下一代行動網路多媒體資訊服務的服務品質保證與資源分配策略之研究

行政院國家科學委員會專題研究計畫 成果報告

下一代行動網路多媒體資訊服務的服務品質保證與資源分

配策略之研究(第 3 年)

研究成果報告(完整版)

計 畫 類 別 ： 個別型

計 畫 編 號 ： NSC 97-2410-H-004-116-MY3

執 行 期 間 ： 99 年 08 月 01 日至 100 年 07 月 31 日

執 行 單 位 ： 國立政治大學企業管理學系

計 畫 主 持 人 ： 郭更生

公 開 資 訊 ： 本計畫可公開查詢

中 華 民 國 100 年 12 月 19 日

中 文 摘 要 ： 多媒體資訊服務是下一代行動網路的關鍵主題，服務品質保

證是下一代行動網路多媒體資訊服務的核心問題。本研究計

畫為期三年，主要研究下一代行動網路中多媒體資訊服務之

服務品質保證機制，設計頻寬資源分配的最佳化機制以達成

服務品質保證。第二年研究主題為對 IEEE802.16e 的媒體接

入控制層，針對移動終端在切換過程中的多媒體資訊服務的

服務品質保證，提出動態資源預留和分配的最佳機制。接

著，考慮各種無線網路整合時，對網路層不同網路之間的切

換問題，提出動態資源預留和分配的最佳機制，以達到多媒

體資訊服務的服務品質保證。

中文關鍵詞： 多媒體資訊服務、服務品質保證、行動網路、IEEE

802.16e、Mesh 網路、下一代行動網路。

英 文 摘 要 ： The multimedia information service is the critical

topic of next-generation mobile networks； the QoS

guarantee is the central issue of the multimedia

information service in next-generation mobile

networks. This is a three-year research project. The

main purpose of this project is to study the

mechanisms on QoS guarantee of the multimedia

information services in next-generation mobile

networks, and to design the optimal mechanisms on

bandwidth resource allocation for achieving QoS

guarantee. In the second year, the research topic is

to first propose the optimal mechanism on dynamic

resource reservation

and allocation for achieving the QoS guarantee of

multimedia information services on the mobile

subscriber stations during the handoff process in

IEEE 802.16e networks. Then, it is to propose the

optimal mechanism on dynamic resource reservation and

allocation for the handoff on Network Layer of the

integrated wireless networks.

英文關鍵詞： Multimedia information services, QoS guarantee,

Mobile networks, IEEE 802.16e, Mesh Networks.

1

行 政 院 國 家 科 學 委 員 會 專 題 研 究 計 畫 成 果 報 告

下一代行動網路多媒體資訊服務的服務品質保證與資源分配策略之研

究(3 年期)

Research on the QoS Guarantee and Resource Allocation Strategy of

Multimedia Information Services in Next-Generation Mobile

Networks

計畫編號：NSC 97-2410-H-004-116-MY3

全程執行期限：97 年 8 月 1 日至 100 年 7 月 31 日

主持人：郭更生 執行機構及單位名稱：國立政治大學

一、中文摘要

多媒體資訊服務是下一代行動網路的

關鍵主題，服務品質保證是下一代行動網

路多媒體資訊服務的核心問題。本研究計

畫為期三年，主要研究下一代行動網路中

多媒體資訊服務之服務品質保證機制，設

計頻寬資源分配的最佳化機制以達成服務

品 質 保 證 。 第 二 年 研 究 主 題 為 對 IEEE

802.16e 的媒體接入控制層，針對移動終端

在切換過程中的多媒體資訊服務的服務品

質保證，提出動態資源預留和分配的最佳

機制。接著，考慮各種無線網路整合時，

對網路層不同網路之間的切換問題，提出

動態資源預留和分配的最佳機制，以達到

多媒體資訊服務的服務品質保證。

關鍵詞：多媒體資訊服務、服務品質保證、

行動網路、IEEE 802.16e、Mesh 網路、下

一代行動網路。

Abstract

The multimedia information service is

the critical topic of next-generation mobile

networks; the QoS guarantee is the central

issue of the multimedia information service

in next-generation mobile networks. This is

a three-year research project. The main

purpose of this project is to study the

mechanisms on QoS guarantee of the

multimedia information services in

next-generation mobile networks, and to

design the optimal mechanisms on bandwidth

resource allocation for achieving QoS

guarantee. In the second year, the research

topic is to first propose the optimal

mechanism on dynamic resource reservation

and allocation for achieving the QoS

guarantee of multimedia information services

on the mobile subscriber stations during the

handoff process in IEEE 802.16e networks.

Then, it is to propose the optimal mechanism

on dynamic resource reservation and

allocation for the handoff on Network Layer

of the integrated wireless networks.

Keywords: Multimedia information services,

QoS guarantee, Mobile networks, IEEE

802.16e, Mesh Networks.

二、研究成果

經由三年的研究工作，我們得到了具

體結果，達到預期成果。現已將其撰寫成

三篇英文論文投稿國際如下。

■已發表

Ming-Xin Hu and Geng-Sheng (G.S.)

Kuo, “Cognitive Radio-Enabled

QoS-Enhanced Scheme for IEEE 802.16e

Networks,” IEEE International Workshop on

Cognitive Radio (IWCR 2010), in Bangalore,

India, on Dec. 20, 2010。

■已被國際期刊接受發表

Lusheng Wang and Geng-Sheng (G.S.)

Kuo, “Mathematical Modeling for Network

Selection in Heterogeneous Wireless

Networks – A Tutorial,” accepted by IEEE

Communications Surveys & Tutorials (the

accept letter is attached to the paper).此國際

期刊的 impact factor 為 3.692，其 5-year

2

考。

■已投稿，現正在審查中

Yahui Hu and Geng-Sheng (G.S.) Kuo,

“Space-frequency Subchannel Allocation and

Adaptive Modulation in SDMA MIMO

OFDM Beamforming Systems with Limited

Feedback,” submitted to IEEE Transactions

on Wireless Communications.

另外，我們現已撰寫完成第四篇英文

論文初稿，準備投稿國際傑出期刊—IEEE

Transactions on Communications。

■準備投稿國際期刊

Guo-Mei Zhu, Geng-Sheng (G.S) Kuo

and Thomas M. Chen, “Convex Quadratic

Optimization Approach for Integrating

Distributed Intrusion Detection in Wireless

Mesh Networks,” submitted to IEEE

Transactions on Communications.

又我們現正撰寫第五和第六篇英文論

文，尚未完成，還須一些時日。

三、結論

本研究計畫全部的研究成果已有具體

結果，就個人自己評估，已達到計畫申請

書的整體預期水準。非常希望能在現有基

礎上繼續研究工作。

"Geng-Sheng Kuo" <[email protected]>

Subject:

[IWCR-10] Your paper 'Cognitive Radio-Enabled QoS-Enhanced Scheme for

IEEE 802.16e Networks'

From:

IWCR-10

To:

Ming-Xin Hu

Cc:

Geng-Sheng (G. S.) Kuo, [email protected]

Date:

10/29/10 05:59 PM

Dear Mr. Ming-Xin Hu:

On behalf of the Technical Program Committee, we are pleased to inform that your paper

# 1569371136 ('Cognitive Radio-Enabled QoS-Enhanced Scheme for IEEE 802.16e Networks')

for IWCR-10 has been accepted for presentation at the International Workshop on Cognitive Radio

in Bangalore, India, 9-11 December 2009.

Publication and presentation of this paper at the conference is contingent upon:

1) Your completion and submission of the camera-ready final version of the technical manuscript for the paper. Camera-ready copy must be submitted according to the instructions provided at

http://www.imsaa.org/callforpapers.html. The

deadline for submission of the final manuscript in PDF format,

after incorporating reviewers' comments if any, is 30 October, 2009. No extension will be given.

All final manuscripts of paper should be submitted only through EDAS portal. Before submitting through EDAS portal of above final paper, it should be IEEE Xplore compliant and verified through IEEE PDF eXpress. Guidelines for PDF eXpress:

i. Please go to the IEEE PDF eXpress at: http://www.pdf-express.org/ ii. Click on the link “New Users – Click Here”

and fill in your information. You will need the Conference ID which is: imsaa09x

2) Receipt of the signed IEEE copyright form through EDAS

www.edas.info under your IMSAA09 Paper(s).

3) Your pre-registration for the conference

http://www.imsaa.org/register.html

The deadline for author pre-registration is 30 October, 2009. For inclusion of the paper in the conference proceedings, every accepted paper must be accompanied by at least one full registration (Not a student registration). Where an author has more than one accepted papers, he/she has to register for each paper separately.

4) The attendance and presentation at the conference of the paper by at least one of the authors of the papers. We reserve the right to withdraw this acceptance for any papers which are not presented in person at the conference.

5) The final manuscript follows IEEE Communications Society guidelines about plagiarism and duplicate submission.

Full details can be found on the conference web site at

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

http://www.imsaa.org/register.html

Reviews of your paper can be found on the EDAS site at the URL at the top of this email under "Reviews", and are also listed below in this email. Please take the reviewers comments into account in preparing your final manuscript.

Thank you for submitting your paper to IMSAA-2009, and we look forward to seeing you in Bangalore at the conference.

Thanks and regards,

John Buford, Vijay Gurbani, Anand Prasad IMSAA 2009 Technical Program Committee Chairs ================================================ ======= Review 1 =======

> *** Summary: Please provide a summary of the paper (the contents and contribution of the paper). (1-3 sentences)

The paper proposes a QoS-enhanced scheme to prevent real-time services in 802.16e from corruption when switching between licensed and unlicensed bands. It also proposes adaptive bandwidth reservation mechanism and simplified switch mechanism to improve QoS performance for 802.16e services.

> *** Strengths: What are the most important reasons to accept this paper

(e.g., technical depth, novelty, creative solution, importance of topic, etc.)? Paper is well written and easy to follow.

> *** Weaknesses: What are the most important reasons NOT to accept this paper? Could address few comments given below to improve the quality of the manuscript. > *** Comments to authors: Please provide detailed comments to the authors. o Authors should explain and justify the simplification they have proposed in sub-section C (page 3).

o Section III and sub-section A have same heading. It would be better to revise this.

o In Section V, paragraph2, line 3: the reference to the figures has a Chinese font, and should be changed.

> *** Relevance: Please rate the relevance of the paper to the main topics of IMSAA08

Moderate importance (3)

> *** Familiarity: Please assess your familiarity with the subject matter of the paper.

Familiar with this area of research (2)

> *** Presentation: Please assess the presentation style of the paper Easy to understand (4)

> *** Recommendation: Please provide your overall recommendation on the

acceptance of the paper. (Final acceptance decisions will also consider literal responses to the questions below.)

Likely accept (4)

======= Review 2 =======

> *** Summary: Please provide a summary of the paper (the contents and contribution of the paper). (1-3 sentences)

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

The paper presents a way to use cognitive techniquesin current technology, and is therefore very relevant.

> *** Strengths: What are the most important reasons to accept this paper

(e.g., technical depth, novelty, creative solution, importance of topic, etc.)? The concept presented is itself not new. However, the simulation results are of interest to the communications community.

> *** Weaknesses: What are the most important reasons NOT to accept this paper? The paper is very poorly presented. Grammar is very poor.

The analysis is very poorly presented, and difficult to understand. The authors should rewrite the entire manuscript to improve readability. Some text appears in chinese..

> *** Comments to authors: Please provide detailed comments to the authors. 1.Rewrite entire paper to improve grammar and readability.

2.Analysis is very vague,and this section should be very carefully rewritten. 3. The analysis is not verified in any meaningful manner in the simulations. > *** Relevance: Please rate the relevance of the paper to the main topics of IMSAA08

Excellent--right on target (5)

> *** Familiarity: Please assess your familiarity with the subject matter of the paper.

Expert and working in this area of research (3)

> *** Presentation: Please assess the presentation style of the paper Hard to understand (1)

> *** Recommendation: Please provide your overall recommendation on the

acceptance of the paper. (Final acceptance decisions will also consider literal responses to the questions below.)

Accept if room (3)

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

Abstract—The IEEE 802.16e network faces the challenge of bandwidth shortage due to the growing demand for high-bandwidth applications. One promising approach is to obtain bandwidth by using cognitive radio (CR) technology, which provides an effective way to achieve incremental bandwidth from unlicensed spectrum. However, the appearance of primary users (PUs) does harm to the performance of secondary users (SUs). In this paper, we propose a QoS-enhanced scheme to prevent real-time services in 802.16e networks from corruption when switching between licensed and unlicensed bands. Adaptive bandwidth reservation mechanism and simplified switch mechanism are proposed to improve QoS performance for 802.16e services. Finally, simulation results show that our scheme has lower connection dropping probability (CDP) at a reasonable cost of connection blocking probability (CBP) than conventional scheme.

Keywords—Cognitive radio, IEEE 802.16e, QoS.

I. INTRODUCTION

IEEE 802.16e network provides last-mile broadband wireless access in metropolitan area, which supports mobile users with better services. Its bandwidth can reach 75Mbps at most [1]. However, it is still insufficient to cope with increasing bandwidth demands brought by multimedia applications, such as VOD, VoIP, etc. Sometimes, bandwidth shortage happens in special scenarios. For instance, when a mobile station (MS) is moving far from base station (BS), the channel between them is becoming worse. In this situation, MS and BS may decrease modulation efficiency to maintain a high reliability of current connections. As a result, channel capacity will suffer a heavy loss. Some applications may be corrupted due to bandwidth shortage. Bad weather (rains or snows) may also result in such loss of network capacity. In certain area, such as railway station or airport, the burst of passenger arrival often brings great pressure to bandwidth allocation. All these can lead to performance degradation of IEEE 802.16e network.

One promising approach is to increase the number of BSs or relay stations (RSs) in above scenarios. But the handover procedure brings extra latency and may cause a high connection dropping probability (CDP) for real-time connections. And, it is not economic to cost much on deployment, especially when the bandwidth pressure just exists in a short period of time. Cognitive radio (CR) provides a better solution, in which BS and MS obtain additional bandwidth from other unused spectrum with less cost. And, it

is feasible since most of the allocated spectrum has low utilization rate in real conditions [2] [3].

CR was first proposed in [4]. It provides high efficiency of spectrum utilization. Non-licensed users are prohibited to access licensed spectrum even though incumbents do not use. The policy of fix spectrum allocation does well in the past. But in recent years, the demands to spectrum increase greatly due to more wireless applications. It is necessary to utilize spectrum in an effective way. CR technology can permit non-licensed users to utilize the licensed spectrum without interfering with incumbents.

SS1 MS1 BS SS2 MS2 MS3 MS4 Licensed channel Unlicensed channel Fig. 1. The IEEE 802.16 network with CR-enabled

enhancement.

IEEE 802.22 standard defines that license-exempt devices can utilize the spectrum allocated to the TV broadcast services by using CR technology with interference avoidance [5]. However, IEEE 802.22 network has difficulty in providing QoS guarantees for real-time services because license-exempt devices should vacate the occupied bandwidth when incumbents appear. IEEE 802.16h also uses CR technology to coexist with incumbents in unlicensed band [6]. It can not promise better QoS parameters for real-time services, either. A special control channel was proposed in [7] and [8] to exchange control information between CR nodes, such as synchronization and negotiation. Our proposed scheme is based on centralized scheduling. Each node is first to access the licensed spectrum and exchange control information with BS. Also, there are many efforts focused on how to improve network capacity [9] [10], and QoS performance of real-time services is also an important issue. [11] proposed an adaptive MAC protocol to enhance throughput in CR networks. When channel status parameter is higher than specific threshold, both licensed and unlicensed channels work on concurrent transmission mode to improve the throughput. Otherwise, the two channels work on error recovery mode to increase channel reliability. However, it needs to work on two RF interfaces simultaneously, which adds the system cost. [12] provided a

Cognitive Radio-Enabled QoS-Enhanced

Scheme for IEEE 802.16e Networks

Ming-Xin Hu and Geng-Sheng (G.S.) Kuo

Beijing University of Posts and Telecommunications, Beijing, China

National Chengchi University, Taipei, Taiwan

can perform a faster and more reliable switch procedure based on licensed channel in the presence of incumbents for IEEE 802.16e network.

The rest of this paper is organized as follows. The background of CR is overviewed in Section II. Section III proposes our scheme for IEEE 802.16e networks. In Section IV, performance analyses are conducted. And simulation results are given in Section V. Finally, conclusions are made.

II. RELATEDWORK

There are two types of users in CR network. One type is primary user (PU), which has the license to operate in a certain spectrum. The other is secondary user (SU), which shares unlicensed spectrum with PU. PU accesses its licensed spectrum freely. But SU should access unlicensed channel with guarantee of no harmful interference to PU. Once PU is detected, SUs must vacate the occupied channel immediately. Therefore, it is important to find a quick and efficient way to detect PUs. Two-stage sensing (TSS) mechanism was provided for SUs to detect PU effectively and quickly. It includes both fast sensing stage and fine sensing stage. A fast sensing stage may include several fast sensing periods. Fast sensing algorithm executes very quickly, usually less than 1ms/channel. A fine sensing stage only includes one fine sensing period, which executes detailed sensing on target channel, such as feature detection, to decrease false alarm probability. So it often needs time in millisecond, e.g., it may cost 69.43ms in 6MHz channel [13]. Fine sensing may be cancelled by BS according to the results of fast sensing. All the sensing operations should be performed in quiet period to avoid interference from neighbor networks. Hence, each BS shall communicate with neighbor BSs to coordinate quiet periods for sensing actions. The mechanism is shown in Fig. 2.

Fig. 2. The two-stage sensing mechanism.

SUs face two choices when PU is detected. One is that SUs switch to other unlicensed band immediately; the other is that SUs still work on current band and do not have any harmful interference with PUs by changing transmission power and modulation means. [14] provided a dynamic spectrum access (DSA) approach to alleviate the spectrum scarcity. In the approach, the PUs have priority in accessing channel. However, both choices should degrade the QoS performance of SUs.

III. PROPOSEDSCHEME

We consider an IEEE 802.16e network with a CR BS (CRBS) and at least one CR MSs (CRMSs). Both CRBS and CRMS have cognitive capacity and reconfigurability. CRBS is equipped with multi-RF interfaces and has the ability

CRMS has only one RF interface and access one channel at a time. So, it costs little on CRMS.

A. Description of Proposed Scheme

1) Initialization and normal operations: When network starts up or available bandwidth provided by licensed band is less than predefined threshold, CRBS and CRMSs work in licensed band just as they do in the legacy IEEE 802.16e network. Here the predefined threshold will be discussed later. Additionally, CRBS coordinates CRMSs to detect the activity of unlicensed spectrum in spare time and maintains a list of unused channels.

2) Switch to unlicensed channels: As on-going connections increase, more and more bandwidths are required. Once available bandwidth provided by licensed band is less than predefined threshold, CRBS will select several available channels from unused channel list and indicate some CRMSs or new CRMSs to perform switch procedure to those channels. Corresponding CRMSs execute sensing procedure after receiving switch indication. If no PU is detected, CRMSs begin to perform switch procedure. However, a CRMS can refuse switch instruction according to its actual situation or user decisions. After switch process, connections still use the same connection IDs in the new channels as before, which helps CRMS to switch fast between licensed and unlicensed channels. CRBS also reserves a certain amount of bandwidth in licensed channel for real-time connections to greatly reduce the dropping rate of real-time connections in unlicensed channels when PU is detected.

We define the ratio of total bandwidth, which has been

allocated to users, to total available bandwidth in licensed channel as γ. Let γr_th be the predefined

threshold. As the demand for bandwidth rises, the value of γ also increases. Once γ ≥γr_th, switch procedure will

be carried out until γ reduces to the value below γr_th.

Here, the derivation of γr_thwill be discussed in part B.

Existing real-time connections should not be corrupted

by high switch latency. So the selection of CRMSs should meet the request of

h_UL m h_L m t t t t ≤ ≤ 1()

here, th_ULdenotes the total switch latency from licensed

band to unlicensed band and th_Ldenotes the total switch

latency from unlicensed band to licensed band. The two parameters are discussed in part IV. tmis the maximum

switch latency that real-time connections can be tolerant. 3) Switch to licensed channel: If a CRMS detects PUs in fast sensing stage, it immediately sends a report to CRBS in next superframe. CRBS will make a decision on whether to carry out switch procedure according to the reports received from all CRMSs. If switch is needed, CRBS broadcasts switch message and reserves bandwidth in licensed channel for influenced real-time connections. If CRBS does not receive any report or data unit from CRMSs after a sensing

channel for related CRMSs. On the other hand, if there is no indication from CRBS after a sensing period, CRMS will consider PUs appear in current channel, and switch to its licensed channel immediately with stored parameters and use allocated bandwidth to send detailed report and bandwidth request for its connections. If no bandwidth is allocated, it uses a special reserved contention bandwidth to send bandwidth request message. Since fast sensing can be executed very fast, it shall do little harm to the QoS performance of real-time connections. After switch procedure, out-band fine sensing is executed to detect incumbent services in original unlicensed channel in spare time. Here, the efficiency of switch procedure is influenced by error rate of fast sensing. A high error rate may cause unnecessary switch operations and degrade the performance of real-time connections. In the actual conditions, we can get a very low error rate by threshold control [15].

B. Threshold of Bandwidth Reservation

We model the appearance of PUs in a channel as ON-OFF period. And, we assume that the ON/OFF periods of each channel are independent. Referring to [16], the probability that PUs appear in channel i, i.e., the utilization of channel i, can be obtained: ( ) ( ) ( ) ( ) [ ] , 1, 2,..., [ ] [ ] ON i pr i ON i OFF i E T p i N E T E T = = + 2()

here, TON(i) and TOFF(i) are the durations of ON and OFF

periods, respectively. And, N is the total number of unlicensed channels which have been used.

Then, the probability that PUs appear in k channels simultaneously can be derived as:

(

1

)

1 2 1 1 ( ) ( ) ( ) 1 1 1 , 1, 2,..., k k N N N o k pr i pr i i i i i i k p p p k N = = + = + − = =

∑ ∑

L

∑

L 3()

The reserved bandwidth in licensed channel can be obtained from the following derivation:

( ) _ 1 ( ) , 1 ,1 ( ) { , 1, 2,..., } res N res CR min i i res o k BW BW n N max k others n P k N φ χ = ≥ =  =  ∈ > =

∑

4()

here, Nres is the number of unlicensed channel for which

licensed channel has reserved bandwidth, and BWres is the

reserved bandwidth in licensed channel. BWCR(min)_i is the

minimum bandwidth required by real-time connections in unlicensed channel i and we assume that BWCR(min)_1 ≥

BWCR(min)_2 ≥ …≥ BWCR(min)_N. χ is used to estimate the

amount of bandwidth reserved for unlicensed channels. Small χ wastes bandwidth reservation, which leads to a low utilization of licensed channel capacity and a high connection blocking probability (CBP). On the other hand, large χ leads to high CDP for the real-time connections in unlicensed

requirements for real-time services in actual deployment. At last, we derive the value of γr_thfrom (5)

_ _ 1 res r th L t BW BW γ = − 5()

where BWL_t is the total available bandwidth provided by

licensed channel.

C. Fast Switch Procedure

Normal handover procedure in IEEE 802.16e networks often includes the stages shown in Fig. 3.

In our proposed scheme, CRMS does not perform handover to a new CRBS. It just switches from one band to another and communicates with the same CRBS all the time. Many stages can be skipped since the CRBS knows most of the required information of CRMSs. If necessary, there is a negotiation between CRBS and CRMS before switch to exchange information. So we can neglect almost all stages except ranging in the switch procedure. And we can even reduce ranging time by assigning an unsolicited ranging opportunity through negotiation. At last, only a non-contention-based ranging stage is needed in our fast switch procedure.

Fig. 3. Handover procedure in IEEE 802.16e networks.

IV. PERFORMANCEANALYSIS

A. CDP

Although we have reserved certain amount of bandwidth in licensed channel for real-time connections in our scheme, some of existing connections will be dropped when PUs are detected in too many non-licensed channels.

Let BWtbe the total bandwidth of real-time connections in

the network and BWCR(i) be the bandwidth of real-time

(

(

)

1 1 2 1 1 1 ( ) 1 1 1 ( ) ( ) _ _ _ 1 1 max 0, res k k k pr i k k N N i i i i i k i CR j res j i pr i t N t r th L t CR i i CDP p C BW BW p BW BW γ BW BW = = = + = + − = = =     −          = +

∑

∑ ∑

∑

∑

∑

L L 6()

It shows the value of CDP is affected by ppr(i)and Nresfrom

above. As ppr(i) increases, CDP becomes higher and our

scheme adjusts the value of Nresto keep CDP at a low value.

If PUs appear very frequently, Nresis set to N and we reserve

the bandwidth for all unlicensed channels. CDP is decreased to 0. However, it also means we get no benefit from unlicensed spectrum at the moment.

B. CBP

The maximum number of real-time connections provided by the network is given by

t ava rt BW N BW   =    7()

here BWrtis the average bandwidth of real-time connections.

Assume the arrival of real-time connections is Poisson process with arrival rate λrt. And, the corresponding lifetimes

follow exponential distributions with rate µrt. So, our scheme

can be modeled as a M/M/m(m) queuing system as shown in Fig. 4.

Fig. 4. State transition diagram of M/M/m(m) system.

0 1 1 1 1 0 ( ) ( 1) , 1, , 1 1 ava ava ava rt rt rt rt i rt i rt i ava rt N ava rt N N i i p p i p p i p i N p N p p λ µ λ µ λ µ λ µ − + − = = + = + + = − = =

∑

L 8()

Let CBP denote the blocking probabilities for real-time connections. We obtain:

ava

N

CBP= p 9()

C. Maximum Switch Latency

The maximum switch latency from licensed band to unlicensed band is composed of three parts: sensing time before switch, switch latency and fine sensing time. Before switch, CRBS and CRMSs should execute sensing on target channel to make sure no PU. In the worst case, CRMS can not

is a fine sensing stage being executed at that time.

CRBS can arrange CRMSs to sensing target unlicensed channels in advance to reduce latency. So we neglect sensing time in the switch process from licensed band to unlicensed band. Although we have a non-contention initial ranging opportunity, it takes at least a superframe duration on ranging for CRMS in unlicensed channel.

So, th _ULis given by

_ 1+

h UL rng fs

t =t t 10() here, trng1 and tfs are the ranging time of CRMS and fine

sensing time in unlicensed channel.

The maximum switch latency from unlicensed band to licensed band is composed of two parts: maximum time of detecting PUs and switch latency from unlicensed band to licensed band. The switch procedure from unlicensed band to licensed band can be simplified to ranging stage by using stored parameters. And, the CRMS also has a non-contention opportunity when re-ranging to licensed band.

Hence, th_Lis given by

_ 2 2

h L s rng

t =t +t 11() here, ts2is the maximum time required by CRMS to detect

PUs and trng2 is ranging time of CRMS when it switches to

licensed channel.

Then, we discuss the estimation of ts2. In the worst case,

PUs appear just after the end of a quiet period. The CRBS and CRMSs are not aware of the presence of PUs until next quiet period that should wait for a transmission time. During the waiting time, the CRBS and CRMSs may not continue normal transmission. At the next quiet period, fast sensing is executed and PUs are detected. CRMSs send report to the CRBS. And, CRBS indicates all related CRMSs to perform switch procedure to licensed band. We assume a fast sensing period follows every L-superframe transmission. ts2is derived

by 2 ( 1) s superframe fs superframe superframe fs t LT T T L T T = + + = + + 12()

here, Tsuperframedenotes a superframe duration in an unlicensed

channel. From (12), L determines the time interval between two fast sensing periods. The smaller L is, the faster we detect incumbent service, and the short switch latency we can get.

V. NUMERICALRESULTS

In this section, we illustrate some simulation results for the performance of our proposed scheme. The simulations are conducted by MATLAB. In the simulations, CRBS works in a licensed band and three unlicensed bands simultaneously. The bandwidth of licensed channel is 1.5Mbps while those of unlicensed channels are 0.4Mbps, 0.5Mbps and 0.6Mbps, respectively. The bandwidth requirement by a real-time service is [25kbps, 30kbps]. The call arrivals of PUs in unlicensed band and end users of 802.16e networks follow Poisson distribution. And, the call arrival rate of end users is 10. The call serving times of PUs and end users are also

0.1 seconds, respectively.

In the simulations, we compared our proposed scheme with conventional one, which has no bandwidth reservation mechanism. From Fig. 5 and Fig. 6, our proposed scheme obtains lower CDP than conventional one. As the arrival rate of PUs rises, the proposed scheme adjusts the reserved bandwidth adaptively to reduce CDP to a lower level. However, the value of CBP increases at the same time due to additional bandwidth reserved for the calls in unlicensed channels. When arrival rate of PUs exceeds 0.6, the proposed scheme reserves bandwidth for all the calls in unlicensed channels. Therefore, no call will be corrupted when PUs appear. However, CBP has the highest value and network capacity reduces to its lowest point at the same time.

0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

Arrival rate of primary users

C

B

P

Conventional Proposed

Fig. 5. CBP of end users by differentiating arrival rate of PUs.

0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

Arrival rate of primary users

C

D

P

Conventional Proposed

Fig. 6. CDP of end users by differentiating arrival rate of PUs.

VI. CONCLUSION

This paper proposes a new scheme to improve the system capacity of IEEE 802.16e network using CR technology in an easy and efficient way. A bandwidth reservation mechanism in licensed channel is used to prevent real-time services from

switch procedure to reduce switch latency as possible as we can. Maximum switch latency is estimated to help real-time services from corruption. Simulation results show that our scheme obtains a lower value of CDP than original IEEE 802.16e network while the CBP is still kept at a normal range.

REFERENCES

[1] IEEE Std 802.16eTM -2005, “IEEE Standard for Local and Metropolitan Area Networks – Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems - Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands,” Feb. 28, 2006.

[2] Staple G, Werbach K, “The end of spectrum scarcity,” IEEE Spectrum, pp.48-52,2004.

[3] Federal Communication Commission, “Facilitating Opportunities for Flexible, Efficient, and Reliable Spectrum Use Employing Cognitive Radio Technologies, NPRM & Order,” Federal Communication Commission, ET Docket no. 03-108, FCC 03-322, 2003.

[4] J. Mitola, “Cognitive Radio: An Integrated Agent Architecture for Software Defined Radio,” Ph.D. Dissertation, Royal Inst. Technol. (KTH), Stockholm, Sweden, 2000.

[5] www.ieee802.org/22/online document: PAR [6] http://wirelessman.org/le/online document: PAR

[7] Xiangpeng Jing and Raychaudhuri, D., “Global Control Plane Architecture for Cognitive Radio Networks,” Proc of IEEE International

Conference on Communications (ICC 2007), pp. 6466 - 6470, Jun. 2007.

[8] Hang Su and Xi Zhang, “Cross-Layer Based Opportunistic MAC Protocols for QoS Provisionings Over Cognitive Radio Wireless Networks,” IEEE Journal on Selected Areas in Communications, Vol. 26, pp: 118 – 129, Jan. 2008.

[9] Johansson.M. and Xiao.L., “Cross-layer optimization of wireless networks using nonlinear column generation,” IEEE Transcations on

Wireless Communications, vol. 5, pp. 435 - 445, Feb.2006.

[10] Juncheng Jia, Qian Zhang and Xuemin Shen, “HC-MAC: A Hardware-Constrained Cognitive MAC for Efficient Spectrum Management,” IEEE Journal on Selected Areas in Communications, Vol. 26, pp: 106 – 117, Jan. 2008.

[11] Byungjoo Lee and Seung Hyong Rhee, “Adaptive MAC Protocol for Throughput Enhancement in Cognitive Radio Networks,” Proc of

International Conference on Information Networking (ICOIN 2008),

Jan. 2008

[12] Chen, Jie, Zhao, Minjian and Tian, Jin, “A QoS MAC Protocol for Cognitive PMP Networks with Rapid Changes of Spectrum Opportunities,” Proc of Vehicular Technology Conference (VTC 2008), pp. 1661 – 1665, May 2008.

[13] Wendong Hu, Willkomm, D. and Abusubaih, M., “COGNITIVE RADIOS FOR DYNAMIC SPECTRUM ACCESS - Dynamic Frequency Hopping Communities for Efficient IEEE 802.22 Operation,”

IEEE Communications Magazine, vol. 45, no. 5, pp. 80 - 87, May 2007.

[14] Geirhofer, S., Lang Tong and Sadler, B.M., “COGNITIVE RADIOS FOR DYNAMIC SPECTRUM ACCESS - Dynamic Spectrum Access in the Time Domain: Modeling and Exploiting White Space,” IEEE Communications Magazine, vol. 45, no. 5, pp. 66 - 72, May 2007.

[15] Wei Zhang, Ranjan K. Mallik and Khaled Ben Letaief, “Cooperative Spectrum Sensing Optimization in Cognitive Radio Networks,” Proc of

IEEE International Conference on Communications (ICC 2008), May

2008.

[16] H. Kim and K. G. Shin, “Adaptive MAC-layer sensing of spectrum availability in cognitive radio networks,” Technical Report, CSE-TR-518-06, University of Michigan, 2006.

"Geng-Sheng Kuo" <[email protected]>

Subject: CST: Decision COMST-00044-2011.R1

From:

[email protected]

To:

[email protected], [email protected], [email protected]

Cc:

[email protected], [email protected]

Date:

12/04/11 10:12 PM

04-Dec-2011

Dear Prof. Geng-Sheng (G.S.) Kuo,

We are happy to inform you that your manuscript COMST-00044-2011.R1

Mathematical Modeling for Network Selection in Heterogeneous Wireless Networks – A Tutorial has been accepted for publication in the IEEE Communications

Surveys and Tutorials. Please find enclosed the reviewer comments.

Instructions for preparation of your publication package can be found at:

http://www.comsoc.org/livepubs/surveys/index.html and then clicking

on "Authorkit for accepted papers." Please contact Sue Lange at

[email protected] if you have any questions on manuscript preparation. Be

advised that although your article will not be published for quite some time, it can be posted early at the IEEE Xplore site. The sooner you get your

materials in to the publications department, the sooner it can be posted there. Please, return

the material for camera ready production at most 1 (one) month after the reception of this notification.

Note: all files should be placed into a folder which should then be compressed using zip, stuffit, tar, or other compression formatting scheme and then

uploaded to the publications ftp site: ftp.comsoc.org, directory: 2011; user id: commpaper; password: 3Em=!!a (case sensitive). Please use an ftp client to upload the files. If you must use Explorer, use the website address

ftp://ftp.comsoc.org/commpaper. If you are unable to ftp your files, contact

Sue Lange for a mailing address where they can be sent. Once you have uploaded the files please email Sue Lange to let her know the name of the file.

I take this opportunity to thank you for submitting your work for consideration to IEEE Communications Surveys and Tutorials. I look forward to seeing your work published in the near future.

Sincerely yours,

Nelson L. S. da Fonseca

IEEE Commun. Surveys & Tut., EiC http://www.comsoc.org/pubs/surveys University of Campinas Institute of Computing [email protected] [email protected] http://www.ic.unicamp.br/~nfonseca --- Reviewer Comments: Reviewer: 1 Recommendation: Accept Comments:

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

The authors have addressed my comments. I am particularly pleased with the significantly improved tutorial value of the paper as well as the significantly increased citations.

Additional Questions:

1. Please provide a one-paragraph description of the content of this

manuscript.: The authors correctly recognize that current and future terminals will be equipped with different wireless technologies. The authors hence

focused in this paper on studying a set of mathematical theories used for modeling the network selection problem in the open literature. It includes GT, NN, and other optimization techniques. As said before, I deem the topic of increasing importance and the tutorial paper timely.

2. Please identify and discuss the contribution of this manuscript. Please include in your discussion items such as the following:<br>

a. Does the paper have significant tutorial content? That is, is there enough background provided so that the generalist in communications can understand its main contributions? Elaborate.<br>

b. Does the paper contain original contributions? What is the nature of the contributions?<br>

c. Is there a description of lessons learned that are given to the reader to help the reader avoid pitfalls in his own work?<br>

d. Is there a need for a paper such as this in the communications community? For example, are there articles that are already available which cover more or less the same topic at about the same depth?: Since this is a re-submission, please, see final summary below.

3. Please discuss the quality of the citations in this manuscript. If you think the citations should be improved, please provide specific references or sources of articles, such as journals or magazines, that should be consulted. : Since this is a re-submission, please, see final summary below.

4. Please comment on the organization of the paper, and offer any suggestions that you think will improve the paper and its readability.: Since this is a re-submission, please, see final summary below.

5. Please comment on the technical correctness of the manuscript in general, identify any specific technical inaccuracies that you find, and make

suggestions for correcting those.: Since this is a re-submission, please, see final summary below.

6. If the manuscript does not require major revision, please provide a list of minor changes, such as spelling or grammatical errors, that need to be made. Please use the format

'p. 7., l. 18 somth ==> smooth' to mean 'on line18 of page 7, correct the spelling from somth to smooth.': Since this is a re-submission, please, see final summary below.

7. Please provide a summary comment on the overall suitability of the paper for publication in IEEE Communications Surveys and Tutorials, assuming the

recommended revisions are made. For example, if this is an outstanding

contribution, please so state. If a major revision is needed, please so state. If the manuscript requires major editing, please so state.: Since this is a re-submission, please, see final summary below.

Reviewer: 2

Recommendation: Accept Comments:

The authors correctly addressed the questions and suggestions I raised during last review round. I believe the paper is now suitable for publication.

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

Additional Questions:

1. Please provide a one-paragraph description of the content of this

manuscript.: The paper proposes a tutorial on mathematical models for network selection.

2. Please identify and discuss the contribution of this manuscript. Please include in your discussion items such as the following:<br>

a. Does the paper have significant tutorial content? That is, is there enough background provided so that the generalist in communications can understand its main contributions? Elaborate.<br>

b. Does the paper contain original contributions? What is the nature of the contributions?<br>

c. Is there a description of lessons learned that are given to the reader to help the reader avoid pitfalls in his own work?<br>

d. Is there a need for a paper such as this in the communications community? For example, are there articles that are already available which cover more or less the same topic at about the same depth?: I believe the authors greatly improved the paper, which now provides useful insights to the problem and enough technical depth to enable the reader to understand the different available solutions for network selection. In its current version, the paper is useful for the communications community.

3. Please discuss the quality of the citations in this manuscript. If you think the citations should be improved, please provide specific references or sources of articles, such as journals or magazines, that should be consulted. :

Citations are ok.

4. Please comment on the organization of the paper, and offer any suggestions that you think will improve the paper and its readability.: The organization of the paper is now acceptable.

5. Please comment on the technical correctness of the manuscript in general, identify any specific technical inaccuracies that you find, and make

suggestions for correcting those.: In general, the manuscript is correct from the technical viewpoint.

6. If the manuscript does not require major revision, please provide a list of minor changes, such as spelling or grammatical errors, that need to be made. Please use the format

'p. 7., l. 18 somth ==> smooth' to mean 'on line18 of page 7, correct the spelling from somth to smooth.': No specific requests.

7. Please provide a summary comment on the overall suitability of the paper for publication in IEEE Communications Surveys and Tutorials, assuming the

recommended revisions are made. For example, if this is an outstanding

contribution, please so state. If a major revision is needed, please so state. If the manuscript requires major editing, please so state.: I believe the paper is now suitable for publication on IEEE Communications Surveys and Tutorials.

Reviewer: 3

Recommendation: Accept Comments:

The manuscript is well written and typeset and I find that you have indeed improved the crispness (and hence the usefulness) of the paper by cleaning up the logical flow and presenting better the progress from the more general to the more specific. The additional of the unifying case study has also helped. Additional Questions:

1. Please provide a one-paragraph description of the content of this

manuscript.: As I said in my original review, the manuscript does a summary of the mathematical models for the optimization of network selection in

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

heterogeneous wireless networks. It then proposes an “integrated solution”. 2. Please identify and discuss the contribution of this manuscript. Please include in your discussion items such as the following:<br>

a. Does the paper have significant tutorial content? That is, is there enough background provided so that the generalist in communications can understand its main contributions? Elaborate.<br>

b. Does the paper contain original contributions? What is the nature of the contributions?<br>

c. Is there a description of lessons learned that are given to the reader to help the reader avoid pitfalls in his own work?<br>

d. Is there a need for a paper such as this in the communications community? For example, are there articles that are already available which cover more or less the same topic at about the same depth?: The paper makes a very thorough survey of the mathematical modeling and optimization

methodologies applicable to the network selection problem. In this sense, it makes a very good survey, if not an excellent tutorial. The tutorial value has been increased after the revision, by working on the overall logical flow and the addition of specific *unifying* case study. There is originality in the fact that there does not appear to be another survey like this one.

3. Please discuss the quality of the citations in this manuscript. If you think the citations should be improved, please provide specific references or sources of articles, such as journals or magazines, that should be consulted. : There is a large number of citations but appropriate for the topic; the authors have done a very thorough job of the literature survey.

4. Please comment on the organization of the paper, and offer any suggestions that you think will improve the paper and its readability.: The organization of the paper and its logical flow had been more uneven in the original manuscript and have been improved in the revised paper.

5. Please comment on the technical correctness of the manuscript in general, identify any specific technical inaccuracies that you find, and make

suggestions for correcting those.: The manuscript appears technically correct. 6. If the manuscript does not require major revision, please provide a list of minor changes, such as spelling or grammatical errors, that need to be made. Please use the format

'p. 7., l. 18 somth ==> smooth' to mean 'on line18 of page 7, correct the spelling from somth to smooth.': Nothing of real consequence.

7. Please provide a summary comment on the overall suitability of the paper for publication in IEEE Communications Surveys and Tutorials, assuming the

recommended revisions are made. For example, if this is an outstanding

contribution, please so state. If a major revision is needed, please so state. If the manuscript requires major editing, please so state.: The manuscript is written well overall and addresses an interesting topic. It does a pretty good job as a survey and after the revision and cleaning up, I now believe the manuscript may be acceptable for publication.

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

"Geng-Sheng Kuo" <[email protected]>

Subject: [IEEE CNOM] Impact Factor of the IEEE Communications Surveys and Tutorials

From:

[email protected]

To:

[email protected]

Date:

10/24/11 02:58 AM

Dear Colleagues,

I am pleased to announce that the very first impact factor of the IEEE Communications Surveys and Tutorials has recently been reported by ISI Thomson Reuters.

It is 3.692 and the 5-year impact fator is 8.462.

http://www.comsoc.org/impact-factor-communications-surveys-and-tutorials

Nelson Fonseca

EiC, IEEE Communications Surveys and Tutorials

_______________________________________________ cnom mailing list

[email protected]

http://listas.inf.ufsc.br/mailman/listinfo/cnom

2011/12/19

http://service.mail.com/callgate-6.50.2.0/render/6.50.2.0/render/getDetails?Authorizat...

Mathematical Modeling for Network Selection in

Heterogeneous Wireless Networks – A Tutorial

Lusheng Wang and Geng-Sheng (G.S.) Kuo

Abstract—In heterogeneous wireless networks, an important task for mobile terminals is to select the best network for various communications at any time anywhere, usually called network selection. In recent years, this topic has been widely studied by using various mathematical theories. The employed theory decides the objective of optimization, complexity and performance, so it is a must to understand the potential math-ematical theories and choose the appropriate one for obtaining the best result. Therefore, this paper systematically studies the most important mathematical theories used for modeling the network selection problem in the literature. With a carefully designed unified scenario, we compare the schemes of various mathematical theories and discuss the ways to benefit from combining multiple of them together. Furthermore, an integrated scheme using multiple attribute decision making as the core of the selection procedure is proposed.

Index Terms—Network selection, heterogeneous wireless net-works (HWNs), utility theory, multiple attribute decision making (MADM), fuzzy logic, game theory, combinatorial optimization, Markov chain.

I. INTRODUCTION

T

He recent development of wireless technologies has totally revolutionized the world of communications. Mul-tiple technologies are evolving simultaneously towards provid-ing users with high-quality services of broadband access and seamless mobility. On one hand, wireless wide area networks (WWANs) evolve from GSM to UMTS and beyond 3G, providing wide coverage and good mobility capabilities. On the other hand, a series of standards of wireless local area networks (WLANs), including IEEE 802.11a, IEEE 802.11b, IEEE 802.11g, IEEE 802.11n, etc., have been established for local-area high-speed economic wireless access. To comple-ment them, wireless personal area networks (WPANs), e.g., Bluetooth and Zigbee, and wireless metropolitan area networks (WMANs), e.g., WiMAX, are developed for short-range and metropolitan coverages, respectively. All the above networks have been deployed with coverage overlapping one another, hence forming a hybrid network for wireless access, which is usually called heterogeneous wireless networks (HWNs).

L. Wang is with Mobile Communications System Department, Institute Eurecom, 06904 Sophia Antipolis, France (e-mail: [email protected]). G.S. Kuo is with National Chengchi University, Taipei, Taiwan (e-mail: [email protected]).

Manuscript received March 12, 2011; revised September 19, 2011; accepted December 4, 2011.

The work of the first author was supported mainly by France Telecom Orange Labs from EC FP7 EuroNF Project and partly by Institute Eurecom. The work of the second author was supported by the National Science Council of Taiwan under Grant NSC 97-2410-H-004-116-MY3.

Digital Object Identifier ...

To access the Internet through HWNs, current terminals, e.g., laptops and cellphones, are usually installed with multiple wireless access network interfaces. One type of terminals widely used nowadays is those with multiple interfaces but no functionality to support IP mobility or multihoming, called multi-mode mobile terminals. The other is with IP mobility and multihoming functionalities, called multi-homed mobile terminals. Mobility means that a terminal can switch between networks without breaking on-going communications. Multi-homing means that a terminal has multiple IP connections to one or multiple networks simultaneously. Multi-homed terminals use multiple interfaces to share load for the same session and support session continuity with low (or no) packet loss during mobility or link break. By contrast, multi-mode terminals can only select and use one interface for certain session at a time.

Both multi-mode and multi-homed terminals require always to rank the access networks and select the best at any time anywhere, which is well known as always best connected (ABC). ABC brings plenty of advantages to users. With ABC functionality, terminals select appropriate access networks to fit for various QoS requirements of applications; terminals avoid selecting a network with high traffic load for avoiding congestion; terminals predict networks’ availability so that they do not connect to networks which disappear soon; and terminals minimize signalling costs by using network selection and handover decision strategies specifically for this purpose. Moreover, ABC benefits operators. Since ABC has the feature of assisting the assignment of traffic load to multiple networks, operators maximize the utilization rate of the resources of the networks they operated, hence maximizing revenue. According to network selection strategies, operators analyze and decide the number of WiFi access points they should deploy to attract users to WLANs. Finally, ABC is suitable to synthetically consider users’ and operators’ benefits, so that a win-win partnership can be achieved.

ABC contains many necessary components [1], such as network discovery, network selection, handover execution, authentication, authorization and accounting (AAA), mobility management, profile handling, content adaptation, etc., in which network selection is a key component and will be extensively discussed in this paper. In recent years, a large number of research works have discussed the selection of the best network. Among them, different mathematical theories have been used for modeling this problem. Although two survey papers on this topic [2], [3] have been published, they were not focused on the mathematical theories used to model

TABLE I

NETWORKS ANDSELECTEDATTRIBUTES IN THEUNIFIEDSCENARIO Bandwidth Price Cell radius Security Power consumption Traffic

WWAN 2 50 2000 3 1/100 X

WMAN 10 20 2000 3 1/100 X

WLAN 54 5 75 1 1/50 X

WPAN 1 1 10 2 1/1000 X

TABLE II

SELECTEDPROPERTIES OF THE16 USERS IN THEUNIFIEDSCENARIO

User No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Application Conversational • • • • Streaming • • • • Interactive • • • • Background • • • •

User Money first • • • • • • • •

Quality first • • • • • • • •

Terminal Battery first • • • • • • • •

Mobility first • • • • • • • •

this problem. Based on our study, the mathematical model used for representing the problem is the first thing and the most important thing we should consider when designing a network selection strategy. It decides the aim of optimization, the utilization of different parameters, and the performance of the selection strategy. Therefore, to fill out this blank, we conduct a serious survey and provide a systematic tutorial on mathematical theories for modeling the network selection problem.

Throughout this paper, we use a unified scenario to help explain schemes using different mathematical theories. On the network side, we consider 4 types of available

net-works (i.e., WWAN, WMAN, WLAN and WPAN) and 6

attributes (i.e., bandwidth, price, cell radius, security, power consumption and traffic), as given in Table I. These attributes are carefully selected, so that there is upward attribute e.g., bandwidth, downward attribute e.g., price, dynamic attribute e.g., traffic, terminal-related attribute e.g., power consumption, application-related attribute e.g., security and mobility-related attribute e.g., cell radius. Note that one attribute could have multiple of these features. Moreover, we design WMAN as a dominant alternative of WWAN, so that we could clearly see the load balancing feature of the schemes with different mathematical theories. On the user side, we consider 4 types

of applications with different QoS requirements including conversational, streaming, interactive and background [4]. For each application type, we consider4 users with different user

preferences (i.e., money first and quality first) and different terminal properties (i.e., battery first and mobility first). To-tally, there are 16 users with different user-side features, as

summarized in Table II.

VHO represents handover between different types of access technologies, which is needed not only for connectivity reason

but also for other ones, such as user preference and network load balancing. In the literature, VHO decision is sometimes confused with the term network selection, so in this paper, we strictly distinguish the two terms: network selection is to rank networks and find the best one, while VHO decision is to decide whether it is worth the handover to the best network or a network better than the current one. VHO decision is not to simply check whether the difference between the two networks is larger than the VHO cost. In fact, this decision takes into account the predicted information of many parameters as long as they are predictable, including the expected time point that a better network will be available, the average duration that a better network can last, the probability density function of a better network’s dwelling time, the utilities of networks, etc. However, since the subject of this tutorial is network selection, we are not going to discuss too much on VHO decision.

The rest of this paper is organized as follows. From Sections II to VII, we systematically discuss the existing studies on network selection using utility theory (cost function), multiple attribute decision making, fuzzy logic, game theory, combi-natorial optimization, Markov chain, respectively. In Section VIII, we compare schemes using different mathematical the-ories, discuss the ways to combine multiple of these theories together, and propose an integrated scheme in the end. Section IX concludes the paper. Finally, Section X and Section XI provides the notations and the glossary.

II. UTILITYTHEORY(COSTFUNCTION)

For making a decision, utility refers to the satisfaction that a goods or service provides to the decision maker [5]. An associ-ated term is utility function which relates to the utility derived by a consumer from a goods or service. Different consumers with different user preferences will have different utility values

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 x Utility(x) exponential 1−e−ax, a=20

logarithmic ln(1+ax)/ln(1+a), a=5 exponential ea(x−1), a=20 linear sigmoidal (x/0.5)a/[1+(x/0.5)a], a=5 sigmoidal (x/0.5)a/[1+(x/0.5)a], a=20 sigmoidal (x/0.75)a/[1+(x/0.75)a], a=20

Fig. 1. Typical utility functions.

for the same product. Thus, the individual preferences should be taken into account in the utility evaluation.

A. Utility functions in network selection

Utilities can be classified into monotonic utilities and non-monotonic ones. The utility is said to be non-monotonic if the measure of satisfaction associated with the attribute shows a monotonic increase and decrease with an increase in attribute value. Otherwise, it is said to be non-monotonic. Normally, monotonic utilities are used, except if the attribute is consid-ered as the nominal-the-best. For a nominal-the-best attribute, instead of considering the best (either the largest or the smallest) as the most desired network, the one that is closest to the service requirement is preferred [6]. When evaluating the utility of an attribute, we should distinguish between the upward and downward attributes. The attributes of which the higher preference relation is in favor of the higher value are called upward attributes. Conversely, the downward attributes encompass various costs. Given an attribute, its utility can be calculated based on certain utility function. And, the utility function of one attribute could be different from that of others. Some examples of common utility functions are shown in Fig. 1. It is important to select the suitable utility function for each attribute. Sigmoidal utility function is considered to be suitable for the network selection problem [7], but the parameters in the sigmoidal function might be different to fit for different attributes’ features.

During the network selection procedure, we consider mul-tiple attributes together, so the utilities of mulmul-tiple attributes are combined as a total utility. It has been pointed out that a valid form to combine these attributes together satisfies the following requirements [7]:          ∂U ∂uj ≥ 0 lim uj→0 U = 0, ∀j = 1, ..., M lim u1,...,uM→1 U = 1 (1)

where U is the total utility of all the attributes and uj is

the utility of attributej. M denotes the number of attributes

throughout this paper.

Cost function is a measurement of the cost caused by using certain network. Usually, the cost of a network can be considered as the inverse of its utility, but the form of this inversion is related with the way to combine multiple attributes. For example, if these attributes are summed up, the total cost is calculated as the cost minus the utility. A general form of cost function for the network selection problem was given in [8], which integrates a large number of attributes, their weights, and furthermore, network elimination factors given by

Fi= X k (Y j ǫkij) X j [fjk(wkj)N (ukij)], (2) whereN (uk

ij) represents the normalized utility of application

k in network i in terms of attribute j. fk

j(wkj) is the weighting

function of attribute j for application k. ǫk

ij is the network

elimination factor, either1 or infinite, to reflect whether current

network conditions are suitable for requested applications. For example, if a network cannot guarantee the delay requirement of certain real-time application, its corresponding elimination factor will be set to infinite. Thus, the corresponding cost becomes infinite, which eliminates this network.

One study that is worth mentioning is the usage of the consumer-surplus concept of microeconomics in [9]. Users always search for cost effective solutions to meet their ex-pectations. If the price is less than the value the user is willing to pay, he saves money. Consumer-surplus represents the difference between the monetary value of the data to the user and its actual price, so the network with the best predicted consumer-surplus, which is also predicted to meet the service completion deadline, will be selected.

B. Attributes in network selection

A lot of studies model the network selection issue with cost or utility functions, but they may consider different attributes and measure them in different manners. A summary of attributes and their usage in different papers is provided in Table III. For types of attributes, we first classify them into upward and downward attributes, then static, dynamic and semi-dynamic attribute. Semi-dynamic attributes are those that are not totally static but not quite dynamic either. For example, bandwidth is sometimes used statically as the total bandwidth of each network, but sometime used dynamically as the average bandwidth a user obtains. Bit error rate (BER), jitter and service completion time are changeable along with the environment and the network condition, but it is difficult to dynamically evaluate their instantaneous values for network selection, so they are classified as semi-dynamic attributes. Moreover, we also consider some other features of attributes, such as mobility-related, QoS-related, terminal-related and inter-network. For lists of references, considering that every study on network selection will use one or multiple attributes as decision criteria and some key attributes are even used by most studies on this issue, so it is tedious to provide complete lists for all the attributes. Instead, Table III just aims to list

TABLE III

KEYATTRIBUTES ANDTHEIRUTILITYFUNCTIONS

Attribute Types References Utility functions

Bandwidth upward/semi-dynamic/QoS-related _{[28], [30], [37], [38], [46], [51]}[7], [8], [10], [20], [21], [23], [24], linear, logarithmic,_sigmoidal Cell radius (diameter) upward/static/mobility-related [38] linear

Security upward/static/QoS-related [10], [21], [23], [24], [51] linear, sigmoidal Battery upward/dynamic/terminal-related [21], [22], [28] linear

SNR/SIR upward/dynamic/QoS-related [21], [22] linear

RSS upward/dynamic/QoS-related [11]–[13], [21], [28], [51] linear Price downward/static [7]–[10], [13], [21], [23], [24], [28],

[34], [38] linear, logarithmic VHO signaling cost downward/static/mobility-related/inter-network [12], [24], [54] linear

VHO latency downward/static/mobility-related/inter-network [12], [27] linear HHO signaling cost downward/static/mobility-related [12], [54] linear HHO latency downward/static/mobility-related [12], [38] linear Handover failure probability downward/static/mobility-related [27] linear Interruption probability downward/static/mobility-related [27] linear Size of unsent messages downward/static/mobility-related [27] linear

Traffic downward/dynamic [7], [11], [24], [34], [37] linear, sigmoidal Power consumption downward/static/terminal-related [24], [38], [51] linear

BER downward/semi-dynamic/QoS-related [21], [23], [24] linear, sigmoidal Delay downward/semi-dynamic/QoS-related [20], [21], [23], [51] linear, sigmoidal Packet loss downward/semi-dynamic/QoS-related [20], [23] linear, sigmoidal Jitter downward/semi-dynamic/QoS-related [20], [21], [23], [24], [51] linear, sigmoidal Response time downward/semi-dynamic/QoS-related [23] linear

Service completion time downward/semi-dynamic/QoS-related [9] linear, polynomial, exponential

some most typical examples of each attribute. For utility func-tions used in the literature, most studies that do not specifically discuss utility functions could be considered as using linear utility functions. While in some recent studies, polynomial, logarithmic, exponential and sigmoidal utility functions are utilized for some attributes, which are summarized in this table.

In the above presentations, we discussed utility functions for various attributes. To avoid a potential misunderstanding, we would like to point out that utility function for a certain attribute could be totally different in different scenarios. For example, the utility of bandwidth should jump to a fixed value after certain thresholds for voice and video applications, but kind of linearly increase for data application [13]. If sigmoidal functions are used, the parametera, as shown in Fig. 1, should

be large for voice and video applications while small for data applications. For voice and video applications, the mid values, corresponding to the thresholds, should be also different.

Moreover, it is important to state clear that other studies on the network selection issue could also evaluate networks based on utility/cost functions which combine multiple attributes. However, those studies focus on other mathematical models, which will be presented in later sections.

C. Case study

We consider the unified scenario presented in Section I with Tables I and II. Since it would be unfair by assuming different networks with different traffic conditions, we assume that they have the same traffic condition, which means that the attribute ‘traffic’ is not considered in this case study. Based on the above studies, sigmoidal utility functions with different configurations of mid value and parametera, as shown in Fig.

1, are used for different attributes under the cases of different user-side properties. For example, user 5 requires streaming

application while user1 requires conversational application, so