General Packet Radio Service Performance Evaluation Based on CIR Calculation, Considering Different Radio Propagation Models

Transcription

1 C38 1 General Packet Radio Service Performance Evaluation Based on CIR Calculation, Considering Different Radio Propagation Models SAMI A. EL-DOLIL and AMIR S. EL-SAFRAWY Dept. of Electronics and Electrical Comm. Eng., Faculty of Electronic Engineering, Menouf, Egypt. Abstract General Packet Radio service (GPRS) is a step on the evolution path towards 3 rd Generation (3G) mobile radio networks. This paper investigates the effect of using varied Coding Scheme (CS) based on Carrier to Interference Ratio (CIR), that is computed using different radio propagation models. A comparison between these models and a fixed CS is evaluated. A simulator model was developed to study the behavior of data transmission between a GPRS server and data terminals via air interface. Simulation results show the convergence between different radio propagation models, and the capability of using a less protective CS. I. INTRODUCTION Migration from the second generation mobile radio network such as the GSM towards the third generation one such as UMTS or CDMA2000 is achieved by a change in the inherent switching technology, from circuit switching to completely packet switching systems. This sharp switching technology conversion could be done smoothly via a hybrid system such as GPRS, known as 2.5 generation. So it is considered as a key step towards the evolution of the third generation. GPRS is a standard from the European Telecommunication Standards Institute (ETSI) as a new bearer service based on GSM architecture [1, 2]. The rule of GPRS is to provide a packet switched service over the existing GSM networks. The service providers use the scarce radio resources in a more efficient way, by assigning it on demand. Wireless access to external internet protocol-based network such as internet, provides mixed service capability for both real-time and non real-time applications such as WWW, ftp, and video conference. GPRS network operator provides subscribers with instant, always on connection, high data rate with theoretical maximum speed up to k bit/s by applying Multi Slot Capability (MSC) against 9.6 k bit/s of SMS over GSM networks. Several analytical and simulation models have been proposed to study the performance of GSM/GPRS. In [3] transmission of data packet in the silent periods of a conversation with voice activity detection was proposed. The effect of using dynamic and fixed channel allocation techniques on the GPRS QoS is presented in [4]. In [5] different types of queues for mobility management related packets was considered to reduce the delay time by reducing the competition time of the Routing Area Update (RAU). A comparison between using a queue and keeping a dedicated channel for GPRS users is introduced in [6]. In [7-9], GPRS performance and capacity evaluation is estimated considering different radio resources allocation strategies. Different from the previous works, the key feature of this work is that: a simulator developed in Matlab environment is used to study the effect of using different radio propagation models on the GPRS system performance. The simulator focuses on the communication over the radio interface between a MS and BS S, as it is one of the most crucial aspect of GPRS operation, and mainly determines the performance of any wireless network. II. GPRS SYSTEM MODEL The proposed model is a GSM/GPRS system, in which the GPRS subsystem is a new bearer service designed to be an extension to the GSM network. The GPRS functionalities was achieved by introducing two GPRS support nodes (GSNs), the serving GPRS support node (SGSN) and the gateway GPRS support node (GGSN). The SGSN delivers packets to and from mobile stations within its service area, it detects new GPRS MS and queries the home location register to get user profile, and it keeps track of mobile stations locations. GGSN links different SGSNs and provides interface between GPRS network and other external Internet Protocol based networks such as (internet-other GPRS networks). GPRS greatly improves the operation of the wireless data communication networks. From the network's point of view, it employs packet switching which enables operators to use the limited spectrum more efficiently as the radio resources are used only when users are actually sending or receiving data so that the system capacity increases. Form the customer's point of view, GPRS offers much higher data rate than conventional GSM system by applying MSC, thus enables new and

2 C38 2 better wireless data applications. So the mobile phone will be changed from the traditional one to a multi media wireless access one. GPRS gets its radio resources by sharing the GSM resources. One of the most crucial question is, how to divide the cell capacity between the traditional GSM and GPRS services. In [7, 8] it is found that a more friendly strategy is, to reserve Q time slots from the N time slots in each GSM/GPRS TDMA frame for data connection preventing their use by voice calls. Consequently guarantee a minimum amount of band width to be shared among all active data connections. The drawback of this strategy is the increase of the number of blocked voice calls due to the reduction in time slots available for voice. In order to guarantee a fixed voice call blocking probability, it is necessary to dimensioning the cell for a reduced offered voice load, therefore a balance between the reduced number of served voice calls and the required increase of data connections has to be performed by the network operator. The remaining N-Q time slots are shared between GSM and GPRS users with full preemption for the voice call over data connection in this shared partition. When a voice call request arrived and there is no available channel, the voice preempt a channel used by data connection and the data terminal (GPRS user data terminal are interchanged) is forced to stop its transmission (immediately released) and stored in a queue, waiting for service. The GPRS user during its data session will send and receive data, that must be converted into bursts. Each burst data bits is convoyed by a time slot each frame period. The conversion of user's data to burst carried out through multiple levels, at first the data is segmented into packets, each packet is split into Logical Link Control (LLC) frames, each LLC is in turn split into a number of Radio Link Control (RLC) blocks as shown in figure (1). The number of data bits contained in each RLC block is 456 coded bits arranged in four consecutive radio bursts. As the size of the RLC block is kept constant, the Coding Schemes (CS1-CS4) determine the a mount of the useful payload convoyed by each RLC block. The RLC block is the Packet Data Channel (PDCH) basic transmission unit, each RLC block requires four time slots in four consecutive time division multiple access frames, hence the RLC block is the unit of transmission. Some previous studies made their calculations considering the RLC block level rather packet level [10]. III. PROPOSED APPROACH This paper focuses on the performance of GPRS data service considering three different radio propagation models in a cluster of seven hexagonal cells cellular mobile network. The key feature of this work is, First, to investigate the convergence of the radio propagation models and introduce a comparison between them with respect to different quality of service parameters such as; the failure probability of new and handoff attempts of data terminals, the RLC block delay and throughput (channel utilization by GPRS load). Second, a comparison between varying CSs according to CIR and applying a fixed CS (CS2, that is used, as it provides a suitable protection and guarantee BLER less than 0.2, [14]. CIR is a crucial parameter in today s network, as the capacity of any wireless network is interference limited. By considering the radio propagation loss and shadowing, both will affect the Carrier to Interference Ratio (CIR) between co-channel cells. The used propagation models are : a- The average propagation loss model is given by [12]; d L ( d ) = log( ) 1000 The path loss exponent model given by [15] L ( d ) = α log( d ) The path loss Okumura-Hata model given by [15] d L( d ) = log[ f c( MHz )] log[ hb] a( hm) + [ log( hb) ] log 1000 where: a h 1.1 log f 0.7 h 1.56 log f 0.8 ( ) [ ( ) ] [ ( ) ] m = c m c f c is the carrier frequency in MHz. d is the distance in meter. α is the path loss h is the BS antenna height (m) (effective); b h is the MU antenna height (m) (above ground); a m h m ( ) is the correction factor;

3 C38 3 The interference between two co-channel cells separated by distance d is; = p L (d ) + λ I tx where λ is the shadowing fading term which is a random variable following normal distribution with zero mean and eight standard deviation, this random is range limited between, [16] 4σ λ 2σ where, σ is the standard deviation Considering the offered traffic per interfered channel is assumed to be constant and equal to 0.4 Erlang, and a constant transmitter power P tx = 33 db [11], a path loss exponent α = 4, standard deviation σ = 8 [16], the frequency of operation f c = 895 MHz, the cell radius R=1000 meter. The number of co-channel interferers cells considered here are the first two interferers tires, so there are at maximum eighteen active interferes and seven interference states (0-18). The total interference strength in db, contributed by these interferers is: It db = 10 log 10 n = 18 n= 0 P n I n wat Where P n is the probability of n active interferers from seven nineteen states (0-18) that can be obtained from the interfering traffic using the Erlang formula. Assume the offered traffic per interfering channel equals.4 Erlang, so the total interfering traffic A is 7.2 Erlang, then P n equals: P A / n! n n = i = 6 i= 0 i A / i! I n wat represents the interference power in watts for n active interferers, The average carrier strength is calculated as; C ( R MS BSS ) = Ptx L( R MS BSS ) where; R MS-BSS is the average intra cell MS-BSS distance that can be approximated by the radius of a circle covering half of the hexagonal cell area, [11] given by: 2 AHex 1.5 3R RMS BSS = = = 643m 2π 2π L (R MS-BSS ) is the propagation loss of the intra cell MS-BSS distance R MS-BSS. The mean carrier to interference ratio CIR mean is; CIR mean = C ( R MS BSS ) I t db The CIR is obtained for each data terminal. It depends on the chosen radio propagation model, and the current location, which assign both the propagation loss and shadow fading term. The value of CIR will control the choice of the coding scheme, and during the user mobility the CIR is updated, according to the following table: Table 1: Switching between different CS's depending on CIR value CIR CIR CIR CIR CIR > Coding Scheme CS1 CS2 CS3 CS4 The switching between these CS's was done according to [10, 17], which present the optimal choice of the coding scheme as a function of the CIR. The value of CIR defines the coding scheme that maximizes the throughput and keep BLER < 0.2. IV. SIMULATION STRUCTURE In this paper a simulator of GPRS at the radio link control level has been carried out and developed using Matlab environment. The simulator focuses on the communication over the radio interface between MS and BSS, because this is one of the most crucial aspects of GPRS operation. It considers a two dimension GSM/GPRS network, in which the coverage area is partitioned into seven hexagonal cells. To allow for a finite number of cells to be simulated while still approximately an infinite system, a warp-around technique is used. This means that opposite sides warp-around, so that the finite size effect is eliminated. The simulator describes the behavior of a single carrier with eight time slots GSM/GPRS system in which one channel (time slot) is reserved for control signaling, another channel is reserved for GPRS data sessions, the remainder six channels are shared by both circuit switched GSM services and GPRS data sessions. The arrival of voice calls and data

4 C38 4 sessions are modeled according to two mutually independent poisson processes, both of them are scheduled to share the radio resources. When a new data session arrives it will be served. If there is no available resource it will be stored in the access queue waiting for a finite time. During this time if resources will be available it will served, if not it will be cleared from the system due to buffer time expiring. When a new voice call request arrives, it takes a free channel, if there is no channel available and the number of voice calls in the service is below six, hence one of GPRS data sessions stops its transmission in order to allocate a channel to the new voice calls. The interrupted data session stored in the suspend queue without limitation on the storing time. When resources are available again, the data sessions in the suspend queue have higher priority to be resources allocated than the new data session or data session in the access queue. The discipline in the access and suspend queue follow the first in first out (FIFO) principle which is the simplest scheduling and queuing method. The average service time of circuit switched calls is assumed to be exponentially distributed with mean 120 seconds. The voice traffic load is set to Erlang to satisfy the general requirements of 2% blocking probability which is a typical target value for GSM operators. In this paper the traffic model is used, that has the following characteristics: Calls and data sessions arrived according to poisson processes. The used coding scheme is CS2. Single carrier, eight time slots. The length of the size is exponentially distributed with a mean of 320 K bit, [18]. Average service time of /13400 second. The data terminal can operate in MSC mode, up to eight time slots were used simultaneously. Mobile stations are assumed to be moving with a uniform truncated speed between maximum value of 60 Km/h (vehicles) and minimum speed of 5 Km/h (pedestrian). Access queue of length 20 and maximum storing time of 2 minutes. Suspend queue with infinite length and storing time. Cell radius set to be 1000 m. The queuing discipline is FIFO. Three radio propagation models were used to consider the effect of environment. As seen, the average service time of GPRS data session is not small, hence during the mobility of the data terminals, handoffs are required for these terminals to complete their data sessions. During the data terminal movement within the cell area, it suffers both propagation loss and shadow fading due to terrain features and building density. The propagation loss is calculated using three radio propagation models. V. SIMULATION RESULTS The developed simulator was intensively utilized to investigate the behavior of the traffic model over GPRS. The simulation results show that the three radio propagation models are converged and there is no great difference between their results. Figure (2) shows that the three propagation models have a converged behavior, the blocking probability of new data sessions differs slightly. The same behavior is also kept for the handoff failure probability of data sessions as shown in figure (3). Figures (4,6) show the RLC block delay, data sessions served in the system per second, both have a semi counterpart results for the three radio propagation models. Percentage of data sessions completed and throughput are shown in figures (7,8), which illustrate that the radio propagation models under consideration are approximately have the same behavior. The above figures prove that the quality of service parameters of the GPRS system are differ slightly for the three radio propagation models, thus mean that they are interchanged. From the figures, it is clear that the Okumura-Hata model has intermediate results between the path loss exponent model and the propagation path loss model. So a comparison has been made between using a fixed coding scheme (CS2) and a varied coding scheme according Hata model. This Hata model used because; First, it has intermediate results. Second, the model depend on the distance, operating frequency, height of base station antenna and the height of mobile terminal antenna. So it's considered a more empirical model. From figures (9,10) which represent the blocking probability of new data sessions and the handoff failure probability, it's clear that the use of varied coding scheme will enhance the system performance for example, at offered traffic of 12 Erlang the blocking probability of new data sessions will reduce from 0.42 to a bout 0.2. The RLC block delay and data sessions served in the system per second, are shown in figures (11,13), again there is a reduction in the RLC block delay and increase in the data sessions served. Percentage of data sessions completed and throughput are illustrated in figures (14,15) respectively. The data sessions enter the system per second in both cases presented in figures (5,12), both figures ensure that all the simulation programs operate under the same offered traffic and thus authenticate the reliability of the

5 C38 5 simulator. It is clear that the use of varied coding scheme enhance the system performance rather than using fixed coding scheme (CS2), thus mean that a lower protection coding scheme of higher data rate such as (CS3,CS4) can be used without missing the global aim, such as maximizing the throughput and maintaining lower BLER(<0.2). VI. CONCLUSION In this paper, a comparison between applying three radio propagation models and a comparison between using a fixed coding scheme (CS2) and a varied coding scheme (Hata model) is investigated. Results were obtained by simulating GSM/GPRS network air interface and traffic model. The contribution of this paper is two folds; First, the simulation results emphasize that the three propagation models have approximately the same behavior and they can be interchanged. It is clear that the Hata model has intermediate results, while the propagation path loss model gives higher blocking probability, handoff failure probability, RLC block delay and lower RLC block served, and the path loss exponent model gives the lowest results. Second, the comparison between using a fixed and a varied coding scheme shows that, using a fixed CS2 wasting the system resources, because the system can operate at higher coding schemes (CS3, CS4), while keep the BLER < 0.2. REFERENCES [1] ETSI, Digital cellular telecommunications system (phase 2+) General Packet Radio Service, (GPRS) service description; stage 2 (3 GPP TS version Release 1998),. [2] G. Brasche and B. Walke, Concepts, services and protocols of the new GSM phase 2+, General Packet Radio Service, IEEE Commun. Mag, pp , Aug [3] J. H. Hung, S. L. Su and J. H. Chen, Design and performance analysis for data transmission in GSM/GPRS system with voice activity detection, IEEE Trans. Veh. Technology vol. 51, pp , July [4] M. D Arienzo, A. Pescapè and G. Ventre, A simulation environment for analysis experimental analysis of multiple traffic scheduling algorithms over GPRS networks, CCCT, July [5] K. Chew and R. Tafazolli, Performance analysis for GPRS with prioritized and non-prioritized mobility management procedures, IEE Third International Conference on 3G Mobile Telecommunication Technologies, London, United Kingdom, pp , 8-10 th May [6] M. Ghanderi and R. Bautaba, Data service performance analysis in GPRS systems, Proceedings of the 15th IEEE international Symposium on Personal, Indoor and mobile radio communications (PRIMC'2004), Barcelona, Spain, September [7] R. G. Garroppo, S. Giordano and S. Lucetti, Capacity evaluation of resource allocation strategies in GPRS system, Academic OPNET Research and Educational Projects, University of Pisa, [8] M. Ermal, K. Müller, J. Schüler and M. Schweigel, Analytical comparison of different GPRS introduction strategies, Proc. 3rd ACM Int. Workshop on Modeling analysis and simulation of wireless and mobile systems, pp. 3-10, Boston, MA, Aug, [9] W. Chen, J. C. Wu and H. Liu, Performance analysis of radio resource allocation in GSM/GPRS networks, Proceedings of IEEE VTC 2002-Fall, vol. 3, pp , Sept [10] O. Queseth, F. Gessler and M. Frodigh, Algorithms for link adaptation in GPRS IEEE Proc. VTC Houston., May, [11] U. Fornefeld and B. Walke, On the CIR gain from location diversity in GPRS networks, Proceedings of the European wireless, vol. 0, pp , Feb [12] ETSI TC-SMG 5, Universal Mobile Telecommunications System (UMTS); selection procedures for the choice of radio transmission technology of the UMTS (UMTS 30.03), Technical Report 3.2.0, European Telecommunication Standards Institute, Sophia Antipolis, France, Apr [13] D. K. Kim, and D. K. Sung, Traffic management in a multicode CDMA system supporting soft handoffs IEEE Trans. Veh. Technology vol. 51, pp , January [14] R. Kalden, I. Meirick and M. Meyer, Wireless internet access based on GPRS, IEEE Commun. Mag, pp. 8-18, April, [15] W. Webb, The complete wireless communications professional, Artech House, Norwood, MA, [16] R.Steele, C.C.Lee, P.gould, GSM cdmaone and 3G systems, Wiley, [17] P. J. A. Gutiérrez, J. Wigard, P. N. Anderson, Performance of link adaptation in GPRS networks, Proceedings of IEEE, VTC'2000-fall, vol. 2, pp , [18] R. Litjens and R. J. Boucherie, Elastic calls in an integrated services network: the greater cell size variability the better the QoS, Performance evaluation, vol. 52, no. 4, pp , 2003.

6 C38 6 User data Packets LLC frames RLC blocks Burst 20 ms Blocking probability of new data sessions. Figure (1): User data to bursts transformation. Figure (2): The blocking probability for three radio propagation models. Handoff failure probability of data sessions. Figure (3): The handoff failure probability for three propagation models. Figure (4): The RLC block delay for three propagation models. Data sessions enter the system per second. Data sessions served in the system per second. RLC block delay [second] Figure (5): Data sessions enter the system for three propagation models. Figure (6): Data sessions served for three propagation models.

7 C38 7 Percentage of data sessions completed. Throughput [normalized channel occupation time]. Figure (7): Percentage of data sessions completed for three propagation models.. Figure (8): Throughput for three propagation models. Blocking probability of new data sessions. Handoff failure probability of data sessions. Figure (9): The blocking probability. Figure (10): The handoff failure probability. RLC block delay [second]. Data sessions enter the system per second. Offered traffic per cell [Erlang] Figure (11): The RLC block delay. Figure (12): Data sessions enter the system.

8 C38 8 Data sessions served in the system per second. Percentage off data sessions completed. Figure (13): Data sessions served. Figure (14): Percentage of data sessions completed. Throughput [normalized channel occupation time]. Figure (15): Throughput.