Você está na página 1de 18

Wireless Pers Commun (2012) 66:595612

DOI 10.1007/s11277-012-0739-7

QoE Model Based Optimization for Streaming Media


Service Considering Equipment and Environment
Factors

Bingjun Han Xin Zhang Yifei Qi Yuehong Gao


Dacheng Yang

Published online: 15 July 2012


Springer Science+Business Media, LLC. 2012

Abstract With the development of smart phones, one of the hot research topics for wireless
personal communication technologies has begun to transit from reaching higher communi-
cation capabilities to better Quality of Experiences (QoE). However, for streaming services,
such as online music or online video, which is one of the mainstream services in the future
wireless network, there is still no breakthrough optimization beyond the Quality of Service
(QoS) hence the optimization results are not satisfactory. Based on our study, this problem
could be relieved by altering the criterion of wireless resource allocation from enhancing
users QoS to enhancing their QoE, which specifically means that the scope of the research
is extended from the so-called last mile between the network and the user equipment to the
last few inches between the user equipment and the user himself. Unfortunately, it seems that
all current QoE researches are still not jointly considering the users environment and user
equipments performance, both of which are paramount factors of the last inches that will
influence users QoE directly and profoundly. Hence, in this paper, we propose a QoE model
involving the environment and equipment factors, and then we use this model to guide the
wireless resource allocation. Simulation results show that this resource allocation method
will increase the effectiveness of network resource utilization as well as improve the average
of satisfaction of users.

Keywords QoE model Environment Equipment Resource allocation

B. Han (B) X. Zhang Y. Qi Y. Gao D. Yang


Wireless Theory and Technology Lab, Beijing University of Posts and Telecommunications,
Beijing 100876, China
e-mail: hanbingjun@gmail.com
X. Zhang
e-mail: zhangxin@bupt.edu.cn
Y. Qi
e-mail: buptqiyifei@gmail.com
Y. Gao
e-mail: yhgao@bupt.edu.cn
D. Yang
e-mail: yangdc@bupt.edu.cn

123
596 B. Han et al.

1 Introduction

With the maturity of 3G networks and the trial of 4G technology, mobile internet business
will advance more rapidly and one of the hot research topics for wireless personal commu-
nication technologies has begun to transit from reaching higher communication capabilities
to better Quality of user Experiences (QoE). The increasing growth of streaming service,
such as online music and video, leads to a high demand on network capacity. This increasing
demand requires a guarantee of adequate network resources and a consistent network oper-
ating environment. Therefore, proper utilization of network resources is equally important
and mutually complementary.
People are the major end-users of a network. Hence the improvement of QoE is the ulti-
mate goal of improving Quality of Service (QoS). Consequently, the main focus on network
resource management has being shifted from QoS to QoE recently. Many articles have already
been published on the subject of QoE(as follows).
In [1], the concept of QoE is defined as general acceptable level of a service, or the user
experience of a certain service. Then, the users QoE is quantified to 5 levels of users sat-
isfaction with the service from level 1 to level 5, which are excellent, good, fair, poor and
bad respectively [2]. In [3], an approach which uses the full reference model is presented
to estimate users QoE of IPTV (Interactive Personality TV) system. Then, the authors of
[4] propose an original and measurable indicator of QoE in EDGE (Enhanced Data Rate
for GSM Evolution) networks, i.e., busy hour throughput per user. Furthermore, a QoS and
QoE correlation model is established in [5], where the QoE evaluation method using QoS
parameters of voice service is studied. Kilkki [6] writes that Quality of Experience includes
everything that really matters. He makes a strong case for a holistic approach to QoE and
presents an initial framework which bridges different research communities from a multidis-
ciplinary perspective. Influenced by Kilkkis advocates, Geerts et al. [7] gathers researchers
from different domains to contribute their findings to an integrated QoE framework, which
consists of four main components, user, (ICT) product, usage process and context.
Various models have been built to measure QoE [811]. Among which the idea Quanti-
fying the QoE by both objective and subjective parameters [12] is widely accepted. Because
quantitative data enable combined analyses of objective and subjective measures for deriv-
ing global QoE ratings [13], researchers integrate objective, technical aspects and subjective,
user-related dimensions in their QoE measurement approach [14]. Norwegian researchers
[15] compare the QoE model as a tree with two branches: measurable and nonmeasurable
parameters.
As the end-to-end QoE assurance is particularly challenging in the multi-vendor, multi-
provider, and multi-network environment of NGN [16], researchers propose some possible
optimization schemes based on their QoE models [17,18]. It is generally agreed now that the
optimization of QoE should be devised by two approaches: service level performance mea-
surements using statistical samples and Network Management System (NMS) using QoS
parameters [19], or simply described as user-centric and technical aspect. The basic rule
for the optimization scheme is that the QoE/QoS reporting components measure QoE and
QoS performances received by users, and then report them to networks and sources. The
QoE management components adjust transport functions and reconfigure application-layer
parameters to maximize QoE. Through the optimization schedule supported by statistical
analysis, operators and service providers will be able to maintain graceful QoE profiles and
optimize their network resources utilization [20,21].
However, in the literatures mentioned above, the User Equipment (UE) and the
environment have not been involved in the QoE model. Nevertheless, based on our

123
QoE Model Based Optimization 597

preliminary conclusion supported by laboratory studies, the two are both key factors affecting
QoE.
Both the experience of life and our preliminary experimental results show that, as for the
user equipment, customers who use a high-performance UE are more likely to differentiate
the high-quality signal and the low-quality signal, while users with low-performance UE
often fail to do this. For example, a user with a high quality pair of headphones can easily
distinguish the difference between lossless music and the highly compressed music, while
people who use a lower-performance headphones or even a phone headset can hardly tell the
difference. Further evidence shows, a user with a big screen laptop can easily distinguish
the High Definition (HD) video signal and the Standard Definition (SD) video signal while
people often ignore this difference on a cellphone screen. These examples show that, to some
degree, we can lower the quality requirement for the signals on some low-performance UE
to save network resources.
Likewise, as for the environment and potential environmental interference, our preliminary
experiment shows that, users tend to have a higher expectation for the signals quality when
they are in a relatively comfortable and undisturbed place. And as the environmental inter-
ference increases, this expectation drops. In this case, even if the users are provided with
high-quality signals, the QoE will still not be that satisfying. For example, in a quiet place,
users will get a better QoE feeling by listening to lossless music than the compressed music.
So we can assume that users then have a high expectation for the quality of signal. However,
when located in a rather crowded environment, lossless music or compressed music seem to
make little difference. That is because the existence of the environmental interference has
lowered users expectation for the signals quality. Similarly, users can easily distinguish the
difference between HD video and standard video signal in a place with comfortable light. Yet
when the light becomes too bright, especially when users couple with physical disturbance
(shaking), such as taking a bus, they will only care about the general meaning of the video
to help them pass some time, regardless of the quality difference between HD video and SD
video.
All in all, we find out that providing higher-quality signals to higher end devices and lower
interference environments will lead to a better QoE. On the contrary, sacrificing the quality
appropriately to users with low-performance UE or in an environment with serious interfer-
ence will not cause the QoE to go further down. It is exactly based on this conclusion that
a new QoE-based method of network resource allocation is proposed. The basic idea in the
method is to test the hardware and environment parameters by UE itself and then adjust the
demand for the quality of network signals according to the result of the test, in order to meet
the needs of users on several different levels. Note that many smart phones have embedded
sensors to detect surrounding noise, light and shaking, which facilitates the implementation
of QoE-based method.
Then, the following parts of the article are organized as follows. Section 2 introduces the
system model, including the QoE model with environment and equipment factors in Sect. 2.1,
the model of online video and online music service in Sect. 2.2, and the model of resource
allocation in Sect. 2.3. Section 3 describes the assumptions and simulation results. The Sect. 4
is about conclusion and future work.

2 System Model

In this section, the system model to be used is described from three aspects. In the QoE
model part, with respect to QoS, environment and equipment factors, we introduce a QoE

123
598 B. Han et al.

Fig. 1 QoE model considering environment and equipment factor

framework as well as a normal QoE model which can be used under this framework. Further-
more, the proposed specific model and the network implementation architecture based on this
model are described. Then, in the service modeling part, we introduce network structure of
streaming media such as online music and video service, and propose a simplified streaming
media service model. At last, in the resource allocation part, the QoE based resource allo-
cation procedure is demonstrated, and an example of resource allocation in a time division
wireless system is given.

2.1 QoE Model

According to our previous description, the QoE model is presented in Fig. 1.


In Fig. 1, the network means a radio access network, e.g. a 3G cellular network, which
can be easily generalized to other types of networks. An original signal on a certain quality
level is transmitted from the network side, arrives at UE of different performance through
wireless channel, and finally reaches the end users from user equipments. In the last mile
from the network to the user equipment, the signal suffers from interference/fading in wire-
less environment, which may lead to frame errors, delay, jitter and so on. Together with that
the quality of original signal may be not fixed at one level, the final signals received by users
have different QoS levels. In the last inches from the user equipment to the user himself, the
interference from surrounding environment, such as noise, light, shaking and so on, may lead
to different QoE levels for the user. Based on our previous analysis, we can see that QoE is
ultimately decided by all the three factors: user equipment, signals quality the UE received
and the environment interference of the UE. Thus, we have
 
L i,u
QoE = f u
QoE L i,u
QoS , L i,u
EI , L i,u
EQ (1)

where L i,u i,u


QoE denotes user is final QoE level of service u. Similarly, L QoS represents the
quantitative parameter reflecting user is QoS. Likewise, L i,u E I is the quantitative parameter
reflecting interference from the environment of user i. L i,u E Q is the quantitative parameter
with respect to user is equipment (referred as U E i later on) performance. f QoE u is a func-
tion mapping to QoE level from QoS, equipment and environment factors. According to this
model, we can integrate the influence of signal quality, equipment and environment factors
on QoE in a comprehensive manner. More specifically, we have
 
u, j
L i,u
QoS = f u
QoS p u,1
QoS , p u,2
QoS , . . . , p QoS (2)

123
QoE Model Based Optimization 599

u, j
where p u,1 u,2
QoS , p QoS , . . . , p QoS are j types of parameters measured by U E i corresponding to
its QoS of service u. For example, p u,1 u,2
QoS represents frame error rate, p QoS represents packet
u, j u , which is
delay, p QoS represents jitter and so on. All these parameters are applied to f QoS
a function to calculate the quantitative value of user is QoS. Apparently, these parameters
vary among different kinds of services or networks. And it also needs to be emphasized that,
these parameters should at least include those who define the quality of the original signals
and who indicate the influence of the wireless interference and fading.
Similarly, L i,u
E I is defined as
 
L i,u u u,1 u,2 u,k
E I = f E I pE I , pE I , . . . , pE I (3)

u, j
where p u,1 u,2
E I , p E I , . . . , p E I are parameters measured by U E i which contribute to environ-
ment interference on service u, while k denotes total kinds of these parameters. For example,
if u represents online music service, then p u,1 E I may represent volume of environmental noise,
p u,2
EI may represent frequency of environmental noise and so on. If u represents online video
service, then p E I may represent the intensity of the visible light accordingly, while p u,2
u,1
EI
represents the shaking of the UE and so on. With the function f Eu I being applied, these
parameters will be eventually transformed into L i,u E I that indicate the severity of the envi-
ronment interference. Accordingly, for different services, the content and the pattern of the
function are also not the same.
Lastly, L i,u
E Q is defined as
 
L i,u u u,1 u,2 u,l
E Q = f E Q pE Q , pE Q , . . . , pE Q (4)

where p u,1 u,2 u,l


E Q , p E Q , . . . , p E Q represents l types of parameters that indicate hardware capa-
bilities of U E i . For instance, p u,1 u,2
E Q may represent the size of the phone screen, while p E Q
represents the brightness of the screen, and p u,lE Q represents the volume of the phone. Sim-
ilarly, f Eu Q is a function that transforms these parameters into the hardware-related overall
level L i,u
E Q of the service u. Likewise, for different services, the involved hardware parameters
and the form of functions are different.
Our preliminary researches show that the eventual QoE is, to a great extent, limited by the
lowest-performance factor among the environment, QoS and equipment. In the meanwhile,
the users expectation for QoE can be adjusted according to the available device and envi-
ronment. Another thing worth mentioning is that these three factors may be not completely
independent. For instance, the QoS and environment factors can lead to different impact,
since UE may have different hardware performance and different anti-interference abilities.
More specifically, high-quality music signals will often get lower QoE on a cellphone handset
than that on a closed headphone because of the low sound distortion and enclosed design of
closed headphone. Similarly, a high-brightness screen can get better QoE when displaying
high-quality video. In conclusion, considering the reasons above, we assume that the eventual
QoE obeys the following equation:
 
L i,u
QoE = min L i,u
QoS , i,u i,u
L EI , L i,u
EQ (5)

where i,u is a ratio between 0 and 1 that indicates the anti-interference ability of U E i .
The bigger i,u is, the stronger anti-interference ability the UE possesses. The point to be
emphasized here is that the value of L i,u i,u i,u i,u
QoS , L E I , L E Q must be delicately designed in

123
600 B. Han et al.

Fig. 2 Framework of UE and network

order to adapt to minimum function. For example, if i,u L i,u i,u


E I and L E Q are defined as level
10, the best QoS that could be displayed need to be defined as level 10 accordingly.
According to the proposed QoE model, when the equipment and environment factor have
already deteriorated and caused the reduction on QoE, providing high-quality signal is actu-
ally a waste of resources. We may lower the signals quality in order to save network resources.
Indeed, the actual QoE model may be more delicate, but it does not hinder us to apply this
simple model to analyze the correctness of this concept and the effectiveness of this method.
Based on this model, we present the UEs new operation framework as Fig. 2 shows.
q
In Fig. 2, pe1 , . . . , pe are all types of parameters measured by a certain UE with respect to
its environment, while ps1 , . . . , psr represent all types of hardware parameters of equipments
detected by UE, where q and r represent the number of types. Based on the type of service, as
well as formula (1), (3) and (4), the analysis module inside UE chooses a part of the parame-
ters, analyzes these parameters and sends a request about QoS level to the network. Based on
this request, together with wireless resource related parameters chosen from pwr 1 , . . . , pv
wr
w
and network operation condition related parameters chosen from pnoc , . . . , pnoc the network
1

finally decides original signal with certain QoS to the UE.

2.2 Service Modeling

In this part, a model of streaming media such as online music and online video is proposed. In
order to get a reasonable model, we first introduce the network structure of wireless streaming
service, then focus on the bit rate and propose a simple model of streaming media. First of
all, the network structure of streaming media service is shown in Fig. 3 [22].
In Fig. 3, the conversion server transforms a certain streaming media into original data
with different quality and stores it in the Content Services (CS). For example, it transform
a video into HD, SD and compressed definition videos and store them in the video content
services. There might be one or more CSs, in charge of storing and delivering data according
to the command from control servers. The control server is in charge of analyzing users
requirements and the network condition, and instructing the CSs to transmit appropriate data
to users, functional equivalent to the network analysis model in Fig. 2. Upon receiving the
instruction, the CSs forward the data to the base stations, which will transmit the data to

123
QoE Model Based Optimization 601

Fig. 3 Network structure of wireless streaming media service

UEs via wireless channels. For instance, according to control servers command, they trans-
mit the HD video to the laptop, SD video to the smart phone usually with large-screen and
compressed video to the ordinary phone. The received data will be stored in the buffer of the
UE, and the playing software in the UE reads the data in the buffer and transforms them into
video or audio signals which are then presented to the user. Before the data in the buffer runs
out, the player works normally; otherwise, the player pauses and waits for buffering.
The quality of steaming media service is usually related to the bit rate (i.e. the data rate
after compression coding). As mentioned before, to adapt to different network conditions,
CSs usually provide video or audio signals with different code rates to get better performance.
In brief, with regard to a certain video or audio, the better the signal quality is, the greater the
bit rate will be, and there will be more content needed to be transmitted to UE in a unit time.
On the contrary, signals of poor quality requires to buffer less content in a unit time. As for
the audio service, we check some typical bit rates and the results are shown in Table 1.
From Table 1, we can see that a bit rate of 24 kbps is able to meet the requirement of
signals of broadcast level quality, while the code rate has to reach 64192 kbps to meet MP3
quality. There exists a great difference between them.
For video services, there has not been a unified standard. So we have to measure different
bit rate and the video size from online video website and network TV, as well as some screen
size parameters from the website. The result is given in Table 2.

123
602 B. Han et al.

Table 1 The bit rate and


Bit rate Audio quality
corresponding audio quality
(kbps)

16 Voice on phone
24 Enhanced voice on phone, short wave broadcast, long
wave broadcast, European standard medium wave
broadcast
40 American standard medium wave broadcast
56 Voice in speech
64 Enhanced voice in speech (the optimal bit rate value of
ringtones and monophonic MP3 player on cell phones)
112 FM stereo broadcast
128 Tape (the optimal value of stereo MP3 player on cell
phones and low-grade MP3 players)
160 Hi-Fi (the optimal value of medium top grade MP3
players)
192 CD (the optimal value of top grade MP3 players)
256 Music studio (suitable for music enthusiasts)

Table 2 The bit rate and


Bit rate Video size Scenarios
video quality
(kbps) (pixels)

1,600 1280*720 Laptop screen, HD video


1,200 1024*768 Tablet computers screen ,VCD
1,000 850*480, 640*480 Smart phone screen, network TV
800 640*360 Smart phone screen,
600 640*360 Common size on video network such as
500 640*360 Youtube
300 400*226, 320*240 Ordinary phone,
200 400*226, 320*240 Poor size on video network
100 320*240
50 176*144 Poor performance phone

Table 3 The QoS level of online music

Level 10 9 8 7 6 5 4 3 2 1

Bit rate (kbps) 256 192 160 128 112 64 56 40 24 16

Table 2 indicates that with regard to laptops, their hardware devices can easily support
a code rate up to 1,600 kbps. However, for general mobile phones with a 320*240 pixels
screen, the maximum code rate supported is only 264 kbps. There is a gap of several times
between them. Furthermore, we find that for videos of the same size, 640*360 for example,
if code rate is decreased by a half, the loss of display effect is not obvious, thus not easy to
be perceived in some condition.

123
QoE Model Based Optimization 603

Table 4 The QoS level of online video

Level 10 9 8 7 6 5 4 3 2 1

Bit rate (kbps) 1,600 1,200 1,000 800 600 500 300 200 100 50

Table 5 Type of users


E & E/WF Good Bad

Good High priority users Condition-constrained users


Bad Resource-constrained users Low priority users

According to the results above, we can relate the streaming media quality to bit rate. For
the original data transformed from a certain streaming media, the higher the bit rate is, the
better its quality will be, and vice versa. The bit rates adopted in this paper for voice service
and video service are given in Tables 3 and 4 respectively.

2.3 Resource Allocation

With consideration of user Environment and Equipment (E&E) factor, network resource
allocation scheme needs corresponding adjustment to get better user experience.All users
can be categorized into four types based on their E&E factor and Wireless Factor (WF) as
listed in Table 5.
Traditional resource allocation approaches do not take E&E factors into account, therefore
some redundant network resources are assigned to condition-constrained users. However,
these network resources contribute little to the improvement of QoE, resulting in a waste
of resources. Considering user E&E factors, we can transfer these underutilized network
resources to resource-constrained users, thus enhancing the efficiency of network resource
utilization.
It is assumed that the max QoE level of service u that user i could get as the E&E con-
strained is denoted by L i,max
QoE,u , which could be calculated by formula (1). And the network
resources allocated to user i are denoted by G i1 , G i2 , . . . , G ih , where G i1 , G i2 , . . . , G ih are dif-
ferent types of network resources, e.g. type 1 is time slot, type 2 is frequency band , . . . , type
h is transmit power. After network resource allocation, the actual QoE experience received
by the user is denoted by L i,actual
QoE,u . The objective of network optimization is to work out the
value of G i1 , G i2 , . . . , G ih to maximize users average L i,actual i,max
QoE,u /L QoE,u , then
 
1  i,actual i,max
m
argmax (L QoE,u /L QoE,u ) (6)
G i ,G i ,...,G i m
1 2 h i=1

where m is the total number of involved UEs. The constraint of formula (6) is given below.


( m Gi )  G1
i=1 1
..
. (7)


m
( i=1 G h )  G h
i

where G 1 , . . . , G h is the total wireless resource of type 1, . . . , type h respectively. It is


straightforward that for every kind of network resource, the total effective resources allo-
cated to all users should be less than the total network resources.

123
604 B. Han et al.

u, j
Users network resources will affect QoS parameters p u,1 u,2
QoS , p QoS , . . . , p QoS in formula
(2), so we have the following formula.
u,1 u,1

p QoS = f QoS,G (G i1 , G i2 , . . . G ih )

p u,2 = f u,2 (G i , G i , . . . G i )
QoS QoS,G 1 2 h (8)

...

u, j u, j
p QoS = f QoS,G (G i1 , G i2 , . . . G ih )
u,1
where f QoS,G is the mapping relationship from the allocated network resource to QoS quality
parameters. Combining the above with formula (1) and (2), we got the relationship between
G i1 , G i2 , . . . , G ih and L i,actual
QoE,u , then the optimization result of G 1 , G 2 , . . . , G h could be
i i i

worked out.
So in practice, according to environment and equipment factors, we can calculate the
best QoE level of service u that a user can achieve. Together with the mapping relationship
between QoE and QoS, and the mapping relationship between QoS and network resources,
we can figure out the optimized configuration scheme.
In the following part, we take the online music service as an example under a time divi-
sion cellular downlink scenario, such as WCDMA HSDPA or CDMA2000-EVDO downlink
system, to illustrate the configuration process of network resources. The QoE formula we
use is (5).
The wireless resource involved here is the number of time slot G islot in a certain time
period. Based on the serving model part, the QoS parameter involved here is the throughput
of user i. Usually, with different wireless conditions, the effects produced by resource G islot
are different. For example, a time slot may bring greater throughput for users with good
channel condition while the same time slot fails to do that for users in the edge of cells.
We use a coefficient ri to analyze the difference in channels. Then ri G islot represents the
throughput of user i, and we have
i
T hraudio = ri G islot (9)
According to Table 3, we can work out QoS level. Combining Table 3 with formula
(5), the relationship between G islot and QoE level could be obtained, so we could find the
optimization method to work out G islot .
However, it is a complicated process. To simplify it, we propose a simple scheduling
mode called QoE Based Scheduling (QBS) to be used in streaming media scenario in detail
as follows. We assume that the total slots in a scheduling period that a base station could
allocate is G total
slot .
1. Based on formula (5), all users compute the maximum QoS level they need according to
their equipment and environment factors. The maximum QoS level of user i is denoted
by L i,max
QoS .
2. Based on formula (9) and Table 3, the users calculate the number of slots they need
to promote one level of QoS based on the current QoS level, the number of slots use i
needs is G ib,b+1 where b is the current QoS level. Note that when b + 1 > L i,max
QoS , the
required slots will be set to infinite.
3. Based on formula (9), the users calculate the raise ratio of QoE b,b+1
i = (L i,b+1
QoE
L i,b i,max i,max
QoE )/L QoE , where L QoE represents user is QoE level when QoS level is b.
4. The network selects the user with the maximum ib,b+1 /G ib,b+1 value, which indicates
the efficiency of that time slot, then the slot planed to be allocated to user i will be added
G ib,b+1 .

123
QoE Model Based Optimization 605

5. Go to step 1 if there are slot resources left in the base station.


6. When the slot resource left is not enough to promote one QoS level for any users, the
users calculate their final QoS level and total slot number planed to be allocated to
them, and the base station schedules the slot resource and QoS based on these results.In
short, the core idea of QBS is that the network is constantly looking for the most effi-
cient resource allocation method, and finally allocates the slot resource based on the
cumulative results.

3 Simulation Results

In this part, we present the simulation assumptions and results, along with some brief analyses.
As described above, we adopt the downlink of CDMA2000 EV-DO Rev.B 3-Carrier system,
with 19 cells, 57 sectors included. The distance between adjacent base stations is 2500m,
and the users in the cells are uniformly distributed. The details of the simulation method are
illustrated in [23]. We compare the proposed QBS with two main scheduling algorithms, i.e.
Proportional Fair Scheduling (PFS) [24,25], and Round-Robin Scheduling (RRS) [26].
Let us illustrate the effect of the resource scheduler. It is assumed that the E&E fac-
tor, min(L i,u i,u i,u
QoS , L E I ), obeys the uniform distribution from level 1 to 10 and the users
are uniform distribution. The resource-constrained users are the users whose E&E level is
larger than 5 and SINR smaller than 0dB. Every sector contains 10 users with online music
service. As the RRS allocates the slots averagely, we only compare the PFS and the QBS.
The relation among users resource allocation, wireless conditions and E&E conditions is
shown in Fig. 4.
In Fig. 4, the x axis represents users E&E level. Higher value means better E&E con-
dition. The y axis represents the users Signal to Interference plus Noise Ratio (SINR). The

Fig. 4 The resource allocation results

123
606 B. Han et al.

Fig. 5 The satisfaction rate results

bigger the SINR is, the better the wireless condition will be. The z axis, slot ratio, represents
the percentage of the allocated time slots in the total time slots resource from the serving base
station. From the figure we can see that, when the PFS is applied, the time slots resource is
mainly allocated to users with better wireless conditions, irrelevant to the E&E condition.
However, when the QBS is applied, the time slot resources begin to move to the resource-con-
strained users defined above, in order to obtain the best resource allocation result. It indicates
that QBS makes the resource allocation incline to resource-constrained users.
To further investigate the effect of these two algorithms, the satisfaction rate results are
plotted in Fig. 5. The x axis represents the E&E level, y axis the SINR, and z axis the satis-
faction rate L i,actual i,max
QoE,u /L QoE,u defined by formula (6). Higher value means greater satisfaction
rate. It can be seen from this figure that the satisfaction rate of resource-constrained users
is very low when PFS is adopted. After the adoption of QBS, the experience of condition-
constrained users does not deteriorate much, while the general satisfaction rate of all users,
especially that of the Resource-Constrained (RC)users, improves to a great extent. It proves
the effectiveness of QBS on resource allocation. The reasons for this satisfaction enhancing
effect is that, when PFS is applied, the users with good SINR is assigned to large effective
resource to accumulate the data in the buffer of streaming media service as much as possible,
even if when the data is enough and unlikely to ran out in a short time. At the same time, the
users with low SINR is assigned to very little resource which could only support a low satis-
faction service. Unlike the PFS, when QBS is applied, the users with high SINR is assigned
fewer resource which could only maintain reasonable buffer data, while the saved resource
is assigned to the resource-constrained users to get a better effectiveness. To investigate the
performance of the three algorithms mentioned above in detail, Figs. 6 and 7 are plotted
below.
In Fig. 6, regarding the effect of online music service, we compared QBS, RRS and
PFS. The horizontal axis denotes the number of users in each sector, while the vertical axis

123
QoE Model Based Optimization 607

100

Satisfaction Rate Of Online Music Service


90

80

70

60

50

40 QBS Average
RRS Average
PFS Average
30 QBS RC Users
RRS RC Users
PFS RC Users
20
0 5 10 15 20 25
Number of UE per Sector
Fig. 6 The satisfaction rate results for online music service
Satisfaction Rate of Online Video Service (%)

100
QBS Average
RRS Average
90 PFS Average
QBS RC Users
RRS RC Users
80 PFS RC Users

70

60

50

40

30

20

10
1 2 3 4 5 6 7 8 9 10
Number of UE per Sector
Fig. 7 The satisfaction rate results of online video services

represents the satisfaction rate. The solid line represents the average satisfaction rate of all
users and the dash line represents the average satisfaction rate of resource-constrained users.
As for the average satisfaction rate shown in the figure, QBS improves the effect by 3 %
compared with RRS, and by 6 % compared with PFS. However, for resource-constrained
users, QBS improves at most by 15 % compared with RRS, and by 30 % compared with PFS.
Considering that the number of resource-constrained users is 21 % of the total users, it is a
significant increase.
There is a notable phenomenon in Fig. 6 that the RRS performs better than PFS. The
reason for this is not hard to understand, the PFS is a scheduling algorithm with the pursuit

123
608 B. Han et al.

80
Online Music
Online Video
70

60
Slot Ratio (%)

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10
Environment Level
Fig. 8 The environmental impact on slot resource allocation

of higher time slot resource utilization efficiency hence the resource is tend to be assigned
to the users with good SINR which can more efficiently use the resource instead of assigned
to resource-constrained users, this imbalance of resource allocation results is a damage on
the users with low SINR. On the contrary, QBS is a schedule algorithm with the pursuit of
fairness and the resource is tend to be assigned to the users equally, so it is not surprising that
the result of QBS is better than PFS.
In Fig. 7, we make a comparison among QBS, RRS and PFS regarding the effect on online
video service. The notation of this figure is similar to that of Fig. 6. As for the average satis-
faction rate shown in the figure, QBS improves the performance by 5 % compared with RRS
and PFS. However, for resource-constrained users, QBS improves at most by 10 % compared
with RRS, and by 20 % compared with PFS.
From Figs. 6 and 7, it is notable that RRS performs better than PFS in online music service
while in online video service these two schedule algorithms are similar. From Tables 3 and 4,
we can conclude the reasons. The needed resource used to improve QoS levels of online video
is much more than online music, there may even has a situation that when all the resource of
a base station is used, the online video service is still on the same level. Therefore, although
some users can obtain more resource, it is still hard to have a better experience as it needs
too much resource to improve QoS levels when they reach a certain level. As a result, the
efficiency of QBS remains the same as PFS in the aspect of online video service.
Figure 8 describes the needed time slots ratio for users at different environment level in
order to get fully satisfaction while the equipment level is fixed. For the purpose of meeting
the maximum satisfaction rate of each user, we only distribute 1 user in each sector. The
x axis represents the environment level of the users, and the y axis represents the time slot
ratio that has been described in Fig. 4, but here it is the needed slot ratio. From the figure we
could see that, with the fixed equipment degree, the slot ratio tend to be increasing while the
environment level is increasing. It is obvious that compared to online music service, online
video service takes use of much more time slots which could be easily understood according
to Tables 3 and 4 as the online video needs higher bit rate.

123
QoE Model Based Optimization 609

100
Online Music
Online Video
90

80

70
Slot Ratio (%)

60

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10
Equipment Level
Fig. 9 The equipment impact on slot resource allocation

Figure 9 describes the required time slots percentage for users with different equipment
level while the environment level is fixed. Making a comparison between Figs. 8 and 9, it
could be concluded that compared to the environment factor, the equipment factor also has
great influence on the needed network resource for the users. Furthermore, from Figs. 8 and
9, we could also see that, the utilized slots ratio decreases sharply when the environment
level or equipment level is lower. Specifically, utilized slots ratio reduces 1020 % when the
environment or equipment level lower by 2 levels on online video service. This means that we
have great potential to save network resource when the QBS and the QoE model considering
environment and equipment factors are used.

4 Conclusion and Future Work

According to our research, a scheduling algorithm called QoE Based Scheduling (QBS) is
proposed in order to make more efficient use of the network resource. To achieve this goal,
a QoE model considering user equipment and environment factors is proposed as well as a
new frame work of user equipment and the network. Simulation results show that the average
satisfaction is improved when QBS is used, especially for the resource-constrained users.
Furthermore, this algorithm has proved to have the potential to save network resource which
can, to a certain extent, alleviate the shortage of the spectrum.
In future work, more in-depth tests of users QoE in various scenarios will be conducted,
which involves the relationship between hardware parameters and user experience, between
environment interference and user experience, and between signal quality and network con-
dition. According to the result of this test, we will optimize our algorithm further.

Acknowledgments The work in this paper is sponsored by 111 Project of China under Grant No. B08004,
BUPT-QUALCOMM Joint Research Program and Fundamental Research Funds for the Central Universities.
The authors would like to thank the QoE testing group led by Yitong Liu for their QoE-related simulation and
verification of our research. The authors would also like to express their gratitude to Hao Wen and Liang Huo.
Finally, the authors would especially like to thank the reviewers of this article for their patience and painstaking
works. Without their consistent and illuminating work, this article could not have reached its present form.

123
610 B. Han et al.

References

1. TR-NGN. (2005). QoS-general aspects of quality of service and networkperformance in the next
generation networks: FGNGN-OD-00129[EB/OL], Geneva: ITUFGNGN [2006-02-04]. http://www.
Itu.int/ITU2T/ngn/fgngn/.
2. Piamrat, K., Ksentini, A., Bonnin, J.-M., & Viho, C. (2009). Q-DRAM: QoE based dynamic rate
adaptation mechanism for multicast in wireless networks. In: Proceedings of the IEEE global
telecommunications conference, pp. 16.
3. Wang, B., Wen, X., Yong, S., & Wei, Z. (2009). A new approach measuring users QoE in the
IPTV. In: Proceedings of the IEEE Pacific-Asia conference on circus, communications and systems,
pp. 453456.
4. Gao, F., Gao, Z., Wen, L., Zhang, B., Tang, D., & Hsu, D. (2009). A research on key performance
indicator of measurable QoE of EDGE network. In: Proceedings of the IEEE international conference
on information engineering and computer science, pp. 14.
5. Kim, H. J., Lee, D. H., Lee, J. M., Lee, K. H., Lyu, W., & Choi, S. G. (2008). The QoE evaluation
method through the QoS-QoE correlation model. In: Proceedings of the IEEE fourth international
conference on networked computing and advanced information management, pp. 719725
6. Kilkki, K. (2008). Quality of experience in communications ecosystem. Journal of Universal Computer
Science, 14(5), 615624.
7. Geerts, D., De Moor, K., Kety, I., Jacobs, A., Van den Bergh, J., Joseph, W., Martens, L.,
& Marez, L. (2010). Linking an integrated framework with appropriate methods for measuring QoE.
In: Proceedings of the IEEE second international workshop on quality of multimedia experience,
pp. 158163.
8. Du, Y., Zhou, W., Chen, B., & Song, J. (2009). A QoE based evaluation of service quality in wireless
communication network. In: Proceedings of the IEEE international conference on new trends in
information and service science, pp. 552557.
9. Volk, M., Sterle, J., Sedlar, U., & Kos A. (2010). An approach to modeling and control of QoE in
next generation networks. IEEE Communications Magazine, 48(8), 126135.
10. Venkataraman, M., Chatterjee, M., & Chattopadhyay, S. (2009). Evaluating quality of experience for
streaming video in real time. In: Proceedings of the IEEE global telecommunications conference,
pp. 16.
11. Gong, Y., Yang, F., Huang, L., & Su, S. (2009). Evaluating quality of experience for streaming
video in real time. In: Proceedings of the IEEE first international conference on emerging network
intelligence, pp. 2932.
12. De Pessemier, T., De Moor, K., Juan, A., Joseph, W., De Marez, L. & Martens L. (2011). Quanti-
fying QoE of mobile video consumption in a real-life setting drawing on objective and subjective
parameters. In: Proceedings of the IEEE international symposium on broadband multimedia systems
and broadcasting, pp. 16.
13. Brooks, P., & Hestnes, B. (2010). User measures of quality of experience: Why being objective and
quantitative is important. IEEE Network, 24(2), 813.
14. Istvn, D. M., Katrien, Ketykand, Wout, J., Luc, M., Lieven, D. M. (2010). Performing QoE-measure-
ments in an actual 3G network. In: Proceedings of the IEEE international symposium on broadband
multimedia systems and broadcasting, pp. 16.
15. Perkis, A., Munkeby, S., & Hillestad, O. I. (2006). A model for measuring quality of experience.
In: Proceedings of the IEEE 7th nordic signal processing symposium, pp. 198201.
16. Zhang, J., & Ansari, N. (2011). On assuring end-to-end QoE in next generation networks:challenges
and a possible solution. IEEE Communications Magazine, 49(7), 185191.
17. Menkovski, V., Exarchakos, G., Liotta, A., & Sanchez, A. C. (2010). Estimations and remedies
for quality of experience in multimedia streaming. In: Proceedings of the IEEE third international
conference on advances in human-oriented and personalized mechanisms, technologies and services,
pp. 1115.
18. Wu, W. W., Wu, M. U., & Wu, B. U. (2009). QoE/QoS improvement in wireless internet.
In: Proceedings of the IEEE first international conference on evolving internet, pp. 125132.
19. Soldani, D. (2006). Means and methods for collecting and analyzing QoE measurements in wireless
networks. In: Proceedings of the IEEE international symposium on world of wireless, mobile and
multimedia networks, pp. 531535.
20. Elkotob, M., Grandlund, D., Andersson, K., & Ahlund, C. (2010). Multimedia QoE optimized man-
agement using prediction and statistical learning. In: Proceedings of the IEEE 35th conference on
local computer networks, pp. 324327.

123
QoE Model Based Optimization 611

21. Reis, A. B., Chakareski, J., Kassler, A., & Sargento, S. (2010). uality of experience optimized
scheduling in multi-service wireless mesh networks. In: Proceedings of the IEEE 17th international
conference on image processing, pp. 32333236.
22. Amram, N., Fu, B., Kunzmann, G., Melia, T., Munaretto, D., Sabine Randriamasy, B. S., Widmer,
J., & Michele, Z. (2011). QoE-based transport optimization for video delivery over next generation
cellular networks. In: Proceedings of IEEE symposium on computer and communication, pp. 1924.
23. Gao, Y., Zhang, X., Yang, D., & Jiang, Y. (2009). Unified simulation evaluation for mobile broadband
technologies. IEEE Communications Magazine, 47(3), 142149.
24. Bender, P., Black, P., Grob, M., Padovani, R., Sindhushayana, N., & Viterbi, A. (2000). CDMA/HDR:
A bandwidth effcient high speed data service for nomadic users. IEEE Communication Maga-
zine, 38(7), 7078.
25. Jalali, A., Padovani, R., Pankaj, R. (2000). Data throughput of CDMA-HDR: A high effciency-high
data rate personal communication wireless system. In: Proceedings of IEEE vehicular technology
conference, pp. 18541858.
26. Shreedhar, M., & Varghese, G. (1996). Efficient fair queueing using deficit round-robin. IEEE/ACM
Transactions on Networking, 4(3), 375385.

Author Biographies

Bingjun Han received a B.Eng. degree in communication engineer-


ing from Beijing University of Posts and Telecommunications (BUPT),
Beijing, China, in 2008. Since 2008, he is working as a Ph.D. student
in Wireless Theory and Technology Lab, BUPT, Beijing. His research
interests include CDMA systems, LTE systems, quality of experience,
and quality of service in comexistence scenario among different com-
munication systems, qualiy of service and resource allocation method
in high-speed railway scenario.

Xin Zhang received his B.Eng. (1997) degree in communications engi-


neering, M.Eng. (2000) degree in signal & information processing
and Ph.D. (2003) degree in communications & information systems
from Beijing University of Posts and Telecommunications (BUPT).
He joined BUPT in 2003, now working in Wireless Theories and Tech-
nologies Lab as an associate professor, and focuses the research mainly
on key technologies and performance analysis of air interface of wire-
less networks.

123
612 B. Han et al.

Yifei Qi received his B.Eng. degree in Beijing University of Posts and


Telecommunications, Beijing, China, in 2010. Currently he is study-
ing at WT & T lab of BUPT for a M.S. degree. His research interests
include CDMA2000 and 3GPP LTE systems, and interference manage-
ment based on QoS parameters in cellular networks.

Yuehong Gao born in 1981, received her B.Eng. degree and D.Eng.
degree from Beijing University of Posts and Telecommunications
(BUPT) in year 2004 and year 2010, respectively. She studied at
Norwegian University of Science and Technology (NTNU) as a joint
phd student during August 2008 to July 2010. She is currently a lec-
turer at BUPT. Her research interests include wireless communication
technologies and application of stochastic network calculus in wireless
systems, particularly in cognitive radio networks.

Dacheng Yang received his M.S. (1982) and Ph.D. (1988) degrees in
circuits and systems from Beijing University of Posts and Telecommu-
nications (BUPT). From 1992 through 1993 he worked at the Univer-
sity of Bristol in the United Kingdom as a senior visiting scholar, where
he was engaged in Project Link-CDMA of the RACE program. In 1993
he returned to BUPT as an associate professor. Currently he is a pro-
fessor at BUPT, working in Wireless Theories and Technologies Lab,
and his research interests are focused on wireless communications.

123

Você também pode gostar