intelligent Converged network consolIdating Radio and ... - … · method, key issues such as synchronization and channel estimation are discussed. Then, for the purpose of implementation,

iCIRRUS Contract No. 644526 1 Jan 2015 – 31 Dec 2017

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 644526 (iCIRRUS)

intelligent Converged network consolIdating Radio and optical access aRound USer equipment

DELIVERABLE: D4.2 - iCIRRUS UE D2D & D2I interfacing

Contract number: 644526

Project acronym: iCIRRUS

Project title: Intelligent converged network consolidating radio and optical access around user equipment

Project duration: 1 January 2015 – 31 December 2017

Coordinator: Nathan Gomes, University of Kent, Canterbury, UK

Deliverable Number: D4.2

Type: Report

Dissemination level Public

Date submitted: 20-05-2016

Editors: Cunhua Pan

Authors / contributors (contributing partners)

Cunhua Pan, Yuan Kai, Huiling Zhu (UniKent), Luz Fernandez del Rosal, Volker Jungnickel (HHI), Stavros Hadjitheophanous (PTEL), Patrik Ritoša (TS), Geza Koczian, Mike Parker (UEssex)

Internal reviewers Ku Yang (UEssex), Onur Sahin (IDCC)


1


Document history

Version 0.0: 22/03/2016

Version 0.1: 06/04/2016

Version 0.2:12/04/2016

Version 0.3:17/04/2016

Version 0.4:22/04/2016 for internal review

Version 0.5:29/04/2016 internally reviewed changes and comments

Version 0.6:02/05/2016 revised after internal review

Version 0.7:06/05/2016 version for checking by internal reviewers

Version 0.8:13/05/2016 for final checking

Version 1.0:19/05/2016 Final, uploaded


2


Abstract This deliverable D4.2 “iCIRRUS UE D2D & D2I interfacing” reports on the results achieved on device-to-device (D2D) and device-to-infrastructure (D2I) communications during the first 16 months in Task 4.1 “D2D and D2I Communications” of Work Package WP4 “Efficient User Equipment Integration” of the iCIRRUS project. Besides, results in Task 4.1, some updated work in Task 4.2 “60 GHz D2D and Mesh Networking” is also included in this deliverable. This report includes four main parts: key D2D technologies investigated, semi-distributed D2D resource management for iCIRRUS infrastructure, device to infrastructure (D2I) communication, and inventory model for the CRAN, D2D and D2I.

The first part provides detailed description of technical contributions and results achieved in Task 4.1 on several important techniques in D2D communications, including the D2D discovery, multiple device to single device communication (MD2D) with content caching, D2D assisted content caching, security issue in D2D communications, and 60 GHz networking for D2D communications.

In Task 4.1, D2D discovery signals were considered to be multiplexed with cellular signals causing in-band emission interference (IEI), which degrades D2D user equipment’s (DUE’s) discovery range and cellular user equipment’s (CUE’s) throughput. In this project, a new discovery method is proposed by applying power control strategy to deal with the adverse effect of IEI. Content caching is another important technical issue in D2D communications. In this deliverable, the system model for a proposed MD2D content caching method is explained and described. As a future work in the iCIRRUS project, the outage performance underlying this model will be analysed for the proposed MD2D content caching method. The second contribution on content caching is that an energy efficient content pushing strategy is investigated among different mobile users (Mus) groups to minimize the total energy consumption in content dissemination. Security is also one of the major concerns for D2D communications. To deal with this issue, the well-known Diffie-Hellman key agreement protocol was adapted to D2D communications. In addition, new contributions and results on 60 GHz networking are reported, where a maximum throughput up to 2.4Gbps has been successfully demonstrated in the laboratory over 20 m between devices, which shows the suitability of applying 60GHz for D2D applications in short range communications.

In the second part, as an important contribution to D2D communication underlaying the iCIRRUS infrastructure, a semi-distributed D2D resource management procedure proposed in Task 4.1 with the objective of achieving high throughput under low D2I overhead is reported, along with the performance analyses and achieved results.

In a previous iCIRRUS deliverable, D4.1, it was assumed that the control information exchanged between end devices and infrastructure has to be sent back to the centralized BBU in order to be fully processed, and then the decision sent back to the end user. In this deliverable, a novel control signalling design is reported with the target of reducing the latency and the performance


3


of D2D links by shifting additional processing capability to the RRHs. To support this signalling method, key issues such as synchronization and channel estimation are discussed. Then, for the purpose of implementation, the system parameters of the proposed semi-distributed resource management are described in detail under the general control signalling structure.

Finally, initial results for the inventory models for the C-RAN, D2D, and D2I are presented. The elements, innovative service attributes, radio connections, and control signaling are presented. . In particular, the D2D communication is modeled as an additional subscriber service on the C-RAN architecture.


4


Executive Summary This deliverable D4.2 “iCIRRUS UE D2D & D2I interfacing” is used to report on the technical outputs on device-to-device (D2D) and device-to-infrastructure communications, which are achieved in the work package WP4 “Efficient User Equipment Integration” of the iCIRRUS project in the first 16 months. The iCIRRUS project focuses on the wireless radio access network segment of future, 5th generation (5G) mobile networks.

This report includes outputs of technologies enabling D2D to work under the infrastructure of C-RAN, e.g., content caching enabled D2D techniques, semi-distributed resource management in D2D communications, innovative techniques in D2I communications, and inventory models for D2D and D2I, which are suitable for future 5G networks. The results are mainly from the current two tasks in WP4: Task 4.1 “D2D and D2I communications” and Task 4.2 “60 GHz D2D and Mesh Networking”.

The technologies presented in D4.2 will continue to be developed in WP4 in the 2nd and 3rd years of the iCIRRUS project. Parallel workpackages WP2 “Centralisation Scenarios and Architecture”, and WP3 “Future Converged Front- (Mid-) Haul Architecture) will also benefit from these WP4 technologies; as indeed, the tasks in WP4 benefit from the parallel WPs, in terms of technology steering, and understanding of application scenarios. During the 2nd year of the project, the results from WP4 will also feed into the design of the test-bed demonstrators and technology validators being developed in WP5.


5


Index of terms

3G 3rd Generation

3GPP 3rd Generation Partnership Project

4G 4th Generation

5G 5th Generation

ATM Asynchronous Transfer Mode

BBU BaseBand Unit (synonym: DU)

BER Bit Error Rate

BS Base Station

CCM Continuity Check Messages

CCM Central Clone Management

CDMA Code Division Multiple Access

CFO Carrier Frequency Offset

CO Central Office

CoMP Coordinated Multi-point

CPRI Common Public Radio Interface

C-RAN Cloud-Radio Access Network (also Centralised-Radio Access Network)

CSI Channel State Information

CWDM Coarse Wavelength Division Multiplexing

D2D Device-to-Device

D2I Device-to-Infrastructure

DDMI Digital Diagnostics Monitoring Interface

DHCP Dynamic Host Configuration Protocol

DU Digital Unit

DWDM Dense Wavelength Division Multiplexing

eNB Evolved Universal Terrestrial Radio Access Network

EPC Evolved Packet Core

ETSI European Telecommunications Standards Institute

G-PON Gigabit-Passive Optical Network


6


GNSS Global Navigation Satellite System

GPS Global Positioning System

IDPC Inter Device Power Management

IM Intelligence Module

IMT-A International Mobile Telecommunications - Advanced

IMT-2020 International Mobile Telecommunications 2020

IPsec Internet Protocol Security

ITU-R International Telecommunications Union - Radio

KPI Key Performance Indicator

LTE-A Long Term Evolution-Advanced

MCN Mobile Cloud Networking

MD2D Multiple Devices to single Device

MEF Metro Ethernet Forum

MIMO Multiple Input Multiple Output

MSA Multi-Source Agreement

NGMN Next Generation Mobile Networks

NG-PON Next Generation Passive Optical Network

OAM Operation Administration and Maintenance (or Management)

OBSAI Open Base Station Architecture Initiative

ODN Optical Distribution Network

ORI Open Radio equipment Interface

OTN Optical Transport Network

PON Passive Optical Network

QoE Quality of Experience

QoS Quality of Service

RAN Radio Access Network

RAT Radio Access Technology

RF Radio Frequency

RFM Radio Frequency Module


7


RM Resource Management

RoF Radio over Fiber

RRH Radio Remote Head (synonym to: RU)

RRU Remote Radio Unit

RSRP Reference Signal Received Power

RSRQ Reference Signal Received Quality

RSTD Radio Signal Time Difference

RTMP Real Time Messaging Protocol

RTP Real-time Transport Protocol

RTT Round Trip Time

RU Radio Unit

SFP Small Form Pluggable

SIP Session Initiation Protocol

SLA Service Level Agreement

SR Scheduling Request

SSH Secure Shell

TCM Tandem Connection Monitoring

TDM Time Division Multiplexing

WDM Wavelength Division Multiplexing

XMPP Extensible Messaging Presence Protocol


8


Contents Document history _____________________________________________________ 1

Abstract _____________________________________________________________ 2

Executive Summary ___________________________________________________ 4

Index of terms ________________________________________________________ 5

1 Introduction _____________________________________________________ 10

2 D2D Technologies Investigated _____________________________________ 12

2.1 D2D discovery _______________________________________________________ 12

2.1.1 Interference management for D2D discovery ______________________________ 12

2.2 Multiple device to single device communication with content caching ________ 20

Introduction of the content caching for multiple device to single device ________ 20

System model ____________________________________________________ 22

2.3 D2D assisted content caching __________________________________________ 23

Introduction of D2D Assisted Content Dissemination ______________________ 23

System Model ____________________________________________________ 24

Problem Formulation ______________________________________________ 28

Solution and Algorithms ____________________________________________ 28

Numerical Results ________________________________________________ 32

Conclusion ______________________________________________________ 35

2.4 Security issue in D2D communications __________________________________ 35

2.5 60 GHz Networking for Low-Latency & High-Throughput ____________________ 36

3 Semi-distributed Resource Management for D2D communication ________ 37

3.1 Local resource management schemes ___________________________________ 40

Best subcarrier CSI (BSCR)-based resource management _________________ 41

Subcarrier achievable data-rate (SAR)-based resource management_________ 41

3.2 Performance evaluation of the local resource management schemes _________ 42

Spectral efficiency ________________________________________________ 42

Effect of coherence bandwidth _______________________________________ 43

3.3 Central spectrum assignment at BBU Hotel _______________________________ 44


9


4 Device to Infrastructure Communication _____________________________ 46

4.1 Control Signaling Design ______________________________________________ 46

Evolved Fronthaul Architecture ______________________________________ 46

Coordinated Data Transmission Between end Users______________________ 47

4.2 D2I based on semi-distributed resource management ______________________ 50

D2D control signaling over C-RAN architecture __________________________ 50

Control signaling between D2D receivers and RRHs ______________________ 50

Control signaling between RRHs and BBU Hotel _________________________ 51

Summarize and future work: ________________________________________ 51

5 Inventory models for the C-RAN, D2D and D2I _________________________ 52

5.1 C-RAN (Cloud Radio Access Network) inventory model _____________________ 52

5.2 D2I (Device to Infrastructure) inventory model ____________________________ 53

5.3 D2D (Device to Device) inventory model _________________________________ 54

6 Conclusions _____________________________________________________ 55

7 References ______________________________________________________ 56

8 List of figures ____________________________________________________ 60

9 List of tables_____________________________________________________ 61


10


1 Introduction This deliverable, D4.2 “iCIRRUS UE D2D & D2I interfacing”, reports the technology innovations developed underlying the iCIRRUS architecture with particular emphasis on the key technologies in device-to-device (D2D) and device-to-infrastructure (D2I) communications. Some recent state-of-the-art innovations have been described in this deliverable, as well as innovations intrinsic to the iCIRRUS project. Some architecture related results that have been developed in the work packages WP2 and WP3, such as the novel modified functional split that with some data processing capability in RRHs, have been leveraged in the development of algorithms and procedures developed in this work. For instance, semi-distributed resource management scheme and D2I control signaling schemes were proposed to be integrated into the overall iCIRRUS architecture.

Chapter 2 describes the key techniques addressed in the iCIRRUS project for D2D communications, which mainly includes the D2D discovery, MD2D with content caching, D2D assisted content caching, security issue in D2D communications, and 60 GHz networking for D2D communications. One novel power control based D2D discovery method is proposed, where the simulation results show that this scheme significantly outperforms the existing discovery method in terms of SINR performance. The system model for multiple devices to single device (MD2D) with content caching has been established and the outage performance will be represented in the future deliverables. One energy efficiency pushing strategy is proposed to minimize the total network power consumption and offload the macrocell traffic to the D2D side, and the simulation results show that the proposed strategy can dramatically reduce the system power consumption. Diffie-Hellman key exchange method is applied to conform with the security requirements in D2D communications, while D2D link is still vulnerable to well-known man-in-the middle attack. Future work will continue to develop efficient method to conquer this attack. . Moreover, under the 60 GHz networking, maximum throughput of 2.4Gbps up to 20 metres have been successfully demonstrated in the lab. This technology can be applied for D2D communications due to the short distance between two D2D devices. . In addition, significant end-to-end latency gains can be achieved by this technology.

In chapter 3, taking the C-RAN architecture in ICIRRUS into consideration, a semi-distributed resource management and the detailed resource allocation algorithms are described. The semi-distributed resource management takes the advantage of multiple RRHs C-RAN architecture to reduce the large CSI collection and transmission signalling as well as computation requirements at BBU hotel. The performance of D2D resource allocation algorithms under semi-distributed resource management is discussed as well.

On the other hand, since the extra fronthaul latency introduced by C-RAN architecture is non-negligible, the increased fronthaul latency has significant impact on the effectiveness of resource allocation, especially when the wireless channel variation between D2D transmitter and receiver is taken into account. Therefore, in Chapter 4, a general D2I communication protocol between D2D users and BBU hotel via RRHs is described. The impact of the fronthaul latency would be eliminated by introducing a new function splitting point between RRHs and


11


BBU hotel The synchronization, channel estimation and rate adaption methods for the newly proposed functional splitting are discussed as well. The key system parameters for D2D semi-resource management under the proposed general D2I communication protocol are described. By adopting the proposed D2I communication protocol, it is also discussed in this deliverable how the semi-distributed resource management schemes could be implemented into the current C-RAN architecture in iCIRRUS.

Another contribution in this deliverable is the inventory models development for C-RAN, D2D, and D2I. Inventory models are part of OSS (Operations Support Systems). They are used by the operators for efficient service provision over the entire network. It contains information about the network physical infrastructure and equipment, services availability and subscriber package information.

In today networks, operators are faced with many different services, technologies and service level demands. To keep control and manage all different network elements it is essential to keep network topology and functionalities up-to-date as much as possible. In this deliverable, the detailed inventory model for D2D is studied, which should be applied in the future 5G networks.

This document is organized as follows. Chapter 1 gives the introductory remarks concerning the contributions in this deliverable, while Chapter 2 offers the detailed technical description of several key techniques developed in iCIRRUS for D2D communications. Chapter 3 provides details of the proposed semi-distributed resource management for D2D communications underlaying iCIRRUS infrastructure. Chapter 4 describes the initial results of the work on the D2I resource management and control signaling design. Chapter 5 summarizes the developed inventory models for D2D, D2I, and C-RAN. Final concluding remarks are given in Chapter 6.

To provide a more clear description for this deliverable, the relationships between the main four sections of this deliverable is described in Figure 1.

Sec-2 D2D Technologies Investigated

Sec-3 Semi-distributed Resource Management

Sec-4 Device to Infrastructure Communication

Sec-5 Inventory models for the C-RAN, D2D and D2I

To/fromWP2,WP3

Deliverable 4.2

To WP6 To WP5,WP6


12


Figure 1 Relationships for the four sections.

2 D2D Technologies Investigated This chapter focuses on the key techniques developed in the iCIRRUS project for D2D communications.

2.1 D2D discovery

2.1.1 Interference management for D2D discovery 1

1) Introduction of the in-band emission interference (IEI) in D2D discovery procedure

Unlike the conventional many-to-one uplink cellular architecture, there exist multiple D2D pairs in a D2D network, which forms a many-to-many network architecture. When D2D users perform the discovery progress, each discovery user receives signals from multiple discovery transmitters and each discovery transmitter will face many receivers. This special structure makes the conventional uplink power control not applicable for D2D discovery network. [1][2][3] assume that the discovery users apply the same fixed transmit power, e.g., the maximum transmit power of 23dBm for discovery signals. On the other hand, according to the agreements of 3GPP LTE, some dedicated uplink cellular physical resource blocks (PRB) are reserved for the D2D discovery purpose.

Hence, due to common resource sharing, the issue of in-band emission interference (IEI) occurs, where the IEI is defined as the interference leakage from the allocated bandwidth to the bandwidth dedicated only for cellular users [4]. The IEI becomes significant when D2D pairs are very close to the base station (BS) as compared to cellular users [5]. For example, in Figure 2, for uplink transmission, the cellular user (CUE) is on the boundary of the cell while D2D users (DUE) is near the BS. Under this scenario, although cellular users and DUEs apply the orthogonal frequency resources blocks, when the DUE transmits discovery signals within current communication systems, there is no coordination for the resource allocation between CUEs and DUEs. Hence, the IEI may have a significant negative effect on the CUEs. One of the proposed solutions to reduce the IEI impact is to control the transmission power of the DUEs.

In order to mitigate IEI caused by D2D discovery while not impairing the D2D discovery efficiency in terms of the probability of successful discovery for DUEs, in this subsection, a new discovery method is proposed by dividing DUEs into different groups and adopting the scheduling information block (SIB) to provide information only for necessary groups to use open-loop power control (OLPC) and orthogonality-conserved symbols (OCSs), where OCSs were defined as the aligned symbols between DRB and physical resource block (PRB) of CUE in a specific sub-frame. In this method, DUEs are classified into two groups based on whether

the received power of a reference signal RSRPP from BS at a DUE is larger than a threshold. If

not, the user will be put in Group-1, otherwise, the user will be put in Group-2. Obviously,

1 The work of this part has been accepted in IEEE VTC Spring, 2016.


13


Group-1 DUEs are the users that have minor influence on the performance of cellular users. In general, the users classified in Group-1 are far away from the BS. On the other hand, cellular users will be significantly affected by IEI from DUEs in Group-2. We alleviate the IEI impact on Group-2 to make it more reliable to CUE by utilizing SIB information, and using OCSs estimation. We apply OLPC if only the cellular user is scheduled to send cellular traffic, and there is no OCSs between the DUEs and cellular users. Our method significantly improves D2D discovery performance as compared to conventional techniques. This proposed method does not affect the cellular user’s received signal at the BS. Furthermore, this method needs less SIB resources than the conventional approach as proposed in [6], which makes it more efficient.

Figure 2: Illustration of In-band emission interference in D2D overlaid in cellular network

2) Proposed BS aided device discovery method

Figure 3 D2D discovery channel structure


14


Before presenting the discovery method, we first provide the discovery channel structure. Figure 3 depicts the D2D discovery channel structure as proposed in 3rd Generation Partnership Project (3GPP) standardization [6] [5] [7]. As shown in Figure 3, the discovery channel is composed of TN sub-frames in time domain. Discovery resource block (DRBs) are frequency multiplexed with cellular PRB. 2 PRB TN N× resource blocks are allocated to the cellular users, while DRB TN N× resource blocks are allocated to the D2DDTs [6] [5] [7]. A DUE transmits discovery signal on its allocated DRB and listens to the remaining DRBs to discover neighbours. To be compatible with the existing LTE system, the time duration of a DRB is one time slot of 0.5 milliseconds. Each DRB consists of 12 subcarriers, each having a bandwidth of 15 kHz.

Figure 4 Proposed discovery model

Based on the above channel structure, we begin to show the idea of our discovery method. Figure 4 describes the main idea of the proposed discovery method. The discovery period (DP) is divided into two time groups, Group-1 and Group-2. In Group-2, a DRB is allocated for each sub-frame in the middle of DP. This DRB is used as SIB. The proposed algorithm is summarized as follows:

Step-1: All DUEs need to be authorized by the network to start the discovery process. DUE sends an authorization request to the network in order to obtain related information such as (Cell-ID, announcing policy, etc.). The BS responds by sending the authorization information to the DUE.

Step-2: DUE calculates the RSRPP as follows

( Shadowing),RSRP TxP P PL= − + where TxP is the discovery transmission power of DUE and PL is the path loss between the BS and the DUE. Shadowing is assumed to follow log-normal distribution. The DUE then transmits the discovery request to the BS.

Step-3: After authorization has been done, the BS prepares discovery response messages to the authorized DUEs. These messages contain the discovery authorization code, a valid timer for the discovery, and DRB allocation to the DUE. The resource allocation of DUEs depends on RSRPP . The DP is divided into two groups in time. Group-1 resources are allocated to DUEs if RSRPP < pre-defined threshold, and Group-2 resources are allocated to DUEs if RSRPP > pre-defined threshold. This step encloses the DUEs that cause IEI in one


15


group. Group-1 presents the DUEs that are likely to be located far from the BS. Group-2 DUEs are likely to be located close to the BS as shown in Figure 1.

Step-4: Each DUE starts the discovery procedure by using the DRBs that has been assigned to them by the network (step-3). The DUEs announce the discovery information and (monitor or listen to) the rest of the DRBs in order to discover neighbours. The DUEs in Group-1 announce the discovery information without applying any power control strategy as they are far from the BS and the IEI impact is negligible. Meanwhile, the DUEs in Group-2 are subject to a power control strategy in order to reduce the IEI, as they are closer to the BS. Group-2 DUEs start to check the SIB information, which are sent by CUEs. This information contains the scheduling information and CP configuration of the CUEs that will help DUEs to decide their discovery transmission power. DUEs can adjust the transmission power to maximize the neighbourhood discovery, in case there is a scheduled CUE. Furthermore, DUEs estimates the OCS locations by using one of following methods: path loss, round trip delay (RTD) information, timing advance command of BS, and the SIB information. The DUEs boost the discovery transmission power in OCS as these symbols is not affected by IEI. Therefore, Group-2 DUEs only apply OLPC if two conditions are met: There is a scheduled CUE user, and there is no orthogonality between CUE and DUE symbols. Otherwise, the DUE can boost its transmission power to announce its discovery signal no more than predefined maximum power. Accordingly, the discovery transmission power of DUE is calculated as follows:

In Group-1 no power control (NPC) is applied. In Group-2, NPC is applied, where there is no scheduled CUE. In this scenario, the power transmitted by the DUE is given by

( )max ,k

DUEP P= (1)

where k is the DRB index of DUE, and maxP is maximum transmission power.

In Group-2, if CUE is scheduled, the transmission power of DUE for each symbol is adjusted as

{ }( )( ) m

max

ax,

min , ,=

,

if OC

S,

if OCS

kRx Thrk

DUE n

P PL nP

P

P n− + ∈

∈/

where n is the DUE symbol index in sub-frame, Rx ThrP − is target interference level that the BS

receives, and ( )kPL is the path loss between the BS and the DUE using the k th DRB. Finally, the total discovery transmission power for one sub-frame is given by

( ) ( ),

1

1 ,m

k kDUE DUE n

nP P

m =

= ∑ (2)

where m is the total number of symbols in one sub-frame.

Step-5: After the DUEs discovery procedures, the discovery report then is sent to the BS. The BS terminates the discovery process once the discovery time runs out.

3) Performance evaluation of the proposed discovery method


16


In this subsection, we evaluate the proposed power control strategy in terms of IEI impact on CUEs and the performance of DUE discovery. We compare our proposed method with the conventional strategies as proposed in [6][5][7]. Case-A and Case-B in Group-2 in this subsection denote the performance of our proposed method. Case-A is used if there are scheduled CUEs defined by SIB. In this case, DUE can maximize discovery transmission power only in OCS symbols. Case-B is used if there is no scheduled CUE. In this case, DUE can maximize transmission power to transmit the discovery signal. Case-B will also show us the upper bound which can be achieved when there is no IEI. Compared with the existing no power control method, our proposed method may incur some additional overhead. However, it is not analyzed here since it is related to the complicated protocols of the whole system, which is left for future work.

A. Simulation parameters

The simulation parameters are set according to 3GPP standardization. All the parameters are shown in Table 1. DUEs are uniformly distributed in a radius of 250 meters. The maximum transmission power is 23 dBm. Small scale fading is modelled by Rayleigh distribution, while log-normal and standard path loss is assumed for large scale fading, respectively [8]. No more than one DRB is allocated to each DUE.

Table 1 Simulation Parameters

Parameter Value

Cluster radius 250 m

UE dropping Uniform random distribution

Noise power density -174 dBm

Path loss exponent 4

System Bandwidth 10 MHz

DRBN 44

Carrier frequency 2GHz

Shadowing standard deviation

7dB

CUE target to BS distance

50 m

DUE to DUE target 3 m

RSRP pre-defined threshold [5]

-67 dBm


17


B. Effect on CUE

The impact of DUE on CUE is evaluated by using signal-to-interference-plus-noise ratio (SINR) of CUE at the BS. We assume there is a targeted CUE which transmits a cellular signal, and is occupying the PRB located directly next to the DP. The evaluation is based upon worst case scenario. Therefore, the targeted CUE is located at the edge area of each group, separately. Further, DRBN DUEs are transmitting their discovery signal in the same sub-frame as the CUE.

DRBN is the maximum number of DUEs allowed to discover neighbours in each sub-frame. The SINR of CUE at the BS is given by

2( ) ( ),( )

( ) 2

·,

j jCUE CUE BSj

CUE j

P hSINR

IEI σ=

+

where ( )jCUEP is the maximum transmission power of targeted CUE,

2( ),

jCUE BSh is the channel gain

between the targeted CUE and BS, j is PRB index of cellular user, and 2σ is the white

Gaussian noise power at the receiver, ( )jIEI in dB for thj PRB, given by

( ) ( ) ( ) ( )

,1

,DRBN

j k k j kdB DUE IEI DUE BS

kIEI P P H−

=

= − +∑

where ( )kDUEP is the discovery transmission power of DUE as calculated in (1) or (2), ( )

,k

DUE BSH is

the channel gain between BS and the DUE using the k th DRB, ( )k jIEIP − is the power difference

between thj PRB of the target CUE and the thk DRB of the interfering DUE, given by

( )( )

0, if in Group-1,

, if in Group-2k j

IEI k jIEI

PP

−−

=

(3)

and ( )k jIEIP − is calculated in [9] as

( )

21 , if 126 , if 2

.31 , if 336 , if 3

k jIEI

dB k jdB k j

PdB k jdB k j

−

− − =− − == − − =− − >

(4)


18


Figure 5 CDF performance comparisons of received cellular signal SINR at the BS [Grouping (Gr.), Proposed

method (Prop.), Group-1 (Gr1), Group-2 (Gr2)]

Figure 5 shows the cumulative distribution function (CDF) performance of the received cellular signal SINR at BS. In this figure, we compare the IEI impact on targeted CUE under different power control strategies. The required CUE’s SINR for successfully signal decoding at the BS is -7.8 dB [10]. From the figure, it can be observed that only two strategies SIB [6] and Group-2 NPC have decoding level less than -7.8 dB. The performance in these cases are worst as IEI will be severe as no power control technique is applied and DUE operate at maximum power of 23 dBm. From the figure, it can be observed that OLPC used in Group-2 and our proposed method in Group-2 have much better performance. In these cases, the CUE signal can be received reliably at the BS. Furthermore, from the figure it can be observed that our proposed method outperforms SIB [59], OLPC [54], and NPC in terms of IEI impact on targeted CUE. Next, we study the impact of IEI on DUE.

C. Effect on DUE


19


Figure 5 CDF performance comparisons of received discovery signal SINR at the DUE [Grouping (Gr.),

Proposed method (Prop.), Group-1 (Gr1), Group-2 (Gr2)]

The impact of CUE on the desired DUE is also evaluated by using SINR as the criterion. We assume there is a target D2D pair (a DUE is in listening mode and the other DUE is in transmitting mode). ( 1DRBN − ) other DUEs are sending their discovery signal and 2 PRBN CUEs are transmitting in the same sub-frame. In the worst case scenario, the CUEs are located near the DUE receiver. The DUE SINR is given by

2( ) ( )2( )

( ) 2

·,

k kDUE D Dk

DUE k

P hSINR

IEI σ=

+ where ( )k

DUEP is the discovery transmission power of target DUE in transmitting mode and

calculated in (1) or (2), 2( )

2k

D Dh is the channel gain between target D2D pair, ( )kIEI is the

interference due to IEI from the CUEs. Figure 6 depicts CDF of received discovery signal’s SINR. The DUE is assumed to successfully receive the discovery signal if the SINR exceeds a given threshold ThrSINR (assumed to be 8 dB in simulation) [10]. From the figure, it can be observed that only SIB [6], OLPC [4] do not exceed the ThrSINR , which means that the DUE receiver will not detect successfully the discovery signal. The performance in these cases are worst as OLPC is applied. From the figure, it can be observed that our proposed method in Group-1, Group-2 (case-A, case-B), and NPC have much better performance. In this case the DUE discovery signal can be received reliably at the DUE receiver. It can be observed that our proposed method outperforms SIB and OLPC in terms of the discovery signal detection at the DUE receiver. Even though NPC has same performance with our proposed method case-B, but achieves much better performance in the cellular side as illustrated in previous subsection.

D. Probability of successful discovery at the DUE receiver


20


Figure 6 Probability of successful discovery at the DUE

The probability of successful discovery is one of the important metrics to evaluate the discovery performance, which is defined as

( )( )kDUE ThrP SINR SINR>

Figure 6 shows the probability of successful discovery at DUE receiver when SIB, OLPC, NPC, and our proposed method Group-1 and Group-2 (case-A and case-B) algorithms are employed. The DUE is assumed to have successfully received the discovery signal if the SINR exceeds a given threshold ThrSINR . From the figure, it can be observed the probability of successful discovery under the worst case scenario, is worst for SIB and OLPC which requires a very low SINR threshold to successfully discover the discovery signal. The performance in these cases is the worst as OLPC is applied. From the figure, it can be observed that NPC, our proposed method Group-1 and Group-2 (case-A, case-B) have better performance than SIB and OLPC. For the same 90% probability of successful discovery, about 40dB gains can be achieved by using the proposed method compared with the SIB and OLPC. Furthermore, the SINR threshold for these methods is higher than SIB and OLPC.

2.2 Multiple device to single device communication with content caching

Introduction of the content caching for multiple device to single device

Global mobile data traffic has grown significantly, and is expected to grow exponentially in the coming years [11]. This constantly growing traffic load is becoming a serious concern for mobile network operators, as the Fourth-Generation (4G) cellular systems have already approached its theoretical capacity limits when serving the growing traffic [12] [13]. Out of this growing mobile traffic more than half of the data represents videos [11]. It has been further revealed that the most data traffic load is caused by duplicated downloads of a few popular


21


short video clips. For instance, 10% of the videos in YouTube account for nearly 80% of viewing [14]. Based on this fact, the peak traffic can be significantly reduced by content caching. With content caching the multimedia contents can be stored and reached from nodes near to terminal devices to reduce the average access latency, offload the network data traffic, and make base station (BS) more scalable [15][16]. Generally, this duplication is accomplished during off-peak hours when network resources are underutilized.

Device to device (D2D) communication is now adopted in Long Term Evolution-Advanced (LTE-A), which allows devices in proximity transmitting their data directly between each other without going through the base station (BS) [17]. Smart phones and tablets with significantly improved storage capacity can now be combined with D2D to unleash the ultimate potential of content caching. One of the first examples of content caching in D2D was presented in [18], where data transmission between femtocell and a user device was considered as D2D communication by assuming the femtocell as a device. The results in [18] showed that a non-vanishing throughput per node was achievable if there was sufficient content reuse. Two caching placement schemes, deterministic (central control) and random caching schemes were studied in [19]. The results showed that the spectral efficiency was improved by one to two orders of magnitude when the central control caching scheme was used in D2D communications. In [20] a fundamental conflict between collaboration distance and interference was considered. A closed form expression of the collaboration distance as a function of the model parameters was derived. The main focus of the study was to minimize the transmission power by optimizing the frequency reuse of the system. It was shown from above research that the key limitations of content caching with D2D communication will depend on

(1) the reliability of channel condition,

(2) whether the user’s cache was not updated properly, and

(3) if a user rejects to communicate with other users. These limitations resulted in highly outage probability.

In the next subsection, we present the system model for video content caching and delivery based on multiple devices to single device (MD2SD) communications for an environment where a high density of users appear, e.g. stadiums and shopping malls. Since this work is ongoing, in the future, we will provide the performance analysis for the outage probability for content delivery in D2D communications.


22


System model

Figure 7 System Model of the conten caching for multiple-device to single-device

Consider a cellular network which consists of a base station (BS) serving N user equipments (UEs). Figure 8 shows a single cell where the main BS is located in the centre and the UEs are modelled as homogeneous Poisson point process (PPP) with a density λ denoted by Φ , where { } 2

1 2, , , Nx x x RΦ = ⊂ , ix is the node location which are identically and independently

distributed (i.i.d) in the Euclidean plane and { }1,2, ,i N= . Each UE is assumed to have a

storage unit ( )Z known as the cache, which is used to store one complete file. The UE will be able to transmit data to another UE via D2D communication when requested. We assume single receiver device at the centre of the shadowing area supported by multiple transmitter devices at the same time inside the shadowing area with radius max( )r , where maxr is the maximum distance from the receiver location at the centre of the showing area to the farthest transmitter who has the desired parts of the file. Denote K as the total number of neighbour transmitters inside the shadowing area, which also follows the PPP distribution. Finally, our model in Figure 8 can be summarized as follows: when a UE requests the file which is stored in the vicinity of radius maxr , neighbours will serve the request via D2D links, otherwise the BS will serve the request. We can get the desired video from multiple devices and SINR can be guaranteed for successful video received. We assume there is no interference inside the


23


shadowing area and all received signals coming from outside the shadowing area is regarded as interference.

Assuming all users in the system have the same transmission power tP , then given the total

number of users K , the SINR at the receiver for this file from different transmitters is given by

2

12

K

tkk

k

K

h r PSINR

I

α

σ

−

==+

∑ (5)

where kr is the distance between the reference user and serving node k , 2

kh is the channel gain independent random variable which follows exponential distribution with unit mean and is denoted as

2 ~ exp(1)kh , krα− is the propagation attenuation between receive and

transmitter nodes,α is the path loss exponent, 2σ is the additive white Gaussian noise power, and I is the cumulative interference from simultaneous transmission of other cellular UEs outside the shadowing area. Since our model is assumed for high density of users and the interference is the summation of large number of independent components, we can model I as a Gaussian distribution with zero mean and variance 2

Iσ , i.e, 2(0, )IG σ based on central limit theorem [22].

In the future work, we will analyze the average outage performance. The outage probability is defined as the complement of success probability that the SINR in (5) exceeds a threshold thγ . Then, the average outage probability for the file is given by

outage th1

( ) (SINR < )Ki

P P K i P γ∞

=

= =∑ (6)

where ( )P K i= denotes the probability when the number of total users is equal to i .

2.3 D2D assisted content caching2

Introduction of D2D Assisted Content Dissemination

One important issue envisioned in D2D communications is how to design effective content offloading schemes [23]. Recently, there have been various D2D offloading schemes proposed in literature. For example, a hybrid cellular and opportunistic communications approach to distribute content chunks was proposed in [24], and an optimal D2D collaborating distance to maximize offloading traffic was proposed in [20]. Due to the limited energy budget in mobile users (MUs), the scheme in [25] focused on minimizing the energy consumption, and the multi-hop content dissemination scheme proposed in [26] considered the energy consumption fairness constraints.

2 The work of this part has been submitted to IEEE Globecom 2016.


24


The prior work overlooked the heterogeneous content preferences and sharing willingness, which resulted in the performance degradation in current offloading schemes [27][28]. To feature the heterogeneity in content preferences, a local popularity instead of global popularity was adopted in [82], and a more realistic model was proposed in [30], which employed the idea of keywords to describe the contents. However, content dissemination process was incomplete in these scheme as the energy required at placing the content is ignored. Furthermore, in the current offloading schemes design, the MUs were regarded to be ``altruistic" and full cooperation was assumed. Given MUs' selfish behaviors in D2D offloading, most researchers focused on the design of incentive schemes [31] [32] [33], which required extra control complexity or overhead for implementation.

Given these problems, we aim to minimize the energy consumption in a more realistic content dissemination process with the consideration of heterogeneous interests and sharing willingness among MUs. The content dissemination process consists of a pushing stage and a transmission stage with D2D offloading. According to different content preferences and D2D sharing willingness, MUs are categorized into groups. By modeling the MU groups as homogeneous Poisson point processes (HPPPs) [34], the total energy consumption is formulated as a function of pushing strategy. We prove that the optimal solution is on the feasible region boundary. However, the complexity to search all the boundary points is exponentially high. To reduce the complexity, an iterative algorithm is proposed, and the convergence is verified. Furthermore, the optimal pushing strategy for the special case of two groups is derived in closed-form. Finally, simulation results show that the iterative algorithm can converge to the optimal solution, but with much reduced complexity. In addition, the proposed algorithm can save much energy compared with the benchmark of all pushing case and no pushing case.

System Model

As shown in Figure 8, we consider a cellular network where MUs can share cached contents with each other by D2D communications. The cellular transmission range and D2D transmission range are denoted by R and r, respectively. For simplicity, we assume that no interference exists among two D2D links in vicinity, and such communications will not interfere with cellular transmissions.


25


R

r

Keyword 2 subscriber without pushingKeyword 1 subscriber with pushingKeyword 2 subscriber with pushing

Keyword 1 subscriber without pushing

Cellular Connection D2D ConnectionCellular Connection RangeR D2D Connection Ranger

user2

user1

Figure 8: An example of a content with two keywords

For simplicity, we only consider one content, which is described by M keywords in set ℳ ={1,2,⋯ ,𝑀𝑀} and the associated weights, which indicate the importance of keywords. To illustrate the idea of keywords and weights, we take the “Harry Potter” movie as an example, which involves the keywords such as “mystery”, “thriller”, “adventure”, “horror” and “romance”. The weight of keyword “romance” is much less than the weight associated with “mystery”, since the movie is mainly about the world of witchcraft.

To feature the different interests of MUs, we assume that each MU subscribes keywords in BS to receive the interested content items. MUs may be interested in more than one keyword, but they are only allowed to choose one with the highest priority. As a result, MUs are classified into disjoint M groups, and the group of MUs subscribed keyword m is denoted by 𝒢𝒢𝑚𝑚. The locations of MUs in group 𝒢𝒢𝑚𝑚 follow a two-dimensional HPPPs with spatial density 𝜆𝜆𝑚𝑚, which is independent from the other groups. In general, the popularity of a content for group Gm is closely related to the weight of keyword m. When the subscribed keyword has a larger weight in this content, there is a higher probability that MUs in this group will be interested in it. The probability that a MU in group Gm is interested in this content is denoted by 𝑤𝑤𝑚𝑚, and 0 ≤ 𝑤𝑤𝑚𝑚 ≤1.


26


BS push a content to MU

An unlucky MU request a content

Yes Request it

No

Download from BS D2D transmission

In the nearby MU’s cache?

Want to share?

Yes

No

Content pushing

Content transmission

Lucky Unlucky Indifferent Normal

No Pushing

Interested

Pushed

No interests

MU State

Interested?

Lucky MUDownload it

Indifferent MURefuse Pushing

Yes

wmcm

∑wmcmρm1-∑wmcm ∑wmcm(1-ρm)

(1-wm)cm

∑ wmcm

No

Figure 9: The flow chart of content dissemination protocol

Another key feature is the willingness of a MU to take part in the D2D communication. Generally, MUs who have subscribed the same keyword share similar characteristics, so that the MUs in the same group are assumed to have same probability to share content with each other. Specifically, the sharing probability for MUs in group 𝒢𝒢𝑚𝑚 is denoted by 𝜌𝜌𝑚𝑚.

Figure 8 shows that a content with two keywords is equally placed in two related groups, where user 1 can get this content via a D2D link, while user 2 has to download it from BS as no helpers are within r. It can be inferred that the condition for D2D offloading is that a subset of MUs initially have been pushed with the content. For a given content, the protocol of content dissemination and the probabilities of events are specified in Figure 10. As shown in Figure 10, the dissemination process consists of two stages; namely, content pushing and content transmission. The details of each stage are illustrated in the following relevant sections.

A. Content Pushing

In this stage, BS pushes a content to MUs who have subscribed the related keywords. As shown in Figure 8, there are four states for MUs in content pushing stage. According to Figure 8, the Lucky MUs will receive content via pushing, and the Indifferent MUs will refuse pushing. The reason is that a MU who has received a content pushing does not necessarily want it, especially when the subscribed keyword has a small weight. Consequently, there are some unlucky Mus who have not received a pushing but they have interests in this content. Finally, some MUs may not be involved in this content dissemination process, and they are called Normal MUs.


27


For a particular content, the pushing probability for MUs in group 𝒢𝒢𝑚𝑚 is denoted by cm. Given the content popularity 𝑤𝑤𝑚𝑚, in group Gm, under the HPPP model, the average number of Lucky MUs from Gm in the cell range is

2( )l m m mN m R w cπ λ= (7)

The average energy consumption of receiving the content from BS for each MU is assumed to be equal, which is denoted as 𝐸𝐸𝑐𝑐, and the average energy consumption at content pushing stage is

2p c l m m m c

m m

E E N (m) πR λ w c E∈ ∈

= =∑ ∑M M

(8)

B. Content Transmission

As shown in Figure 10, in content transmission stage, Unlucky MUs make requests for the content to D2D helpers and BS. Given the pushing strategy 𝒄𝒄 = [𝑐𝑐1, 𝑐𝑐2,⋯ , 𝑐𝑐𝑀𝑀] due to the sharing probability 𝜌𝜌𝑚𝑚 in different groups, the average number of MUs from all the groups who will join in the content offloading within D2D communication range r is

2 .f m m m mm

C r w cπ λ ρ∈

= ∑M

(9)

Note that 𝐶𝐶𝑓𝑓 can also be interpreted as the average availability of this content in D2D offloading. Given the HPPP distribution, the probability that there are n MUs in area A is

( ),!

n AeP n An

λλ −

= (10)

where 𝜆𝜆 is the density of MUs in unit area. In our case, 𝜆𝜆 = ∑ 𝜆𝜆𝑚𝑚𝑤𝑤𝑚𝑚𝜌𝜌𝑚𝑚𝑐𝑐𝑚𝑚 𝑚𝑚∈ℳ and 𝐴𝐴 = 𝜋𝜋𝑟𝑟2. As a result. As a result, the probability that an Unlucky MU can download the content by D2D offloading is

21 (0, ) 1 fCd P r eπ −= − = −P (11)

From the HPPP distribution, the average number of Unlucky MUs from in the cell range is

2( ) (1 )u m m mN m R w cπ λ= − (12)

As a result, the average energy consumption in content transmission stage is

((1 ) ) ( )t d c d d um

E E E N m∈

= − +∑M

P P (13)

where 𝐸𝐸𝑑𝑑 is the average energy consumption of D2D links, and 𝐸𝐸𝑑𝑑 ≪ 𝐸𝐸𝑐𝑐 due to short transmission distance.

Remark: From the content dissemination protocol, we can infer that the energy consumption relies on the pushing strategy. In order to show it more clearly, we consider two extreme cases; namely, all pushing case and no pushing case. In the all pushing case, BS pushes the content to all MUs who have subscribed the relevant keywords, but only interested MUs will download it. While no pushing effort is made by BS in no pushing case, and only interested MUs will request it from BS. Obviously, the energy consumptions in these two cases are equal, where all the energy is consumed by means of cellular transmission, and D2D transmissions are not utilized to reduce the energy consumption. As a result, a good content push strategy should take full advantage of the D2D transmission to minimize the total system energy consumption.


28


Problem Formulation

Based on the energy consumption for each stage in Section II, we aim to optimize the pushing probability for each group 𝒢𝒢𝑚𝑚 to minimize the total energy consumption. The optimal problem is formulated as

total1: min

s.t. 0 1,m

p cc

m

E E E

c m

= +

≤ ≤ ∀ ∈

P

M (14)

Substituting equation (8) and equation (13) into the objective function of Problem (14), the total energy consumption is rewritten as

2 Δ (1 )fCtotal m m c m m m

m m

E R w E w c E eπ λ λ −

∈ ∈

= − −∑ ∑M M

(15)

where Δ𝐸𝐸 = 𝐸𝐸𝑐𝑐 − 𝐸𝐸𝑑𝑑 , 𝑐𝑐𝑚𝑚 = 1 − 𝑐𝑐𝑚𝑚. The first summation term in is a fixed value, which can be regarded as the energy consumption in all pushing case (same as no pushing case), and the variables 𝑐𝑐𝑚𝑚 are only involved in the second summation term, which represents the energy saving by D2D offloading. As a result, 𝒫𝒫1 can be simplified to the following equivalent problem 𝒫𝒫2, where the energy saving by D2D offloading is maximized.

( )1

2 : max 1

s.t. 0 1,

m m mm

m

B t c

s m mc m

m

E t c e

c m

ρ∈

− −

∈

∑ = −

≤ ≤ ∀ ∈

∑ M

MP

M (16)

where 𝑡𝑡𝑚𝑚 = 𝜆𝜆𝑚𝑚𝑤𝑤𝑚𝑚 , which can be regarded as the average number of MUs in unit area from group 𝒢𝒢𝑚𝑚 attracted by this content, and B represents the D2D cooperation area 𝜋𝜋𝑟𝑟2. It is easy to verify that problem 𝒫𝒫2 is non-convex, which is difficult to solve.

Given the optimal solutions 𝑐𝑐𝑚𝑚∗ and 𝐸𝐸𝑠𝑠∗ of 𝒫𝒫2, the optimal solutions 𝑐𝑐𝑚𝑚∗ and 𝐸𝐸𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡∗ of 𝒫𝒫1 can be

easily obtained by

* *1m mc c= − (17)

* *Δtotal all sE E EE= − (18)

s.t. 0 1,mc m≤ ≤ ∀ ∈M (19)

where 𝐸𝐸𝑡𝑡𝑡𝑡𝑡𝑡 = � 𝜋𝜋𝑅𝑅2𝜆𝜆𝑚𝑚𝑤𝑤𝑚𝑚𝐸𝐸𝑐𝑐𝑚𝑚∈ℳ . For the convenience of analysis, we will focus on solving 𝒫𝒫2 in the following part.

Solution and Algorithms

Although Problem 𝒫𝒫2 is non-convex, in the following theorem, we prove that the optimal solution must be on the feasible region boundary.

Theorem 1: If sharing probability 𝜌𝜌𝑚𝑚 of 𝒢𝒢𝑚𝑚 is different from each other, the optimal solution 𝒄𝒄∗ = �𝑐𝑐1

∗, 𝑐𝑐2∗ ,⋯ , 𝑐𝑐𝑀𝑀

∗ � is on the feasible region boundary. Specifically, only one group has the pushing probability that 0 ≤ 𝑐𝑐𝑚𝑚

∗ ≤ 1 and for all 𝑗𝑗 ≠ 𝑚𝑚, 𝑐𝑐𝑗𝑗∗ = 0 or 𝑐𝑐𝑗𝑗

∗ = 1.

Although Theorem 1 shows that the optimal solution is on the boundary, the complexity of searching all the boundary points is exponentially high. In the following subsection, we develop


29


an iterative algorithm with much reduced complexity. However, the simulation results show that this algorithm achieves the optimal solution.

1) Algorithm design for M groups

For each group m , if the pushing strategy of other groups j m∀ ≠ are given, the optimal solution can be achieved. Based on this, we adopt the alternate optimization algorithm to solve problem 𝒫𝒫2. The details of the algorithm is given in Algorithm 1. In the following, we derive the optimal solutions for each group, given the pushing strategy of the other groups.

The second derivative of 𝐸𝐸𝑠𝑠(c) in 𝒫𝒫2 with respect to cm is

∂2Es(c)∂cm

2 = −Btm2 ρm(2tmB + St)Bρm)eBSc < 0

where St = ∑ tmcmMm=1 , Sc = � tmρm(1− cm).m∈ℳ As a result, we can claim that Es �cm�C

~m�

is concave with respect to cm, where 1 1 1[ , , , , , ].m m m Mc c c c− +=C

By taking the derivative of 𝐸𝐸𝑠𝑠(𝒄𝒄) with respect to 𝑐𝑐𝑚𝑚, we have the stationary point 𝑐𝑐~𝑚𝑚 at ∂𝐸𝐸𝑠𝑠

∂𝑐𝑐𝑚𝑚=

0 as

𝑐𝑐~𝑚𝑚 = 𝑓𝑓𝑚𝑚(𝑐𝑐1,⋯ , 𝑐𝑐𝑚𝑚−1, 𝑐𝑐𝑚𝑚+1⋯ , 𝑐𝑐𝑀𝑀)

=1

𝐵𝐵𝜌𝜌𝑚𝑚𝑡𝑡𝑚𝑚(𝒲𝒲(𝑒𝑒𝐵𝐵𝑝𝑝𝑡𝑡+𝐵𝐵𝑞𝑞𝑐𝑐+1) − 𝐵𝐵𝜌𝜌𝑚𝑚𝑝𝑝𝑚𝑚 − 1)

where

1, 1,1, ( ),

M MM

t m m c j j m j m j jm j m j mj j

p t q t c p t cρ ρ ρ= =≠= ≠

= = − =∑ ∑ ∑

Obviously, the objective value of 𝒫𝒫2 in terms of energy saving is monotonically increasing after optimizing each 𝑐𝑐𝑚𝑚, which means the total energy consumption is decreasing after each step in each iteration. As a result, the convergence of the proposed algorithm is guaranteed.

For our proposed Algorithm, the computation in each iteration is composed of closed-form expressions. Thus, the total computation complexity is 𝒪𝒪(𝐼𝐼𝑀𝑀) where I is the total number of iterations. However, the complexity of searching all the feasible region boundary points is 𝒪𝒪(𝑀𝑀2𝑀𝑀−1).

Table 2 A Low-complexity Iterative Algorithm for M group

Algorithm 1 Iterative Algorithm for general number of groups

Input: MUs densities mλ ; Keywords weights mw ;

Desire precision δ ; D2D cooperation area B ;

sharing willingness probabilities mρ ;


30


1. Initialize 0 0, m m= =c 0 V 0 ; 2. In iteration 1,2,i = 3. Repeat for group 1,2,m M= 4. 1 1

1 1 1[ , , , , , ]i i i i im m m Mc c c c− −

− +=V 5. Update i

mc 6. IF 1i

mc > 7. 1i

mc = ; 8. ELSIF 0i

mc < 9. 0i

mc = ; 10. ELSE 11. i i

m mc c= ; 12. ENDIF 13. Update i

mc , ( )i is mE c ;

14. Until 1| ( ) ( ) |i i i is m s mE E δ−− <c c and i

mc is on boundary; 15. RETURN { ( )i i

s mE c , imc };

Furthermore, we have derived the closed-form optimal solution for the special case of two groups, which is shown as follows.

2) Optimal Pushing strategy of Two Groups Case A. The same sharing probability

If the sharing probabilities are equal, i.e. 𝜌𝜌1 = 𝜌𝜌2 = 𝜌𝜌, we define a new group 0 so that

0 1 2 0 0 1 1 2 2, .t t t t c t c t c= + = + (20)

Substituting (20) into the objective function of problem 𝒫𝒫2 yields

max𝑐𝑐0

𝑡𝑡0𝑐𝑐0(1− e−𝐵𝐵𝑡𝑡0𝜌𝜌(1−𝑐𝑐0))

s.t.0 ≤ 𝑐𝑐0 ≤ 1

It is easy to verify that the objective function is concave, and the optimal solution 𝑐𝑐0∗ is obtained

as

𝑐𝑐0∗ = [

1𝐵𝐵𝜌𝜌𝑡𝑡0

(𝒲𝒲(e−𝐵𝐵𝜌𝜌𝑡𝑡0+1)− 1)]01

where 𝒲𝒲 is the Lambert-W function, and [𝑥𝑥]01 represents that if 𝑥𝑥 <0, its value is 0, if 𝑥𝑥 >1, its value is 1, and if 0 ≤ 𝑥𝑥 ≤ 1,its value is 𝑥𝑥.

According to (20), there may exist multiple pairs of �𝑐𝑐1∗, 𝑐𝑐2

∗� that satisfy the following condition.

𝑡𝑡1𝑐𝑐1∗ + 𝑡𝑡2𝑐𝑐2

∗ = 𝑐𝑐0∗𝑡𝑡0

In this case, the optimal solution for 𝒫𝒫2 is not unique.

B. Different sharing probabilities

Define two functions


31


𝑓𝑓1(𝑐𝑐2) =1

𝐵𝐵𝜌𝜌1𝑡𝑡1(𝒲𝒲(exp(𝑠𝑠1 + 𝑐𝑐2Δ1)− 𝐵𝐵𝜌𝜌1𝑡𝑡2𝑐𝑐2 − 1),

𝑓𝑓2(𝑐𝑐1) =1

𝐵𝐵𝜌𝜌2𝑡𝑡2(𝒲𝒲(exp(𝑠𝑠1 + Δ2𝑐𝑐1)) −𝐵𝐵𝜌𝜌2𝑡𝑡1𝑐𝑐1 − 1),

where

𝑠𝑠1 = 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 + 𝐵𝐵𝑡𝑡2𝜌𝜌2,Δ1 = 𝐵𝐵𝑡𝑡2(𝜌𝜌1 − 𝜌𝜌2)),Δ2 = 𝐵𝐵𝑡𝑡1(𝜌𝜌2 − 𝜌𝜌1).

Based on Theorem 1, the globally optimal solution is chosen as the best among all locally optimal solutions for the following four cases. The locally optimal solutions are characterized as follows:

• Case-1: 0 ≤ 𝑐𝑐1∗ ≤ 1, 𝑐𝑐2

∗ = 0 If the condition that 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 ≥ 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2 is satisfied, the optimal solution is given by

𝑐𝑐1∗ = 𝑓𝑓1(0), 𝑐𝑐2

∗ = 0. If 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 < 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2, then

𝑐𝑐1∗ = 1, 𝑐𝑐2

∗ = 0.

• Case-2: 0 ≤ 𝑐𝑐1∗ ≤ 1, 𝑐𝑐2

∗ = 1 If the condition that 1 + 𝐵𝐵𝑡𝑡2𝜌𝜌1 ≤ 𝑒𝑒𝐵𝐵𝑡𝑡1𝜌𝜌1 is satisfied, the optimal solution is given by

𝑐𝑐1∗ = 𝑓𝑓1(1), 𝑐𝑐2

∗ = 1. If 1 + 𝐵𝐵𝑡𝑡2𝜌𝜌1 > 𝑒𝑒𝐵𝐵𝑡𝑡1𝜌𝜌1, then

𝑐𝑐1∗ = 0, 𝑐𝑐2

∗ = 1.

• Case-3: 0 ≤ 𝑐𝑐2∗ ≤ 1, 𝑐𝑐1

∗ = 0 If the condition that 1 + 𝐵𝐵𝑡𝑡2𝜌𝜌2 ≥ 𝑒𝑒𝐵𝐵𝑡𝑡1𝜌𝜌1 is satisfied, the optimal solution is given by

𝑐𝑐1∗ = 0, 𝑐𝑐2

∗ = 𝑓𝑓2(0). If1 + 𝐵𝐵𝑡𝑡2𝜌𝜌2 < 𝑒𝑒𝐵𝐵𝑡𝑡1𝜌𝜌1, then

𝑐𝑐1∗ = 0, 𝑐𝑐2

∗ = 1.

• Case-4: 0 ≤ 𝑐𝑐2∗ ≤ 1, 𝑐𝑐1

∗ = 1 If the condition that 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌2 ≤ 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2 is satisfied, the optimal solution is given by

𝑐𝑐1∗ = 1, 𝑐𝑐2

∗ = 𝑓𝑓2(1). If 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌2 > 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2, then

𝑐𝑐1∗ = 1, 𝑐𝑐2

∗ = 0.

In this section, we only present the derivation of Case-1 as illustration, since other cases follow the same procedure. In Case-1, substituting 𝑐𝑐2 = 0 into the objective function of problem 𝒫𝒫2 yields

𝑚𝑚𝑚𝑚𝑥𝑥𝑐𝑐1

𝐸𝐸𝑠𝑠 = 𝑡𝑡1𝑐𝑐1(1− e−𝐵𝐵𝑡𝑡2𝜌𝜌2−𝐵𝐵𝑡𝑡1𝜌𝜌1(1−𝑐𝑐1))

s.t. 0 ≤ 𝑐𝑐1 ≤ 1


32


It is easy to know that the objective function is concave, so that the optimal solution is obtained as

𝑐𝑐1∗ = [𝑓𝑓1(0)]01

= [1

𝐵𝐵𝜌𝜌1𝑡𝑡1(𝒲𝒲(exp(1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 + 𝐵𝐵𝑡𝑡2𝜌𝜌2)− 1))]01 .

Denote function 𝑢𝑢(𝑥𝑥) = 𝒲𝒲(𝑒𝑒𝑥𝑥+𝑡𝑡)− 𝑥𝑥 . We have 𝑢𝑢′(𝑥𝑥) = −(𝒲𝒲(𝑒𝑒𝑥𝑥+𝑡𝑡) + 1)−1 < 0 , ∀𝑥𝑥 > 0 . Thus, function 𝑢𝑢(𝑥𝑥) is strictly decreasing function and 𝑢𝑢(𝑥𝑥) ≤ 𝑢𝑢(𝑒𝑒𝑡𝑡) = 0,∀𝑥𝑥 ≥ 𝑒𝑒𝑡𝑡. As a result, when 𝑓𝑓1(0) ≤ 1, the following condition can be obtained.

1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 ≥ 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2

Since exp(𝐵𝐵𝑡𝑡1𝜌𝜌1 + 𝐵𝐵𝑡𝑡2𝜌𝜌2) > 1, the following equality always holds.

𝑓𝑓1(0) =1

𝐵𝐵𝜌𝜌1𝑡𝑡1(𝒲𝒲(exp(1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 + 𝐵𝐵𝑡𝑡2𝜌𝜌2)− 1)) > 0

Overall, if 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 ≥ 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2, 𝑐𝑐1∗ = 𝑓𝑓1(0), and if 1 + 𝐵𝐵𝑡𝑡1𝜌𝜌1 < 𝑒𝑒𝐵𝐵𝑡𝑡2𝜌𝜌2 , 𝑐𝑐1

∗ = 1.

Numerical Results

Table 3: General simulation parameters

Parameters Values

Content size 10 M

Cellular Down-link rate 1 Mbps

D2D rate 10 Mbps

Energy consumed via cellular downlink 1.8 Joules/sec

Energy consumed in D2D transmitter 1.425 Joules/sec

Energy consumed in D2D receiver 0.925 Joules/sec

Cell range 𝜋𝜋𝑅𝑅2 1 km2

D2D communications range 𝜋𝜋𝑟𝑟2 50 m2

user density of each group 𝜆𝜆𝑚𝑚 0.2 MU per unit area

In this section, we illustrate the numerical results for evaluating the proposed iterative algorithm performance. All the optimal solutions are achieved by searching the feasible region boundary except the case of two groups. The general system parameters are listed in Table 2, where the energy consumption values are obtained according to [35].


33


Figure 10: Energy consumption for different popularities with M=20

Figure 11 and Figure 12 shows the energy consumption performance versus the number of iterations for different popularities and different sharing probabilities, respectively. To give an insightful observation, the energy consumption of all pushing case (same as no pushing case) is also presented in the figure marked as 𝐸𝐸𝑡𝑡𝑡𝑡𝑡𝑡 . The content popularities and sharing probabilities are drawn from the standard uniform distribution on the intervals [0,2𝑤𝑤𝑡𝑡] and [0,2𝜌𝜌𝑡𝑡], respectively.

In Figure 11, the average sharing probability 𝜌𝜌𝑡𝑡=0.5, while the average popularity 𝑤𝑤𝑡𝑡=0.5 in Figure 12. From Figure 11 and Figure 12, we can see that the energy consumption of proposed algorithm is monotonically decreasing during the initial iterative procedure, and 8 iterations are sufficient for all the considered cases to converge to the optimal value. In addition, Figure 11 shows that increasing the average popularity leads to an increase of the total power consumption, where the proposed algorithm shows larger power savings compared with the all pushing case. At the same time, Figure 12 shows that the power consumption monotonically decreases for proposed algorithm with the increase of average sharing probability due to the growth of D2D offloading. Moreover, when the average sharing probability is quite small (5%), the proposed algorithm still can provide a considerable (60%) energy reduction.

0 1 2 3 4 5 6 70

0.5

1

1.5

2

2.5

3

3.5

4

4.5x 10

7

Iterations

Ene

rgy

cons

umpt

ion

(Jou

les)

Eall wa=0.2

Epro wa=0.2

Eopt wa=0.2

Eall wa=0.3

Epro wa=0.3

Eopt wa=0.3

Eall wa=0.5

Epro wa=0.5

Eopt wa=0.5


34


Figure 11: Energy consumption for different sharing probabilities with M=20

Figure 12: Energy consumption performance under different similarities between 𝑤𝑤𝑚𝑚 and 𝜌𝜌𝑚𝑚 with M=6

To study the impact of heterogeneity in preferences and sharing willingness, we adopt the normalized probabilities 𝑤𝑤𝑚𝑚 and 𝜌𝜌𝑚𝑚 to describe the popularity and sharing willingness in the following part. Figure 13 shows the energy consumption performance under different similarities when sharing probabilities and popularities are modeled as Zipf distributions [36]. The Zipf exponent of sharing willingness is denoted by 𝛾𝛾𝜌𝜌. Intuitively, a larger 𝛾𝛾𝜌𝜌 indicates a higher sharing reuse scenario, where the majority of sharing activities are accounted for the first a few groups. Thus, the similarity between 𝑤𝑤𝑚𝑚 and 𝜌𝜌𝑚𝑚 are defined as

sim(𝝆𝝆,𝒘𝒘) =� 𝑤𝑤𝑚𝑚 𝜌𝜌𝑚𝑚

𝑀𝑀

𝑚𝑚=1

�� 𝜌𝜌𝑚𝑚2𝑀𝑀

𝑚𝑚=1�� 𝑤𝑤𝑚𝑚

2𝑀𝑀

𝑚𝑚=1

.

0 1 2 3 4 5 6 7 8 90.5

1

1.5

2

2.5

3

3.5

4x 10

7

Iterations

Ene

rgy

cons

umpt

ion

(Jou

les)

Eall

Epro ρa=0.05

Eopt ρa=0.05

Epro ρa=0.1

Eopt ρa=0.1

Epro ρa=0.3

Eopt ρa=0.3

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 12

2.2

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8x 10

6

Similarity between keywords weights and sharing will

Ene

rgy

cons

umpt

ion

(Jou

les)

Epro γρ=0.3

Epro γρ=0.8

Epro γρ=1.3

Eall


35


As expected, it can be seen that the energy consumption is decreasing with the growth of similarity, while a larger 𝛾𝛾𝜌𝜌 achieves a lower energy consumption under the same similarity. It means the D2D offloading system can benefit from the heterogeneous MUs' interests and sharing willingness rather than regard them as drawbacks.

Conclusion

This subsection studied the energy efficient pushing strategy to minimize the total energy consumption in content dissemination, where MUs were categorized into groups by different popularities and sharing probabilities. First, the optimal pushing strategy was proved to be on the feasible region boundary. However, the complexity of searching all the boundary points are exponentially high, so that we proposed an iterative algorithm to reduce the computation complexity. Next, the optimal pushing strategy for the special case of two groups was derived in closed-form. Finally, by using the numerical examples, we have verified the superiority of our proposed algorithm to the benchmarks of all pushing case and no pushing case.

2.4 Security issue in D2D communications

D2D communication potentially allows two devices to communicate without a base station (BS) involvement. Even if the BS is involved in the communication, its involvement is limited. This form of cooperative communication is expected to be implemented in the 5G ecosystem, and therefore presents a security threat since user data might be routed through other UEs. A possible solution could be to establish a trusted environment (closed access) for devices operating under D2D.

In general, the security-based implementations for 5G must be lightweight with minimal overheads, as this might have an effect on quality of service and seamless mobility. For example, a model was proposed for 4G networks, where argument was presented for a separation between the control and data planes. The control plane in this case performs functions relating to QoS signalling and security [37][38].

D2D communications are based on wireless broadcast and hence the communication is considered vulnerable to a variety of attacks. Among others the most popular attacks include message modification, node impersonation, surreptitious eavesdropping and man-in-the middle attack. An attacker that is passively listening to a communication between two devices can gain critical or privacy information such as identity related information of critical information or trade secrets.

As a result, security is one of the major concerns for D2D communications. One way for enhancing the security between two end users of a D2D link is to establish a shared secrete key. Diffie-Hellman key exchange is a well-known method of securely exchanging cryptographic keys between two parties (that have no prior knowledge of each other) over a public (insecure) channel. This jointly established shared secret key can then be used to


36


encrypt subsequent communications using a symmetric key cipher. Although Diffie-Hellman key agreement is a non-authenticated key-agreement protocol, it can be considered as the basis for a variety of authenticated protocols. In D2D when two parties (A and B) would like to agree on a shared secret key [39]. Assume p and g are large values, that are already shared between Party A and Party B. The process follows the following steps:

• Party A and Party B generate their private keys a and b respectively. • Party A calculates A = ga mod p and sends it to Party B. • Party B calculates B = gb mod p and sends it to Party A. • Party A calculates s = Ba mod p • Party B calculates s = Ab mod p • Party A and B now they share the same secrete key s

This is based on the following mod operations Ab mod p = gab mod p = gba mod p = Ba mod p => (ga mod p)b mod p = (gb mod p)a mod p

However, even with the Diffie-Hellman key agreement protocol, the D2D link is still vulnerable to well-known man-in-the middle attack. This attack could be implemented with an active adversary who can make independent connections with both parties. The adversary initiates individual communications with the two parties and makes them believe that they are talking directly to each other. In the future work, we will study the methods to deal with this attack.

2.5 60 GHz Networking for Low-Latency & High-Throughput

In iCIRRUS D4.1, we have already successfully demonstrated D2D connectivity up to 20 metres distance, with a maximum throughput of 2.4Gbps (full-duplex). Ad-hoc mesh networking between multiple devices, and mobile distributed caching (MDC) has also been achieved up to Gigabit speeds. However, up to now, our solution has only provided point-to-point connectivity between a Dell laptop (TX) and the docking station (RX).

With regard to D2I (i.e. connectivity to fixed infrastructure), this has not been demonstrated by us up to now. As such, in that context, we have recently been focusing on how to integrate the 60-GHz (802.11ad) system into the overall iCIRRUS network architecture. To that end, Figure 14 below shows a likely scenario that we want to demonstrate, with a 60-GHz wireless link between the RRH and Base Station. Wireless D2I functionality (i.e. tablets/smartphones to the RRH) is demonstrated, with wireless (60 GHz) fronthaul functionality also enabled.


37


Figure 13 A scenario with a 60-GHz wireless link between the RRH and Base Station

To verify the above architecture, a test bed has been constructed at UEssex where the RRH node (laptop with RF modem) receives data from a device (in this case, a server via the Gigabit Ethernet port) and retransmits this wirelessly upstream over 60-GHz to the Base Station (docking station). The packet delay between the RRH and Base Station was measured with 0.2 ms. . Measurements of the overall end-to-end latency from the UE to the Cloud Server are also possible, i.e. the latency can be observed between any two points using the iPerf, WireShark or similar tools. Multi-gigabit wireless speeds are also possible for the various network links by link aggregation (i.e. multiple RRHs to Base Station connections, and also aggregation of the BS links, as indicated). This gives the flexibility of either having aggregated bandwidth or redundancy in the various D2D, D2I, and fronthaul scenarios.

3 Semi-distributed Resource Management for D2D communication 3

D2D communication in ICIRRUS underlaying a cellular network allows two devices in proximity to set up direct transmission link, so the data can be transmitted through direct link instead of going through the base station (BS). Unlike the conventional local communications supporting direct communication networks (e.g. WiFi, BlueTooth, etc.) are limited by user-based authentication mechanism and the severe communication environment of unlicensed spectrum. In contrast, the cellular D2D communication allows two users in proximity to form a D2D pair and communicate with each other directly by using the licensed spectrum under the assistance of the BS. Therefore, the quality of service (QoS) in the cellular D2D communication could be guaranteed as well as total system throughput

3 Part of this work has been sumbitted to IEEE Globecom 2016.


38


In order to exploit the advantage of D2D communications, resource allocation is very important in system design. However, the D2D resource management is difficult to be fully centralized, since centralized BBU hotel needs to collect not only the channel state information (CSI) from all active D2D pairs, but also from any two devices if frequency reused is allowed, which would inflate the control signaling and computation requirements at BBU hotel. Furthermore, the fronthaul latency of the transmission packets under C-RAN architecture would cost inaccuracy of both CSI transmission as well as resource allocation results feedback, therefore, deduct the benefit of centralized resource management. In this chapter, a semi-distributed resource management scheme will be detailed explained. The control signalling protocols, CSI transmission and fronthaul latency issues are investigated in Chapter 4.

For distributed resource allocation, the active users will form the D2D links if the communication meets the proximity requirement and there is a traffic demand for which D2D communication is suited. The network assigns a generic frequency allocation for D2D use and D2D devices select individual resource in an autonomous way. The network has no control on the resource utilization of specific D2D links, and thus the intra-cell interference can only be managed in a statistical way resulting in non-optimal system performance [40].

Therefore, as mentioned in deliverable 4.1, semi-distributed resource management is proposed in ICIRRUS under the C-RAN architecture. It is noticed that in a small area surrounding a D2D link, the allocated spectrum should not be reused at same time to avoid strong co-channel interference. Thus, the active D2D users within each cell can be separated into different groups according to their geographical locations and reuse the spectrum based on the frequency reuse pattern among different groups. As shown in Figure 15, different from the traditional network layout, in a C-RAN architecture, multiple RRHs are deployed within the coverage of one urban macro-cell. For each cellular user, one RRH will typically be used for accessing the network. Thus, the location of the users could be perceived by the RRH identification and can be used to group the D2D users.

The semi-distributed resource management for D2D communication under C-RAN architecture is separated into two layers as shown in Figure 16.

The top layer is performed at BBU hotel, which will focused on the spectrum assignment for each RRH based on the feedback from each RRH. Since the feedback from each RRH to BBU Hotel via fronthaul link will affected by the transmission delay, some statistical system parameters are transmitted instead of instantaneous CSI and resource allocation results. The feedback parameters include the number of permitted D2D pairs within the coverage of the RRH and the maximum transmit power of each D2D device.


39


O1

O2

O6

O5

O4

O3

Transmitter of a D2D pair

Receiver of a D2D pair

Base Station

O0

(b) C-RAN architecture

O0

RRH

(a) traditional cellular architecture

D2D data link

Fronthaul link

Figure 14 System architectures for cellular assist D2D communication


40


The bottom layer is performed at each RRH, where local resource management is performed for the D2D communications under its coverage by using the assigned spectrum from BBU hotel at top layer. In order to avoid strong interference from the nearby D2D pairs, in the coverage of one RRH, there is no spectrum reuse among the D2D pairs. In other words, the spectrum will be assigned to each D2D pair orthogonally according to the CSI feedback.

3.1 Local resource management schemes

For a reference RRH, it is assumed that the assigned spectrum from BBU hotel could be separated into N subcarriers, and there are K D2D pairs within its coverage. Since the local resource management is performed by RRH, and intra-cell interference between different RRHs are considered in top layer when BBU Hotel perform spectrum assignment. Therefore, at each RRH, the intra-cell interference is considered as a statistical value and omitted for the sake of simplicity.

The resource allocation problem of maximizing the achievable capacity of the reference RRH can be formulated as

Figure 15 Block diagram for semi-distributed D2D resource allocation in C-RAN architechture

)1log(max1 1

2

2,

2,

,∑∑= =

−⋅⋅+⋅⋅=

K

k

N

n

aknknk

nk

dhPwBR

σ

Top layer: Central control for revenue frequency sharing factor and pattern design (frequency reuse factor, number of subcells)

Bottom layer: Local resource allocation for D2D communication offloading

Opt. of RRH 1(Number of D2D pairs k1,

transmit power)

Opt. of RRH M(Number of D2D pairs km,

transmit power)

D2D {1,1} D2D {1,2} D2D {1,k1}

Grouping RRHs and Subcarriers {C1...Cn}

{km}{SCc1} Signalling overhead and delay

BBU Pool


41


s.t. },...,1{},,...,1{},1,0{, NnKkw nk ∈∈∀∈

∑=

∈∀≤N

nnk Kkw

1, },...,1{,1

∑=

∈∀≤K

knk NnPP

1max, },...,1{,

Here, the wk,n denotes the subcarrier indicator, which wk,n=1 indicates subcarrier n is assigned to D2D pair K. Pmax indicates the maximum transmit power of each D2D transmitter.

Best subcarrier CSI (BSCR)-based resource management

Greedy resource allocation algorithm is commonly used to maximize the total system throughput in OFDMA downlink resource allocation, in which BS is the transmitter for the downlink signals, thus only the total power constraint need to be satisfied . For subcarrier allocation, one subcarrier should be assigned to the D2D pair which have best CSI on this subcarrier. In other words, the subcarrier should be allocated to the D2D pair which have best subcarrier achievable data-rate and is only related with CSI. And then, global water-filling algorithm is used to assign the power to each subcarrier.

As each device will have a unique maximum power constraint, global water-filling power allocation algorithm is no longer suitable. Instead, local water-filling power allocation is used to maximize the data-rate can be achieved for individual D2D link. Therefore, this RM method is called best subcarrier CSI (BSCR)-based resource allocation algorithm.

Subcarrier achievable data-rate (SAR)-based resource management

It is note that, for one D2D pair D2Dk, the number of subcarriers allocated, Nk, is a variable. Please note, if the total transmit power is given for D2Dk, the power level after performing water-filling for D2Dk will decrease with increasing Nk. Figure 17 illustrates how the power allocation (PA)and power level (PL) are changed when Nk = 9 and Nk = 10 subcarriers are allocated to D2Dk. When the total power level for all the subcarriers allocated to D2Dk decreases, the total data-rate for D2Dk may decreases. For the BSCR- based resource allocation scheme, the number of subcarriers allocated to one D2D pair is ignored, which may cause the problem of allocating quite large number of subcarriers to one D2D pair, thus degrades the system throughput. Therefore, with the consideration of simplifying the global searching method and addressing the power level changes caused by the change of number of subcarriers allocated to one D2D pair, a two-step sub-optimal resource allocation algorithm is proposed in iCIRRUS project [41].


42


Figure 16 Power level and Power allocation for D2Dk when Nk=9 and 10.

Based on the CSI feedback from the receiver of each D2D pair, the resource allocation algorithm allocates the subcarriers to each D2D pair adaptively. The subcarrier achievable data-rate (SAR)-based algorithm is summarized as follows:

1) Calculate the increment of the total system capacity rn’,k when one unallocated subcarrier n’ is assigned to D2D pair k

2) Choose the maximum value of rn’,k over all unallocated subcarriers to all D2D pairs. 3) Repeat 1) and 2) until all subcarriers are allocated. 4) For each D2D pair, the water-filling algorithm is used to assign the transmit power of

D2Dkto all the subcarriers assigned to D2Dk.

3.2 Performance evaluation of the local resource management schemes

The performance of proposed joint power and subcarrier allocation algorithm is shown in this subsection via simulations. The simulations are carrier out in a single cell scenario with four D2D pairs. The variance of Rayleigh fading for each D2D pair is E[h2

k,n] = 1. Unless otherwise specified, the system has 8 subcarriers, each with subcarrier spacing (bandwidth) 15kHz. The coherence bandwidth of all the channels is set to 15 kHz, and link distance between each D2D pair are set to 1 meter, i.e. dk = 1m. The noise density of each receiver is N0 = −111dB/Hz.

Spectral efficiency

In Figure 18, it is clearly demonstrated that the SAR-based algorithm is close to the system optimal result which obtained via exhaustive global searching, and outperforms the BSCR-


43


based algorithm in terms of spectral efficiency, which is defined as the total system throughput divided by the bandwidth of spectrum allocated to each RRH (with unit bits/s/ Hz).

BSCR-based algorithmProposed SAR-based algorithmGlobal searching result

Figure 17 Spectrum efficiency of different algorithms in the case of varying the ratio of transmit signal power to the noise power per subcarrier

Effect of coherence bandwidth

As the impact of coherence bandwidth is non-negligible in OFDMA systems. It is necessary to look into the effect of coherence bandwidth to the spectral efficiency. In this simulation, we enlarge the total number of subcarriers to be 40, to show the effect of the coherence bandwidth varying from 15 kHz to 52.5 kHz. The simulation results of spectral efficiency for both proposed SAR-based algorithm and BSCR-based algorithm are shown in Figure 19. Regarding the impact of coherence bandwidth to the frequency selective fading, it can be seen from Figure 19 that for both algorithms, the spectral efficiency of the system decreases when the coherence bandwidth enlarges. However, the variation on coherence bandwidth shows less impact to the spectral efficiency of SAR-based algorithm compare to the BSCR-based algorithm.


44


BSCR-based algorithmSAR-based algorithm

Figure 18 The effect of coherent bandwidth on the spectrum efficiency

3.3 Central spectrum assignment at BBU Hotel

It is assumed that the location and coverage of each RRH are generally fixed and known by the BBU hotel. Therefore, the statistics of intra-cell interference between two D2D pair in different RRH’s coverage could be estimated under given frequency reuse pattern, as shown in Figure 20. With frequency reuse factor f=1, the available spectrum is assigned to all RRHs, however, the intra-cell interference from all of the other RRHs’ coverage will significant reduce the spectrum efficiency. Therefore, by reduce the frequency reuse factor from 1 to 1/3, the strong interference links from adjacent RRH’s D2D pair is eliminated.


45


O0

O1

O2

O6

O5

O4

O3

RSub

Cell Transmitter of a D2D pair

Receiver of a D2D pair

Cross sub-cell Interferece

RRH Base Station

O0

O1

O2

O6

O5

O4

O3

RSub

(a) frequency reuse factor f=1

(b) frequency reuse factor f=1/3

Optical Fiber link

Figure 19 Different frequency reuse pattern under hexagonal RRH topology


46


So, for the central spectrum assignment at BBU Hotel, based on the feedback of number of D2D pairs from each RRH, the transmit power on each subcarrier and thus the intra-cell interference could be estimated according to the local resource management.

4 Device to Infrastructure Communication The control signaling of D2D communication in ICIRRUS underlaying a C-RAN network structure is different from traditional cellular D2D communication networks, since the fronthaul delay has great impact on the performance of D2D resource allocation, as discussed in Chapter 3. The general control signaling structure between D2D link and Infrastructure is then discussed in this chapter by considering the fronthaul delay issue. The system parameters of semi-distributed resource management are described in detail under the general control signaling structure.

4.1 Control Signaling Design

Evolved Fronthaul Architecture

In the iCIRRUS project, fronthaul over Ethernet is proposed together with a modified functional split between the baseband unit (BBU) and a remote radio head as enablers for both statistical multiplexing and more efficient resource utilization when connecting the RRH to the BBU via a shared fixed access network. Additionally, D2D communication is considered as a substitute for traditional D2I communication in iCIRRUS in order to reduce latency, power consumption and to improve spectral efficiency of the radio transmission as described in previous sections. As discussed in previous deliverables, such as iCIRRUS D3.1, one of the most promising points for realising this modified functional split would be at the upper physical layer, right after the forward error correction (FEC) for both the down and the uplink, in most cases as it is depicted in Figure 21 for the single input single output (SISO) case in the downlink.

Figure 20 Possible split points for the fronthaul in the downlink for a SISO scenario

This modified functional split has been included in the simplified iCIRRUS architecture shown in Figure 22 together with the main architecture components as well as the data and control planes.


47


Figure 21 iCIRRUS architecture: functional split with data and control planes

Coordinated Data Transmission Between end Users

The previous iCIRRUS deliverable, D4.1, discussed how wireless D2D data links need the assistance of the fixed network for a coordinate data transmission by the exchange of control information. The aim of controlling the D2D link via the fixed infrastructure is to assign specific radio resources to the D2D links. Then, reliable data transmission can be achieved without interference. Interference-free transmission can be established by using the synchronization already in place from the links between the end users and the fixed infrastructure.

The control information exchanged between end devices and infrastructure has to be sent back to the centralized BBU in order to be fully processed and then back to the end user. The quality of the D2D link can be negatively affected by the latency experienced on the control information for this transmission. In order to reduce the latency and thus improve the performance of the D2D link, one should reduce the latency for the control signalling associated with this. By shifting any processing capability closer to the architecture edge, which means moving it closer to the end user, a reduction on the associated latency would be achieved. Figure 23 shows a modification in the function distribution between the BBU and RRH that would lead to a lower latency in the control signalling exchange for coordinating D2D links.


48


Figure 22 Refined iCIRRUS architecture: functional split with data and control planes

Since the data and control plane are logically separated, the processing functionalities of the control plane, could be completely performed at the RRH. That means mainly moving the FEC from the BBU to the RRH as shown in Figure 23. Going a step further, lower latency can be achieved if different FEC algorithms were used for the different planes. A lower overhead in the FEC for the control plane will also require less processing time at the RRH, especially at decoding. In the following sections it is discussed how this new function distribution between BBU and RRH lead to a lower latency in the coordination of D2D links.

Synchronisation

One fundamental requirement for efficient radio resource utilization is synchronization. Synchronization is needed both in time and frequency what can be achieved with the aid of a preamble in both cases. First, the preamble is used by the receiver to get aligned to the transmitted frame structure and bits. Moreover, in order to get an interference-free transmission in the D2D link, the coordinator should know the accuracy of such synchronisation for timing advance. The synchronization accuracy in the D2D link is estimated at the UE and is then fed back to the infrastructure over the control plane. The coordinator can then calculate the correspondent correction for each UE and send it over the control plane to the UEs. The UEs would then correct themselves as indicated by the coordinator. This synchronisation process supported by the fixed infrastructure is depicted in Figure 24 for the case when he BBU coordinate the process (links) and the lower latency approach where the D2D control processing is moved to the RRH (right). Please note that there it can be more than one D2D link in the same RRH coverage area, thus precise time synchronization will be of critical importance.


49


Figure 23 Control signaling for synchronization in the D2D link: initial proposal (left) and low latency proposal (right)

Second, to achieve frequency synchronization CFOs in the D2D link should be estimated and corrected. Here once again, the fixed network will support the D2D link. When the reference oscillator serving as clock source in the UE is the same for both the transmission part in the uplink and the reception part in the downlink, the CFOs of both frequencies (in a FDD mode) are directly related by a constant. That means that it is enough to estimate the CFO for one frequency in order to compensate for both. That it is extensible to the D2D link, probably operating at mm-wave frequencies, if the same clock oscillator source is used as for the D2I links; in that case the CFO of the D2D link will be also directly related to the CFO of the D2I links and can be compensated based on the estimation already done for the D2I links. This principle is also depicted in Figure 24 and shows that there is no difference in terms of latency for the original proposal (left) and the new for a lower latency (right).

Channel estimation and rate adaption

Finally, control signalling is also needed for channel estimation and adaptive data rate adaption. Channel estimation is usually performed with the aid of a preamble. Here once again, the coordinator will also be informed of the channel quality in order to decide, for example, whether the data rate should be adapted to the current channel as depicted in Figure 25. The schema on the left shows the exchange of control information when the BBU is the coordinator and the schema right shows the latency reduction that can be achieved when shifting the coordination for D2D to the RRH.


50


Figure 24 Control signaling for channel estimation and rate adaptation in D2D link: initial proposal (left) and low latency proposal (right)

4.2 D2I based on semi-distributed resource management

To be able to guarantee the quality of service of D2D communications and maximize the system performance, the D2D communication should be managed centrally and assisted by the cellular network. The control signaling for semi-distributed resource management is described in this section.

D2D control signaling over C-RAN architecture

The traditional D2D control signaling for centralized resource management requires the CSI feedbacks from all the D2D transmitter to D2D receiver. Therefore, if the total number of D2D pairs within in one cell is N, the total number of CSI feedback needed is proportional to N2. As the number of D2D pairs may be very large in practical systems, centralized resource management will increase the control signaling transmission and computation significantly. Furthermore, in ICIRRUS architecture, with extra latency introduced by fronthaul architecture, the inaccuracy of CSI feedback will deduce the improvement of centralized resource management. Therefore, the general control signaling between D2I for D2D communication consists of two parts, which are D2D to RRH via wireless link and RRH to BBU Hotel via fronthaul link. Based on the semi-resource management mentioned in Chapter 3, the control signaling parameters are described as follow subsections.

Control signaling between D2D receivers and RRHs

The control signaling between D2D receivers and RRHs is used for local resource management, which can be represented by two key parameters.

1) CSI: the channel state information of the direct D2D communication link (Estimated at D2D receiver). The D2D transmitter should generate the pilot signal which contains the D2D communication identifier and broadcast it through the pilot channel. The surrounding devices which have same D2D communication identifier would estimate


51


the CSI and send it to RRH accordingly. Please note that the D2D communication identifier of one D2D link is assumed to be decided in D2D discovery process and known by both of the transmitter and receiver of D2D communication.

2) Resource allocation results: the allocated subcarriers and permitted transmit power

for the D2D communication. The local resource allocation algorithm is performed at each RRH and the resource allocation result including assigned subcarrier sets and transmit power will be send to the D2D links together with D2D communication identifier.

The latency of local resource allocation is same as D2D communication in cellular network without C-RAN architecture, since there is no direct communication needed from BBU hotel. Thus, it could be performed in the same time scale as cellular resource management.

Control signaling between RRHs and BBU Hotel

The control signaling between RRHs and BBU Hotel is used for centralized resource assignment, which can be represented by two key parameters:

1) K: Number of D2D pairs within the coverage of RRH. Each D2D pair would need to communicate with one RRH by D2D communication identifier during the local resource allocation process. Thus, the number of D2D pairs within the coverage of RRH could be obtained by each RRH and send to BBU Hotel via fronthaul link.

2) Nl: Amount of spectrums assigned to the lth RRH for D2D communication. Based on

the feedback from RRHs, BBU Hotel would decide the frequency reuse factor and assign the spectrum to each RRH to be able to control the intra-cell interference, and guarantee the quality of service for D2D communication.

Since the number of D2D pairs is relatively stable compare to the resource allocation period, it is not necessary to update the spectrum assigned to each RRH frequently. Furthermore, the variation of number of D2D pairs within one RRH is predictable in some cases and could be further estimated with the help of Self-Optimizing Network (SON). For instance, the number of D2D pairs within a shopping mall during the weekend at daytime should be much higher than other working days.

So, the central spectrum assignment process could be performed in longer term compare to local resource management. Therefore, the extra fronthaul latency in C-RAN architecture has less influence on the resource management.

Summarize and future work:

In this section, the relevant system parameters for D2I based control signaling of semi-distributed resource allocation are described in detail. Also, there are some related ongoing works as shown below.


52


1. The centralized resource assignment at BBU Hotel may jointly be considered with D2D discovery and mobile cloud networks and will be discussed in deliverable D4.4.

2. The packet form of the control signaling and the transmission channel between D2I will be discussed in the deliverables of WP5 in the future.

3. The control signaling for D2D discovery is under investigation and will be jointly discussed in the deliverables of WP5 in the future.

5 Inventory models for the C-RAN, D2D and D2I Inventory models are part of OSS (Operations Support Systems). They are used by the operators for efficient service provision over the entire network. It contains information about the network physical infrastructure and equipment, services availability and subscriber package information.

In today networks, operators are faced with many different services, technologies and service level demands. To keep control and manage all different network possibilities, it is essential to keep network topology and functionalities updating as much as possible.

Currently, it is not possible to completely describe the 5G mobile network inventory model, because some of the target technologies are still under investigation. As far as we currently expect, the models should follow the proposed scheme.

5.1 C-RAN (Cloud Radio Access Network) inventory model

Figure 26 shows that the C-RAN model consists of some RAN elements: BBU pool, radio towers and several RRU element depending on site locations and cell number. Elements are physically interconnected using optical connections, over which an Ethernet technology is assumed. For switching (aggregation) functionality, Ethernet switches are used. For different services and capacity demands, various VLANs could be configured and managed.


53


Figure 25: C-RAN inventory model

5.2 D2I (Device to Infrastructure) inventory model

D2I is a conventional mobile concept. It consists of a Base station, its cells and UE on the other end. Mobile installation path represents radio connectivity from UE to base stations cell. Over mobile installation path we deliver 3-play or any other mobile services to end costumers. Each service is modeled with specific service attributes and supplemental features.

Figure 26: D2I inventory model

AMO LTEBase Stat ion

MOBILE INSTALLATION path

Radio leg

celica 1 na BP

User capacity 1

Free capacity

Cell modelled as »Network object«, category »MBT_CELICA«, fixed Assignment.

N cell BS

Channel

SERVICE 1 SERVICE 2 SERVICE 3

AMO Terminal Equip

SERVICE N

User capacity 2

User capacity 3


54


5.3 D2D (Device to Device) inventory model

Figure 27: D2D communication concept

According to the D2D general concept depicted on Figure 28, the D2D inventory model is updated based on D2I model to support additional D2D communication. To support the D2D communication, from the technological part, on the wireless network access side there are no major changes need. The main technical upgrade is on user equipment (UE). It should be equipped with the appropriate communication hardware to support D2D. Network should take care for the signalization (D2I and D2D).

From the inventory point of view, due to the major functionality resources on the UE side, the D2D communication is modeled as an additional subscriber service or just another additional feature on existing mobile service (yet to be defined).

Figure 28: D2D inventory model

5GBase station

D2D

wire

less

com

mun

icatio

n

cell1 on BBU (RRH)

User capacity 1User capacity 1User capacity 1

Free capacity

Cell modelled as »Network object«, category »MBT_CELICA«, fixed Assignment.

N cell on BBU (RRH)


55


6 Conclusions In this deliverable, D4.2, we have presented the results that have arisen out of the second year’s work of the work package WP4 “Efficient User Equipment Integration” of the iCIRRUS project. We have described some results of the technologies that we have been researching to enable D2D to work under the C-RAN architecture, which will be applicable in the future 5G networking.

In particular, we have reported on the key D2D communication technologies, which include the D2D discovery, MD2D with content caching, D2D assisted content caching, security issue in D2D communications, and 60 GHz networking for D2D communications. A novel D2D discovery method by adopting power control is presented to deal with the IEI issue, which improved the SINR performance of the D2D system. The model for MD2D with content caching has been established, the outage performance of which will be left for future work. One energy efficient pushing strategy for D2D content caching scenario is proposed, which can achieve large energy savings. Significant offloading can be realized by using D2D assisted content caching transmission. The Diffie-Hellman key agreement protocol is applied to D2D communication systems. However, the D2D link is still vulnerable to well-known man-in-the middle attack under this protocol. This issue will be considered for future work. Then, we reported on the 60 GHz (mm-wave) communication results that may underpin D2D communications, or as a means to create alternative D2I networking topologies. A maximum throughput of 2.4Gbps up to 20 metres distance have been successfully demonstrated. Future work will demonstrate the 60-GHz system into D2I systems.

As an important contribution to D2D communication underlaying the iCIRRUS infrastructure, a semi-distributed D2D resource management proposed in iCIRRUS with the objective of achieving high throughput under low D2I overhead was reported in this deliverable along with the performance analyses and achieved results.

We have also reported the control signalling for D2I communications. One novel signalling method for D2I system has been proposed, which shifts the signal processing task originally performed at the BBU pool to the RRHs. In this case, the RRHs are more near to the users, and can reduce the latency. In addition, to make this scheme work, some additional techniques, such as the synchronization issue, channel estimation and rate adaption, have been discussed. Based on the signalling platform, how to implement the proposed the semi-distributed resource management scheme into to the iCIRRUS infrastructure was discussed.

Finally, the Inventory models for the C-RAN, D2D and D2I has been investigating, and initial results and strategies for D2D signalling has been presented. These inventory models are part of OSS (Operations Support Systems). They are used by the operators for efficient service provision over the entire network.

The technologies presented here will continue to be investigated into the 3th year of the iCIRRUS project within WP4. In addition, some results developed in this deliverable are closely related to the parallel workpage WP2 “Centralisation Scenarios and Architecture”, and WP3 “Future Converged Front- (Mid-) Haul Architecture). For example, the control signaling for D2I communications requires the RRHs to have some capabilities of signal processing, which is the main research topic for WP2 and WP3 concerning function splitting between BBU


56


pool and RRHs. In the 3th year of the project, the results from WP4 will be focused on the design of the test-bed demonstrators and technologies for D2D and D2I, providing inputs to WP5.

7 References

[1] T. Kwon and J. W. Choi, "Spatial Performance Analysis and Design Principles for Wireless Peer

Discovery," in IEEE Transactions on Wireless Communications, vol. 13, no. 8, pp. 4507-4519, Aug.

2014.

[2] Z. J. Yang et al., "Peer discovery for device-to-device (D2D) communication in LTE-A

networks," Globecom Workshops (GC Wkshps), 2013 IEEE, Atlanta, GA, 2013, pp. 665-670

[3] M. Simsek, A. Merwaday, N. Correal and İ Güvenç, "Device-to-device discovery based on 3GPP system

level simulations," Globecom Workshops (GC Wkshps), 2013 IEEE, Atlanta, GA, 2013, pp. 555-560.

[4] D. Li and Y. Liu, "In-Band Emission in LTE-A D2D: Impact and Addressing Schemes," Vehicular

Technology Conference (VTC Spring), 2015 IEEE 81st, Glasgow, 2015, pp. 1-5

[5] D. Kim, Y. Kwak, J. Oh, Y. Kim and J. Lee, "Discovery resource grouping for D2D discovery for

mitigation of in-band emission in LTE-Advanced," Globecom Workshops (GC Wkshps), 2014, Austin,

TX, 2014, pp. 869-874.

[6] W. Hwang, D. Lee and H. J. Choi, "A new channel structure and power control strategy for D2D

discovery in LTE cellular network," Communications (APCC), 2014 Asia-Pacific Conference on,

Pattaya, 2014, pp. 150-155.

[7] J. H. Song, D. H. Lee, W. J. Hwang and H. J. Choi, "A selective transmission power boosting method

for D2D discovery in 3GPP LTE cellular system," Information and Communication Technology

Convergence (ICTC), 2014 International Conference on, Busan, 2014, pp. 267-268.

[8] J. Li, J. B. Song and Z. Han, "Network Connectivity Optimization for Device-to-Device Wireless System

With Femtocells," in IEEE Transactions on Vehicular Technology, vol. 62, no. 7, pp. 3098-3109, Sept.

2013.

[9] Y. Kwak, S. Ro, S. Kim, Y. Kim and J. Lee, "Performance Evaluation of D2D Discovery with eNB

Based Power Control in LTE-Advanced," Vehicular Technology Conference (VTC Fall), 2014 IEEE

80th, Vancouver, BC, 2014, pp. 1-5.

[10] 3GPP TR 36.888 , “Study on provision of low-cost Machine-Type Communications (MTC) User

Equipments (UEs) based on LTE (Release 12),” June 2013.

[11] T. Cisco, “Cisco visual networking index: Global mobile data traffic forecast update, 2012–2017,” Cisco

Public Information, 2013.


57


[12] Y. Cao, T. Jiang, and C. Wang, “Cooperative device-to-device communications in cellular networks,”

IEEE Wireless Communications, vol. 22, no. 3, pp. 124–129, March 2015.

[13] E. Bastug, M. Bennis, and M. Debbah, “Living on the edge: The role of proactive caching in 5G wireless

networks,” IEEE Communication Magazine, vol. 52, no. 8, pp. 82–89, August 2014.

[14] M. Cha, H. Kwak, P. Rodriguez, Y.-Y. Ahn, and S. Moon, “Analyzing the video popularity

characteristics of large-scale user generated content systems,” IEEE/ACM Transactions on Networking

(TON), vol. 17, no. 5, pp. 1357–1370, May 2009.

[15] U. Niesen, D. Shah, and G. W. Wornell, “Caching in wireless networks,” IEEE Transactions on

Information theory, vol. 58, no. 10, pp.6524–6540, October 2012.

[16] K. Shanmugam, N. Golrezaei, A. G. Dimakis, A. F. Molisch, and G. Caire, “Femtocaching: Wireless

content delivery through distributed caching helpers,” IEEE Transactions on Information Theory, vol.

59,no. 12, pp. 8402–8413, December 2013.

[17] S. Sun, Q. Gao, W. Chen, R. Zhao, and Y. Peng, “Recent progress of long-term evolution device-to-

device in third-generation partnership project standardisation,” IET Communications, vol. 9, no. 3, pp.

412–420, March 2015

[18] N. Golrezaei, A. Dimakis, and A. Molisch, “Wireless device-to-device communications with distributed

caching,” in 2012 IEEE International Symposium on Information Theory Proceedings (ISIT), July 2012,

pp. 2781–2785.

[19] N. Golrezaei, P. Mansourifard, A. F. Molisch, and A. G. Dimakis, “Base-station assisted device-to-

device communications for high-throughput wireless video networks,” IEEE Transactions on Wireless

Communications, vol. 13, no. 7, pp. 3665–3676, Aug. 2014.

[20] N. Golrezaei, A. Dimakis, and A. Molisch, “Scaling behavior for device-to-device communications with

distributed caching,” IEEE Transactions on, Information Theory, vol. 60, no. 7, pp. 4286–4298, July

2014.

[21] M. Cha, H. Kwak, P. Rodriguez, Y.-Y. Ahn, and S. Moon, “I tube, youtube, everybody tubes: analyzing

the world’s largest user generated content video system,” in Proceedings of the 7th ACM SIGCOMM

conference on Internet measurement, ACM, 2007, pp. 1–14.

[22] A. Goldsmith, Wireless Communications. Cambridge University Press, 2005.

[23] F. Rebecchi, M. Dias de Amorim, V. Conan, A. Passarella, R. Bruno, and M. Conti, “Data offloading

techniques in cellular networks: a survey,”IEEE Communications Surveys & Tutorials, vol. 17, no. 2,

pp. 580–603,2015.

[24] V. Sciancalepore, D. Giustiniano, A. Banchs, and A. Hossmann-Picu,“Offloading cellular traffic through

opportunistic communications: Analysis and optimization,” IEEE Journal on Selected Areas in

Communications,vol. 34, no. 1, pp. 122–137, 2016.


58


[25] Y. Wu, J. Chen, L. P. Qian, J. Huang, and X. Shen, “Energy-aware cooperative traffic offloading via

device-to-device cooperations: An analytical approach,” IEEE Transactions on Mobile Computing, vol.

PP, no. 99, pp. 1–1, 2016.

[26] L. Al-Kanj, H. V. Poor, and Z. Dawy, “Optimal cellular offloading via device-to-device communication

networks with fairness constraints,” IEEE Transactions on Wireless Communications, vol. 13, no. 8,

pp.4628–4643, 2014.

[27] P. Sermpezis and T. Spyropoulos, “Effects of content popularity on the performance of content-centric

opportunistic networking: An analytical approach and applications,” arXiv preprint arXiv:1601.05266,

2016.

[28] Y. Guo, L. Duan, and R. Zhang, “Cooperative local caching and file sharing under heterogeneous file

preferences,” arXiv preprint arXiv:1510.04516, 2015.

[29] J. Jiang, S. Zhang, B. Li, and B. Li, “Maximized cellular traffic offloading via device-to-device content

sharing,” IEEE Journal on Selected Areas in Communications, vol. 34, no. 1, pp. 82–91, 2016.

[30] Y. Li, M. Qian, D. Jin, P. Hui, Z. Wang, and S. Chen, “Multiple mobile data offloading through

disruption tolerant networks,” IEEE Transactions on Mobile Computing, vol. 13, no. 7, pp. 1579–1596,

2014.

[31] Y. Zhang, L. Song, W. Saad, Z. Dawy, and Z. Han, “Contract-based incentive mechanisms for device-

to-device communications in cellular networks,” IEEE Journal on Selected Areas in Communications,

vol. 33,no. 10, pp. 2144–2155, 2015.

[32] M. N. Soorki, M. Yaghini, M. H. Manshaei, W. Saad, and H. Saidi, “Energy-aware optimization and

mechanism design for cellular deviceto-device local area networks,” arXiv preprint arXiv:1601.06762,

2016.

[33] A. Antonopoulos, E. Kartsakli, and C. Verikoukis, “Game theoretic d2d content dissemination in 4g

cellular networks,” IEEE Communications Magazine, vol. 52, no. 6, pp. 125–132, June 2014.

[34] S. N. Chiu, D. Stoyan, W. S. Kendall, and J. Mecke, Stochastic geometry and its applications. John

Wiley & Sons, 2013.

[35] K. Mahmud, M. Inoue, H. Murakami, M. Hasegawa, and H. Morikawa, “Energy consumption

measurement of wireless interfaces in multiservice user terminals for heterogeneous wireless networks,”

IEICE transactions on communications, vol. 88, no. 3, pp. 1097–1110, 2005.

[36] L. A. Adamic and B. A. Huberman, “Zipfs law and the internet,”Glottometrics, vol. 3, no. 1, pp. 143–

150, 2002.

[37] W. Shen, W. Hong, X. Cao, B. Yin, D. M. Shila and Y. Cheng, "Secure key establishment for Device-

to-Device communications," 2014 IEEE Global Communications Conference, Austin, TX, 2014, pp.

336-340.


59


[38] Fu, Xiaoming, et al. "QoS and security in 4G networks." First annual global mobile congress, shanghai,

China. 2004.

[39] Merkle, Ralph C. "Secure communications over insecure channels."Communications of the ACM 21.4

(1978): 294-299.

[40] Q. Ye and A.-S., “Distributed Resource Allocation in D2D Enhanced-cellular Networks,” IEEE

Transactions on Communications, Feb. 2015, pp.441-454.

[41] Kai, Y and H. Zhu, “Resource Allocation for Multiple-pair D2D Communications in Cellular

Networks,” IEEE International Conference on Communications (ICC) 2015, London, 2015.


60


8 List of figures Figure 1 Relationships for the four sections......................................... Error! Bookmark not defined.

Figure 2: Illustration of In-band emission interference in D2D overlaid in cellular network ..... 13

Figure 3 D2D discovery channel structure ...................................................................................... 13

Figure 4 Proposed discovery model ................................................................................................. 14

Figure 5 CDF performance comparisons of received cellular signal SINR at the BS [Grouping (Gr.), Proposed method (Prop.), Group-1 (Gr1), Group-2 (Gr2)] ... Error! Bookmark not defined.

Figure 6 CDF performance comparisons of received discovery signal SINR at the DUE [Grouping (Gr.), Proposed method (Prop.), Group-1 (Gr1), Group-2 (Gr2)] ............................................. 19

Figure 7 Probability of successful discovery at the DUE .............................................................. 20

Figure 8 System Model of the conten caching for multiple-device to single-device .............. Error! Bookmark not defined.

Figure 9: An example of a content with two keywords ......................... Error! Bookmark not defined.

Figure 10: The flow chart of content dissemination protocol .............. Error! Bookmark not defined.

Figure 11: Energy consumption for different popularities with M=20 ........................................... 33

Figure 12: Energy consumption for different sharing probabilities with M=20 ............................ 34

Figure 13: Energy consumption performance under different similarities between 𝒘𝒘𝒘𝒘 and 𝝆𝝆𝒘𝒘 with M=6 ....................................................................................................................................... 34

Figure 14 A scenario with a 60-GHz wireless link between the RRH and Base Station .............. 37

Figure 15 System architecture for D2D underlaid cellular networks ............................................. 39

Figure 16 Block diagram for semi-distributed D2D resource allocation in C-RAN architechture ...................................................................................................................................................... 40

Figure 17 Power level and Power allocation for D2Dk when Nk=9 and 10. .................................... 42

Figure 18 Spectrum efficiency of different algorithms in the case of varying the ratio of transmit signal power to the noise power per subcarrier ...................................................................... 43

Figure 19 The effect of coherent bandwidth on the spectrum efficiency ..................................... 44

Figure 20 Different frequency reuse pattern under hexagonal RRH topology ............................. 45

Figure 21 Possible split points for the fronthaul in the downlink for a SISO scenario .............. 46

Figure 22 iCIRRUS architecture: functional split with data and control planes ........................... 47

Figure 23 Refined iCIRRUS architecture: functional split with data and control planes ............ 48

Figure 24 Control signaling for synchronization in the D2D link: initial proposal (left) and low latency proposal (right) .............................................................................................................. 49

Figure 25 Control signaling for channel estimation and rate adaptation in D2D link: initial proposal (left) and low latency proposal (right) ....................................................................... 50

Figure 26: C-RAN inventory model ................................................................................................... 53

Figure 27: D2I inventory model ......................................................................................................... 53

Figure 28: D2D communication concept .......................................................................................... 54


61


Figure 29: D2D inventory model ........................................................................................................ 54

9 List of tables Table 1 Simulation Parameters ......................................................................................................... 16

Table 2 A Low-complexity Iterative Algorithm for M group ......................................................... 29

Table 3: General simulation parameters ........................................................................................... 32

Documents

intelligent Converged network consolIdating Radio and ... - … · method, key issues such as synchronization and channel estimation are discussed. Then, for the purpose of implementation,