contributions to the optimization of radio access networks ... · proiect posdru/107/1.5/s/76903...

67
FONDUL SOCIAL EUROPEAN Investește în oameni! Programul Operațional Sectorial pentru Dezvoltarea Resurselor Umane 2007 2013 Proiect POSDRU/107/1.5/S/76903 Valorificarea capitalului uman din cercetare prin burse doctorale (ID76903) UNIVERSITY POLITEHNICA OF BUCHAREST Faculty of Electronics, Telecommunications and Information Technology Telecommunications Department Contributions to the optimization of radio access networks in 4th generation communication systems A Dissertation submitted for the degree of Doctor of Philosophy Author: Alexandru VULPE PhD supervisor: Simona HALUNGA Abstract Bucharest, 2014

Upload: others

Post on 08-Sep-2019

13 views

Category:

Documents


0 download

TRANSCRIPT

FONDUL SOCIAL EUROPEAN

Investește în oameni!

Programul Operațional Sectorial pentru Dezvoltarea Resurselor Umane 2007 – 2013

Proiect POSDRU/107/1.5/S/76903 – Valorificarea capitalului uman din cercetare prin burse doctorale (ID76903)

UNIVERSITY POLITEHNICA OF BUCHAREST

Faculty of Electronics, Telecommunications and Information Technology

Telecommunications Department

Contributions to the optimization of radio access networks in 4th generation communication systems

A Dissertation submitted for the degree of Doctor of Philosophy

Author: Alexandru VULPE PhD supervisor: Simona HALUNGA

– Abstract –

Bucharest, 2014

i

Table of contents Table of contents ........................................................................................................................ i

Introduction ............................................................................................................................... 2

Chapter 1 4th generation communication systems ............................................................... 4

1.1 LTE Rel. 10 characteristics ............................................................................................ 4

Chapter 2 Radio Resource Management ............................................................................... 6

2.1 Introduction .................................................................................................................... 6

2.2 RRM algorithms ............................................................................................................. 6

2.3 Resource scheduling and link adaptation ....................................................................... 6

2.4 Scheduling algorithms.................................................................................................... 6

Chapter 3 Solutions for Radio Access Network Optimization ............................................. 9

3.1 Introduction .................................................................................................................... 9

3.2 Wi-Fi offloading ............................................................................................................. 9

3.3 Uncoordinated deployment of small cells .................................................................... 11

3.4 Conclusions .................................................................................................................. 13

Chapter 4 Resource scheduling and admission control algorithms in LTE-Advanced ....... 14

4.1 Introduction .................................................................................................................. 14

4.2 Scheduling algorithms for systems with spectrum aggregation ................................... 14

4.3 Multi-Carrier Scheduling Algorithm (MCSA) ............................................................. 16

4.4 Enhanced MCSA algorithm (E-MCSA) ...................................................................... 17

4.5 Conclusion ................................................................................................................... 17

Chapter 5 Results ................................................................................................................ 18

5.1 Main results .................................................................................................................. 18

5.2 Conclusions .................................................................................................................. 18

Chapter 6 Conclusions and future work .............................................................................. 20

6.1 General conclusions ..................................................................................................... 20

6.2 Future work .................................................................................................................. 20

6.3 Main contributions of the author .................................................................................. 20

6.4 List of publications....................................................................................................... 21

Bibliography............................................................................................................................ 24

2

Introduction

The continuous evolution of mobile communication systems makes the number of users and

the request for high data rate services to continuously rise. Therefore, 3rd generation mobile

communication systems have evolved towards 4th generation ones, with different requirements.

End users expect data rates that are similar to the ones received via fixed Internet, while operators

expect high capacity and low data traffic costs. In order to meet these demands, version 8 of the

3GPP specifications, known as Long Term Evolution (LTE) has been developed.

However, LTE does not meet ITU requirements for a 4G technology (or IMT-Advanced

requirements). These demand higher peak data rates, enhanced spectral efficiency as well as

increased user throughput. In order to meet these demands and also remain backwards compatible

with Rel. 8, 3GPP Release 10, also known as LTE-Advanced, has been developed. LTE-Advanced

supports the technique known as Carrier Aggregation, which allows to put together different

frequency bands in order to obtain a larger aggregated band, even though operators do not own such

large chunks of continuous spectrum.

The present dissertation focuses on radio resource management in LTE-Advanced systems.

The author proposes new resource scheduling and admission control algorithms for LTE-Advanced.

The algorithms leverage the multiple dimensions that come from using carrier aggregation and

optimize resource allocation as well as enhance LTE-Advanced user throughput. The performances

of these algorithms are compared to traditional scheduling algorithms, that run independently on

each component carrier, as well as to some of the early developed scheduling algorithms for LTE-

Advanced. The analysis is done both in the case of a macro-only network and in the case of a 3GPP

heterogeneous network (HetNet). The different algorithms offer different performances, according

to the scenarios in which they are being used. The thesis is organized as follows.

Chapter 1 gives a quick background on LTE (3GPP Rel. 8/9) and LTE-Advanced (3GPP Rel.

10 and beyond). The focus is on the changes to LTE brought by LTE-Advanced.

Chapter 2 described the main RRM functions, with a focus on scheduling algorithms for LTE

and how they can be adapted for LTE-Advanced.

Chapter 3 focuses on two of the most popular techniques for capacity increase in mobile

networks: Wi-Fi offloading and small cell deployment. A hardware and software platform for Wi-Fi

offloading is proposed. The platform is suitable to be used in a crowded environment both in fixed

and mobile scenarios. Also, the uncoordinated deployment of small cells in similar scenarios is

studied. The conclusion is that scheduling and admission control algorithms have to be distributed

and minimize the need for coordination.

Chapter 4 proposes new or enhanced scheduling algorithms for carrier aggregation systems

(LTE-Advanced). First, a simple scheduling algorithm which optimizes the cross-CC resource

allocation, without taking into account CSI is developed. It is based on the definition of a convex

optimization problem with the target of obtaining an average user throughput that is higher than the

one offered by using traditional scheduling algorithms. Then, based on the same idea, an algorithm

that takes into account user QoS requirements has been proposed. It is again based on a convex

optimization problem, but with different constraints.

Chapter 5 presents the results obtained from evaluating the performance of the algorithms in

Chapter 4. Different scenarios (macro-only, HetNet) were used. The simulations conducted showed

3

that the proposed algorithms have better performances, especially in HetNet scenarios. They have

poor performances, however, in macro-only scenarios (with a few exceptions). In the end, some

recommendations are made with regards to which scheduling algorithm is more suited according to

the scenarios in which it is being used.

4

Chapter 1 4th generation communication systems

The last years have seen an increase of Internet data traffic via mobile broadband connections.

The cause of this unprecedented growth is mainly the user willingness to „be online” at any time

and any place. The High Speed Packet Access (HSPA) și Long Term Evolution (LTE) systems have

made mobile broadband a reality. Users now own a variety of portable devices with high data rate

capabilities – smartphones, tablets, USB modems – and a large number of applications that allow

access to social networks, audio and video file sharing etc.

Recent surveys have shown that the mobile data traffic is still rising at an extremely high rate

and there is more and more the need for scalable communication solutions, especially in nomadic

areas. According to [1], mobile data traffic has increased in 2012 by 70%. Out of this total, 51%

was represented by video traffic video (video calls, video conferences, video streaming etc.). Also,

the survey shows that, although 4G connections represent only 0.9% out of today’s mobile

connections, it generates 14% of the total mobile data traffic.

Also, 33% of the total mobile data traffic has been offloaded, either via access points based on

the IEEE 802.11 standard(Wi-Fi offloading), or via low power cellular base stations, owned by the

users themselves (femto-cells). This means that, without offloading, mobile data traffic would have

risen in 2012 with 96% and not just 70%.

The same survey show a 13-fold increase of the global mobile data traffic in 2017, and that

65% of the mobile data traffic will be offloaded. Also, two thirds of the traffic will be video traffic,

meaning a 16-fold increase compared to 2012.

These figures show, without a doubt, the need for local, low-power, low-coverage acces

networks, that would complement traditional cellular networks. These would provide better indoor

coverage, enhanced user experience as well as higher performance/cost ratios as shown in [2]

Release 10 of the 3GPP specifications, also known as LTE-Advanced, is an evolution of LTE

that brings enhancements by using Carrier Aggregation and supporting heterogeneous networks

(defined by 3GPP as a mix of macro-cells and low-power, low coverage cells – small cells), with

the purpose of fulfilling and even surpassing IMT-Advanced (or 4G) requirements. Among these

requirements is the possibility to provide data rates of up to 1 Gbps for low mobility scenarios like

nomadic access.

1.1 LTE Rel. 10 characteristics

1.1.1 Carrier Aggregation

The first version of LTE already supported various spectrum assignments with bandwidths

ranging from 1.4 MHz to 20 MHz, both in paired and unpaired spectrum. However, since IMT-

Advanced requirements had to be met, Carrier Aggregation was used. Using this technique, several

LTE carriers (known as component carriers) are aggregated to form one aggregated carrier with a

large bandwidth that can be used in its entirety by a UE to transmit and receive. Up to 5 CCs can be

aggregated, each with different bandwidths, and a carrier with an aggregated bandwidth of at most

100 MHz can be obtained. Likewise, each CC has an LTE Rel. 8 carrier structure, thus ensuring

backwards compatibility. For a Rel. 8/9 terminal, each component carrier will appear like a Rel. 8

5

carrier, while a CA enabled terminal will be able to use the larger bandwidth obtained from

aggregating the CCs, thus enabling higher data rates. The number of aggregated component carriers

on the uplink and the downlink can differ.

1.1.2 Relaying

Relaying means that the UE communicates with the network via a relay node which is

connected to a donor cell via the LTE air interface. From the UE point of view, the relay appears as

a regular cell. This hold the advantage of simplifying the UE implementation and ensuring

backwards compatibility with LTE Rel. 8/9. Basically, the relay is a low power base station

wirelessly connected to the rest of the network.

1.1.3 Heterogeneous Networks

Heterogeneous networks (HetNets), can be defined as a mix of cells with different

transmission powers that, partially, operate in the same frequency bands and have overlapping

coverage. A typical example is a pico cell that is placed within the coverage area of a macro cell.

Although such scenarios were already supported in Rel. 8, Rel. 10 introduced enhanced Inter-Cell

Interference Coordination focused on scenarios where there are large differences between the cell

transmission powers [3].

6

Chapter 2 Radio Resource Management

2.1 Introduction

The role of the RRM block is to ensure an efficient utilization of the radio resources and to

serve users according to their configured QoS requirements.

2.2 RRM algorithms

The RRM algorithms at the eNodeB have different Layer 1 to Layer 3 functions. Layer 3

algorithms manage QoS, admission control and semi-static scheduling. RRM. Layer 1 and 2 RRM

algorithms such as HARQ management, dynamic packet scheduling and link adaptation are

dynamic functions that are executed each 1 ms TTI [4], [5].

The Layer 1 CQI manager processes the received CQI reports (for the downlink) and SRS

reports (for the uplink) from the active users of the cell. Each CQI and SRS is used by the eNodeB

for scheduling and link adaptation decisions [5], [6].

3GPP specifies RRM signaling [6], but RRM algorithms are not specified by 3GPP, and can

be operator or vendor independent.

2.3 Resource scheduling and link adaptation

Dynamic packet scheduling and link adaptation are functions which ensure a high spectral

efficiency, at the same time providing the required QoS level in the cell.

The packet scheduler makes scheduling decision each TTI, by allocating RBs to the users,

but also by controlling transmission parameters, including modulation and coding scheme. The

selected RBs and the transmission parameters are signaled to the scheduled users via PDCCH. The

target of scheduling algorithms is to maximize the capacity, at the same time ensuring that the QoS

requirements are fulfilled and there are also resources available for Best Effort services. The

scheduling decisions are taken for each user, even though it might have several ongoing data

streams. In order to differentiate between them, each data stream is identified by LCID, and the

MAC layer will decide on the amount of data to be transmitted for each LCID[3], [5].

2.4 Scheduling algorithms

Resource allocation is a way to provide quality of service for scenarios where there are many

users who want to be served simultaneously with a certain level of quality. Such an allocation of

resources is desired to maximize the spectral efficiency ensuring also the QoS. This efficiency

should be made by studying the channel conditions at different moments, sometimes in real-time.

There are several scheduling algorithms and a large number of variations. The main classes

are described in the following sections.

2.4.1 Round Robin scheduling

Round Robin is a simple scheduling technique in which mobile devices are served without

7

taking into account CSI information. Users are allocated resources without discrimination and by

default. Because it does not take into account the current or previous radio channel, there will be a

variation of the throughput from one user to another, due to the variation of radio channel. This type

of scheduling is not suited for situations involving users with significant differences in radio

conditions experienced.

u=1

u>N?

Nu

Da

START

STOP

Alocă N/NPRB

blocuri PRB

utilizatorului u

u=u+1

Figure 2.1 Round Robin scheduling

n=1

u=1

n>NPRB?

u>N?

Nu

Utilizatorul u cu

cel mai mare

index RSIZu,n va

fi planificat pe

blocul PRB de

ordin n

Da

Calcul RSIZ pentru

utilizator u și bloc

PRB de ordin n:

RSIZu,n

START

Stația BS primește

rapoartele CQI de

la UE atașate

STOP

n=n+1

u=u+1

Figure 2.2 MAX C/I scheduling

A diagram of the Round Robin algorithm is given in Figure 2.1.

2.4.2 MAX C/I scheduling

MAX C/I allocates resources to UEs with the best channel quality. Therefore UEs with higher

SINR will have the scheduling priority over other UEs. This type of scheduling maximizes the cell

throughput but may lead to some users never being served. [5].

MAX C/I scheduling is shown in Figure 2.2.

2.4.3 Proportional Fair scheduling

PF scheduling, in its simplest form, allocates resources to the user that maximizes a function

for each PRB, defined in (2.1) [4,5].

A function called Proportional Fair index is defined as the ratio between the instantaneous

user throughput and the historical average user throughput.

i

ti

tiD

DF

,

,

(2.1)

where:

tiD , is the instantaneous throughput for user i, for PRB t.

8

iD is the historical average throughput for user i.

A diagram of the Proportional

Fair algorithm is given in Figure 2.3.

There are several variations of this

algorithm which the interested reader

can fiind in [7]–[9].

n=1

u=1

n>NPRB?

u>N?

Nu

Utilizatorul u cu

cel mai mare

index Fu,n va fi

planificat pe

blocul PRB de

ordin n

Da

Calcul index de

alocare

proporțională

pentru utilizator u și

bloc PRB de ordin

n: Fu,n

START

Stația BS primește

rapoartele CQI de

la UE atașate

STOP

n=n+1

u=u+1

Figure 2.3 Proportional Fair scheduling

9

Chapter 3 Solutions for Radio Access Network

Optimization

3.1 Introduction

There are higher and higher pressure on cellular network operators to deliver guaranteed

throughput to their subscribers. Some solutions for this problem can be the new technologies that

promise enhanced spectral efficiency. Therefore, mobile operators should use the capacity of the

radio access network to the maximum, by gaining knowledge of where there is the need for

additional radio resources and how it is better to deploy them (for instance, small cells bring more

capacity if they are deployed in areas with a relatively low spectral efficiency).

The present chapter focuses on two of the most popular techniques for capacity increase in

mobile networks: Wi-Fi offloading and small cell deployment.

3.2 Wi-Fi offloading

The following proposes to define a simple, ready to be deployed framework for Wi-Fi

offloading from 3G/4G networks in clearly defined scenarios. These scenarios are depicted in

Figure 5.1. This architecture will be called Application to Offload Peak Traffic in 3G/4G

Networks (ALO34)

The main component is a common hardware platform that is called the ALO34 central unit. It

contains all the hardware necessary for ensuring the proper offloading of traffic. This hardware

platform will slightly differ according to the scenario in which it is being used. In the fixed

scenario, the unit will be deployed in crowded areas such as shopping malls, restaurants, stadiums,

concert areas etc. It will be connected to the Internet via the wired infrastructure available on the

site, and will be named the ALO34 fixed unit. In the mobile scenario, the unit will be deployed in

high-speed vehicles such as trains, buses, or even cars. It will be connected to the 3G/4G network

via the air interface and will act just like a regular UE, being named the ALO34 mobile unit. The

main difference between this unit and a regular UE, from the air interface point of view is that it

will not have the battery constraints that are present in end-user UEs. It will be possible for this unit

to have multiple antennas (MIMO) for throughput boost and even higher gain antennas.

Here it is assumed that the WiFi hotspot coverage overlaps the already offered coverage

provided by an LTE eNodeB and does not extend beyond the LTE coverage.

Following the above prerequisites the following use cases, which could be interesting from

the operators’ point of view, are imagined:

1. All the users have data sessions opened on LTE or Wi-Fi and experience good

connections. At some point one or more existing users ask for more resources that cannot be

accommodated only on the LTE network. Three sub-cases are possible:

a. The WiFi network can accommodate all the extra traffic from LTE at per user level –

traffic will be routed only via Wi-Fi

b. The WiFi network can accommodate all the extra traffic from LTE at per session

level – Wi-Fi offloading

c. The WiFi network cannot accommodate all the extra traffic at per user or session

10

level. In this sub-case, the algorithms controlling offloading and network aggregation will also

attempt to optimize the network load between the WiFi and LTE network but it is clear that we are

dealing with congestion and a future upgrade should be decided.

2. All the users have data sessions open on LTE or Wi-Fi and experience good

connections. At some point one or more new users ask for more resources that cannot be

accommodated only on the LTE network. Subcases 1b and 1c apply here. Also, a new sub-case is

identified when considering that the Wi-Fi network can accommodate all the extra traffic from LTE

at per user level. The difference from 1a is that no Wi-Fi offloading is needed here as some of the

users will connect directly on WiFi and the rest will connect directly on LTE

a) Fixed scenario b) Mobile scenario

Figure 3.1 Usage scenario for WiFi offloading

It can be seen that in all the use cases it is envisioned that LTE data connection is preferred. In

case no resources are available on the LTE network, then the extra traffic is moved to the Wi-Fi

network. However it should be emphasized that the offloading should be bidirectional (from Wi-Fi

to LTE, from LTE to Wi-Fi). Moreover, it should be noted that not only the users asking for more

resources are viable to offloading but also some already running applications (non-real time ones

like web browsing or e-mail download) can be moved from one access network to another in order

to use at maximum the capacity provided by WiFi and LTE.

It is thus mandatory to define offloading algorithms based on quality aware real-time metrics

that can predict, observe and compare the load of the two access network types (Wi-Fi and 3G/4G).

Moreover, there is the possibility of aggregating WiFi and 3G/4G access networks mainly to

enhance throughput [23]. This aggregation refers to the case in which it would be possible for a user

running for example two services simultaneously (i.e mail and video streaming) to have one service

transported over WiFi and one transported over 3G/4G. The decision for each of the service to be

routed on one access network or the other will be performed using offloading algorithms in case the

needs of the users cannot be accommodated by one single access network.

The code operating on the rules for offloading will reside both in the end-user device, and in

the fixed/mobile unit, meaning it is divided between the two. In order for this to happen, when the

user connects to the hotspot for the first time, he will be prompted to download and install a custom

application that, among other features, will ensure the necessary control for the UE to offload

11

traffic. This will be called the ALO34 Wi-Fi offloading client.

3.3 Uncoordinated deployment of small cells

This section proposes to study the uncoordinated deployment of small cells, with a main

research focus on access network deployment and optimization based on dynamic network

configuration and process automation. Small cells are supposed to provide additional capacity and

improved coverage, but this capacity gets partially lost due to the increased interference and the

inability of the network operator to manually configure the smaller cell to be properly detected by

mobile devices and utilized. In the currently proposed LTE heterogeneous scenarios/networks

(HetNet), small cells can be deployed with minimal pre-planning in order to fulfil temporary needs

for service. In an uncoordinated deployment, neither the position, nor the number of cells is known

a priori. Also, the simple HetNet base stations (HeNB) do not require interface links (e.g., X2) to

share information and coordinate actions with other cells.

The main vision is to enable additional capacity, improved coverage, and energy/cost-efficient

networks by the design, analysis, development and evaluation of procedures for fast network

deployment and dynamic reconFiguretion of small cells in emergency and vehicular scenarios. The

processes needed to deploy small cells economically (how to deploy, where to deploy, how to deal

with the increased number of small cell sites, etc.) have to be studied.

The following scenarios are considered:

a. Emergency and congestion scenario

While the deployment of temporary mobile networks and other wireless equipment following

disasters has been successfully accomplished by governmental agencies and network providers in

the aftermath of previous disasters, there appears to be little optimization effort involved with

respect to maximizing key performance measures of the deployment or minimizing overall cost to

deploy [24]. One method for ensuring that wireless communication in a region affected by a disaster

is available is to utilize portable mobile network base stations that can be deployed when conditions

are appropriate (Figure 5.2). Uncoordinated deployment of small cells can provide a fast, efficient

and non-intrusive way of realizing a communication network that will benefit both first-responders

as well as affected citizens/users or simply to ease off congestion in critical traffic loads (e.g., sport

events) (Figure 5.3). The focus is on modelling an optimal deployment of uncoordinated small cells

that can be utilized prior to, during, and immediately following an emergency or disaster or simply

during very high traffic loads (e.g., sport events).

Figure 3.2 Emergency deployment scenario (the dashed cell in the right figures crashed and is

replaced with newly deployed cells)

12

Figure 3.3 Congestion deployment scenario (the users’ concentration in a small area requires a newly

deployed cell)

b. Mobile cells scenario

In order to ensure a good connectivity for most of the users, the deployment of small cells in a

high-speed vehicle (train, bus, car etc.) is an option (Figure 5.4). In this particular case the mobile

cell movement imposes a very fast and unpredictable reconFiguretion of the access network and

also of the network at the S1 interface level. A good optimization of the access network will

significantly decrease the spectrum and energy used by these types of vehicular access to the

network

Figure 3.4 Mobile cells scenario (traffic generated by the users’ concentration in a vehicle is ensured

efficiently by a new deployed mobile cell)

c. Traffic loading and traffic management between BSs to save energy

In the heterogeneous scenarios of 3GPP LTE (Long Term Evolution) the load imbalances

degrade performance. It is widely accepted that the primary approach for fair user service, while

maximizing network-wide throughput is to balance the load between cells [25]. Load balancing

algorithms are usually defined as resource allocation problems with signal quality restrictions.

Published research [26] shows that whenever possible, the BS could be shut down and the traffic

could be moved to another BS, or sometimes it is possible to shut a sector down and use

omnidirectional antennas to save energy (Figure 5.5). A good reconfiguration of the network radio

resources will lead to a significant economy in terms of energy and spectrum usage.

13

Figure 3.5 Reduced power and sleep mode scenario (the inexistence of active users in a cell sector gives the

possibility to activate a sleep mode for that sector, at least for one aggregated band)

3.4 Conclusions

The present chapter has proposed two solutions for radio access network optimization in LTE-

Advanced, Wi-Fi offloading and small cell deployment. A hardware and software platform for Wi-

Fi offloading has been proposed. The platform is suitable to be used in a crowded environment both

in fixed and mobile scenarios. Also, the uncoordinated deployment of small cells in similar

scenarios has been studied. We conclude that scheduling and admission control algorithms have to

be distributed and minimize the need for coordination.

14

Chapter 4 Resource scheduling and admission control

algorithms in LTE-Advanced

4.1 Introduction

Scheduling can be useful for enhancing user experience by providing higher throughput or

minimizing the interference. Likewise load balancing, which is usually associated with admission

control in conventional cellular networks, can play a role in ensuring that mobile subscribers

maintain their QoS requirements. [2,7].

4.2 Scheduling algorithms for systems with spectrum aggregation

4.2.1 Algorithms based on Proportional Fair

The algorithms presented in Chapter 2 can be adapted for spectrum aggregation systems in

the following way:

The PF index defined in (2.1) can be computed independently for each Component Carrier

and each user is scheduled independently according to the PF index value. This is termed as

the independent PF scheduling algorithm.

When computing the historical average user throughput, all the Component Carriers can be

considered [10]. Thus, the index will be computed as in (4.1)

N

j

ji

tji

tji

D

DF

1

,

,,

,,

(4.1)

where:

Di,j,t is the instantaneous throughput for user i, on carrier j and PRB t;

𝐷𝑖,𝑗 is the historical average throughput for user i and carrier j. N is the number of

aggregated carriers.

In this variant, the algorithm is the same on each component carrier, but now the

instantaneous user throughput will be divided by the aggregated throughput achieved on the

aggregated carrier. This variant is termed joint PF scheduling algorithm.

The only change required for implementing this algorithm is to sum the historical average

user throughput for each carrier. Since, at eNodeB level, these numbers are already known, all that

remains is a simple sum operation.

4.2.2 Algorithms based on Round Robin

For Round Robin, the following are considered:

The PRBs can be allocated equally to users on each component carrier independently. The

Round Robin algorithm will be the same, only it will be run independently on each CC. This

algorithm can be termed independent Round Robin scheduling algorithm.

The aggregated carrier can be considered and the total number of PRBs can be allocated

equally to the entire number of users. Here, we have to observe that the total number of

15

users is not necessarily the sum of users connected to each component carrier, because some

users may not be capable of CA or they only use one carrier. This is where a number of

optimizations can be performed.

4.2.3 Scheduling algorithm for LTE-Advanced based on optimizing the resource

allocation on the aggregated carrier (Cross-CC Round Robin)

The following proposes a scheduling algorithm based on the theoretical considerations in

4.2.2. Because it does not take into account Channel State Information, it has been included in the

Round Robin family of algorithms.

The following Profit Function is proposed in order to maximize the total throughput of the

user

,:)(1 1

cN

c

N

u

cuxPF (4.2)

In the above equation, xcu is the allocation variable. Also Nc is the numbered of configured

component carriers (CCs) in the cell and N is the number of users attached to the cell. The

allocation variable xcu reflects the resource allocation (i.e. allocated PRBs) for a specific user u on a

specific component carrier c. We can therefore define the allocation variable as

:

cu

cunnxcu

carrier component on user toallocated are PRBs no if,0

carrier component on user toallocated are PRBs if,. (4.3)

Therefore xcu is an NxNc matrix and highlights the cross-CC resource allocation (PRBs) to

users.

The above problem is a convex optimization problem because xcu is a constant for each

given TTI [11]. In order to solve it, the following constraints are defined:

1. Allocation Constraint: the sum of allocated PRBs for each user u has to be equal to the

number of PRBs available for each CC. This can be written as

,...,,2,1,1

c

N

u

PRBcu NcNxc

,

(4.4)

where cPRBN represents the number of PRBs available for component carrier c.

2. User constraint: the allocated PRBs on each component carrier c for each user u are

forcefully 0, if the user is not configured on the respective component carrier. This can be

written as

cuxcu carrier component toattachednot is user if,0 (4.5)

The objective is, therefore, to maximize the Profit Function, so, for each component carrier c

and user u, a number of PRBs x*cu is allocated, according to:

.maxarg* PFxcux

cu (4.6)

Therefore, the resource allocation or scheduling will be the solution of the above optimization

problem.

16

4.3 Multi-Carrier Scheduling Algorithm (MCSA)

4.3.1 Introduction

Carrier aggregation adds a new dimension for the scheduling of the users in LTE-Advanced

and raises an optimisation problem for the best use of network resources. The following Profit

Function is proposed in order to maximize the total throughput of the cell [12]

(𝑃𝐹 − 𝑀𝐶𝑆𝐴) ∑ ∑ 𝑤𝑐𝑢𝑥𝑐𝑢

𝑁

𝑢=1

𝑁𝑐

𝑐=1

, (4.7)

where xcu is the allocation variable and wcu is a normalized metric. Also Nc is the number of

configured component carriers (CCs) in the cell and N is the number of users attached to the cell.

The allocation variable xcu reflects the resource allocation (i.e. allocated PRBs) for a specific

user u on a specific component carrier c. We therefore define the allocation variable as

𝑥𝑐𝑢 = {0, if no PRB is allocated𝑛, if 𝑛 PRBs are allocated to user 𝑢 on carrier 𝑐

. (4.8)

The normalized metric wcu should reflect the requested user data rate according to a specific

service class and also the average throughput experienced by user u. The normalized metric is

defined as

𝑤𝑐𝑢 =

𝑇𝑢

𝑅𝑢, (4.9)

where Tu is the current estimated throughput for user u and Ru is the average received throughput

for user u.

The optimization problem is defined as

max𝑥𝑐𝑢

∑ ∑ 𝑤𝑐𝑢𝑥𝑐𝑢

𝑁

𝑢=1

,

𝑁𝑐

𝑐=1

(4.10)

meaning the values xcu have to be found such that the function is maximized.

4.3.2 Defining constraints

In order to solve this problem we need to define a number of constraints on the allocation

variable [11]:

1. Allocation Constraint: each user u with an application belonging to a certain class, ku, has

to be allocated a minimum number of PRBs 𝑁𝑃𝑅𝐵𝑚𝑖𝑛𝑘𝑢 in order to meet the service

requirements for that class. At the same time, the number of allocated PRBs must not exceed

an upper bound such that all users belonging to the same class and attached to the same cell

as user u can be served

𝐿𝐵 ≤ ∑ 𝑥𝑐𝑢 ≤ 𝑈𝐵

𝑁𝑐

𝑐=1

, ∀𝑢 ∈ 𝐷(𝑁), (4.11)

where

𝐿𝐵 = 𝑁𝑃𝑅𝐵𝑚𝑖𝑛𝑘𝑢, (4.12)

and

17

𝑈𝐵 = ∑ [𝑁𝑃𝑅𝐵𝑐

𝑁𝑢𝑠𝑒𝑟𝑠𝑘𝑐𝑢

] .

𝑁

𝑐=1

(4.13)

With 𝑁𝑃𝑅𝐵𝑐 representing the number of available PRBs per component carrier c, 𝑁𝑢𝑠𝑒𝑟𝑠𝑘𝑐𝑢

representing the number of users with the same QoS priority as user u and attached to carrier

c. Also D(N) represents the subset of users running the same delay sensitive application.

2. Bandwidth constraint: the total number of active users on each carrier is upper bounded by

the maximum normalized load that can be handled on that carrier, 𝐿𝑐𝑚𝑎𝑥 ∈ [0,1]

∑𝑆𝑟𝑎𝑡𝑒𝑢

𝑅𝑐𝑒𝑙𝑙𝑐𝑢

∙ 𝑥𝑐𝑢 ≤ 𝐿𝑐𝑚𝑎𝑥

𝑁

𝑢=1

, ∀𝑐 ∈ {1, … , 𝑁𝑐}. (4.14)

4.4 Enhanced MCSA algorithm (E-MCSA)

4.4.1 Introduction

The MCSA algorithm can be enhanced if we consider that the value of 𝑛𝑃𝑅𝐵𝑚𝑖𝑛 , as previously

defined in MCSA, is fixed. This does not guarantee that the user that is scheduled 𝑛𝑃𝑅𝐵𝑚𝑖𝑛

number of PRBs will get his guaranteed throughput.

4.4.2 Deriving 𝒏𝑷𝑹𝑩𝒎𝒊𝒏 from the minimum guaranteed throughput

Let’s denote 𝑇ℎ𝑚𝑖𝑛 as the minimum guaranteed user throughput. According to the Shannon-

Hartley theorem we can write

𝐶 = 𝐵𝑊 ∙ log2(1 + 𝑆𝐼𝑁𝑅), (4.15)

or

𝐵𝑊𝑢 =

𝑇ℎ𝑚𝑖𝑛

log2(1 + 𝑆𝐼𝑁𝑅𝑒𝑓𝑓𝑢 )

(4.16)

Finally, the minimum number of PRBs can now be determined by simply dividing the user

bandwidth to the bandwidth of a PRB (180 kHz for LTE [18])

𝑛𝑃𝑅𝐵𝑚𝑖𝑛𝑢

= ⌊𝐵𝑊𝑢

𝐵𝑊𝑃𝑅𝐵⌋ + 1, (4.17)

Here 𝐵𝑊𝑃𝑅𝐵 represents the bandwidth of a PRB and ⌊∙⌋ represents the floor function.

The enhanced algorithm will, thus, be similar to the one defined in 4.3.1 with one exception.

Now 𝑛𝑃𝑅𝐵𝑚𝑖𝑛𝑢 will not be fixed but will vary for each user u. We will call this improvement the E-

MCSA (Enhanced MCSA) algorithm.

4.5 Conclusion

The present chapter proposed several scheduling algorithms for LTE-Advanced. These have

been defined as convex optimization problems, for which the necessary constraints have been

defined. Also proof that the proposed problems are feasible have been given.

18

Chapter 5 Results

5.1 Main results

The main results are given in the Annexes, as reprint of papers published.

5.2 Conclusions

The present chapter has analyzed the algorithms proposed in Chapter 4 and this enables to

draw some conclusions.

In the case of a macro-only scenario, it is not justified to use an algorithm that is

computationally intensive like Cross-CC RR instead of Round Robin because the performances are

similar. We can notice that Cross-CC PF which has been presented in the literature as an algorithm

for LTE-Advanced, has worst performance than RR.

The MCSA and E-MCSA algorithms do not increase the user throughput in a macro-noly

scenario. In most cases, the PF scheduling algorithm offers the highest user throughput.

Things change though, in the case of a HetNet scenario. Here, we can notice that for users

with QoS requirements, E-MCSA is the best choice, but for users with no QoS requirements it is the

worst choice. We can say that MCSA offers a gain in throughput for users with QoS requirements,

although lower than the one offered by E-MCSA, but also a throughput that is closet o the one

offered by traditional scheduling algorithms (PF, RR). Table 5.1 sums up the considerations

expressed before.

Table 5.1 Recommended scheduling algorithms according to the number of users in a cell and scenario

Scenario Number of users in cell Recommended algorithm

Macro-only Low Round Robin

High Proportional Fair

Macro-only, LTE Rel. 8 users Low E-MCSA

High Round Robin

Macro-only, LTE Rel. 10 users - Proportional Fair

Macro-only, users with no QoS

requirements

Low Round Robin

High Proportional Fair

Macro-only, users with QoS

requirements

Low E-MCSA

High Proportional Fair

Macro-only, LTE Rel. 8 users

with no QoS requirements - Cross-CC PF

Macro-only, LTE Rel. 8 users

with QoS requirements - E-MCSA

Macro-only, LTE Rel. 10 users

with no QoS requirements

Low MCSA

High Proportional Fair

Macro-only, LTE Rel. 10 users

with QoS requirements

Low Round Robin

High Proportional Fair

HetNet - MCSA

HetNet, users with no QoS

requirements

- MCSA

HetNet, users with no QoS

requirements

- E-MCSA

HetNet, LTE Rel. 8 users with

no QoS requirements

- MCSA

19

HetNet, LTE Rel. 8 users with

QoS requirements

- E-MCSA

HetNet, LTE Rel. 10 users with

no QoS requirements

- Proportional Fair

HetNet, LTE Rel. 10 users with

QoS requirements

- MCSA/E-MCSA

20

Chapter 6 Conclusions and future work

6.1 General conclusions

The thesis proposes several optimizations of the radio access network, with regards to

capacity and throughput. The main categories of scheduling algorithms in LTE have been outlined

and modifications that the use of the Carrier Aggregation technique brings have been proposed and

discussed.

Next, two solutions for radio access network optimization in LTE-Advanced systems have

been described and theoretical and practical considerations have been given. Finally it is concluded

that scheduling and admission control algorithms have to be distributed and minimize the need for

coordination.

Starting from the idea that carrier aggregation introduces an additional dimension in user

scheduling, a number of resource scheduling algorithms were proposed. Because some of them

were defined as a convex optimization problem, admission control algorithms had to be developed,

in order to avoid that the optimization problem becomes unfeasible or unbounded.

The results that were obtained by comparing the algorithms between them have shown that

there are some scenarios in which the algorithms that take into account QoS requirements have

results that are superior to the ones that are traditionally used. Also, the users with QoS

requirements are more favored by these types of algorithms,. However, in macro-only scenarios, the

proposed algorithms give rather poor results, and the performances are similar to the traditional

ones.

6.2 Future work

The use of CA and the further evolution of LTE paves the way for a great number of

optimizations that can be done not only in resource scheduling. Future work could include: ●

Evaluating the performances of the proposed algorithms in scenarios where more than 2 LTE

carriers are aggregated. The code that the author has written allows the use of 1 to 5 CCs, but some

implementation issues have limited the number of CCs to only 2. ● Extending the LTE-Advanced

simulator for inter-band carrier aggregation. This can also serve as basis for experimenting with

different other algorithms ● Evaluating the proposed MCSA/E-MCSA algorithms in scenarios

where there are users with different QoS requirements. This can be done with some modifications

of the source code. ● Physically implementing the algorithms on a configurable hardware platform

like FPGA.

6.3 Main contributions of the author

The main contributions of the author are as follows: ● Developing a simple multi-carrier

scheduling algorithm for LTE-Advanced that takes into account the existence of multiple CCs and

optimises the resource allocation ● Implementing an LTE-Advanced system in a simulator, starting

from an LTE simulator ● Implementing the proposed algorithm in the developed simulator ●

Modifying existent scheduling algorithms, previously used for LTE, to be deployed in LTE-A and

implementing them in the developed simulator. ● Implementing a scheduling algorithm for LTE-

Advanced (found in the literature) in the developed algorithm ● Evaluating the performances of the

21

simple algorithm in comparison to the traditional algorithms ● Developing a second scheduling

algorithm for LTE-Advanced that takes into account the existence of multiple aggregated carriers

and the user QoS requirements ● Developing an admission control algorithm, necessary for the

above mentioned scheduling algorithm to work ● Implementing the second proposed scheduling

algorithm in the developed simulator ● Implementing the proposed admission control algorithm in

the developed simulator ● Evaluating the performance of the second proposed algorithm compared

to the traditional scheduling algorithms ● Enhancing the second proposed algorithm as a result of

the performance evaluation ● Evaluating the performance of the enhanced algorithm compared to

the traditional scheduling algorithms and the initial algorithm ● Evaluating the performance of all

the proposed algorithms compared to the traditional scheduling algorithms and compared between

themselves ● Highlighting the scenarios where the algorithms have superior performances, as well

as drawing conclusions about the situations and scenarios where each algorithm is recommended to

be used.

Besides the research activity within the doctoral studies, the author has also contributed to

proposal writing and has worked in projects coordinated by Prof. Dr. Ing. Simona Halunga and

Prof. Dr. Ing. Octavian Fratu. Among these projects the most relevant are the following: ● Wireless

Hybrid Access System with Unique Addressing (SAWHAU). Contract no. 12-126/01.10.2008,

PNCDI II partnership research project (2008-2011) ● Evolution, implementation and transition

methods of DVB radiobroadcasting using efficiently the radio frequencies spectrum. Contract no.

106/2011, Ministry of Communications and Information Society of Romania project (2011-2013) ●

Scalable Radio Transceiver for Instrumental Wireless Sensor Networks (SaRaT-IWSN). Contract

no. 20/2012, UEFISCDI Romania project (2012-2015) ● Optimization and Rational Use of Wireless

Communication Bands (ORCA), NATO project no. SfP-984409 (2013-2015) ● „eWALL for Active

Long Living (eWALL)”, FP7 project no. 610658 (2013-2016).

6.4 List of publications

1. Vulpe, A.; Obreja, S.; Fratu, O., "Interoperability procedures between access technologies

using IEEE 802.21," Wireless Communication, Vehicular Technology, Information Theory

and Aerospace & Electronic Systems Technology (Wireless VITAE), 2011 2nd International

Conference on , vol., no., pp.1,5, Feb. 28 2011-March 3 2011, doi:

10.1109/WIRELESSVITAE.2011.5940886 (IEEEXplore)

2. Vulpe, A.; Fratu, O., "A MIH-enabled emulated network for streaming applications,"

Telecommunication in Modern Satellite Cable and Broadcasting Services (TELSIKS), 2011

10th International Conference on , vol.2, no., pp.658,661, 5-8 Oct. 2011, doi:

10.1109/TELSKS.2011.6143198 (IEEEXplore)

3. Fratu, O.; Popovici, E.C.; Vulpe, A.; Halunga, S.V., "Heterogeneous wireless access

networks analysis from simulation to implementation," Telecommunication in Modern

Satellite Cable and Broadcasting Services (TELSIKS), 2011 10th International Conference

on , vol.2, no., pp.481,488, 5-8 Oct. 2011, doi: 10.1109/TELSKS.2011.6143248

(IEEEXplore)

4. Vulpe, A.; Fratu, O.; , "An Evaluation of an UMTS/WLAN Interworking Architecture using

IEEE 802.21," Information, Communication and Energy Systems and Technologies (ICEST),

2012 47th International Scientific Conference on , pp.126-129, 28-30 Jun. 2012, ISBN: 978-

22

619-167-002-4

5. Vulpe, A.; Fratu, O.; Craciunescu, R., "Performance evaluation of heterogeneous

interworking using IEEE 802.21," Telecommunications Forum (TELFOR), 2012 20th , vol.,

no., pp.498,501, 20-22 Nov. 2012, doi: 10.1109/TELFOR.2012.6419256 (ISI, IEEExplore)

6. Vulpe, A.; Fratu, O.; Mihovska, A.; Prasad, R., “A Multi-Carrier Scheduling Algorithm for

LTE-Advanced,” Wireless Personal Multimedia Communications (WPMC), 2013 15th

International Symposium on, pp. 1-5, 24-28 Jun. 2013, ISBN 978-87-92982-52-0 (ISI,

IEEEXplore)

7. Marțian, A.; Crăciunescu, R.; Vulpe, A.; Fratu, O.; Marghescu, I., “Perspectives on Dynamic

Spectrum Access Procedures in TV White Spaces”, Global Wireless Summit (GWS), 2013 –

Special Session on Recent Advances in Spectrum Measurements and Modelling towards

Flexible Spectrum Usage, pp. 1-5, 24-28 Jun. 2013, ISBN 978-87-92982-52-0 (ISI,

IEEEXplore)

8. Mihai, G.; Vulpe, A.; Crăciunescu, R.; Fratu, O., „LTE Interference Analysis in an Urban

Area,” 11th International Conference on Electronics, Telecommunications and Applied

Informatics (ETAI 2013), pp.1-6, 26-28 Sept. 2013, ISBN 978-9989-630-68-2

9. Cojocaru, C.A.; Vulpe, A.; Crăciunescu, R.; Fratu, O., „The Planning of a Large Scale

Sensor Network using ICS-Telecom,” 11th International Conference on Electronics,

Telecommunications and Applied Informatics (ETAI 2013), pp.1-4, 26-28 Sept. 2013, ISBN

978-9989-630-68-2

10. Fratu, O.; Vulpe, A.; Halunga, S.; Crăciunescu, R., „Interference Analysis for Inter-Band

Carrier Aggregation in LTE-Advanced, ” 11th International Conference on

Telecommunications in Modern Satellite, Cable and Broadcasting Services (TELSIKS 2013),

pp. 1-10, 16-19 Oct. 2013, ISBN 978-1-4799-0900-1 (IEEEXplore)

11. Crăciunescu, R.; Voicu, C.; Vulpe, A.; Halunga, S., “Performance analysis of MC-CDMA

system when image transmission is involved,” Communications (COMM), 2014 10th

International Conference on, pp. 377-380, 29-31 Mai 2014, ISBN 978-1-4799-2385-4 (ISI,

IEEEXplore)

12. Vulpe, A.; Fratu. O.; Halunga, S., “Downlink Packet Scheduling in LTE-Advanced

Heterogeneous Networks”, University Politehnica of Bucharest Scientific Bulletin, accepted

(CNCSIS B+)

13. Fratu. O.; Vulpe, A.; Crăciunescu, R.; Halunga, S., “Small Cells in Cellular Networks –

Challenges of Future HetNets”, Wireless Personal Communications, Springer, accepted

(ISI)

14. Popovici, E.; Fratu, O.; Vulpe, A.; Halunga, S., “New Developments in Vertical Handover

based on Media Independent Handover Standard”, Wireless Personal Communications,

accepted (ISI)

15. Vulpe, A.; Fratu, O.; Mihovska, A.; Prasad, R., ”Admission Control and Scheduling

Algorithm for Multi-Carrier Systems,” Wireless Personal Communications, submitted (ISI)

16. Suciu, G.; Vulpe, A.; Fratu, O.; ” Future Networks for Convergent Cloud and M2M

Multimedia Applications,”, Global Wireless Summit, pp. 1-5, Aalborg, Denmark, May 11-

14, 2014, (IEEEXplore)

17. Vulpe, A.; Fratu, O.; Halunga, S., ” QoS-Aware Downlink Scheduling in Multi-Carrier

23

Communication Systems,” Global Wireless Summit, pp. 1-5, Aalborg, Denmark, May 11-14,

2014, (IEEEXplore)

ACKNOWLEDGEMENT

Results presented in this paper were supported by the Sectoral Operational Programme

Human Resources Development 2007-2013 of the Romanian Ministry of Labour, Family and Social

Protection through the Financial Agreement POSDRU/107/1.5/S/76903

24

Bibliography

[1] “Cisco Visual Networking Index: Global Mobile data Traffic Forecast Update 2012–2017,”

Cisco Public Information, Febr., 2013.

[2] ETSI MCC, “Report of 3GPP TSG RAN IMT-Advanced Workshop,” 2008.

[3] E. Dahlman, S. Parkvall, and J. Skold, 4G: LTE/LTE-Advanced for Mobile Broadband.

Oxford, UK: Academic Press, 2011.

[4] H. Holma and A. Toskala, Lte for umts: evolution to lte-advanced. 2011.

[5] 3GPP TS 36.321, “Technical Specification Group Radio Access Network; Evolved Universal

Terrestrial Radio Access (E-UTRA); Medium Access Control (MAC) protocol specification

(Release 8),” 2010.

[6] 3GPP TS 36.133, “Technical Specification Group Radio Access Network; Evolved Universal

Terrestrial Radio Access (E-UTRA); Requirements for support of radio resource

management (Release 8),” 2010.

[7] E. C. Strinati, A. De Domenico, and A. Duda, “Ghost femtocells: A novel radio resource

management scheme for OFDMA based networks,” 2011 IEEE Wirel. Commun. Netw.

Conf., pp. 108–113, Mar. 2011.

[8] Q. Sun, Y. Zhang, S. Jin, and X. Gao, “Statistical Channel State Information Aided

Proportional Fair Scheduling Scheme for Highly Transmit Correlated Channels,” Wirel.

Pers. Commun., vol. 70, no. 4, pp. 1261–1283, Jul. 2012.

[9] A. Pokhariyal, K. I. Pedersen, G. Monghal, I. Z. Kovacs, C. Rosa, T. E. Kolding, and P. E.

Mogensen, “HARQ Aware Frequency Domain Packet Scheduler with Different Degrees of

Fairness for the UTRAN Long Term Evolution,” 2007 IEEE 65th Veh. Technol. Conf. -

VTC2007-Spring, 2007.

[10] Y. Wang, K. I. Pedersen, T. B. Sørensen, and P. E. Mogensen, “Carrier load balancing and

packet scheduling for multi-carrier systems,” Wirel. Commun. IEEE Trans., vol. 9, no. 5, pp.

1780–1789, 2010.

[11] S. Boyd and L. Vandenberghe, Convex optimization. 2004.

[12] A. Vulpe, O. Fratu, A. Mihovska, and R. Prasad, “A multi-carrier scheduling algorithm for

LTE-advanced,” Int. Conf. Wirel. Pers. Multimed. Commun. (WPMC), 2013 16th, pp. 1–5,

2013.

Wireless Pers CommunDOI 10.1007/s11277-014-2218-9

Admission Control and Scheduling Algorithmfor Multi-carrier Systems

Alexandru Vulpe · Albena Mihovska · Octavian Fratu ·Simona Halunga · Ramjee Prasad

© Springer Science+Business Media New York 2014

Abstract LTE-Advanced aims to provide a transmission bandwidth of 100 MHz by usingcarrier aggregation to aggregate LTE Rel. 8 carriers. In order to increase the system capac-ity, resource allocation becomes a very good tool, and, in the context of the existence ofmultiple component carriers in LTE-Advanced becomes a complex optimization problem.This paper proposes a Multi-Carrier Scheduling Algorithm that takes into account the user’sQoS requirements and also targets the maximization of the user throughput. The algorithmis evaluated in a scenario with both macro and femto base stations (i.e. a HetNet scenario)that respects the 3GPP specifications. Numerical results show that this algorithm has betterperformances than the traditional round robin and proportionally fair resource schedulingalgorithms.

Keywords 4G · Dynamic scheduling · Convex optimization · Simulation

1 Introduction

In multi-user scenarios, many users share a limited amount of resources, these resourcesbeing the medium through which the users communicate. Scheduling and resource allocation

A. Vulpe (B) · O. Fratu · S. HalungaFaculty of Electronics, Telecommunications and Information Technology, POLITEHNICA Universityof Bucharest, Splaiul Independentei, nr. 313, Sector 6, 060042 Bucharest, Romaniae-mail: [email protected]

O. Fratue-mail: [email protected]

S. Halungae-mail: [email protected]

A. Mihovska · R. PrasadCenter for TeleInfrastruktur (CTIF), Aalborg University, Aalborg, Denmarke-mail: [email protected]

R. Prasade-mail: [email protected]

123

A. Vulpe et al.

become essential components of wireless data systems because different users experiencedifferent fading conditions at the same time [1,2].

Dynamic scheduling means how physical layer resources are allocated to each user, in eachgiven time slot (time transmission interval), and how this is optimized. It is a very complexproblem, because it has to take into account different variables like the user radio conditons,the user’s traffic pattern or the user Quality of Service (QoS) and try to optimize the allocationof radio resources from the whole system point of view. Essentially, one has to decide whenand how the available resources in the cell are assigned to each of the users, where theresources depend on the access method. For users in an LTE or LTE-Advanced system, theusers are allocated physical resource blocks (PRBs). PRBs are defined as units consisting of12 consecutive OFDM subcarriers in the frequency domain and two consecutive time slotsor one LTE subframe in the time domain, which constitutes the minimum scheduling unit forLTE and LTE-Advanced [3].

In order for scheduling to be efficient, the awareness of the channel needs to be specified.In wireless systems available today, channel quality feedback is available, thus enabling adynamic adaptation of the scheduled users and the allocation of physical layer resources. Thishas led to a great deal of interest in the research community on scheduling and resource allo-cation algorithms. Some well-known scheduling algorithms include the max-C/I schedulingstrategy, round-robin scheduling and the proportional-fair scheduling. While the max-C/Ischeduling algorithms provides the highest cell throughput, it can lead to users that do nothave favourable channel conditions never being served. Likewise, the round robin schedulingalgorithm provides the “blind” allocation of users, without taking into account channel stateinformation (CSI), meaning the cell resources are equally divided amongst users.

Proportional fairness proposes a good tradeoff between maximising throughput and pro-viding data rate fairness. This type of scheduling has as objective the maximization of the longterm received throughput. Several variations of the algorithm have been proposed, includingmaximising the instantaneous throughput [4–6], taking into account QoS [7,8], consideringpower control [8]. However, all of these works have been targeted towards OFDM-basedsystems. With the proposal of carrier aggregation, these proportionally fair algorithms haveto be rethought and adapted for the existence of multiple component carriers (CCs).

For instance, authors in [9] proposed an Integrated Common Radio Resource Manage-ment framework that used spectrum aggregation. While their work dealt with aggregatingspectrum in different frequency bands, it was targeted towards HSDPA networks, not LTE orLTE-Advanced. An interesting sub-channel allocation algorithm in HetNets was proposed in[10]. The authors proposed an intelligent physical resource block (PRB) allocation as a solu-tion to mitigate the downlink intra-small cell interference as well as the inter-tier interferencein OFDM-based systems. The allocation of the PRBs was formulated as a graph coloringproblem, and solved using an ant colony optimization (ACO)-based approach. Their workhowever targeted power allocation and not packet scheduling. Finally, authors in [11] devel-oped a Best Minimum Summation (BMS) scheduling algorithm that dealt with schedulingtable in different dimensions (PRB-wise and UE-wise). Their work, however targeted onlyLTE and was restricted to these two-dimensions.

The main contribution of this paper is in formulating a joint two-dimensional (PRB-wiseand CC-wise) optimization of the radio resource scheduling that can be applied in LTE-Advanced systems, that is loosely based on proportional fair scheduling and takes into accountthe existence of user QoS requirements. Also, we propose admission control algorithm inorder to enable this scheduling algorithm. This work is based on preliminary work proposedin [12], where we defined the optimization problem and performed preliminary simulation

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

for validating the algorithm. Here we extend this work to include some refinements of thealgorithm and to detail the admission control algorithm used.

The paper is organized as follows: Sect. 2 describes the system model used to evaluate thealgorithm, Sect. 3 addresses the optimization problem posed by the Multi-Carrier SchedulingAlgorithm and Sect. 5 formulates some assumptions and describes the simulation scenarioused for performance evaluation. Sect. 6 discusses the results obtained with the proposedalgorithm and, finally, Sect. 7 makes the concluding remarks.

2 System Model

LTE Rel. 10 specifies the aggregation of up to 5 LTE Rel. 8 carriers, also known as componentcarriers (CCs), in order to achieve an overall bandwidth of 100 MHz. This work proposes analgorithm which enables the best user allocation over any number of CCs, with the objectiveof maximising the total user throughput, and maintaining QoS requirements.

Although the algorithm proposed in this work enables the scheduling of users on up tofive aggregated CCs, in this paper we chose to evaluate the algorithm on two CCs in the samefrequency band (the case of contiguous intraband carrier aggregation). Depending on thecapabilities of the user equipment (UE), each user may be allocated to a single LTE carrier,or simultaneously to two carriers. This does not restrict however the generality of the definedproblem.

The LTE-Advanced network is deployed with hexagonal cells and with frequency reusepattern one, in order to evaluate the worst-case scenario in terms of interference. All thecells deploy two CCs, with each CC transmitted from different antenna connectors. In thiscase, when the network is not coordinated, for each cell, the adjacent cells become a sourceof interference (but only on each CC) which leads to a deterioration of the user receivedsignal-to-interference-plus-noise (SINR) ratio and, consequently, a drop in user throughput.

Each time transmission interval (TTI) is associated with the duration of one of the tensubframes that makes up an LTE frame (which is 10 ms long), which corresponds to a sub-frame duration of 1 ms. For every TTI, the scheduler will allocate users depending on theused scheduling algorithm.

The Radio Resource Management framework allocates the available radio resources (i.e.PRBS) to the users in order to satisfy the user QoS requirements, while maximizing cell anduser throughput. It takes into account, as mentioned in the next sections, the CSI of the UE,in order to improve throughput and spectral efficiency.

3 Multi-Carrier Scheduling Algorithm

Carrier aggregation adds a new dimension for the scheduling of the users in LTE-Advancedand raises an optimisation problem for the best use of network resources. The followingProfit Function is proposed in order to maximize the total throughput of the cell:

Nc∑

c=1

N∑

u=1

wcu xcu (1)

where xcu is the allocation variable and wcu is a normalized metric. Also Nc is the numberedof configured component carriers (CCs) in the cell and N is the number of users attached tothe cell. The allocation variable xcu reflects the resource allocation (i.e. allocated PRBs) for a

123

A. Vulpe et al.

specific user u on a specific component carrier c. We therefore define the allocation variableas

xcu ={

n, if n PRBs are allocated to user u on component carrier c

0, if no PRBs are allocated to user u on component carrier c(2)

The normalized metric wcu should reflect the requested user data rate according to aspecific service class and also the average throughput experienced by user u. We define thenormalized metric as:

wcu = Tu

Ru(3)

where Tu is the current estimated throughput for user u and Ru is the average receivedthroughput for user u.

The following assumption is made: the users are running two different services withdifferent data rate requirements. One class of services is a delay sensitive application i.e. theuser has to have a minimum amount of resources (PRBs) allocated every TTI, and the otheris a delay tolerant application i.e. the users do not have to be allocated PRBs every TTI.

Therefore, the optimization problem becomes

maxxcu

Nc∑

c=1

N∑

u=1

wcu xcu (4)

This means that we must find the values of xcu that maximize the proposed profit function.

3.1 Constraint Definitions

The above problem is a convex optimization problem (Proof in Appendix 1) for which thefollowing constraints are defined:

1. Allocation constraint: each user u with an application belonging to a certain class hasto be allocated a minimum number of PRBs, NP RBminku

, in order to meet the servicerequirements for that class. At the same time, the number of allocated PRBs must notexceed an upper bound such that all users belonging to the same class and attached tothe same cell as user u can be served:

L B ≤Nc∑

c=1

xcu ≤ U B, ∀u ∈ D(N ) (5)

whereL B = NP RBminku

(6)

and

U B =Nc∑

c=1

[NP RBc

Nuserskcu

](7)

with NP RBc representing the number of available PRBs per component carrier c andNuserskcu representing the number of users with the same QoS priority as user u andattached to carrier c. Also D(N ) represents the subset of users running the delay sensitiveapplication.

D(N ) = {u ∈ 1, ..., N |u is running delay-sensitive application

and is attached to carrier c} (8)

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

2. Bandwidth constraint: the total number of active users on each carrier is upper boundedby the maximum normalised load that can be handled on that carrier, Lmax

c ∈ [0, 1]:N∑

u=1

Srateu

Rcellcu

· xcu ≤ Lmaxc ,∀c ∈ {1, ..., Nc} (9)

where the first term represents the requested service data rate for user u per PRB nor-malized with the maximum data rate that the cell can offer on component carrier c foruser u.

3.2 Deriving Upper Bounds

One must take into account the fact that there can be users that are connected to one or severalCC, therefore the number of total users that are running the same delay sensitive applicationattached to the cell may be lower than the sum of users connected to each of the componentcarriers of the cell.

Nusersku≤

Nc∑

c=1

Nuserskcu, ∀u ∈ D(N ) (10)

We can note also that:

Nusersku≥ max{Nuserskiu

|i ∈ {1, ...Nc}} (11)

Also, the number of users with delay sensitive application on an individual CC is upperbounded by:

Nuserskcu≤

[NP RBc

NP RBminku

](12)

Taking into account the fact that L B < U B the following formula yields (Proof inAppendix 2):

Nusersku<

NP RBtot

NP RBminku

(13)

Equations 12 and 13 give the upper bounds for the number of users that can be connectedto a single component carrier and to the entire aggregated carrier, respectively.

We must note that this will not be valid if there exists only one delay tolerant applicationclass.

3.3 Deriving the Minimum Number of PRBs

The MCSA algorithm defined in 3 can be enhanced if we consider that the value of NP RBminkuas defined therein is a fixed one. This does not guarantee that the user that will be scheduledthis amount of PRBs will receive his minimum guaranteed throughput. Therefore we mustderive the minimum number of PRBs taking into account the SINR value experienced by theuser.

We can denote T hmin as the minimum user guaranteeed throughput. Taking into accountwhat the used simulator can compute, we use the downlink SINR for each PRB and eachcomponent carrier. Knowing the SINR for each PRB, we can compute the effective SINR ofa user for a component carrier by using the Exponential Effective SINR mapping (EESM) as:

123

A. Vulpe et al.

SI N Ref f = −λ · ln(

1

NP RB

NP RB∑

i=1

e−SI N Ri

λ

)(14)

where λ is a constant that depends on the Modulation and Coding Scheme (MCS) used,NP RB is the total number of PRBs available for the component carrier, and SI N Ri is theSINR for the PRB with the ith index.

With this knowledge, we can find out the necessary bandwith to achieve guaranteedthroughput for user u from the Shannon-Hartley theorem.

BWu = T hmin

log2

(1+ SI N Ru

ef f

) (15)

Finally, the minimum number of PRBs can now be determined by simply dividing theuser bandwidth to the bandwidth of a PRB (180 kHz for LTE)

n P RBminu=

⌊BWu

BWP RB

⌋+ 1 (16)

Here, BWP RB represents the bandwidth of a PRB and �·� represents the floor function.The MCSA algorithm will therefore be the same with the only difference being that the

lower bound defined in Eq. 6 will vary for each user on each CC, at each TTI and will beequal to the value defined in Eq. 16. We will call this improvement the Enhanced MCSA(E-MCSA) algorithm.

4 Admission Control Algorithm

Starting from the upper bounds derived in Sect. 3.2, an admission control algorithm has tobe developed. This is because, otherwise, the optimization problem defined in Sect. 3 canbecome unbounded or unfeasible. With this in mind we proposed algorithm 1.

Algorithm 1 Admission control for macro-only scenarios1: Get cell that offers highest power for UE: C1

2: if u ∈ D(N ) and Nusersku<

NP R BtotNP R Bminku

then

3: U E ← C14: STOP5: else6: Remove C17: Go to 18: end if

Also, the choice of which cell a user should be attached to must be made taking intoconsideration the cell that can better serve each user. One simple strategy could be to choosethe cell that offers the highest nominal power at the UE position. However, due to the bigdifference of total available transmit power between macro BSs and femto BSs, the choicewould almost always lie in the macro-cell since the Pathloss difference would not cover thetransmit power difference.

Taking these into account, algorithm 2 is proposed and used. Please note that some nota-tions from Sect. 3 are used in both algorithms.

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

Algorithm 2 Admission control for HetNet scenarios1: Get cell that offers highest power for UE: C12: Get cell that offers second highest power for UE: C2

3: if (C1 is femto and u /∈ D(N )) or u ∈ D(N ) and Nusersku<

NP R BtotNP R Bminku

then

4: U E ← C15: STOP6: else7: if (C2 is macro and u /∈ D(N )) or u ∈ D(N )) and Nusersku

<NP R Btot

NP R Bminku

then

8: U E ← C19: STOP10: else11: Compute SINR if only C1 and C2 exists in the network and UE connected to C1 (macro)

12: if (SI N R > 10d B and u /∈ D(N )) or Nusersku<

NP R BtotNP R Bminku

then

13: U E ← C114: STOP15: else16: Estimate distance between C2 and UE: r17: if r < 40m and u /∈ D(N ) then18: U E ← C119: STOP20: else21: U E ← C222: STOP23: end if24: end if25: end if26: end if

5 Simulation Scenario and Assumptions

The network scenarios that have been analysed were chosen taking into account the carrieraggregation scenarios presented by 3GPP. We designed our own MATLAB LTE simulator,and enhanced it to support carrier aggregation. Following the recommendations of 3GPP[13] the worst interference-case scenarios were considered, where there is frequency reuse1 between the femto-cells and macro cells.

The scenario simulated here represents an urban environment with 19 sites with 3 macroBS per site. The macro BSs have a height of 30 m and an electrical downtilt of 15◦ following3GPP specifications. For each macro BS there is a femto BS randomly placed in its coveragearea. Fig. 1 shows the scenario topology.

There are a number of users per femto and macro cells, that are randomly placed insidethe coverage area of each type of cell. This does not necessarily mean that the femto userscannot connect to a macro cell or the other way aroud. Which cell the user connects to isdecided by the algorithms described in Sect. 4

The PathLoss Model is Model 2 proposed by 3GPP, which considers different PathLossesfor Line of sight (LoS) and Non-LoS (NLoS) conditions, where the transmitter and thereceiver do not have a direct channel between them [13].

To solve problem 1 we used CVX, a package for specifying and solving convex programs[14,15].

Table 1 lists the main simulation parameters used for this work. All other parameters arealso taken from the 3GPP specifications.

123

A. Vulpe et al.

−1500 −1000 −500 0 500 1000 1500−1500

−1000

−500

0

500

1000

1500

12

3

45

6

78

91011

12

1314

15 1617

18

1920

21

2223

24

2526

27

2829

303132

333435

36

3738

39

4041

42 4344

45 4647

48

4950

51

5253

54

5556

571

2

3

4

5

6

78

9

101112

1314

1516

17

18

1920

21

22

23

24

2526

27

2829

303132

333435

36

37

38

39

404142

4344

45

464748

4950

51

5253

54

555657

Fig. 1 Simulated scenario topology

Table 1 Main simulationparameters

Parameter Value

Topology Hexagonal grid, 19 macro BSs,3 sectors per BS, 19 femto BS

Carrier frequency 2 GHz

Component carrier bandwidth 20 MHz

Number of component carriers 2

Duplexing FDD

Antenna pattern for macro BS As described in [13]

Antenna pattern for femto BS Omnidirectional

Inter-site distance 500 m between macro

Distance between macro and femto 75 m

Distance between femto 40 m

Tx power for femtocell 30 dBm

Tx power for macrocell 49 dBm

Path loss As described in [13]

Shadowing standard deviation 8 dB

Penetration loss 20 dB

Number of users per femtocell 10

Number of users per macrocell 20

User distribution in site area Uniform

Traffic model Full buffer

User guaranteed bit rate 2 Mbps

6 Results

Figure 2 shows the average cell throughput obtained with two variants of the proposedscheduling algorithm and compared with the round robin and proportionally fair algorithms.

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

Fig. 2 Comparison of macro and femto cell throughputs between the different scheduling algorithms

Fig. 3 Comparison of users with and without QoS requirements between the different scheduling algorithms

It can be seen that there are only marginal differences between them both in macro cellsand femto cells. This has to be expected, since the objective of our proposed algorithm is tomaximize the user throughput.

Figure 3 shows the average user throughput for both users that have and do not have QoSrequirements. Here it can be seen that MCSA variant of the proposed algorithm provides theusers with QoS requirements with a gain in throughput, compared to the Proportionally Fairalgorithm, i.e. a gain of about 25.8 %. The same figure shows that the algorithm performspoorly for users that do not have the QoS requirements, which had to be expected, giventhat the users with no QoS requirements, although their channel conditions may be morefavorable, are scheduled fewer PRBs than by using the Proportionally Fair scheme. We donotice also that E-MCSA provides a higher throughput than MCSA for users with no QoS

123

A. Vulpe et al.

Fig. 4 Comparison of Rel. 8 users with and without QoS requirements between the different schedulingalgorithms

Fig. 5 Comparison of Rel. 10 users with and without QoS requirements between the different schedulingalgorithms

requirements, which can be explained by the fact that now the users get only the neededamount of PRBs, which can be lower or higher than the fixed value used for MCSA.

Figures 4 and 5 provide a breakdown of the user throughput results into Rel. 8 users, withno carrier aggregation, and Rel. 10 users, aggregating two CCs. It can be seen that, for Rel.8 users with QoS requirements, both variants of the algorithm provide a gain in throughputcompared to the traditional scheduling algorithms. E-MCSA does actually perform betterthan MCSA for both users with QoS requirements, and no QoS requirements. But, if we

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

Fig. 6 Cell edge throughput comparison of Rel. 8 users with and without QoS requirements between thedifferent scheduling algorithms

Fig. 7 Cell edge throughput comparison of Rel. 10 users with and without QoS requirements between thedifferent scheduling algorithms

look towards Rel. 10 users, we see that MCSA provides the highest throughput gain ascompared to the other algorithms 12.7 % higher than MCSA and 29.4 % higher than thetraditional algorithms. The downside is that it provides the lowest throughput out of allthe algorithms used. A better choice would be here its variant E-MCSA, because it stillprovides a high enough throughput for users with QoS requirements while maintaining anacceptable throughput for users with no QoS requirements. If comparing Figs. 4 and 5, wesee throughput gains of 3–4, when aggregating a second carrier. However, theoretically, the

123

A. Vulpe et al.

Fig. 8 User with QoS requirements throughput CDF

expected throughput gain should be 2. This can be explained by the fact that there is animbalance between macro and femto users, Rel. 8 and Rel. 10 users. There are more macrousers than femto users with Rel. 8 capabilities, and likewise, there are more femto users thanmacro users with Rel. 10 capabilities. These specifics of the macro and femto users and otherexternal factors (also users coming and leaving the BS coverage zone) do not allow to reachthe theoretical expectations.

In order to have a better image of the performances of the proposed algorithms, wehave analyzed also the cell edge user throughput. Figures 6 and 7 show the cell edge userthroughput for Rel. 8 users and Rel. 10 users respectively. By looking at the results we cansee that cell edge users with QoS requirements that run E-MCSA experience the highestthroughput. We can also see that cell edge users with QoS requirements that run MCSAhave the lowest throughput. An interesting result is also that cell edge users that are capableof carrier aggregation with no QoS requirements that run E-MCSA have also the highestthroughput, compared to the traditional scheduling algorithms and MCSA.

Figures 8 and 9 show the CDF of the average normalized overall (i.e. both LTE Rel.8 and Rel. 10) user with QoS requirements and LTE Rel. 10 user with QoS requirementsthroughput, respectively, for the scheduling algorithms considered. We can see that the MCSAcurve exhibits a higher throughput than each of the other scheduling algorithms in the overallcase, which means it would be more suited in a mixed scenario of both LTE and LTE-A users.On the other hand, if we consider that there are mainly QoS users that are LTE-Advancedcapable, E-MCSA would be a better choice.

7 Conclusion

This paper presented and evaluated a scheduling algorithm for LTE-Advanced that takes intoaccount the existence of multiple aggregated LTE Rel. 8 component carriers and the user radioconditions and QoS requirements, and targets the maximization of user throughput. Also, an

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

Fig. 9 LTE Rel. 10 User with QoS requirements throughput CDF

admission control algorithm was defined, in order to enable the scheduling algorithm to berun. Two variants were derived, a basic one, and an enhanced one, and both were avaluated.Results showed that, by using one or the other variants of the algorithm, the average userthroughput is greater for the users with QoS requirements compared to the traditional roundrobin and proportionally fair scheduling algorithms. This same conclusion was valid if wetook into consideration cell edge user throughput. The algorithm stands to be extendedby introducing multiple QoS classes for users and also by developing an optimum powerallocation algorithm.

Acknowledgments This research activity was supported by UEFISCDI Romania under the Grant No.20/2012 “SaRaT-IWSN”, by the Ministry of Communications and Information Society of Romania underthe Grant No. 106/2011 “Evolution, implementation and transition methods of DVB radiobroadcasting usingefficiently the radio frequencies spectrum” and by the Sectoral Operational Programme Human ResourcesDevelopment 2007–2013 of the Romanian Ministry of Labour, Family and Social Protection through theFinancial Agreement POSDRU/107/1.5/S/76903

Appendix 1: Proof of Convexity of the Optimization Problem

From the optimization problem in Sect. 3 the following utility function is obtained:

− u(x11, x12, ..., x1Nc , x21, ..., xNc Nc ) = −Nc∑

c=1

N∑

u=1

wcu xcu (17)

where each weight wcu is the ratio between the estimated throughput of user u and thehistorical average data rate for the same user.

wcu = cu Du Tu

Ru(18)

123

A. Vulpe et al.

where Tu represents the potentially achievable data rate and Ru the historical average datarate.

Tu is calculated with the help of the following equation:

Tu =∑

n

log2(1+ At

nc,u · SI N Rn

c,u

) =∑

n

log2

(1+ At

nc,u ·

PnT X,c,u · |hn

c,u |2σ 2 +∑C

j =c Pnj · |hn

j,u |2)

(19)The historical average data rate is calculated as in the following equation:

Ru,t = γ · Ru,t−1 + (1− γ ) · Tu,t−1 (20)

It is obvious that Tu depends only on the SINR experienced by the user u attachedto cell c. At the moment (the current TTI) the scheduler makes its decision, all thereceived powers, as well as the channel gains are constants. We can therefor say that

C1 = ∑n log2

(1+ At

nc,u ·

PnT X,c,u ·|hn

c,u |2σ 2+∑C

j =c Pnj ·|hn

j,u |2)= const. Likewise, the historical average

data rate is also a constant, since it depends on the average data rate, and the experiencedthroughput in the last TTI. Consequently, C2 = Ru = const.

By rewriting Eq. 17, we obtain:

− u(x11, x12, ..., x1Nc , x21, ..., xNc Nc ) = −Nc∑

c=1

N∑

u=1

f cu (21)

Where f cu is an affine function:

f cu(x) = b + x · aa = C1

C2, b = 0, x = xcu

(22)

Since an affine function is also a convex function and since the sum of convex functions pre-serves the convexity of the function [16], we obtain that−u(x11, x12, ..., x1Nc , x21, ..., xNc Nc )

is a convex function.

Appendix 2: Proof of the Upper Bound for Number of Users of the Cell

Considering Eq. 12 and knowing that

x − 1 < [x] ≤ x (23)

we obtain

Nuserskcu≤ NP RBc

NP RBminku

(24)

From Eq. 10, considering also that the component carriers have equal bandwidths(NP RBc = NP RB∀c ∈ {1, ..., Nc}), we obtain:

Nusersku≤

Nc∑

c=1

Nuserskcu≤

Nc∑

c=1

NP RBc

NP RBminku

= 1

NP RBminku

·Nc∑

c=1

NP RBc =Nc · NP RB

NP RBminku

= NP RBtot

NP RBminku

(25)

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

References

1. Marcu, I., Halunga, S., Fratu, O., & Vizireanu, D. (2011). Multiuser systems implemen-tations in fading environments. InTech. [Online]. http://www.intechopen.com/books/applications-of-matlab-in-science-and-engineering/multiuser-systems-implementations-in-fading-environments

2. Craciunescu, R., Halunga, S., & Fratu, O. (2012). Guard Interval effects on OFDM /BPSK transmissionsover fading channels. In Telecommunications Forum (TELFOR), 20th November 2012 (pp. 471–474).

3. Dahlman, E., Parkvall, S., & Skold, J. (2011). 4G: LTE/LTE-advanced for mobile broadband. Oxford,UK: Academic Press.

4. Song, G., & Li, Y. (2005). Utility-based resource allocation and scheduling in OFDM-based wirelessbroadband networks. IEEE Communications Magazine, 43(12), 127–134.

5. Girici, T., Zhu, C., Agre, J. R., & Ephremides, A. (2010). Proportional fair scheduling algorithm inOFDMA-based wireless systems with Qos constraints. Journal of Communications and Networks, 12(1),30–42.

6. Nguyen, T.-D., & Han, Y. (2006). A proportional fairness algorithm with QoS provision in downlinkOFDMA systems. IEEE Communications Letters, 10(11), 760–762.

7. Parag, P., Bhashyam, S., & Aravind, R. (2005). A subcarrier allocation algorithm for OFDMA usingbuffer and channel state information. In IEEE 62nd Vehicular Technology Conference Fall VTC-2005-Fall(Vol. 1, pp. 622–625).

8. Mohanram, C., & Bhashyam, S. (2007). Joint subcarrier and power allocation in channel-aware queue-aware scheduling for multiuser ofdm. IEEE Transactions on Wireless Communications, 6(9), 3208–3213.

9. Cabral, O., Meucci, F., Mihovska, A., Velez, F., Prasad, N., & Prasad, R. (2011). Integrated commonradio resource management with spectrum aggregation over non-contiguous frequency bands. WirelessPersonal Communications, 59(3), 499–523. doi:10.1007/s11277-011-0242-6

10. Siddavaatam, R., Anpalagan, A., Woungang, I., & Misra, S. (2013). Ant colony optimization based sub-channel allocation algorithm for small cell hetnets.Wireless Personal Communications, 1–22. doi:10.1007/s11277-013-1513-1

11. Rafea Ahmed, Z., Subramaniam, S., Ahmad Zukarnain, Z., & Othman, M. (2014). Best minimumsummation scheduler for long term evolution. Wireless Personal Communications, 1–19. doi:10.1007/s11277-014-1637-y

12. Vulpe, A., Fratu, O., Mihovska, A., & Prasad, R. (2013). A multi-carrier scheduling algorithm for LTE-advanced. In 16th International Symposium on Wireless Personal Multimedia Communications (WPMC),2013 (pp. 1–5).

13. 3GPP. (2010). Evolved Universal Terrestrial Radio Access (E-UTRA): Further advancements for E-UTRAphysical layer aspects, 3rd Generation Partnership Project (3GPP), TS 36.814, March 2010. http://www.3gpp.org/ftp/Specs/html-info/36814.htm

14. I. Cvx Research. (2012). CVX: Matlab software for disciplined convex programming, version 2.0.Retrieved August 2012, from http://cvxr.com/cvx

15. Grant, M., & Boyd, S. (2008). Graph implementations for nonsmooth convex programs. In V. Blondel,S. Boyd, & H. Kimura (Eds.), Recent advances in learning and control, ser. Lecture Notes in Control andInformation Sciences (pp. 95–110). Berlin: Springer. http://stanford.edu/boyd/graph_dcp.html

16. Boyd, S., & Vandenberghe, L. (2004). Convex optimization. Cambridge: Cambridge University Press.

Alexandru Vulpe received the Ph.D. degree in Electronics, Telecom-munications and Information Technology from the UniversityPOLITEHNICA of Bucharest, Romania in 2014. His research inter-ests include, among others, Mobile Communications, Wireless SensorNetworks, Quality-of-Service, Radio Resource Management, MobileApplications. His publications include more than 30 papers pub-lished in journals or presented at international conferences. He par-ticipated as a researcher in a number of national or internationalprojects such as “Reconfigurable Interoperability of Wireless Com-munications Systems (RIWCoS)”, NATO Science for Peace Researchproject (2007–2010), “Wireless Hybrid Access System with UniqueAddressing (SAWHAU)”, Romanian “Partnerships in priority fields”research project (2008–2011) “eWALL—eWall for Active Long Liv-ing” (FP7 project, 2013–2016), “Optimization and Rational Use ofWireless Communication Bands (ORCA)”, NATO Science for Peaceproject (2013–2015).

123

A. Vulpe et al.

Albena Mihovska has a PhD degree in mobile communications fromAalborg University, Aalborg, Denmark, where she is currently an Asso-ciate Professor and Head of Standardisation and Head of Teachingat the Center for TeleInfrastruktur (CTIF). From 2004–2009, she wasinvolved in the EU-funded under FP6 projects WINNER and WINNERII that developed the current concept for LTE-Advanced. Currently, sheis involved with research related to innovative research concepts for 5Gcommunication systems, the design and implementation of eHealth ser-vices (EU project eWALL) and to optimizing and supporting reliableand high performance intensive data rate communications as requiredby the Internet of Things.

Octavian Fratu received the Ph.D. degree in Electronics and Telecom-munications from the University POLITEHNICA of Bucharest, Roma-nia in 1997. He achieved postdoctoral stage as senior researcher in3rd generation mobile communication systems, based on a researchcontract between CNET-France, ENS de Cachan—France and Uni-versite Marne la Vallee—France. He is currently an associated pro-fessor in Electronics, Telecommunications and Information TheoryFaculty. His research interests include Digital Mobile Communica-tions, Radio Data Transmissions, Mobile Communications and Wire-less Sensor Networks, Radio communication networks. His publica-tions include more than 100 papers published in national or interna-tional scientific journals or presented at international conferences. Heparticipated as director or collaborator in many international researchprojects such as “Optimization and rational use of Wireless Commu-nication Bands (ORCA)”, NATO Science for Peace Research project(2013–2016), “Reconfigurable Interoperability of Wireless Communi-

cations Systems (RIWCoS)”, NATO Science for Peace Research project (2007–2010), “REDICT—RegionalEconomic Development by ICT/New media clusters” (FP7 CSA project, 2008–2009), “ATHENA—DigitalSwitchover: Developing Infrastructures for Broadband Access”, (FP6 STREP project, 2004–2006) and other.

Simona Halunga received the M.S. degree in electronics and telecom-munications in 1988 and the Ph.D. degree in communications fromthe University POLITEHNICA of Bucharest, Bucharest, Romania, in1996. Between 1996–1997 she followed postgraduate courses in Man-agement and Marketing, organized by the Romanian Trade and Indus-try Chamber and Politehnica University of Bucharest, in collaborationwith Technical Hochschule Darmstadt, Germany, and in 2008-post-graduate courses in Project Management—Regional Centre for Con-tinuous Education for Public Local Administration, Bucharest She hasbeen Assistant Professor (1991–1996), Lecturer (1997–2001), Asso-ciate Professor (2001–2005) and from 2006 she is a full professorat in Politehnica University of Bucharest, Electronics, Telecommuni-cations and Information Theory Faculty, Telecommunications Depart-ment. Between 1997 and 1999 she has been a Visiting Assistant Profes-sor at Electrical and Computer Engineering Department, University ofColorado at Colorado Springs, USA. Her domain of interest are Multi-

ple Access Systems & Techniques, Satellite Communications, Digital Signal Processing for Telecommuni-cation, Digital Communications-Radio Data Transmissions, Analog and Digital Transmission Systems andDigital Signal Processing for Telecommunications.

123

Admission Control and Scheduling Algorithm for Multi-carrier Systems

Ramjee Prasad is currently the Director of the Center for Tele-Infrastruktur (CTIF) at Aalborg University, Denmark and Professor,Wireless Information Multimedia Communication Chair.Ramjee Prasad is the Founding Chairman of the Global ICT Standard-isation Forum for India (GISFI: www.gisfi.org) established in 2009.GISFI has the purpose of increasing of the collaboration between Euro-pean, Indian, Japanese, North-American and other worldwide standard-ization activities in the area of Information and Communication Tech-nology (ICT) and related application areas. He was the Founding Chair-man of the HERMES Partnership—a network of leading independentEuropean research centres established in 1997, of which he is now theHonorary Chair.He is the founding editor-in-chief of the Springer International Journalon Wireless Personal Communications. He is a member of the edito-rial board of other renowned international journals including those ofRiver Publishers. Ramjee Prasad is a member of the Steering, Advisory,

and Technical Program committees of many renowned annual international conferences including WirelessPersonal Multimedia Communications Symposium (WPMC) and Wireless VITAE. He is General Chair ofGlobal Wireless Summit 2014 (http://gws2014.org/). He is a Fellow of the Institute of Electrical and Elec-tronic Engineers (IEEE), USA, the Institution of Electronics and Telecommunications Engineers (IETE),India, the Institution of Engineering and Technology (IET), UK, and a member of the Netherlands Elec-tronics and Radio Society (NERG), and the Danish Engineering Society (IDA).Along with several eminent awards, he is a Knight (“Ridder”) of the Order of Dannebrog (2010), a distin-guished award by the Queen of Denmark.

123

U.P.B. Sci. Bull., Series C, Vol. 77, Iss. 3, 2015 ISSN 2286-3540

DOWNLINK PACKET SCHEDULING IN LTE-ADVANCED HETEROGENEOUS NETWORKS

Alexandru VULPE1, Octavian FRATU2, Simona HALUNGA3

In order to increase the system capacity in next generation wireless networks, optimizing packet scheduling and resource allocation can be an approach. In the context of the existence of multiple Component Carriers in LTE-Advanced, it becomes a complex optimization problem. This paper presents a scheduling algorithm, designed for use in an LTE-Advanced system. It is formulated as a convex optimization problem and is compared to traditional algorithms such as Round Robin and Proportional Fair. Results show that in some situations it provides better performances than the previous algorithms.

Keywords: LTE-Advanced, resource allocation, scheduling algorithm, convex optimization

1. Introduction

One of the problems that wireless cellular network face today is that many users share a limited amount of resources, these resources being the medium through which the users communicate (i.e. radio spectrum). Scheduling and resource allocation, thus, are essential components of wireless systems because different users experience different fading conditions at the same time [1].

Dynamic scheduling implies two aspects: how physical layer resources are allocated to each user, in each given time slot (Time Transmission Interval - TTI), and how this is optimized [2]. This is a very complex problem that has to take into account different variables like the user radio conditions, the user's traffic pattern or the user Quality of Service (QoS) and has to try to optimize the allocation of radio resources from the whole system point of view. In a nutshell, one has to decide when and how the available resources in the cell are assigned to each of the users. In the case of LTE or LTE-Advanced system, the resources are Physical Resource Blocks (PRBs). PRBs are defined as units consisting of 12 consecutive OFDM subcarriers in the frequency domain and two consecutive time slots or one

1 PhD student., Telecommunications Dept., University POLITEHNICA of Bucharest, Romania, e-

mail: [email protected] 2 Prof., Telecommunications Dept., University POLITEHNICA of Bucharest, Romania, e-mail:

[email protected] 3 Prof., Telecommunications Dept., University POLITEHNICA of Bucharest, Romania, e-mail:

[email protected]

86 Alexandru Vulpe, Octavian Fratu, Simona Halunga

LTE subframe in the time domain, which constitutes the minimum scheduling unit for LTE and LTE-Advanced [3].

Some well-known scheduling algorithms include the max-C/I scheduling strategy, round-robin scheduling and the proportional-fair scheduling. While the max-C/I scheduling algorithm provides the highest cell throughput, it can lead to users that do not have favorable channel conditions never being served.

Proportional fair algorithm also proposes a good tradeoff between maximising throughput and providing data rate fairness. This type of scheduling has as objective the maximization of the long term received throughput.

Likewise, the Round Robin scheduling algorithm provides a “blind” allocation of users, without taking into account Channel State Information (CSI), meaning the cell resources are equally divided amongst users. It can be seen as fair in the sense that each communication link receives the same amount of radio resources (the same amount of time). Another advantage is that it does not involve too much computational effort for the scheduler and it seems the mobile operators prefer it, for its simplicity[3].

The emergence of LTE-Advanced with the Carrier Aggregation technique has opened new research possibilities for scheduling and resource allocation. Now, one has to take also into account the existence of multiple Component Carriers (CCs), which gives another degree of freedom in formulating scheduling and resource allocation problems.

Only recently, algorithms for joint scheduling have started to appear. One such example is given in [4], where the authors propose a modified Proportional Fair (PF) algorithm. Here, the authors compute the PF index for each user, by taking into account the past user throughput over all aggregated CCs. They find that their algorithm improves the fairness for Rel. 8 users and provides a gain in coverage for Rel. 10 users, with no degradation in the average cell throughput. Authors in [5] propose an improved PF scheduling algorithm, based on a PF algorithm that runs independently over each CC. Their algorithm is found to balance the throughput of LTE and LTE-A users and enhance the system fairness, while having a lower complexity than the algorithm in [4]. Their simulations, however, were conducted only with macro cells, without any pico cells. Finally, a joint component carrier assignment and packet scheduling approach is proposed in [6]. Authors formulate a combinatorial optimization problem with constraints. Their results show a lower average delay and a more flat distribution of the aggregated throughput. However, the simulations were conducted with only one eNodeB and 10 users, which, in the authors’ opinion, does not provide sound results.

The main contribution of this paper is in formulating a joint two-dimensional (PRB-wise and CC-wise) optimization of the radio resource scheduling that can be applied in LTE-Advanced systems, that is loosely based on

Downlink packet scheduling in LTE-Advanced heterogeneous networks 87

the idea of round robin scheduling. It does not take into account the channel state information and is a simple to deploy scheduling mechanism, and minimizes the overhead needed to transmit CSI.

2. System model

3GPP LTE-Advanced specifies the aggregation of up to 5 LTE Rel. 8/9 carriers, also known as component carriers (CCs), in order to achieve the overall bandwidth of 100 MHz. This work proposes a simple algorithm, which leverages the existence of multiple CCs and enables the user allocation over any number of CCs, with the objective of maximising the total network throughput. This is an extension of the Round Robin algorithm, in the sense that it does not take into account the user channel conditions and “blindly” allocates resources to users of the base station.

Although the algorithm proposed in this work enables the scheduling of users on up to 5 aggregated CCs, in this paper we chose to evaluate the algorithm on two CCs in the case of Contiguous Intraband Carrier Aggregation, i.e. the aggregated carriers are adjacent to each other. Depending on the capabilities of the User Equipment (UE), each user may be allocated to a single LTE carrier, or simultaneously to two carriers, meaning LTE and LTE-A users coexist. This does not restrict however the generality of the defined problem.

The LTE-Advanced network is deployed with hexagonal cells and with frequency reuse factor one, in order to evaluate the worst-case scenario in terms of interference, as mentioned in the 3GPP specifications. All the cells deploy two CCs, each CC being transmitted from different antenna connectors. In this case, for each cell, the adjacent cells become a source of interference (but only on each CC), which leads to a deterioration of the user received Signal-to-Interference-plus-Noise (SINR) ratio and, consequently, a drop in user throughput. The network topology is illustrated in Fig. 1.

The scheduler allocates packets to users at the beginning of each Time

Transmission Interval (TTI). Each TTI is associated with the duration of one of the ten subframes that makes up an LTE frame (which is 10 ms long), which corresponds to a subframe duration of 1 ms.

88 Alexandru Vulpe, Octavian Fratu, Simona Halunga

Fig. 1. Network Topology

The Radio Resource Management framework allocates the available radio resources (i.e. Physical Resource Blocks - PRBS) to the users according to the solution of the scheduling problem defined in the next section.

3. Cross-CC Round Robin Scheduling Algorithm

Carrier aggregation adds a new dimension for the scheduling of the users in an LTE-Advanced system and raises an optimisation problem for the best use of network resources. The following Profit Function is proposed in order to maximize the total throughput of the user.

∑∑= =

cN

c

N

ucuxPF

1 1:)( (1)

In the above equation, xcu is the allocation variable. Also Nc is the numbered of configured component carriers (CCs) in the cell and N is the number of users attached to the cell. The allocation variable xcu reflects the resource allocation (i.e. allocated PRBs) for a specific user u on a specific component carrier c. We can therefore define the allocation variable as

Downlink packet scheduling in LTE-Advanced heterogeneous networks 89

⎩⎨⎧

=cu

cunnxcu carrier component on user toallocated are PRBs no if,0

carrier component on user toallocated are PRBs if, (2)

The above problem is a convex optimization problem because xcu is a

constant for each given TTI [7]. In order to solve it, the following constraints are defined:

1. Allocation Constraint: the sum of allocated PRBs for each user u has to be equal to the number of PRBs available for each CC. This can be written as:

{ }cN

uPRBcu NcNx

c...,,2,1,

1∈∀=∑

= (3)

where

cPRBN represents the number of PRBs available for component carrier c.

2. User constraint: the allocated PRBs on each component carrier c for each user u are forcefully 0, if the user is not configured on the respective component carrier. This can be written as:

cuxcu carrier component toattachednot is user if,0= (4) The objective is, therefore, to maximize the Profit Function, so, for each

component carrier c and user u, a number of PRBs x*cu is allocated, according to :

PFx

cuxcu maxarg* = (5)

4. Simulator description, simulation scenario and assumptions

The network scenarios that have been analysed were chosen taking into account the carrier aggregation scenarios presented by 3GPP. We designed our own MATLAB LTE simulator, and enhanced it to support Carrier Aggregation.

90 Alexandru Vulpe, Octavian Fratu, Simona Halunga

Fig. 2. Simulator flow

Downlink packet scheduling in LTE-Advanced heterogeneous networks 91

The flow of the simulator is depicted in Fig. 2. Following the recommendations of 3GPP [8] the worst interference-case scenarios were considered, where there is Frequency Reuse 1 between the femto-cells and Macro cells.

The scenario simulated here represents an urban environment with 19 sites with 3 Macro BS per site. The Macro BSs have a height of 30 meters and an electrical downtilt of 15° following 3GPP specifications.

The PathLoss Model is Model 2 proposed by 3GPP, which considers different PathLosses for Line of sight (LoS) and Non-LoS conditions, where the transmitter and the receiver do not have a direct channel between them [8].

To solve problem (5) we used CVX, a package for specifying and solving convex programs [9, 10].

Table 1 lists the main simulation parameters used for this work. All other parameters are also taken from the 3GPP specifications.

Table 1 Main simulation parameters

Parameter Value

Topology Hexagonal grid, 19 macro BSs, 3 sectors per BS, 19 femto BS

Carrier Frequency 2 GHz Component Carrier Bandwidth 20 MHz Number of Component Carriers 2 Duplexing FDDAntenna pattern for macro BS As described in [8]Antenna pattern for femto BS Omnidirectional Inter-site distance 500 m between macro Distance between macro and femto 75 m Distance between femto 40 m Tx Power for Femtocell 30 dBmTx Power for Macrocell 49 dBmPath loss As described in [8] Shadowing standard deviation 8 dB Penetration loss 20 dB Number of users per Femtocell 10 Number of users per Macrocell 20User distribution in site area UniformTraffic model Full buffer

5. Results

Fig. 3 shows the average cell throughput obtained with the proposed scheduling algorithm and compared with the Round Robin and Proportionally Fair algorithms broken down into femto cells and macro cells. It can be seen that the

92 Alexandru Vulpe, Octavian Fratu, Simona Halunga

proposed algorithm achieves slightly better results for Macro Cells than the other existing algorithms. For the Femto Cell case, compared to the Round Robin algorithm, it is superior, marking an increase in cell throughput of about 3.8%.

Fig. 3. Average cell throughput for different scheduling algorithms Fig. 4 provides a breakdown of the user throughput results into Rel. 8

users, with no Carrier Aggregation, and Rel. 10 users, aggregating two CCs. It can be seen that, as expected, the proposed algorithm performs marginally poor in terms of user throughput for Rel. 10 users. As we may have intuited, the Proportional Fair algorithm, ran separately for each CC, gives users a better throughput, since it takes into account channel conditions.

Fig. 4. Average user throughput for different scheduling algorithms

Downlink packet scheduling in LTE-Advanced heterogeneous networks 93

An interesting result is for macro cell users, where PF gives the worst throughput. This can be attributed to the fact that the existence of femto cells implies an increase in Inter-Cell Interference, which in turn means a lot of users

Fig. 5. Cell Throughput CDF

Fig. 6. User throughput CDF

that have a low PF index, leading to some users never being scheduled, or being scheduled very few times.

Fig. 5 and Fig. 6 show the Cumulative Distribution Function (CDF) of the average normalized cell and user throughput, respectively, for the three scheduling algorithms considered.

94 Alexandru Vulpe, Octavian Fratu, Simona Halunga

We can see that the proposed algorithm has slightly better results for the cell throughput case. This mean that, for a specified value of the normalized DL cell throughput, there is a higher probability of achieving that throughput when using the proposed algorithm, compared with the others. Similarly, in the case of user throughput, Cross-CC RR has better results for lower throughput values, being surpassed by PF for higher values.

6. Conclusions

This paper presented and evaluated a scheduling algorithm for LTE-Advanced that takes into account the existence of multiple aggregated LTE Rel. 8 Component Carriers and targets the maximization of allocated resources. Results showed that, by using this algorithm, the average user throughput is greater compared to the traditional Round Robin algorithm but lower than using Proportionally Fair. The algorithm can be used as an alternative to Round Robin, without having to use Channel State Information that can provide unnecessary overhead for network signaling.

R E F E R E N C E S

[1]. I. Marcu, S. Halunga, O. Fratu, and D. Vizireanu, Multiuser Systems Implementations in Fading Environments. InTech, 2011. [Online]. Available: http://www.intechopen.com/books/applications-of-matlab-in-science-and-engineering/ multiuser-systems-implementations-in-fading-environments

[2]. Ofuji, Y.; Morimoto, A.; Abeta, S.; Sawahashi, M., "Comparison of packet scheduling algorithms focusing on user throughput in high speed downlink packet access," Personal, Indoor and Mobile Radio Communications, 2002. The 13th IEEE International Symposium on , vol.3, pp.1462,1466 vol.3, 15-18 Sept. 2002

[3]. E. Dahlman, S. Parkvall, and J. Skold, 4G: LTE/LTE-Advanced for Mobile Broadband. Oxford, UK: Academic Press, 2011

[4]. Y. Wang, K. I. Pedersen, T. B. Sørensen, and P. E. Mogensen, “Carrier load balancing and packet scheduling for multi-carrier systems,” Wirel. Commun. IEEE Trans., vol. 9, no. 5, pp. 1780–1789, 2010.

[5]. L. Lin, Y. Liu, F. Liu, G. Xie, K. Liu, and X. Ge, “Resource scheduling in downlink LTE-Advanced system with carrier aggregation,” J. China Univ. Posts Telecommun., vol. 19, no. 1, pp. 44–123, Feb. 2012.

[6]. G. Gupta and P. Mohapatra, “Joint carrier aggregation and packet scheduling in LTE-advanced networks,” 2013 IEEE Int. Conf. Sensing, Commun. Netw., pp. 469–477, Jun. 2013.

[7]. S. Boyd and L. Vandenberghe, Convex optimization. Cambridge University Press, 2004 [8]. 3GPP, “Evolved Universal Terrestrial Radio Access (E-UTRA); Further advancements for E-

UTRA physical layer aspects,” 3rd Generation Partnership Project (3GPP), TS 36.814, Mar. 2010. [Online]. Available: http://www.3gpp.org/ftp/Specs/html-info/36814.htm

[9]. I. CVX Research, “CVX: Matlab software for disciplined convex programming, version 2.0,” http://cvxr.com/cvx, Aug. 2012.

[10]. M. Grant and S. Boyd, “Graph implementations for nonsmooth convex programs,” in Recent Advances in Learning and Control, ser. Lecture Notes in Control and Information Sciences, V. Blondel, S. Boyd, and H. Kimura, Eds. Springer-Verlag Limited, 2008, pp. 95–110, http://stanford.edu/ boyd/graph dcp.html.

Wireless Pers Commun (2014) 78:1613–1627DOI 10.1007/s11277-014-1906-9

Small Cells in Cellular Networks: Challengesof Future HetNets

Octavian Fratu · Alexandru Vulpe ·Razvan Craciunescu · Simona Halunga

Published online: 15 July 2014© Springer Science+Business Media New York 2014

Abstract Due to their low cost and easy deployment, small cells provide a viable and cost-effective way of improving the cellular coverage and capacity both for homes and enterprises,both in metropolitan and rural areas. Stimulated by their attractive features and advantages, theongoing development and deployment of small cells by manufacturers and mobile networkoperators have seen a surge in recent years. Together with macro-cells, they form, what arecalled Heterogeneous Networks or HetNets. However, the successful rollout and operationof small cells are still facing significan issues. In this paper the need for, challenges andsolutions of small cell deployments are analyzed. This analysis is conducted with respect toself-organizing features, interference coordination, energy efficiency and spectrum efficiency.The analysis is complemented with numerical results based on system simulations in Macro-only and HetNet scenarios and also on real measurements performed on an mobile operatornetwork. Results show the clear improvement that a HetNet brings in term of user throughputand also the amunt of spectrum waste that is present in nowadays’ operator networks.

Keywords 4G · Small cells · HetNet · Future networks

O. Fratu · A. Vulpe (B) · R. Craciunescu · S. HalungaFaculty of Electronics, Telecommunications and Information Technology,Politehnica University of Bucharest, Splaiul Independentei nr. 313,Sector 6, 060042 Bucharest, Romaniae-mail: [email protected]

O. Fratue-mail: [email protected]

R. Craciunescue-mail: [email protected]

S. Halungae-mail: [email protected]

123

1614 O. Fratu et al.

1 Introduction

Recent years have seen a surge in the growth of bandwidth-hungry applications. This hasput pressure on Mobile Network Operators (MNOs) to increase the capacity of their net-works. The choices made by MNOs seem to be the mass adoption of Long Term Evolution(LTE) as a cellular technology and the deployment of small cells. The main characteristic ofnowadays mobile operators network is that almost 80 % of all data traffic is indoor. Thus, themain challenge is how to penetrate through wall, roofs, sky-buildings. Studies have shownthat deploying small cells in hotspots significantly increases the network capacity and cov-erage as the networks become layered and dense. Also small cell deployment benefits fromincreased spectrum efficiency, and all these advantages apply equally to home, office or out-side deployment. The largest small cell deployments have already reached 1 million activecells [4].

It is estimated that the small cell market will generate US$22 billion during 2016, 73 %of which will be driven by public area small cells (see Fig. 1). The number of small cellsdeployed overtook the total number of macrocells in late 2012. The small cell technology isused by a number of 56 different operators. 44 of them are offering residential small cells and33 of them are offering enterprise, urban and rural small cells. Hence, the forecast predictsthat the non-residential small cell market will grow up to 10 billion US$, and the residentmarket up to 400 million US$ per year in 2018 [3].

Small cells are wireless infrastructure equipment that operate in licensed bands with lowpower, have an intelligent nature and characteristics like auto-configuration, environmentsense, self-organizing features etc. They consist of:

– Femtocells: Primarily deployed in consumer and enterprise environments;– Picocells: Deployed in indoor public areas (airports, train stations, shopping areas);– Microcells: usually deployed in urban areas or in cases where the footprint of a macrocell

is not necessary.

Small cells are deployed either in an autonomous network either in association with macro-cells in order to form a heterogeneous network (HetNet), either in specific areas in order toaddress local capacity requirements. In outdoor environments the small cell base stations(BS) are placed at low heights (3–10 m) and in an indoor environment the BSs are deployedsimilar to WiFi networks. Moreover, deployed small cells have WiFi and 3G compatibility,and as the LTE technology is gaining groud, small cell LTE compatibility is also required.

Fig. 1 Small cell market revenue over the following years

123

Small Cells in Cellular Networks 1615

The 3G and LTE compatibility for small cells deployment depends on the size of the operator.For a large operator (e.g. AT&T, Vodafone, etc.) it is essential to deploy primary 3G smallcells, as they have a lot of subscribers with 3G handsets and a growing subscriber base, inorder to increase the network capacity. Beside the 3G small cells capabilities they could alsodeploy 4G small cells in a cost efficient way. For small operators, that are looking their wayon the market it would be better to reduce costs by deploying directly 4G small cells. The 3Gand 4G operators are using licensed spectrum. As small cells deployment could also meanWiFi compatibility, the WiFi Alliance introduced the Hotspot 2.0 so that a subscriber couldautomatically join the operator WiFi service whenever he is in a Hotspot 2.0 area. Hotspot2.0 is based on the IEEE 802.11u standard and enables cellular like roaming [2].

Deployment and operation of small cells comes with a number of challenges:

– Self-organization: The number of small cells to be deployed is expected to be large. Itwill not be feasible for operators to optimize small cell deployments using traditionalnetwork planning. Therefore, 3GPP has introduced the Self-Organizing Network (SON)concept since Rel. 8;

– Increased InterCell Interference: The rollout of user-deployed cells overlying macro-cells will create new cell boundaries, in which end users will suffer from strong intercellinterference;

– Energy efficiency: The deployment of a large number of small cells will make the overallenergy consumption to increase;

– Cost efficiency: The deployment of small cells, especially at the users location, wouldlead to a cost reduction in terms of energy bill at the operator site and would transfer thecosts to the user side;

– Spectrum efficiency: There will be more and more cells using the same amount ofspectrum.

The next sections deal with each of the aforementioned issues and give insight into thechallenges faced for each issue. Section 2 briefly describes the main characteristics of self-organizing networks and highlights challenges and proposed solutions for SON algorithmsin future HetNets. Section 3 describes the issue of interference coordination. Inter-Cell Inter-ference Coordination (ICIC) and enhanced ICIC (eICIC) schemes are briefly described alongwith challenges and solutions for future HetNets. A comparison between coordinated anduncoordinated small cell deployments is made highlighting the impact on (e)ICIC techniques.Next, Section 4 raises the issues of energy efficiency in current and future radio access net-works (RANs). The energy consumption will sky-rocket in the next few years especially sincemore and more small cells are deployed and operated. Finally, Section 5 deals with spectrumefficiency in HetNet deployments. As an example, spectrum occupancy measurements of amobile operator network are shown highlighting the spectrum inefficiency of current cellularnetworks. Section 6 draws the conclusions.

2 Self-Organization

Traditionally, self-organization consists of three phases:

1. Self-configuration: This is performed at start-up, where initial parameters are configured;2. Self-optimization: This phase is performed on a regular basis. It tunes parameters accord-

ing to fluctuations in network, traffic, channel3. Self-healing: This phase automatically detects network failures and corrects or mitigates

them.

123

1616 O. Fratu et al.

There are multiple ways to achieve self-organisation. Usually, there are three SON schemes.The first scheme adopts a centralized global entity in charge of a given number of cells,

responsible for their optimization (typically an OAM system). Small cells can forward theirparameters to the centralized server which optimizes them based on overall informationprovided as input by the cells.

The second scheme is a distributed architecture where each cell optimizes its own para-meters based on local sensing. The SON algorithms are located in eNodeBs, thus small cellscan learn about neighbouring cells and fine tune their parameters faster.

The third scheme is a hybrid one. In practical deployments, the two architecture alternativesabove are not mutually exclusive and could coexist for different purposes.

2.1 Challenges

There are several challenges that affect the self-organization features of small cells. Onechallenge is that the expected massive small cell deployment requires additional (optimized)auto-configuration functionality. Manually adjusting parameters is clearly not an option, dueto the large number of parameters to be configured/optimized.

Another challenge is how to deploy Hybrid SON with minimum impact on the cellularnetwork and minimum costs. Also, the aplicability of SON algorithms has to be extended tothe most part of network planning, deployment and operation, such that only small parts ofthese need human intervention.

Finally, taking into account that there will be a large number of parameters to tune (comingfrom the large number of deployed small cells), there is the need for a greater optimizationof SON algorithms. Here, a challenge is how to develop simple SON algorithms that canbe implemented on devices with limited computing power for optimum or near optimumnetwork performance.

2.2 Solutions

There are some proposed solutions for some of the challenges that were mentioned before.First, SON use cases can be enhanced to treat differently UE groups or clusters. The UEscan be grouped into high/low speed UEs or Release 8/Release 10 UEs/Machine to machine(M2M) UEs or any such other grouping.

Another solution would be the addition of position information as input for the SONalgorithms. This would allow, for instance, dynamically adjusted handover thresholds alongcells. This would yield a more balanced load between cells. Another solution along the sameline would be to use Adaptive Antenna Systems (AAS) to adapt coverage of the base stations[20]. Using this, in conjunction with the position information of the UE would allow the cellto “follow the user” and adapt to users’ actual traffic and demands.

Finally, another solution would be to optimize spectrum allocation between different RadioAccess Technologies (RATs);

A solution would be to enable cost efficient small cell deployments. This would requirethat SON algorithms determine the optimum placement of a small cell network such as toachieve the best performance of the communications taking place according to whichevermetric is used.

It is noteworthy to mention that the SON techniques are a work in progress, and SON-related functionality will continue to expand through the subsequent releases of the LTEstandard, Release 10 and beyond, to cover all key aspects related to network management,troubleshooting and optimization in heterogeneous networks.

123

Small Cells in Cellular Networks 1617

3 Interference Coordination

Because LTE operates with frequency reuse one, the need for interference coordination andcancellation (ICIC) schemes arose. Interference is the main bottleneck against enhancedspectrum efficiency. In ICIC based schemes, cell resources are divided into frequency bandsand transmit power proles to reduce intercell interference. The X2 interface is used to share theinformation between the eNBs. The ICIC is used either in time or frequency domain [6,21].

3.1 Interference Coordination Techniques

To support small cell deployment ICIC is necessary. By using ICIC in order to reduce inter-ferences, the small cell uses one set of carriers or one set of time slots. In the remaining set ofcarriers or time slots only the macro cell can transmit (the small cell suppresses its transmis-sion) so that the macro cell can protect the vulnerable UE (e.g. that are at the cell-edge). The3GPP Rel. 8 standard did not include the capability of varying the reference signal power,according to the sub-band. In a dense heterogeneous network we cannot find a sufficientnumber of PSD patterns. Hence, the macro or the small cells on a specific sub-band or timeslot have to be powered down, so that the interferences are mitigated. The problem is thatthe macro and the small cell loose capacity [21].

So, enhanced ICIC (eICIC) techniques have been proposed in 3GPP Rel. 10. This eICICproposal mitigates interference on traffic and control channels (as opposed to only trafficchannels in ICIC). It uses power, frequency and also time domain to mitigate intra-frequencyinterference in heterogeneous networks.It also improves cell-edge user throughput, coverageand deployment flexibility [6].

One important feature of eICIC time domain is the concept of Almost Blank Subframes(ABS). These are used especially when the small cell users are strongly affected by inter-ference from the Macro cell. In order to assure a reliable communication for the small cellusers, the macro cell periodically mutes some subframes (i.e. the macro cell is not allowed totransmit in those subframes). However, these frames arent totally muted. Low power signalsare still transmitted, such as Common Reference Symbols (CRS), Primary and SecondarySynchronization Signals (PSS and SSS), Physical broadcast channel (PBCH), System Infor-mation Block-1 and paging with their associated PDCCH. The mentioned signals must betransmitted in order to assure the backward compatibility or to avoid link failure. eNodeBsexchange ABS specific configuration using the X2 interface [6].

Another important feature of eICIC is the Flexible User Association and Cell selectionbias (CSB) technique. This prevents underutilizing or overloading a small cell. In order toassociate itself to the macro cell or to the small cell, the UE measures the reference signalstrength from the two. As the small cell reference signal is a low power signal, and as the UEis not very close to the small cell base station, there is a high probability that the macro cellreference signal strength is higher than that of the small cell. Thus, the UE associates withthe macro cell and the small cell is underutilized. In order to prevent such cases the CBStechnique is used. Thus, every cell is given a power bias, pi . The bias is smaller or largeras a cell is underutilized or overutilized. As the cell reference power is Pi the UE associatesitself with the cell that has the maximum (Pi + pi ), thus obtaining an efficient utilization ofthe small/macro cell.

Another interference coordination scheme is CoMP (Coordinated MultiPoint). InLTE/LTE-A, it requires close coordination between a number of geographically separatedeNBs. These eNBs dynamically coordinate in order to provide joint scheduling and trans-missions and also to provide joint processing of the received signals. In this way a UE at the

123

1618 O. Fratu et al.

edge of a cell is able to be served by two or more eNBs. This improves signal reception andtransmission and increases throughput particularly under cell edge conditions. This schemeis used for the enhancement of SON LTE networks and for the mobility enhancement of theeNB and HeNB [10].

Beside the above schemes, a set of receiver based techniques are being discussed, likeInterference Rejection Combining (IRC) that is used for interference suppression.

IRC is used to reduce the interference from the adjacent cells and increase the user through-put at the cell-edges, thus providing a better user experience. Using this technique the receiverregenerates the transmitted signal based on the spatial estimated data from the previous recep-tions using spatial separation features. This technique is, however, very sensitive to channelestimation errors [13].

3.2 Challenges and Solutions

Despite the above mentioned techniques, there are future enhancements planned for the nextgeneration LTE-A small cell systems. These enhancements are required in order to increasethe capacity and performances of the actual small cell systems thus resulting in an increaseduser experience. In the following a couple of key approaches are proposed [9].

Hence, for the next generation LTE-A small cell system, advanced transmitter beamform-ing techniques could be used. These techniques can use either 3D-MIMO techniques that,enhanced with active antenna systems (AAS), could control the beamforming in both hor-izontal and vertical directions. In order to implement 3D-MIMO AAS systems, more thaneight antenna elements are needed. Beside the number of antenna elements, interferencesuppression should be investigated.

Also, in terms of MIMO systems, massive antenna technologies could be used. As theantenna elements are getting smaller and smaller and the transmitted frequency gets higher, atwo dimension massive antenna array could be implemented. As this massive antenna array isimplemented, a very narrow beamforming is expected, thus compensating for the propagationloss [9].

As discussed in section 3.1, receiver cancellation techniques are required for a better userexperience, especially at cell-edges. For the next generation LTE small cell systems, advancedreceiver cancellation techniques [14] have to be implemented for inter-cell interference andnon-orthogonal multiple access. In order to achieve these goals, more receiver antennas haveto be implemented for IRC techniques (from two in LTE Release 11 to four or eight). Besidethe IRC techniques, successive interference cancellation, at the UE receiver side, can be usedas a nonlinear interference cancellation technique. Also, the next generation small cell LTEsystems should efficiently use the path loss difference among users especially in wide areas[8].

In terms of power use, cognitive power sleep mechanisms have to be implemented. Thismeans that small cells have to predict traffic peak hours and off-peak hours, based on thetraffic history. In the traffic off-peak hours, the small cell can shut down (e.g. enter in sleepmode) and all the users would use the macro cell as their primary cell. This small cell “shutdown” will result in power reduction and cost reduction. The energy efficiency aspects arediscussed in detail in the next section.

3.3 Coordinated versus Uncoordinated Depolyments

In the context of interference management and SON algorithms there is an architecturalchoice between coordinated and uncoordinated deployment of small cells. As the name

123

Small Cells in Cellular Networks 1619

Table 1 Advantages and disadvantages of different small cell deployments

Advantages Disadvantages

Coordinated Control information exchange amongdifferent cellsMore accurate interference mitigation

Excess signaling in the networkComputational burden for macro BS

Uncoordinated No physical interface between BSsNo complex signaling

Heavy inter-cell interferencePoor system performance for cell-edgeusers

suggests, a coordinated deployment architecture requires constant information sharing (i.e.coordination) between base stations in the coverage area, which can be applied to small celldeployment. Uncoordinated deployment means that the base stations are deployed withoutprior network planning, and there is no control for their placements. This is the case, forinstance, with user deployed home eNodeBs (HeNB). Table 1 depicts some advantages anddisadvantages of both architectures.

Two clear disadvantages to the coordinated approach seem to be the excess of signaling inthe network that creates a substantial overhead and the computational burden on the macroBS executing the Radio Resource Management algorithms.

In HetNet scenarios, the uncoordinated deployment has slightly different meaningsdepending on the type of cell, but always the same objective of providing additional capacityto users. Uncoordinated small-cell deployment is done with minimal pre-planning. Sincethese BSs operate either on a permanent or temporary basis in the macro BSs coverage area,constant optimization should be performed in order to take into account user distribution,Inter-Cell Interference (ICI) (mainly from the macro BS and neighboring pico BSs) andperhaps for energy saving reasons as well—not all BSs need to transmit in the entirety ofthe spectrum. In a completely uncoordinated scenario, this management is more complexseeing that neither position nor number of BSs is known a priori, because they are deployedby home and office users, and not by the operator. SON optimization capabilities can be usedto limit ICI and congestion in the coverage area.

We should note that it is mandatory for small cells in the uncoordinated HetNet to haveself-organizing features of their own. This way, they will attempt to adapt to the macroenvironment they find themselves in. Neither adapting the macro configuration on its ownwith the static information from currently deployed small cells, nor adapting the small cellconfiguration with fixed macro configuration will give the optimal solution. The best solutionrequires a joint approach.

In order to illustrate the throughput gains achieved when deploying HetNets as com-pared to macro-only networks in LTE-A networks, we conducted link-level simulations andcompared these two scenarios using three different scheduling algorithms. We chose two ofthe traditional scheduling algorithms Round Robin (RR) and Proportional Fair (PF) [12],that were adapted for a Carrier Aggregation scenario, and we also used a third algorithm,MCSA [19], designed to optimize user throughput for users with QoS requirements acrossComponent Carriers and the aggregated carrier.

In the macro-only scenario, 30 users are uniformly distributed inside the coverage area ofthe cell. In HetNet scenarios, there is one small cell per macro (i.e. inside the macro coveragearea), placed randomly with some constraints. 20 users are still connected to the macro cell,but 10 users are connected to the small cell. There are 50 % LTE users and 50 % LTE-A users.The proportion of users with QoS requirements vs. users with no QoS requirements is alsothe same.

123

1620 O. Fratu et al.

Fig. 2 Overall average user throughput gain of HetNet deployments over Macro-only

Fig. 3 Breakdown of average user throughput gain of HetNet deployments over Macro-only

Figure 2 shows the average user throughput gain of HetNet deployments over macro-onlynetworks. It can be noticed that there are fairly large throughput gains (of around 80-120 %)for almost all types of users. We can also see that MCSA yields higher throughput gainsthan the other algorithms for all the scenarios. This show that developing an algorithm thatjointly optimizes the user throughput can bring better throughput gains than using tradi-tional scheduling algorithms. The biggest throughput gain is for LTE-A users. This resultshows that deploying small cells, coupled with the use of Carrier Aggregation techniques(i.e. LTE-A technology) can bring very big improvements in term of user throughput andexperience.

To make our statements more sounde, Fig. 3 illustrates a breakdown of the results for mixedtypes of users (with LTE or LTE-A, with QoS requirements or not). We can see that, as wewould have expected from the results in Fig. 2, LTE-A users benefit from higher throughput,and also, that QoS users are advantaged by the MCSA algorithm. For users runing RR andPF algorithms, performance gains are not very different between user types. We must alsonotice that MCSA brings less gain for LTE users when compared to the other algorithms, soits use would not be justified in a scenario with only LTE users.

This enables us to observe that using small cells in an LTE-Advanced network and couplingthis with a scheduling algorithm that leverages the existence of users with different QoSrequirements can improve the user throughput and experience.

123

Small Cells in Cellular Networks 1621

4 Energy Efficiency

The green aspects of radio access technologies have become more relevant in the last years.Both users and engineers have been sensible to the convenience aspects of the mobile access,but in the last years it became clear that the wireless access networks consume a lot of energyand their efficiency for this point of view is very poor. The Base Stations (BSs) are responsibleof about 80 % of the total energy consumption in cellular networks [15]. About 75–80 % ofbase stations are not used or are used far to their capacity. The perception is changing nowin terms of network energy efficiency [11].

Figure 4 shows the evolution of spectrum and energy efficiency in future networks [11].It can be seen that in the future, networks will consume much more energy (due to thedeployment of many small cells), and this requires enhancements of network energy saving.

There are several methods to reduce energy consumption in cellular networks [18]:

– Minimization of the number of base station sites This request is related to the fact thatthe traffic is mainly localized. An optimal deployment of the base stations would requireto choose sites placed in the center of the areas with high-traffic demands. The energyefficiency will thus increase when transmitters are brought closer to the receivers.

– Use of Femtocells inside buildings It is well known that most of the traffic is generatednowadays by users with low mobility, located in urban areas, especially in buildings [1].

– Use of MIMO techniques to increase spectrum efficiency The so-called “massiveMIMO” allows also to enable precise focusing of emitted energy on the intended users,resulting in a higher energy efficiency [5].

– Use of smart antennas These would be used to increase the transmission distance andwould have the same impact as massive MIMO deployment.

– Use of self-aware and self-configurable network elements By developing SON algo-rithms with the purpose of saving energy, the network would gain the ability of self-optimization from an energy efficiency point of view.

Fig. 4 Traffic volume and energy consumption evolution

123

1622 O. Fratu et al.

1900 1950 2000 2050 2100 2150 2200-140

-120

-100

-80

-60

-40

-20

freq [MHz]

Period of the day: 06 - 14 hourssi

gnal

leve

l [dB

m]

UMTS UL UMTS DL

maximum signal level

minimum signal level

average signal level

Fig. 5 Signal level within UMTS bands during morning and noon (measurements performed in 2010 in thebuilding of the Faculty of Electronics, Telecommunications and Information Technology, Bucharest, Romania)

– Building of intelligent networks using scalable elements, able to extend, configure andoptimize themselves.

– Optimization of data routing– Building of software routines to manage idle, sleep or shutdown of BSs.

A major challenge in access network energy efficiency is the signaling policy. The 2Gand 3G systems, which are not aware on the energy efficiency in the access network, uselogical channel signaling with continuous transmission. This conducts to a massive energyuse for transmitting the control signals, which may be observed in the large differencesbetween the power spectral density in the downlink bands in comparison to the uplink ones(see Fig. 5 [7]). So the permanent transmission through control channels must be reduced oravoided. A possible solution is to use burst transmissions of control signals. Another solutionis to transmit the control signals only in the macrocells and to have assisted small cells bymacrocells (solution considered also by 3GPP in its Rel. 12).

Studies related to radio propagation at different frequencies highlighted differences inradio attenuation in different frequency bands (increased propagation losses at high fre-quency). There is a need of new planning procedures to increase the network efficiency:

– Avoid the use of high frequency bands in macrocells and use them mainly for smallcells (i.e. 3.5 GHz band);

– Avoid carrier aggregation of bands with important frequency differences especiallyin case of collocated transceivers;

– Predict the traffic more accurately and reconfigure the network accordingly Usethe behavior and traffic habits of the users (most of them with low mobility) and extendthe deployment of the small cells in order to increase the traffic capabilities in the areawhere the users are placed (we can say that the access network must follow the users);

– Use macrocells for a continuous coverage but with lower traffic capabilities and enhancethese capabilities especially in the area where there are users with high mobility.

The use of relays and remote radio heads may increase the energy efficiency. Someimprovements are, however, necessary for a better performance [22]:

– Low latency connection between the macrocell BS and these radio units;

123

Small Cells in Cellular Networks 1623

– Design low-latency wireless/mobile backhaul links to cover such connections as partof a reconfigurable and scalable network;

– Improved radio propagation prediction to adapt the offered capacity to the real oper-ating conditions;

– Improve the CoMP procedures in terms of energy efficiency.

5 Spectrum Efficiency

The traffic structure is changing very fast. Data traffic is increasing (expected to reach about87 % of the total traffic at the end of the decade) and voice and SMS traffic is decreasing.The traffic has been growing yearly faster (1.6 times/year in the first decade of this century,now it is doubling, 12 times from 2011 to 2015 and 500 times from 2011 to 2020) [17].

The spectrum dedicated to cellular networks is not enough (about 320 MHz total bandwidthnow). There are plans to extend it (Digital Dividend 2, additional bands in the frequency areaof 1.8, 2.1 GHz (2013), 2, 2.6, 3.5 GHz (2016), additional 200 MHz bandwidth till 2020). Itis unrealistic to think that the spectrum may be extended with more than 600 MHz (3 timesin the period 2011–2020). There is a pressure to identify new frequency bands. For example,the European Union is looking through the new Horizon 2020 Research framework for atechnical solution to use spectrum in the 60 GHz band for such access.

The last generation of cellular networks, especially the LTE one, brings a significantincrease of the spectral efficiency [1] (see Fig. 6) and it is expected to continue with morespectrally efficient technologies, based on extensive use of MIMO.

Also there is a competition for the use of licensed and unlicensed bands. Based on theexperience of 802.11-type systems, some authors consider that the use of unlicensed spectrumconducts to better spectrum efficiency [16].

The extended bandwidths that are foreseen to be used need a better management and usage.In order to have a unitary approach for slow or fast applications, including ones that demandlarge bandwidth, it is necessary to implement Extended Carrier Aggregation procedures (i.e.more than two bands of 20 MHz each in LTE-A). Special attention (to achieve good spectralefficiency) has to payed to choose closed frequency bands for aggregation [8]. Asymmetricdeployment is also very important in order to enhance the spectrum utilization.

Fig. 6 Spectral Efficiency in downlink of 3G and beyond-3G cellular systems in comparison with HSDPA3.6

123

1624 O. Fratu et al.

The use of cognitive radio methods will increase the spectrum efficiency as well. Thesemethods have to be considered in conjunction with the introduction of dynamic access pro-cedures in the current use in cellular systems, doubled with the spectrum sharing betweendifferent applications. The exploitation of the white-space islands for other applications willalso enhance the spectrum efficiency.

As a consequence, the main resources to cover the traffic evolution are:

– Improved Radio Resource Reuse (expected 56 time increase in 2011–2020), with twomain components: the intensive deployment of small cells, in order to bring the networkcapabilities closer to the user demand, and the development of hybrid access networks,heterogeneous both in terms of radio access technology and cell dimension;

– Improved Spectral Efficiency (expected 6 time increase in 2011- 2020), based on higher-order MIMO and AAS, as well as Cooperative MultiPoint (CoMP) techniques.

6 Conclusions

Massive deployment of small cells is undergoing and cannot be reversed, especially becausemost of the users and usage scenarios are with low mobility and the users stay in the samecell or are moving between a reduced number of neighbouring cells. This trend will helpto optimize the access network and will bring it closer to the users’ demands, but also willbring a number of challenges for cellular network technologies. For example, it is necessaryto add self-organizing features and they will need to be enhanced as the number of deployed(small) cells keeps growing. Another challenge is related to interference coordination andcancellation, which will likely be realized by a joint approach: transmitter-based techniquesand receiver based techniques.

Reducing signaling, shutting down or idling unused BSs, adapting the network to specifictraffic patterns and several other techniques can be seen as solutions to reduce the interferenceand, in general, to optimize the access networks and to enhance their performances.

The main challenge for any network is to follow the users and their trends and demands.For mobile access this will be a must and the use of small cells may be an essential instrument.The entire network and especially the access network must be reconfigurable, according tothe traffic evolution. The social face of the users may have a higher impact on the small cellcharacteristics and challenges, as well as on the procedures of dynamic network reconfigu-ration.

Acknowledgments This research activity was supported by UEFISCDI Romania under the Grant No.20/2012 “SaRaT-IWSN”, by the Ministry of Communications and Information Society of Romania underthe Grant No. 106/2011 “Evolution, implementation and transition methods of DVB radiobroadcasting usingefficiently the radio frequencies spectrum” and by the Sectoral Operational Programme Human ResourcesDevelopment 2007–2013 of the Ministry of European Funds through the Financial Agreements POS-DRU/107/1.5/S/76903 and POSDRU/159/1.5/S/132397.

References

1. The economics of the thousand times challenge: Spectrum, efficiency and small cells. http://techneconomyblog.com/2012/11/02/the-economics-of-the-thousand-times-challenge (2012).

2. How interworking works: A detailed look at 802.11u and Hotspot 2.0 mechanisms. Tech. rep., RuckusWireless (2013).

3. Small cell market status. Tech. Rep. SCF050, Small Cell Forum (2013).4. Small cells - big ideas. Tech. Rep. SCF030, Small Cell Forum (2014).

123

Small Cells in Cellular Networks 1625

5. Björnson, E., Kountouris, M. & Debbah, M. (2013). Massive MIMO and small cells: Improving energyefficiency by optimal soft-cell coordination. CoRR abs/1304.0553.

6. Deb, S., Monogioudis, P., Miernik, J., & Seymour, J. (2014). Algorithms for enhanced inter-cell inter-ference coordination (eICIC) in LTE HetNets. IEEE/ACM Transactions on Networking, 22(1), 137–150.doi:10.1109/TNET.2013.2246820.

7. Fratu, O., Halunga, S., Perju, C., Martian, A. & Marcu, I. (2010) On the availability of CDMA channelsfor secondary users. In: Applied sciences in biomedical and communication technologies (ISABEL),2010 3rd international symposium on, pp. 1–5. doi:10.1109/ISABEL.2010.5702775.

8. Fratu, O., Vulpe, A., Halunga, S. & Craciunescu, R. (2013). Interference analysis for inter-band CarrierAggregation in LTE-Advanced. In: Telecommunication in modern satellite, cable and broadcasting ser-vices (TELSIKS), 2013 11th international conference on, vol. 01, pp. 199–208. doi:10.1109/TELSKS.2013.6704919.

9. Kishiyama, Y., Benjebbour, A., Nakamura, T., & Ishii, H. (2013). Future steps of LTE-A: Evolutiontoward integration of local area and wide area systems. Wireless Communications, IEEE, 20(1), 12–18.doi:10.1109/MWC.2013.6472194.

10. Liu, M., Li, B.& Tong, Z. (2013). Cell clustering and resource allocation scheme for CoMP SU-MIMO.Wireless Personal Communications 69(4), 1399–1411. doi:10.1007/s11277-012-0641-3. http://dx.doi.org/10.1007/s11277-012-0641-3.

11. Ltd., H. (2012). Views on enhancement of system capacity and energy efficiency toward release12 andonward. 3GPP RAN workshop on release 12 and onwards, Ljubljana, Slovenia.

12. Nguyen, T. D., & Han, Y. (2006). A proportional fairness algorithm with QoS provision in down-link OFDMA systems. Communications Letters, IEEE, 10(11), 760–762. doi:10.1109/LCOMM.2006.060750.

13. Ohwatari, Y., Miki, N., Asai, T., Abe, T. & Taoka, H. (2011). Performance of advanced receiver employ-ing interference rejection combining to suppress inter-cell interference in LTE-Advanced downlink. In:Vehicular technology conference (VTC Fall), 2011 IEEE, pp. 1–7. doi:10.1109/VETECF.2011.6093196.

14. Priyanto, B., Kant, S., Rusek, F., Hu, S., Chen, J. & Wugengshi, C. (2013). Robust ue receiver withinterference cancellation in lte advanced heterogeneous network. In: Vehicular technology conference(VTC Fall), 2013 IEEE 78th, pp. 1–7 (2013). doi:10.1109/VTCFall.6692396.

15. Richter, F., Fehske, A. & Fettweis, G. (2009). Energy efficiency aspects of base station deploymentstrategies for cellular networks. In: Vehicular technology conference fall (VTC 2009-Fall), 2009 IEEE70th, pp. 1–5 . doi:10.1109/VETECF.2009.5379031.

16. Syputa, R. (2011). Maravedis: Achieving capital, spectral efficiency beyond 4G. http://www.fiercewireless.com/tech/story/maravedis-achieving-capital-spectral-efficiency-beyond-4g/2011-04-08.

17. Telecom, S. (2012). Efficient spectrum resource usage for next-generation N/W. 3GPP RAN workshopon release 12 and onwards, Ljubljana, Slovenia.

18. Vukovic, A. (2011) Making LTE greener & energy-efficient to enable cost-reduction in the network. 7thLTE world summit, Amsterdam.

19. Vulpe, A., Fratu, O., Mihovska, A. & Prasad, R. (2013) A multi-carrier scheduling algorithm for LTE-advanced. In: Wireless personal multimedia communications (WPMC), 2013 16th international sympo-sium on, pp. 1–5.

20. Yilmaz, O.N., Hämäläinen, J. & Hämäläinen, S. (2013). Optimization of adaptive antenna sys-tem parameters in self-organizing lte networks. Wireless Network 19(6), 1251–1267. doi:10.1007/s11276-012-0531-3. http://dx.doi.org/10.1007/s11276-012-0531-3.

21. You, C., Yoon, G., Seo, C., Portugal, S., Park, G., Jung, T., Liu, H. & Hwang, I. (2011). Intercell interferencecoordination using threshold-based region decisions. Wireless Personal Communications 59(4), 789–806.doi:10.1007/s11277-011-0265-z. http://dx.doi.org/10.1007/s11277-011-0265-z.

22. Zhioua, G., Labiod, H., Tabbane, N. & Tabbane, S. (2013). LTE Advanced relaying standard: A survey.Wireless Personal Communications 72(4), 2445–2463. doi:10.1007/s11277-013-1157-1. http://dx.doi.org/10.1007/s11277-013-1157-1.

123

1626 O. Fratu et al.

Octavian Fratu received the PhD degree in Electronics and Telecom-munications from the University POLITEHNICA of Bucharest, Roma-nia in 1997. He achieved postdoctoral stage as senior researcher in3rd generation mobile communication systems, based on a researchcontract between CNET-France, ENS de Cachan - France and Uni-versite Marne la Vallee - France. He is currently an associated pro-fessor in Electronics, Telecommunications and Information TheoryFaculty. His research interests include Digital Mobile Communica-tions, Radio Data Transmissions, Mobile Communications and Wire-less Sensor Networks, Radio communication networks. His publica-tions include more than 100 papers published in national or interna-tional scientific journals or presented at international conferences. Heparticipated as director or collaborator in many international researchprojects such as “Optimization and rational use of Wireless Commu-nication Bands (ORCA)”, NATO Science for Peace Research project(2013–2016), „Reconfigurable Interoperability of Wireless Communi-

cations Systems (RIWCoS)”, NATO Science for Peace Research project (2007–2010), „REDICT –RegionalEconomic Development by ICT/New media clusters” (FP7 CSA project, 2008-2009), “ATHENA - DigitalSwitchover: Developing Infrastructures for Broadband Access”, (FP6 STREP project, 2004–2006) and other.

Alexandru Vulpe received the PhD degree in Electronics, Telecom-munications and Information Technology from the UniversityPOLITEHNICA of Bucharest, Romania in 2014. His research inter-ests include, among others, Mobile Communications, Wireless SensorNetworks, Quality-of-Service, Radio Resource Management, MobileApplications. His publications include more than 30 papers pub-lished in journals or presented at international conferences. He par-ticipated as a researcher in a number of national or internationalprojects such as „Reconfigurable Interoperability of Wireless Com-munications Systems (RIWCoS)”, NATO Science for Peace Researchproject (2007–2010), “Wireless Hybrid Access System with UniqueAddressing (SAWHAU)”, Romanian “Partnerships in priority fields”research project (2008-2011) „eWALL – eWall for Active Long Liv-ing” (FP7 project, 2013–2016), “Optimization and Rational Use ofWireless Communication Bands (ORCA)”, NATO Science for Peaceproject (2013–2015).

Razvan Craciunescu received the Engineer’s degree and the Mas-ter’s degree in electronic engineering and telecommunications fromthe University POLITEHNICA of Bucharest, Bucharest, Romania,in 2010 and 2012, respectively. He is currently a Ph.D. student,in the field of communication systems and optimization, at Uni-versity POLITEHNICA of Bucharest, Romania. Since 2012 he isTeaching Assistant at the Faculty of Electronics, Telecommunicationsand Information Technology within the University POLITEHNICAof Bucharest. His research interests include Mobile Communications,Wireless Sensor Networks and System Optimization. His publicationsinclude more than 10 papers published in nationally and abroad or pre-sented at international conferences. He participated as researcher innational and international projects such as „Optimization and Rationaluse of wireless Communication bAnds (ORCA)”, NATO Science forPeace Research project (2013–2016), „eWALL –eWall for Active LongLiving” (FP7 project, 2013–2016).

123

Small Cells in Cellular Networks 1627

Simona Halunga received the M.S. degree in electronics and telecom-munications in 1988 and the Ph.D. degree in communications fromthe University POLITEHNICA of Bucharest, Bucharest, Romania, in1996. Between 1996–1997 she followed postgraduate courses in Man-agement and Marketing, organized by the Romanian Trade and Indus-try Chamber and Politehnica University of Bucharest, in collaborationwith Technical Hochschule Darmstadt, Germany, and in 2008- postgraduate courses in Project Management - Regional Centre for Con-tinuous Education for Public Local Administration, Bucharest She hasbeen Assistant Professor (1991–1996), Lecturer (1997 - 2001), Asso-ciate Professor (2001–2005) and from 2006 she is a full professorat in Politehnica University of Bucharest, Electronics, Telecommuni-cations and Information Theory Faculty, Telecommunications Depart-ment. Between 1997-1999 she has been a Visiting Assistant Professorat Electrical and Computer Engineering Department, University of Col-orado at Colorado Springs, USA. Her domain of interest are Multiple

Access Systems & Techniques, Satellite Communications, Digital Signal Processing for Telecommunication,Digital Communications -Radio Data Transmissions, Analog and Digital Transmission Systems and DigitalSignal Processing for Telecommunications.

123