Guard Capacity Implementation in OPNET Modeler WiMAX Suite

Transcription

1 Guard Capacity Implementation in OPNET Modeler WiMAX Suite Natalia Vassileva 1,2, Yevgeni Koucheryavy 2, Francisco Barcelo-Arroyo 1 1 Department d Enginyeria Telematica, Universitat Politecnica de Catalunya (UPC) Barcelona, Spain 2 Department of Communications Engineering Tampere University of Technology (TUT) Tampere, Finland natalia@entel.upc.edu, yk@cs.tut.fi, barcelo@entel.upc.edu Abstract Admission Control (AC) algorithms in mobile cellular networks support continuity of active communication sessions by guaranteeing availability of free resources. In this paper, we first describe the factors that condition AC design and show that AC in broadband mobile networks presents new challenges compared to AC in telephone cellular networks. We tailor the guard channel admission control strategy to the characteristics of mobile WiMAX networks. For AC validation and performance evaluation we use OPNET Modeler WiMAX Suite. To this end, we create in the simulator a framework for AC implementation. Then, we incorporate the Guard Capacity (GC) algorithm, which is the preferred choice of telephone cellular networks operators as GC is simple yet efficient. The accuracy of implementation is validated through simulation. Keywords- cellular networks; IEEE ; mobile WiMAX; admission control (AC); guard capacity. I. INTRODUCTION Mobility support plays a central role in wireless cellular networks as it provides untethered access and free movement of subscribers as well as for the creation of new services such as location-based ones. An aspect of the multifaceted mobility problem is delivering continuous service to Mobile Stations (MSs) with on-going calls. MS with active transport connections requests service at each of the Base Stations (BSs) that lie on MS s trajectory. If all system resources are busy, MS s connections can not be served and as a result are interrupted. Thereby, lack of free BS capacity leads to forced session termination, which is against the intrinsic, for the considered cellular networks, mobility concept. Admission Control (AC) algorithms support continuity of active communication sessions by guaranteeing availability of free resources. AC logic is executed at each BS to give priority to pre-existing over new connections. Mobile cellular networks are intended to provide the same QoS as their wired counterparts while supporting free movement of subscribers. Hence, as in the wired networks, the system must support uninterrupted service and maintain low ratio of blocked new traffic. QoS in this context is measured by Probability of Blocking (PB) of a new communication session (that is, probability of rejecting a new request) and Probability of Failure (PF) of a communication session handed over from neighbour BS (probability of interrupting an active session). Resource reservation and handover traffic prioritization is done at the price of blocking a portion of the new arrival traffic. Therefore, when AC is executed the Carried Traffic (CT) is less than the CT when all connection requests are accepted unconditionally. PB is a metric of interest to network operators because it is a measure of the efficiency of system utilization and hence of operator s revenue, whereas PF is a measure of MS s perceived QoS. Decrease in one of the two probabilities usually leads to some increase in the other. Consequently, the efficiency of AC algorithms is commonly evaluated by the achieved trade-off between PB (rejected traffic that otherwise can be carried by the system) and PF (interrupted traffic due to lack of free resources). Many contributions have been made in the area of AC for the traditional telephone networks (see for instance [1-6] and references therein). However, there are not sufficient AC studies for the emerging mobile cellular networks with Broadband Wireless Access (BWA) that are the focus of this work. It was shown in [7] that the problem of resource reservation/call prioritization for mobility support of active sessions in BWA networks is different from the one in voiceoriented networks and that traditional AC algorithms should be adapted to the specifics of the broadband technology. We concentrate on Worldwide interoperability for Microwave Access (WiMAX), based on IEEE family of standards, as it is one of the new generation BWA technologies that already has several practical implementations in Europe, Asia and America. WiMAX offers high data rates and multimedia services with a wide range of QoS metrics. The latest IEEE standard [8] embraces both fixed (last-mile) and mobile point-to-multipoint broadband wireless scenarios. The latter facilitates the release of broadband mobile cellular networks. In this paper, we discuss the factors that condition AC in BWA networks and point out new challenges that must be successfully addressed in order to design a pertinent AC for WiMAX. Then, we introduce the guard capacity concept. The idea of reserving portion of system capacity (guard channels) to

2 high priority traffic is not new. The guard channel methods proposed in the past, however, had been designed for telephone cellular networks. In the work presented here, we have tailored the algorithm to the characteristics of the broadband wireless networks, WiMAX in specific. Our goal is to examine system performance and validate algorithm s efficiency under timevarying system and service capacity typical conditions for WiMAX environment. To this end, we have first created a framework for AC implementation in OPNET Modeler WiMAX Suite and then we have implemented the Guard Capacity (GC) scheme. We show that if simplicity of implementation is sought, GC method is a valid approach that can be efficiently used in mobile WiMAX networks. The reminder of the paper is organized as follows. Section II compares broadband to telephone mobile cellular networks to show that the technology implemented in the network conditions AC design and lists the new AC challenges encountered in BWA. Section II comprises also scheme that represents AC unit and its interplay with the system. Section III is devoted to GC algorithm concept definition, framework for AC implementation in OPNET and GC implementation in the simulator. Section IV comprises the simulation verification, experiment and results. In Section V we briefly discuss the traffic conditions for which AC is essential for achieving required QoS levels. Section VI concludes the paper and describes future work. II. ADMISSION CONTROL A. Broadband vs. Telephone Mobile Cellular Networks The features of telephone and broadband mobile systems that are relevant to AC design are discussed next. The discussion shows that whereas a simple and abstract model can be created for the conventional systems, such model is not suitable for BWA systems. Comparison of those networks (see Table 1) demonstrates also that there is not a direct, immediate correspondence between AC in voice-oriented and AC in multimedia cellular networks. The comparison points out the need for tailoring AC algorithms to the specifics of the broadband technology. Traditional cellular networks are voice-oriented. They are characterized with hard capacity, i.e., capacity that does not change with time. Moreover, the resources that are assigned to a call are also fixed, that is, remain constant for the duration of the call. Since there is only one type of service (voice), each accepted connection uses the same amount of resources. The system (i.e., a cell/bs) thereby, naturally lends itself to the notion of channel. Channel refers to the resources assigned to a call in service. It is used to abstract the underlying technology (TDMA/FDMA) and allows for a simple model, which represents BS capacity by a number of channels, C, as shown in Fig. 1. This model greatly facilitates the design of AC algorithms because at any instance of time system state can be conveniently represented by the number of free and busy channels. Broadband wireless systems are designed to support a wide range of services. The IEEE standard [8] for instance defines five scheduling services to accommodate all the PD probability of dropping a handover request in a system, i.e., at a BS PF probability of forced termination of a handover call in the network PB probability of blocking a new call request Figure 1. Simple abstract model of a BS in a telephone cellular network. existing and future application services. Each of these classes has different bandwidth and delay requirements. In addition, the resources used by a call 1 are time-varying. Different system resources might be requested for the duration of a call mainly due to two factors. Change in the requested application bit rate causes change in required system resources. However, even constant bit rate services such as those served by the Unsolicited Grant Service (UGS) scheduling class might have varying resource needs defined by PHY channel conditions. Adaptive Modulation and Coding (AMC) scheme is introduced in WiMAX to combat adverse or adjust to good channel conditions. When MS is experiencing good CINR (carrier to interference noise ratio) for instance, a less robust Modulation and Coding (MC) scheme is used, which results in higher bit rate per slot and in a release of some of the previously assigned resources (slots). The abstract term channel used to denote a fixed number of resources assigned to a call is not applicable in a system where the resources required by a call are timevarying. B. New Challenges Broadband environment, WiMAX in specific, defines new challenges with regard to AC design (see Table 1): System capacity computation System capacity can be expressed in number of slots (the basic unit in OFDMA-based systems). However, slot capacity (bytes per slot) depends on the MC scheme used, therefore system capacity can not be expressed in number of calls that can be simultaneously served by the system as it is for conventional voice networks. Call prioritization/capacity reservation under timevarying resource conditions The time-varying nature of the needed resources complicates the computation of the resources that have to be allocated to a call to meet its call-level and packet-level QoS requirements.

3 TABLE I. BROADBAND VS. TELEPHONE MOBILE CELLULAR SYSTEMS Resources Type Capacity Services per per call service Telephone hard voice const const Broadband soft Voice, video, data, remote control, etc. depend on application and PHY conditions variable Whereas the question of resource reservation in conventional cellular networks concerns only handover calls, in next generation networks it is extended to resource reservation for pre-existing calls comprising handover and calls initiated in the same cell. We highlighted earlier the time-varying nature of the resources used to serve a call in broadband networks. A communication session that has been started while MS experienced good channel conditions will require additional resources if the channel conditions are worsened later. As a result of the changed PHY channel, the BS will issue more robust MC scheme and eventually more OFDMA slots will need to be provided to the call to support its packet-level QoS requirements. If there is not sufficient BS capacity to accommodate the newly set resource requirements the call will be dropped. Therefore, in BWA resource reservation for delivering uninterrupted service should consider both newly set and handed over calls. C. AC Scheme Fig. 2 presents the AC unit and its interaction with the other functional blocks of the system. When a call is received at a BS, it is first classified. After defining the scheduling class to which the call belongs, the number of required resources to guarantee given QoS is defined. The computation of the required service capacity is based on the service class and MS s current physical layer parameters. Then, AC performs a simple capacity check to see if it can accommodate the call. Free capacity check involves computation of required slots accounting for control in addition to user data slots as well as defining used and free slots. Next, AC block checks if QoS of present sessions is maintained after new call acceptance, a task typical for fixed access networks (e.g., WiFi). The AC typical for conventional cellular systems, i.e., prioritization of active calls, is performed too. The scheduler orders the slots so that QoS metrics of every accepted session are met. New call is blocked if no resources are available or due to active call prioritization. Existing calls are dropped only if there is no place for them at the system. AC unit and scheduler are executed at the beginning of each frame to dynamically react to system changes as explained before. AC assigns slots per call whereas scheduler orders the slots to meet QoS requirements. III. GUARD CAPACITY IMPLEMENTATION A. Guard CapacityConcept The simplest strategy for call prioritization is to set aside portion of system capacity for high priority traffic. The common part of the capacity is shared between high and low priority traffic. The capacity used only by high priority calls is called guard because it protects these calls from blocking when the system is overloaded, i.e. when the common resources are busy. There are different strategies with regard to the access to the guard resources common capacity first, guard capacity first or random access until the common capacity is used that can be applied to achieve different level of prioritization. The guard channel scheme was first proposed by Hong and Rappaport [2] to address the mobility problem related to lack of resources and consequently call disconnection. The guard channel concept has been extensively studied analytically and through simulation (e.g., [6]) for the conventional cellular networks because it is simple yet efficient and easy to implement in practice. There are few studies, however, that consider the mobility problem in WiMAX. In [9], the guard channel concept is applied to WIMAX environment. The algorithm proposed in [9] sets four thresholds to give priority in descending order from UGS scheduling class (highest) to BE (lowest). There is PB probability of blocking a new call request PD probability of dropping an active call Figure 2. AC unit and its interaction with the system

4 no threshold set for handover calls as these are accepted as long as there is free system capacity. However, [9] has one major flaw although the scheduling services are taken into account, the variability of the system and service resources in WiMAX are not considered. The authors of [9] assume that the system can be divided into a number of channels, where the resources needed for service are called channel (as in voice-oriented networks). Moreover, the variable resource needs of a call are not considered, which facilitates the analysis, design and evaluation of the system but the system model then is not realistic. The assumption of fixed service resources implies as well that the probability of dropping a new call (on-going call initiated in the cell) is omitted. System capacity can be expressed in a number of slots. However, the slot capacity depends on the modulation and coding scheme used as explained before. Hence, it is not straightforward to compute the number of connections that can be served by a system at a given moment. Therefore, we propose the following simple approach. A certain percentage of each subframe capacity is reserved for on-going connections. The percentage of reserved subframe capacity depends on the arrival handover and carried traffic. We follow a simple strategy for computing in advance the resources that are needed to serve a call. The strategy uses the service requirements and the most robust MC scheme for calculating the maximum number of slots that can be requested. Later, when the call arrives or its resource requirements change (e.g., more favourable PHY channel), the slots that are left free (if any) are assigned to BE or to delay-tolerant traffic to achieve better bit rate. The calculation of the control capacity depends on the scheduling type of the call and on the PHY metrics. B. AC Framework in OPNET WiMAX Module Suite Two modifications have been made to OPNET WiMAX Suite in order to create a framework for AC implementation. Call classification First, a mechanism for call classification has been introduced. A basic operational metric used by AC algorithms is the type of the call. AC algorithms apply different criterion for call acceptance/rejection based on whether a new call request or a request for handover is received. communication mechanism allows TBS to take AC decision about the service flows of MS before MS starts the actual handover to the target BS. This mechanism prevents/postpones the instant when forced handover termination due to fully loaded TBS is experienced because the load information is provided to the SBS and eventually to MS in advance, before MS starts initial ranging procedure with TBS. The MS in that case can use the scanning information about neighbour BSs gathered during the scanning phase to check for other suitable TBS. If such TBSs are available MS can handover to them. The MS context information is interchanged between SBS and TBS via ASN and we set the status of the MS handover as a flag in the control messages. We also introduced a mechanism for call request classification when ASN mobility is not supported. According to the standard, when MS performs a hard handover MS disconnects from the SBS (sends control MOB_HO_IND message to SBS) before connecting to the TBS. When hard handover is executed the procedure of MS association with TBS is the same as the one for initial network entry. Therefore, some indication of the nature of the request was required. To this end, we have included in the RNG_REQ control message (after CDMA allocation has been received at the MS) a flag that indicates the mobility status of the MS (see Fig. 3). Metrics for performance evaluation AC performance is commonly evaluated by PB and PF. We have extended the output statistics of the simulator (which supports packet-level measures besides others) by introducing the following metrics that are recorded during simulation execution: total number of served/blocked new/handover connections. These metrics are used to calculate the probabilities of interest as follows: # new _ connections _ blocked PB =, (1) total # new _ connections # existing _ connections _ dropped PD =, (2) total # existing _ connections where total # i _ connections = # i _ connections _ served + # i _ connections _ dropped (3) Figure 3. Ranging request control message The ASN anchored handover that is available in the simulator enables the interchange of MS-context information between the Serving BS (SBS) and Target BS (TBS) transfer of service flows from MS and transfer of data queues. This PF probability of interruption of an active call on the move due to lack of free resources in some of the cells that MS will visit while communicating can be computed if the mobility ratio (α), i.e. the average number of handovers (visited cells) per call, is known. PF is calculated then according to the following equation: PF = 1 (1 PD) α. (4)

5 C. GC Implementation in OPNET In order to evaluate guard capacity concept in mobile WiMAX environment, the BS node in OPNET has been modified accordingly. The guard capacity was designed as a user-defined metric by modifying the BS attribute Reserved XL Subframe Capacity. The attribute is used in OPNET to partially block downlink and/or uplink subframe capacity. For the purpose of GC algorithm implementation we modified this attribute to serve as a threshold that restricts the acceptance of new call requests. When MS sends a call request, its mobility status is flagged in the request. When the call request is received, the AC first checks whether it is of new or handover type. If the connection request is received from a moving subscriber with active connections it is treated preferentially as long as there are enough resources (part of the) MS s connections are accepted and served by the BS. If the received request is from new MS, BS checks whether the threshold has been crossed or not. In the former case the request is accepted in the system; in the latter it is rejected. Fig. 5 specifies the steps that are followed by the GC logic when a call request arrives at the system. Our modification concerns threshold setting and AC logic implementation accept new requests if below the threshold, otherwise block them; handover requests are accepted whenever free resources are present. The computation of required service symbols is part of OPNET WiMAX Suite and has not been modified by us. IV. SIMULATION EXPERIMENT AND VERIFICATION The following simple experiment was set to verify GC implementation in OPNET. The simulation scenario consists of three cells (Fig. 6). Each cell has at least one stationary MS that requests VoIP service. There is one mobile subscriber that also requests VoIP connection. MS moves from the coverage of one BS (BS_1) to the coverage of a neighbour BS (BS_2) for the duration of the simulation. MS uses AMC scheme and ertps scheduling service, whereas the stationary MSs use fixed modulation and coding scheme (16-QAM 1/2) and UGS scheduling service. The threshold of BS_2 is set to 90%, that is, the reserved capacity for handover connections of downlink (DL)/and uplink (UL) subframes is 90% from the total subframe capacity. The threshold was purposefully set to an elevated value. High threshold (respectively low common subframe portion) forces blocking of new connections even if there is free room in the guard portion of the subframe. Therefore, we expect to see blocked new connections in BS_2 even in light load scenarios. We examine system performance when GC is used and compare it to the case when the capacity of the visited BS is completely shared, i.e., common for both new and handover calls. When no reservation for handling handover connections is made all the connections of the stationary MS_3 (in the coverage area of BS_2) as well as all the connections (transport and data) of the moving MS_1 are accepted in BS_2. However, when guard capacity is used for guaranteeing uninterrupted service to mobile connections, both UL and DL voice Figure 4. Example of input values of BS s Reserved XL Subframe Capacity attribute. connections of MS_3 are rejected and those of MS_1 accepted as expected. The described simulation experiment was used to check the accuracy of GC implementation. Additional simulations are carried out to evaluate the efficiency of GC in WiMAX environment in terms of PB, PD, PF and carried traffic as pointed out in Section VI. V. ENVIRONMENT DISCUSSION A relevant question concerning admission control implementation in broadband wireless networks is under what conditions AC algorithms are needed. As discussed in the introductory part of the paper telephone and multimedia (the later refers to BWA as their role is to support a broad spectrum of services) cellular networks provide different environment for AC implementation and condition the AC design. There are also substantial differences with regard to the traffic conditions under which AC is relevant in next generation networks. compute free_capacity_symbols free_capacity_symbols = num_symbols_free_per_xl num_symbols_reserved_per_xl; if free_capacity_symbols > 0 then compute required_service_symbols determine scheduling service type compute needed resources based on the scheduling service (required data rate), use the most robust MC scheme add MAC overhead add PHY overhead convert the computed capacity needed to serve the connection into symbols if required_service_symbols < free_capacity_symbols else reject connection else reject connection then accept connection request Figure 5. Pseudo code of GC algorithm implementation in OPNET WiMAX Suite.

6 Figure 2. Simulation scenario. Traditional cellular networks are voice-oriented, which characteristic besides the simplifications mentioned before (see Section II) yields to yet another one. Under moderate and heavy load the system is busy most of the time. Consequently, the probability of interrupting a call in progress (PF) is high. The need for AC that regulates the acceptance rate of new calls guarantees free capacity for on-going calls - is evident, as AC in these traffic conditions is the only approach for providing QoS support. Are these conditions relevant to BWA networks however? BWA are expected to accommodate the whole range of real and non-real time services as well as best effort traffic; that is, broadband traffic that behaves very differently from voice traffic. The capacity portions used by different services are as important in BWA as the current load of the system. Traffic distribution is a factor of major influence because BE traffic for instance can be used to accommodate high-priority traffic. Consequently, even if the cell is fully loaded the BS can still dynamically respond to capacity changes (due to varying PHY layer conditions or arrival of handover requests) when the portion of BE traffic is large. Therefore, AC for BWA sets yet another new question what are the traffic conditions under which AC is required? The most reliable answer can be provided by statistical data, however, WiMAX has not been in exploration for long enough time (nor has its main competitor LTE). The only metropolitan broadband cellular network, WiBro, in Seoul, South Korea dates from It can be argued however, that the peak values of voice traffic that are evidenced in telephone networks during the busy hour will be maintained in BWA that offer voice service on competitive prices and that voice traffic will be a substantial part of the arrival traffic. Under these conditions there is a clear need for AC execution for providing call and packet level quality guarantees. VI. CONCLUSION AND FUTURE WORK Admission Control (AC) for delivering uninterrupted service to MSs with active communication sessions in broadband mobile networks has been the focus of this paper. We have pointed out the factors that condition AC design and have shown that existing AC proposals for conventional telephone cellular networks need to be adapted to the specifics of the broadband technology. Then, we have tailored Guard Capacity (GC) algorithm to serve mobile WiMAX networks. GC algorithm has been implemented in OPNET Modeler WiMAX Suite to allow its performance evaluation. We have validated the accuracy of the implementation through simulation experiment. In future work, we intend to evaluate the efficiency of the GC algorithm under realistic traffic conditions. GC performance evaluation includes call-level and packet-level QoS metrics as well as volume of lost traffic due to resource reservation. Another relevant question that we intend to address by simulation experiments is detecting traffic conditions under which AC are indispensable for guaranteeing QoS to active communication sessions. In future work, we also intend to introduce our AC solution that deals with call dropping probability of pre-existing calls, i.e., handover as well as newly originated on-going calls. ACKNOWLEDGMENT The authors thank Dr. Alexandar Sayenko and Dr. Dirk Staehle for the constructive discussions and especially for pointing out the open question of defining traffic conditions under which AC implementation is relevant in mobile cellular networks with broadband access. This research was funded by the Spanish Government and FEDER through the Plan Nacional de I+D (TEC /TCM). REFERENCES [1] E. C. Posner, R. Guerin, Traffic policies in cellular radio networks that minimize blocking of handoff calls, in Proc. 11th Int. Teletraffic Congress, pp , [2] D. Hong, S. S. Rappaport, Traffic model and performance analysis for cellular mobile radio telephone systems with prioritized and nonprioritized handoff procedures, IEEE Trans. Veh. Tech., no. 3, pp.77-92, [3] R. Ramjee, R.Nagarajan, D. Towsley, On optimal call admission control in cellular networks, in Proc. IEEE INFOCOM, pp , [4] Y.-B. Lin, S. Mohan, A. Noerpel, Queueing priority channel assignment strategies for PCS hand-off and initial access, IEEE Trans. Veh. Tech., no.3, pp , [5] M. Kulavaratharasha, A. Aghvami, Teletraffic performance evaluation of microcellular personal communication network (PCNs) with prioritized hand-off procedures, IEEE Trans. Veh. Tech., pp , [6] A. E. Xhafa, and O. K. Tonguz, Handover performance of priority schemes in cellular networks, IEEE Trans. Vehicular Technology, vol. 57, vol.3, no.1, pp , [7] N. Vassileva, Y. Koucheryavy, and F. Barcelo-Arroyo, Admission Control for Supporting Active Communication Sessions in Mobile WiMAX Networks, in Proc. 3 rd ERCIM Workshop on emobility held in conjunction with WWIC 09, Enschede, The Netherlands, May [8] IEEE Std : Air Interface for Fixed and Mobile Broadband Wireless Access Systems. [9] K. F. Tsang, L. T. Lee, H. Y. Tung, R. Lam, Y. T. Sun and K. T. Ko, Admission Control Scheme for Mobile WiMAX Networks, in Proc. IEEE Int. Symposium on Consumer electronics (ISCE 07), Dallas, TX, USA, June 2007.