MASTER'S THESIS. Improved Power Control for GSM/EDGE

Transcription

1 MASTER'S THESIS 2005:238 CIV Improved Power Control for GSM/EDGE Fredrik Hägglund Luleå University of Technology MSc Programmes in Engineering Department of Computer Science and Electrical Engineering Division of Signal Processing 2005:238 CIV - ISSN: ISRN: LTU-EX--05/238--SE

2 Improved Power Control for GSM/EDGE Fredrik Hägglund 11th February 2005

3 Abstract In GSM (Global System for Mobile Communication), a tradeoff between different goals is necessary to achieve the optimal system performance. Generally, high speech quality, high capacity and low power consumption are major goals. Power control is one of several techniques used to reach these goals. Power control regulates the signal strength with the aim to reduce the overall interference. Since radio environment and mixture of different user requirements may vary, there is an interest in making the setting of the power control target value automatically and dynamically. The target value should adapt to the environment and the situation. The method used is to extend the current power control with an outer loop that could adjust the quality parameter qdes. An attempt to use EMR data to adjust qdes is shown to have effects that eliminate the essential principle of power backoff to avoid the so called party effect, and is therefore not recommended. However, a method using the distribution of the transmitted power indicates more promising results. It is shown that there is high correlation between the number of satisfied users and the number of users within the regulating window, i.e. the number of users not limited by the maximum or minimum power levels.

4 Acknowledgement This Master s Thesis is the final part of my Master of Science Degree in Signal Processing at Luleå University of Technology. The work has been carried out at Ericsson Research in Luleå during fall First I would like to thank my supervisor Magnus Thurfjell at Ericsson for great guidance and valuable support during the Thesis. I would also like to thank my examiner James LeBlanc at Luleå University of Technology for his work. i

5 Contents 1 Introduction General Purpose Method Delimitations Outline Background Evolution of cellular networks The GSM network Radio network Impairments to radio transmission Fast fading Path loss Shadow fading Time dispersion Co-channel interference Power control General Current power control algorithm EMR Problem statement Simulator description Simulation model Basic functionality Propagation model Fast fading Frequency hopping Discontinuous transmission Quality estimation ii

6 5 Tested algorithms General Parameter settings Outer loop based on EMR Outer loop based on frame erasure rate Outer loop based on coefficient of variation Summary Outer loop based on power distribution Power distribution analysis Simulated algorithm Discussion Conclusions Further studies A List of Abbreviations 42 iii

7 List of Figures 2.1 Simple schematic view over the GSM hierarchy A 4/12 frequency reuse pattern. The gray cells are using the same frequency group and hence mobiles from these cells using the same frequency could interfere with each other Transmission power p 1 and p 2 from base stations to mobile stations. Each mobile experience a carrier signal power C and interference I Principle for down regulation. The values rxlev and rxqual are the measured values before any exponential filtering Schematic view of the FHRUNE simulator Format of the frequency hopping matrix The steps in the mapping from C/I to FER The lookup table for mapping mean and standard deviation to FEP shown as a figure The existing power control in GSM extended with the outer loop Block scheme for the existing power control algorithm extended with an outer loop based on EMR The outer loop when the control algorithm is based on FER Satisfied users at different traffic loads BER and CV of BER is plotted against each other. Every dot gives the BER and CV of BER for each mobile at each measurement time. An approximation of a level curve for FER_target = 1% is also in the figure A plot over how CV_BEP changes around the average line for one single mobile during a call. The call lasts for nearly 7 seconds and it is impossible to see a trend for the CV_BEP values Histograms for two different occasions. In the left plot the system has experienced a relatively low traffic load and in the right a relatively high traffic load iv

8 5.8 Number of satisfied users for a specific condition for different qdes The usage of power control and the amount of satisfied users plotted for different qdes. The traffic load is fixed and low The plot shows the regulating fraction of the transmitted powers and the number of satisfied users for a high traffic load The plot shows the regulating fraction of the transmitted powers and the number of satisfied users for a high traffic load The difference between number of mobiles transmitting on maximum effect and the number of mobiles transmitting on minimum effect for some different traffic loads A block scheme of the outer loop based on the usage of power control The qdes value is adjusted to a proper value, depending on the current environment in the cell The number of satisfied users with and without an outer loop in the power control. The result with no power control is also displayed v

9 Chapter 1 Introduction This section will give a brief introduction to the subject handled in this work. The purpose, method and the delimitations are also in this section. Finally, an outline for the report is presented. 1.1 General GSM (Global System for Mobile Communication) is a mobile system that is globally used. In GSM, and in other traditional mobile systems a tradeoff between different goals is necessary to achieve the optimal system performance. Generally, high speech quality, high capacity and low power consumption are major goals in cellular radio communication systems. Power control is one of several techniques used to achieve these goals. Power control regulates the signal strength to reduce the overall interference. One of the parameters that control the power level are a target value controlling the desired quality. Instead of setting this target value for each separate case, for example different radio environment or speech codecs, there are some benefits in making the control of this target value automatic. How the target value should be set may depend on many variables, such as the interference from other cells, the distance between the mobile station and the base station, the traffic load, the total amount of radiated power etc. Therefore, there is an interest in making the setting of the target value automatic and dynamic. 1.2 Purpose The purpose of this work is to find out whether a method, based on the additional information from for example the new standardized measurement report, could improve the current power control by reducing the hard parameter predefinitions. This could be done by making it possible to set the target value automatically. The method should improve, or at least preserve the 1

10 performance of the existing power control. The results of the work includes recommendations on algorithm design, parameter tuning and comments of performance in general. 1.3 Method The method chosen to fulfill the purpose was to automatically adjust the target value in an outer loop, using additional information from for example the enhanced measurement report. Initially some analysis was done, principally about how information concerning speech quality should be used, but also what kind of information to use. Several different fundamental algorithms have been developed and a simple evaluation determines whether the algorithms have been excluded or if they were relevant for further analysis. A model has been made for each of the relevant algorithms, which shows how the additional information controls the desired target value. Adjusting the signal strength to track this target value sets the desired speech quality. Input values, which are decided from the outer loop, that adjusts the signal strength have been tested and evaluated. This means that simulations have been done to verify the choices of input values in different situations. It should be easy to change the input values. The simulation tool that has been used is an advanced existing simulator at Ericsson Research called FHRUNE. 1.4 Delimitations The work was delimited to only look at an outer loop that controls the quality target value, which is an input to the existing power control algorithm. This means that there are possibilities that other solutions could give better performance for the outer loop. But such algorithms have not been examined in this work. Another limitation was to only look at speech transmission and not packet data transmission. 1.5 Outline In chapter two the background to the subject will be presented. Chapter three describes the existing power control algorithm, and a problem statement that will more precisely describe the problem. In chapter four the simulation model is described and how the simulations where done to model the different approaches of the outer loop will be presented in chapter five. The results of the simulations and an analysis of the results will also be presented in chapter five. Finally, in chapter six a discussion is presented with the conclusions and some ideas about future work. 2

11 Chapter 2 Background The background theories about mobile systems, the GSM system, radio network and impairments to radio transmission are presented briefly in this section to give an initial understanding. 2.1 Evolution of cellular networks The first generations cellular networks (1G) were analogue systems launched in the beginning of the eighties. Examples are NMT450 and NMT900 (Nordic Mobile Telephone), TACS (Total Access Communication System) in United Kingdom and AMPS (Advanced Mobile Phone Service) in USA and Canada. All these systems where using Frequency Division Multiple Access (FDMA). In FDMA each user has a dedicated frequency and hence only one user per channel is allowed. In the beginning of the nineties the second generation cellular systems (2G) were developed. Examples of systems are GSM (Global System for Mobile Communication), D-AMPS (Digital-Advanced Mobile Phone Service) in America and PDC (Personal Digital Cellular) in Japan. The 2G systems were improved with both digital transmission and Time Division Multiple Access (TDMA). In TDMA users are sub-divided into a number of time slots for each carrier frequency. For example, the GSM system uses both FDMA and TDMA, where users are still assigned to a discrete slice of the frequency spectrum, but these are divided into eight time slots (full-rate GSM). GSM was designed for voice, but with data capabilities. [2] The third generation cellular system (3G) was launched first in G are designed for data, but with voice capabilities and allow different multimedia services at very high bit rates. GPRS (General Packet Radio Services) is a developed version of GSM and the first step towards the third generation. Other systems developed against 3G are GSM/EDGE (Enhanced Data rates for GSM Evolution) and WCDMA (Wideband Code Division Multiple Access). 3G are planned to be the key system for future multimedia 3

12 communication worldwide. 2.2 The GSM network The GSM system was introduced to the European market in It is the most widely spread of the 2G standards. GSM is active on the frequency band around 900 MHz. As said before, GSM uses both FDMA and TDMA. The GSM system is using a number of 200 khz FDMA channels, each one divided into eight TDMA timeslots for speech transmission [4]. Each of the eight time slots has duration of ms. Binary Gaussian Minimum Shift Keying (GMSK) is used as the modulation technique, which provides one information bit per symbol. This could be compared with 8 PSK (Phase Shift Keying) that is used in for example EDGE that provides three information bits per symbol. EDGE is developed from GSM and GPRS to support much higher data rates. [2] In the transmitter, the analog speech is digitized and divided into segments of 20 ms, called a frame. GSM digitize the speech by using an 8 khz sampling frequency and 8 bits resulting in a bit rate of 64 kbps. The bit stream has been compressed and quantized by using a mix of a vocoder and a waveform coder, resulting in a 13 kbps bit stream. There are also some burst formatting, channel coding and interleaving before the information is transmitted on a time slot. A sequence of bits sent during one time slot is called a burst and each speech frame is spread over eight bursts. The GSM radio network is built up with Base Transceiver Stations (BTS) in a cellular structure. The BTS consists of a group of transmitters and receivers that communicate with the mobile stations (MS) located in the cell that it controls. The communication to different mobile stations takes place in different channels, divided up as described above. The next level in the network is the Base Station Controller (BSC), which communicates with one or more BTS. In other words, the BTS has the radio equipment for example antennas and transmitters that makes it possible for the BSC to communicate with the mobile stations. The BSC is in charge of hand overs, what transmit powers to use and other higher level tasks. The next step in the traditional GSM hierarchy is the Mobile services Switching Center (MSC). The MSC sets up, supervises and releases calls. This is a big switch that interfaces several BSC, via a Gateway MSC (GMSC), to other telephony and data systems such as PSTN (Public Switched Telephone Network) or other networks. In Figure 2.1 there is an overview of the GSM hierarchy. [5] 4

13 PSTN GMSC MSC BSC MS MS BTS BTS MS MS Figure 2.1: Simple schematic view over the GSM hierarchy. 2.3 Radio network The basic components in a radio network are the mobile station, the base station that communicates with the mobile station, and the mobile switching center, which sets up, controls and releases calls. All base stations control different areas, also known as cells. Each cell has one base station that mobile stations in that cell could connect to. However, several base station are usually placed at one single site depending on the cellular structure. Normally a cell structure of three base stations per site is used. A cell can be of any size from a radius of tens of meters to a radius of tens of kilometers. [1] The transmission between the base station and the different mobiles in a cell takes place in a set of channels, for example based on the number of frequencies. Mobile systems only have a limited amount of bandwidth. Each user requires a certain amount of that bandwidth. If each frequency only were used once, also the number of users would be limited. This means that 5

14 frequencies used in one cell need to be reused in another cell at a certain distance away. Users that use the same frequency will interfere with each other. A group of cells that together uses all the available frequencies in the system is called a cluster of cells. Clusters repeated over and over again forms a cellular network. A schematic illustration of a cellular network is seen in Figure 2.2, where the cells are represented as hexagons for simplicity. The frequencies used in a cluster are divided into frequency groups. How many frequencies there are in each group is dependent on the total number of available frequencies and the required reuse factor. Different reuse patterns can be formed but some examples are 4/12 or 3/9. For example, 4/12 means that all the available frequencies are divided into 12 frequency groups, one for each cell, which is located at 4 sites with 3 base stations each. In a system like this the reuse factor is 12. As seen in the figure below each base station site has three cells and is using directional antennas. [2] Figure 2.2: A 4/12 frequency reuse pattern. The gray cells are using the same frequency group and hence mobiles from these cells using the same frequency could interfere with each other. It is preferable to keep the number of base stations down, to decrease costs of the BTS hardware and expensive sites when establishing new base stations. The existing base stations should be used in the most efficient way to avoid establishing new ones, including considerations about the limited amount of bandwidth. The base stations that use the same set of frequencies should be placed at sufficient distance apart from each other to decrease interference. However, to keep the capacity at an acceptable level the base stations have to be relatively close to each other. So there will always be some interference. The mobile stations in the gray cells in the Figure 2.2 above could experience co-channel interference because of the reuse of frequency carrier. Cells that simultaneously uses the same carrier frequency interfere with each other if they are close enough. To measure the amount of interference at 6

15 a connection one measurement that could be used is the carrier to interference ratio (C/I). This measurement is the relation between the carrier signal power C and the interference power I. The carrier signal power, in db, for one connection is defined as C(t) = p(t) + g(t), (2.1) where p(t) is the transmitted power in the downlink and g(t) is the (negative) gain between the mobile and the base station. The interference I(t) contains both of the interference carrier power from other base stations and the thermal noise, see [3]. However, the interfering power from neighboring cells with the same carrier frequencies are above the noise floor. This means that the system is interference limited rather than noise limited. The C/I, in db, for each mobile is defined as C/I = C(t) I(t) = p(t) + g(t) I(t). (2.2) Suitable ranges for this parameter are decided by the desired quality profile. For example acceptable speech quality or required data bit rate. However, a high C/I-value corresponds obviously to a high quality in the specific connection [3]. 2.4 Impairments to radio transmission The problem with radio transmission is that it is impossible to control the transmission environment. The impairments are known, but their effect as a function of time is unpredictable and hence, it is difficult to accurately model the transmission dynamically. There are several factors that affect radio transmission conditions, in addition to noise. Some of the problems with radio transmission are described below [5] Fast fading One effect that could occur is that the reflected signals could arrive at the receiver with such an unfavourable phase that they cancel out each other. This cancelling effect, which is also referred to as fast fading or Rayleigh fading, is however dependent on the position of the transmitter or receiver, so that often it is already sufficient to change position by less than half a wavelength. Moreover, fading also depends on the transmitter frequency which influences the phase of the signals at the reception site. This means that the fading dips will appear at different places for different frequencies. This fading effect is most common where the reflecting is high. 7

16 2.4.2 Path loss The amplitude of a signal diminishes when a signal travel further away from the transmitter. This phenomenon is referred to as path loss. Path loss can make it difficult to get enough signal strength in a large cell. An advantage with path loss is the natural decrease in interference from other cells. This is the principle on which the reuse in a cellular system is built Shadow fading Another fading impairment, except for fast fading, is shadow fading or sometimes called slow fading or log normal fading. Obstacles, that are shadowing the radio path between the transmitter and receiver, will cause slow variations in signal strength. Anything interrupting the free line of sight could be considered as an obstacle Time dispersion A reflecting object far away, such as a mountain, reflects signals and the mobile station will hence receive both a direct radio signal and a fairly strong reflected signal from the reflecting object. These two signals will arrive at different times which could cause individual bits to overlap with each other and disturb the overall received signal. This effect is called Inter Symbol Interference (ISI) Co-channel interference Distant radio transmitters transmitting on the same frequency as the one used by a special radio link will disturb. Even if they are very far away and their amplitude has been attenuated due to path loss a disturbing effect will be noticed in the receiver. This impairment is known as co-channel interference and is an important part in mobile networks. Co-channel interference is described more deeply in other parts in this report. 8

17 Chapter 3 Power control This chapter describes power control in general and the existing power control algorithm. A description of the EMR (Enhanced Measurement Report) is also presented. The chapter ends with a discussion about the problem and how the problem might be solved. 3.1 General Power control (PC) refers to the strategies or techniques required to adjust the transmitted power. Power control regulates the transmitted power to achieve a desired signal strength. A mobile far away from a base station requires a stronger transmitted signal than a mobile close to a base station. If the speech quality is better than necessary for one mobile the signal strength for that specific mobile will be decreased. This implies that the system will be improved, because of the reduction in interference. The single mobile will also experience a decrease in battery consumption when transmitting to the base station. This is the main idea with power control. Power control is used both in uplink and in downlink between base station and mobile. In this work only downlink is described. The principle is however equal. The transmission power p from the base station to the mobile, Figure 3.1, should be controlled to optimize the system. The power should be high enough to achieve a sufficient carrier signal power C at the mobile station and low enough to minimize interference I at other mobiles. The transmitted powers from base stations to mobiles are controlled by the power control algorithm developed for the GSM network. 9

18 p 1 p 2 C C I I Figure 3.1: Transmission power p 1 and p 2 from base stations to mobile stations. Each mobile experience a carrier signal power C and interference I. A common strategy to utilize available resources in cellular radio systems is therefore to control the transmitter powers described above. The motivation is to maintain an acceptable quality throughout the lifetime of a connection, in other words an acceptable C/I. That is, the aim with power control is to increase the number of mobile stations with a C/I on an acceptable level. The power control will optimize the transmitted power, and thus increase the number of satisfied users if traffic is maintained, or keep the the number of satisfied users if traffic is increased [6]. When power control is used the total amount of radiated power is reduced compared to when it is not used. Keeping an acceptable level of quality should hold despite varying channel conditions and presence of disturbing interference from other users. The existing power control algorithm considers the system quality and not the quality for single mobiles as the main regulate factor. This means that some single mobiles have to accept slightly worse quality to improve the total system quality. However, when applying power control to real systems, some challenges are prevalent. Available information in measurement reports is crude, highly quantized and constrained to physical limits. So one challenge is to issue relevant power levels based on this information to obtain an acceptable quality. [3] 3.2 Current power control algorithm In the power control algorithm, quality and signal strength is both considered. Bad quality as well as low signal strength will increase the transmitted power. For each measurement period (480 ms) two variables, rxqual and rxlev, are reported based on measurement from all bursts during that measurement period. The variables rxqual and rxlev stands for received quality and received signal strenght. These variables are used to adjust the transmitted power. Predefined values for quality respectively signal strength defines values for controlling rxqual and rxlev in the regulation process. The 10

19 variables, rxqual and rxlev are filtered with nonlinear exponential filters in order to eliminate variations of temporal nature. The measurement reporting causes a delay that typically is three periods. The controlling parameter for rxqual in the regulation is qdes, and for rxlev the parameter ssdes. The qdes value is the target value that specifies the desired quality. Internally, qdes and rxqual are converted to C/I-valus, expressed in db according to Table 3.1. Linear interpolation is used to realize C/I. Table 3.1: The conversion between qdes, rxqual and C/I. The unit dtqu stands for deci-transformed quality units. qdes [dtqu] rxqual C/I [db] The instruction for the change in power for the regulation is given by pu = α (ssdes rxlev filtered ) + β (qdesdb rxqualdb filtered ) (3.1) where α and β is the path loss respective quality compensation and qdesdb and rxqualdb filtered are the qdes and rxqual mapped to C/I as in table 3.1 [6]. The power level down regulation order is then given by PL = INT( pu 2 ) (3.2) where INT truncates the power level to a higher value. PL could have values from 0 to 15, and that represent a down regulation of 0 to 30 db, which could be seen in the final output power level by the BTS, BTS output power = p max 2 PL (3.3) where p max correspond to full power. The existing power control algorithm accepts lower quality in single connections if the whole system experiences a gain in quality. It considers the system quality and not the quality for single mobiles as the main regulate factor. If a mobile experience lower signal quality than the target value, the transmitted power to that mobile will be increased. However, it will only increase the power so that the signal quality changes towards the target value and not actually achieve the target value. This regulation is to prevent party effect. The party effect can be describes as, if one mobile experience low signal quality, the transmitted signal power will be increased to that mobile. This results with an increment of the interference at other 11

20 mobiles. These mobiles then require higher signal power and the interference at the first mobile is increased. Finally all mobiles will be transmitting with maximum power, and this effect it is called the party effect. The existing power control algorithm prevents this by just regulating towards the target value. This is done by setting β < 1 in equation 3.1 [6]. To get a good understanding in the power regulation, knowledge about how much the output power will be down regulated for certain signal strength or quality is necessary. Hence, the dependence between, signal strength, quality and down regulation is important. A way of studying these quantities is in a plot describing the behavior of the algorithm. This could be seen in Figure 3.2. How great the down regulation is, depends on the values of rxqual and rxlev. Figure 3.2: Principle for down regulation. The values rxlev and rxqual are the measured values before any exponential filtering. In the figure above it is shown how the power is down regulated. The controlling values for the desired signal strength and quality, ssdes and qdes, are set to define the point where the two separate planes of the algorithm meet, point marked 1 in Figure 3.2, and the positions of the planes, marked 2 and 3 in Figure 3.2. Point 1 is at approximately rxqual = 3, which is equal 12

21 to qdes = 30, and rxlev = 14. Plane 2 regulates mainly against the signal strength to avoid lower power than the noise floor and plane 3 regulates mainly towards quality. 3.3 EMR The EMR (Enhanced Measurement Report) is a standardized measurement report that contains additional information compared to the earlier measurement report. Like the old measurement report it contains information of the performance of the transmission, for example rxlev and rxqual. The additional information in the EMR are the mean and the CV (coefficient of variation) of the bit error probability (BEP). These values are called MEAN_BEP and CV_BEP, and they are calculated as an average over the frames in a measurement period. The CV_BEP has a general definition as the standard deviations divided by the mean value. 3.4 Problem statement The existing power control algorithm adjusts the transmitting powers to track a predefined quality value, to keep the system quality at an acceptable level. A fixed value of target BER, qdes, is used as the predefined value for quality control. However, there are some problems with using qdes as the predefined quality value. First, BER might not be a good measurement for quality, which implies that quality might change but BER does not. The other problem is the usage of a fixed target value for all different situations. This means that a pre-study to determine the proper fixed value need to be done, and the value could not be changed in an ongoing system. However, the main problem with the existing power control algorithm is the amount of parameters that needs to be predefined. Reducing these parameters makes the algorithm less complex and more intuitive. A method that automatically sets and adjusts this fixed target value if the environment or other factors changes is desirable. This would be a way to avoid predefining a number of parameters and thus make the power control more intuitive. The problem with setting the value initially will also be solved due to the automatic tracking. 13

22 Chapter 4 Simulator description In this chapter the simulation model is described, for example how the simulation environment looks like and how some functions works. The simulator has been developed at Ericsson Research and is named FHRUNE and it has been used in all simulations. 4.1 Simulation model The simulation model is based on the real GSM system network. The model contains an environment that takes things like propagation and thermal noise into account. The model also gives opportunity to set parameters that control how the GSM system should work. In this work, the possibility to adjust the parameters in the power control algorithm is of special interest. This implies that there are possibilities to adjust parameters to change the behavior of the system. Like the real GSM system the simulation model also uses a measurement report to extract data from the environment. FHRUNE uses three different time interval levels because updating all parameters at the shortest time interval would be both unnecessary and inefficient. Parameters like power and mobile positions change only at defined measurement period intervals, 480 ms in GSM. The longest time interval is therefore represented by the measurement period. The next time interval represents the length of a speech frame or a block, which contains 20 ms of information. Each measurement period has 26 blocks. 24 blocks is used to carry the information and signaling and empty bursts constitute together the two extra blocks. The shortest time interval is for the parameters changing fastest and is the burst level. A schematic view of the main function in FHRUNE is shown in Figure 4.1. Before entering the main loop all variables and system parameters are declared and initiated. In the main loop new mobiles are created and added to the system. They are given an initial position and speed within the cell plan. Path loss between all base stations and all mobiles are calculated 14

23 and allocation for new mobiles or hand-over for already existing mobiles are also taken care of in the beginning of the main loop. Finally, before entering the inner loop a number of parameters for both mobiles and base stations are updated, for example the transmit powers. In the inner loop some packet scheduling are done and then on burst level implemented by matrix operations the C/I values for all mobiles are calculated. The C/I values are used to estimate the quality for both speech and data users. When the inner loop is finished data is extracted and logged. Finally in the main loop, users with low speech quality are removed together with completed calls and all remaining mobiles are given a new position and speed. Measurement period Initiation Frame period Create traffic Path loss calculations Allocation and blocking Initiate and update user specific data Packet scheduling Calculate C/I and quality (Burst level) Extract and log data Complete and drop calls Move mobiles Figure 4.1: Schematic view of the FHRUNE simulator. 15

24 4.2 Basic functionality The functions in the real GSM system are modeled by the system simulator FHRUNE. Some of the basic functionalities are described below. The simulator must both simulate the environment and the GSM system. The propagation model and the fast fading are typical parts used to model the environment. Frequency hopping, discontinuous transmission (DTX) and power control are all functions in the GSM system and strives to improve the system performance Propagation model Once for each measurement period a G-matrix is calculated. The G-matrix has a row for each mobile station and a column for each base station. A value in the matrix includes path loss, antenna gain and slow fading and each value represents a mobile/base station pair. The matrix updates once a measurement period. The innermost level inside the inner loop calculates different fast fading values for each burst and add them to the G-matrix. This means that for each step of the inner loop four updated matrices are generated. The values in the G-matrices are used to calculate the resulting C/I at the receivers in the system for each measurement period. The up- and downlink calculations are performed separately and hence there are different matrices for each direction Fast fading The fast fading is caused by multiple reflections close to the receiver producing a Rayleigh distributed fading pattern. The fading values can vary considerable because of the dependence of both the used frequency and the position of the receiver. In the simulator a Rayleigh fading map models the fast fading. The rows of this matrix represent the available carrier frequencies and each column represents a distance of the Rayleigh fading path. The fading pattern of the Rayleigh path is defined by the used frequency and the coherence bandwidth. The frequency used for each burst defines which row to use. Different maps will be used for different values of the coherence bandwidths. The maps are pre-generated, because of the complexity of the Rayleigh fading model, and a parameter defines which map to use. In FHRUNE, a fading path that represents a distance of 50 meters with separate values at each millimeter is used. The length 50 meters is chosen to avoid correlation with distant values. Due to the separate frequency bands for up- and downlink in GSM, two separate maps for each value of the coherence bandwidth, are used in parallel. 16

25 4.2.3 Frequency hopping Frequency hopping is an important option in GSM systems by which network performance can be enhanced. Consider co-channel interference between different connections. Not all of the slots are in use on all of the physical channels on each site where they are reused. If we can take each caller on a particular sector and jump them from frequency to frequency, then each user runs a far lower risk of suffering from co-channel interference. This is because the co-channel interference is shared by many users. The simulator gives a possibility to choose between GSM pseudo-random sequences or ideal sequences from the MATLAB random number generator. In Figure 4.2 an example of frequency hopping sequences for some calls and some bursts during a measurement period is shown. In the figure, GSM pseudo-random sequences are used. For example, call 1 is transmitting on channel 7 in the first burst, but changes to transmit on channel 3 in the second burst. This hopping between which carrier frequency to transmit on continues throughout the call. call bursts (104 columns) Channel numbers Figure 4.2: Format of the frequency hopping matrix Discontinuous transmission Discontinuous transmission (DTX) means that the base station instructs the mobile station to shut down the transmission during the silent periods in a conversation. This is done to avoid unnecessary transmission and save energy. The most important part in DTX is the voice detection, which has to separate the voice from the background sounds. The DTX is modeled in the simulator on speech frame level as a two state machine, active respectively inactive. The switching between states is controlled by parameters that indicate whether the state is active or inactive. 17

26 4.2.5 Quality estimation The C/I-values that is calculated in the inner loop for each burst is one kind of quality measurement used in FHRUNE. The C/I-values describes the relationship between the signal carrier power and the interference power. However, the simulation model is designed to have a number of different quality measurements available. In the sections below some definitions of quality measurements used in this work are presented. A better quality measure than the C/I-values is one that is based on a method that maps the C/I-values to frame error probabilities for speech. The C/I-values are used as input to a process that decides if each frame is successfully received or not. These mappings are the result of link level simulations and the process could be seen in Figure 4.3 and is described below. C/I table lookup BER group and calculate µ, σ two dim. table lookup FEP random process FER Figure 4.3: The steps in the mapping from C/I to FER. In the first step of the process all the C/I-values are mapped to bit error probabilities for each burst, in other words each individual C/I-value correspond to an individual bit error probability. The mapping is implemented as a one-dimensional lookup table. The bit error probability values are grouped in speech frames, and the mean µ, and the standard deviation σ, are calculated for each frame as µ = 1 n n BEP i, (4.1) i 18

27 σ = 1 n (BEP i µ) 2. (4.2) n 1 i The values BEP i are the bit error probabilities for each burst. The calculations are done per frame and since each frame consists of eight bursts, n is equal to 8. The mean and standard deviation are then used in the two-dimensional lookup table, Figure 4.4, to get FEP. One value of error probability is extracted from each pair of mean and standard deviation. [7] The lookup table as a figure FEP std mean Figure 4.4: The lookup table for mapping mean and standard deviation to FEP shown as a figure. The FEP values are used in a random process to decide if each frame is erroneous or not. In the random process a uniformly distributed random vector is compared to the FEP according to f rameerror = random(size(f ep)) < f ep, (4.3) where frameerror is a vector containing ones and zeros where a one indicates a frame error, random(size(fep)) is a uniformly distributed random vector and f ep is the frame error probability. The frame error is used to calculate the frame erasure rate (FER). 19

28 Chapter 5 Tested algorithms The proposed method for dynamically regulating qdes is by adding an outer loop to the existing power control. The implementation of the outer loop in the simulator is included as a function in the main part of the existing simulator FHRUNE. The outer loop has the purpose to calculate a qdes value based on some information for the controlling of the inner loop. This target value could be extracted in different ways. The following sections describe the method in general as well as three different approaches, using different information, to calculate qdes. This chapter also presents the result from the simulations of the three different approaches and an analysis of the result. Comparison with the theories is included to verify the results. Results from both unsuccessfully and successfully algorithms are presented. 5.1 General The motivation for including an outer loop to the existing power control algorithm is that bit error rate (BER) is not necessarily well correlated to quality. BER is the measurement used in the current power control to calculate rxqual. Instead, for example the percentages of lost frames are more relevant, since the effect of modulation, coding and interleaving is included. The information that should be used, and how it should be used, depends on the design of the outer loop. One possibility is the frame erasure rate (FER). However, the objective with the outer loop is to assign a qdes value for the existing power control algorithm to track. In Figure 5.1 the block scheme over the existing power control in GSM is extended with a block for the outer loop. The existing power control could be seen as an inner loop. The outer loop may use information from the EMR or other measurements to produce a qdes value for the inner loop to track. 20

29 Target qdes Current PC algorithm Power Environment Algorithm Additional information RxQual Inner loop Measurement report Outer loop Figure 5.1: The existing power control in GSM extended with the outer loop. The purpose of the outer loop is to serve the inner loop with a dynamic qdes value that changes automatically. The measurement report should give the outer loop additional information about the current quality in the system. The mean and the standard deviation of the bit error rate is additional information in the enhanced measurement report and the outer loop may use these measurements to adjust the qdes value. A couple of different approaches of how to use the information are tested in this work. Algorithms are set up, simulated and evaluated. 5.2 Parameter settings Some of the parameters that describe the environment and also have an important part in how to set the power control parameters are presented in Table 5.1. These parameters were held constant throughout all simulations. 21

30 Table 5.1: Fixed parameters that is used in the simulations. Name Value Description Frequency band [MHz] 900 Could be 900 or Number of frequencies 27 Number of 200 khz bands. Frequency groups 3 The reuse factor. Sectors per site 3 Could be 1 or 3. Cell radius [m] 500 Size of each cell. Number of time slots 1 One instead of 8 for simplicity. Simulator time step [s] 0.48 Length of measurement period in GSM. 5.3 Outer loop based on EMR This method uses the additional information from the EMR as the input to the algorithm in the outer loop. The MEAN_BEP and CV_BEP are in the simulator calculated for each frame. In this model they are calculated as the average value over a measurement period, in other words as an average of 24 values. Target qdes Current PC algorithm Power Environment Algorithm MEAN_BEP CV_BEP RxQual EMR Figure 5.2: Block scheme for the existing power control algorithm extended with an outer loop based on EMR Outer loop based on frame erasure rate This method is based on the frame erasure rate (FER). FER is defined as the percentage of erroneous frames. A target value, F ER_target, is used as the input value. In this method the target value is 0.8%. Basically, the qdes value is increased when FER is smaller than F ER_target and decreased when FER is greater than FER_target. The idea in this method is that 22

31 a specific qdes value is used for each specific mobile. Hence, qdes could be different for each mobile depending on the quality of each connection. The outer loop strives to increase quality in each specific mobile. In Figure 5.3 the blocks in the outer loop with the proper input values are displayed. FER_target Comparison qdes change qdes FER_filtered qdes old Filtering FER Extracting quality measurement MEAN_BEP CV_BEP Figure 5.3: The outer loop when the control algorithm is based on FER. The input values to the outer loop are partly estimated from the EMR and partly defined by a user as a fixed value. From the measurement report information to define the FER is used, as described in section above. The MEAN_BEP and the CV_BEP could be converted to the mean and std in Figure 4.4 and hence be used to extract FER in the simulator. The mean is equal to MEAN_BEP, and the std is calculated as, σ = CV _BEP µ (5.1) where σ and µ are the mean and the std. However, in the simulator a value of FER calculated for each frame is used. FER is than used to calculate a value that is compared to the fixed target FER value, FER_target = 0.8%, defined by the user. The fixed target FER corresponds to the percentage of useless frames that could be accepted. An exponential filtering will be done on the FER for each single mobile, FER_filtered n = α FER + (1 α) FER_filtered n 1. (5.2) where α = 0.5. This result is then compared to the FER_target to get the difference and then multiplied with a constant C to get the difference in the proper unit. The conversion constant is also used to minimize the effect of quick changes in FER and will also decrease the change in qdes. Finally, the change is added to the old qdes values, 23

32 where qdes change is defined as qdes = qdes old + qdes change, (5.3) qdes change = C (FER_filtered FER_target). (5.4) This algorithm should slowly adjust the values of qdes, for each mobile to hopefully strive against a FER lower than target FER. The results from the simulations of this algorithm show no improvement of the system. This could be described by the fact that the outer loop has too much influence and disturbs the existing power control algorithm. Basically, the outer loop takes away the handling if the party effect. This is because the outer loop contradicts to the inner loop and will change qdes until the desired FER is achieved. The FER value could be seen as a measure of quality and when FER is high for a single mobile the outer loop strives to lower the FER by regulate the qdes value for that specific mobile. The goal is no longer to increase the whole system quality but just single mobiles. Figure 5.4 shows the amount of satisfied users at different traffic loads for system with or without an outer loop. The number of satisfied users is defined as the amount of mobiles with the average FER, during the lifetime of a connection, lower than one percent. The equation is satisfied users = 1 M FER i < 1%, (5.5) M i=1 where M is the total number of mobiles and FER is the frame erasure rate for each mobile. This definition of the number of satisfied users are just one of many. 24

33 100 FER_target = 0.8%, alpha = 0.5 PC without OL PC with OL no PC 95 Satisfied users [%] (FER < 1%) Traffic load [Average users per cell] Figure 5.4: Satisfied users at different traffic loads. In the figure above it is obvious that there are more satisfied users when the system uses a power control algorithm without an outer loop based on FER. Although, the performance is better with the outer loop compared to when no power control is used. This means that the effect of the power control algorithm is decreased when this outer loop is included Outer loop based on coefficient of variation This method is based only on the changes of CV_BEP. Removing the dependence of the MEAN_BEP, removes the ignorance of the party effect, as was presence in the outer loop based on FER. The benefit is that the outer and the inner loop no longer contradicts each other. The idea is that the values of CV_BEP are directly mapped into qdes values. More precisely, the equivalent values of CV_BEP at the desired target FER level curve will be found and depending on the position, the qdes is set to an appropriate value. The qdes value is thus not totally dynamic, thou it changes between fixed values depending on the value of CV_BEP. Although, the changes in qdes are automatic. The fixed changes are performed on mobile level, in other words each mobile have specific qdes. The interesting part in Figure 4.4 for the lookup table is that for higher 25