr. If total resource units in a frame duration are equally divided to HeNBs in a femto block without overlapping, then we assume every HeNB would utilize at most 15 resource units for data transmission in a frame duration.
Besides, we assume that MeNB has utilized total bandwidth, and divides its total
30
transmission power fairly on the bandwidth. On the other hand, if MUE that needs service can detect the highest average SINR on bandwidth among HeNBs whose serving UE number is less than the maximum number of serving UEs, we assume the MUE would automatically hand over to that HeNB in the beginning of frame.
However, to evaluate the throughput gain from the deployment of HeNBs, we will consider the pure macro network. The pure macro network is similar to the macro-femto networks but only one difference, which is that all HeNBs are disabled and MeNB would use total bandwidth to serve all HUEs and MUEs.
Table 4.1: Parameters of the macro-femto networks.
Parameters Value
Carrier frequency 2 GHz
Frame duration 10 ms
Number of sub-frames in a frame duration (L) 10
Radius of macrocell 1000 m
Radius of femtocell 10 m
Total transmission power of MeNB 43 dBm
Total transmission power of HeNB (P*) 20 dBm Pre-determined probability of false-alarm ( ) 0.1 Pre-determined probability of mis-detection (Pm*) 0.1
Thermal noise density -174 dBm/Hz
4.2 Traffic Model and QoS Requirements
Since the voice traffic is modeled as a two-state voice activity, the state update is made at the speech encoder every 20 ms with changing probability from active to inactive state of 0.1, and that from inactive to active state of 0.1. During the active period, the encoder generates a 40 bytes packet to UE every 20 ms. During the inactive period, the encoder generates a 15 bytes packet to UE every 160 ms. Thus, the arrival rate of voice traffic is 8.375 Kbps.
In the video traffic, the video frame arrives every 100 ms, and contains 8 packets,
*
Pf
31
while the size of packet and the inter-arrival time between packets are both truncated Pareto distributed. The parameters of the size of packet and the inter-arrival time are given in Table 4.2, and the arrival rate of video traffic is 64 Kbps.
Table 4.2: Video traffic model parameters.
Component Distribution Parameters
packets in a frame Truncated Pareto
Min. = 2.5 ms, Max. = 12.5 ms, Mean = 6 ms, α= 1.2
In the HTTP traffic, the reading duration and the parsing duration are exponen- tial distributed with mean of 30 s and 0.13 s, respectively. The sizes of main and embedded objects are truncated log-normal distributed. The number of embedded objects is truncated Pareto distributed. The parameters of the sizes of main and embedded objects, as well as the number of embedded objects are given in Table 4.3, and the arrival rate of HTTP traffic is 14.178 Kbps.
Table 4.3: HTTP traffic model parameters.
Component Distribution Parameters
Main object size Truncated Log-normal
Min. = 100 bytes, Max. = 2 Mbytes, Mean = 10710 bytes, std. dev. = 25032 bytes
Embedded object size Truncated Log-normal
Min. = 50 bytes,
32
In the FTP traffic, the FTP file size is truncated log-normal distributed, and the inter-arrival time between two successive files is exponential distributed with mean of 180 s. The parameters of the file size are given in Table 4.4, and the arrival rate of FTP traffic is 88.89 Kbps.
Table 4.4: FTP traffic model parameters.
Table 4.4: FTP traffic model parameters.