Cellular Mobile Radio Networks Design

Save this PDF as:
 WORD  PNG  TXT  JPG

Size: px
Start display at page:

Download "Cellular Mobile Radio Networks Design"

Transcription

1 Cellular Mobile Radio Networks Design Yu-Cheng Chang Ph. D. Candidate, Department of Technology Management Chung Hua University, CHU Hsinchu, Taiwan Chi-Yuan Chang CMC Consulting, Inc. Taichung, Taiwan Ching Chang Department of Transportation Technology and Logistics Management Chung Hua University, CHU Hsinchu, Taiwan Guang-Yu Tu Physical Education Ling Tung University, LTU Taichung, Taiwan Abstract Two aspects of cellular radio systems design are covered in this work. First, is a way of evaluating a cellular system with a given configuration, including coverage, cochannel interference, and traffic capacity considerations. Second, based on this evaluation scheme, Simulated Annealing is used to improve the system performance by varying system parameters such as site channel assignment, site power, antenna azimuth, and antenna down-tilt. The main improvement of this work over previous reports is two-fold. First, calculations of the co-channel interference are direct. In comparison, a common practice is to use a compatibility matrix or similar measure for interference. A single number is then used to characterize the interference between a pair of sites / cells, which is an over simplification because interference varies for different points in the cell served by a site. It is the collective results of all other cells using the cochannel and not only just between the two sites. Furthermore, interference in both directions, uplink and downlink, need be included. Uplink is the weaker link in general and is commonly ignored in the compatibility matrix. The second improvement is that the entire area distribution of traffic is used in this work as opposed to the common practice of using total traffic at each site. Keywords: coverage; co-channel interference; cellular system I. INTRODUCTION In real-world cellular systems, the use of hexagonal cell cluster deployment is standard [1], yet impractical. Site location limitations and irregular terrain render the standard cell configuration inaccurate, however, the basic principles of cellular design are unchanged. The goal is to design a system that provides sufficient radio coverage, tolerable co-channel interference, and highest possible capacity. These three factors are interrelated and sometimes contradictory to each other. Along with system traffic distribution, they make the system design a complex problem. Among all the factors, co-channel interference is the most complex and difficult to deal with. Co-channel interference exists in both uplinks and downlinks and is dependent upon the RSSI (Received Signal Strength Indicator) distribution, traffic distribution, site locations, and the location of the point under consideration. Some researchers choose a simplified treatment by using a compatible matrix [2-5]. The shortfall with this type of simplification is that one cannot use just one number to characterize the interference between a pair of sites / cells, let alone the meaning of that number. Interference varies at different points in a cell. Furthermore, even at the same point, interference is different on different channels. It has many dimensions and cannot be sufficiently represented by a single number. Some authors try to minimize the number of interfering base stations [6]. The problem is that lower number of interfering sites does not necessarily mean lower interference. In both cases, only the downlink seems to be included. It is well known, that the uplink is often the weaker one and cannot be ignored in the model. In this work, the interference is calculated for the entire service area, uplinks and downlinks; and for each possible channel for that point, using the geographic traffic distribution. When this interference is lowered, the communications quality is improved. One practical example is used to demonstrate different aspects of the modeling work and calculations. Also, since the current configuration of this example is known, any improvement can be compared against it. A sector and a cell are synonymous for the rest of this paper unless otherwise stated. II. THE EXAMPLE A deployed system in KeeLung is used because of its geographic isolation by the shoreline and hilly terrain surrounding the city. This is shown in Fig. 1 below. Figure 1 Terrain Map of the Example System The largest polygon depicts the service area. There are 88 sites shown by the circles, and each site has up to 5 sector antennas, for a total of 202 sectors. Sixty-six channels are used and reused among the 202 sectors. Each channel has 8 TDMA time slots, with one of them used for control and the other 7 slots are for traffic. There are 15 types of antennas /11/$ IEEE

2 with different horizontal and vertical patterns. Some patterns are shown below. Figure 2 Example Horizontal Pattern Figure 3 Example Vertical Pattern Figure 4 Current System Figure 4-1 Zoomed-in View of Current System The current system is shown in Fig. 4 The spikes show the sector antenna s horizontal orientations. The smaller polygons represent increasingly denser areas and will use finer grids in the calculations. Fig. 4-1 shows the smaller polygons in more details. The minimum required RSSI for communication is 104 dbm for the uplink and 100 dbm for the downlink. The minimum S/I ratio is 15 db for both links. The EIRP of all sector antennas are given in the original data file in dbm. The maximum EIRP for the MS units is 500 mw. This EIRP is lowered by the system power control as the MS moves closer to the BS. The loss due to the site cable etc. is assumed to be 3 db. The loss at the BS unit is assumed to be 10 db. The building/environment loss is set at 12 db. This loss is in addition to the terrain propagation loss. The traffic file indicates that the 22 nd hour is the busiest hour and the traffic for that hour is used for analysis. For calculation efficiency purposes, three tiers of grids are used. The grid sizes are 300, 800, and 3000 points per arc degree respectively. Propagation losses from each sector antenna to each grid point are calculated using TIREM [7]. In order to apply the vertical antenna pattern, the depression angles are also calculated [8]. Sector antenna heights are given in the original data file. Mobile antenna height is set at 5 feet. Three-second terrain elevation data are used. Since the original data file contains only combined traffic at each sector, the traffic need be spread out to obtain an area traffic distribution. This is done by first finding out the extent of each cell. Then the traffic of each sector is evenly distributed over the cell it serves. The cell extent is found by assigning each grid point to the sector, which has the highest RSSI at the point. The traffic distribution is uniform at this point in each cell. Later on when system parameters are varied, cell extent will also vary and will no longer be uniform in each cell. An (Objective Function, OF) for the system is evaluated in the following way for a given system configuration. For each grid point, if the RSSI in either uplink or downlink direction is below the threshold, the traffic in that cell is excluded from further consideration. Next, interferences in both directions are calculated assuming a full load situation, i.e. all assigned channels at all cells are being used. This is performed for all possible channels at that point, i.e. all channels assigned to the home sector. If the SIR in either direction is lower than the 15 db threshold, traffic from that grid point has excessive interference on that channel. Overall, the amount of interfered traffic is proportional to the number of channels with excessive interference. For the downlink, the total interference is simply the summation of signals from all other sectors using the same channel. For the uplink, the total interference from all other cells using the same channel are averaged and summed. Averaging is necessary because the mobile unit can be anywhere in the cell. Traffic is used as the weighting factor in the averaging process. The total traffic in each cell is now compared against the capacity of all assigned channels at the sector and excess traffic is simply blocked. In this work, queuing is not included. The contribution to the Objective Function from the grid point under consideration has two parts. The first part is positive and consists of the carried traffic. The second part is negative and is composed of the traffic, which experience excessive interference as calculated above. So, first the OF has all

3 carried traffic. Second, any traffic which experience excessive interference at full load in either direction cause a penalty and decrease the OF value: OF = T i W N i (1) i Here T i is carried traffic from grid point i; i N i is interfered traffic at point i; W is a weighting factor. The summations are over all grid points. W represents the importance of the full load interference. W is set to 1 in this work. It can be set to other values as needs arise. During evaluation of the OF, the blocked traffic and the traffic excluded because of low RSSI are also obtained. These two parts are implied by the first term of the OF. This OF provides a measure for the system communications quality and is used in the improvement process. III. IMPROVEMENT PROCESS System performance can be improved by varying system parameters to achieve a higher value of the OF. Simulated Annealing is used for that purpose [9]. Overall algorithm is shown in below: Initialize system. Calculate OF old. Set Initial temperature t. While t < final t do: Do until equilibrium is reach: Randomly find a neighborhood point in the configuration space. Calculate OF new of the new point. If OF new > OF old or exp ((OF new - OF old )/t ) > random, adopt new point. t is decremented. A. Temperature And Equilibrium The initial temperature is set such that a new configuration is adopted with high probability regardless of the OF. In this work, the initial temperature is calculated as follows: Initial t = 200 / (log (1/ar ) (2) Where ar is a high acceptance rate and log denotes the natural logarithm. A value of is used for ar in this work. During the initialization, a seed system configuration is established as a starting point. The current configuration is used as a seed in this work. Which configuration is used is really not critical. The final temperature is a positive number very close to 0. A 1.0E-8 is used here. For equilibrium purposes, the number of transitions at a particular temperature depends on the nature of the problem and the number of variables. Stable results are obtained with 200 of transition tries for this example. Temperature is cooled each time by a factor of 0.9. The random number, random, is between 0 and 1. B. Configuration Space The configuration space neighborhood structure is determined by the action taken at each transition try. There are 5 possibilities. For each transition try, a sector is randomly selected. Each possible action occurs with equal probability: Add a channel: If a sector already has 4 channels, skip this action. Otherwise a channel in the available channel pool is randomly selected such that the selected channel number is different from any assigned channel number by at least 2. This is to avoid assigning a channel twice to the same sector and to avoid adjacent channel assignment. Drop a channel: If a sector only has 1 channel, skip this action. Otherwise a channel is randomly selected and dropped. Vary the azimuth: The azimuth is incremented or decremented by 5 degrees with equal probability. If the azimuth is at 0 degree it can only be incremented and it can only be decremented if it is 355 degrees. Vary the down-tilt: The down-tilt is from 0 to 9 degrees and is varied 1 degree in either direction at a time. At 0 / 9, it can only increase / decrease. Vary EIRP: EIRP can be varied up to plus or minus 4 db s, 1 db at a time. The EIRP factor is varied by 1 or 1 at a time. C. Calculations Efficiency Aside from using the heterogeneous grid system, calculations in the SA process need also be streamlined. RSSI from all sites to all points in the service area are calculated beforehand. An uplink interference table is also calculated before the SA improvement process. This table contains interference from every cell to every point in the service area. As SA proceeds, cells change, home sectors also vary, RSSI change, etc. So the pre-calculated RSSI and the uplink interference tables change and need updates constantly. With each transition in the SA process, it is necessary to determine which quantities of which sectors are altered and make the updates. The changes are all collected and updated. This is necessary in order to cut down the computation. One invaluable check we have is for any configuration and the objective function OF 1, either during or after the SA process, we can start from scratch and calculate all tables and the objective function OF 2. Equality of OF 1 and OF 2 means that we have made all necessary updates to the tables. IV. SA RESULTS The above SA scheme is applied to the example location in two parts. In the first part, the current traffic is used and in the second part, the traffic is enhanced by a factor of 3. Improved system performance is compared with that of the

4 current configuration in both cases. The results for the current traffic are shown below in Table 1. TABLE I. RESULTS FOR CURRENT TRAFFIC LOAD, ENTRIES ARE IN ERLANGS Objective Function Carried Low RSSI Blocked Interfered Current Configuration SA Improved Improvement Percentage Improve N/A 99.6 Since the current system is lightly loaded, the improvement on the carried traffic is only 1.3 percent. The Improvement on the excluded traffic because of low RSSI is 15.2 percent. The more impressive item is the interference: nearly all interferences are eliminated! One thing worth mentioning is that, for the current configuration, close to 80 percent of the interfered traffic have their interference in the uplink. After SA Improvement, the remaining interferences are almost all on the downlink. Results for the system with three times of the current traffic are shown in Table 2 below: TABLE II. Current Configuration RESULTS FOR THREE TIMES CURRENT TRAFFIC LOAD, ENTRIES ARE IN ERLANGS Objective Function Carried Low RSSI Blocked Interfered SA Improved Improvement Percentage Improve We have shown a general scheme for evaluating a cellular mobile radio system. An objective function is introduced based on this scheme. Furthermore, we have shown the application of SA to enhance the system performance by improving the objective function. In this work, only some of the variables are altered. One could conceivably include site cost considerations in the scheme. Also, the number of sectors can be increases and more channels can be allocated at each sector. When all these possibilities are included, system performance can be improved further. REFERENCES [1] Mehrotra, A. Cellular Radio Performance Engineering, Artech House, [2] Wong, W., and C.K. Rushforth, An Adaptive Local-Search Algorithm for the Channel-Assignment Problem (ACP), IEEE Trans. on Vehicular Technology, Vol. 45, No. 3, Aug. 1996, pp [3] Kim, S., and S-L. Kim, A Two-Phase Algorithm for Frequency Assignment in Cellular Mobile Systems, IEEE Trans. On Vehicular Technology, Vol. 43, No. 3, Aug 1994, pp [4] Catedra, Manuel F., and Perez-Arriaga, Jesus, Cell Planning For Wireless Communicaions, Artech House, [5] Duque-Anton, M, Kunz D., and Ruber, B., Channel Assignment for Cellular Radio Using Simulated Annealing, IEEE Trans on Vehicular Technology, Vol. 42, No. 1, Feb. 1993, pp [6] Hurley, S., Planning Effective Cellular Mobile Radio Networks, IEEE Trans. On Vehicular Technology, Vol. 51, No. 2, Mar. 2002, pp [7] IEEE Trans. On Vehicular Technology, Vol. 37, No. 1, Feb [8] Giger, A. J., Low-Angle Microwave Propagation: Physics and Modelling, Artech House, [9] Laarhoven, P. J. van, and Aarts, E. H. L., Simulated Annealing: Theory and Applications, D. Reidel Publishing Company, The purpose of the second part is to demonstrate the effect of loading the system to a significantly higher traffic level. The SA improved configuration shows much better system performance than the legacy system. The carried traffic is increased by 15.8 percent. The low RSSI traffic is reduced by 12.1 percent. The most impressive improvements come from the last two columns: the blocking and interference are almost all gone. As in the first part, while the legacy system has most of its interference in the uplink, the remaining interferences after SA improvement are mostly in the downlink. The improvement on the carried traffic is limited because the total numbers of channels, which can be allocated to a sector are no more than four. Furthermore, the number of sectors at each site cannot be altered. The low RSSI traffic / coverage improvement is limited by the current site locations. Coverage can be further improved by putting more sites where coverage is poor. This is not explored in the current work. V. CONCLUSION