Upload
krishnan-venkataraghavan
View
67
Download
1
Tags:
Embed Size (px)
DESCRIPTION
This document provides interference mitigation techniques in LTE- A good read
Citation preview
Interference Management in
LTE-Advanced Heterogeneous Networks
Using Almost Blank Subframes
HISHAM EL SHAER
Master's Degree Project
Stockholm, Sweden
XREESB 2012:006
1
Interference Management In LTE-Advanced Heterogeneous Networks
Using Almost Blank Subframes
Hisham El Shaer
March 2012
Degree Project in Signal Processing
Stockholm, Sweden 2012
2
Abstract
Long term evolution (LTE) is the standard that the Third-generation Partnership Project (3GPP)
developed to be an evolution of UMTS. LTE offers higher throughput and lower latency than UMTS and
this is mainly due to the larger spectrum used in LTE but in terms of spectrum efficiency LTE does not
offer a lot of improvements compared to UMTS. The reason for that is that current technologies such as
UMTS and LTE are approaching the theoretical boundaries in terms of spectral efficiency. Since
spectrum has become a scarce resource nowadays, new ways have to be found to improve the network
performance and one of the studied approaches to do that is to enhance the network topology.
The concept of heterogeneous networks has attracted a lot of interest recently as a way to improve the
performance of the network. The heterogeneous networks approach consists of complementing the Macro
layer with low power nodes such as Micro or Pico base stations. This approach has been considered a way
to improve the capacity and data rate in the areas covered by these low power nodes; they are mostly
distributed depending on the areas that generate higher traffic.
Since cell selection for the users is based on the downlink power level and due to the transmitting power
differences between Macro and Pico nodes, Pico nodes might be under-utilized, meaning that a low
number of users are attached to the Pico nodes. As a solution to this problem an offset to the received
power measurements used in cell selection is applied allowing more users to be attached to the Pico
nodes, this solution is called Range Extension which refers to the extended coverage area of the Pico
nodes.
The problem with Range Extension is that it drastically increases the interference that the Macro nodes
impose on the Pico nodes users in the Range Extension area in terms of data and control channels.
Enhanced Inter-Cell Interference Coordination (eICIC) schemes have been proposed to combat the heavy
interference in the Range Extension case ranging from frequency domain schemes like carrier aggregation
to time domain schemes like Almost Blank Subframes (ABS).
The focus of this thesis will be on the ABS solution which consists of reserving a group of subframes
during which the Macro nodes are partially muted allowing the users in the range extension area to be
served with lower interference.
The objective of this thesis work is to introduce a closed form expression to calculate the Almost Blank
Subframes allocation in order to maximize the normalized cell-edge users throughput. The derivations are
carried out for a simplified model of a telecommunications network. The expression will be validated
with simulations involving different users and Pico nodes distributions and different channel models (ITU
channel models and Spatial Channel Models). Another goal is to try to have a deeper understanding and
concrete conclusions about the different heterogeneous deployments.
3
Acknowledgment
This work would not have been possible to complete without the support of many people to whom I want
to show my gratitude. First of all I would like to dedicate this thesis to my soon to be born daughter.
I would like to thank my family and my wife for their continuous support and patience. I also want to
thank my supervisor at Ericsson Niklas Wernersson and my manager Maria Edvardsson for their help and
guidance during the project. Finally I want to thank my supervisor at KTH Mats Bengtsson for his
support before and during the thesis.
4
Table of Contents
1. Introduction .......................................................................................................................................... 6
Brief about LTE .............................................................................................................................. 6 1.1
1.1.1 LTE requirements .................................................................................................................. 7
1.1.2 LTE downlink transmission scheme ...................................................................................... 7
1.1.3 Cyclic-Prefix insertion ............................................................................................................ 8
1.1.4 Spectrum flexibility ............................................................................................................... 9
1.1.5 Physical resources ............................................................................................................... 10
1.1.6 Enhancements introduced in LTE advanced (Release 10) .................................................. 12
Introduction to Heterogeneous Networks (HetNets) ................................................................. 14 1.2
1.2.1 Motivation and description of HetNets .............................................................................. 15
Goal of the thesis ........................................................................................................................ 16 1.3
2. Range extension and associated problems ......................................................................................... 17
Range extension introduction ..................................................................................................... 17 2.1
Range extension advantages ...................................................................................................... 18 2.2
Interference effects associated to range extension ................................................................... 18 2.3
3. Inter-cell interference available solutions .......................................................................................... 19
Frequency domain multiplexing inter-cell interference coordination scheme .......................... 19 3.1
Time domain multiplexing inter-cell interference coordination scheme (Almost Blank 3.2
Subframes) .............................................................................................................................................. 20
4. Range extension with almost blank sub-frames (ABS) ....................................................................... 22
Common reference signals (CRS) interference ........................................................................... 22 4.1
Proposed formula to calculate the ABS ratio to maximize the performance. ............................ 23 4.2
4.2.1 General model ..................................................................................................................... 24
4.2.2 Simulations validating the previous results ........................................................................ 30
4.2.3 Example to validate the general model results .................................................................. 37
4.2.4 Example to validate the general model results (without the assumption of Ptotal_Pico) ...... 40
4.3 Summary ........................................................................................................................................... 43
5. System simulation results ................................................................................................................... 44
The Raptor simulator .................................................................................................................. 44 5.1
System simulation assumptions .................................................................................................. 45 5.2
Simulation results ....................................................................................................................... 45 5.3
5
5.3.1 Who wins and who loses in terms of throughput in a heterogeneous network
deployment? ....................................................................................................................................... 45
5.3.2 Simulations demonstrating the benefits of using ABS ........................................................ 48
5.3.3 Simulations validating the ABS ratio formula for different users and Pico-eNBs
distributions. ....................................................................................................................................... 53
5.3.4 Does having a high range extension give a better performance? ...................................... 68
6. Conclusions ......................................................................................................................................... 71
7. Future work ......................................................................................................................................... 71
8. List of Acronyms .................................................................................................................................. 72
9. References .......................................................................................................................................... 73
6
1. Introduction
In this section an introduction about Long Term Evolution (LTE) will be presented focusing only on the
downlink since the thesis work mainly focuses on the downlink1 transmission, then an explanation of the
heterogeneous networks (HetNets) concept, its motivation and its different types will be introduced.
Finally the goal of the thesis and the contributions done in it will be introduced.
Brief about LTE 1.1
Through the past few years the mobile broadband technology was released making it possible for
applications such as live streaming, online gaming and mobile TV to be used on mobile handsets.
However, the data rate requirements for these applications have grown exponentially.
The Third-generation Partnership Project (3GPP)2 started working on solutions to fulfill the need for high
data rates and came up with HSPA3 which is currently used in 3G phones for the before mentioned
applications.
In order to ensure the competitiveness of its standards in the future, 3GPP developed the Long Term
Evolution (LTE) to be the 4th generation of mobile telephony. LTE as defined by the 3GPP [12] is the
evolution of the 3rd
generation of mobile communications (UMTS). The main goal of LTE is to introduce
a new radio access technology with a focus on high data rates, low latency and packet optimized radio
access technology, LTE is also referred to as E-UTRAN (Evolved UMTS Terrestrial Radio Access
Networks).
In December 2008, the LTE specification was published as part of Release 8 and the first implementation
of the standard was deployed in 2009. The first release of LTE, namely release 8, supports radio network
delay less than 5ms and multiple input multiple output (MIMO) antenna techniques which allow
achieving very high data rates.
Later on in December 2009 release 9 has been introduced with extensions to various features that existed
in release 8 such as Closed Subscriber Group (CSG) and Self Organizing Network (SON). It added also
new features such as Location Services (LCS) and Multimedia Broadcast Multicast Services (MBMS).
Finally release 10 has been introduced in March 2011 which is also called LTE-Advanced and it added
new features such as carrier aggregation, relaying and heterogeneous deployments which will be all
discussed in details later.
The rest of this LTE introduction will focus on the LTE requirements, the downlink transmission scheme
and the spectrum flexibility.
1 Downlink refers to the communication from the base station to the mobile user.
2 3GPP is a collaboration between groups of telecommunications associations with the goal of standardizing,
developing and maintaining of a globally 3rd
generation mobile phone system. 3 HSPA is short for High Speed Packet Access which is an amalgamation of the 2 protocols High Speed Downlink
Packet Access (HSDPA) and High Speed Uplink Packet Access (HSUPA)
7
1.1.1 LTE requirements
The main requirements for an LTE system were identified in [1] and the most important ones can be
summarized in the following points.
- Data rate: Peak data rates of 100 Mbps (downlink) and 50 Mbps (uplink) for a 20 MHz spectrum
allocation.
- Throughput: The target downlink average user throughput per MHz is enhanced 3 to 4 times
compared to release 64. The target for uplink average user throughput per MHz is enhanced 2 to 3
times compared to release 6.
- Bandwidth: Scalable bandwidths of 5, 10, 15 and 20 MHz shall be supported. Also smaller
bandwidths smaller than 5 MHz shall be supported for more flexibility like 1.4 MHz and 3 MHz.
- Interworking: Interworking with existing UTRAN/GERAN5 and non-3GPP systems.
- Mobility: The system should be optimized for low mobile speeds (0-15 km/h) but should also
support higher mobile speeds including high speed train environments.
- Coverage: The targets stated above should be met for 5 km cells6 and some degradation in
throughput and spectrum efficiency for 30 km cells. Finally 100 km cells and larger are not covered
by the specifications.
1.1.2 LTE downlink transmission scheme
The LTE downlink transmission scheme is based on Orthogonal Frequency Division Multiplexing
(OFDM) where the available spectrum is divided into multiple carriers called subcarriers. Data symbols
are independently modulated and transmitted over orthogonal subcarriers where modulation schemes such
as QPSK, 16QAM and 64 QAM are used. The subcarriers being orthogonal means that there is no
interference between the subcarriers. OFDM transmission is a block based transmission where during
each OFDM symbol interval N modulation symbols are transmitted in parallel.
In practice an OFDM signal can be generated using IFFT (Inverse Fast Fourier Transform) digital signal
processing which is an efficient way to generate an OFDM signal. Figure 1 illustrates an OFDM
transmitter where OFDM modulation is done by means of IFFT processing [5].
As a first step the bits from the encoder are modulated into symbols, then these symbols are passed to a
serial to parallel converter to be able to process the N symbols through the IFFT modulator
4 3GPP standards are structured as releases, release 6 added mainly HSUPA and MBMS.
5 UTRAN and GERAN are responsible for the specifications of the Radio Access part of UMTS (3G) and
GSM/EDGE (2G) respectively. 6 A cell is the term used to describe the coverage area of a single base station and is usually illustrated by a
hexagonal shape.
8
simultaneously, then the IFFT samples are passed to a parallel to serial converter and to a digital to analog
converter which sends the signal to the up-converter to be transmitted.
Figure 1: OFDM modulation by means of IFFT processing
1.1.3 Cyclic-Prefix insertion
The main advantage of an OFDM signal is that it can be demodulated without any interference between
the subcarriers due to the orthogonality between them.
However, considering a time dispersive channel7, the orthogonality between the subcarriers will, at least,
be partly lost. This loss of orthogonality in the time dispersive channel is due to the fact that the
demodulator correlation interval of one path will overlap with the symbol boundary of another path as
shown in Figure 2.
Figure 2: Time dispersion and received signal timing
Cyclic-prefix insertion implies that the last part of the OFDM symbol is copied and inserted at the
beginning of the OFDM symbol as shown in Figure 3, so cyclic-prefix basically increases the length of
the OFDM symbol from Tu to Tu+Tcp, where Tcp is the length of the cyclic-prefix which in turn reduces
the OFDM symbol rate.
7 Time dispersive channels are channels where multi-path exists and it is characterized by its time delay spread
which is the total time interval during which reflections with significant energy reach the receiver.
Parallel to
serial
converter
9
Figure 3: Cyclic prefix insertion
Cyclic-prefix preserves the orthogonality between the subcarriers in the case of a time dispersive channel
as long as the span of the time shift or time difference between symbols is shorter than the cyclic prefix
length.
The problem with cyclic-prefix is that only a part of the received signal power is utilized by the OFDM
modulator, so there is a power loss. Also there is a loss in terms of bandwidth as the symbol rate is
reduced due to the insertion of the cyclic-prefix. One way to combat this loss of bandwidth is to reduce
the subcarrier spacing. A detailed description of OFDM and cyclic-prefix is given in [2] and [18].
1.1.4 Spectrum flexibility
Spectrum flexibility is one of the main characteristics of LTE radio-access technology. The main reason
of this spectrum flexibility is to allow for the deployment of LTE radio-access in different frequency
bands with different sizes since spectrum has become a scarce resource. This flexibility includes 2 main
areas as follows.
1.1.4.1 Flexibility in duplex arrangements
One important aspect of LTE is the possibility to operate in both paired and unpaired spectrum. Paired
frequency bands mean that the uplink and downlink transmissions use separate frequency bands while
unpaired spectrum means that uplink and downlink transmissions share the same frequency band.
LTE supports both frequency and time division based duplex arrangements.
Frequency-Division Duplex (FDD), as shown in Figure 4, implies that uplink and downlink transmissions
take place in different and sufficiently separated frequency bands.
Time-Division Duplex (TDD), as shown in Figure 4, implies that uplink and downlink operate in different
non-overlapping time slots.
10
Figure 4: TDD and FDD operation
1.1.4.2 Bandwidth flexibility
Another important aspect in the LTE operation is the possibility to operate in different transmission
bandwidths in uplink and downlink. The reason for that is that the amount of spectrum available for LTE
deployment can vary a lot between frequency bands and also depending on the operator. Also this
bandwidth flexibility gives the possibility for gradual frequency bands migration from other radio-access
technologies.
1.1.5 Physical resources
1.1.5.1 LTE time domain structure
Downlink transmissions are organized in (radio) frames of length 10 ms which, in turn, are divided into
10 equally sized subframes of 1ms duration each. As illustrated in Figure 5, each subframe consists of 2
time slots of length Tslot=0.5 ms, where each time slot consists of a number of OFDM symbols including
cyclic prefix.
11
Figure 5: LTE frame structure
1.1.5.2 LTE frequency domain structure
A resource element is the smallest physical resource in LTE and it consists of one subcarrier during one
OFDM symbol, resource elements are grouped into resource blocks. A resource block has a duration of
0.5 ms (one slot) and a bandwidth of 180 KHz (12 subcarriers) so each resource block consists of 12x7 =
84 resource elements in the case of normal cyclic prefix and 12x6 = 72 in the case of extended cyclic
prefix. The LTE physical layer specification allows for a carrier to consist of any number of resource
blocks in the frequency domain, ranging from a minimum of 6 resource blocks up to a maximum of 110
resource blocks which can be translated in frequency to a range between 1 MHz and 20 MHz with very
fine granularity that allows for the spectrum flexibility discussed before.
The time-frequency physical resources in LTE are shown in Figure 6.
12
Figure 6: LTE frequency domain structure
1.1.6 Enhancements introduced in LTE advanced (Release 10)
The most important target for LTE release 10 was to be able to fulfill the IMT-Advanced requirements.
IMT is a global broadband multimedia international mobile telecommunication system that the ITU
(International Telecommunication Union) has been coordinating along with governments, industry and
private sector. IMT-Advanced is the term that ITU uses to describe radio-access technologies beyond
IMT-2000.
Some of the IMT-Advanced requirements are listed as follows [4]:
- Support for channel bandwidth up to 40 MHz.
- Peak spectral efficiencies of 15 bit/s/Hz in downlink (corresponding to peak rate of 600 Mbit/s).
- Peak spectral efficiencies of 6.75 bit/s/Hz in uplink (corresponding to peak rate of 270 Mbit/s).
- Control plane latency of less than 100 ms.
- User plane latency of less than 10 ms.
The main reason for LTE release 10 to be called LTE-Advanced is that its radio-access technology is
fully compliant with the IMT-advanced requirements.
In the following we introduce some of the most important enhancements and features introduced in LTE-
Advance.
Time
Freq
uen
cy
13
1.1.6.1 Carrier aggregation
As mentioned before the previous releases of LTE have introduced a lot of flexibility in terms of
bandwidth as it allows operating in bandwidths ranging from 1 MHz to 20 MHz in both paired and
unpaired modes. In LTE release 10 the transmission bandwidth can be further extended using carrier
aggregation.
The main idea is to aggregate several component carriers and jointly use them for transmission to and
from single terminals. Up to 5 transmission components can be aggregated whether they belong to the
same frequency range or not and this feature allows the transmission bandwidth to reach 100 MHz, it also
allows to make use of the fragmented spectrum, as operators with fragmented spectrum can use this
feature to offer high data-rates by combining all the small spectrum fragments into a sufficiently large
component.
1.1.6.2 Extended multi antenna transmission
In LTE release 10, downlink spatial multiplexing has been expanded to support up to 8 transmission
layers so together with carrier aggregation a downlink data rate of up to 30 bit/s/Hz can be achieved.
In terms of uplink, spatial multiplexing of up to 4 layers is supported by release 10, this allows for an
uplink data-rate of 15 bit/s/Hz.
1.1.6.3 Relaying
Relaying implies that the mobile node is connected to its serving cell through a relay node that is
wirelessly connected to the serving node using the LTE radio-interface technology.
From a mobile node perspective the relay node is invisible as the mobile node can only see that it is
connected to the serving base station. This feature has the advantage of improving the coverage especially
in indoor environments.
1.1.6.4 Heterogeneous deployments
Heterogeneous deployments refer to deployments where we have base stations with different transmission
powers and coverage areas sharing, fully or partially, the same set of frequencies and having an
overlapping geographical coverage. An example of Heterogeneous networks is having a Pico-eNB8
placed in the coverage area of a Macro-eNB9.
Heterogeneous networks, also called HetNets, were supported by release 8 and 9 but release 10
introduced improved inter-cell interference handling making HetNet scenarios more robust. The rest of
this report will focus on HetNets and the Enhanced Inter-Cell Interference Coordination (eICIC) used by
release 10 to combat the interference caused by the Macro-eNBs to the Pico-eNB users.
8 Pico-eNB is a low transmitting power base station that has limited coverage and will be explained in details later.
9 Macro-eNB is the normal base station which is called eNB (short for evolved node B.) in LTE.
14
Introduction to Heterogeneous Networks (HetNets) 1.2
Mobile broadband traffic has been growing very fast through the past few years; it surpassed voice traffic
and is expected to grow much faster in the future. This growth is mainly driven by new services and the
evolution of terminals capabilities. Annual traffic is predicted to double annually during the next five
years so that by 2014 the average user traffic will be about 1 GB of data per month compared to 100 or
200 MB now [5].
The mobile industry has been striving to improve data rates indoors and outdoors to be able to meet the
evolution of mobile services. There are several options that can be considered to increase the network
capacity and meet traffic and data rates demands such as:
- Improving the Macro layer: Upgrading the radio access of existing sites whether HSPA or LTE
would increase the data rates, this can be done by adding more spectrum which can notably enhance
the downlink data rates although the enhancement is negligible in the uplink.
Another option would be to add more antennas or enhance the processing within and between the
nodes. But at some point the capacity and data rates enhancements introduced by improving the radio
access of the nodes would be insufficient.
- Densifying the Macro layer: Increasing the number of Macro sites in urban and dense areas has
been a popular approach taken by operators to combat the traffic increase, it has the advantage of
decreasing the distance between the user and the serving base station so the uplink data rate is largely
enhanced and of course it has a big effect on the downlink data rates as well. The problem with this
approach is that it is very expensive to add more Macro sites in terms of cost, finding suitable
locations to deploy new sites and interference as we are placing high power nodes closer to each
other.
- Heterogeneous networks: This approach consists of complementing the Macro layer with low power
nodes such as Micro and Pico base stations. This approach has been considered a way to improve the
capacity and data rate in the areas covered by these low power nodes; they are mostly distributed in
an unplanned manner depending on the areas that generate higher traffic.
Through the rest of the report we will focus on Heterogeneous networks and specifically on the Pico base
stations deployments that will be referred to as Pico-eNB for the rest of the report and will be described in
details in the following section.
15
1.2.1 Motivation and description of HetNets
The concept of Heterogeneous networks has attracted a lot of interest recently to optimize the
performance of the network. Spectral efficiency of current systems like WCDMA and LTE is
approaching theoretical boundaries [13], we can see that from the fact that LTE release 8 does not offer a
lot of improvements in terms of spectral efficiency compared to UMTS, instead LTE improves system
performance by using more spectrum and since spectrum has been a scarce resource in the past few years
a different approach must be considered to improve network performance.
The main approach to enhance the performance is to improve the network topology. This is done in the
scenario of Heterogeneous networks by overlaying the planned network of high power Macro base
stations with smaller low power Pico base stations that are distributed in an unplanned manner or simply
in hotspots where a lot of traffic is generated. These deployments can improve the overall capacity and
the cell edge users10
performance. [2]
Figure 7: Heterogeneous network using Pico-eNBs
1.2.1.1 Properties of Pico base stations:
1) They have a transmission power of 1W.
2) They can be deployed to eliminate coverage holes.
3) Offer high data rate and capacity where they are deployed.
4) Offloading the Macro-eNBs by serving some users that used to belong to the Macro-eNBs, which
allows the Macro-eNB to serve better its users.
5) Due to their low transmission power and small physical size they can offer flexible site acquisitions.
In the following section we will explain how to optimize the performance of Pico-eNBs and the problems
that face this approach.
10
Cell-edge users, in this report, are defined to be the worst 5% of the total number of users in terms of capacity or
throughput.
16
Goal of the thesis 1.3
The goal of this thesis is to find a closed form expression for the Almost Blank Subframes (ABS)
allocation that optimizes the network performance in terms of cell edge users throughput. Most of the
previous work has focused on the allocation of ABS depending on the ratio between the number of Macro
users per cell and the number of range extension users per Pico cell as in [14] and [17] which basically
means that the ABS allocation depends on the ratio of the number of the Macro users to the number of
range extension users belonging to each Pico-eNB or just choosing the Pico-eNB with the maximum
number of range extension users and applying that to all the Pico-eNBs. Through this thesis we will
deduce a formula, theoretically and using simulations, for the ABS allocation that depends on the ratio of
the number of Macro users to the total number of range extension Pico users in a cell and it will be proven
that it gives a better performance in terms of cell edge users throughput.
The main contributions of this thesis can be summarized in the following points.
1. Running system simulations in order to have solid conclusions about HetNets, concerning the users
who experience an increase or decrease of throughput after adding the Pico layer and the reasons
behind that. Also extract some conclusions about the Almost Blank Subframes as a TDM interference
coordination scheme in terms of its advantages and the winners and losers in this scenario.
2. Deduce a closed form expression for the ABS allocation that optimizes the performance in terms of
cell edge users throughput while keeping a fair level of normalized throughput. The deduction will be
done theoretically and will be validated using system level simulations.
3. Implement a graphical interface for the Raptor system simulator, which is the simulator I am working
on in Ericsson. This graphical interface will be used to illustrate a hexagonal cellular network
featuring the Macro-eNBs, Pico-eNBs and the simulated users. It allows focusing on a specific user or
group of users and studying their statistics. An example of this graphical interface will be presented in
the simulations section.
17
2. Range extension and associated problems
Range extension introduction 2.1
Cell selection in LTE is based on terminal measurements of the received power of the downlink signal or
more specifically the cell specific reference (CRS) downlink signaling.
However; in a heterogeneous network we have different types of base stations that have different
transmission powers including different powers of CRS. This approach for cell selection would be unfair
to the low power nodes (Pico-eNBs) as most probably the terminal will choose the higher power base
stations (Macro-eNBs) even if the path loss to the Pico-eNB is smaller and this will not be optimal in
terms of:
- Uplink coverage: as the terminal has a lower path loss to the Pico-eNB but instead it will select the
Macro-eNB.
- Downlink capacity: Pico-eNBs will be under-utilized as fewer users are connected to them while the
Macro-eNBs could be overloaded even if Macro-eNBs and Pico-eNBs are using the same resources
in terms of spectrum, so the cell-splitting gain is not large and the resources are not well utilized.
- Interference: due to the high transmission power of the Macro-eNBs, then the Macro-eNB
transmission is associated with a high interference to the Pico-eNB users which denies them to use
the same physical resources.
As a solution for the first 2 points cell selection could be dependent on estimates of the uplink path loss,
which in practice can be done by applying a cell-specific offset to the received power measurements used
in typical cell selection. This offset would somehow compensate for the transmitting power differences
between the Macro-eNBs and Pico-eNBs; it would also extend the coverage area of the Pico-eNB, or in
other words extend the area where the Pico-eNB is selected. This area is called Range Extension and is
illustrated in Figure 8.
Figure 8: range extension area illustration
18
Range extension advantages 2.2
1) Applying range extension would maximize the achievable uplink SINR which in turn maximizes the
uplink data rate.
2) The terminal transmit power would be reduced as the path loss to the Pico-eNB is lower than the one
to the Macro-eNB so the interference to other cells would be reduced and the uplink system
efficiency would be improved.
3) It also allows more users to be connected to the Pico-eNB, thus increasing the cell splitting gain.
4) Since the Macro-eNB transmits to fewer users then the interference it applies on the Pico-eNB is
reduced and the Pico-eNBs can reuse the resources more efficiently so the downlink system
efficiency is maximized as well.
Interference effects associated to range extension 2.3
Due to the difference in transmission powers of the Macro-eNBs and the Pico-eNBs, in the range
extension area, illustrated in Figure 8, where the Pico-eNB is selected by the terminal while the downlink
power received by that terminal from the Macro-eNB is much higher than the power it receives from the
Pico-eNB, this makes the users in the range extension area more prone to interference from the Macro-
eNB.
So along with the benefits of range extension comes the disadvantage of the high inter-cell interference
that the Macro layer imposes on the users in the range extension area of the Pico layer. Figure 9 illustrates
the comparison of 2 users connected to the Pico-eNB where:
- User 1 is placed close to the Pico-eNB so we will call it center Pico user, this is not affected very
much by the Macro-eNB interference as the downlink received power from the Pico-eNB is higher
than the one received from the Macro-eNB.
- User 2 is placed farther from the Pico-eNB, in the range extension area, and as discussed before this
user endures a severe interference from the Macro-eNB.
Solutions for the high interference levels in the range extension area will be discussed in the next section.
Figure 9: range extension interference
19
3. Inter-cell interference available solutions
The enhanced Inter-Cell Interference Coordination (eICIC) in heterogeneous networks introduced in
LTE-Advanced has been a hot topic lately as without an efficient inter-cell interference scheme the range
extension concept loses its advantage and efficiency. The problem with ICIC schemes in releases 8 and 9
was that they were only considering data channels and did not focus on the interference between control
channels, so LTE release 10 solves this problem with the solutions in the following subsections.
The solutions are mainly divided into frequency domain solutions such as carrier aggregation and time
domain solutions such as almost blank subframes (ABS), and they will be discussed in details in the
following.
Frequency domain multiplexing inter-cell interference coordination 3.1
scheme
The main FDM interference cancellation method used in LTE-Advanced is carrier aggregation; this
feature has been discussed in section 1.1.6.1 which is one of the most important features of LTE-
Advanced and it basically enables an LTE-Advanced user equipment (UE) to be connected to several
carriers simultaneously.
Carrier aggregation not only allows resource allocation across carriers but also allows scheduler based
fast switching between carriers without time consuming handovers, which means that a node can schedule
its control information on a carrier and its data information on another carrier.
An example of that concept in a HetNet scenario is to partition the available spectrum into, for example, 2
separate component carriers, and assign the primary component carrier (f1) and the second component
carrier (f2) to different network layers at a time as shown in Figure 10 .
Figure 10: Illustration of eIIC based on carrier aggregation
In the example we have 2 component carriers f1 and f2 where 5 subframes are shown in each carrier.
There are 2 cases, the case of Macro layer usage and the case of Pico layer usage; the subframes are
distributed in control part, the blue part, and data part. The control part in the example only illustrates the
PDCCH, PCFICH and PHICH11
at the beginning of the subframes.
11
See list of acronyms.
20
As shown Figure 10 the Macro layer can schedule its control information on f1 but can still schedule its
users on both f1 and f2 so by scheduling control and data information for both Macro and Pico layers on
different component carriers, interference on control and data can be avoided.
It is also possible to schedule center Pico-eNB users12
data information on the same carrier that the Macro
layer schedules its users as shown in the third subframe in Figure 10, as the interference from the Macro
layer on center Pico-eNB users can be tolerated, while Pico-eNB users in the range extension areas are
still scheduled in the other carrier where the Macro-eNB users are not scheduled.
The disadvantage of carrier aggregation with cross carrier scheduling is that it is only supported by
release 10 terminals and onwards so this feature cannot be used by release 8 and 9 terminals.
Time domain multiplexing inter-cell interference coordination scheme 3.2
(Almost Blank Subframes)
In this approach transmissions from Macro-eNBs inflicting high interference onto Pico-eNBs users are
periodically muted (stopped) during entire subframes, this way the Pico-eNB users that are suffering from
a high level of interference from the aggressor Macro-eNB have a chance to be served.
However this muting is not complete as certain control signals are still transmitted which are:
- Common reference symbols (CRS) which will be explained later
- Primary and secondary synchronization signals (PSS and SSS)
- Physical broadcast channel (PBCH)
- SIB-113 and paging with their associated PDCCH.
These control channels have to be transmitted even in the muted subframes to avoid radio link failure or
for reasons of backwards compatibility, so muted subframes should be avoided in subframes where PSS,
SSS, SIB-1 and paging are transmitted or in other words subframes #0, #1, #5 and #9. Since these muted
subframes are not totally blank they are called Almost Blank Subframes (ABS).
The basic idea is to have some subframes during which the Macro-eNB is not allowed to transmit data
allowing the range extension Pico-eNB users, who were suffering from interference from the Macro-eNB
transmission, to transmit with better conditions. The outline of ABS has been specified by the 3GPP in
[15].
ABS have specific patterns that are configured and communicated between the eNBs over the X2
interface. These patterns are signaled in the form of bitmaps of length 40 subframes, i.e. spanning over 4
frames and they can be configured dynamically by the network using self-optimizing networks (SON)
feature to optimize the ABS ratio according to some criterion that can be the cell-edge users throughput or
load balancing for instance and of course keeping in mind the above mentioned subframes that should be
avoided.
12
Center Pico-eNB users are the users connected to the Pico-eNB but that are not in the range extension area. 13
See acronyms list.
21
Figure 11: Illustration of TDM ICIC
As shown in Figure 11, TDM ICIC using ABS causes a lot of variation in terms of interference between
the subframes, this fact can be used in the sense that the users that suffer from a high level of interference
should be served during these ABS while the users that are closer to the transmitting node or that are not
very much affected by interference can be served during the non-ABS subframes.
So for a Pico-eNB cell, users are categorized into 2 groups in terms of ABS usage this time:
- Users in the range extension area and these users suffer from a high level of interference as explained
before so these users should only be served during the ABS.
- Users closer to the Pico-eNB that are called center Pico users and they are not heavily affected by the
interference from the Macro-eNB due to the good channel they maintain with their serving node. So
these users can be served by any subframe whether ABS or non-ABS.
One of the properties of LTE release 10 is that it allows eNBs to restrict the channel measurements done
by the users attached to them to a specific set or pattern of subframes. The reason for that is that if the
channel state information (CSI) measurements which are responsible of reporting the channel conditions
were to be done jointly for ABS and non-ABS, they will not accurately reflect the interference of either
type of subframes. So the terminals are configured with different CSI-measurement subsets corresponding
to the subframes that the terminal is allowed to use.
Users belonging to the range extension area are only allowed to report CSI measurements for the ABS as
they are only allowed to transmit during these subframes.
Users belonging to the center Pico-eNB area transmit 2 different subsets of the CSI measurements, one
for the ABS and another for the non-ABS as they are allowed to transmit through all the subframes. CRS
interference will be discussed in the following section.
22
4. Range extension with almost blank sub-frames (ABS)
Common reference signals (CRS) interference 4.1
Common reference signals (CRS) are transmitted in every downlink subframe and in every resource block
in the frequency domain, so they cover the entire cell bandwidth. CRS can be used by the terminal for
channel estimation for coherent demodulation of downlink physical channels [2].
They can also be used by the terminal to acquire channel state information (CSI) which is used as the
basis for cell selection and handover decisions.
Figure 12: Structure of CRS within a pair of resource blocks
As shown in Figure 12, the structure of a single cell-specific reference signal consists of reference
symbols of predefined values inserted within the first and third last OFDM symbol of each slot, so within
each resource block pair there are 8 reference symbols, also the number of different reference signals in a
cell corresponds to the number of antenna ports available in the cell.
CRS is considered as the most important cause of interference in ABS as CRS exists in every resource
block as shown in Figure 12. CRS can be eliminated with different strategies that are explained in [16]:
- Using Multicast-Broadcast Single Frequency Network Subframe: which is a specific subframe where
CRS is not transmitted in the data part but is still transmitted in the control part.
- Interference cancellation of CRS from Macro-eNB cells: Using techniques to cancel the CRS effect
such as successive interference cancellation.
- Puncturing of resource elements in which Macro-eNB transmits CRS: which means not considering
the resource elements where CRS is present.
Throughout the rest of the report we will consider perfect CRS interference cancellation and we will
focus on optimizing the ABS ratio.
23
Proposed formula to calculate the ABS ratio to maximize the performance. 4.2
In this section we will deduce a closed form expression for the ABS (Almost Blank Subframes) allocation
percentage or ratio14
that maximizes the performance of the network in terms of cell-edge users capacity.
As was stated before the ABS configuration is communicated between the nodes using a 40 subframes
pattern, so by optimizing the ABS ratio we mean optimizing the number of subframes that are considered
as ABS in this pattern.
In the following example a round robin scheduler is considered where Macro-eNB users and center Pico-
eNB users are only allowed to be scheduled in the non-ABS while the range extension Pico-eNB users are
only allowed to be scheduled in the ABS. The constraint on the center Pico-eNB users is introduced for
simplicity and to allow the range extension users some fairness in using the ABS because in reality ABS
are shared between center and range extension Pico-eNB users and it becomes harder to determine which
users are scheduled in the ABS. First we start by an introduction about round robin scheduler and why it
is used in this example.
Round robin is a simple scheduling method that is based on assigning the resources to the terminals in
turn, one after another, which means that all the users have equal chances to be scheduled without
considering their CQI (channel quality indicator) which is explained in the flow chart in Figure 13.
Figure 13: Flow chart explaining the round robin scheduler
14
By this we mean the number of subframes that are used as ABS out of the total number of subframes in the pattern, so if we use 10 subframes out of 40 as ABS the ratio would be 0.25.
24
The reason for using round robin scheduling is its simplicity and that it is very convenient to use in a
theoretical example to ensure that all the users have the same chance of being scheduled and then
comparing users in terms of capacity and throughput for instance.
The rest of this section will be divided into 3 parts; the first one is a general model that is used to deduce a
general formula for calculating which is the ABS ratio, and the second part consists of simulations that
validate the theoretical results and finally an example with a specific setup of the model in the first part,
which means specifying the path loss model, transmitting power and position for each node, which is also
used to validate the results.
4.2.1 General model
Considering a simple setup having a 1 cell network with the following features:
a. This cell contains 1 Macro-eNB and a certain number Npico of Pico-eNBs. The Pico-eNBs are
randomly distributed in the cell.
b. The users are randomly distributed throughout the cell area.
c. All Pico-eNBs have the same number of users in the range extension area.
d. Round robin scheduler is used as explained in the previous section
If we consider a channel model15
that is only impaired by additive white Gaussian noise (AWGN) and
interference, then the ith user capacity
16 will be according to the following equation
(1)
where hi is the channel gain, SINRi is the signal to interference and noise ratio and BW is the bandwidth
which is considered to be 1 Hertz through the whole example for simplicity, also the number of subframes
is assumed to be 1. The following notation will be used in the deduction.
Macro-eNB transmission power P1
Pico-eNB transmission power P2
Channel gain from Macro-eNB to the ith user (hm_ue)i
Channel gain from the kth Pico-eNB to the i
th user (hp_ue)k,i
number of ues per Macro-eNB Nm
number of Pico-eNBs Npico
number of center Pico-eNB ues per Pico-eNB Np_c
number of range extension ues per Pico-eNB Np_re
Almost blank subrames ratio (Alpha)
the noise in the system N0
Table 1
15
Here every user has a different channel to each node (Picos and Macro) so each user has a vector (Npico+1) long
of channels that are only impaired by AWGN and interference. 16
Channel capacity is defined as being the tighter upper bound of the amount of information that can be transmitted
over a communication channel.
25
As explained before, cell selection is based on the downlink reference signal power measurements so the
users attached to the Macro-eNB (Nm) have a higher downlink power coming from the Macro-eNB than
the Pico-eNBs, While center Pico-eNB users (Np_c) receive the reference signals from the Pico-eNB with
a higher power than the signals coming from the Macro-eNB. Finally for the range extension Pico-eNB
users (Np_re), although they receive the reference signals from the Macro-eNB with a higher power but
due to the range extension offset, that was explained before, these users are attached to the Pico-eNB.
So using the above notation the capacity for the users attached to the different nodes can be formulated as
follows starting by the ith Macro-eNB user capacity in equation (2).
(2)
. (3)
Then the capacity of the ith center Pico-eNB user attached to the k
th Pico-eNB
(4)
. (5)
And finally the ith range extension Pico-eNB user attached to the k
th Pico-eNB
(6)
. (7)
We can plot the users capacity in equations (2), (4) and (6) as a function of , so by choosing one user
from each group (Macro, center Pico and range extension Pico) and specifying values for the different
parameters (channel gains, P1, P2, Nm, Np_c and Np_re) we get the plot in Figure 14.
26
Figure 14: Plot of the capacity of Macro-eNB, center Pico-eNB and range extension users against
So in order to maximize the cell edge users capacity
17 we need to find the intersection point between the
lowest range extension capacity line, corresponding to the range extension user having the lowest
capacity, and the first line it intersects with which is the lowest Macro-eNB or center Pico-eNB user
capacity line, corresponding to the Macro-eNB or center Pico-eNB user having the lowest capacity.
So we can define the intersection point, which is basically found by a search over , using the following
criterion:
{ } (8)
In this case we will not consider the center Pico-eNB capacity line, so we will only focus on the range
extension and Macro-eNB users as in reality center Pico-eNB users are not affected by the ABS ratio, but
here we assume that center Pico-eNB users are only allowed to transmit during non-ABS to make the
scheduler simpler and giving the Macro-eNB user and Pico-eNB range extension user an equal chance to
be scheduled.
We will denote the Macro-eNB user having the lowest capacity by user m having the following
capacity
(9)
. (10)
17
In this model we maximize the worst user (0% worst user) capacity instead of the cell edge users (5% worst users)
capacity for simplicity.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
10
20
30
40
50
60
alpha
capacity
macro user capacity
center pico user capacity
range extension user capacity
Cap
acit
y (b
its/
sec)
27
We will denote the range extension user having the lowest capacity by user n and assuming that this
user belongs to the kth Pico-eNB with the following capacity
(11)
. (12)
The intersection point can be acquired analytically by equating equations (9) and (11) in order to find the
optimum alpha that maximizes the cell edge capacity as follows
(13)
And by reordering the previous equation we get the following equation which can be considered as the
optimal value of in order to optimize the 0% worst user throughput.
(14)
Since the mth
Macro-eNB user capacity is given by eq. (9) so considering that only this Macro-eNB user
gets all the resources all the time then the capacity would be given by the following expression, i.e.
putting the number of users to 1 in eq. (1)18
.
. (15)
Which we can call the maximum Macro-eNB user capacity, so is the same as but
only assuming that the Macro-eNB is only serving this user m, this is why it is called
because this is the maximum capacity that this user can reach. And doing the same for the nth range
extension Pico-eNB user
(16)
Then can be expressed as
. (17)
From this equation we can clearly see that alpha depends on 2 factors:
1. The ratio between the number of Macro-eNB ues to the number of range extension ues per Pico-eNB.
2. The ratio between the maximum capacity of a range extension user and the maximum
capacity of a Macro-eNB user .
18
This is exactly as if we have only one Macro-eNB user so this user will use the available resources (subframes) all
the time.
28
Focusing on the second factor and trying to simplify it, starting with the maximum Macro-eNB user
capacity
(18)
Since the noise value is very small we can neglect it also assuming the value of P1 to be very large so
(
P1) is much bigger than the term in the denominator then we can approximate the previous
equation to
. (19)
Normally most users attached to the Macro-eNB are placed close to it, although some Macro-eNB users
are placed very close to the Pico-eNB due to the high transmission power of the Macro-eNB but we will
consider only the users closer to the Macro-eNB, who are the majority, and assuming that the interference
to these users is dominated by one or at most two Pico-eNBs while the rest cause negligible interference.
Under this assumption we can approximate the interference term
with a constant (I)
since it is assumed to be independent on the number of Pico-eNBs and is dominated by the interference
caused by the closest 1 or 2 interferer Pico-eNBs.
) . (20)
Since is assumed to be independent on Npico so it can be considered as a constant and can be
denoted by C1.
Now focusing on the second term which is .
. (21)
Inserting the SINR3 expression
(22)
Assuming that we have a very large Npico then N0 can be neglected, considering that P2 0, and the
interference term in the denominator would be larger than the numerator so the previous equation can be
approximated to
(23)
where k is the serving Pico-eNB for the range extension user.
29
Since the Pico-eNBs are distributed randomly in the cell so and can be considered
as independent and identically distributed (IID) random variables. Also since we are trying to optimize
the capacity and we are assuming a large Npico so optimizing would be the same as optimizing its
expected value so we can replace by as follows
. (24)
Since all the values of can be considered as independent identically distributed (IID) random
variables having the same mean value and can be expressed as
19
. (25)
can be considered as a constant value so
( )
(26)
and finally the term
can be considered as a constant and can be denoted by C2 and since
Npico is assumed very large so and can be expressed as
. (27)
Finally
. (28)
So can be expressed as
(29)
where Nre*Npico is equal to the total number of range extension users which can be denoted by Nre_total..
Finally is expressed by
. (30)
19
It is known that
but we will use this approximation anyway to simplify the problem. Also the
variance of the values has been found to be very small, in the order of 10-14
, which verifies the approximations
done in this equation.
30
So if the values of C2 and C1 are assumed to be approximately equal, which will be shown in the following
sections, then we can introduce which is considered, according to simulations, to be the optimized
value that gives the optimal or suboptimal value of and is expressed by:
. (31)
This means that the ABS ratio is proportional to the ratio between the number of users attached to the Macro-eNB and the total number of range extension users attached to the Pico-eNBs.
4.2.2 Simulations validating the previous results
In this section a small MATLAB system simulator that performs Monte Carlo20
simulations [11] will be
introduced to verify the results in the previous section specifically equations (29) and (31) as they are
considered the most important results in the deduction. The simulations consist of a 1 cell network with a
Macro cell at a predefined position and a specific number of Pico-eNBs and users are dropped randomly
throughout the cell area.
The path loss is calculated according to 2 models, the ITU channel model and the Spatial Channel Model
(SCM) which will be explained in details in the following.
- ITU channel model: we will use the urban Macro-eNB (UMa), for Macro-eNB users, and urban
micro (UMi), for Pico-eNB users, models in [6].
Assuming that all users have line of sight to the serving base station so the path loss in dB for Macro-eNB
users will be calculated according to
for d < 160 m (32)
for d > 160 m (33)
where d is the distance between the user and the node, hBS = 24m, hUT = 0.5 m and fc=1 GHz.
And for Pico-eNB users the path loss is given as
for d < 120 m (34)
for d > 120 m (35)
where d is the distance between the user and the node, hBS = 9m, hUT = 0.5 m and fc=1 GHz.
20
Monte Carlo method is a class of computational algorithms that depends on repeated random sampling to compute
its results which in our case means to drop the users and Pico-eNBs repeatedly and in a random way to compute the
end result.
31
- Spatial channel model: This model will be calculated according to the equations in [10] and
assuming no line of sight for both Macro-eNB and Pico-eNBs.
For the Macro-eNB users the path loss in dB is given by
10 10
10 10
[ ] 44.9 6.55log log ( ) 45.51000
35.46 1.1 log ( ) 13.82log ( ) 0.7
bs
ms c bs ms
dPL dB h
h f h h C
(36)
where hbs is the base station antenna height in meters, hms is the MS antenna height in meters, fc the carrier
frequency in MHz, d is the distance between the BS and the user in meters, and C is a constant which is
equal to 3dB for urban Macro-eNB. These parameters are set to hbs = 32m, hms = 1.5m and fc=1900MHz.
And the path loss for Pico-eNB users is given by
PL = -55.9 + 38*log10(d) + (24.5 + 1.5*fc/925)*log10(fc) (37)
where fc = 1900 MHz.
The idea is to use the Monte Carlo method to compare the optimum alpha, given by equation (14), with
the deduced alpha in (29) and (31). In order to do that, an average of 100 drops21
, with a random
realization for the positioning of the Pico-eNBs and users for each drop, will be used to calculate an
average value of alpha and this process will be repeated 500 times so that we will have 500 calculated
alpha for each equation at the end then we compare the results.
4.2.2.1 Validating the alpha expression:
In this section we will validate the expression given by equation (29), the idea is to calculate the value
of according to equations (29) and the optimum value of according to equation (14), this process will
be iterated 500 times, as explained before, so at the end we will have 2 vectors of , each consisting of
500 values, that we can compare and if the values in both vectors are approximately equal, then equation
(29) can be validated to give an optimal value for .
Since in the deduction we assume having a large number of Pico-eNBs, we will drop 100 Pico-eNBs and
200 users randomly and alpha will be calculated according to equations (14) and (29) and both values will
be compared, listed in Table 2 are the parameters used in this simulation.
Cell area22
50m x 50m
Macro-eNB position X:0 Y:25
Pico-eNBs positions Random but keeping a minimum distance of 10 m
from the Macro-eNB.
Users positions Random
Macro-eNB transmitting power 40 W
Pico-eNB transmitting power 1 W
Number of drops 100
Table 2
21
A drop is defined as one simulation run over a certain time period. 22
The reason for having a very small cell area is to decrease the distance between the Pico-eNB-eNBs to increase
the interference between them to fulfill the assumption of having a very big number of Pico-eNB-eNBs in the cell.
32
The only problem in equation (29) is that the value of in C2 is not known; also the
assumption that all the Pico-eNBs have the same number of range extension users is not present in this
simulation so we will go back 1 step to equation (25) which was given by
23
(38)
and will consider to be the constant and will be denoted by so the value of alpha will be
given by
(39)
4.2.2.1.1 ITU channel model
We start by the ITU channel model. Figure 15 represents the PDFs of the 500 alpha values calculated
from equations (14) and (39). And it shows that the PDFs are concentrated at very close values.
Figure 15: PDFs of Alpha according to eq (14) (Blue) and Alpha according to eq (39) (Green)
Figure 16 represents a plot of the alpha value in both cases for 500 iterations; each iteration is an average
of 100 drops. If we compare both values at any of the 500 measurements we will see that the difference
between them is always less than 0.1 which means that the value of alpha calculated in equation (39)
gives the optimal or the suboptimal value of the ABS ratio24
. It can be seen from these results that the
result from equation (39) can be validated to give the optimal or suboptimal ABS ratio for the ITU
channel model.
24 It will be shown in the simulations section that if the formula gives a solution that is 0.1 less or more than the
optimal one this solution is the suboptimal one, which means that it is the second best solution, and is very close to
the optimal solution.
0.65 0.7 0.75 0.8 0.850
20
40
60
80
100
120
Alpha
Optimal Alpha according to eq (14)
Alpha according to eq (29)
33
Figure 16: Plot of the 100 values of Alpha according to eq (14) (Blue) and Alpha according to eq (40) (Green)
4.2.2.1.2 Spatial channel model
In this part we will repeat the previous simulation but using the Spatial Channel Model instead of the ITU
channel model. Figure 17 represents the PDFs of the results from equations (14) and (39). And it shows
that the pdfs are concentrated at very close values.
Figure 17: PDFs of Alpha according to eq (14) (Blue) and Alpha according to eq (39) (Green)
Figure 18 represents the plot of the alpha value in both cases for 500 iterations; each iteration is an
average of 100 drops. If we compare both values at any of the 500 measurements we will see that the
difference between them is always less than 0.1 which means that the value of alpha calculated in
equation (39) gives the optimal or the suboptimal ratio of ABS. It can be seen from these results that the
result from equation (39) can be validated to give the optimal or suboptimal ABS ratio for the spatial
channel model.
0 50 100 150 200 250 300 350 400 450 500
0.65
0.7
0.75
0.8
0.85
Itteration number
Alp
ha
Optimal Alpha according to eq (14)
Alpha according to eq (31)
0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.750
20
40
60
80
100
120
140
Alpha
Optimal Alpha according to eq (14)
Alpha according to eq (29)
34
Figure 18: Plot of the 100 values of Alpha according to eq (14) (Blue) and Alpha according to eq (40) (Green)
In this section equation (39) which is the same as equation (29) has shown to be giving results very close
to those of equation (14) which, in turn, shows that equation (29) gives the optimal ABS ratio in terms of
cell edge users throughput in the case of the ITU channel model and spatial channel model. It is also
worth noting that the difference between the alpha values according to equation (29) and equation (14) is
higher in the case of Spatial Channel Model compared to the ITU channel model and this is due to the fact
that the path loss in the case of SCM is lower than in the case of ITU channel model, which means that
the interference in the ITU case is higher, so by putting the values of the channel gains according to SCM
in equation (14) we get a larger value of alpha.
4.2.2.2 Validating the alpha expression:
In this section we will validate expression given by equation (31), the idea is to calculate the value of
according to equations (31) and the optimum value of according to equation (14), this process will be
iterated 500 times, as explained before, so at the end we will have 2 vectors of , each with 500 values,
that we can compare and if the values in both vectors are close enough then equation (31) can be
validated to give an optimal value for .
For this part we use a more realistic example where we drop 6 Pico-eNBs placed randomly in the cell, in
addition 200 users are dropped randomly throughout the cell area. The simulation parameters are listed in
Table 3.
Cell area 500m x 500m
Macro-eNB position X:0 Y:250
Pico-eNBs positions Random but keeping a minimum distance of 70
m from the Macro-eNB and the other Pico-eNBs.
Users positions Random
Macro-eNB transmitting power 40 W
Pico-eNB transmitting power 1 W
Number of drops 100
Table 3
0 50 100 150 200 250 300 350 400 450 5000.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
Itteration number
Alp
ha
Optimal Alpha according to eq (14)
Alpha according to eq (31)
35
4.2.2.2.1 ITU channel model
We start by the ITU channel model. Figure 19 represents the PDFs of the results from both equations.
And it shows that the PDFs are almost coinciding.
Figure 19: PDFs of Alpha according to eq (14) (Blue) and Alpha according to eq (31) (Red)
Figure 20 represents the plot of the alpha value in both cases for 500 iterations, each iteration is a 100
drops.
Figure 20: Plot of the 100 values of Alpha according to eq (14) (Blue) and Alpha according to eq (31) (Red)
0.5 0.52 0.54 0.56 0.58 0.6 0.62 0.64 0.66 0.680
20
40
60
80
100
120
140
Alpha
Optimal Alpha according to eq (14)
Alpha according to eq (31)
0 50 100 150 200 250 300 350 400 450 5000.48
0.5
0.52
0.54
0.56
0.58
0.6
0.62
0.64
0.66
0.68
Itteration number
Alp
ha
Optimal Alpha according to eq (14)
Alpha according to eq (31)
36
These results show that the result in equation (31) is very close to the optimum value given by (14) when
using the ITU channel model therefore it can be validated.
4.2.2.2.2 Spatial Channel Model (SCM)
In this part we will repeat the previous simulation but using the spatial channel model instead of the ITU
channel model. Figure 21 represents the PDFs of the results from equations (14) and (31). And it shows
that the PDFs are concentrated at very close values.
Figure 21: PDFs of Alpha according to eq (14) (Blue) and Alpha according to eq (31) (Red)
Figure 22 represents the plot of the alpha value in both cases for 500 iterations, each iteration is an
average of 100 drop and as seen the values resulting of both equations are very close.
Figure 22: Plot of the 100 values of Alpha according to eq (14) (Blue) and Alpha according to eq (31) (Red)
0.58 0.6 0.62 0.64 0.66 0.68 0.7 0.72 0.740
20
40
60
80
100
120
Alpha
Optimal Alpha according to eq (14)
Alpha according to eq (31)
0 50 100 150 200 250 300 350 400 450 5000.54
0.56
0.58
0.6
0.62
0.64
0.66
0.68
0.7
0.72
0.74
Itteration number
Alp
ha
Optimal Alpha according to eq (14)
Alpha according to eq (31)
37
These results show that the result in equation (31) is very close to the optimum value given by (14) when
using the spatial channel model therefore it can be validated. Same as the previous case, the difference
between the alpha values according to equation (31) and equation (14) is higher in the case of Spatial
Channel Model compared to the ITU channel model and this is due to the fact that the path loss in the
case of SCM is lower than in the case of ITU channel model, which means that the interference in the
ITU case is higher, so by putting the values of the channel gains according to SCM in equation (14) we
get a larger value of alpha. To summarize, it has been shown that the theoretical deductions in equations
(29) and (31) can be validated to give optimal or suboptimal results for the ABS ratio using Monte-Carlo
simulations. In the following section an example that is a special case of the general model used in the
deduction will be introduced to elaborate more on the theoretical results.
4.2.3 Example to validate the general model results
Through this example we will apply the previous theoretical model into a more practical scenario using
specific path loss distributions, Pico-eNB distributions and nodes transmitting powers.
4.2.3.1 Defining the network topology and new parameters used in this example
In this subsection the network topology and different parameters used in the example are stated.
a) We use a 1 cell network which contains 1 Macro-eNB and 2 groups of Pico-eNBs where group 1 are
the Pico-eNBs closer to the Macro-eNB and group 2 are the Pico-eNBs farther from the Macro-eNB
and we will start the example by 4 Pico-eNBs, 2 in each group, as shown in Figure 23 and then we
will generalize the model for any number of Pico-eNBs (Npico).
b) Same as the general model we will consider only the Macro-eNB user that has the lowest capacity
and the range extension Pico-eNB user that has the lowest capacity.
c) We will assume, for simplicity, that the Pico user has the same path loss from all the other Pico-
eNBs. This can be the case when we have only 4 Pico-eNBs as they are equidistant, see Figure 23,
but we will assume that this can be extended to any number of Pico-eNBs which is a strong
assumption but it can be motivated due to the fact that we are not considering inter-cell interference in
this example but in reality if we have a large number of Pico-eNBs (Npico), as we will assume later,
then the Pico user placed at the cell border suffers from a larger inter-cell interference than the Pico-
eNB user placed in the cell center for instance, in that sense we can assume a close interference value
for all the Pico-eNB users.
d)
path loss from group 1 Pico-eNBs to Macro-eNB user hp1_ue
path loss from group 2 Pico-eNBs to Macro-eNB user hp2_ue
Table 4
38
e) Transmission powers for the different nodes.
P1 40 W
P2 1 W
Ptotal_Pico 4 W
Table 5
f) Defining the distances between the different nodes.
Distance between Macro-eNB and Macro-eNB user 40m
Distance between Pico-eNB and center Pico-eNB user 10m
Distance between Pico-eNB and range extension Pico-eNB user 20m
Distance between Macro-eNB and group 1 Pico-eNBs center user 110m
Distance between Macro-eNB and group 2 Pico-eNBs center user 150m
Distance between Pico-eNBs and other Pico-eNBs center user 40m
Distance between Pico-eNBs and other Pico-eNBs range extension user 30m
Distance between group 1 Pico-eNBs and Macro-eNB user 80m
Distance between group 2 Pico-eNBs and Macro-eNB user 120m
Table 6
Figure 23: Macro-eNB and Pico-eNBs in a cell
4.2.3.2 Calculating the values of C1 and C2 according to the example.
The path loss is calculated according to the urban Macro-eNB (UMa), for Macro-eNB users, and urban
micro (UMi), for Pico-eNB users, which belong to the ITU channel model in [6] and they were explained
in details in 4.2.2.
After defining the different parameters for this example we calculate the values of C1 and C2 25
according
to this example to find a closed formula for . We start by C1 which is given by
25
C1 and C2 are the same as the ones deduced in section 4.2.1 but adapted to the scenario of the example.
39
). (40)
Putting the value of the interference according to the specifications of the example, the
becomes
. (41)
And since we have 2 groups of Pico-eNBs as shown in Figure 23 we can assume that the number of Pico-
eNBs increases to form 2 circles for group 1 and 2 to maintain the distance from the Macro-eNB as
shown in Figure 2426
.
Figure 24: Increasing the number of picos in the cell
So the previous equation can be rewritten as follows
(
) . (42)
Assuming that there is a specific constant budget for the total power transmitted by all the Pico-eNBs
which can is denoted by so this equation can be re-written as follows
(
) . (43)
Now considering C2 which was equal to
, but since and have specific
values in this example then they are no longer random variables and C2 can be expressed as
26
The reason for this distribution of Pico-eNBs is to simplify the equations by having only 2 channel gain values (one for the first group of Pico-eNBs and the other for the second group of Pico-eNBs).
40
. (44)
Now calculating the values of C1 and C2 according to the path loss and transmitting powers stated above.
C1 = 8.7025 C2 =8.6194
So
So this example shows that the optimum value of can be expressed according to equation (45).
(45)
4.2.4 Example to validate the general model results (without the assumption of Ptotal_Pico)
In this example the previous example is repeated but without the constraint of Ptotal_pico, so we try to
generalize the validation of the result for by removing the assumption that we have a budget for the
Pico-eNBs transmitting power so we go back to equation [19] and rewrite it according to our example as
follows
(
) . (46)
And introducing the values for the power and path loss stated before.
. (47)
Also for the range extension user maximum capacity given by
( ) . (48)
It is simplified to
( ) . (49)
So the ratio
can be expressed in terms of as follows
( ) . (50)
In order to understand this expression we plot it against Np and plot with it
for comparison.
41
Figure 25: Comparison between
and
From this figure we see that
is very close to
even for small Npico, so this validates the
expression for that we deduced before which is given by eq. (30).
This result shows that is dependent on the total number of range extension users in a cell. But logically it should be dependent on the number of range extension users per Pico-eNB instead of the total number
since the resources are reused for each Pico-eNB but this can be explained in the next figure where we
plot the maximum range extension user capacity and also the Macro-eNB user maximum capacity against
the number of Pico-eNBs, which means that we plot the user capacity while changing the number of Pico-
eNBs in the cell and see how the capacity behaves.
Figure 26: Plot of the range extension user capacity against the number of picos.
0 10 20 30 40 50 60 70 80 90 1000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
number of picos (Np)
C-re-max/C-macro-max
1/Np
2 3 4 5 6 7 80
20
40
60
80
100
120
140
number of picos (Np)
capacity
range extension user capacity against the numner of picos
Cap
acit
y (b
its/
sec)
Range extension user capacity against the number of picos
42
Figure 27: Plot of the Macro user maximum capacity against the number of picos.
From Figure 26 we see that the range extension user capacity decreases when increasing the number of
Pico-eNBs in the cell. This means that the range extension user capacity is interference limited and that
the capacity depends very much on the interference coming from other Pico-eNBs which in turn depends
on the number of Pico-eNBs. Also from Figure 27 it is obvious that the Macro-eNB user capacity is
almost not affected by the number of Pico-eNBs or in other words the interference caused by the Pico-
eNBs to the Macro-eNB users is not significant.
So this explains the dependence of the alpha calculations on the number of Pico-eNBs or more generally
the total number of range extension users in the cell as will be shown in the next section.
Now we validate the result by trying different user distributions for the same Pico-eNBs distribution
(4 Pico-eNBs) and compare the alpha we get by simulation and that we get using equation (31). As
shown in the following example:
Considering case1, for example, we have 36 Macro-eNB users, 4 range extension users and 10 center
Pico-eNB users.
2 3 4 5 6 7 80
50
100
150
200
250
number of picos (Np)
capacity
macro user maximum capacity against the number of picos
Cap
acit
y (b
its/
sec)
43
Figure 28
The optimized alpha according to simulations, see the intersection point in the figure, is 0.085 while
calculating the alpha value according to the formula gives 0.1. ( =1/ (1+ (36/4)) =0.1). The rest of the
results are listed in the following table
Nrof_Macro-eNB_users Nrof_re_users/Pico-eNB Sim_alpha Calculated_alpha( )
36 1 0.085 0.1
32 2 0.176 0.2
28 3 0.27 0.3
24 4 0.363 0.4
20 5 0.462 0.5
16 6 0.563 0.6
12 7 0.671 0.7
8 8 0.78 0.8
4 9 0.887 0.9
Table 7
As seen from Table 7 the simulation results for are very close to the value of calculated from (31).
4.3 Summary
As a conclusion, from the last subsection, the value of from equation (31) is applicable in
interference limited situations, i.e. situations where Pico-eNBs are causing interference to each
other. Through section 4 a closed form expression for has been deduced and it has been tested
to be valid in the case of the ITU channel model and the Spatial Channel Model (SCM), but it
might not be the best solution in cases where there is no interference between Pico-eNBs.
The equation in (31) will be tested more in the next section where simulations are conducted using
more realistic channel models and bigger networks.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
5
10
15
20
25
30
35
40
45
alpha
capacity
macro user capacity
center pico user capacity
range extension user capacity
Cap
acit
y (b
its/
sec)
44
5. System simulation results
This section is for the simulation results, we start by introducing the simulator used in this section which
is the Raptor simulator, then listing the assumed simulation parameters used and finally presenting the
simulations in different subsections.
The Raptor simulator 5.1
All the simulations in this project are performed using a simulator called Raptor which is a property of
Ericsson. Raptor is an LTE-Advanced system simulator which means that it performs physical layer
simulations.
The simulator is divided into 3 parts:
1) Input parameter files: MATLAB files containing all the simulation parameters that will be used as
input to the simulator.
2) Main simulator: the main body of the simulator which is developed in C++ and this simulator
generates MATLAB result files.
3) Graphical interface: MATLAB graphical interface that processes the MATLAB result files to
illustrate the results in the form of CDFs, bar charts and scatter plots as will be shown in the next
section.
I contributed mainly in creating my own input parameter files and optimizing the graphical interface to
show more illustrative plots.
45
System simulation assumptions 5.2
The criterion that we focus on optimizing is the cell-edge users throughput
27 while keeping a fair level of
average throughput.
The simulation assumptions are listed in the following table28
:
Parameter Description
Network topology 21 cell network (i.e. 7 three-sector sites)
Number of ues 30 ues per cell
Number of Pico-eNBs From 2 to 10 per cell depending on the tested scenario and all the Pico-
eNBs are outdoors and located at predefined locations.
Deployments Configuration 129
and 4b30
[7]
Traffic model Full buffer 31
Range extension offset From 0 to 18 dB depending on the scenario
Downlink scheduling Proportional fair scheduler [2]
Carrier frequency 2 GHz
Path loss mode ITU Channel Model and Spatial Channel Model (SCM)
Downlink link adaptation Ideal link adaptation32
CRS interference modeling Assuming perfect CRS interference cancellation.
Total bandwidth 20 MHz
Antenna tilting According to TR36.819- 12 degrees for Macro-eNB, 0 degrees for Pico-
eNB
Table 8
Simulation results 5.3
5.3.1 Who wins and who loses in terms of throughput in a heterogeneous network
deployment?
Who are the winners and losers in terms of throughput in a Macro-Pico heterogeneous deployment is a
very crucial question, we mean by winners or losers the users who experience an increase or decrease of
throughput when adding the Pico layer to the Macro layer. To answer this question we will compare the
following 2 network deployments:
1) Macro-eNB only deployment: we only have 1 Macro-eNB per cell. 2) Macro-eNB + Pico-eNB deployment: we have 1 Macro-eNB and 4 Pico-eNBs per cell, with no range
extension applied to the Pico-eNBs.
27
Cell edge users