Pico Cell Densification Study in LTE Heterogeneous Networks

Transcription

1 M.Sc. Thesis report Pico Cell Densification Study in LTE Heterogeneous Networks Supervisors: Fredric Kronestedt Systems & Technology Development Unit Radio Ericsson AB Ming Xiao Communication Theory Lab School of Electrical Engineering KTH

2

3 Abstract Abstract Heterogeneous Network (HetNet) deployment has been considered as the main approach to boost capacity and coverage in Long Term Evolution (LTE) networks in order to fulfill the huge future demand on mobile broadband usage. In order to study the improvement on network performance, i.e. capacity, coverage and user throughput, from pico cell densification in LTE HetNets, a network densification algorithm which determines the placement locations of the pico sites based on pathloss has been designed and applied to build several network models with different pico cell densities. The study has been taken based on a real radio network in a limited urban area using an advanced Matlab-based radio network simulator. The simulation results show that the network performance generally is enhanced by introducing more pico cells to the network. Keywords: HetNet, LTE, pico cell, densification, capacity, coverage, throughput. ii

4 Acknowledgements Acknowledgements This thesis was performed in System & Technology, DU Radio, Ericsson AB in cooperation with Communication Theory lab, EES, KTH. First of all, I would like to express my appreciation and gratitude to my supervisor from Ericsson AB Fredric Kronestedt for his great guidance and kindness during the last six months. I would also like to thank my supervisor from KTH Prof. Ming Xiao who gave me a great deal of assistance and provided me with support throughout the entire thesis. I must express my thanks to Dirk Gerstenberger and Fredric Kronestedt for giving me this opportunity to spend six months in Ericsson AB to accomplish my master thesis. My sincere gratitude is also directed to my examiner Prof. Mikael Skoglund for his time and for his help when applying for this thesis. I would also like to thank the following people from Ericsson AB for their support and kindness during the last six months: Gunther Auer, Jason Chen, Peter Björkén, Stefan Ström, Thomas Chapman and Tomas Lundborg. iii

5 Table of Contents Table of Contents Abstract... ii Acknowledgements... iii List of Figures... vi List of Tables... viii Terminology... ix 1 Introduction Background and problem motivation Overall aim Scope Problem statement Outline Theory Heterogeneous networks in LTE Low power nodes deployment Cell selection Two packet traffic models Full buffer model Equal buffer model Comparison Related Tools TEMS CellPlanner Elin LTE Astrid Models Studied Network Simulation parameters Propagation models Urban model ITU indoor propagation model Implementation Work flow Pico sites densification algorithm iv

6 Table of Contents Results Network densification Comparison of Pico sites densification algorithms Comparison of RSRP and Biased RSRP Network Performance Capacity and Coverage User throughput An extreme case Comparison of pico and macro cells served area Conclusions References v

7 List of Figures List of Figures Figure 2.1: Heterogeneous Network... 3 Figure 2.2: Cell selection in 3GPP LTE HetNets... 6 Figure 2.3: Users with different radio link condition... 8 Figure 2.4: Average bitrate in full buffer model... 8 Figure 2.5: Average bitrate in equal buffer model... 9 Figure 3.1: Analyzed area Figure 4.1: Work flow... 2 Figure 4.2: Flow chart of pathloss based pico sites densification algorithm Figure 5.1: Densified networks using pathloss based pico sites densification algorithm (chapter 4.2) Figure 5.2: Densified networks using randomly distributed pico sites densification algorithm (chapter 4.2) Figure 5.3: Pathloss based pico sites densification algorithm Figure 5.4: Randomly distributed pico sites densification algorithm Figure 5.5: Uplink comparison Figure 5.6: Downlink comparison Figure 5.7: Uplink comparison... 3 Figure 5.8: Downlink comparison Figure 5.9: Uplink mean user throughput vs uplink subscriber capacity Figure 5.1: Uplink cell edge user throughput vs uplink subscriber capacity Figure 5.11: Improvement of uplink capacity Figure 5.12: Improvement of uplink coverage Figure 5.13: Downlink mean user throughput vs downlink subscriber capacity Figure 5.14: Downlink cell edge user throughput vs Downlink subscriber capacity Figure 5.15: Improvement of downlink capacity Figure 5.16: Improvement of downlink coverage Figure 5.17: Uplink user throughput maps Figure 5.18: Improvement of uplink mean user throughput... 4 Figure 5.19: Downlink user throughput maps Figure 5.2: Improvement of downlink mean user throughput Figure 5.21: Downlink geometry distribution Figure 5.22: Uplink throughput vs capacity vi

8 List of Figures Figure 5.23: Downlink throughput vs capacity Figure 5.24: Average uplink utilization of pico and macro cells with the target of 1 Mbps uplink cell edge user throughput Figure 5.25: Average downlink utilization of pico and macro cells with the target of 1 Mbps downlink cell edge user throughput Figure 5.26: Pico and Macro cells served area Figure 5.27: Comparison of Pico and Macro cells served area vii

9 List of Tables List of Tables Table 2.1: Guidelines for LPNs deployment [3]... 5 Table 2.2: Comparison of Full and Equal buffer model... 1 Table 3.1: LPNs deployment Table 3.2: Simulation parameters viii

10 Terminology Terminology Abbreviations LTE LPN HetNet 3GPP CSG RRU UE RSRP RSRQ eicic TCP GSM CDMA WCDMA CPLM ACP SINR ABR MRC IRC Long Term Evolution Low Power Node Heterogeneous Network 3rd Generation Partnership Project Closed Subscriber Group Remote Radio Unit User Equipment Reference Signal Received Power Reference Signal Received Quality Enhanced Inter-cell Interference Coordination TEMS CellPlanner Global System for Mobile Communications Code Division Multiple Access Wideband Code Division Multiple Access Composite Pathloss Matrix Automatic Cell Planning Signal to Interference plus Noise Ratio Achievable Bitrate Maximal Ratio Combining Interference Rejection Combining ix

11 Terminology ISD Inter Site Distance x

12 1 Introduction Introduction 1.1 Background and problem motivation Mobile broadband usage has increased dramatically the last couple of years due to new types of terminals such as smart phones and tablet computers. According to [1], in 21, wireless devices only accounted for 37% of IP traffic; but by 215, wireless devices is estimated to consume 54% of IP traffic while wired devices will only consume 46% of IP traffic, which means that traffic from wireless devices will exceed wired devices. To support the huge future demands as both the number of users and the user demand will increase, it is essential to enhance the network capacity and coverage. But with the knowledge that the deviation between Long Term Evolution (LTE) link level performance and Shannon capacity is very small which limits the potential to increase spectrum efficiency [2], forcing us to find other means to meet the future demands. A key method to fulfill the future needs is network densification through adding smaller low power nodes (LPNs) in traditional high power macro nodes, namely Heterogeneous Network (HetNet), which is expected to boost capacity and coverage beyond what is available in current LTE networks [3]. With the knowledge that HetNet deployment has a large potential to improve the network capacity and coverage, the influence of pico cell densification on the network performance is obviously of large interest. 1.2 Overall aim In order to see how effective the method of HetNet is to solve the problem of the huge future demand on network performance, the project s overall aim is to investigate how the network performance will be affected by deploying more and more pico cells in the network. 1.3 Scope The study has its focus on the impact of pico cell densification on the network performance, the cost of the pico sites deployment is ignored; 1

13 1 Introduction as a result, the number of pico sites in the network models that have been built in this project might be unrealistic. The study will be taken based on a real radio network in a limited urban area in a dense major European city, the results might vary for different area and cities due to different terrain features and macro sites deployment. 1.4 Problem statement To achieve the overall aim stated in chapter 1.2, the study has an objective to respond to the following questions: Is it true that adding more and more pico cells will result in larger and larger capacity and coverage? Is there any upper limit on how many pico cells can be added to a dense urban macro network and still improves the network performance? To achieve this objective, it is also desired to design a strategy of pico sites deployment: How should the pico sites be deployed to achieve better network performance from capacity and coverage point of view? 1.5 Outline Chapter 2 provides the related theory concerning some aspects of HetNet in LTE and some tools that have been used during this thesis work. In chapter 3, the network model and simulation parameters of this project are presented, in addition with a brief discussion of an indoor propagation model which has been applied in a part of this study. Chapter 4 describes how the network models with different pico cell densities have been built and simulated. A pico sites densification algorithm to decide where to place the pico sites has been designed and is also discussed in this chapter. In chapter 5, the simulation result of this work will be presented. The conclusions and possible future work will be presented in chapter 6. 2

14 2 Theory Theory 2.1 Heterogeneous networks in LTE Heterogeneous network (HetNet) has been identified as a key method to fulfill the huge future demands on mobile broadband usage as both the number of users and the user demand will increase. In 3GPP LTE HetNets, traditional high power macro nodes are complemented with low power nodes (LPNs) which cover small areas and offer very high capacity and data rates in these areas [3], as illustrated in Figure 2.1. Figure 2.1: Heterogeneous Network Besides simply adding LPNs to the existing macro networks, there are some other approaches to expand network capacity and coverage as well, such as improving macro cells by allocating more spectrum and densifying the macro sites. Compared with these two methods, adding LPNs performs the same in the downlink and better in the uplink [3]. These three approaches can of course be combined together to meet higher demand Low power nodes deployment LPNs deployment is a real challenge in HetNets, many aspects need to be considered [3]: Demand: traffic volumes, traffic location, target data rates Supply: macro cell coverage, site availability, backhaul transmission, spectrum and integration with the existing macro network 3

15 2 Theory Commercial: technology competition, business models In [3], some guidelines for LPNs deployment are provided: Open or closed access Open access means LPNs are available for all subscribers to access. Open access should be chosen for public systems deployed by operators. Closed access refers that LPNs belong to a Closed Subscriber Group (CSG), that is to say access is only available for users in CSG. Closed access is always used in user-deployed cases (by individual, enterprises). Indoor or outdoor deployment Deploying indoor LPNs is suitable in cases when traffic is concentrated to specific indoor locations such as shopping malls. Outdoor LPNs deployment that also covers indoor areas is preferable in cases such as local traffic hotspots cover a wide area including several buildings or the macro cells in the existing networks are too sparse to meet indoor service demand. Type of LPNs There are several types of LPNs: Remote Radio Units (RRUs), conventional pico nodes, relay nodes. RRUs are suitable for networks with low-latency and high-capacity backhaul; otherwise stand-alone pico base stations should be preferable. Deploying relay nodes is a preferred option for networks without wire backhaul. Frequency reuse The HetNets can be seen as composed of two layers: macro cell layer and pico cell layer. The two layers can use different frequency band or share the same band. Reusing the frequency band of the macro cell layer for the pico cell layer is of course spectrum-efficient. When spectrum is scarce or capacity is the diver, frequency should be reused. However, due to the inter-layer interference, elaborate cell planning and interference management technique is needed. 4

16 2 Theory Table 2.1 summarizes the rules addressed above. Choices Guidelines Access Deployment Type of LPNs Frequency reuse Open access Closed access Indoor Outdoor RRU Stand-alone pico base stations Relay nodes Reuse macro spectrum Separate spectrum Operator-deployed User-deployed Concentrated large indoor hotspot Outdoor hotspot or many smaller indoor hotspot Low-latency and high-capacity backhaul (fiber) High-latency and low-capacity backhaul (copper/microwave) No wire backhaul Spectrum is scarce / Capacity is driver CSG Table 2.1: Guidelines for LPNs deployment [3] Cell selection Conventionally, cell selection is based on the downlink received signal strength which means mobile users will connect to the site from which the received power is strongest. For example, in 3GPP LTE, cell selection is performed according to two parameters measured by a User Equipment (UE): Reference Signal Received Power (RSRP) and Reference Signal Received Quality (RSRQ) [4]. Reference signal received power (RSRP), is defined as the linear average over the power contributions (in [W]) of the resource elements that carry cell-specific reference signals within the considered measurement frequency bandwidth [4]. RSRQ is calculated based on RSRP which provides additional information and ensures a reliable cell selection decision when RSRP is not sufficient. In homogeneous networks, RSRP based cell selection guarantees good channel conditions in both downlink and uplink. But in HetNets, since the transmission power in the downlink is different between the LNPs and the macro nodes and this transmission power difference doesn t 5

17 2 Theory present in the uplink, RSRP based cell selection only guarantees good downlink channel conditions. As illustrated in Figure 2.2: in the grey area, the macro node is selected based on the downlink RSRP, but for the uplink the LPN is better since the transmission power is the same and the pathloss is lower towards the LPN [3]. That is to say, a better cell selection for the uplink is minimum pathloss cell section [5]. Downlink RSRP cell boundary Uplink minimum pathloss cell boundary Biased RSRP cell boundary (RSRP + offset) Figure 2.2: Cell selection in 3GPP LTE HetNets As shown in Figure 2.2, the optimal cell boundary of downlink and uplink is not identical. To solve this problem, the RSRP cell boundary of the LPN should be extended. The most straightforward way is to extend the RSRP cell boundary by increasing the LPN transmission power, but this method reduces site availability since it affects the site size and the cost [3]. Another means without increasing the output power is to add an offset to the RSRP from the LPNs which will affect the cell selection and increase the pico cell range Biased RSRP cell selection [3][5][6], see Figure 2.2. Biased RSRP cell selection mechanism could of course improve the uplink performance. However, it causes higher downlink interference for users in the extended cell range area. Some interference management techniques have been developed to solve this problem such as enhanced Inter-cell Interference Coordination (eicic) [7]. Without this kind of interference management, there should be a tradeoff between downlink and uplink performance. Biased RSRP cell selection with a modest offset 3-4dB performs the best in many cases [6]. 6

18 2 Theory Two packet traffic models In the simulation of LTE networks, there are mainly two packet traffic models: full buffer model and equal buffer model. These two models are both implemented in the simulator used in this project Full buffer model Definition of full buffer: Static traffic: the number of active users in the system is fixed, no arrivals and departures; Infinite sessions: each active user s session lasts forever creating infinite data volume; Best effort: each active user fully utilizes the radio link. In full buffer model, on one hand, the radio links are always utilized since active users are fixed and last forever. On the other hand, users with different radio conditions spend same time in the system (infinite session), that is to say, users with poor radio conditions generate less data. As shown in Figure 2.3, user A and B are always sending data, as a result A and B contributes the same to the system throughput as illustrated in Figure 2.4. As a conclusion, full buffer model gives optimistic performance estimation which may deviate from reality. 7

19 2 Theory User A: 1 Mbps User B: 1 Mbps Figure 2.3: Users with different radio link condition Bitrate average t Figure 2.4: Average bitrate in full buffer model Equal buffer model Definition of equal buffer: Dynamic traffic: the number of active users in the system is not fixed, new session arrivals and complete session departures; Finite sessions: each active user s session ends when all data have been transmitted, all sessions have the same volume of finite data to send; Best effort: available link bitrate is utilized. In equal buffer model, on one hand, radio links will be idle when all users are inactive. 8

20 2 Theory On the other hand, users with different radio conditions have same finite data volume to send (finite session), that is to say, users with poor radio conditions spend more time in the system. As shown in Figure 2.3, user A and B are sending the same volume of data, say 1 Mbits, user A needs 1s and user B needs 1s. As a result, B contributes 1 times more than A to the system throughput as illustrated in Figure 2.5. As a conclusion, equal buffer model brings down the system capacity but it is more realistic. As a result, equal buffer model is chosen for this study and the simulation results shown in chapter 5 are all from equal buffer model. Bitrate average Figure 2.5: Average bitrate in equal buffer model t Comparison Some main differences between full buffer model and equal buffer model are listed in Table

21 2 Theory Full buffer Equal buffer Queue Volume per user Session time Fixed active users, no arrivals and departures Infinite Infinite Random dynamic active users, arrivals and departures Finite, fixed Depend on radio condition of the user Stability Always utilization < 1% User arrival No arrivals e.g., Poisson arrival Table 2.2: Comparison of Full and Equal buffer model 2.3 Related Tools As mentioned before, the study in this project will be taken based a real radio network in a limited urban area. To achieve this, several tools will be used during the whole process of this project TEMS CellPlanner In some studies of radio network, the simulations based on the calculated on-grid site locations and simplified propagation models are not sufficient compared with the realistic site data and propagation prediction. TEMS CellPlanner (TCP), a commercial cell planning tool, can be used to export such realistic data [9]. TEMS CellPlanner is a graphical PC-based application for designing, implementing, and optimizing mobile radio networks. It assists you in performing complex tasks, including network dimensioning, traffic planning, site configuration, and frequency planning, and network optimization [1]. TCP supports multiple mobile technologies such as GSM (Global System for Mobile Communications), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access) and WiMax. The main features of TCP that will be used in this project are pathloss analysis and Composite Pathloss Matrix (CPLM) analysis. The propagation models TCP used for these analyses are [11]: 9999 model Urban model Okumura-Hata model 1

22 2 Theory Walfish-Ikegami model Pathloss analysis in TCP is performed for each individual sector: a bin matrix is built for each sector with the pathloss from the sector antenna to each bin within the prediction radius set manually. And TCP gives pathloss predictions very close to realistic data based on the inputs of TCP such as [11]: Terrain data: elevation, etc. Land use data: buildings, trees, open area, etc. System data: technology, frequency, etc. Site data: site location, etc. Antenna data: height, gain, etc. CPLM analysis in TCP merges the pathloss results from individual sectors into a composite dataset based on the following parameters [11] [12]: Max. pathloss (db): largest pathloss which is considered to be relevant and allowed for the calculation. Any pathloss higher than this value is not included in the resulting CPLM calculation; Max delta pathloss (db): largest difference between lowest and largest pathloss values in the resulting CPLM calculation. The lowest pathloss value is searched. Any value higher than lowest pathloss + maximum delta pathloss is not included in the calculation; Max. number of cells: maximum number of cells from which the pathloss values are considered in CPLM calculation at any one bin. Another useful module of TCP is Automatic Cell Planning (ACP) which optimizes the network performance in a limited area by modifying the parameters of the sector antennas. ACP runs under some user-defined performance targets on coverage, quality and capacity, as well as some user-defined configuration constraints on the sector antennas, e.g., constraints on power, electrical tilt, mechanical tilt and azimuth. 11

23 2 Theory Elin Elin is a program package which can be integrated with the TCP installation. Elin is used to export system data, site data and calculated propagation data (see chapter 2.3.1) from TCP into some wireless network simulators (e.g., LTE Astrid which will be introduced in the following chapter) [9]. The exported data is stored in a so called proj mat-file (Matlab data file) created by Elin. The pathloss values are stored in a g-matrix in proj. The pathloss values are sorted to have the best values of each bin in the first column of the g- matrix, second best in the second column and so on. Another matrix, namely cellno-matrix, with the same size of the g-matrix contains the cell numbers corresponding to the pathloss values in the g-matrix [9]. These two matrixes together can be used by the wireless network simulators to calculate received signal strength, Signal to Interference plus Noise Ratio (SINR) and so on, and to do cell selection for each bin during the simulation LTE Astrid LTE Astrid is a static LTE network simulator in Matlab which conducts Monte Carlo method [13] in the simulation. The realistic network data exported from TCP via Elin can be analyzed with respect to capacity and coverage by LTE Astrid. The entire simulation in LTE Astrid comprises of several Monte Carlo runs. Each run is a network snapshot with randomized distributed users in the area [14]. The number and locations of the users follows the predefined parameters, i.e., traffic distributions and utilization. LTE Astrid simulates the network using RSRP based cell selection. With LPNs deployed, the simulator should be modified to apply Biased RSRP cell selection with an offset added to the RSRP from LPNs as discussed in chapter The two packet traffic models, full and equal buffer models, addressed in chapter 2.2 are both implemented in LTE Astrid. Achievable bitrate (ABR) is the main output from LTE Astrid Monte Carlo simulation which is the way to calculate system throughput (STP) for full and equal buffer model: Full buffer model estimate: STP FB = mean( ABR) u 12

24 2 Theory Equal buffer model estimate: STP EB = 1 / mean(1/ ABR) u STPFB STPEB Here, u means utilization which defines the probability that a cell has a user to schedule. In LTE Astrid, STP is actually calculated in a similar but more advanced approach. One problem in wireless network simulation is border effects, i.e., the number of cells in an area might be very large which makes the simulation of all the cells infeasible, but simulating a part of them would underestimate interference for the users in the border cells. One method to avoid this problem is called wrap-around which models the system as homogeneous and connects the edges of the simulation area in a torus fashion. But this method is not realistic in HetNets for analyzing real networks. Another solution is to analyze a small area while simulate a relatively larger area. The smaller area is denoted active area while the larger one is denoted supporting area. The cells in the active area are active cells while surrounding cells are supporting cells [14]. These two areas should be defined in TCP by polygons in advance and network data in the supporting area should be exported via Elin. 13

25 3 Models Models 3.1 Studied Network As mentioned before, the study will be taken based on a real radio network in a limited urban area in a dense European city as shown in Figure x [m] [m] x 1 5 Figure 3.1: Analyzed area The studied network models are built based on existing WCDMA 2.1 GHz site grid: Macro cells: 228 cells in total (83 sites) Macro inter site distance(isd): m 6% of study area is indoor area Macro sites average antenna height ~33 m 14

26 3 Models Macro sites average antenna electrical tilt: ~5.15 deg Macro sites average antenna mechanical tilt: ~1.46 deg Analyzed area: Area: ~1.4 km 2 Macro cells: 24 cells (9 sites) High resolution maps are used: 3D building data bases 5x5m bin resolution 3.2 Simulation parameters According to chapter 2.1.1, the choices of LPNs deployment in this project are listed in Table 3.1. Lamp post deployment means the LPNs are deployed several meters away from the buildings. And in the project, they are deployed 3 meters away. Access Open access Deployment Outdoor (lamp post deployment, ~3 m) Type of LPNs Frequency reuse Table 3.1: LPNs deployment RRU Reuse macro spectrum All the results demonstrated in this report, unless otherwise indicated, are obtained based on simulation parameters in Table

27 3 Models Parameter Carrier frequency Value 2.6 GHz Network Bandwidth Cell selection Traffic distribution 2 MHz Biased RSRP cell selection with 4dB offset (Chapter 2.1.2) 8% of the traffic is generated from indoor area (all buildings in the area), the rest 2% is distributed outdoor Utilization 1, 5, 1, 2, 3, 5, 7, and 95 % Macro Sites Pico Sites UE Packet traffic model Output power Average antenna height Average antenna gain Average antenna tilt Tx/Rx Diversity combining Output power Antenna height Antenna gain Antenna half power beam width (HPBW) Tx/Rx Diversity combining Max output power Min output power Antenna height Antenna gain Tx/Rx Equal buffer model 6 W ~33 m ~16 dbi ~5.15 o (Electrical tilt) ~1.46 o (Mechanical tilt) 2 Tx/2 Rx Maximal Ratio Combining (MRC) in uplink 5 W 5 m 12 dbi 63 o (Horizontal) 28 o (Vertical) 2 Tx/2 Rx directional antenna MRC in uplink 21 dbm -4 dbm 1.5 m -1 dbi 1 Tx/2 Rx omni antenna Body loss Diversity combining Table 3.2: Simulation parameters 3 db Interference Rejection Combining (IRC) in downlink 16

28 3 Models Propagation models Urban model As addressed in Chapter 2.3.1, there are several propagation models which are implemented in TCP. Since this study will be taken in an urban environment, the urban model is the one which will be mainly used for pathloss analysis. In an urban environment, radio wave propagation has two dominant paths which are over the rooftops and along the street. The first path dominants when the UE is far from the site, while the second path dominants when the UE is near to the site [15]. The urban model is valid under the following conditions [1][15]: Frequency from 45 MHz up to 22 MHz; Receiving antenna at distance to the base station antenna from m up to (at least) 5 km; Base station antenna heights between 5 m and 6 m and antennas placed below as well as above rooftops; Large receiving antenna height from 1.5 m up to 5 m. The urban model consists of three wave propagation algorithms [15]: Half-screen model: calculates propagation above the rooftop and generates pathloss Labove; Recursive micro cell model: calculates propagation between buildings, i.e. along the street, and generates pathloss Lbelow; Building penetration model: calculates propagation from an outdoor base station antenna to an indoor UE and generates pathloss Linside; The urban model pathloss is expressed as: L = min( L urban above, L below ) 17

29 3 Models As mentioned before, the radio waves propagation has two dominant paths and the received signal strength is the sum of both. In most situations, one of the paths will dominate, so Lurban which takes the minimum is justified [15]. The building penetration model pathloss is determined by [15]: L inside = Loutside + Lw + d sα in which Loutside is the pathloss from the base station antenna to a point just outside the external wall; Lw is the penetration loss through the external wall; ds is the distance inside building [m]; α is the building penetration slope [db/m]. In this project, Lw=12 db and α =.8 db/m ITU indoor propagation model ITU (International Telecommunication Union) indoor propagation model estimates the pathloss of radio propagation in the indoor environments. This model is applicable to frequency from.9 up to 5.2 GHz and to buildings with 1 to 3 floors [8]. According to ITU indoor propagation model, the indoor propagation pathloss is [8]: L = 2log f + N log d + P ( n) 28 indoor f, in which, f is the transmission frequency [MHz]; d is the transmission distance [m]; N is the distance power loss coefficient; Pf(n) is the floor loss penetration factor; n is the number of floors via transmission. 18

30 3 Models In this project, ITU indoor propagation model is applied in pico sites placement algorithm which will be addressed in chapter 4.2. It is just used for rough estimation of the pathloss from the newly added pico site. After pico cell placement is done, the pathloss will be calculated in TCP. Since only rough estimation is needed, the floor loss penetration factor Pf(n) is ignored here. And the distance power loss coefficient N is chosen to be 28. Taking into account of the exterior wall loss Lwall=12 since the pico sites will be placed outdoor, the total pathloss estimation can be expressed as: L = L + L = 2log f + 28log d 16 indoor wall 19

31 4 Implementation Implementation 4.1 Work flow 4 pico cells/macro cell 8 pico cells/macro cell 12 pico cells/macro cell... Desity TCP Project (WCDMA) Elin Project Pico Sites densification (Matlab) New Cells TCP ACP Elin Project LTE Astrid (Matlab) Simulation Results Performance Evaluation Figure 4.1: Work flow The whole project process is organized as Figure 4.1: 1. Export Elin project of the original network with macro cells only from TCP; 2. Based on the Elin project, densify the network with new pico sites in Matlab using the pico sites densification algorithm which will be discussed in chapter 4.2. Network densification should be done with several pico cell densities; 3. Import the new pico sites to the original TCP project, run ACP to optimize the network and export Elin projects of the networks after densification; 4. Run LTE Astrid for the networks with different pico cell densities; 5. Evaluate simulation results. 2

32 4 Implementation Pico sites densification algorithm As mentioned in chapter 1.4, how should the pico sites be deployed to achieve better network performance is of the interest of this project as well, a pico sites densification algorithm based on pathloss is designed. Briefly speaking, the core of the algorithm is as following: 1. Find the indoor bins with worst pathloss; 2. Find the closest outside-wall bin; 3. Find an outdoor bin 3 meters away from this outside-wall bin (lamp post deployment); 4. Place the pico site in this outdoor bin and point the sector antenna to the worst pathloss indoor bin. The details about how the algorithm works are illustrated in Figure 4.2: 1. Find the worst pathloss indoor bin based on the first column of the g-matrix from the Elin project file exported from TCP; 2. Find the closest outside-wall bin to this indoor bin, then find an outdoor bin 3 meters away from this outside-wall bin; 3. Check if this outdoor bin fulfills the distance constrain predefined, minpico2pico: if not, redo the previous steps for the next worst pathloss indoor bin; if yes, continue; 4. Place a pico site in this outdoor bin and point the antenna to the worst pathloss indoor bin currently considered; 5. Calculate pathloss vector for the indoor bins within 1 meters to the new pico site; 6. Compare the new pathloss vector with the first column of previous g-matrix and modify the pathloss values of the indoor bins covered by the new pico site; 7. Check if the picodense target is fulfilled: if not, redo the previous steps to find the next pico site position; if yes, stop searching. 21

33 4 Implementation Start Find the worst pathloss indoor bin Find the next worst pathloss indoor bin Place a pico in this outdoor bin Find the closest outside-wall bin Point the sector antenna to the worst pathloss indoor bin Find an outdoor bins 3 meters away from this outside-wall bin Calculate pathloss for the indoor bins within 1 meters to the new pico site 2 Modify pathloss values of the indoor bins covered by the pico site pico2pico no >minpico2pico 1 yes picodense 3 fulfilled? no yes End 1. minpico2pico: minimum distance between pico sites, a constraint to reduce interference between pico sites. 2. Pathloss calculated based on ITU indoor propagation model. 3. picodense: pico cell density target, e.g., 4 pico cells/macro cell. Figure 4.2: Flow chart of pathloss based pico sites densification algorithm In this project, the parameter minpico2pico has been chosen to be 3 meters. This might not be a perfect choice. And the parameter picodense has been chosen to be 4, 8, 12, 16 and 2 in order to build network models with different pico cell densities; an extreme case with picodense being 4 has also been studied. In the following chapter, the network performance will be compared between the network where pico sites are placed using this pathloss 22

34 4 Implementation based algorithm and the network where pico sites are placed randomly. In the randomly distributed algorithm, the pico sites are also deployed outdoor (lamp post deployment), but the placement positions are randomly selected without the constraint of the parameter minpico2pico. 23

35 5 Results Results In this chapter, the results of pico sites densification will be shown followed by the simulation results for the network models with different pico cell densities. 5.1 Network densification Figure 5.1 and Figure 5.2 illustrates the densified networks: the results of the pathloss based pico sites densification algorithm are shown in Figure 5.1 while the results of randomly distributed algorithm are presented in Figure 5.2. Compare these two figures, it can be seen that some of the pico sites in Figure 5.2 are very close to each other since there is no constraint on the minimum distance between pico sites (minpico2pico) in the randomly distributed algorithm. In Figure 5.1 and Figure 5.2, only network models with pico cell density up to 2 picos/macro are shown since the sites in the 4 picos/macro case are too dense and won t be very clear to be shown on the map here. 24

36 Macro only 4 picos/macro cell 5 Results picos/macro cell 12 picos/macro cell 16 picos/macro cell 2 picos/macro cell Figure 5.1: Densified networks using pathloss based pico sites densification algorithm (chapter 4.2) 25

37 Macro only 4 picos/macro cell 5 Results picos/macro cell 12 picos/macro cell 16 picos/macro cell 2 picos/macro cell Figure 5.2: Densified networks using randomly distributed pico sites densification algorithm (chapter 4.2) 26

38 5 Results Figure 5.3 shows the average cell density (per km 2 ) and average ISD of the networks where pico sites are placed using the pathloss based algorithm, while Figure 5.4 shows the average cell density and ISD of the networks where pico sites are placed randomly. Here one extreme case with pico density to be 4 picos/macro is also illustrated. As shown in these two figures, the average cell density presents no large difference; however the ISD of the pathloss based algorithm is longer since there is a constraint on the minimum distance between pico sites (minpico2pico) in this algorithm. 7 Average cell number per km 2 35 Inter site distance [m] Average cell number [/km 2 ] ISD [m] Pico cell density (picos/macro) Pico cell density (picos/macro) Figure 5.3: Pathloss based pico sites densification algorithm 7 Average cell number per km 2 35 Inter site distance [m] Average cell number [/km 2 ] ISD [m] Pico cell density (picos/macro) Pico cell density (picos/macro) Figure 5.4: Randomly distributed pico sites densification algorithm 27

39 5 Results Comparison of Pico sites densification algorithms In order to choose a better pico sites densification algorithm for the main parts of this project, in this chapter the network performance is compared between the pathloss based algorithm and the random one. Figure 5.5 and Figure 5.6 show the user performance (y-axis) versus capacity (x-axis) for both uplink and downlink. Two types of user performance measures are used: mean and 1% worst user throughput. In this project, the cell edge user throughput is defined as the 1th percentile of the user throughput in the analyzed area. Mean user throughput [Mbps] Uplink indoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Mean user throughput [Mbps] Uplink outdoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Throughput 1% worst uers [Mbps] Uplink indoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Throughput 1% worst uers [Mbps] Uplink outdoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Figure 5.5: Uplink comparison As illustrated in Figure 5.5, the two pico sites densification algorithms result in almost the same uplink performance for lower pico cell densities, e.g., 4 picos/macro. That is to say, for low pico cell densities the uplink performance is mainly related to the number of cells in the network rather than the position of cells. However for higher pico cell densities, e.g. 2 picos/macro, the pathloss based algorithm performs better than the random one according to Figure 5.5. As a conclusion, for the uplink, the advantage of the pathloss based algorithm compared 28

40 5 Results with the random one becomes clearer for higher pico cell densities. The difference between the algorithms increases with pico cell density. Mean user throughput [Mbps] Downlink indoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Mean user throughput [Mbps] Downlink outdoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Throughput 1% worst uers [Mbps] Downlink indoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Throughput 1% worst uers [Mbps] Downlink outdoor 4 picos/macro(pathloss based) 4 picos/macro(random) 12 picos/macro(pathloss based) 12 picos/macro(random) 2 picos/macro(pathloss based) 2 picos/macro(random) Figure 5.6: Downlink comparison Figure 5.6 illustrates the downlink performance comparison. It is obvious that the pathloss based algorithm results in better performance than the random one on the downlink. 29

41 5 Results Comparison of RSRP and Biased RSRP According to the results in chapter 5.2, the pathloss based pico sites densification algorithm will be chosen for the following simulation and discussion. As shown in Figure 5.7 and Figure 5.8, biased RSRP cell selection with 4dB offset presents a better improvement than RSRP based cell selection on both uplink and downlink performance. But according to some previous studies, biased RSRP cell selection with higher offset will degrade the downlink performance. Mean user throughput [Mbps] Uplink indoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Mean user throughput [Mbps] Uplink outdoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Throughput 1% worst uers [Mbps] Uplink indoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Throughput 1% worst uers [Mbps] Uplink outdoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Figure 5.7: Uplink comparison 3

42 5 Results Mean user throughput [Mbps] Downlink indoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Mean user throughput [Mbps] Downlink outdoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Throughput 1% worst uers [Mbps] Downlink indoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Throughput 1% worst uers [Mbps] Downlink outdoor 4 picos/macro(rsrp+4db) 4 picos/macro(rsrp) 12 picos/macro(rsrp+4db) 12 picos/macro(rsrp) 2 picos/macro(rsrp+4db) 2 picos/macro(rsrp) Figure 5.8: Downlink comparison 5.4 Network Performance Based on the comparisons in chapter 5.2 and 5.3, the following simulation results are obtained from the network models where pico sites are placed using the pathloss based algorithm, and biased RSRP cell selection with 4dB offset has been applied during the simulation. In this chapter, the uplink and downlink performance for the indoor and outdoor users will be analyzed separately. 31

43 5 Results Capacity and Coverage As illustrated in Figure 5.9 and Figure 5.1, for the uplink, with the same target of mean user throughput or the same target of cell edge user throughput, the subscriber capacity keeps increasing with pico cell densification. Mean user throughput [Mbps] Uplink indoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Mean user throughput [Mbps] Uplink outdoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Mean user throughput [Mbps] Uplink all area (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Figure 5.9: Uplink mean user throughput vs uplink subscriber capacity 32

44 5 Results Throughput 1% worst uers [Mbps] Uplink indoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Throughput 1% worst uers [Mbps] Uplink outdoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Throughput 1% worst uers [Mbps] Uplink all area (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Figure 5.1: Uplink cell edge user throughput vs uplink subscriber capacity Figure 5.11 illustrates one example of the improvement of uplink capacity: with the same target of 1 Mbps uplink cell edge user throughput, uplink capacity improves with pico cell densification. For the network with pico cell density less than 4 picos/macro, the cell edge throughput can never reach 1 Mbps, thus the capacity is shown to be. 33

45 5 Results Uplink subscriber capacity [GB/month/subscriber] Uplink cell edge throughput=1mbps, RSRP+4dB offset Pico cell density (picos/macro) Figure 5.11: Improvement of uplink capacity From Figure 5.1, it is also possible to see that with the same target of uplink capacity, the cell edge user throughput grows with densification, that is to say the uplink coverage of the network has been improved. An example is shown in Figure 5.12 with a target of 3.1 GB/month/subscriber uplink capacity, the improvement of cell edge user throughput is significant especially for the indoor users which presents an exponential growth manner. For the uplink, pico cell densification benefits the indoor users much more than the outdoor users from the coverage point of view. 34

46 5 Results Uplink indoor, Capacity=3.1GB/month/subscriber, RSRP+4dB offset 4 Uplink outdoor, Capacity=3.1GB/month/subscriber, RSRP+4dB offset 3 35 Cell edge throughput [Mbps] Cell edge throughput [Mbps] Pico cell density (picos/macro) Pico cell density (picos/macro) 5 Uplink all area, Capacity=3.1GB/month/subscriber, RSRP+4dB offset Cell edge throughput [Mbps] Pico cell density (picos/macro) Figure 5.12: Improvement of uplink coverage According to Figure 5.13 and Figure 5.14, the same result with the uplink can be drawn for the downlink. Compare Figure 5.15 with Figure 5.11 and Figure 5.16 with Figure 5.12, the enhancement of the downlink capacity and coverage from pico cell densification is not as significant as the uplink capacity and coverage. Furthermore, the capacity in Figure 5.15 seems to increase linearly; and the increase of cell edge user throughput in Figure 5.16 slows down with higher pico cell densities which indicates an upper limit. 35

47 5 Results Mean user throughput [Mbps] Downlink indoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Mean user throughput [Mbps] Downlink outdoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Mean user throughput [Mbps] Downlink all area (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Figure 5.13: Downlink mean user throughput vs downlink subscriber capacity 36

48 5 Results Throughput 1% worst uers [Mbps] Downlink indoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Throughput 1% worst uers [Mbps] Downlink outdoor (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Throughput 1% worst uers [Mbps] Downlink all area (RSRP+4dB offset) Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 2 picos/macro Figure 5.14: Downlink cell edge user throughput vs Downlink subscriber capacity Downlink subscriber capacity [GB/month/subscriber] Downlink cell edge throughput=1mbps, RSRP+4dB offset Pico cell density (picos/macro) Figure 5.15: Improvement of downlink capacity 37

49 5 Results Cell edge throughput [Mbps] Downlink indoor, Capacity=7.8GB/month/subscriber, RSRP+4dB offset Cell edge throughput [Mbps] Downlink outdoor, Capacity=7.8GB/month/subscriber, RSRP+4dB offset Pico cell density (picos/macro) Pico cell density (picos/macro) Downlink all area, Capacity=7.8GB/month/subscriber, RSRP+4dB offset 4 35 Cell edge throughput [Mbps] Pico cell density (picos/macro) Figure 5.16: Improvement of downlink coverage User throughput From Figure 5.9 and Figure 5.1, it can be seen that both indoor and outdoor users uplink throughput is improved when more pico cells are added to the network. With the same target of 3.1 GB/month/subscriber uplink capacity, the uplink user throughput maps for the networks with different pico cell densities are shown in Figure As can be seen, there is some noticeable improvement from the original case to the 4 and 8 picos/macro cases; after that, the throughput improvement is not very clear according to the throughput maps. 38

50 5 Results Uplink Capacity=3.1GB/month/subscriber, picos/macro Uplink Capacity=3.1GB/month/subscriber, 4 picos/macro (Mbps) > Uplink Capacity=3.1GB/month/subscriber, 8 picos/macro Uplink Capacity=3.1GB/month/subscriber, 12 picos/macro (Mbps) > Uplink Capacity=3.1GB/month/subscriber, 16 picos/macro Uplink Capacity=3.1GB/month/subscriber, 2 picos/macro (Mbps) > Figure 5.17: Uplink user throughput maps Figure 5.18 illustrates the average uplink user throughput within the analyzed area. It can be seen that, the gain for indoor users in general is larger than for outdoor users. 39

51 5 Results Uplink indoor, Capacity=3.1GB/month/subscriber, RSRP+4dB offset 5 Uplink outdoor, Capacity=3.1GB/month/subscriber, RSRP+4dB offset 45 Mean user throughput [Mbps] Mean user throughput [Mbps] Pico cell density (picos/macro) Pico cell density (picos/macro) 35 Uplink all area, Capacity=3.1GB/month/subscriber, RSRP+4dB offset 3 Mean user throughput [Mbps] Pico cell density (picos/macro) Figure 5.18: Improvement of uplink mean user throughput From Figure 5.13 and Figure 5.14, it can be seen that both indoor and outdoor users downlink throughput is improved when more pico cells are added to the network. Figure 5.19 shows the downlink user throughput maps with the same target of 7.8 GB/month/subscriber downlink capacity. It is obvious that the improvement is significant from the reference case to the 4 picos/macro case. This dramatic increase of mean user throughput can also be noticed in Figure 5.2. After 4 picos/macro, the enhancement is not so noticeable. As illustrated in Figure 5.21, the downlink geometry keeps being strengthened with the increase of pico cell density, but the increase slows down with higher density. 4

52 5 Results Downlink Capacity=7.8GB/month/subscriber, picos/macro Downlink Capacity=7.8GB/month/subscriber, 4 picos/macro (Mbps) > Downlink Capacity=7.8GB/month/subscriber, 8 picos/macro Downlink Capacity=7.8GB/month/subscriber, 12 picos/macro (Mbps) > Downlink Capacity=7.8GB/month/subscriber, 16 picos/macro Downlink Capacity=7.8GB/month/subscriber, 2 picos/macro (Mbps) > Figure 5.19: Downlink user throughput maps 41

53 5 Results Downlink indoor, Capacity=7.8GB/month/subscriber, RSRP+4dB offset 8 Downlink outdoor, Capacity=7.8GB/month/subscriber, RSRP+4dB offset Mean user throughput [Mbps] Mean user throughput [Mbps] Pico cell density (picos/macro) Pico cell density (picos/macro) Downlink all area, Capacity=7.8GB/month/subscriber, RSRP+4dB offset 8 7 Mean user throughput [Mbps] Pico cell density (picos/macro) Figure 5.2: Improvement of downlink mean user throughput Geometry distribution,downlink Capacity=7.8GB/month/subscriber, RSRP+4dB) 1 Macro only.9 4 picos/macro 8 picos/macro.8 12 picos/macro 16 picos/macro.7 2 picos/macro.6 CDF Geometry [db] Figure 5.21: Downlink geometry distribution 42

54 5 Results An extreme case In order to answer the question in chapter 1.4: Is there any upper limit on how many pico cells can be added to a dense urban macro network and still improves the network performance? An extreme network model with 4 picos/macro has been investigated, both the pathloss based pico sites densification algorithm and the randomly distributed algorithm have been studied; the results are shown in Figure 5.22 and Figure As can been seen from the figures, for both the two algorithms, the network keeps performing better with higher pico cell density up to 4 picos/macro. Figure 5.22 also confirms the result in chapter 5.2 which is for the uplink the pathloss based algorithm performs better than the random one for higher pico cell densities. The difference between the algorithms increases with pico cell density. Mean user throughput [Mbps] Uplink indoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Mean user throughput [Mbps] Uplink outdoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Throughput 1% worst uers [Mbps] Uplink indoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Throughput 1% worst uers [Mbps] Uplink outdoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Figure 5.22: Uplink throughput vs capacity 43

55 5 Results Mean user throughput [Mbps] Downlink indoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Mean user throughput [Mbps] Downlink outdoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Throughput 1% worst uers [Mbps] Downlink indoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Throughput 1% worst uers [Mbps] Downlink outdoor (RSRP+4dB offset) Macro only 2 picos/macro(pathloss based) 4 picos/macro(pathloss based) 2 picos/macro(random) 4 picos/macro(random) Figure 5.23: Downlink throughput vs capacity Figure 5.24 illustrates the average uplink utilization of pico and macro cells separately with the same target of 1 Mbps uplink cell edge user throughput; while Figure 5.25 illustrates the average downlink utilization of pico and macro cells separately with the same target of 1 Mbps downlink cell edge user throughput. From these two figures, it can be seen that with the target of 1 Mbps uplink and 1 Mbps downlink cell edge user throughput, even for the densest network model with 4 picos/macro, the average utilization of pico and macro cells is below 5%. The probability of interference will hence be rather low. Also, pico cells have much lower utilization than macro cell since each pico cell covers smaller area. In the figures, the capacity is not the same for each case since the pico cell densities are different and the cell edge user throughput is kept the same. 44

56 5 Results Uplink cell edge throughput=1mbps, RSRP+4dB offset.45 Uplink cell edge throughput=1mbps, RSRP+4dB offset.16.4 Average uplink utilization of pico cells Average uplink utilization of macro cells Pico cell density (picos/macro) Pico cell density (picos/macro) Figure 5.24: Average uplink utilization of pico and macro cells with the target of 1 Mbps uplink cell edge user throughput.2 Downlink cell edge throughput=1mbps, RSRP+4dB offset.4 Downlink cell edge throughput=1mbps, RSRP+4dB offset Average downlink utilization of pico cells Average downlink utilization of macro cells Pico cell density (picos/macro) Pico cell density (picos/macro) Figure 5.25: Average downlink utilization of pico and macro cells with the target of 1 Mbps downlink cell edge user throughput 5.5 Comparison of pico and macro cells served area In order to see the level of pico cell densification in this project, Figure 5.26 shows the pico cells served area (green area) and the macro cells served area (red area) separately; Figure 5.27 compares these two areas. As can be seen from Figure 5.26 ad Figure 5.27, with an extensive number of pico cells, the macro cells will cover diminishing area. And for the 4 picos/macro case, the macro cells will run mostly idle since they cover really small fraction of the area. This is definitely not realistic in real networks, but this indicates that the study has reached an extreme level with very high pico cell density but the network performance still seems to keep being improved. 45

57 5 Results picos/macro cell 8 picos/macro cell 12 picos/macro cell 16 picos/macro cell 2 picos/macro cell 4 picos/macro cell Pico cells Macro cells Figure 5.26: Pico and Macro cells served area 46