WLAN Call Admission Control Strategies for Voice Traffic over Integrated 3G/WLAN Networks

Transcription

1 WLAN Call Admission Control Strategies for Voice Traffic over Integrated 3G/WLAN Networks Alessandro Bazzi, Marco Diolaiti, Claudio Gambetti, Gianni Pasolini IEIIT-BO/CNR, DEIS-University of Bologna, v.le Risorgimento 2, 4136 Bologna, Italy. Siemens s.p.a., via Monfalcone 1, 292 Cinisello Balsamo (Milano) Abstract Heterogeneous networks integration and interworking, with particular reference to third generation (3G) cellular networks and wireless local area networks (WLANs), is a challenging issue for the early future wireless communications world. In our researches we consider the opportunity to convert those circuit switched voice flows that would be blocked by UMTS for resource saturation into voice over IP streams to be served through the WLAN connection. In order to increase voice users satisfaction when WLAN available resources are critically falling off, in this paper we investigate, by means of a simulation approach, different WLAN call admission control strategies within the context of 3G-WLAN integrated networks. I. INTRODUCTION Heterogeneous networks integration and interworking, with particular reference to third generation (3G) cellular networks and wireless local area networks (WLANs), is a challenging issue for the early future wireless communications world. Provided that the WLAN hot spot is within the 3G network coverage and that the final user is equipped with a dual mode terminal, integrating the two technologies, thus increasing the pool of available resources, would considerably increase both users satisfaction and networks utilization efficiency. In the last few years several projects were dedicated to the issue of 3G/WLAN interworking networks (e.g., [1]) and many papers appeared, investigating specific aspects mainly related to architectural/signalling issues (e.g., [2]) and vertical handover (e.g., [3]). Besides the data users exchange between UMTS and WLAN, in this paper we focused our attention on the possibility to serve by means of the WLAN technology those voice calls that within the hot spot region cannot be served by the 3G network (and would therefore be blocked) in case of saturation of its radio resources. Obviously a format conversion from circuit-switched (CS) voice flows into packet-switched (PS) Voice over IP (VoIP) flows (and vice versa) is needed in order to allow the radio network switching (see [4] for details on the network architecture). The benefits of 3G/WLAN interworking are here investigated with reference to the Chinese TD-SCDMA UMTS This work was carried out in the framework of VICOM Project funded by MIUR. technology [5] and to the upcoming IEEE82.11e [6] WLAN technology. In this paper, in particular, the users satisfaction level provided in a realistic scenario by the integrated network is assessed by means of a dedicated simulation platform. Three simple but effective call admission control (CAC) strategies for the WLAN are implemented in our simulation platform in order to prevent WLAN overloads. Their performance have been derived and compared. The paper is outlined as follows. In section II the WLAN CAC strategies are described, in section III the scenario under investigation is introduced; in section IV and in section V the simulations settings and the related numerical results are reported and finally, in section VI the final conclusions are drawn. II. WLAN CALL ADMISSION CONTROL STRATEGIES To prevent WLAN resource saturation, thus avoiding the consequent poor performance level provided to WLAN users, a call admission control strategy is needed. Here we investigated three CAC strategies based on a centralized evaluation of the current network congestion level. This is done monitoring the channel occupation through the assessment of the Channel Occupation Rate parameter, C O [7], defined as the ratio between the amount of time the medium is busy, T B, and the related observation time T : C O = T B T. The evaluation of the channel occupation rate is particularly simple to be implemented in existing Access Points (APs), owing to their carrier sensing capability. The AP simply adds the busy medium sensed time T SB to its transmission time T AP (T B = T AP + T SB ). In order to improve the accuracy of the estimation of T B, also the mandatory idle times SIFS and AIFS [6] have to be considered in the assessment of T AP and T SB. The CAC strategy denoted in the following as two thresholds CAC [8] foresees that when a congestion state is detected, that is, when C O exceeds a given congestion threshold C T, incoming users are blocked or, if user redirection is admitted,

2 6 5 Potential Voice users Potential Data users Voice satisfied, no CAC Voice satisfied, Voice satisfied, Voice satisfied, + Potential Voice Users No CAC Potential Data Time (sec) Fig. 1. Performance of the WLAN CAC strategies. TABLE I SET OF PARAMETERS ADOPTED FOR IEEE 82.11E. Voice WWW Ftp AIFS CWmin CWmax AIFS CWmin CWmax AIFS CWmin CWmax they are redirected to UMTS. The congestion state ends when C O goes below a decongestion threshold D T, with C T >D T. In case of traffic redirection, in order to give to VoIP users (coming from UMTS) a priority higher than data users, a four thresholds CAC can also be adopted. In this case two different congestion events are defined: when a first congestion threshold (C T 1 ) is reached, only VoIP users can be admitted to the WLAN, while data users are redirected to UMTS, which, on its own, will perform its CAC strategy according to its available resources. When C O exceeds a second critical congestion threshold, C T 2 (with C T 2 >C T 1 ) all further incoming users are redirected. The two congestion states end when C O goes below the decongestion thresholds D T 1 and D T 2, respectively, with D T 1 <C T 1 <D T 2 <C T 2. In the following we will denote this CAC strategy as four thresholds CAC. Optionally, each time the WLAN enters in the second congestion state (when all users are blocked), the data user experiencing the lowest throughput can be redirected to UMTS, thus leaving room for VoIP users. While being in the heaviest congestion state, a further user is also redirected each time C O exceeds C T 2 for 1, even non consecutive, T time intervals. This strategy will be denoted in the following as four thresholds CAC with data user redirection (in figures: +). In all cases an user redirected to a system and further blocked is definitively rejected. In figure 1 the effectiveness of the above described CAC strategies is investigated simulating a single IEEE82.11e WLAN hot-spot. Since IEEE82.11e specifications [6] define only the MAC level strategies, in this paper IEEE82.11a [9] is assumed at the physical layer. Parameters of table I are adopted for IEEE82.11e MAC level protocol (refer to [6] for their meanings). Fig. 2. Simulated scenario. In figure 1 the two curves denoted as Potential Voice and Potential Data represent the number of voice and data users (which are supposed to be all at 2m from the AP) requiring the service, plotted as a function of the time. As can be observed we considered a dynamic traffic scenario with some sessions naturally terminated. The curve denoted as - no CAC represents the number of satisfied voice users when no CAC is adopted. As can be seen, if no CAC is performed VoIP users are satisfied until their number reaches a given threshold (which depends also on the physical level performance and, therefore, on the distance from the AP), afterwards the network is overloaded and a satisfaction breakdown of all users is observed. The three dashed curves in figure 1 represent the amount of satisfied VoIP users for each one of the three CAC strategies above described. As can be noted, no performance breakdown is observed in all cases; moreover the four thresholds CAC with data users redirection proved to be very efficient, maximizing the number of satisfied VoIP users. To derive the curves of figure 1 and those reported in the numerical results section we assumed D T =.6 and C T =.75 for the two thresholds CAC and D T 1 =.6, C T 1 =.65, D T 2 =.7, C T 2 =.75 for both the four thresholds CACs. T =.25s in all cases. III. SCENARIO, SERVICES, TRAFFICS AND NETWORKS INTERWORKING STRATEGIES Hereafter we reported the assumptions we made in order to reproduce realistic operating conditions in our simulations. Scenario. The scenario investigated is depicted in figure 2 and consists of 18 UMTS Nodes-B with tri-sectorial antennas over an area of 372x2775 m 2 which also includes a WLAN hot spot. The WLAN AP is assumed co-located with a Node- B (one of the central ones) and is equipped with a dipole antenna offering an omnidirectional coverage in the horizontal plane; to avoid border effects all merit figures investigated in the numerical results section refer to an area of 1x1m 2 centered in the AP (the grey square in figure 2). Services. In order to reproduce the variety of services provided in a real scenario, here we assumed that users could perform voice calls as well as web browsing (hereafter, WWW)

3 and file transfer sessions (hereafter, FTP), thus generating three different kinds of traffic whose statistical description have been carefully reproduced, as detailed in the following sections. Traffic distribution. As for the geographical distribution of traffic, here we considered the superimposition of a background traffic generated by users uniformly distributed in the whole 372x2775 m 2 region, and of a hot spot traffic generated by users uniformly distributed in a circular area (the hot spot region) centered in the AP, with a radius of 25m, which could represent a highly crowded area, e.g., an airport gate, where the WLAN hot spot is deployed (see figure 2): Background traffic. It is generated by background users that are assumed to perform phone calls (generating voice traffic) or WWW sessions. Hot spot traffic. It is the further traffic contribution which is added to the background traffic only in the hot spot region and it is constituted by an additional amount of voice, WWW and FTP traffics. Networks Interworking Strategies. Here we assumed that WWW and FTP users within the hot spot are preferably served by the WLAN; in case the connection request is refused by the WLAN CAC, WWW and FTP users are redirected to UMTS (which can, on its own, refuse the connection according to its CAC strategy). Voice calls within the hot spot are, on the other hand, served by UMTS, if there are available resources, otherwise they are converted into VoIP flows and redirected to the WLAN (which can refuse the connection according to its CAC strategy). Outside the hot spot voice and data users are obviously served by UMTS and are blocked in case of resource saturation. When considering, for comparison purpose, the case of separately working networks, no traffic redirection is performed, hence in case of lack of resources in one of the networks the related service requests are simply blocked. The above reported assumptions are summarized in the first three columns of table II, while in the fourth column the transport level protocols we adopted in our simulations are given. A. Merit figures IV. SIMULATIONS SETTINGS For each kind of service the main merit figure investigated is the percentage of users experiencing a satisfactory service level, which is assessed on the basis of performance indicators depending on the particular kind of service; let us define, in particular: average perceived throughput: it is the average throughput perceived by users at application level; delivery delay: it is the time interval occurring from the generation of an application-level packet in the transmitter to its successful reception; outage: an outage event for circuit switched voice calls (i.e., those served by UMTS) occurs when the average value of the bit error rate (that is, BER) exceeds the threshold BER out = 1 2. B. Quality requirements and traffic settings Since voice users generate either CS traffic (when served by UMTS) or packet switched (PS) VoIP flows (when served by the WLAN) in our simulations we considered four classes of traffic, whose characteristics, references and requirements are detailed in table III. As for the satisfaction requirement of VoIP traffic, according to [1] we assume that the maximum tolerable one-way transmission time is 15ms. Since the voice/voip conversion is assumed to be performed by a G.729 codec (see [1]) which takes around 12ms (including coding, decoding, bufferization delay, etc.) and assuming that the CS Core Network introduces a negligible delay, the maximum tolerable delay introduced by the WLAN is 3msec, as reported in table III. C. Network assumptions WLAN. As for the MAC protocol of IEEE 82.11e, here we considered the contention-based access mode (Enhanced Distributed Coordinated Function EDCF, see [6] for details). Let us recall that IEEE82.11e differs from IEEE82.11a, as well as from IEEE82.11b/g, since it introduces the concept of Quality of Service in the frame delivery procedure at the MAC level. According to table I the VoIP traffic is given the highest priority, while FTP traffic is given the lowest priority. UMTS. At top of Layer 2, Radio Resource Control (RRC) block implements call admission control: a new radio link is successfully setup provided that the necessary spreading codes are available, the estimated interference is less than a given threshold and the initial power required in Node-B is available. Two classes of traffic are supported: CS Adaptive Multi Rate speech at 12.2 Kbps and PS Best Effort at 64/384(uplink/downlink)Kbps. Network delay. In order to carry out realistic throughput evaluations, the impact of the packet switched wired network (the Internet) should not be neglected, especially in terms of packet delivery delays; here we assumed that a constant delay of 2ms is introduced on WWW and FTP packets by the Internet in both directions. This value has been chosen as the average delay we measured in a number of Internet connections. Since VoIP packets are converted into a CS flow, thus perceiving a negligible core network delay, no further delay is added to VoIP packets transmissions other than the previously mentioned coding/decoding delay. V. NUMERICAL RESULTS In order to derive the performance of the UMTS-WLAN integrated network, we developed a simulation tool aimed at reproducing the behavior of the two interworking technologies (as detailed in [4]). In the following all merit figures are evaluated as a function of the rate of arrival of new voice calls in the 1x1 m 2 area under investigation (see figure 2), hereafter denoted as R c, having fixed the rates of arrival of voice and WWW background traffics (Voice backgr., WWW backgr. ) and the rates

4 TABLE II TRAFFIC SCENARIO (CS STANDS FOR Circuit Switched, HS FOR hot spot) Traffic Area Served by Transport level Average arrival rate (Poissonian distribution) Voice backgr. whole scenario UMTS (otherwise WLAN in the HS) CS (UDP if VoIP) 2.7 calls/s WWW backgr. whole scenario UMTS out of the HS, preferably WLAN in the HS New Reno TCP 1.8 sessions/s Voice HS hot spot UMTS (otherwise WLAN) CS (UDP if VoIP) varying WWW HS hot spot WLAN (otherwise UMTS) New Reno TCP.4 sessions/s FTP HS hot spot WLAN (otherwise UMTS) New Reno TCP.12 sessions/s TABLE III ADOPTED TRAFFIC CLASSES: PARAMETERS AND REQUIREMENTS FOR SATISFACTION. Class Characteristics Refer to Satisfaction thresholds CS Voice Poissonian duration, 12 sec in average Natural conclusion with outage lower than 5% in each direction VoIP WWW FTP CBR traffic, 2 bytes packets with rate 8 kbit/s per direction 1 to 8 packet calls of 7 kbytes in average, divided into 1 to 3 packets each; 6 sec average reading time 1 to 6 packet calls of 5 kbytes in average; 18 sec average reading time [1] (codec G.729) 97% of packets received in less than 3 ms (wireless link) in each direction [11] with α =.6, average reading time halved Each packet call follows [12], with µ = 13.6 and σ =.12 9% of packets received in less than 5 sec Average downlink throughput of 1 kbit/s Percentage of satisfied voice users Fig R c (calls/sec) Voice in the investigated 1x1 m 2 area: users satisfaction. of arrival of the additional FTP and WWW traffics in the hot spot region (WWW HS, FTP HS ) according to the values reported in the fifth column of table II. Figures 3 to 5 show the impact on users satisfaction of voice calls redirection in the investigated area. The case of no redirection, that is of simple call blocking, is also investigated for comparison purpose. A voice user is assumed not satisfied whether its call is blocked or it perceives a poor quality of service (in the terms defined in the fourth column of table III). Please note that in the case of call redirection, with the expression voice users we denote indistinctly both voice users served by UMTS and VoIP users served by the WLAN. All results were obtained simulating over 48 seconds of network activity (simulation outcomes related to the first 6 simulated seconds were not considered to avoid transitory effects). In figure 3 the percentage of satisfied voice users is reported as a function of R c ; the solid curve refer to the case of no traffic redirection in which all voice users are served, if there are sufficient resources, by UMTS, otherwise they are blocked; the three dashed curves refer, on the contrary, to the cases of traffic redirection and are related to the cases of two thresholds CAC, four thresholds CAC and four thresholds CAC with data users redirection, respectively. As is intuitive, an increase of voice service requests leads to a decrease of the percentage of satisfied voice users; it can be noted, however, that redirecting to the WLAN those voice calls rejected by UMTS allows to serve with a satisfactory performance level up to 45% more voice users. As expected, from the viewpoint of voice users the CAC strategies with four thresholds outperforms the one with only two thresholds, which do not differentiate between VoIP and data users. What is less intuitive at the first glance, is that the gap between the satisfaction curves related to the two CACs with four thresholds (with and without data users redirection) decreases as R c increases, until the convergence of the two curves is reached. This behavior is due to the fact that for high values of R c most of the WLAN traffic is constituted by VoIP users hence no significant gain can be achieved redirecting data users. In order to get a complete picture of the impact of traffic redirection on the performance of the integrated network, also the satisfaction level experienced by data users has been investigated as a function of R c ; the outcomes of this investigation are reported in figures 4 and 5. In figure 4, in particular, the percentage of satisfied WWW users is reported as a function of R c. Looking at the uppermost curve, which is related to the case of no redirection, we can observe that in the considered operating conditions the amount of traffic load for the WLAN is such that all WWW users perceive a satisfactory performance level. This choice has

5 Percentage of satisfied web browsing users Average Ftp throughput (Mbps) R c (calls/sec) R (calls/sec) c Fig. 4. WWW in the investigated 1x1 m 2 area: users satisfaction. Fig. 5. FTP average throughput. been performed in order to clearly highlight the performance degradation introduced by the redirection of voice users from UMTS to the WLAN. In particular, observing the three lowermost curves of figure 4, which refer to the case of traffic redirection, we can argue that the redirection of voice calls leads to a degradation of the percentage of satisfied WWW users (which, let us recall, are preferably served by the WLAN); it can be observed, moreover, that the adoption of the four thresholds CAC leads to a significant performance degradation with respect to the case of two congestion thresholds. This is the drawback consequent to the higher priority given by the four thresholds CACs to the voice traffic incoming in the WLAN. Focusing the attention on the curves related to the four thresholds CACs (the two lowermost curves), the previously highlighted reduction of the impact of data users redirection as R c increases is still observed, since with both CACs for high values of R c almost all WWW users have already been redirected to UMTS. In figure 5, finally, we reported the average (over time and users) downlink throughput perceived at application level by FTP users. Here we can observe once again the impact of the different CAC strategies on the perceived performance. The rapid degradation emerging for increasing values of R c with the four thresholds CACs is mainly due to the fact that, as R c increases, more and more FTP users are refused by the WLAN and are redirected to UMTS, where they are assigned a 64/384 Kbit/s bearer. Considering that the 384 Kbit/s downlink data rate provided by UMTS is by far lower than the data rates achievable with the WLAN, it is straightforward to understand the rapid degradation observed in figure 5 for high values of R c. However, it has to be observed that with the assumed satisfaction threshold of 1Kbit/s for the average downlink throughput perceived at application level, the percentage of satisfied FTP users is almost 1% for all CAC strategies. Obviously increasing the satisfaction requirement would lead to a degradation of the percentage of satisfied FTP users; it has to be observed, however, that UMTS cannot provide at physical level downlink data rate higher than 384Kbit/s, hence a satisfaction requirement considerably higher than the assumed 1kbit/s of average application level throughput would not be reasonable. VI. CONCLUSIONS In this paper the issue of UMTS and WLANs interworking has been faced considering the Chinese TD-SCDMA UMTS technology and the forthcoming IEEE82.11e WLAN. Provided that UMTS CS voice flows can be converted into VoIP streams and redirected to the WLAN, here we investigated three different WLAN call admission control strategies aimed at increasing the number of voice calls that an integrated UMTS-WLAN network can manage with a satisfactory performance level. The simulation outcomes show that all CAC strategies are effective and provide different trade-offs between the improvement in terms of percentage of satisfied voice users and the performance degradation observed by data users. REFERENCES [1] [2] K.Pahlavan et al., Handover in Hybrid Mobile Data Networks, IEEE Personal Communications, April 2. [3] M.Lott et al., Interworking of WLAN and 3G systems, IEE Proceedings-Communications Volume 151, n.5, 24 Oct. 24. [4] O.Andrisano, A.Bazzi, M.Diolaiti, C.Gambetti, G.Pasolini, UMTS and WLAN Integration: Architectural Solution and Performance, IEEE PIMRC 25. [5] 3GPP, TS 25 series. [6] IEEE WG, IEEE 82.11e/D5. [7] D.Gu, J.Zhang, A new measurement-based admission control method for IEEE82.11 wireless local area networks. IEEE PIMRC 23. [8] A.Bazzi, M.Diolaiti, G.Pasolini, Measurement based Call Admission Control Strategies in Infrastructured IEEE82.11 WLANs, IEEE PIMRC 25. [9] IEEE Std 82.11a 1999 [1] Bur Goode, Voice Over Internet Protocol (VoIP), Proceedings of the IEEE, vol.9, No.9, sept. 22. [11] 3GPP TR V3.2., Technical Report Universal Mobile Telecommunications System (UMTS); Selection procedures for the choice of radio transmission technologies of the UMTS [12] N.K.Shankaranarayanan, Traffic Model for IEEE 82.2 MBWA System Simulations (Baseline Draft). IEEE 82.2 Working group on Mobile Broadand Wireless Access, July 23.