This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
Performance Enhancement for Unlicensed Users in Coordinated Cognitive Radio Networks Via Channel Reservation Xu Mao† , Hong Ji† , Victor C.M. Leung‡ , Ming Li† Laboratory of Wireless Communications and Networking, Beijing University of Posts and Telecommunications, Beijing, 100876, China ‡ Dep. Electrical and Computer Engineering, The University of British Columbia, Vancouver, BC, Canada V6T 1Z4 Email:
[email protected],
[email protected],
[email protected] and
[email protected] †
Abstract— In common models of cognitive radio networks (CRNs), priority in spectrum access of primary users (PUs) must be guaranteed, but the quality of service (QoS) for secondary users (SUs) is mostly ignored. Subject to meeting PUs’ blocking probability objective, this paper proposes a channel reservation scheme to enhance QoS for SUs. The proposed scheme employs a centralized control manager (CCM) to coordinate the dynamic spectrum access among PUs and SUs by dividing the spectrum into two parts: the normal access channels (NACs) that may be taken back anytime by PUs, and the reserved secondary channels (RSCs) that are locked by occupied SUs until their call sessions are complete. A Markov chain model is developed to analyze the proposed spectrum access scheme and compare its performance to that of a CRN without reserved channels. Furthermore, the number of reserved channels is optimized based on SUs’ utility functions. Numerical results show that by optimally selecting the number of reserved channels to adapt to various traffic loads, the proposed scheme can significantly improve SUs’ QoS while meeting PUs’ QoS objectives.
I. I NTRODUCTION The proliferation of wireless systems has brought about a huge demand for more bandwidth. While the traditional approach of fixed spectrum allocation to authorized networks has lead to that there exists no vacant spectrum to meet these newly increasing requirements of high-rate data services. However, in recent studies especially by the Federal Communications Commission (FCC), some licensed spectrum is under-utilized, with reported utilization as low as 15% [1]. Cognitive radio (CR), which allows secondary users (SUs) to temporarily share the same frequency bands with primary users (PUs) under certain spectrum usage constraints, provides a promising approach to increase the availability of spectrum to new services and enhance the utilization of licensed frequency bands [2], [3]. Resource management in cognitive radio networks (CRNs) is more difficult than in conventional wireless systems. On one This work was jointly supported by State Key Program of National Natural Science of China (Grant No. 60832009), Natural Science Foundation of Beijing, China (Grant No. 4102044), Innovative Project for Young Researchers in Central Higher Education Institutions, China (Grant No. 2009RC0119) and New Generation of Broadband Wireless Mobile Communication Networks of Major Projects of National Science and Technology, China (Grant No. 2009ZX03003-003-01).
hand, PUs are typically protected by designing the secondary system subject to interference power constraints [4], [5]. Furthermore, SUs need to be able to effectively share the spectrum left over by the primary system [6], [7]. On the other hand, SUs’ behavior should be transparent to PUs. That is, SUs may opportunistically access segments of the spectrum left vacant by the licensed PUs, and if a PU decides to access the primary channel, all SUs using the corresponding secondary channels must relinquish their transmissions immediately [8]. In general, these unfinished secondary transmissions may be either handed over to other available sub-channels to continue their transmissions or simply discarded. This is effective in protecting PUs from interference by SUs. However, in the studies mentioned above, quality of service (QoS) for the SUs, whose communications are subject to interruptions and disruptions, is rarely considered. It is therefore of interest to develop new spectrum management concepts [9], [10] that enable more efficient usage of the spectrum while enhancing the QoS experienced by SUs, which is the main objective of this paper. Recently there is growing interest in dynamic spectrum management that does not have total transparency between SUs and PUs [10], [11], e.g., a cellular mobile system of SUs sharing the same set of channels with a cellular mobile system of PUs. In this paper, we develop a new coordinated dynamic spectrum access scheme for the above scenario, where the PUs and SUs share a pool of channels. PUs have pre-emptive priority access to most of the channels to ensure that their channel access meets some specified blocking probability objective. We present a centralized control manager (CCM) model for coordinated spectrum access. Based on this, we propose a channel reservation scheme to optimize SUs’ utility function, while satisfying the blocking probability objective of PUs. In our proposed scheme, the spectrum band is divided into normal access channels (NACs), in which PUs can pre-empt SUs, and reserved secondary channels (RSCs) reserved for use by SUs. Finally, the users’ performance with or without channel reservation is systematically analyzed. Simulation results show that the proposed scheme can greatly enhance the QoS of SUs. Thus it can be applied to improve the QoS of SUs under the
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
constraint that the QoS objectives of PUs are met. The rest of this paper is organized as follows. A general system model of pre-emptive sharing between PUs and SUs is presented and analysed in Section II. Section III presents the proposed channel reservation scheme. Simulation results are analysed in Section IV. Section V concludes the paper. II. P RE - EMPTIVE S HARED C HANNEL ACCESS A. System Overview Fig. 1 shows the coordinated dynamic spectrum access model under consideration. We assume that K channels are shared by two radio systems, designated as the primary and secondary systems, respectively. The primary system serves PUs and the secondary system serves SUs. For simplicity, we assume that the radio systems have perfect information on channel usage, and serve users in the same coverage area. Concurrent transmissions over the same channel are not permitted. A PU arrival is allocated a free channel if available, and if all channels are occupied, it is allocated a channel occupied by a SU by pre-empting the SU, which suffers a forced termination. The PU arrival is blocked if all channels are occupied by other PUs. A SU arrival is allocated a channel only if a free channel is available; otherwise it is blocked.
channel reservation in the CR systems can be cataloged into two parts: When at least one free channel is available, the state belongs to Γ1 = {(i, j) |i + j < K} in Fig. 2A. In this case, an arriving PU or SU can be allocated a channel, and no pre-emption occurs. We have [i μs + λs +(i + 1) μp + λp ] π (i, j) = λs π (i − 1, j) + (i + 1) μs π (i + 1, j) (1) + λp π (i, j − 1) + (i + 1) μp π (i, j + 1) , where i ≥ 0, j ≥ 0 and i + j < K. If all channels are occupied, then the state belongs to Γ2 = {(i, j) |i + j = K} in Fig. 2B. The arrival of a PU may lead to the pre-emption of a SU, and the respective states probabilities satisfy [i μs +j μp + λp ] π (i, j) = λs π (i − 1, j) + λp π (i, j − 1) , where i ≥ 0, j ≥ 0 and i + j = K.
iPs
Os
( j 1) P p
Op
*1
Os , P s
Op , P p
(2)
Os (i 1) P s jP p
Op
^(i, j ) i j K `
Fig. 2.
iPs
Op
*2
Os
Op
jP p
^(i, j ) i j
K`
Markov chain without channel reservation.
The steady state probability distribution {π (i, j) | (i, j) ∈ Γ = Γ1 ∪ Γ2 } can be obtained by solving the equations above with the normalization constraint (i,j)∈Γ π (i, j) = 1. C. Performance Analysis
Fig. 1. Model of pre-emptive sharing between PUs and SUs in coordinated CR systems.
B. System Model We assume that the arrival processes of PUs and SUs are Poisson with arrival rates λp and λs , respectively. The service times for PUs and SUs follow exponential distributions with means 1/μp and 1/μs , respectively. Let (i, j) represents the system state, where i and j are the number of SUs and PUs receiving service in the respective systems. The state space Γ = {(i, j)|i + j ≤ K} . Let π (i, j) be the steady state probability distribution for state (i, j) ∈ Γ . The states without
For resource allocation at the session level, the blocking probability [12] is an important QoS parameter. Denote the blocking probability of PUs and SUs by Pbp and Pbs , respectively. A PU arrival is blocked when all channels are occupied by other PUs, where i = 0 and j = K; therefore, we have p Pb = π (0, K) . However, a SU arrival is blocked when all the channels are occupied; therefore, s π (i, j). (3) Pb = (i,j)∈Γ2
When all channels are occupied, the arrival of a PU may pre-empt a SU that is being served, causing the SU to suffer a service interruption. The probability that a SU’s service is interrupted by the next PU arrival is π (i, j) . (4) PInt = (i,j)∈{ i+j=K,i>0,j≥0} i
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
In this paper, we assume that a service interruption lead to a forced service termination. The probability that a SU suffers a forced termination of its service due to a PU arrival is given by ∞ r r−1 (5) tp,u (1 − PInt ) Pr t s > PInt , Pf t =
PU is blocked. The pre-empted SU would perform a spectrum handoff and occupy an available RSC xi ∈ {CR | φr = 0} ; however, if all RSCs are occupied, the SU would be forced to terminate service. 2) SU arrival An available NAC xi ∈ {CN | φn = 0} is allocated to the arriving SU. If no NAC is available, an available RSC xi ∈ {CR | φr = 0} is allocated. If all NACs r=1 u=1 and RSCs are occupied, the SU is blocked. where ts denotes the service time of a SU, tp,1 stands for 3) PU or SU departure from NAC As soon as a PU or SU the duration between a SU’s arrival and the arrival of the completes service in a NAC, if there are SUs being served in next (first) PU, and tp,u , u = 2, 3, ... represent the interthe RSCs, one of them would be switched to the vacated NAC arrival times between the second and first PUs, third and r by spectrum handoff; that is, one more RSC becomes free. On second PUs, etc. Let ξr = tp,u ; then, ξr is an Erlang the other hand, if all RSCs are vacant, the NAC becomes free u=1 distributed random variable with probability density function for allocation to the next arriving PU or SU. r−1 λp (λp t) 4) SU departure from RSC The RSC becomes free for fξr = (r−1)! e− λp t . Therefore, (6) becomes allocation to a SU that has been pre-empted from a RSC or ∞ ∞ r−1 has newly arrived and found all RSCs busy. λp (λp t) e− λp t dt (1 − PInt )r−1 PInt . (6) Pf t = The system dynamics is quite different compared with the (r − 1)! r=1 0 conventional system described in Section II. Fig. 3 shows Adopting the result in [13], we have Pf t = μs λ+p λPpInt . the state transitions, where the states transitions are described PInt Further, the service completion probability of a SU can be under the following three scenarios: obtained by Pcom = 1 − [Pb + (1 − Pb ) Pf t ] .
(7)
iPs ( j 1) P p
Another important QoS parameter for SUs is the service completion rate ρ, which is given by iπ (i, j)μs Pcom . (8) ρ= (i,j)∈Γ
Op O
p i, j-1, 0
'1
Os ( j 1) P p
Os Os i+1, j, 0
i, j, 0
i, j+1, 0
A)
III. P RE - EMPTIVE S HARED C HANNEL ACCESS WITH R ESERVATION
i-1, j, 0
i, j, l+1
(i l 1) P s ( j l )Ps
i, j, l
i-1, j+1, l+1
Op O p
(i 1) P s jP p
i, j, l-1
O j P ps
i+1, j-1, l-1
^(i, j, l ) i j J , l
0`
'2
B)
^(i, j, l ) i j
J , 0 l L`
0, J, l-1
Os J Pp
iPs (l 1) P s
In this section, we give a description of our proposed channel reservation scheme as well as derive the optimum number of reserved channels based on SUs’ utility function.
0, J, l+1
C)
'3
1, J-1, l-1
0, J, l
Os
^(i, j, l ) i
Op 0, j
J , 0 l L`
A. Channel Reservation Scheme Description To reduce forced terminations of SUs due to arrivals of PUs, the set of available channels C in the system is divided into two parts: CR = {1, 2, ..., L} and CN = {1, 2, ..., J }, where J and L are the numbers of RSCs and NACs, respectively, and → − L + J = K. We use an indicator vector φ = {φ1 , ..., φK } to represent the channel occupation status, where φk = 0 means that channel k is free, while φk = 1 means that the channel is occupied. Let (i, j, l) represents the system state with i, j and l representing, respectively, the numbers of SUs, PUs being served by NACs and the number of SUs being served by RSCs. The state space is Δ = {(i, j, l)|i + j ≤ J, l ≤ L} . Let π (i, j, l) be the steady state probability distribution for the state (i, j, l) ∈ Δ. The proposed scheme can be described under different scenarios: 1) PU arrival When a PU arrives, if some free NACs are available, one of them xj ∈ {CN | φn = 0} would be allocated to the PU. If all NACs are busy and some of them are occupied by SUs, one NAC is released by pre-empting a SU. In the worst case, if all NACs are occupied by other PUs, the new
Fig. 3.
Markov chain with channel reservation.
When one or more free RSCs are available (Fig. 3A), Δ1 = {(i, j, l) |i + j < J, l = 0}, a NAC can be allocated to the next arriving PU or SU, and no blocking or forced termination occurs. We have [i μs + λs +(i + 1) μp + λp ] π (i, j, 0) = λs π (i − 1, j, 0) + (i + 1) μs π (i + 1, j, 0) + λp π (i, j − 1) + (i + 1) μp π (i, j + 1) ,
(9)
where i ≥ 0, j ≥ 0 and i + j < J. For the states satisfying = Δ2 {(i, j, l) |i + j = J, 0 < l < L} in Fig. 3B, here, all NACs are occupied either by PUs or SUs while the RSCs are partly occupied by SUs. We have [λs + (j + l) μs +j μp + λp ] π (i, j, l) = (i + l + 1) μs π (i, j, l + 1) + λs π (i, j, l − 1) + λp π (i + 1, j − 1, l − 1) + (j + 1) μp π (i − 1, j + 1, l + 1) ,
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
(10)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
Especially, in the worst case where all NACs are occupied by PUs and all RSCs are occupied, Δ4 = {(i, j, l) |i = 0, j = J, l = L} , all new arrivals, whether SU or PU, are blocked (L μs +J μp ) π (0, J, L) = (12) λs π (0, J, L − 1) + λp π (1, J − 1, L − 1) . Applying the normalization constraint 4 Δi , to jointly (i,j)∈Δ π (i, j, l) = 1 , where Δ = i=1
solve (10)-(13) yields the steady state probability distribution {π (i, j, l) | (i, j, l) ∈ Δ}. The set of linear equations can be solved using the XOR iterative method [14]. B. Channel Reservation Optimization Under our proposed channel reservation scheme, an arriving PU would be blocked when all NACs are occupied by other PUs; thus, p π (i, j, l). (13) Pb (L) = (i,j,l)∈Δ3 ∪ Δ4
As for SUs, an arriving SU would be blocked when all the channels, including NACs and RSCs, are occupied; therefore, s π (i, j, l). (14) Pb (L) = (i,j,l)∈Δ4
The probability that the service of a SU is interrupted by the next PU arrival is π (i, j, l) . (15) PInt (L) = (i,j,l)∈Δ4 i Following the same relationship between interruption probability and forced termination probability in Section II, we also have λp PInt (L) . (16) Pf t (L) = μ s + λp PInt (L) Similarly, the service completion rate for SUs defined in (9) becomes ρ (L) = iπ (i, j)μs Pcom (L), (17) (i,j)∈Δ
where Pcom (L) = 1 − [Pb (L) + (1 − Pb (L)) Pf t (L)]. We use the sigmoid utility function [15], which is also used in [16], to represent quantitatively the satisfaction on SUs’ service completion rate ρ. This utility function U is defined as follow U (ρ (L)) =
1 , 1 + exp {−g [ρ (L) − h]}
(18)
where g and h are the parameters of the sigmoid function. We seek to optimize the number of reserved channels, L, to maximize the utility of SUs while keeping the blocking p probability of PUs within a given constraint PbM AX . The optimization problem can be formulated as: L∗ = arg max U
(19)
p s.t., Pbp (L) < PbM AX , L ≤ K,
where L∗ denotes the optimal number of reserved channels. As the potential optimal number of reserved channels is limited to integers chosen from 1 to K, this optimization problem can be solved by enumerative algorithms [17]. IV. S IMULATION R ESULTS AND D ISCUSSIONS Simulation results in Matlab are presented in this section to evaluate the performance of the proposed channel reservation scheme in the previous sections. We compare dynamic spectrum access by pre-emptive sharing, with and without the proposed channel reservation method. We consider a practical network, where a primary access point (AP) and a secondary AP are accessed by multiple PUs and SUs, respectively. Both PUs and SUs share K = 30 channels in a pool of the CCM. Unless otherwise specified, the following parameters are applicable: μs = 1.2, μp = 0.8, p g = 1, and h = 0. In addition, λp , λs and PbM AX are chosen from the intervals (0.1, 0.5), (0.2, 1) and (0.01, 0.05), respectively. Fig. 4 depicts the optimal number of RSCs vs PU traffic with various given SU traffic loads and PU blocking constraints. As the arrival rate of PU increases, the optimum number of RSCs decreases, in order not to go beyond the blocking probability limit of PU. Given the same blocking probability limit of p PU, e.g., PbM AX = 0.05, we can see that, when a smaller SU’s arrival rate is adopted, the optimum number of RSCs decreases. Note that given the same arrival rate of SU, the optimum number of RSCs is also increased as the blocking probability limit of PU is increased from 0.01 to 0.05. 7 λs=0.8, PbMAX=0.05 6
λ =0.6, P
=0.05
λ =0.8, P
=0.01
λ =0.6, P
=0.01
s
Optimum number of RSCs
where i > 0, j ≥ 0 and i + j = J. Note that a departure from a NAC may trigger a spectrum handoff by a SU from a RSC. For the states satisfying Δ3 = {(i, j, l) |i = 0, j = J, l < L} as depicted in Fig. 3C, all NACs are used only by PUs, and SUs are served in RSCs; thus, (λs +l μs +J μp ) π (0, J, l) = (l + 1) μs π (0, J, l + 1) + λs π (0, J, l − 1) (11) + λp π (1, J − 1, l − 1) .
bMAX
s
5
bMAX
s
bMAX
4 3 2 1 0 0.1
0.15
0.2
0.25 0.3 0.35 Arrival rate of PU (λ )
0.4
0.45
0.5
p
Fig. 4. Optimum number of RSCs vs PU traffic with various given SU traffic loads and PU blocking constraints.
Fig. 5 illustrates the optimal number of RSCs vs SU traffic with various given PU traffic loads and PU blocking
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
constraints. Given certain arrival rate of PU and blocking probability limit of PU, as the arrival rate of SU increases, the optimum number of RSCs also increases until the blocking probability of PU reaches the maximum admitted value. Given p the same blocking probability limit of PU, e.g., PbM AX = 0.01, as PU’s arrival rate increases from 0.1 to 0.4, the optimum number of RSCs decreases to meet PUs’ blocking constraint. 6 λ =0.1, P p
Optimum number of RSCs
5
λ =0.1, P p
4
=0.05
bMAX
λp=0.4, PbMAX=0.05 =0.01
bMAX
R EFERENCES
2
1
0.3
0.4
0.5 0.6 0.7 Arrival rate of SU (λ )
0.8
0.9
1
s
Fig. 5. Optimum number of RSCs vs SU traffic with various given PU traffic loads and PU blocking constraints.
SUs’ utility vs SU traffic to compare the case with and without RSCs given various PU traffic loads are compared in Fig. 6. Given certain arrival rate of PU, e.g., λp = 0.1, p and blocking probability limit of PU, e.g., PbM AX = 0.05, it is obvious to see that, when the arrival rate of SU is 0.8, the utility of SUs can be enhanced using our proposed channel reservation scheme as much as 30% compared with the method without RSCs. Note that our proposed scheme is more effective as the arrival rate of SU is increased. Moreover, even in the case that a heavy PU traffic (λp = 0.4), the utility of SUs can also be largely increased by RSCs than the case that a light PU traffic (λp = 0.1) without channel reservation. 10 9
With optimum RSCs, λ =0.1, P p
Without RSCs, λ =0.1, P
=0.05
7
Without RSCs, λ =0.4, P
=0.05
p
Utility of SUs
=0.05
bMAX
With optimum RSCs, λp=0.4, PbMAX=0.05
8
p
bMAX bMAX
6 5 4 3 2 1 0.2
In this paper, we have considered QoS enhancements for SUs in CRNs where PUs and SUs share a common pool of channels on a pre-emptive basis. We have proposed a channel reservation scheme to increase the average service completion rate represented by a satisfaction function for SUs while keeping the blocking probability of PUs within a specified constraint, and determined the optimal number of reserved channels under various conditions. Simulation results demonstrate that our proposed scheme can greatly improve SUs’ QoS while satisfying the required QoS constraint for PUs.
λp=0.4, PbMAX=0.01
3
0 0.2
V. C ONCLUSIONS
0.3
0.4
0.5 0.6 0.7 Arrival rate of SU (λ )
0.8
0.9
1
s
Fig. 6. SUs’ utility vs SU traffic to compare the case with and without RSCs given various PU traffic loads.
[1] FCC, “Spectrum policy task force report,” Rep. ET Docket, no. 02-155, Nov 2002. [2] J. M. Peha, “Sharing spectrum through spectrum policy reform and cognitive radio,” Proceedings of the IEEE, vol. 97, no. 4, pp. 708–719, Apr. 2009. [3] I. F. Akyildiz, W. Y. Lee, M. C. Vuran, and S. Mohanty, “A survey on spectrum management in cognitive radio networks,” IEEE Communications Magazine, vol. 46, no. 4, Apr. 2008. [4] T. Peng, W. Wang, Q. X. Lu, and W. B. Wang, “Subcarrier allocation based on water-filling level in OFDMA-based cognitive radio networks,” in IEEE WiCom’07, pp. 196–199, China, Sept. 2007. [5] P. Wang, M. Zhao, L. M. Xiao, S. D. Zhou, and J. Wang, “Power allocation in OFDM-based cognitive radio networks,” in IEEE Globecom’07, pp. 4061–4065, Nov. 2007. [6] A. T. Hoang and Y. C. Liang, “A two-phase channel and power allocation scheme for cognitive radio networks,” in IEEE PIMRC’06, pp. 1–5, Finland, Sept. 2006. [7] J. Acharya and R. D. Yates, “A framework for dynamic spectrum sharing between cognitive radios,” in IEEE ICC’07, pp. 5166–5171, Scotland, Jun. 2007. [8] E. W. M. Wong and C. H. Foh, “Analysis of cognitive radio spectrum access with finite user population,” IEEE Communications Letters, vol. 13, no. 5, pp. 294–296, May. 2009. [9] S. Geirhofer, L. Tong, and B. M. Sadler, “A cognitive framework for improving coexistence among heterogeneous wireless networks,” in IEEE Globecom’08, pp. 1–5, USA, Dec. 2008. [10] P. K. Tang, Y. H. Chew, W. L. Yeow, and L. C. Ong, “Performance comparison of three spectrum admission control policies in coordinated dynamic spectrum sharing systems,” IEEE Transactions on Vehicular Technology, vol. 58, no. 7, pp. 3674–3683, Sept. 2009. [11] M. M. Buddhikot and K. Ryan, “Spectrum management in coordinated dynamic spectrum access based cellular networks,” in IEEE DySPAN’05, pp. 299–307, USA, Apr. 2005. [12] Y. Zhang, “Dynamic spectrum access in cognitive radio wireless networks,” in IEEE ICC’08, pp. 4927–4932, China, May 2008. [13] J. Bae, E. Beigman, R. Berry, M. Honig, and R. Vohra, “Sequential bandwidth and power auctions for distributed spectrum sharing,” IEEE Journal on Select Areas in Communication, vol. 26, no. 7, pp. 1193– 1203, Sept. 2008. [14] R. Cooper, , “Introduction to queueing theory,” New York: NorthHolland, 1981. [15] R. Cooper, M. Xiao, N. B. Shroff, and E. K. P. Chong, “A utilitybased power-control scheme in wireless cellular systems,” IEEE/ACM Transactions on Networking, vol. 11, no. 2, pp. 210–221, Apr. 2003. [16] D. Niyato and E. Hossain, “A hierarchical model for bandwidth management and admission control in integrated IEEE 802.16/802.11 wireless networks,” in IEEE WCNC’07, pp. 3763–3767, Hong Kong, Mar. 2007. [17] M. L. Balinski, “Integer programming: methods, uses, computation,” Management Science, vol. 12, no. 3, pp. 253–313, Nov. 1965.
978-1-4244-5638-3/10/$26.00 ©2010 IEEE