CELL ZOOMING TECHNIQUES FOR POWER EFFICIENT BASE STATION OPERATION. A Thesis. Presented to the. Faculty of. San Diego State University

Transcription

1 CELL ZOOMING TECHNIQUES FOR POWER EFFICIENT BASE STATION OPERATION A Thesis Presented to the Faculty of San Diego State University In Partial Fulfillment of the Requirements for the Degree Master of Science in Electrical Engineering by Ramapriya Balasubramaniam Summer 2012

2

3 iii Copyright 2012 by Ramapriya Balasubramaniam All Rights Reserved

4 iv DEDICATION This thesis is a dedication to my family.

5 v ABSTRACT OF THE THESIS Cell Zooming Techniques for Power Efficient Base Station Operation by Ramapriya Balasubramaniam Master of Science in Electrical Engineering San Diego State University, 2012 With the enormous growth in telecommunication industry, energy efficiency has become a critical issue. Base stations account for a significant portion of the energy budget of a cellular network. Traditional energy saving techniques switch some base stations off completely during light loads to save energy. This creates problems for the backhaul network and also for quickly returning to full capacity when demand increases. In this paper, we propose a novel technique called cell zooming to reduce energy consumption at base stations. With cell zooming, base stations dynamically adjust their coverage radius and hence their transmit powers based on user locations. The transmit power is set to the minimum required level depending on user locations and signal to interference and noise ratio (SINR) required. Base stations are never completely switched off. Simulations show that the proposed cell zooming algorithm reduces the energy consumption of base stations by up to 40% without compromising the quality of service (QoS) for users compared to traditional static-coverage-area base stations.

6 vi TABLE OF CONTENTS PAGE ABSTRACT...v LIST OF TABLES... vii LIST OF FIGURES... viii ACKNOWLEDGEMENTS...x CHAPTER 1 INTRODUCTION FUNDAMENTALS OF CELL ZOOMING PROPOSED CELL ZOOMING TECHNIQUES Definition of Parameters Continuous Cell Zooming Method Discrete Cell Zooming Method Linear Division Method (LDM) Equal Area Division Method (EADM) Equal Power Division Method Fuzzy Cell Zooming Method Framework of Discrete Cell Zooming Implementation of Discrete Cell Zooming SIMULATION RESULTS AND ANALYSIS CONCLUSION...30 REFERENCES...31

7 vii LIST OF TABLES PAGE Table 4.1. System Parameters...22

8 viii LIST OF FIGURES PAGE Figure 2.1. Scenario with BS 'E' supporting two voice users and the corresponding optimal network reconfiguration after cell zooming....6 Figure 2.2. Scenario II depicting BS E supporting one voice and one video user and the corresponding network configuration after cell zooming....7 Figure 3.1. Definition of parameters for path loss model Figure 3.2. Linear division method Figure 3.3. Equal area division method Figure 3.4. Equal power divison method Figure 3.5. Fuzzy discrete model Figure 3.6. Framework of cell zooming...15 Figure 3.7. Cell zooming operation in cellular networks Figure 3.8. Flowchart represents discrete cell zooming methods Figure 3.9. Flowchart to calculate BS transmit power Figure 4.1. Ratio of dynamic over static power for continuous, discrete and fuzzy algorithms Figure 4.2. Ratio of dynamic over static power for different values of the number of discrete levels Z...23 Figure 4.3. Ratio of dynamic over static power for continuous and discrete cell zooming algorithms Figure 4.4. Ratio of dynamic to static power consumption versus average inter-arrival time of times...25 Figure 4.5. Average received SINR with and without cell zooming as a function of the number of users Figure 4.6. Ratio of dynamic over static power for 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for discrete algorithm Figure 4.7. Ratio of dynamic over static power for 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for fuzzy algorithm Figure 4.8. Average received SINR vs 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for Equal Power Division Method....28

9 Figure 4.9. Ratio of dynamic over static power for 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for fuzzy algorithm with more zones ix

10 x ACKNOWLEDGEMENTS It is a pleasure to thank all who have made this thesis possible. Foremost, it is difficult to put in words my gratitude towards my thesis advisor, Dr. Mahasweta Sarkar. Without her support, advice, great explanation skills, guidance and continued patience, this would not have been possible. I would like to take this opportunity to thank Dr. Santosh Nagaraj and Dr. Vuskovic Marko for having co operated and making this thesis successful. Thanks to all my friends, for the informal support and help I received in the form of discussions through sleepless nights, insightful comments, challenging questions and support when I was lost. Lastly and importantly, I would like to thank my family for providing me moral, emotional and financial support to make this possible.

11 1 CHAPTER 1 INTRODUCTION In recent years Green Communication, an energy efficient communication in cellular networks is of more concern to network operators because of the rising energy costs and carbon foot print on earth. This trend stimulated the interest in researchers to design, develop and investigate the various metrics, models and associated science needed for combined energy efficiency and network optimization. Nowadays, the whole world of telecommunications and information communities is facing a more and more serious challenge, namely on one side the transmitted multimediarich data are exploding at an astonishing speed and on the other side the total energy consumption by the communication and networking devices and the relevant global CO 2 emission are increasing terribly. Research shows the importance of energy efficiency metrics which are indicators of efficiency, understanding those metrics provides us a better view on how energy efficiency can be achieved in wireless systems/networks. Learning these metrics will help us build a better understanding on energy consumption problems. The study of green communications will require investigation in several areas such as power efficient RF hardware, efficient MAC protocols, networking, and integration of renewable energy with communications equipment, frequency reuse deployment strategies, and spectrum policy. While each area individually contributes to energy consumption, researching the interaction across separate layers will provide the truly transformational discoveries. Today, with the enormous growth of communication and wireless technologies, energy consumption has increased globally to a great extent. The unexpected increase in energy consumption and mobile communication made the cellular industry to deploy more than 4 million base stations for mobile users, each consuming about 25Mwh per year and the number of base stations are expected to be double by 2012 [1]. Research shows that 3% of the world s annual electrical energy consumption and 2% of CO 2 emissions are caused by the information and communication technology (ICT) infrastructure [2].

12 2 Studies show that for the first time in history worldwide data traffic was more dominant than the voice traffic. In 2020, mobile data is expected to dominate all mobile traffic with a decreasing share of voice and is predicted to account for more than 10 percent of all IP traffic in 2020, mostly due to the smart phones [3]. In addition to the environmental costs, energy costs increases the overall operational costs for the network operators. Typically to connect a base station to electrical grid will cost about $3000 per year to operate [4]. In 2010, it corresponded to 60 billion kwh of electricity usage and about 40 million metric tons of CO 2 emissions each year. Most of the cellular operators focus on fulfilling the needs of the consumer, exploiting the available energy resources. An increase of the global number of mobile subscribers will increase the energy consumption of the networks. In recent years with the growth in new technologies such as Android and iphone devices, the use of ipad and kindle demanded for increase in data usage. By designing energy efficient base stations, an economical solution to the energy shortage problem is provided while contributing to a greener environment simultaneously. Prior research has focused on energy efficiency of electrical power generators and other supporting systems for Base Stations such as cooling and heating systems. It has been estimated that every year, about 120,000 new Base Stations are deployed to serve millions of new consumers around the world which significantly increase the carbon footprint even more on earth [2] given that Base Stations contribute to about 60%-80% of the total energy consumption. Moreover, it is estimated that ICT energy consumption is rising at 15-20% per year, thereby doubling every five years [5]. A significant portion of the operational expenditure of a cellular network goes to pay the electricity bill. It has been estimated that the mobile network budget for electricity globally is more than 10 billion dollars today [6]. Implementation of green communication protocols globally at base stations not only plays a vital role in energy conservation but it is also an economically significant issue. Hence cellular network operators have been exploring ways to increase energy efficiency in all components of cellular networks, including mobile devices, base stations and core (backhaul) networks. The need to develop green wireless communication systems turns out to be more and more urgent as wireless networks are becoming ubiquitous [3]. This realization has led to a push towards green wireless

13 3 communications that strives for improving energy efficiency as well as reducing environmental impact. Network planning, cell size and capacity are usually fixed based on the estimation of the peak traffic load. However traffic nature in the network can vary by both spatial and temporal fluctuations and bursty in nature for many data applications. The traffic load in the network depends on the time of the day, location and many other factors. If the cell size is fixed, then some cells will always operate under heavy or light load. Clearly to ensure scenario-specific end user demands and to satisfy the increasing demands of the users, the required wireless resources need to be consumed in a energy efficient way. The increase use of resources for maximizing QOS performance is beneficial for wireless users but harmful to the environment. Therefore, a reasonable way for operators is to support the On-Demand service for users. Supporting the On-Demand service for users not only satisfies the requirement of users, but also consumes the minimum resources which may minimize the CO2 emission and the cost of operators. We propose one such On- Demand technique to dynamically adjust the cell sizes depends on the distance of the user. To conserve energy at the base station, many switching on/off schemes have been proposed in the literature [7]. Switching off a few cells in the network reduces the power consumption of the overall network. In this paper, we propose a new concept called cell zooming. Cell zooming is an adaptive method in which the cell sizes, i.e., coverage radius of the base stations are dynamically reconfigured. Base stations adjust their transmit power to cater to the current traffic profile (user locations and data rates) with minimal power consumption. This is in contrast to the static cell size technique which consumes maximum power at all times [8]. Cell zooming algorithms help in achieving greener communication in a cost efficient way, since the base station transmits at the minimum required power level by dynamically switching to the minimum required cell radius. Simulation results show that cell zooming leads to substantial energy savings of up to 40% while maintaining coverage during off-peak hours. Furthermore, when demand increases, the base station can quickly return to its full coverage state. In this paper, we focus on three different cell zooming algorithms: continuous, discrete and fuzzy methods. We compare and contrast the power efficiencies of these three cell zooming techniques. However, there are many challenges in the deployment of energy efficient network such as

14 4 network architecture, avoidance of electromagnetic interference and requirement of power efficient devices for operation. It is important to note here that cell zooming is different from power control. With power control, transmit powers for individual users are varied while maintaining a static coverage radius [9]. Furthermore, power control is rarely implemented on the downlink. The base station always transmits at the maximum power level, i.e., static transmission parameters. Cell zooming offers a useful balance between a static maximum power transmission strategy and switching base stations off completely. This paper is organized as follows. In Chapter 2, we introduce the Fundamentals of cell zooming. In Section 3, we discuss the system model in detail along with the three proposed cell zooming techniques. In Section 4, we present and analyze simulation results that validate our claims of power savings in cellular networks. We finally conclude the paper in Section 5.

15 5 CHAPTER 2 FUNDAMENTALS OF CELL ZOOMING Energy consumption has become one of the most important issues in the world, as large number of base stations contribute to major energy consumption. To overcome this problem, a new concept cell zooming is introduced in this paper. Cell Zooming has the capacity to dynamically adjust the cell size without switching off or putting the base station to sleep completely. When a base station is in its working mode, the energy consumption of the processing circuits and coolers will take up to 40% of the total consumption. Therefore by merely controlling the transmit power, the effect of energy saving is marginal and Switching on/off schemes will add more overhead to the network, which leads to increase in energy consumption. Obviously, cell could not to zoom to infinity due to constraint of power. Depending on the traffic load and distant user location, base station transmit power can be calculated and cells will be zoom in or zoom out based on the transmission power [10]. Cell zooming is a load balancing scheme that can be used to satisfy user requirements as far as possible. Intuitively, Cell Zooming technique can be applied only during the low traffic period, because at full load cells will zooms out to the maximum or transmit at maximum power. Since receive power falls rapidly with distance from the base station, the smaller the cell size, the lower should be the total power consumption of the access network. Exploiting the spatial and temporal fluctuation of traffic through a 24-hour period in any cellular network, as an effective means of scavenging energy is a concept that has not received its due attention from the research community [11]. In this paper, we address that very issue. We utilize the traffic fluctuation of a network at different times of the day to our advantage and propose a cellular network infrastructure which exploits this traffic fluctuation to harvest significant energy conservation an idea and concept that has not been explored previously. Some of the cellular network energy saving schemes which advocate BSs to be shut off depending on traffic profile do so by inherently assuming that every user is running the same application on their cell phones and hence has the same QoS requirements.

16 6 Moreover, the specific location of users has largely been ignored in framing these optimization problems [12]. In contrast, we emphasize that every user is unique in terms of its location and QoS requirements. Thus a currently active BS with resources to accommodate say one user, may or may not be able to accommodate a user being currently served by another BS at its required QoS level if the first BS continues to operate at its current system parameters. Now let us consider a specific traffic scenario during off-peak hours. Let us especially focus on cell site E which has only two voice users in its cell, located as denoted in Figure 2.1. Let us further assume that BS A and D have adequate resources to accommodate a voice user each. Moreover, from an energy perspective, if BSs A and D need to hike up their transmission powers to serve these users and if the total increase in power (not just in terms of increased coverage but also in terms of QoS) is less than the power required to keep BS E active and of service to these users then it is justified to make BS E shrink its cell size to zero (thereby consuming very low power) and have BSs A and D increase their cell size and capacity to accommodate these users. This scenario helps us recognize the potential for energy saving when employing cell zooming. Figure 2.1. Scenario with BS 'E' supporting two voice users and the corresponding optimal network reconfiguration after cell zooming. Let us consider another scenario as depicted in Figure 2.2. The spatial distribution of the users in the cell sites is identical to that in Figure 2.1 but now one of the two users in cell site E is a video user. Under the resource and power constraints required to support a video

17 7 Figure 2.2. Scenario II depicting BS E supporting one voice and one video user and the corresponding network configuration after cell zooming. user, it might be the case that the network configuration solution will be different though the number and location of the users in the network might not have changed. This might be because BSs A and D already support video users and may not have available resources to accommodate the QoS requirement of another resource intensive application like video. On the other hand, BS B which currently serves only voice users might have resources available to support both the video and voice users originally covered by BS E. Thus the network will be reconfigured in a way where BS B would expand its cell size to accommodate the users of BS E who can then shrink its cell size to zero. The other BSs might also shrink their cell sizes to various degrees so that the network can operate with the minimal possible energy. Some of the cellular network energy saving schemes which advocate BSs to be shut off depending on traffic profile does so by inherently assuming that every user is running the same application on their cell phones and hence has the same QoS requirements [12]. These algorithms are solely based on the number of users that a base station serves. Moreover, the specific location of users has largely been ignored in framing these optimization problems. In contrast, we emphasize that every user is unique in terms of its bandwidth and other QoS requirements. Thus a currently active BS with resources to accommodate say one user, may or may not be able to accommodate a user being currently served by another BS at its required QoS level if the first BS continues to operate at its current system parameters.

18 8 To illustrate with an example, let us assume that BS A has resources to accommodate one user engaged in a voice call. Let us further assume that BS B is currently serving only one user who is engaged in a streaming video application. Current methodologies to reconfigure the cellular network will propose that BS B be shut off and BS A takes over the user served by BS B. However, BS A clearly cannot satisfy the QoS requirements of the streaming video user. This aspect has been ignored in existing work but something that had addressed in this paper, In traditional planning, cells cannot be configured dynamically based on the resource requirements but in cell zooming guarantee of coverage and quality of service can be achieved by adjusting cell configurations to extend coverage [8]. Cells cannot zoom to maximum as there is a constraint with transmit power. After dividing the cell radius into discrete level, cells can zoom to the particular discrete level based on the user distribution.

19 9 CHAPTER 3 PROPOSED CELL ZOOMING TECHNIQUES We consider a cell with a single BS which can transmit at the maximum power of over a maximum coverage radius of. We assume that all the users in the cell require the same data rate. Further, the distribution of the users is considered to be uniform over the cell area and the traffic arrival pattern is assumed to be Poisson distributed over the time period for which we study the system. We next describe three algorithms for cell zooming. 3.1 DEFINITION OF PARAMETERS The two basic propagation models (free space loss and plane earth loss) are the mostly used predictive models to calculate the path loss [1]. However, to use such models would require detailed knowledge of the location, dimension and constitutive parameters of every tree, building and terrain feature in the area to be covered. According to design perspective the overall coverage area is more important than considering the individual area which is complex. One appropriate way of accounting for these complex effects is via an empirical model [6]. To create such a model, an extensive set of actual path loss measurements is made, and an appropriate function is fitted to the measurements, with parameters derived for the particular environment, frequency and antenna heights so as to minimize the error between the model and the measurements. The following terms will be used in defining path loss models as shown in Figure 3.1: h m - mobile station antenna height above local terrain height [m], often taken as1.5 m d m - distance between the mobile and the nearest building [m] h 0 - typical (usually the mean) height of buildings above local terrain height [m] h b - base station antenna height above local terrain height [m] r - circle distance between base station and mobile [m]

20 10 Figure 3.1. Definition of parameters for path loss model. The simplest empirical propagation path loss model is given by: or in decibels L = 10nlogr+K (3.1) Where and are the effective isotropic transmitted and predicted isotropic received Powers. L is the path loss; r is the distance between the base station and the mobile. The path loss at a point is defined as the ratio of transmitted power at r 0,,over the received power at r,. For free-space propagation, the path loss can be simply expressed as: L db 10log 10log λ (3.2) π Where and are the gains of the transmitting antenna (T x ) and receiving antenna (R x ), respectively, d is the distance between T x and R x, and is the wavelength in free space. K = -10log10k and n are constants of the model. Parameter k can be considered as the reciprocal of the propagation loss that would be experienced at one meter range (r = 1m). Models of this form will be referred to as power law models: (3.3) Where & are the feeder losses at the transmitter and receiver and constant From Equation (3.1) and (3.3): (3.4)

21 11 (3.5) Where k is the constant P r /c. The parameter n is called the path loss exponent (typically 2-4) and depends on the channel model. 3.2 CONTINUOUS CELL ZOOMING METHOD Continuous method of cell zooming technique is based on a BS transmitting at the power level that is just adequate to reach its farthest user. Thus a BS dynamically grows (up to ) or shrinks (potentially to zero) its cell radius to just accommodate the farthest user within its boundaries. With cell zooming, we assume to be a fixed required value for the farthest user (at a distance ) and the BS required transmission power is therefore given by: (3.6) The transmitter power is proportional to the nth power of the distance of the farthest user in the cell. Continuous transmit-power adaptation is the most energy efficient cell zooming approach, but, implementation of this method is challenging because of high user mobility and consequent stringent location feedback requirements. 3.3 DISCRETE CELL ZOOMING METHOD With discrete zooming, the BS transmit power is chosen from only a discrete set of allowable values. The cell area is divided into Z number of zones with being the radius of the discrete zone and i ranging from 1 to Z. Suppose that the farthest user is located between two discrete levels and 1. The BS chooses to transmit power based on the higher discrete level of radius 1 to provide coverage to all users including the edge users in that particular zone. The advantage of discrete cell zooming is in the reduced location feedback requirements. The mobile user need not report its location information to the BS until it crosses one discrete zone to the next. By increasing the number of zones Z, the BS increases its energy savings at the cost of increased feedback complexity. We propose three different methods of obtaining the allowable discrete radius values r(i). A detailed description of each method is discussed in the following sections.

22 Linear Division Method (LDM) In linear division method, the maximum cell radius is divided into equally spaced levels. In this method, the area of each zone (and hence the number of users covered) increases with the level i as shown in Figure 3.2. The discrete level of radius using LDM is calculated by using the equation: (3.7) The transmission power of the BS using Linear Division Method can be calculated using the equation: (3.8) Here is the power to be transmitted by the BS using a linear division of the cell and is the radius of the zone with the farthest user. Figure 3.2. Linear division method Equal Area Division Method (EADM) In this method, the cell is divided into Z discrete zones with equal area in each of the zones as shown in Figure 3.3. Since the area covered by each zone is equal, the user distribution will be equal (on average) in every zone. In this method, the discrete level of radius r(i) is obtained by using the equation: (3.9) The transmission power of the BS using EADM can be computed using the equation: (3.10) where, r(i x ) is the radius of the zone with the most distant user. Since this method divides the cell into equal area zones, radii of the first few zones will be more when compared to the

23 13 Figure 3.3. Equal area division method. Linear Division Method and hence power consumption is also more when the farthest user is still not far from the BS Equal Power Division Method In this method, the cell is divided into discrete zones with equal BS transmission power increments for each zone. The transmission power for the zone is equal to /. The area of each zone is not the same as compared to the Equal Area Division Method as shown in Figure 3.4. Since we have used a path loss exponent n=4. From the propagation equation (3.1), the discrete level of radius is calculated by using the equation: / / (3.11) The transmission power of the BS using Equal Power division method can be calculated using the equation: (3.12) where, is the radius of the zone with the most distant user. In this method, the number of zones increases as we move towards to the edge of the cell. The area of the first zone is higher when compared to the Equal Area Division Method and so there is an increase in level of power consumption when compared to the other methods at low demand. 3.4 FUZZY CELL ZOOMING METHOD Fuzzy cell zooming method is an extension of the discrete cell zooming method with a small (about 10% to 20%) increase in the range of coverage at each discrete level and a slight compromise in received signal to interference and noise ratio (SINR) for the users

24 14 Figure 3.4. Equal power divison method. located beyond the corresponding discrete level of radius. Fuzzy cell zooming technique is based on checking the boundary conditions of the users along with an extension of coverage range as explained next. In Figure 3.5, the shaded portion shows the fuzzy region with a radius of about 10% or 20% in excess of the corresponding. In fuzzy cell zooming method, when the farthest user is located within the specified range of the current discrete level, the BS chooses to transmit power at the current discrete level instead of transmitting at the next higher discrete level of radius 1. The transmission power of the BSs using fuzzy model can be calculated using the equation: (3.13) Here, is the power transmitted by BS using fuzzy model and is the radius of the zone which contains the farthest user within its 10% (or 20%) fuzzy region. Fuzzy cell zooming method performs better than the discrete cell zooming method since the BS transmits at the current discrete level of radius instead of switching to the next higher level. The received SINR, however, is slightly lower than the desired value in the fuzzy region and hence has to be compensated for by using more powerful error correction coding techniques. 3.5 FRAMEWORK OF DISCRETE CELL ZOOMING Implementation of cell zooming in the mobile and wireless world needs some changes in the current architecture and addition of new components for cell zooming. The framework of cell zooming is shown in Figure 3.6. There is a server at the centre which

25 15 Figure 3.5. Fuzzy discrete model. Figure 3.6. Framework of cell zooming.

26 16 controls the network and coordinates between the network information and network operation. The channel first sense the network information such as user location, channel conditions, division method used for discrete cell zooming, percentage of fuzzy region and number of levels required. Decision making and analysis in the network for cell zooming either can be implemented as centralized or distributed in the BSs.After collecting all the sensing information, the central server will analyze and gives feedback to the network. The server will provide the discrete level of radius and fuzzy percentage region information to the Base station. Then the BS will coordinate with the network and zoom in or zooms out to the corresponding discrete level of radius provided. Base station will adjust the transmit power according to the corresponding discrete level of radius. Power transmitted by the base station varies and depends on the type of discrete division method used. The network with more data users requires more power compared to the normal distribution of voice users. Many techniques can be used to implement cell zooming as described in [13]. Energy consumption in cellular networks can be achieved by zooming in or zooming out the cells based on the current traffic distribution. By adjusting the transmit power of the base station, power can be saved and can be varied dynamically. Methods like physical adjustment and relaying to increase the transmit power requires high overhead and implementation will be very complex. When a Base station is working in sleeping mode, other energy consuming equipments can be switched off and energy can be saved, but the possibilities of signal coverage in that area is less which will lead to call drops and handover. Base station cooperation is an effective technique in which BS s will cooperatively transmit and receive from mobile units (UE). The cell size will be larger compare to the normal cell size, since the cell size is the sum of the original size of the BS s in cooperation. Cells can zoom in or zoom out to improve the coverage but the overhead will be more which will increase the design complexity. In the proposed discrete cell zooming method, cell zooms in or zooms out based on the network information which will reduce the overhead by reducing the frequent switching between the discrete levels. The fuzzy region adds more flexibility to the design and user

27 17 located in that region will also experience the same power level instead of switching to the next higher level. Consider a scenario as shown below in Figure 3.7 which shows the cell zooming operation in cellular networks. Let us assume that mobile unit A is located at the inner discrete level of radius, user B is located at the interior fuzzy region of the extreme discrete level of radius and user C is located at the exterior fuzzy region. A BS B r2 r1 C Figure 3.7. Cell zooming operation in cellular networks. Initially the entire coverage area is divided into different discrete level using linear division, equal area or equal power method. Base station will transmit at the power level of zone 1 which is calculated by using the discrete level of radius r1 to user A even though the user is beyond the discrete level. Since the user A is in the fuzzy region of discrete zone 1, BS station can reach the user A with a small compromise of upto ±3db of the actual power. Though the user A receives lesser SINR compare to other users inside the zone 1, power will be saved by avoiding the switching to the discrete level of radius r2. Similarly user B receives a power lesser than the power level of zone 2 which is calculated by using the discrete level of radius r2, since the user is in the interior fuzzy region and the power level will be approximately -3db lesser than the actual power level

28 18 transmitted. Base station will transmit at the power level of zone 2 to the user C which is calculated by using the discrete level of radius r2. User C is located beyond the discrete level r2 but within the fuzzy region, so BS station will transmit at the same power level instead of switching to the next higher level. The area of fuzzy region can be varied and depends on the user distribution. By increasing the fuzzy region, power savings will be more with a little compromise on the SINR. Cell zooming can provide various benefits in cellular networks. Cell zooming can be used for load balancing by transferring traffic from cells under heavy load to cells under light load. Secondly, cell zooming can be used for energy saving. Resources can be allocated to satisfy the user requirements and based on the distribution of users. Cell zooming will improve the network throughput, coverage, supports more user, reduces the unnecessary switching of BS power transmission and so on. Since the frequent upgrade of the network can be avoided, the network operational cost will be reduced. Cell zooming is different from other power control techniques, because it focuses on saving network power consumption of the whole network rather than controlling the transmit power. There exists many challenges in implementing cell zooming. To make the cell zooming more efficient, the accurate information of traffic load and distribution is very important, since there will be spatial and temporal fluctuations in the network. Current cellular networks are not designed to support cell zooming, so implementation of cell zooming might have compatibility issues. It requires a change in the existing architecture and components used. Sometimes cell zooming might produce coverage holes when cells zoom in or zoom out. Cell zooming technique cannot be used during the high traffic load in the network. 3.6 IMPLEMENTATION OF DISCRETE CELL ZOOMING There are many steps involved in the implementation of cell zooming. Flowchart shown in Figure 3.8. represents the ways to divide the entire coverage area into discrete levels by using three different methods. Detailed description of each step is as follows: Step1: Initialize with the coverage area r max, number of zones, type of division method used and fuzzy percentage region and radius of the distant user r dist. Step2: Choose the type of mode to be used (i.e) type of division method used to divide the coverage area.

29 Figure 3.8. Flowchart represents discrete cell zooming methods. 19

30 Step3: Calculate the radius of the each discrete level using various discrete division methods. The number of discrete levels depends on the number of zones to be divided. Step4: Check for the condition that if the radius of the distant user is within the discrete level say (i), then radius of the discrete level will be equal to the radius of the user. Step5: If the mode used is fuzzy, then check whether the radius of the distant user is greater than the radius of the current discrete level say (i) and lesser than next discrete level (i+1) in addition to the fuzzy boundary region. If it is true then the radius of the user will be equal to the radius of the next discrete level. Step6: If it is not true then check whether the radius of the distant user is greater than the radius of the next discrete level say (i+1) and lesser than next discrete level (i+1) in addition to the fuzzy boundary region. If the condition is true then the radius of the user will be equal to the radius of the next discrete level otherwise radius will be equal to next higher discrete level. Step7: As shown in Figure 3.9 the average number of users in the network can be calculated based on the simulation time used. The random distribution of both the data and voice users are computed. Step8: The Poisson distribution can be computed based on the inter arrival time and number of users in the network. The arrival call arrival instance, hold time and termination are computed based on the Poisson distribution. Step9: If the user arrival time is greater and the termination time is lesser than the simulation time, then BS will provide service to that user. Step10: BS transmit power can be computed based on the rzone value calculated in Figure 3.8. using discrete and fuzzy cell zooming methods. The value of the rzone various for different type of division method used and hence the BS transmit power varies. 20

31 Figure 3.9. Flowchart to calculate BS transmit power. 21

32 22 CHAPTER 4 SIMULATION RESULTS AND ANALYSIS The proposed cell zooming methods were evaluated in a scenario with Poisson distributed user traffic for both voice and data users. The inter arrival time, which is the average time between the arrival two successive users, is varied dynamically based on the average number of users per hour. The hold time of the users is assumed to be Gaussian distributed with mean µ and standard deviation σ and it is the time duration during which the user stays in the network. For the fuzzy algorithm, it is assumed that the BS can extend its coverage by 10% or 20% of the specified range. The maximum cell radius of the BS and the various simulation parameters are summarized in Table 4.1. Table 4.1. System Parameters Maximum cell radius 500m Simulation time 1 Hour Inter-arrival time 360s Mean Hold Time (µ) 100s Std. Dev. Of Hold Time (σ) 20s Number of zones 5 Fuzzy range Variable (10% to 20%) Figure 4.1. Shows the variation of the ratio of dynamic to static power consumption with the average number of users per hour for cell zooming based on continuous, discrete, and, fuzzy method based on discrete boundaries. Results show that the fuzzy method performs slightly better than the discrete method, since in the fuzzy method, switching to the higher discrete level is avoided by accommodating the users which are above the specified discrete level but within the specified excess range of 10%. Our simulation results showed that the loss in SINR for the boundary users was about 1 db. Figure 4.2. Shows the variation of dynamic to static power consumption with an increase in the number of discrete zones. As expected, our results show that with an increase in the number of zones, the BS requires lesser transmitter power. With increase in the number of discrete zones, the discrete zooming method gets closer to the continuous method.

33 Ratio of dynamic power over static power Continuous vs discrete radius Avg number of users continuous discrete fuzzy Figure 4.1. Ratio of dynamic over static power for continuous, discrete and fuzzy algorithms Continuous vs various no of zones 0.8 Ratio of dynamic power over static power Continuous 4 zones 6 zones 8 zones Avg number of users Figure 4.2. Ratio of dynamic over static power for different values of the number of discrete levels Z.

34 24 However, as the number of zones increases, location feedback information will need to be sent more frequently as the mobile users will cross the zone boundaries more frequently. In Figure 4.3, the ratio of power consumed by a dynamic (cell zooming) base station to a static (traditional) base station has been plotted against the average number of users for various cell zooming methods. Results show that the continuous method performs the best as expected, but, with a significant feedback overhead. Among the discrete methods, Equal Power Division Method is the most energy efficient. It consumes higher power than the continuous method but is easier and much more flexible to implement. The user needs to provide just one bit feedback to the BS as to whether the zone index i has increased or decreased due to mobility. The base station will adjust its transmission parameters accordingly. 1 Continuous vs discrete Radius 0.9 Ratio of dynamic power over static power Continuous Linear Divsion 0.1 Equal area Division Equal power Division Avg number of users Figure 4.3. Ratio of dynamic over static power for continuous and discrete cell zooming algorithms. Figure 4.4. Shows the ratio of dynamic power to static power consumption as the inter-arrival time of users is varied while maintaining hold time statistics at µ=100 s and σ=20 s. Once again, cell zooming benefits are best seen at lighter loads when the inter-arrival times are high.

35 Ratio of dynamic power over static power Continuous vs discrete radius Inter-arrival time continuous discrete fuzzy Figure 4.4. Ratio of dynamic to static power consumption versus average inter-arrival time of times. Figure 4.5. Shows the variation in the average received SINR as a function of the number of users. The static BS inefficiently provides a much higher average SINR than what is required as the BS always transmits at full power and users that are not on the cell edge always receive SINR much higher than that required. The dynamic BS efficiently provides a lower average SINR but still above that which is required. Figure 4.6. Shows the variation of dynamic over static power consumption for different combination of voice and data users for discrete algorithm. In this scenario, the network is considered with 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users. Since the received power requirement of data users is more for 1/3voice+2/3data user combination in the network consumes more power than the other combination of voice and data users. Figure 4.7. Shows ratio of dynamic over static power for different combination of voice and data users for fuzzy algorithm. The power transmitted for the combination of 2/3voice+1/3 data users in the network using fuzzy method is more when compared with the 25

36 26 60 Users vs SINR SINR (db) dynamic discrete fuzzy static Threshold Avg number of users Figure 4.5. Average received SINR with and without cell zooming as a function of the number of users /2voice+1/2data 1/3voice+2/3data 2/3voice+1/3data Discrete Method 0.8 Ratio of dynamic power over static power Average number of users Figure 4.6. Ratio of dynamic over static power for 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for discrete algorithm.

37 /2voice+1/2data 1/3voice+2/3data 2/3voice+1/3data Fuzzy Method 0.8 Ratio of dynamic power over static power Average number of users Figure 4.7. Ratio of dynamic over static power for 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for fuzzy algorithm. discrete method, since the frequent switching between the zones is minimized for the variation in user location and received SINR requirement. The SINR provided to the data users is more compared to the voice users, since they require more power for data usage. Though the power transmitted is high but it is still within the minimum required power level and satisfies user requirements. By increasing the percentage of fuzzy region, better results can be achieved. Figure 4.8. Shows the plot between the average received SINR and number of users for equal power division method. The average received SINR gradually increases as the number of users in the network increases. Since the requirement of SINR is more data users, network with 1/2voice+1/2data and 1/3voice+2/3data user distribution receives almost same level of SINR. The flexibility of the fuzzy discrete algorithm makes the network to perform better as well as satisfy user SINR requirements. As the number of users increases the average

38 28 70 Equal Power Division Method(fuzzy) SINR(dB) /2voice+1/2data 1/3voice+2/3data 2/3voice+1/3data Average number of users Figure 4.8. Average received SINR vs 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for Equal Power Division Method. received SINR becomes constant at a certain level and remains constant even though the user count increases. Figure 4.9. Shows the plot for fuzzy discrete algorithm with more number of zones. The number of zones in the network can be varied and depends on the user distribution and traffic load. To support more number of users in the network, the total coverage area can be divided into more number of zones. As the number of zones increases, the average required power to transmit is less and network throughput increases. With the increase in the number of zones, fuzzy region increases and more power can be saved. Though the result shows that power consumption is lesser for continuous method, discrete and fuzzy methods also perform comparatively better with less overhead. The coverage holes are minimized with less switching between the discrete zones. The results show that proposed method performs better than the static algorithm. The proposed cell

39 /2voice+1/2data 1/3voice+2/3data 2/3voice+1/3data Equal Power Division Method(fuzzy with more no of zones) 0.8 Ratio of dynamic power over static power Average number of users Figure 4.9. Ratio of dynamic over static power for 1/2voice+1/2data, 1/3voice+2/3data, 2/3voice+1/3data users for fuzzy algorithm with more zones. zooming method is more flexible as they can freely leverage the trade-off between energy consumption and overhead.

40 30 CHAPTER 5 CONCLUSION In this paper, we proposed a novel concept called cell zooming which dynamically adjusts the transmission power and hence coverage area of the base station depending on the location of the farthest user. From an implementation perspective, cell zooming is much simpler to implement in existing systems than base station switching off/on to conserve energy. The base station can quickly return to full coverage and capacity when demand increases. Three different cell zooming methods were suggested and compared for performance and complexity. Our results showed that nearly 40% reduction in power consumption can be saved at the base stations with cell zooming which can achieve green efficient communication in cellular networks.

41 31 REFERENCES [1] S. Bhaumik et al., Breathe to stay cool: Adjsuting cell sizes to reduce energy consumption, in Proc. ACM Mobicim, Special Workshop on Green Networking, New Delhi, India, 2010, pp [2] A. Amanna et al., Metrics and Measurement Technologies for Green Communications. Gaithersburg, MD: National Institute of Standards and Technology, [3] Z. Hasan et al., Green cellular network : A survey, some research issues and challenges, IEEE Commun. Surveys & Tutorials, vol. 13, pp. 1-16, Sept [4] A. Fehske et al., The global footprint of mobile communication: The ecological and economic perspective, IEEE Commun. Mag., vol. 49, pp , Aug [5] G. Fettweiss and E. Zimmermann, ICT energy consumption-trends and challenges, in Proc. of IEEE WPMC, Lapland, Finland, 2008, pp [6] D. Willkomm et al., Primary user behavior in cellular networks and implications for dynamic spectrum access, IEEE Commun. Mag., vol. 47, no. 3, pp , Mar [7] X. Weng et al., Energy-efficient cellular network planning under insufficient cell zooming, in Proc. 73rd Vehic. Technol. Conf., Budapest, Hungary, 2011, pp [8] S. Zhou et al., Green mobile access network with dynamic base station energy saving, in Proc. ACM Mobicom, Beijing, China, 2009, pp [9] J. T. Louhi, Energy efficiency of modern cellular base stations, in Proc. of IEEE INTELEC 07, Rome, Italy, 2007, pp [10] G. Miao et al., Energy-efficient transmission in frequency-selective channels, in Proc. IEEE Globecom 2008, Las Vegas, NV, 2008, pp [11] O. Arnold et al., Power consumption modeling of different base station types in heterogeneous cellular networks, in Proc. ICT Mobile Summit, Florence, Italy, 2010, pp [12] E. Oh and B. Krishnamachari, Energy savings through dynamic base station switching in cellular wireless access networks, in Proc. IEEE Globecom, Miami, FL, 2010, pp [13] Z. Niu et al., Cell zooming for cost-efficient green cellular networks, IEEE Commun. Mag., vol. 48, no. 11, pp , Nov