UPTEC F 19056
Examensarbete 30 hp
Oktober 2019
Energy Consumption Optimizations
for 5G networks
Teknisk- naturvetenskaplig fakultet UTH-enheten Besöksadress: Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0 Postadress: Box 536 751 21 Uppsala Telefon: 018 – 471 30 03 Telefax: 018 – 471 30 00 Hemsida: http://www.teknat.uu.se/student
Abstract
Energy Consumption Optimizations for 5G networks
Martina Tran
The importance of energy efficiency has grown alongside awareness of climate change due to the rapid increase of greenhouse gases. With the increasing trend regarding mobile subscribers, it is necessary to prevent an expansion of energy consumption via mobile networks.
In this thesis, the energy optimization of the new radio access technology called 5G NR utilizing different sleep states to put base stations to sleep when they are not transmitting data is discussed. Energy savings and file latency with heterogeneous and super dense urban scenarios was evaluated through simulations with different network deployments.
An updated power model has been proposed and the sensitivity of the new power model was analyzed by adjusting wake-up time and sleep factors. This showed that careful implementation is necessary when adjusting these parameter settings, although in most cases it did not change the end results by much.
Since 5G NR has more potential in energy optimization compared to the previous generation mobile network 4G LTE, up to 4 sleep states was implemented on the NR base stations and one idle mode on LTE base stations. To mitigate unnecessary sleep, deactivation timers are used which decides when to put base stations to sleep. Without deactivation timers, the delay could increase significantly, while with deactivation timers the delay increase would only be a few percent.
Up to 42.5% energy could be saved with LTE-NR non-standalone deployment and 72.7% energy with NR standalone deployment compared to LTE standalone deployment, while minimally impacting the delay on file by 1%.
ISSN: 1401-5757, UPTEC F 19056
Sammanfattning
Med tanke p˚a den ¨okande medvetenheten om klimatet och den stigande koldioxidhalten, vill man minska energianv¨andandet. Ju mer energi som anv¨ands, desto s¨amre ¨ar det f¨or klimatet d˚a skapandet av energi via f¨orbr¨anning av fossil leder till ¨okad koldioxidhalt. Dessutom ¨okar antalet mobilanv¨andare f¨or varje ˚ar och d¨arf¨or ¨ar det viktigt att inte energianv¨andandet p˚a grund av det forts¨atter ¨oka.
Genom att anv¨anda sig av ”heterogeneous deployment” och en sovalgorithm kan energi sparas och samtidigt p˚averka anv¨andare minimalt. Tidigare arbeten har gjorts p˚a detta, d˚a hur mycket energi man kan spara in som mest unders¨oktes, utan att ta h¨ansyn till hur anv¨andare p˚averkas. I detta arbete har energioptimisering av NR med olika djupa s¨omntillst˚and implementerats p˚a basstationer. Basstationer kan s¨attas i s¨omn n¨ar det inte i finns n˚agra aktiva anv¨andare i.e. anv¨andare som vill ladda ner filer. Utv¨arderingar gjordes p˚a energianv¨andandet och filf¨ordr¨ojningar via simulationer i stadsmilj¨oer med olika mobiln¨at.
Tack vare ”the ultra-lean design” kan 5G NR sova djupare och l¨angre j¨amf¨ort med det tidigare mobiln¨atet 4G LTE. Den implementerade sovalgoritmen anv¨ander sig av 4 s¨omnstadier m¨ojliga p˚a NR basstationer och ett vilol¨age p˚a LTE basstationer. En viktig faktor n¨ar man anv¨ander sig av sovalgoritmen ¨ar att veta n¨ar det ¨ar som mest optimalt att s¨atta basstationer i s¨omn. F¨or tidig s¨omn p˚averkar anv¨andare negativt och f¨or sen s¨omn p˚averkar energibesparingar negativt. F¨or att minska onn¨odig s¨omn anv¨andes deaktiverinstider som best¨ammer n¨ar basstationer s¨atts i s¨omn och g¨or det gradvis. Deaktiveringstiderna ¨ar viktiga att anv¨anda och optimera beroende p˚a vad det ¨
ar f¨or trafik. De har visats minska effekten p˚a filf¨ord¨ojningen och samtidigt fortfarande maximerat energibesparingen.
En uppdaterad energimodell har redovisats eftersom den tidigare energimodellen antas vara f¨or optimistisk. Sensitiviteten av den nya modellen har analyserats genom att justera aktiveringstiden och energianv¨andandet f¨or varje s¨omnstadie. Analysen visar att f¨or de flesta fall f¨or¨andras inte resultatet mycket, men det ¨ar fortfarande viktigt att t¨anka p˚a hur parametrarna best¨ams.
Acknowledgements
The master thesis job was done at Ericsson R&D in Kista, Stockholm.
Firstly, I want to give my biggest gratitude to my supervisor Richard Tano at Ericsson. Thank you for your continuous support and guidance working with this project. I learnt a lot and I could not have asked for a better supervisor. Your patience allowed me to do the best I could with the time I had at Ericsson.
A special thanks to Gustav Wikstr¨om and everyone else at the section who I had the pleasure to meet, and for making me feel welcomed. Thank you for the great fikas and the fun after work activities.
Additionally, I would like to thank P˚al Fregner for providing important knowledge about energy optimizations and for the comments during my presentation.
Contents
List of Figures vi
List of Tables viii
1 Introduction 1
1.1 Background . . . 2
1.1.1 Approaches to Energy Efficiency . . . 3
1.1.2 5G NR Network . . . 4 1.1.3 Thesis Focus . . . 5 1.1.4 Delimitation . . . 6 1.1.5 Previous Work . . . 7 2 Theory 8 2.1 Base station . . . 8 2.2 Heterogeneous deployments . . . 8
2.2.1 Macro cells and Small cells . . . 9
2.3 LTE and NR basics . . . 9
2.3.1 Orthogonal frequency division multiplexing, OFDM . . . 9
2.3.2 Time-domain and Frequency-domain structure . . . 10
2.3.3 Reference signals . . . 11
2.4 Dual Connectivity . . . 11
2.5 Duplex Schemes . . . 13
2.6 Scheduling . . . 13
2.7 File transfer protocol, FTP . . . 14
2.8 Power models . . . 15
2.8.1 Idle mode power consumption . . . 15
2.8.2 LTE base station power model . . . 16
2.8.3 NR base station power model . . . 17
2.9 Energy and performance metrics . . . 18
3 Methodology 19 3.1 Performance evaluation . . . 19
3.1.1 Round Trip Time, RTT . . . 19
3.1.2 Downlink user throughput . . . 19
3.2 Energy evaluation . . . 19
3.2.1 Daily traffic load . . . 19
3.2.2 Power per unit area . . . 20
3.2.3 Average daily power consumption . . . 20
3.3 Sleep mode algorithm . . . 20
3.3.1 Configurations of sleep states . . . 24
3.5 Scenario setup . . . 25
3.6 Method . . . 26
4 Simulation Results 27 4.1 Throughput vs Latency . . . 27
4.2 Different deactivation timers . . . 27
4.3 Performance Evaluations . . . 29
4.3.1 LTE-NR dual connectivity deployment . . . 29
4.3.2 NR standalone deployment . . . 30
4.3.3 10th and 50th percentiles . . . 32
4.3.4 90th percentile . . . 34
4.4 Energy Evaluation . . . 37
4.4.1 LTE-NR dual connectivity deployment . . . 37
4.4.2 NR standalone deployment . . . 40
4.4.3 Percentage spent in sleep states . . . 41
4.5 Sensitivity analysis . . . 43
5 Discussion 47 5.1 Deactivation timers’ impact . . . 47
5.2 Performance evaluation discussion . . . 47
5.2.1 Best case users and the median discussion . . . 48
5.2.2 Worst case users discussion . . . 49
5.3 Energy evaluation discussion . . . 50
5.4 Sensitivity analysis discussions . . . 51
6 Conclusion 53
7 Future work 55
List of Figures
1.1 The increase of global mobile connections (excluding IoT) between year
2014 and 2024 [3]. . . 2
1.2 A timeline of the mobile network evolution [5]. . . 2
1.3 The increase of base stations between the years 2007-2012 [8], [9]. . . 3
1.4 Figure shows from left to right, the power consumption in mobile networks, and the power consumption distribution in a base station [8]. . . 4
2.1 A heterogeneous network deployment with hexagonal cells, macro base stations, micro base stations and users. . . 9
2.2 OFDM subcarrier spacing ∆f [15]. . . 10
2.3 Time-domain structure of LTE and NR [15], [11]. . . 10
2.4 Frequency-domain structure of LTE and NR [15], [11]. . . 11
2.5 Example of dual connectivity. . . 12
2.6 Bearer concept of dual connectivity with LTE and NR [16]. . . 13
2.7 Duplex schemes in time-frequency structure for uplink and downlink [15]. 13 2.8 The congestion avoidance [18]. . . 15
2.9 The idle mode power consumption of LTE and NR [13]. . . 16
2.10 Power consumption dependency on relative RF output power in base sta-tion [19]. . . 17
2.11 The four defined sleep state modes with the sleep factors and total power consumption, P . . . 18
3.1 The average daily traffic profile in Europe [19]. . . 20
3.2 Flowchart of the activation mode. . . 22
3.3 Power model with a factor 2 decrease in power consumption for each 10-fold increase in wake-up time. . . 23
3.4 Flowchart of the sleep state algorithm with the four sleep modes. . . 24
3.5 The process of obtaining the simulation results. . . 26
4.1 The performance curves of 10 Mb file download without deactivation timers. 27 4.2 Average delay of 10 Mb file download with optimized deactivation times and without deactivation times. . . 28
4.3 The average delay of multiple 1 Mb FTP file downloads. . . 29
4.4 The average delay of multiple small file transmissions with deactivation timers from Table 4.2. . . 30
4.5 The average delay with fixed SS configuration on macro layer with varying SS configurations on micro layer. . . 31
4.6 The average delay where the micro layer cells have fixed SS3. . . 32
4.8 The 10th and 50th percentile delay curves of LTE-NR deployment with multiple 1 Mb file downloads. . . 33 4.9 The 10th and 50th percentile delay with fixed SS configuration on macro
layer with varying SS configurations on micro layer. . . 34 4.10 The 90th percentile delay curves of LTE-NR deployment with 10 Mb file
downloads. . . 35 4.11 The 90th percentile delay curves of LTE-NR deployment with 1 Mb file
downloads. . . 35 4.12 The 95th percentile delay of multiple 1 Mb file downloads with optimized
deactivation timers. . . 36 4.13 The 90th percentile delay with fixed SS configuration on macro layer with
varying SS configurations on micro layer. . . 37 4.14 The total energy consumption of the scenarios in Figure 4.2. . . 38 4.15 The daily energy consumption curves of the scenarios in Figure 4.14. . . . 38 4.16 Zoomed in version of Figure 4.15 with the three scenarios with the lowest
energy usage. . . 39 4.17 Total energy consumption of multiple small file transmissions with DA1
and DA2. . . 39 4.18 The total energy consumption with fixed SS3 on micro layer. . . 40 4.19 The total energy consumption with varying SS configurations on micro layer. 40 4.20 10 Mb file transmission with LTE-NR deployment. . . 41 4.21 NR standalone deployment with fix SS4 on macro layer cells. . . 42 4.22 Average state percentage of cells for small file transmissions with baseline
DA and DA2. . . 42 4.23 Average state percentage of cells for small file transmissions with DA1 and
DA2. . . 43 4.24 Sensitivity analysis of delay when adjusting wake-up time. . . 44 4.25 Sensitivity analysis of total energy consumption with A=5 and A=15. . . 44 4.26 Sensitivity analysis of the total energy consumption with different sleep
List of Tables
2.1 Example of a 3-sector LTE macro 2x2 LTE power model parameters. . . . 17
2.2 Example of a 3-sector NR macro 2x2 NR power model parameters. . . 17
3.1 Power model configurations, with the different cell states (columns) in different configurations (rows). . . 22
3.2 The summarized simulation setup. . . 26
4.1 The baseline deactivation timers and the corresponding sleep state. . . 28
4.2 Deactivation timers for 1 Mb file transmissions. . . 28
4.3 10 Mb file transmission delay increase with deactivation timers. . . 29
4.4 1 Mb file transmissions delay increase with optimized deactivation timers. 29 4.5 10 Mb file transmission delay increase with NR deployment. . . 30
4.6 Energy savings in percentage for each configuration from Figure 4.14. . . 38
4.7 The increase of energy savings in percentage between using DA1 and DA2. 39 4.8 Energy savings in percentage for each configuration from Figure 4.19 and Figure 4.18. . . 41
4.9 The updated wake-up times and sleep factor used in sensitivity analysis. . 43
4.10 The energy savings in percentage between configurations from adjusting wake-up time. . . 45
4.11 The energy savings in percentage between configurations from Figure 4.11 46 6.1 Summarized delay and energy saving results. . . 53
List of Abbreviations
3GPP Third Generation Partnership Project CDMA Code-Division Multiple Access
CRS Cell Specific Reference Signal
DA Deactivation timers
EARTH Energy Aware Radio and neTwork tecHnologies EDGE Enhanced Data Rates for GSM Evolution
FFT Fast Fourier Transform
FTP File Transfer Protocol
FWA Fixed Wireless Access
GSM Global System for Mobile Communications
HSPA High Speed Package Access
ICT Information and Communications Technology IEEE Institute of Electrical and Electronics Engineers
IoT Internet of Things
ISD Inter-site Distance
LTE Long Term Evolution
MAC Media Access Control
MIMO Multiple Input Multiple Output
NR New Radio
PBCH Physical Broadcast Channel PSS Primary Synchronization Signal
RF Radio Frequency
SDU Super Dense Urban
SI System Information
SSS Secondary Synchronization Signal
SS Sleep State
TD-SCDMA Time Division Synchronous Code Division Multiple Access
w/o without
w/ with
Chapter 1
Introduction
New Radio (NR) is the radio access network for the fifth generation (5G) mobile system defined by Third Generation Partnership Project (3GPP). 3GPP is the standard orga-nization that develops the protocols for mobile networks. The early implementation of 5G is already upon us and released for commercial use in the US and South Korea. By the end of 2019, 16 more major countries will have launched 5G. However, it is expected that it will not be until after 2020 that the majority of countries will deploy 5G networks [1]. With the increasing usage of mobile networks and internet of things (IoT), there is also an increase of energy consumption to satisfy the large data traffic volumes. This is a major concern when developing next generation mobile systems. A study about the global footprint of mobile communications [2] predict that between year 2007 and 2020 the increase of CO2 emissions will increase by a factor of three.
By the end of the year 2024, 1.9 billion mobile connections are expected by Ericsson to be provided by 5G. This corresponds to over 20% of the global mobile connections (excluding IoT) at that time [3]. The increase of mobile connections between the years of 2014 and 2024 is illustrated in Figure 1.1. In terms of IoT connections, it is expected to increase from 8.6 billion to 22.3 billion between the years of 2018 to 2024.
The impact on the climate due to the energy consumption depends on the source of energy. The most common energy source is fossil fuels. Burning fossil fuels releases energy which could be used for electrical power. The downside of using fossil fuels are the release of CO2 which traps heat. As more CO2 is released in the atmosphere the
more the global temperature rises and affects the climate. As there is a rise of attention and awareness of climate change and the rising global temperature, it is vital to save energy by increasing energy efficiency. There are many ways to save energy, however in this thesis the focus will solely be placed on energy efficiency enhancements using sleep cycle algorithms.
Figure 1.1: The increase of global mobile connections (excluding IoT) between year 2014 and 2024 [3].
1.1
Background
Mobile networks have gone through five generations from, 1G to 5G, shown in Figure 1.2 below. 1G was the start of mobile networks where it was possible to make mobile voice calls. With the introduction to 2G text messaging via SMS was made possible. 3G was the beginning of mobile web browsing and when 4G was released, data speed and capacity increased significantly which then lead to video streaming via the mobile phone. Finally, with 5G comes a variety of new features like improved network connection and lower latency. It opens up new use case opportunities and allows for more connected devices which was previously not possible.
One of the major points of concern is the increase of energy consumption of Information and Communications Technology (ICT) as it is quickly changing and developing. Between the years 2010 and 2015, the ICT energy consumption has increased by 31%. Additionally during that time the operational carbon emission grew by 17%. According to Ericsson, the total annual operational carbon emission in 2015 of ICT networks was estimated to be 0.53% (32 Gtonnes) of the global CO2 emissions related to energy, or 0.34% (40
Gtonnes) of all CO2 emissions.
The annual increase of energy consumption and operational CO2 emission is
approxi-mately linear, and the reason for the increase is mainly due to the expansion of mobile networks [6]. By the end of 2024 it is estimated that 5G is going to provide coverage to more than 40% of the world’s population. Between the years 2018 and 2024 the total mobile traffic data is expected to grow by a factor of 5 [7].
The largest power draining part in mobile networks are the base stations, which take up about 57% of the total energy consumption. Additionally, as seen in Figure 1.3 the total increase of base stations has been roughly 95%. Most of the power optimization can therefore be done in the base stations [8], [9]. The energy consumption of other parts in mobile networks are presented in Figure 1.4. Apart from reducing the environ-mental footprint by reducing energy consumption, it also reduces the operational energy costs for the network operators. The reduced cost could help with providing mobile communications in developing countries by improving the communication infrastructure there.
Figure 1.3: The increase of base stations between the years 2007-2012 [8], [9].
1.1.1 Approaches to Energy Efficiency
There are a few approaches to take when optimizing energy efficiency [9]: • Improving energy efficiency of base station hardware components • Turning off components selectively e.g. sleep mode techniques • Optimizing energy efficiency of the radio transmission process • Planning and deploying heterogeneous cells
Figure 1.4: Figure shows from left to right, the power consumption in mobile networks, and the power consumption distribution in a base station [8].
figure also shows that the power amplifier part in the base station is the biggest part of the energy consumption, most of it dissipated through heat. Improving this aspect comes with a higher equipment cost as it needs to be replaced and upgraded. Better power amplifier could provide higher power efficiency. The cost of the hardware upgrade increases as more power efficiency is desired.
Selectively turning off components e.g. by using sleep cycle algorithms depending on the traffic load can be a good implementation. In dense areas, like super dense urban (SDU) areas, with many base stations or small cells this option could be especially good for energy savings since there will be times of low load in which many base stations or small cells will not be transmitting to any active users. Sleep cycle algorithms could also be applied in less dense areas, like rural areas with less base stations. In those scenarios there are lower loads leading to more sleep opportunities. This approach could come at the cost of reduced performance.
Ways to optimize energy efficiency of the radio transmission process could be done by using techniques like multiple-input and multiple-output (MIMO) technique, cognitive radio transmissions, cooperative relaying, channel coding, and resource allocation for signaling. These are all methods to increase efficiency of utilizing resources in time, frequency and spatial domains. This category of optimizing energy comes at the cost of reduced performance.
Deploying small cells (low powered base station coverage area) in an efficient way could also reduce energy consumption. This is called heterogeneous deployment and is further discussed in Section 2.2. The small cells help with improving the coverage between the user and base station by increasing pathgains, and increasing capacity. The drawback is having too many ”unnecessarily” deployed small cells could instead increase energy consumption and radio interference between cells when compared to homogeneous de-ployments. Heterogeneous deployments could be combined with sleep mode techniques to mitigate the drawback of heterogeneous deployments.
1.1.2 5G NR Network
• Massive system capacity • High data rates
• Low latency
• Ultra-high reliability and availability • Low device cost and energy consumption • Energy-efficient networks
• Interoperability with existing wireless networks
NR has the potential to allow for more energy savings compared to previous generations of mobile networks as the base stations can sleep for longer and allows for deeper sleep thanks to the ultra-lean radio access design. Previous mobile networks transmit some signals periodically even when there are no active users in the base station coverage area i.e. the base station cell. These signals are not data user related but are for channel estimations, base station detection, and broadcast of system information. Those signals could also be called always-on signals. The always-on signals in the LTE case, does normally not add much to the total energy consumption as they are designed to only have a minor impact. However, in dense deployments where not all network nodes will always be utilized to its fullest potential, it results in a low average traffic load per node which altogether adds up to more energy impact. The always-on signals will then be a bigger part of the overall transmissions. The ultra-lean design aims to minimize the always-on signals [10], [11]. Instead of transmitting the always-on signals at regular intervals it should only be transmitted when necessary. More about the always-on signal impact is further discussed in Section 2.8.
NR introduces a vast range of frequencies in which it could operate. As of 3GPP Release 15 two new frequency bands has been introduced, high frequency band and low frequency band, which are used for different purposes. The low bands include the range 0.45-6 GHz and the high bands include the range 24.25-52.6 GHz. The higher frequencies provide higher data rates over a smaller area while the lower frequencies provide lower data rates but higher coverage.
Since NR now can operate in the higher frequency bands, it also has a much smaller wavelengths compared to LTE. Although this allows the NR base stations to be smaller in size, the wavelengths are blocked more easily. One of the solutions to this problem could be countered with using many more base station cells to not lose the area coverage of the signal, like when using heterogeneous deployment. However, the increase of cells will also lead to an increase in energy consumption. To counter this problem an option is to take advantage of the fact that there is lower traffic during some periods of the day and areas. For example, in not very densely populated areas it is not necessary to always have all the base stations active and ready to transmit as there is less activity in every cell [11].
1.1.3 Thesis Focus
The main goal of this thesis is to investigate and optimize the energy savings of using sleep cycle algorithms and investigate its impact on the users by looking at the performance in different scenarios with different network deployments. The new NR features allows for more energy savings by using sleep cycle algorithms with deeper sleep states compared to the previous generations’ LTE.
standalone deployment. The LTE-NR non-standalone means 4G and 5G will interwork while NR standalone only deploys 5G. There is also the third deployment option which is LTE standalone where only 4G is deployed. This deployment is the one deployed currently before the transition. All three deployments are simulated in this thesis. LTE standalone deployment is used in the baseline reference case.
Evaluations for energy savings and user impact when deploying LTE-NR dual connec-tivity is done by having NR cells added to the existing LTE cells. Simulations of the scenarios will be done with an internal java-based simulator which implements an algo-rithm that puts NR cells to different degrees of deep sleep.
The following subjects are investigated:
• How much energy could be saved by implementing sleep states to base station cells while minimally impacting the users
• Comparisons between the different network deployments with an implemented sleep cycle algorithm
• How much latency and energy usage are changed by implementing different degrees of deep sleep
• How different traffics are affected by the sleep cycle algorithm
• How the increase of traffic load affects the latency and energy usage with sleep states
• A sensitivity analysis of the power model that has been used in the evaluations to analyze how the sleep factor and the activation time affects the model
• User performance impact with sleep states from different perspectives, like worst user case and best user case
• What factors are important when optimizing and implementing the sleep state algorithm
1.1.4 Delimitation
The scenario setup for the simulations has several different parameters and settings that could be changed to fit a desired scenario. For the thesis these parameters and settings have been set according to what 3GPP or the supervisor has recommended. The SDU scenario has been used for the simulations, which is the scenario with the densest deploy-ments and highest traffic load. This scenario was chosen because it simulates the highest traffic load allowing a better view of the impact of sleep states. Other scenarios have not been included in this thesis due to time limitations. More about the scenario setup are described in Section 3.5.
The traffics investigated are large file transmission and multiple small file transmissions. Large files are transmitted in bursts with a set time interval in-between, this is explained more in Section 2.7. However, in the case of small file transmissions, the whole file is sent in one instance without being split up. With this traffic, time between the small file transmissions can be controlled and configured.
In this thesis it is assumed that when base stations enter sleep they do not lose connection and thus there is no need to re-establish contact to the users again. The deactivated components in the base station when it is in sleep does not affect the connection between the user and the base station.
simulated. Also [12] has defined 4 sleep modes as possible depending on the capabilities of hardware components at the base station.
When evaluating the user performance impact, the main impact that has been inves-tigated is the latency. Another subject that could be investigated is the download throughput. However, when it comes to what scenario shows worse or better results, both latency and throughput show almost the same results, but inverted as shown in Section 4.1. Therefore, to delimit only the latency is investigated.
1.1.5 Previous Work
Chapter 2
Theory
2.1
Base station
The base station communicates with the core network and the users. It is responsible for transmitting and receiving radio signals and creating a cell area.
The parts that are essential for a base station are:
• Antenna elements, the part that receives and transmits data to and from the base station. The more antenna elements used the more capacity could be increased. • A tall structure, i.e. a tower or something else tall that the antenna could be
mounted on as the signals need the height to not be blocked by objects like trees or buildings.
• The hardware, the electronics that support the operation of the base station. The power amplifier is housed in this and helps with amplifying the transmission signal. • A mobile switching station, that connects the wireless calls to the main telephone
system.
2.2
Heterogeneous deployments
Figure 2.1: A heterogeneous network deployment with hexagonal cells, macro base stations, micro base stations and users.
2.2.1 Macro cells and Small cells
The macro cells are cells with the widest coverage area and use higher transmission power than small cells. Therefore, macro cells are mainly responsible for the baseline coverage and small cells are placed within the macro cell to increase capacity, coverage, and data rates.
Since NR must fulfill high data rates, reliability, and low latency, it must either utilize denser macro deployment or take advantage of small cells. There are different types of small cells that could be used with heterogeneous deployments. They are all categorized by the size of the coverage area and the number of users that these cells can service. There are micro cells, pico cells, and femto cells, that vary in heights from tallest to shortest respectfully. In comparison to the others which are suited for indoor use, macro cells are suitable for the outdoors because of it’s higher coverage area [14]. The size of the coverage area from the cells varies depending on the frequency and bandwidth of the signals. Higher frequency and more bandwidth available the bigger the coverage area. The coverage area that the cells can provide is also related to its output power. The wider the coverage area, the higher the output transmission power is from the cell. Since macro cells are taller it means that the antenna is at a higher height. The higher the antenna is placed, the less likely it will be that the signals will be hindered by obstacles such as trees and buildings thus coverage area is not limited by the obstacles.
2.3
LTE and NR basics
2.3.1 Orthogonal frequency division multiplexing, OFDM
Orthogonal frequency division multiplexing (OFDM) is a kind of multicarrier transmis-sion where subcarriers are closely packed with a subcarrier spacing of ∆f , as seen in Figure 2.2. On the transmission side, subcarriers are modulated with data during one symbol duration which is one OFDM symbol duration. To produce the OFDM symbol an Inverse FFT transform was performed on the subcarriers. Guard intervals are then introduced between each OFDM symbol to prevent interference between them. They are then concatenated to create radio frames of data to be transmitted. The receiver can then execute a FFT on the OFDM symbols to retrieve the data [15], [11].
Figure 2.2: OFDM subcarrier spacing ∆f [15].
2.3.2 Time-domain and Frequency-domain structure
In the time-domain, both LTE and NR transmissions are organized into frames that are 10 ms. Each frame consists then of 10 subframes that are 1 ms each. For LTE, one subframe then consists of two slots. For NR, one subframe consists of one or more slots depending on the subcarrier spacing, as seen in Figure 2.3. When NR subcarrier spacing is 15 kHz, one NR slot has the same structure as an LTE subframe. The slots then are divided into several OFDM symbols. Each LTE slot either consists of 6 or 7 OFDM symbols and each NR slot consists of 14 OFDM symbols [15], [11].
Figure 2.3: Time-domain structure of LTE and NR [15], [11].
a user can be scheduled. The resource elements are then grouped together to create resource blocks. There is a difference between NR and LTE resource blocks. An NR resource block is one-dimensional, spanning the frequency-domain only, while the LTE resource block is two-dimensional with 12 subcarriers in frequency-domain and one slot in time-domain. NR resource blocks are more flexible in time-domain thanks to this. The frequency-domain structure is shown in Figure 2.4.
Figure 2.4: Frequency-domain structure of LTE and NR [15], [11].
2.3.3 Reference signals
Reference signals in the physical layer are predefined signals, which have different pur-poses and occupies specific resource elements. Some of those signals and its purpur-poses are [15], [11]:
• The primary synchronization signal (PSS): used for radio frame synchroniza-tion by the user in cell search. It is the first signal the user device will look for when it enters the cell and can also find the cell identity of the cell identity group. • The secondary synchronization signal (SSS): used for radio frame synchro-nization by the user and is the second signal the user look for after PSS. It can find the cell identity group of the detected cell.
• The system information (SI): system information is the non-user-specific in-formation that is necessary to properly operate in the network. In LTE case, SI is transmitted periodically over the entire cell. In NR case, SI is split into two parts. The first part are transmitted with longer periodicity (160 ms) and the second part can be transmitted when necessary.
• The physical broadcast channel (PBCH): necessary for the users to access the network and carries some of the system information
• The cell-specific reference signal (CRS): used for channel quality estimation, downlink demodulation, and time-frequency tracking. NR does not use CRS and use user-specific reference signal for channel estimation, which means that these signals are not transmitted unless there is data to transmit.
2.4
Dual Connectivity
Figure 2.5: Example of dual connectivity.
The primary cell and secondary cell operates on different carrier frequencies and are con-nected via a non-ideal backhaul, meaning there is a certain latency and limited capacity in the backhaul. In LTE-NR non-standalone mode, the primary cell uses an LTE base station and an NR base station for secondary cell. The connection to the core network in dual connectivity is split between the user plane and control plane. Both primary cell and secondary cell have direct contact with the core network in the user plane, but in the control plane only the primary cell has direct contact with the core network. The user plane carries the user data, while the control plane is responsible for transmitting system information and controlling the user connectivity [16].
The user plane and control plane protocol stack consists of the physical (PHY), medium access control (MAC), radio link control (RLC), and the packet data convergence protocol (PDCP) layers. The control plane additionally consists of the radio resource control (RRC). The key functionalities of the different layers are described as follows [17]:
• The PHY layer transmits information from MAC transport channels and handles functions like power control, link adaptation, and cell search.
• The MAC layer handles the mapping between logical channels and transport chan-nels, scheduling information reporting, error correction, priority handling between users, ets.
• The RLC layer transfers protocol data units (PDU) to upper layer, error correction, segmentation and re-segmentation, etc.
• The PDCP layer is responsible for transferring user data, header compression, du-plication detection, packet dudu-plication, etc. In the control plane this layer performs ciphering and integrity protection.
• The RRC layer handles the establishment, configuration, maintenance, and release of data radio and signaling radio bearers, broadcast of system information, mobility handling, etc.
Figure 2.6: Bearer concept of dual connectivity with LTE and NR [16].
2.5
Duplex Schemes
There are two different duplex schemes that could be used for both LTE and NR, the frequency division duplex (FDD), and the time division duplex (TDD) [11], [15]. The duplex schemes provide spectrum flexibility and allow the base station to transmit on both uplink (from user to base station) and downlink (from base station to user) simul-taneously. Both LTE and NR support separation of uplink and downlink transmissions. Both duplex schemes are illustrated in Figure 2.7.
FDD allows the uplink/downlink transmissions to use different frequencies and can trans-mit simultaneously. The downside of this duplex scheme is that it requires more frequency bands compared to TDD.
TDD allows the uplink/downlink transmissions to use the same frequency but are sepa-rated by time. The downside of this is that it is more susceptible to interference.
Figure 2.7: Duplex schemes in time-frequency structure for uplink and downlink [15].
2.6
Scheduling
The scheduler decides how and when the data is transmitted between the users and base station depending on the channel variations. Best time to transmit is when the conditions are advantageous to not risk losing packets, increase retransmissions, and overall to utilize radio resources effectively. The operation of the scheduler used in the simulations is called dynamic scheduling, this is where the scheduler makes a decision for every subframe. The scheduler decides the resource allocation for users before it transmits the data depending on conditions like the channel quality, and buffer status and priorities of different data flows.
The scheduling strategy is implementation specific. This means that there is no standard but it is chosen depending on what is desired. The scheduling strategy used in simulations is round-robin scheduling. In round-robin scheduling users take turns to utilize the shared resources, without taking into account the channel condition. The radio resources are then equally distributed.
Link adaptation is closely related to scheduling. Link adaptation tries to adjust trans-mission parameters to compensate for variations in the instantaneous channel conditions. When channel quality is too poor, the error probability increases when transmitting data. Either power at the transmitter could be controlled to keep a constant data rate or the data rate could be adjusted to compensate for the varying channel conditions. The latter is better for packet-data traffic where a constant data rate is not necessary.
Before transmission, both the scheduling, and link adaptation try to best adapt to the channel variations. However, since the radio-link quality varies quite a lot, it is impossible to have perfect adapation to it and could cause lost packets during transmission. The variations could be caused by several reasons like shadow fading, which is when obstacles affect the signal, or distance-dependent path loss. This is where hybrid automatic repeat request (HARQ) comes in. It requests retransmissions of erroneous data packets and handles random errors like noise in the receiver [15].
2.7
File transfer protocol, FTP
FTP use TCP as its transport layer protocol and allow a client to download files from a server. When a connection is created between an ftp client and an ftp server, a TCP connection is established and provides reliable transmission. Another characteristic of FTP files is that they are packaged into smaller packets before being transmitted, they are then reassembled after the transmission, resulting in transmissions with a small interval in-between them.
When a client or a server wants to send large amounts of data over TCP, the data is limited by a window. The window is the amount of data that can be sent without receiving a confirmation that the data has been received on the other side. For example, when the client begins transmitting data and the server starts processing the data, if the client starts transmitting more data when the server has yet to finish the processing, then the window closes for the client. When the window is closed, the client cannot continue transmitting. After the server has finished processing the data, it opens the window and the transmissions continue. The window gradually moves as data is acknowledged until everything has been sent.
slow network, then the source side will want to send more data than the destination side can handle, thus risking data loss. The congestion window gradually increases until a threshold is reached where the network congestion would most likely occur. Thus, the source side must send data that does not exceed both the window on the destination side, and the congestion window.
Another characteristic of TCP is its’ slow start. The slow start is what defines the maximum congestion window. It starts off as one data segment transmission and waits for confirmation. If it receives the confirmation then two segments of data is sent. This process repeats itself by increasing the segment size with a factor of two until either the source has reached the window size, or congestion occurs. If congestion is reached, then the congestion window will be halved, and the value is saved as the congestion threshold. After the congestion window has been halved the slow start repeats itself. During the second time the slow start process starts, if the congestion window is less than or equal the threshold, then it is possible to send double the data. However, if the congestion window is larger than the threshold, then doubling the data would most likely cause congestion. In that case the congestion window is only increased by a small amount [18]. This is called congestion avoidance and is also shown in Figure 2.8.
Figure 2.8: The congestion avoidance [18].
2.8
Power models
2.8.1 Idle mode power consumption
The idle mode is the mode in which base stations have no active users in cell and does not transmit data. The power consumption in idle mode is different between using either LTE base stations or NR base stations [13]. Thanks to the ultra-lean design, as described in Section 1.1.2, NR base stations can have several deep sleep states while, LTE can only use micro sleep. The LTE micro sleep is a sleep state that is even shallower in sleep compared to NR sleep states, meaning that less energy can be saved in micro sleep. The power consumption when LTE base stations and NR base stations are in idle mode can be seen in Figure 2.9, which shows that LTE cells has to ”wake up” many more times in a shorter time period compared to NR cells. This also shows that LTE cells in idle mode consumes more energy overall compared to NR cells.
LTE case, all subframes transmit the CRSs. From Figure 2.9, at subframe 0 there is an increase in the power consumption and it contains PSS, SSS, and PBCH. At subframe 5 there is a period of time of about 1 ms where the energy consumption increases and it contains SI, PSS, and SSS. In between the peaks of power consumption the LTE base station can sleep for a few OFDM symbols i.e. micro sleep. Before each CRS transmission which is one OFDM symbol (71.4µs), the LTE base station has to start waking up, thus there is a small increase of energy consumption right before the peak.
In the NR case, because of the ultra-lean design the signals are not transmitted as frequently as in the LTE case, and signals could also be transmitted simultaneously. Thus, minimizing the number of transmissions. The power consumption in idle mode for NR base stations depends on the number of synchronization signal blocks (SSB) in a cell, the subcarrier spacing, and more. For comparison sake, the subcarrier chosen is the same as LTE case. In NR, SSB contains PSS, SSS, and PBCH. It is assumed that transmission of both the SSB and SI could take place in the same symbol and consumes 25% of the total base station power. The SSB transmission periodicity is 20 ms. In between the SSB and SI transmissions, the NR cell can sleep longer and deeper. Although the peaks are higher in NR case, it consumes less energy overall because of the less frequent transmissions in the idle mode.
Figure 2.9: The idle mode power consumption of LTE and NR [13].
2.8.2 LTE base station power model
Power consumption at variable load of an LTE base station cell configuration was based on the EARTH Model [19]. The relation between radio frequency (RF) output power, and the base station power consumption are almost linear, therefore a linear approximation was made. The power model is defined in Eq.(2.1), where Ptx is the relative RF output
power, δ is the sleep factor, Pcell is the base station power consumption, P0 is the power
consumption at no load RF output power, and ∆P is the slope of the load dependent
power consumption from the linear model parameter. The ∆P slope can be seen in
2.1, which was used with the LTE idle mode power consumption. Pcell = Ntransmitter× ∆pPtx+ P0, if Ptx> 0 (BS transmitting) P0, if Ptx= 0 (BS no data to transmit) δP0, if Ptx= 0 (BS in a sleep state) (2.1)
Figure 2.10: Power consumption dependency on relative RF output power in base station [19].
BS type Ntransmitter Ptx [W] ∆p P0 [W] δ
Macro cell 2 20 4.7 130 0.84
Table 2.1: Example of a 3-sector LTE macro 2x2 LTE power model parameters.
2.8.3 NR base station power model
For NR base station, the power model that is used is the power consumption model of large scale antenna systems with N antenna elements [20]. Eq.(2.2) shows the NR power model which uses four sleep states, shown in Figure 2.11. The parameters in Eq.(2.2) are N , which is the number of antenna elements, is the efficiency of the power amplifiers, P0 is the no load power consumption when not transmitting, Pc is the circuit power per
antenna branch, δ is the sleep factor, and Ptx is the radiated power per cell. Table 2.2
is an example of the power model parameters, which are also the parameters used with the NR idle mode power consumption.
It should be noted that the power model proposed in this thesis are more conservative compared to the power model in [12]. It is believed that there are further limitations on the base station which could impact the activation time and power usage. The limitations are taken into consideration in the updated model, therefore the power model presented in this thesis is thought to be more realistic.
Pcell = Ptx + N Pc+ P0, if Ptx> 0 (BS transmitting) N Pc+ P0, if Ptx= 0 (BS no data to transmit) δP0, if Ptx= 0 (BS in a sleep state) (2.2) Ptx [W] P0 [W] Pc Sleep factor, δ BS cell 40 0.25 260 1 0.50, 0.25, 0.125, 0.0625
Figure 2.11: The four defined sleep state modes with the sleep factors and total power consumption, P .
2.9
Energy and performance metrics
To quantify the network performance, some performance metrics are necessary to define. The performance metrics are used to study the benefits of different technology solutions, like sleep states.
For the energy evaluations, the metric needed to relate the energy consumption to the coverage area to quantify the energy consumption. The EARTH model defines the power per area unit metric, as described in Section 3.2. When designing an energy efficient network, it is important to consider both the energy consumption metric and the user performance metric. The goal is to minimize the energy consumption, while not decreas-ing the performance of the network [19].
When evaluating the user performance in radio networks, some of the performance metrics that are important to investigate, are the perceived average user throughput and the percentiles of the user throughput, as recommended by 3GPP [21]. With the perceived average user throughput, the average from all perceived throughput for all transmitted data bursts are intended for the user.
To evaluate the user performance in this thesis, latency has been used instead of through-put. More explained in Section 4.1.
Chapter 3
Methodology
3.1
Performance evaluation
3.1.1 Round Trip Time, RTT
The Round Trip Time (RTT) has been calculated according to Eq.(3.1), where tf r is the
time of file reception, and trq is the time when the file request was sent.
tRT T = tf r− trq, [s] (3.1)
This is the latency that is being investigated in the simulations. The RTT delay includes amongst others the delay that comes from queuing, activating, transmitting, processing, and propagating. It is necessary to include all these delays from file request until the end of the file reception as it will otherwise be difficult to compare the latency of different scenarios with different degrees of deep sleep.
3.1.2 Downlink user throughput
The downlink user throughput is calculated using RTT, as seen in Eq.(3.2). N rOf Bytes refers to the number of bytes downloaded by the user.
T hroughput = N rOf Bytes 1 · 106· t
RT T
, [M b/s] (3.2)
The throughput is the transfer rate between the user and base station. It is the actual data transmitted and successfully received during a specific amount of time.
3.2
Energy evaluation
3.2.1 Daily traffic load
the given daily traffic profile, an estimation can be done of the traffic for every hour as a fraction of the maximum traffic load.
Figure 3.1: The average daily traffic profile in Europe [19].
The total traffic load is calculated by taking the downlink data rate divided by the coverage area, A, as described in Eq.(3.3). This is used for when plotting the performance evaluation figures in Section 4.3 as it is used in the x-axis of the figures.
Xtraf f ic =
DataRate
A , [M b/s/km
2] (3.3)
3.2.2 Power per unit area
The power per unit area is defined as in Eq.(3.4), where P is the average power con-sumption and A is the coverage area. This metric is closely related to the CO2 emission
[19].
Parea =
P
A, [W/m
2] (3.4)
3.2.3 Average daily power consumption
Energy performance is defined as the daily average area power consumption [19]. The daily power consumption is defined by Eq.(3.5), where Pactive and Pidle are the power
consumption of either the LTE or NR power model depending on what base station was used, η is the daily traffic profile point, and Ncells are the number of cells.
EP = 1 24 P24 t=1 PNcells
i=1 [(Pactive)ηti+ Pidle(1 − ηit)]
A , [kW/km
2] (3.5)
3.3
Sleep mode algorithm
savings. Depending on the number of deactivated parts, the longer the time to activate said parts. While it may be possible to quickly shut down the components in the nodes to go into deeper sleep, it will however take time to activate them. Some will take longer to activate depending on factors such as, if there are components which needs to be warmed up before operating, or if shuffling of data between memory parts in the base station is necessary.
Problematic for most sleep cycle algorithms is knowing when to put base station cells to sleep and when to activate them. If the base station cells wake up too late, the user performance will be impacted negatively. If they wake up too early or they start sleeping too late, the energy efficiency will not be as optimal as it could be. Therefore, in the presented algorithm, deactivation timers are introduced. The deactivation timers puts the cells gradually to sleep to mitigate unnecessary sleep.
The implemented sleep cycle algorithm used in this thesis has six states, with four sleep states (SS) used with NR base station cells, which is the number being used in the evaluations:
• Awake mode: The base station cell is active and transmitting data to users. • Sleep State 1 (SS1): The base station cell is in a very light sleep, but still ready to
wake up and transmit when active users arrive in the cell. It has the lowest sleep factor and shortest activation time. This is the first sleep state the cell enters after awake mode.
• Sleep State 2 (SS2): The base station cell is in a deeper sleep than SS1 with a little longer activation time and higher sleep factor compared to the standby mode. The cell enters this sleep state after SS1.
• Sleep State 3 (SS3): The base station cell is sleeping with an even higher sleep factor and activation time compared to SS2. The cell enters this sleep state after SS2.
• Sleep State 4 (SS4): The base station cell is in deep sleep and takes the longest to activate but has the lowest sleep factor and uses the least energy. The cell enters this sleep state after SS3.
• Activating mode: The base station cell is in the process of waking up and cannot transmit data yet. This mode is activated if cell is in one of the sleep states and suddenly needs to wake up. After waking up and leaving this mode the BS cell enters an awake mode.
LTE base station cells use only two states. Those states being:
• Awake mode: Cell is awake and active. Consumes more power compared to the awake mode of NR cell.
• Micro sleep (SS0): This is the only sleep state of LTE cells. It is actually not really a sleep state as it does not require activation that causes an additional delay. The wake-up times or activation times, and power consumption that belongs to each of these sleep states are summarized in Table 3.1. The power consumption is presented in percentage of overall power usage. The sleep factor of the four sleep states are also presented in Table 2.2. LTE0 is the configuration for SS0 and NRX are the configurations for the different deep sleeps SS1-SS4.
It is clear from Table 3.1, what power usage and activation time should be associated with which sleep state. However, when a cell is in activation mode, the activation time depends on what sleep mode the cell was in when it enters that mode.
Mode configuration LTE0 NR1 NR2 NR3 NR4 Awake Activation Wake-up time (ms) 0 1 10 100 1000 0 1-1000 Power usage (%) 84 50 25 12.5 6 100 6-50 SS0 X SS1 X X SS2 X X X SS3 X X X X SS4 X X X X X
Table 3.1: Power model configurations, with the different cell states (columns) in different configurations (rows).
activated. For example, if cell was in SS4 and then is triggered to wake up, it will start with setting the activation time to the corresponding 1000 ms and the energy usage in that state is 6% of the full energy usage. The cell though will only spend 900 ms (1000-100 ms) in SS4 before it enters SS3 where the activation time left will be (1000-100 ms and the energy usage is 12% of the full energy usage. For every subframe the activation timer is decreased, and the cell enters lighter sleep until it is fully awake. This process is also described in the flowchart in Figure 3.2. Additionally, illustrated in Figure 3.3 is the gradual activation process of cell with wake-up times.
Figure 3.3: Power model with a factor 2 decrease in power consumption for each 10-fold increase in wake-up time.
The baseline sleep cycle algorithm utilizes the data buffer of the scheduler. Whenever there are active users in the cell, and data arrive at the buffer, it triggers the base station to start waking up. If the base station is already sleeping, then it enters activation mode where an ”activation timer” starts. The base station does not start transmitting data until the activation timer expires and the base station enters an awake mode. If no active users enter the base station cell then the base station will successively enter each sleep mode until it reaches the deepest sleep state. The algorithm will be executed for each cell for every subframe as the scheduler allocates the resources. If the base station is already in a sleep state, or there are no active users, the ”sleep timer” increments for every subframe. If the base station is in an activation mode, the activation timer decrements for every subframe until it expires and starts transmitting. The flowchart for the sleep mode algorithm implemented is shown in Figure 3.4.
Figure 3.4: Flowchart of the sleep state algorithm with the four sleep modes.
If the deactivation timers are set to zero it means no deactivation timer is used. No deactivation timers mean that the cell will enter the next sleep state without waiting for the sleep timer to increase. For example, the deactivation time for SS1 is 0, which means the cell enters SS1 as soon as there are no data to transmit. If the deactivation time is 0 for both SS1 and SS2, it would mean that the cell enters SS2 as soon as there is no data to transmit.
The first algorithm, which is the baseline sleep mode algorithm used in simulations, utilizes macro cells for coverage, while small cells are mainly used for increasing data rates. In the simulations, the small cells are on the high frequency band which provides more bandwidth, thus providing higher capacity. Deeper sleep states are implemented on micro cells while the deepest sleep on macro cells is micro sleep. This algorithm is implemented for LTE-NR non-standalone deployments.
The second algorithm used in simulations, assumes that both the macro layer cells and the micro layer cells can utilize deep sleep, this means that both layers use NR cells. Utilizing deeper sleep on both macro layer and micro layer would possibly allow for more energy savings. This algorithm is implemented for NR standalone deployments.
3.3.1 Configurations of sleep states
SS ) + (micro layer cell SS ). Those mainly refer to the NR-NR deployment. SS0 or SS0+SS0 (LTE standalone) is the baseline configurations where no sleep states has been implemented.
3.4
Simulator
For the simulations a java-based simulator provided by Ericsson was used and post-processing of the data obtained by the simulator was done in MATLAB. In the simulator there are many different possibilities for parameters and traffic models that could be used to fit the simulation to the desired scenarios.
The event-driven simulator is a large scale radio system, which can model radio networks and users. It models deployments, propagation, protocols and applications for several different radio access technologies, including LTE and NR. The simulator additionally allows for logging of data, like delay and throughput.
3.5
Scenario setup
Simulation parameters Values
Scenario SDU
Peak hour traffic 720 Mbps/km2
Macro cell layer Hexagonal 3-sector
Micro cell layer Omnidirectional single sector
Macro antenna elements 2 at 25 m height Micro antenna elements 64 at 10 m height Transmit power macro/micro 40 W / 20 W
Propagation model 3GPP spatial channel model
Frequency band Micro cells, 3.5 GHz
Macro cells, 0.9 GHz
Bandwidth Micro cells, 40 MHz
Macro cells, 10 MHz
Duplex scheme FDD, subcarrier spacing 15 kHz
User antenna height 1.5 m
Indoor probability 80%
File downloads 10 Mb FTP file & 10 of 1 Mb FTP files
ISD 380 m
Table 3.2: The summarized simulation setup.
3.6
Method
The whole process of obtaining the simulation results has been an iterative process start-ing with definstart-ing the algorithm. After that updatstart-ing the simulator and teststart-ing it follows, see Figure 3.5. The process has been split into two parts, the system simulation part and the post processing part. The system simulation part simulates the scenarios with the algorithm. Following the system simulation is the post processing part, where simulation results are obtained. After evaluating the results, parameters and settings are changed to optimize the results and investigate other setups.
Chapter 4
Simulation Results
4.1
Throughput vs Latency
As seen from Eq.(3.2), the downlink throughput has an inverted correlation to the delay. The inverted correlation means if throughput decreases, the delay increases and vice versa. Therefore, for performance evaluations the throughput show about the same results as when evaluating latency but inverted.
Figure 4.1 shows the downlink throughput and average delay of the same scenario with increasing traffic load. Comparing them shows that they both have decreasing perfor-mance as the traffic load and number of used deep sleep states possible on cell increases. Therefore, to not increase redundancy and delimit, only the latency is evaluated when it comes to performance.
4.2
Different deactivation timers
(a) The average downlink throughput curves. (b) The average delay curves.
Figure 4.1: The performance curves of 10 Mb file download without deactivation timers.
The baseline deactivation timers used with the algorithm is presented in Table 4.1, and has been optimized for 10 Mb FTP file downloads.
Sleep States Deactivation timers (ms)
SS1 0
SS2 100
SS3 300
SS4 600
Table 4.1: The baseline deactivation timers and the corresponding sleep state. Figure 4.2 shows the average delay impact with increasing traffic load, where both sce-narios with using the defined baseline deactivation timers and not using them has been simulated. It is clear that using sleep states on cells without deactivation timers increase the delay. The overall delay also increases with increasing traffic load, especially for the configuration of using SS4.
Figure 4.2: Average delay of 10 Mb file download with optimized deactivation times and without deactivation times.
Simulations of multiple small file transmissions were done with the baseline deactivation timers, and with optimized deactivation timers for the specific traffic scenario (DA2), seen in Figure 4.3. When comparing using and not using optimized deactivation timers, the delay is smaller with the deactivation timers. The optimized deactivation timers (DA2) used in Figure 4.3b are presented in Table 4.2.
Configurations DA1 (ms) DA2 (ms)
NR1 0 0
NR2 200 500
NR3 700 700
NR4 1000 1000
(a) Deactivation timers not optimized. (b) Deactivation timers optimized.
Figure 4.3: The average delay of multiple 1 Mb FTP file downloads.
4.3
Performance Evaluations
4.3.1 LTE-NR dual connectivity deployment
Large file transmission
Figure 4.2 shows the performance result from large file transmission. The resulting delay increase has been summarized in Table 4.3 below. The increase in delay for SS1-SS3 was consistent despite the increase in traffic load as the baseline delay increases with the load. Generally, increasing the number of sleep states also increases the delay, and by extension, decreases the throughput.
Sleep configuration Average delay increase
SS1 1%
SS2 1%
SS3 2%
SS4 31% - 90%
Table 4.3: 10 Mb file transmission delay increase with deactivation timers.
Multiple small file transmissions
Figure 4.3b shows the performance result from multiple small file transmissions and is summarized in Table 4.4.
Sleep configuration Average delay increase
SS1 1%
SS2 1.6%
SS3 1.6%
SS4 6.6% - 14%
Figure 4.4 shows the comparison of the performance results of small file transmissions between using DA1 and DA2 from Table 4.2. Both DA1 and DA2 are optimized deacti-vation timers for this traffic but have different times before entering SS2. This is used to investigate how different optimized deactivation timers change performance and energy. Energy impact is further discussed in Section 4.4. The increase of delay is between∼1-2%
when using DA1 compared to the sleep configurations using DA2.
Figure 4.4: The average delay of multiple small file transmissions with deactivation timers from Table 4.2.
4.3.2 NR standalone deployment
Large file transmissions
Figures 4.5 shows the average delay results from 10 Mb file transmission with different combinations of number of sleep states on both the macro and micro layer. The results have been summarized in Table 4.5. Throughout all the scenarios one can conclude that using the 4th sleep state gives a rather large delay increase.
Sleep configurations
(macro+micro) Average delay increase
SS1+SS1/SS2/SS3 1% SS2+SS1/SS2/SS3 1% SS3+SS1/SS2/SS3 5% SS1/SS2/SS3+SS4 29% - 102% SS4+SS1/SS2 12% SS4+SS3 180% SS4+SS4 290%
(a) SS1 on macro layer (b) SS2 on macro layer
(c) SS3 on macro layer (d) SS4 on macro layer
Figure 4.5: The average delay with fixed SS configuration on macro layer with varying SS configurations on micro layer.
Figure 4.6: The average delay where the micro layer cells have fixed SS3.
4.3.3 10th and 50th percentiles
10th and 50th percentiles show the best case user performance and the median. The 10th percentile showed that 90% of the users have either the delay shown in the curve or higher. The 50th percentile curves are dashed, while the 10th percentile curves are solid. Figure 4.7 depicts the 10th and 50th percentile delay of 10 Mb file download of LTE-NR deployment with deactivation timers and without. It can be observed that the 10th and the 50th percentile delay curves are rather close to its corresponding baseline when utilizing SS1 to SS3 on the cells. The difference between the 10th and 50th percentile curves is about 8% at high loads when using deactivation timers, and up to about 12% when not using deactivation timers, not including SS4. SS4 on cells show a bigger increase in delay for both 10th percentiles and 50th percentiles. Without deactivation timers, the performance is worse when observing the 10th and 50th percentiles compared to using them.
(a) With deactivation timers (b) Without deactivation timers
Figure 4.7: The 10th and 50th percentile delay curves of LTE-NR deployment with 10 Mb file downloads.
(a) With optimized deactivation timers (b) Without optimized deactivation timers(DA2)
Figure 4.8: The 10th and 50th percentile delay curves of LTE-NR deployment with multiple 1 Mb file downloads.
Figure 4.9 shows the 10th and 50th percentile of NR standalone deployment with 10 Mb file download. Figure 4.9a and Figure 4.9b show similar results. Overall, the more sleep states used on either macro or micro layer increases even the best user case delay. The 10th percentile curves display that increasing SS on micro layer increases the delay, at most, about 33% for all cell configurations on the macro layer, with the most delay increase when utilizing SS4 on both macro and micro layer. The median delay results show similar results to the average delay.
The 10th percentile curves when observing Figure 4.9a, Figure 4.9b, and Figure 4.9c, prove the 10th percentile curves and 50th percentile curves do not have a bigger delay difference than about 17%. Figure 4.9d indicates the biggest delay difference between 10th percentile delay and 50th percentile delay is between SS4+SS4 (∼317%). Otherwise,
(a) SS1 on macro layer (b) SS2 on macro layer
(c) SS3 on macro layer (d) SS4 on macro layer
Figure 4.9: The 10th and 50th percentile delay with fixed SS configuration on macro layer with varying SS configurations on micro layer.
4.3.4 90th percentile
Investigations of the 90th percentile was done to study the worst case user performance. The 90th percentile curves show that 10% of the users have either the delay as seen in the curve or higher.
(a) With deactivation timers (b) Without deactivation timers
Figure 4.10: The 90th percentile delay curves of LTE-NR deployment with 10 Mb file downloads.
Figure 4.11 shows the 90th percentile delay of 1 Mb multiple small file download of LTE-NR deployment with optimized deactivation timers and without. The 95th percentile for the optimized deactivation timers is shown in Figure 4.12. While Figure 4.11a depicts that the worst users in all configurations have a small delay increase compared to the baseline. Figure 4.11b show that with SS2 and SS3, the delay does not deviate too far away from the baseline delay, but SS4 have a bigger delay increase.
(a) With optimized deactivation timers (b) Without optimized deactivation timers
Figure 4.12: The 95th percentile delay of multiple 1 Mb file downloads with optimized deactivation timers.
(a) SS1 on macro layer (b) SS2 on macro layer
(c) SS3 on macro layer (d) SS4 on macro layer
Figure 4.13: The 90th percentile delay with fixed SS configuration on macro layer with varying SS configurations on micro layer.
4.4
Energy Evaluation
4.4.1 LTE-NR dual connectivity deployment
Figure 4.14: The total energy consumption of the scenarios in Figure 4.2. Configurations Energy savings (%)
SS1 25 SS2 w/ DA 43 SS3 w/ DA 51 SS4 w/ DA 52 SS2 w/o DA 43 SS3 w/o DA 50
Table 4.6: Energy savings in percentage for each configuration from Figure 4.14.
Figure 4.15 and Figure 4.16 are the daily energy consumption.
Figure 4.16: Zoomed in version of Figure 4.15 with the three scenarios with the lowest energy usage.
The total energy consumption of multiple file transmissions using DA1 and DA2 can be seen in Figure 4.17, as well as the corresponding energy consumption of scenarios from Figure 4.4. The corresponding energy savings between using DA1 and DA2 is found in Table 4.7.
Configuration Energy savings (%)
SS2 DA1-DA2 1.3
SS3 DA1-DA2 1.3
SS4 DA1-DA2 1.2
Table 4.7: The increase of energy savings in percentage between using DA1 and DA2.
4.4.2 NR standalone deployment
Figure 4.18 depicts the total energy consumption of the scenarios from Figure 4.6, with the red bar representing the macro layer and the blue bar for micro layer. The corre-sponding energy savings in percentage is summarized in Table 4.8. Figure 4.19 shows the total energy consumption of the scenarios from 4.5.
Figure 4.18: The total energy consumption with fixed SS3 on micro layer.
(a) SS1 on macro layer (b) SS2 on macro layer
(c) SS3 on macro layer (d) SS4 on macro layer
Configuration Energy savings (%) Configuration Energy savings (%) SS1+SS1 39 SS3+SS1 51 SS1+SS2 55 SS3+SS2 66 SS1+SS3 59 SS3+SS3 76 SS1+SS4 59 SS3+SS4 76 SS2+SS1 46 SS4+SS1 54 SS2+SS2 64 SS4+SS2 73 SS2+SS3 73 SS4+SS3 81 SS2+SS4 73 SS4+SS4 81
Table 4.8: Energy savings in percentage for each configuration from Figure 4.19 and Figure 4.18.
4.4.3 Percentage spent in sleep states
Figure 4.20 shows the average percentage that the micro cell has spent in the different NR cell states for 10 Mb file download with LTE-NR deployment.
(a) The average total state percentages of the cells.
(b) Total time cells spent in activating state in percentage.
Figure 4.21 shows the average time the cells spend either in awake mode or in sleep mode for 10 Mb file download NR standalone as traffic load increases.
(a) The average time spent in awake state (b) The average time spent in a sleep state
Figure 4.21: NR standalone deployment with fix SS4 on macro layer cells. Figure 4.22 shows the total average percentage the micro cells spends in different cell states excluding the activation state. This shows the comparison between using opti-mized deactivation timers and not using optiopti-mized deactivation timers with LTE-NR deployment.
