Abstract: In this study, we describe various approaches that can be used to estimated radio and network resources in order to find optimum configuration parameters for the existing General Packet Radio Service (GPRS) networks. In our simulation we have used a sn2 network simulator. The proposed approaches, can be applied for reserving a fixed number of Packet Data Channels (PDCH) for GPRS services and dynamically allocate the rest of the channels for voice and GPRS services
INTRODUCTION
General Packet Radio Service (GPRS) uses the same frequency spectrum as GSM for Uplink (UL) and for Downlink (DL). In GPRS Mobile Stations (MSs) compete for access of Base Station (BS) via Frequency Duplex Division (FDD) link with Time Division Multiple Access (TDMA). GPRS consists of 124 radio frequency duplex chmels of 200 KHZ each; some of these chmels are allocated to GPRS and some to GSM Timeslots of TDMA available for GPRS users are called packet data channels. These chmels are allocated toMS only when it needs to send data, more than one timeslot can be given to a single MS resulting to a maximum of data rate of 172 kbits sec -1 . One timeslot (physical channel) is 0.577 ms with asymmetric data traffic, meaning that different number of timeslots for UP and DL. In a typical GSM network with fixed data rate services, data users are permanently allocated two timeslots for every TDJ\.1A, one for UL and another for DL. Since data packets are burst in nature, the fixed time allocation scheme is not reliable, to solve this problem GPRS implements a dynamic timeslot allocation for data traffic, that is available radio chmels may be concmently shared among several users by dynamically allocate timeslots (Kari, 2003). In order to guarantee the Quality of Service (QoS) requirements, we need to optimize this dynamic resource allocation. To allocate system resource efficiently it is necessary to provide optimum parameter settings. Therefore, this study presents various approaches that can be used to estimated radio and network resources in order to find optimum parameter settings for the existing GPRS networks.
QOS GUARANTEE OVER GPRS
GPRS enables the use of a Packet based interface over the exiting Circuit Switched (SC) GSM networks. GPRS allows efficient utilization of bandwidth (BW) because radio resources are used only when packets are sent or received. CS data are usually given priority over packet switched (PS) data and there are few seconds when no data is being transferred.
The GPRS mobile subsystem consists of a number of MSs generating traffic in accordance with negotiated QoS profiles, the negotiated profile is stored in Packet Data Protocol (PDP) context Figure 1 show a typical QoS profile information.
The GPRS QoS profile is defined in terms of the following classes; precedence, delay, reliability, means and peak throughputs. QoS involve solutions that allow networks to meet performance requirement of different application in terms of delay, BW, network availability and packet transfer reliability.
The channel encoding used depends on the transfer speed requirements, each coding scheme has different amormt of error correcting coding that is optimum for different radio envirornnent (Ericsson, 2003).
| Fig. 1: | QoS profile information element |
| Table 1: | GPRS coding schemes |
The least robust (but fastest) encoding scheme (CS-4) is used when the MS is near BS, while the most robust encoding scheme (CS-1) is used when MS is further away from BS. The basic transmission nnit is called a radio block, the structure and the number of the payload bits of a radio block depends on the coding schemes (Table 1 ).
Once the negotiated resources are made available, the GPRS networks strive to provide the negotiated resources to MS while the MS must honor the use of resources allocated to it. To achieve these performance requirements the GPRS system should implement; Admission Control (AC), Resource Reservation scheme, Traffic Scheduling (IS) and Policing to negotiate network traffic.
GPRS NETWORK RESOURCES ALLOCATIONS
GPRS transmission plane overview: GPRS transmission plane illustrated in Fig. 2 consists of layered protocol structure. The transmission plane provides user data transfer along with the associated information transfer control procedure such as flow control, error detection and error recovery.
Sub Network Dependent Convergence Protocol (SNDCP) transmits data between MS and SGSN.
Urn is air interface between MS and Base Service Station (BSS). Gb interfacing BSS to SGSN while Gn interfacing SGSN with Gateway GPRS Support Node (GGSN). Logical Layer Control (LLC) provides reliable information between layer 3 (ISO 7 layers reference model) in the MS and the Service GPRS Support Node (SGSN), LLC is independent of the nnderlying radio interface segments LLC frame into RLC data blocks. The Media Access Control (MAC) sub-layer decides which PDU to be sent during each Time Transmission Interval (TTl). Since in GPRS each timeslot is allocate to user only when they need to sent data many users can use the same timeslot.
Upon request, temporary radio resources may be allocated to MS. the Temporary Block Flow (TBF) may then be established to facilitate the transfer of LLC frames on the Packet Data Channels (PDCHs). A TBF is physical connection and is identified by the Temporary Flow Identifier (TFI). the IF!is included in every transmitted block, which allows multiplexing of blocks originated from different MS on the same PDCH. The Traffic Channels (TCH) for GSM that are not used are made available for GPRS use, if the number of PDCHs does not exceed the sum of the maximum allowed number of permanent and dynamic allocated PDCHs. The choice of data transmission rate depends on the number of GPRS users multiplexed on the same PDCH. There are different configurations on the number of TBFs that PDCH can have in DL and UP, for example Ericsson technology up to 6 UL and 8 DL TBFs are allocated per PDCH (Mimva et al., 2001 ).
A TCH can be allocated for either CS voice connection or PDCH. In our simulation, the Channel Management Module (CMM) in the NS2 simulator communicates with the CS traffic generator and the GPRS radio resource management instance that is located in the RLCIMAC layer. The state of this module is defined by the number ofTCH available to a radio cells.
Channel allocation schemes: Since data and voice user in a cell share common transport media which has limited capacity, the resource for GS1.1/GPRS traffic can be statistically, dynamically or a combination of both can be used.
| Fig. 2: | GPRS transmission plane protocol stack |
Different cell capacity partition scheme can be defined as follows
Complete Partition (CP): Total cell capacity is partitioned into 2 fixed parts for voice and data traffic.
Partial Sharing (PS): Data users have their exclusive bandwidth but they can also use the available BW of voice service with pre-emptive service priority for voice calls.
Complete Sharing (CS): Total capacity is shared between 2 types of users with pre-emptive priority for the voice service. However, this method does not provide fair competition for data packets as they are likely to be blocked.
In our simulation we have used Partial Sharing scheme as suggested by Baynat and Eisenmann (2004) and Lindemann and Thumber (2003) since it guarantees an amount of BW to be shared between all the active data users. We have divided this (Partial Sharing) scheme into 3 parts because partial sharing is a combination two schemes, permanent allocation (which provides exclusive BW to data user) and dynamic resource allocation (the capacity available for the voice is shared between data users and voice users). In order to guarantee QoS requirements we need to optimize resource allocation, which is the combination of scheduling mechanism and efficient resource utilization.
To achieve this goal we first need to define the QoS level required (for example throughput, acceptable blocking probability), then estimate the number ofPDCH channels required to provide the defined QoS level for data traffic. We can define the QoS level by investigating the dependency of IP throughput by the user in the cell. We have reconfigured the parameters and modified some scripts on the Network Simulator NS-2 (Jain, 2001) so that the BS is interfaced with wired nodes using hierarchical routing to simulate the GPRS network. The transmission and network protocols TCPIIP are implemented as in (Jain, 2001). The MAC protocol instances are operating with random access sub channels per 52frame. LLC windows size is fixed at 16 frames. Outgoing TBFs in UP and DL are served with Round robin scheme with a Round Robin depth of 10 radio blocks. All conventional MAC request are scheduled with a FIFO format. Internet sessions consists of the applications World Wide Web (WWW) and electronic mail (e-mail) running the TCPIIP protocol stack. The parameters of Internet traffic models are updated by parameters given ETSI (European Telecommunications Standards Institute) or 3GPP (3rd Generation Partnership Project) suppositions for the behavior of mobile Internet users. The e-mail model describes the traffic arising with the transfer of message downloaded from a mail server by an electronic mail user. The only parameter is the amount of data per e-mail. The value of 12000 byte as the e-mail size is chosen, since it is assumed that no emails with large attachments will be downloaded on mobile devices. Distribution of e-mail size is log 2-normal. A constant base quota of 300 byte is added to this size. WWW sessions consist of requests for a number of pages (geometric distribution with mean 5). Intervals between pages are distributed with negative exponential distribution (mean 12 sec). These pages consist of a number of objects with a dedicated object s1ze. Another characteristic parameter is the delay between two pages depending on the user’s behavior to surf around the Web. The number of objects per page (2.5 objects) and the small object size (3700 byte) were chosen with geometric and log2-Erlang-k distribution respectively. The Internet traffic is composed of different proportions one consist of 600/o e-mail sessions and 40% WWW sessions and the other consist of 40% of e-mail sessions and 60% of WWW sessions. The inactive period between two sessions is set to 15 sec.
Permanent resource allocation: The Choice of number of timeslots permanently reserved for GPRS (Cd) and timeslot reserved as TCH channels for GSM (Cv) can influence the QoS perceived by users. Since among the T available timeslots, then a number T-Cd-Cv is timeslots shared between PDCH and TCH users.
A new incoming GPRS data user is accepted only if the number of current active GPRS users is below maximum acceptable value, otherwise it is blocked. Simulation in this section can be done by downloading a Web page or E-mail on DL and evaluate the DL throughput for each user in a given TDMA frame as PDCH. The maximum IP throughput in DL is taken as 24 kbit sec-1 and average IP throughput per user is IP throughput in downlink measured during period of transmission.
Throughput is the most important parameter measured in user’s point of view. This measure is evaluated statistically by counting the amount of IP bytes transmitted in each TDMA frame period for each user, if a packet train is running. The throughput is averaged over active periods only. From our defmition a small value of Cv means increased number of voice blocked calls and small value of Cd will affect the QoS requirement for PDCHs. Therefore it’s important to define the QoS averaged over DL throughput that can balance between the call blocking probability and QoS of service that guarantees PDCH data traffic to attain this goal, other parameters need to be considers are:
| Fig. 3a: | Fixed PDCH (E-mail60%. web 60%) |
| Fig. 3b: | Fixed PDCH (E-mail40%. web 60%) |
| • | Approximated number ofiP user per cell (in this case is about 11 0 users) |
| • | The offered IF traffic per user (we have approximated to be 55 kbytes h-1 or can be expressed as 0.12 kbits sec -1). |
| • | Total offered IP traffic per cell ( is given by 11 0 x 0.12 kbits 13.2 kbits sec -1) |
Using the above parameters we can estimate the number of PDCH required and can be found by following equation.
| (1) |
Using graph we first locate the operating point Q defined by the QoS target and total offered traffic. Q is a reference point used to judge whether the corresponding curve falls within the expected region or not. This is done by choosing the proper PDCH curve that lies just above and adjacent to the Q point. The Q point in our simulation is the intersection between 13.4 kbit sec-1 on DL IF thronghput (y-axis) axis and 3.6 kbit sec -1 on Offered load traffic (x-axis) axis of Fig. 3a-b.
| Fig. 3c: | DL traffic with Web 60%. E-mail40% |
| Fig. 3d: | Offered traffic (kbit sec-1) |
Result of simulations: From the simulation result above, we observe that when the percentage composition of Web (60%) sessions is higher than that of E-mail sessions (40%), causes faster decrease of average DL IF throughput. If E-mail curve and Web curve are considered separately their curves appear to be very close (Fig. 3c) to one another, this is due to fact that Web’s TBFs have longer duration compared to that of E-mail’s, so Web sessions tends to saturate channel reserved for FDCH.
When the percentage composition of E-mail (60%) is higher than that of Web (40%). the 2 curves (Fig. 3d) are at distance apart with low E-mail IF throughput. Because E-mail TBFs are shorter so they are created and released very fast, the resources released by E-mail may be taken and occupied by Web TBFs causing E-mail session to starve. However, in our Simulation both case 4 FDCHs and 5 FDCHs lie above point Q; therefore for optimum resource utilization 4 FDCHs are still sufficient to meet the required QoS level. Furthermore, 4 FDCH Curve cuts very close to the Q point (when Web is 60%) results also show different thronghput under different PDCHs.
Dynamic resource allocation: The voice calls are independent ofGFRS connections; the GFRS service time is shorter than the mean call duration. In this section we would like to analyze the GFRS performance and study on how the GFRS connections interact with voice calls. Usually the GRPS users can be dynamically allocated available BW reserved for voice services depending upon the desired level of blocking probability. mean throughput and radio resource utilizations. The simulation was carried out using combination of Web and Email do\Vllloading over GPRS on dynamic chmel allocation scheme. As in the Permanent allocation scheme, in this case too, we have considered PDCHs with 3 TRX and MS with same multi slot capability. To accomplished our simulation we have first established the require operating point Q which is defined by QoS target and total offered traffic for a given blocking probability Pb (Pb 2% in our case).
Let us consider the total capacity of the cell (C). the number of static PDCH charmels (Cs). the number of PDCH charmels dynamically allocated (Cd) and the maximum offered voice traffic (Vm) that corresponding to 2% blocking probability (2% GOS).
Erlang table is used to determine the air interface in GSM. Considering 600 calls per hour with average time duration of call is 90 sec, then by using Erlang table the number ofErlang can be obtained as following
| (2) |
Let P(j) be a steady state probability that j users are in active transfer
and the average overall throughput XT of the resource is obtained as
| (3) |
Where r(j) is effective BW for each user, for j = 1 ... nmax,nmax , is the maximum of PGRS usersin active transfer.
The Blocking Probability is given by
| (4) |
Where
It’s assumed that, the voice calls arrive as a Poisson process at the rate λj.
Adding a TCH charmel for Broadcast Control Charmel (BCCH) and one TCH for the Standalone Dedicated Control Charmel (SDCCH)/8 means 3 TRX units are required (one TRX 8 charmels). Using Erlang table means 22 TCH chmels are needed. Fifteen or more chmels may be reserved for GSM users at any one time to prevent a level of call blocking during rush hours.
| Fig. 4: | DL IP throughputs perMS |
| Fig. 5: | DL IP throughputs perMS (Dynamic Allocations) |
Since 15 channels can provide the voice service required by GSM, then the remaining 7 channels can be reserved for GPRS services. Implementing
CS-2 coding scheme, 90 kbits sec-1 can be achieved without increasing
the number of channels.
From Fig. 4 we observe that as the number of user’s increases, the average performance degrades slowly. The IF throughput depends on the characteristic nature of the traffic generator, for example low data volume such as Email have shorter TBF duration while high data volume such as Web have longer TBF duration thus occupies radio resource for long period of time causing throughput performance degradation. Therefore to achieve optimum network configuration, Q point must be define and traffic load should be carefully examined, then parameters like number of fixed PDCH. max number ofPDCHs that can be dynamic allocated and maximum number of user in a cell. Figure 5 indicates that the Pb = 2% curve lies above the Q point, which means this blocking probability is acceptable.
Combination of static and dynamic allocation: Ithas been observed that dynamic allocation does not provide efficient utilization of chmels due to implementation complexity and signalling effects likewise static allocation result into inefficient resource utilizations because one party may be overloaded while the other is rmder utilizing the resources reserved for it. Thus a combination of both schemes may provide an optimum solution. For operator to guarantee network accessibility, must choose proper fixed number of GPRS and dynamically allocate the rest efficiently.
CONCLUSIONS
In this study we investigate methods that would enable proper estimation of necessary resources, namely number of PDCHs cbonnels. in the GSM/GPRS network taking into accmmt parameters of QoS. We have considered the user’s IF throughput as a basic parameter for QoS requirement. The procedures on how to estimate necessary resources are also given, we have used static, dynamic and combination of both static and dynamic scheme to support the proposed methods. Through our simulation results, we have demonstrated that in order to achieve resources optimization then proper traffic scheduling methods should be implemented, therefore in this paper after careful investigation we have proposed criterion that can be applied for reserving a fixed number ofPDCH for GPRS services and dynamically allocate the rest of the channels for voice and GPRS services. Besides throughputs and blocking probability other factors need to be considered are composition of offered traffic, loss of packets, latency, jitter and the influence of multi-slot capability on network traffic.