Call Admission Control Schemes for Next Generation Mobile and Wireless Networks: Review

Study of Guard-Channel-Based Call Admission Control Schemes for 4G Cellular Networks

Faisal Iradat Center for Computer Studies, Institute of Business Administration, Karachi [email protected]

Dr. Sayeed Ghani Center for Computer Studies, Institute of Business Administration, Karachi [email protected]

Abstract traditional and normalized approaches to the The success of next generation mobile and guard channel based CAC schemes. wireless networks will significantly depend on the performance of Call Admission I. Introduction Control (CAC) schemes which contributes With the rapid advancement in the next to the overall network performance. Due to generation of mobile and wireless networks, the limited availability of resources, future telecommunication networks are mobility of users, and quality of service targeted at providing a rich set of new (QoS) provisioning for applications whose wireless data and multimedia services along demand and nature are highly with the support of traditional voice service. heterogeneous, CAC schemes now play a The designers of these telecommunication more central role in QoS provisioning for networks face significant challenges due to next generation mobile and wireless the strict quality of service (QoS) networks such as 3G, B3G, WLAN, requirements. These challenges are further WiMAX and others. In this paper, we exasperated due to the typical mobile and briefly review existing CAC schemes and wireless issues such as limited availability of examine the analytical methods employed resources, user mobility, fading, shadowing, by these schemes for computing noise etc. new/handoff call blocking/dropping probabilities. The normalized approach for In order to cope with these challenges, the determining the new/handoff call next generation of wireless technologies will blocking/dropping probabilities for the New- have to incorporate radio resource Call Bounding Scheme is further management (RRM) mechanisms that investigated in terms of the expected number efficiently utilize the available resources. of new/handoff call packets in the system in RRM plays a critical role in the provisioning steady state. A performance comparison of of QoS in wireless systems. The these techniques is carried out via simulation performance of these schemes in turn affects and compared with analytical results. the overall network performance. Many On the basis of this study we also suggest researchers have proposed several radio that the performance evaluation of an resource management techniques but later “effective holding time” approach of the on discovered through analysis and guard-channel (GC) based CAC scheme be simulation that a sub area within RRM, reexamined/addressed in a 4G cellular called call admission control (CAC) is to be network. This is due to the higher efficiency addressed more critically than RRM [1,2]. and lower computational complexity of the said approach, as compared to other

1 Traditionally arriving new/handoff calls are network (WiMAX) etc. Congestion control either accepted or denied access to the techniques for these sort of wireless systems network by the CAC scheme based on call are expected to be far more complex than handoff dropping probability or cell traditional cellular network systems in the overload probability [3]. A CAC scheme sense that the proposed CAC scheme should controls the amount of traffic entering the now be capable of also selecting the best network by either managing the number of access network and at the same time call connections into the network or supporting QoS requirements and user reducing the overall network load thus mobility. enabling the network to provide the desired QoS to new/handoff call connections. For such heterogeneous networks, several solutions have recently been studied such as Various CAC schemes have been proposed the WLAN voice manager [9], ranking in the past for different types of networks. schemes [2], call admission and rate control Following are some of the prominent scheme (CARC) [10], Distributed Call schemes that have been proposed for 3G, Admission Control scheme for non-uniform beyond 3G (B3G) and Asynchronous traffic [11], thinning schemes [12], Transfer Mode (ATM) networks. Differentiated Treatment to Multimedia Traffic at Link Layer Framework [13], For cellular networks such as 3G and B3G Dynamic Programming Based Approach in systems, several CAC schemes have been Polynomial Time [14], and Stable Dynamic proposed such as the distributed call CAC (SDCA) [15] for call admission admission control [3], handoff density [4], control purposes. two-dimensional Markov approximation approach [5], Hoeffding Bounds [6], Future telecommunications networks such as Tangent at Peak [7], Tangent at Origin [7], the 4G cellular networks aim to provide Measure CAC[7], and Aggregated traffic differentiated services such as traditional envelops [7] in order to reduce the call voice, data, and multimedia with the desired handoff probability or cell overload QoS to service providers. From the call probability. management perspective, in such type of networks normally handoff calls are Whereas Non-statistical allocation, Peak assigned higher priority over new calls as bandwidth allocation, and statistical call dropping (which occurs when the allocation call admission control schemes network is unable to provide resources to a have been proposed for Asynchronous call for it to be handed off to another cell) is Transfer Mode (ATM) networks [8]. more intolerable to mobile users than call blocking. In the above networks, CAC schemes are only concerned with accepting or rejecting Handoff priority-based CAC schemes is the calls based on QoS requirements. On the area where most of the recent research on other hand in upcoming wireless CAC for wireless networks has been studied infrastructures such as the 4G ubiquitous in depth in the recent past as compared with wireless systems, there will be other schemes discussed previously heterogeneous wireless access networks [5,12,16,17]. In this paper we explore these such as wireless local area network CAC schemes briefly and examine the (WLAN), wireless metropolitan area analytical methods employed by these

2 schemes for computing new / handoff call In the new call bounding scheme, a blocking / dropping probabilities. The threshold is enforced on the number of normalized approach for determining the new calls accepted into the cell. If the new/handoff call blocking/dropping number of new calls in a cell is less than probabilities for the New-Call Bounding a threshold K than a new call will be Scheme is further investigated in terms of admitted. the expected number of new/handoff call Cutoff Priority Scheme [18,19,20]: packets in the system in steady state [En(L) In the cutoff priority scheme, new calls & Eh(L)]. are admitted only if the number of busy Finally, suggest that the performance channels is less than a threshold m. evaluation of the “effective holding time” Fractional GC Scheme [17]: approach of the guard-channel based CAC This is also called as the Thinning scheme be reexamined / addressed in a 4G Scheme, in which new calls will be cellular network. selectively blocked when the cell traffic increases. More specifically, new calls

The rest of this paper is organized as are admitted with probability  i when follows. In Section II we briefly present the the number of busy channels is i. Based different Guard-Channel (GC) schemes. on this probability a new call is either Next, in section III we study the admitted or blocked. performance of the New-Call Bounding Scheme in detail. Results of simulation and By assigning different values to  this analysis are presented in this section. In i section IV we review the “effective holding scheme can be reduced to the cutoff time” approach of the guard-channel based priority scheme [17] that has been CAC scheme. Finally section V summarizes discussed earlier. As this scheme is quite our findings and discusses future research similar to cutoff priority scheme, it will areas. not be presented in detail, in this paper.  nc1 h    q(c 1)  cq(c)  n h  II. Brief Overview of Guard-Channel c=1,….,C (1) Based CAC Schemes Here q(c) denotes the equilibrium In a wireless mobile / cellular network channel occupancy probability when environment, call dropping is more exactly c channels are occupied in a cell intolerable than call blocking from the and   and   are the arrival and mobile users aspect. As a result most of the h, h n, n CAC schemes proposed, give higher priority departure rates for handoff calls and new to handoff calls than to news calls. Handoff calls respectively. c is the number of priority-based CAC schemes proposed in occupied channels in a cell. [5,12,16,17], are classified in two broad Rigid Division Based Scheme [21]: categories, as described below. In this scheme all channels are divided into two groups: one for common calls 1. Guard Channel (GC) schemes: and the other for handoff calls. This In guard channel based schemes some scheme does not efficiently utilize the channels are reserved for handoff calls. available resources and mentioned for Following are the four different types: reference only. New Call Bounding Scheme [5]:

3 2. Queuing Priority (QP) schemes: different distributions and average holding In these schemes if channels are free times [30,31]. than calls will be accepted. When all channels are busy either new calls are To avoid such complexities associated with queued and the handoff calls are multidimensional Markov chain models, blocked, or new calls are blocked and approximations have been used that have handoff calls are queued, or all arriving high accuracy and low computational calls are queued with certain complexity with the condition that the rearrangements in the queue approximation is based on a one- [22,23,24,25,26]. dimensional Markov chain model with These schemes are restrictive to the GC exponentially distributed holding times for schemes presented earlier [17]. new / handoff calls however with different Secondly, they cannot be compared with average values [5,17]. other GC schemes as the protocol is quite different [17]. III. Performance of New Call Bounding Scheme using In recent research [5,17] the above Simulation and Analysis mentioned GC based CAC schemes were In this section we will briefly discuss the examined using the Markov chain models. performance of the new call bounding scheme using the traditional and the In a traditional circuit switched network, the normalized analytical methods proposed in one-dimensional Markov chain model was [5,17]. The assumptions made are that the used under the assumption that all arrivals arrival process of the new/handoff calls are are modeled as Poisson and the channel modeled as Poisson, and the average holding time and channel residence time are channel holding times are different, and are modeled as exponential with predetermined exponentially distributed. Resulats of a channel bandwidth [27,28]. However with simulation written in C++ are given and the rapid evolution of cellular networks, the compared to analytical results as well. Markov chain model used for obtaining blocking/dropping probabilities for In the traditional approach, the new/handoff calls becomes more complex as blocking/dropping probabilities are derived the new and handoff calls may have for new/handoff calls with the assumption different average holding times and worse, that the new/handoff calls have the same different distributions. This problem may be average values and channel holding time solved with use of multidimensional Markov distributions. This system is approximated chain model but these suffers from the curse as one-dimensional Markov chain model of dimensionality thus making them much with a fixed channel holding time for the more complex as compared with the one- total cell traffic [5]. dimensional Markov chain models [29]. Many researchers attacked this problem with In the normalized approach [5], the results the assumption that the channel holding are improved by normalizing the average times for new/handoff calls are identically service times for new/handoff calls to unity distributed with the same average values. rather than equal average channel holding However this assumption is inappropriate as time. new and handoff calls may have quite The performance of the new call bounding

4 scheme is discussed as follows: we get the following transition diagram (Fig. 1) which is modeled by a two-dimensional As presented earlier in section II, if we use Markov chain model (Please refer to the call admission control scheme given in appendix A for nomenclature). section II for the new call bounding scheme, Handoff calls (n2)

New Calls (n1)

Fig. 1. State transition diagram for the new call bounding scheme (Source [5]) The steady state probability p(n1, n2) can be blocking/dropping probabilities of the new determined as given below (2) by solving call bounding scheme for the normalized the global balance equations from the state approach) [5,17]. Whereas, the traditional transition diagram as shown in Fig. 1. n1 and approach uses fixed average channel holding n2 represents new calls and handoff calls. time  av for total cell traffic is given by:  n1  n2 n h 1 n 1 h 1 n  h p(n1, n2 )  p(0,0) (2)    n ! n ! (3) 1 2 av n  h n n  h h n  h

n h 1 1 Where  n  and  h  represents the By replacing and in the formulas for  n  h  n  h traffic intensities for new/handoff calls new/handoff call blocking/dropping (please refer to appendix A for probabilities of the normalized approach nomenclature). (please refer to appendix B for Due to the inherent complexity associated blocking/dropping probabilities of the new with the two-dimensional Markov chain call bounding scheme for the normalized model, in the normalized approach the approach) we get the traffic intensities for new/handoff calls blocking/dropping new/handoff calls in the resulting one- a a probabilities (  nb and  nh ) are dimensional Markov chain model for the determined under the assumption that the traditional approach (please refer to service rate of new/handoff calls equals to 1 appendix B for blocking/dropping (please refer to appendix B for

5 probabilities of the new call bounding scheme for the traditional approach).

In order to compare the normalized versus traditional approaches a simulation was also conducted, written in C++. The simulation made the same assumptions as noted above for traffic and call admission control. Results of the simulation and analytical Fig. 4. Average number of New/Handoff calls in expressions are discussed below. a cell in steady state * When the new call bounding scheme is * N and T denotes the Normalized and investigated under fixed handoff holding Traditional Approaches, n and h denote times and varying new call holding times new/handoff, S and A denote Simulation / (K=15, C=30, 1/λn=30, 1/µn=changes from Analytical 200 to 600, 1/λh=30, 1/µh=450), it is apparent from Fig. 2 that the traditional On the other hand when the new call approach overestimates the normalized bounding scheme is investigated under approach when handoff call traffic load is varying handoff holding times and fixed higher than new call traffic load whereas the new call holding times (K=20, C=30, traditional approach underestimates handoff 1/λn=30, 1/µn=300, 1/λh=30, 1/µh= changes call blocking probability of the normalized from 100 to 1200), it is observed from Fig. 5 approach as shown in Fig. 3. Figure 4 shows and Fig. 6 that the new call blocking the graph of average number of new/handoff probability and handoff call blocking calls in a cell in steady state for different probability are similar in value. This makes sense because due to the heavy traffic load new call traffic loads ( n ). incurred from handoff calls the new calls will not be able to reach a bound. However, the traditional approach is close but not the same as the normalized approach. . Figure 7 shows the graph of average number of new/handoff calls in a cell in steady state for

different handoff traffic loads ( h ).

Fig. 2. New call blocking probability in the new call bounding scheme*

Fig. 5. New call blocking probability in the new * Fig. 3. Handoff call blocking probability in the call bounding scheme new call bounding scheme*

6 chain model. All that is required is a better approximation method.

In the “effective holding time approach”, equation (1) is simplified by replacing the average holding times for new/handoff calls with the average channel holding time for the total traffic also referred to as average Fig.6. Handoff call blocking probability in the new call bounding scheme * 1 * effective channel holding time . In this N and T denotes the Normalized and eff Traditional Approaches, n and h denotes 1 1 new/handoff, S and A denotes Simulation / case we would replace and with   Analytical n h 1 1 rather than as in the traditional eff  avg 1 approach. We cannot use the average  avg since the average channel holding times for total cell traffic have different blocking probabilities for new / handoff calls. The Fig. 7. Average number of New/Handoff calls in 1 a cell in steady state * average is obtained from the idea eff * N and T denotes the Normalized and proposed in [29] which is also referred to as Traditional Approaches, n and h denotes the knapsack problem approach (please refer new/handoff, S and A denotes Simulation / to appendix B for locking/dropping Analytical probabilities of the cutoff priority scheme for the “effective holding time” approach) IV. Review of the “effective holding [17]. time” approach of the guard- channel based CAC scheme In order to evaluate the performance, the From the results presented in the previous numerical results obtained from this scheme section we can determine that the i.e. the “effective holding time approach” normalized approach performs better than was compared with those obtained from the the traditional approach [17]. However in existing traditional approach, normalized order to further improve the results while approach, and the direct approach. The exact keeping the complexity low, a new approach method that is used for determining the called the “effective holding time approach” actual values of a multidimensional Markov was proposed which produced better results chain model using the “LU decomposition” when compared with the existing is referred to as the direct approach [17]. approaches discussed earlier [17]. Through the comparison (Fig. 8 and Fig. 9) To keep the computational complexity low it it was observed that the traditional approach is useful to reduce the two-dimensional and normalized approaches deviate Markov chain model to a one-dimensional significantly from the direct and the “effective holding time approach”. The

7 “effective holding time approach” results are analyze exactly due to, the asymmetric very close to the actual results obtained from nature of the states in a Markov chain the direct method and seem as the best fit as transition diagram. Hence it was noted that it a future GC based CAC scheme for wireless is useful to attempt reducing/approximating network. the two-dimensional Markov chain model with a one-dimension Markov chain model. All GC scheme approximate a two- dimensional Markov chain model with a one-dimensional Markov chain model using different approaches referred to as the traditional approach, normalized approach, and the effective holding time approach.

It has been shown that the “effective holding Fig. 8. New call blocking probability in the time approach” outperforms the traditional effective holding time approach [17]. and normalized approaches in terms of accuracy and low computational complexity. Hence in order to evaluate CAC schemes for a 4G network environment, which will require handling of multiclass services we suggest that the “effective holding time” approach be used and further studied.

References [1] Hossam Ahmed, “Call Admission Fig. 9. Handoff call blocking probability in Control in wireless Networks: A the effective holding time approach [17]. Comprehensive Survey”, IEEE Communications Survey & Tutorials, In summation, the traditional approach Volume 7, No. 1, First Quarter 2005. overestimates whereas the normalized [2] Robert Achieng and H. Anthony Chan, approach underestimates the direct “A QoS Enabling Queuing Scheme for approach. 4G Wireless Access Networks”, ACM – V. Summary and Future Work IWCMC, July 2006. In this paper we critically reviewed existing [3] M. Naghshineh, and M. Schawrtz, guard-channel based CAC schemes “Distributed Call Admission Control in proposed in [5,17] in order to determine the Mobile/Wireless Networks”, approach that gives more accurate results International Symposium on Personal when compared with other approaches. It Indoor Mobile Radio Communications, was learnt that the two-dimensional Markov IEEE, September 1995. chain model solves the problem when the [4] Q. Gao and A. Acampora, “Performance average channel holding times for the new Comparison of Admission Control and handoff calls are significantly different. Strategies for Future Wireless However as the dimensions of the Markov Networks”, Proc. IEEE wirelss chain model increase due to distinct traffic Communication and Network type’s e.g. video, voice, and packetized data, Conference, WCNC2002, vol. 1, Mar. the computational complexity increases and 2002, pp. 317-21. at times becomes virtually impossible to

8 [5] Y. Fang and Yi Zhang, “Call Admission Schedule in Cellular Networks”, IEEE Control Schemes and Performance INFOCOM 2003, 2003. Analysis in Wireless Mobile [15] S. Wu, K. Wong, and B. Li, “A Networks”, IEEE Trans. Veh. Technol., Dynamic Call Admission Policy with Vol 51, no. 2, pp. 371-382, Mar. 2002. Precision QoS Guarantee using [6] G. Razzano and A. Curcio, Stochastic Control for Mobile Wireless “Performance Comparison of Three Networks”, IEEE Transactions on Call Admission Control Algorithms in a Networking, Vol 10, No. 2, April 2002. Wireless Ad Hoc Network”, Proc. Int’l [16] I. Katzela and M, Naghshineh, Conf. Commun. Tech. (ICCT 2003), “Channel assignment schemes for vol. 2., Apr. 2003, pp. 1332-36. cellularmobile telecommunication [7] L.Breslau and S. Jamin, “Comments on systems: A comprehensive survey”, the Performance of Measurement-Based IEEE Personal Commun., vol. 3, pp. 10- Admission Control Algorithms”, Proc. 31, June 1996. IEEE INFOCOM 2000, vol. 3, Mar. [17] A. Yavuz and C. M. Leung, 2000, pp. 1233-41. “Computationally efficient method to [8] H. G. Perros, and K. M. Elsayed, “Call evaluate the performance of Guard- Admission Control Schemes: A Channel-Based call admission control review”, IEEE Communications in cellular networks”, IEEE Trans. on Magazine, Nov. 1996, vol. 34., no. 11, Veh. Techno.., vol. 55, no. 4, July 2006. pp. 82-91. [18] R. A. Guerin, “Queuing–blocking [9] Yi Qian, Qingyang Hu, Hsiao-Hwa system with two arrival streams and Chen, “A Call Admission Control guard channels,” IEEE Trans. Framework for Voice over WLANS”, Commun., vol. 36, no. 2, pp. 153–163, IEEE wireless communications, Feb. 1988. February 2006. [19] D. Levine, I. Akyildiz, and M. [10] H. Zhai, X. Chen, and Y. Fang, “A Call Naghshineh, “A resource estimation and Admission and Rate Control Scheme call admission algorithm for wireless for Multimedia Support over IEEE multimedia networks using the shadow 802.11 Wireless LANs”, QSHINE 04, cluster concept,” IEEE/ACM Trans. 2004. Netw., vol. 5, no. 1, pp. 1–12, Feb. [11] J. Kim, E. Jeong, and J. Cho, “Call 1997. Admission Control for Non-uniform [20] C. Oliveira, J. B. Kim, and T. Suda, Traffic in Wireless Networks”, “An adaptive bandwidth reservation Electronic letters, January 2000. scheme for high-speed multimedia [12] Y. Fang, “Thinning Schemes for Call wireless networks,” IEEE J. Sel. Areas Admission Control in Wireless Commun., vol. 16, no. 6, pp. 858–874, Networks”, IEEE transactions on Aug. 1998. Computers, Vol. 52, No. 5, May 2003. [21] M. D. Kulavaratharasah and A. H. [13] R. Jayaram, S. K. Das, S. K. Sen, “A Aghvami, “Teletraffic performance Call Admission and Control Scheme for evaluation of microcell personal Quality of Service (QoS) Provisioning communication networks (PCN’s) with in Next Generation Wireless prioritized handoff procedures,” IEEE Networks”, Baltzer Journal, 2000. Trans. Veh. Technol., vol. 48, no. 1, pp. [14] S. Ganguly, B. Nath and N. Goyal, 137–152, Jan. 1999. “Optimal Bandwidth reservation

9 [22] R. A. Guerin, “Queuing–blocking networks,” IEEE Trans. Commun., vol. system with two arrival streams and 47, no. 7, pp. 1062–1072, Jul. 1999. guard channels,” IEEE Trans. [31] Y. Fang, I. Chlamtac, and Y. B. Lin, Commun., vol. 36, no. 2, pp. 153–163, “Channel occupancy times and handoff Feb. 1988. rate for mobile computing and PCS [23] E. D. Re, R. Fantacci, and G. networks,” IEEE Trans. Comput., vol. Giambene, “handover queuing 47, no. 6, pp. 679–692, Jun. 1998. strategies with dynamic and fixed channel allocation techniques in low Appendix – A earth orbit mobile satellites systems”, C Number of channels in a cell. IEEE Transc. Commun., vol. 47, no. 1, c Number of occupied channels in a cell. pp. 89-102, 1999. K Threshold for new call bounding [24] C. H. Yoon and C. K. Un, scheme. “Performance of personal portable radio m Threshold for the cutoff priority scheme. telephone systems with and without n Arrival rate for new calls.

guard channels”, IEEE J. Select. Areas h Arrival rate for handoff calls. Commun., vol. 11, no. 6, pp. 911-917, 1  Average channel holding time for new 1993. n calls. [25] C. Chang, C. J. Chang, and K. R. Lo, 1  Average channel holding time for “Analysis of a hierarchical cellular h system with reneging and dropping for handoff calls. waiting new calls and handoff calls,” 1 eff Average effective channel holding time.

IEEE Trans. Veh. Technol., vol. 48, no.  n Traffic intensities for new calls. 4, pp. 1080–1091, Jul. 1999.  Traffic intensities for handoff calls. [26] V. K. N. Lau and S. V. Maric, h “Mobility of queued call requests of a Pnb Blocking probability for new calls.

new call queuing technique for cellular Phb Dropping probability for handoff calls. systems,” IEEE Trans. Veh. Technol., P a Blocking probability for new calls from vol. 47, no. 2, pp. 480–488, May 1998. nb the normalized approximation. [27] R. Guerin, “Equivalent capacity and its a application to bandwidth allocation in Phb Dropping probability for handoff calls high-speed networks,” IEEE J. Sel. from the normalized approximation. t Areas Commun., vol. 9, no. 7, pp. 968– Pnb Blocking probability for new calls from 981, Sep. 1991. the traditional approximation. [28] Q. Ren and G. Ramamurthy, “A real- t Phb Dropping probability for handoff calls time dynamic connection admission from the traditional approximation. controller based on traffic modeling, P eff Blocking probability for new calls from measurement, and fuzzy logic control,” nb IEEE J. Sel. Areas Commun., vol. 18, the effective holding time approximation. eff no. 2, pp. 184–196, Feb. 2000. Phb Dropping probability for handoff calls [29] A. Gersht and K. J. Lee, “A bandwidth from the effective holding time approximation.

management strategy in ATM  i Admission probability for new calls in networks,” GTE Laboratories, fractional guard channel (GC) scheme Waltham, MA, 1990. Tech. Rep. q(c) Equilibrium occupancy probability. [30] Y. Fang and I. Chlamtac, “Teletraffic  q(c) Approximate equilibrium occupancy analysis and mobility modeling for PCS probability.

10 Appendix – B

1. New Call Bounding Scheme Normalized Approach CK  K  K K 1  n1  Cn1  n h   n h a n2 0 K! n2! n1 0 n1! (C  n1 )! Pnb  n n K  1 Cn1  2  n  h n1 0 n1! n2 0 n2!

K  n1  Cn1  n h a n1 0 n1! (C  n1)! Pnb  K  n1 Cn1  n2  n  h n1 0 n1! n2 0 n2! Traditional Approach

 n n  n   ( n   h ) av n  h

 h n  n   ( n   h ) av n  h

Replace the given traffic intensities in the equations for normalized approach (new call bounding approach) given above (1.1) to get the probabilities for the traditional approach. 2. Effective Holding Time Approach C 1   n. j .q( j) Peff  1 j0 nb C   n.q( j) j0 C 1   h. j .q( j) Peff  1 j0 hb C   h.q( j) j0