Application of Support Vector Machines to Bandwidth Reservation in Sectored Cellular Communications∗ Chenn-Jung Huanga#, Wei Kuang Laib, Rui-Lin Luoc & You-Lin Yanb a
Institute of Learning Technology, National Hualien Teachers College, Hualien, Taiwan 97043 b
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung, Taiwan 80424
c
Department of Computer Science, National Tsing Hua University, HsingChu, Taiwan 30055
Abstract Many mechanisms based on bandwidth reservation have been proposed in the literature to decrease connection dropping probability for handoffs in cellular communications. The handoff events occur at a much higher rate in sectored cellular networks than in traditional cellular systems.
An efficient bandwidth reservation mechanism for the neighboring cells is
therefore critical in the process of handoff during the connection of multimedia calls to avoid the unwillingly forced termination and waste of limited bandwidth in the sectored cellular communications, particularly when the handoff traffic is heavy.
In this paper, a
self-adaptive bandwidth reservation scheme, which adopts support vector machines technique, is proposed to reduce the forced termination probability. Meanwhile, a channel borrowing technique is used to decrease the new call blocking probability of real-time traffic.
The
simulation results show that the proposed scheme can achieve superior performance than the representative bandwidth-reserving schemes in sectored cellular networks in the literature when performance metrics are measured in terms of the forced termination probability and the new call blocking probability. Keywords: Bandwidth reservation, wireless networks, sectored antenna, quality of service, support vector machines ∗ This research was partially supported by National Science Council under grant NSC 92-2213-E-026-001 # Contact author. Email:
[email protected] 1
1. Introduction With the increasing demand for the provision of the multimedia services in wireless networks, a great deal of attention is being paid to resource allocation for providing seamless multimedia access in the next generation mobile communication networks [1-2].
However,
how to allocate and use the limited bandwidth in a wireless environment efficiently is a challenge issue due to the existence of some bandwidth intensive multimedia applications and client mobility. There are two important Quality-of-Service (QoS) parameters considered in wireless networks, namely the handoff dropping probability (CDP) and new call blocking probability (CBP). Handoff is a mechanism that a mobile host (MH) is transferred from one base station (BS) to another during an ongoing call and the desired bandwidth should be allocated in the new cell in order to provide QoS guarantee for multimedia traffic. The CDP denotes the likelihood that an ongoing call is forced to terminate during a handoff process when the allocated resources in the new cell are degenerated to an unacceptable level, while the CBP represents the possibility that a new connection request is denied admission into the cellular networks. Accordingly, one of the most important QoS issues in providing multimedia traffic in wireless networks is to reduce handoff drops caused by lack of available bandwidth in the new cell while maintaining high bandwidth utilization and low new call blocking rate. This QoS issue is even more important in dense urban areas whereby microcellular networks or the BS sectorization is widely employed to provide greater traffic capacity in a given area than traditional macrocellular networks or nonsectorized layouts [3-4] due to much higher frequencies for handoff events.
In recent years, a variety of resource reservation algorithms
have been proposed to process handoff to ensure satisfactory reception quality in cellular networks [5-15].
Among them, Oliviera et al. suggested reserving some bandwidth in the
target cells and the neighboring cells at the same time. However, their scheme was unable to
2
adapt to the abrupt oscillation of bandwidth requirement and bandwidth utilization was deteriorated as well [8].
Levine et al. presented a shadow cluster scheme to reserve
resources with neighboring cells by exchanging information related to the movement pattern and position [9]. However, the scheme introduces too many communication overheads among the BSs of the cellular system.
In [10], a scheme based on max-min fairness
protocol to provide QoS guarantees in wireless multimedia network is proposed.
In spite of
potentially improving both the CBP and the CDP in this scheme, the users might be subjected to significant bandwidth fluctuations.
Lee et al. presented a handoff management scheme
using simultaneous multiple bindings that reduces packet loss and generates negligible delays due to handoff in IP-based third-generation cellular systems [11].
The CDP is probably
reduced whereas the bandwidth levels of ongoing multimedia traffic are also degraded. Kuo et al. took use of the knowledge of staying time, available time, and the class of the MH to develop a resource semi-reservation scenario and it turns out to be idealistic since the speed of the MH is difficult to detect accurately [13].
In [14], the traffic in a wireless system is
first divided into two classes, which are voice calls and video calls, respectively. Then a channel borrowing scheme is proposed to allow voice calls to borrow channels from those pre-allocated to video calls temporarily.
Although the CBP for the voice calls is reduced,
the issue of improving the CDP during the handoff is not addressed.
The work proposed by
Ei-Kadi et al. borrowed bandwidth from multimedia connections for supporting the new calls or handoff connections because multimedia connections can tolerate and gracefully adapt themselves to transient fluctuations in QoS [15].
The borrowed bandwidth is returned to the
original connections as soon as possible to satisfy the QoS requirements. There is 15% of bandwidth reserved exclusively for multimedia handoff connections. Thus, if a new call or handoff call requests for bandwidth, the scheme in [15] tries to borrow bandwidth from other existing connections first. If the borrowed bandwidth is insufficient for the request, the connection will try to use the bandwidth in the reservation pool. If there is no enough 3
bandwidth, the connection will be dropped. Although much attention has been paid to process the handoffs by using bandwidth reservation schemes in the traditional cellular systems, sectorized layouts in the future cellular systems, to our best knowledge, have never been considered in the literature.
In
sectored cellular systems, directional antennas are used to divide a cell into several sectors. In 3-sectored cellular systems, three directional antennas are used at the center of each cell to provide 360° of coverage.
Although the three directional antennas can be angled to improve
traffic cover and meet other network design objectives, the azimuths of three antennas are assumed to be 0, 120, 240, respectively, in this work to simplify the design of the proposed algorithm.
Fig. 1 shows an example of 120° antenna cellular system, where the alphabets A,
B, C, D, E, F, and G denote the seven cells in a cluster.
g1
b1
G
B g2
b2
a1
f1
F f3
b3
c1
A f2
C
a3
a2
e1
e3
E
c3
v
g3
c2
d1
D
e2
d3
d2
Fig. 1. A cluster of seven cells in 3-sectored cellular systems. Support vector machines (SVM) have been successfully applied in many areas, such as time series prediction [16-19], Internet traffic prediction, call classification for AT&T’s natural dialog system, multi-user detection and signal recovery for a code division multiple 4
access (CDMA) system, etc [20-24].
Moreover, there are lots of solutions on VLSI chips
which allow the SVM to be hardware-computed, and high-speed low cost SVM chips have been introduced recently, the implementation of SVM by hardware thus becomes feasible nowadays [25-27].
This work thus attempt to employ the SVM technique to estimate the
amount of reserved bandwidth in the neighboring cells. The novelties of the proposed scheme are: (1) it is the first design of the multimedia handoff management scheme from the aspect of resource reservation mechanism in the sectored antenna environment; (2) the bandwidth reservation algorithm is based on support vector machines, which is used to predict the moving direction of a MH by accessing the existing records kept in the base station; (3) the computation is performed in the local cell online and periodically to adapt to the volatile traffic conditions and exclude the signaling overhead occurring in most bandwidth reservation schemes in the literature; (4) it is well common that the reservation schemes provide small blocking rate but high drop rate, while the proposed scheme can improve both new call blocking probability and handoff call dropping probability without deteriorating bandwidth utilization. The proposed approach is also compared with another well-known machine learning technique, fuzzy logic system, which is renowned by its mathematical framework to deal with real world imprecision, and allows decision making with estimated values under incomplete or uncertain information [28]. The remainder of the paper is organized as follows.
A primitive bandwidth reservation
scheme for the sectored cellular systems is introduced in Section 2.
Then Sections 3 and 4
introduce the fuzzy logic system and support vector machines techniques, respectively, to replace the core portions of the bandwidth-reserving estimator proposed in Section 2 for better performance achievement.
Section 5 is the simulation results, which compare the
proposed approach with two major algorithms.
5
Conclusions are given in Section 6.
2. An Adaptive Resource Reservation Scheme The traffic in cellular networks is usually categorized into the following two classes in the literature [8].
Class I traffic denotes real-time multimedia traffic, such as interactive
audio and video, while Class II is non-real-time data traffic, such as images and text.
The
scheme presented in [8] anticipates that a Class I connection request will make a handoff into one of its neighboring cells in the future and thus try to reserve some bandwidth in surrounding cells before the connection request is admitted.
The Class I connection is
forced to dropped during handoff if its minimum acceptable bandwidth requirement cannot be satisfied in the entering cell.
As for Class II traffic, a handoff is always accepted as long
as there is any free bandwidth available.
Notably, the adaptive resource reservation scheme
[8] will accept a Class I handoff connection if bandwidth reservation succeeds in all new neighboring cells, whereas reject the handoff if the reservation fails in any of the six surrounding cells.
Indeed the above-mentioned scheme can effectively lower the CDP in
traditional macrocell wireless networks, nevertheless, it will introduce too many overheads among the BSs and waste much bandwidth in microcell or sectored cellular wireless systems due to excessively frequent reservation process. The resource reservation scheme proposed in this section aims at reducing overheads among the BSs and reserving bandwidth in an effective manner, while keeping the CDP and bandwidth utilization at a reasonable level. In a three-sectored cellular system, each cell is divided into three sectors. For example, cell A is split up into sectors a1, a2, and a3 as shown in Fig. 2. Suppose that a MH is located at sector a1, then the BS in cell A will realize the MH is in the upper portion of the cell, and reserve some bandwidth for possible handoff in the neighboring cells, which are cells B, G, C, and F given in Fig. 2.
In case the MH moves to sector a2, the BS will release
the entire reserved bandwidth in cell G, keep portion of reserved bandwidth in cell B, and turn to reserve some bandwidth for the new neighboring cells D and E.
6
The amount of the reserved bandwidth is determined by the following three factors: z
The probability that the MH will move to a neighboring cell will be larger if the neighboring cell is a hot cell.
z
The current reserved bandwidth for the six neighboring cells.
The probability of
moving to a neighboring cell is proportional to the bandwidth that the neighbor cell reserves. z
The sector that the MH is located at. There are more chances that the MH will move to two neighboring cells of the sector that the MH resides in. Based upon the above considerations, the bandwidth reserved in cell B for the MH
located at sector a1 when the new connection is accepted as shown in Fig. 2, can be derived as follows. BRB = BWMH ⋅ PB ⋅ PH,
(1)
where BWMH denotes the minimum bandwidth requested by a MH at sector a1, PB represents the probability that the MH moves to cell B, and PH stands for the possibility that the MH moves out of the current cell. Next this work suggests amending the amount the reserved bandwidth over a fixed span reflecting the alteration of the moving direction for the MH. The correction of the reserved bandwidth can be obtained by ∆ (BR B ) = BW MH ⋅ PB ⋅ ∆ (D MH ) ⋅ PH ,
where ∆(DMH) denotes the change of the MH's moving direction.
7
(2)
Fig. 2. The bandwidth reservation for the neighboring cells of sector a1.
3. A Fuzzy Bandwidth-Reserving Estimation Scheme The fuzzy logic technique has been used to solve several connection admission control and channel assignment problems efficiently in ATM and wireless networks in the literature [29-31].
In this section, fuzzy logic controller concept is first applied to estimate the
reserved bandwidth in the neighboring cells as shown in the scheme presented in the previous section. The fuzzy bandwidth-reserving estimator embedded in the proposed resource reservation scheme is encompassed in the dotted frame as shown in Fig. 3.
The basic functions of the
components employed in the fuzzy bandwidth-reserving estimator are described as follows. z
Fuzzifier: The fuzzifier performs the fuzzification function that converts three types of input data from the bandwidth reservation scheme into suitable linguistic values which are needed in the inference engine.
z
Fuzzy rule base: The fuzzy rule base is composed of a set of linguistic control rules and the attendant control goals. 8
z
Inference Engine: The inference engine simulates human decision-making based on the fuzzy control rules and the related input linguistic parameters.
The max-min inference
method is used to associate the outputs of the inferential rules [28], as described later in this subsection. z
Defuzzifier: The defuzzifier acquires the aggregated linguistic values from the inferred fuzzy control action and generates a non-fuzzy control output, which represents a ratio of the estimated amount of reserved bandwidth to the minimum bandwidth requested by the MH.
The Mamdani defuzzification method is employed to compute weighted average
of the aggregated output of the inferential rules due to its simplicity in computation [28].
d
Pm
Pl
Fuzzifier
RBW
Defuzzifier
BW,w Inference Engine
Fuzzy Rule Base
Fuzzy Logic Bandwidth-Reserving Estimator
Fig. 3. The fuzzy logic based bandwidth-reserving estimator. Notably, the input to the fuzzifier d represents the probability that the MH will move to a
9
neighboring cell will be larger if the neighboring cell is a hot cell.
The input Pm denotes the
normalization of the current reserved bandwidth for the six neighboring cells, and Pl the possibility that the MH moves out of the current cell. Fig. 4 shows the mapping of the inputs/output into some appropriate linguistic or membership values, which are expressed by the values within the range of 0 and 1. The input d is mapped into three linguistic term sets, “cold”, “intermediate” and “hot” to give the hotness measure of the target cell; whereas the other two inputs Pm and Pl and the output parameter of the inference engine, BW, are mapped into three linguistic term sets, “low”, “medium” and “high”.
Fig. 4. Membership function for the inputs/output parameters. The input and output fuzzy sets are correlated to establish the inferential rules of the fuzzy bandwidth-reserving estimator as listed in Table 1. By way of illustration, rule 1 can be interpreted as: IF the target cell is a “hot” cell, AND the level of current reserved bandwidth for the target cell is “low”, AND the closeness of the MH and the target cell is “low”, 10
THEN the weighting factor of the reserved bandwidth for the target cell is “medium”. Table 1 The fuzzy rule base for bandwidth-reserving estimator. Rule
d
Pm
Pl
BW
1
hot
low
low
medium
2
hot
low
medium
medium
3
hot
low
high
high
4
hot
medium
low
high
5
hot
medium
medium
high
6
hot
medium
high
high
7
hot
high
low
high
8
hot
high
medium
high
9
hot
high
high
high
10
intermediate
low
low
medium
11
intermediate
low
medium
medium
12
intermediate
low
high
medium
13
intermediate
medium
low
medium
14
intermediate
medium
medium
medium
15
intermediate
medium
high
high
16
intermediate
high
low
high
17
intermediate
high
medium
high
18
intermediate
high
high
high
19
cold
low
low
low
20
cold
low
medium
low
21
cold
low
high
low
22
cold
medium
low
medium
23
cold
medium
medium
medium
24
cold
medium
high
medium
25
cold
high
low
medium
26
cold
high
medium
medium
27
cold
high
high
medium
11
Min µ
µ
µ
µ
1
Pm
d
BW
Pl
Max
Min µ
µ
µ
µ 1
Pm
d
BW
Pl
µ
1
BW RBW
Fig. 5. The reasoning procedure for Mamdani defuzzification method. Fig. 5 illustrates the reasoning procedure for a two-rule Mamdani fuzzy inference system.
The composition of minimum and maximum operations is employed in the
evaluation of the two fuzzy rules as shown in Fig. 5.
The non-fuzzy output of the
defuzzifier can then be expressed by the following algebraic expression:
RBW =
∫ (µ (BW )⋅ BW )dBW , ∫ µ (BW )dBW A
(3)
A
where µ A (BW ) denotes the membership function of the aggregated output. The estimated bandwidth reserved in the target cell, such as cell B shown in Fig. 2, can be derived as follows: BRB = RBW ⋅ BWMH ,
(4)
where RBW stands for the non-fuzzy output, and BWMH is the minimum bandwidth requested by the MH.
12
4. Bandwidth Reservation Mechanism Using Support Vector Machines Support vector machines (SVM) have recently gaining popularity due to its numerous attractive features and eminent empirical performance [32-35].
The main difference
between the SVM and conventional regression techniques is that it adopts the structural risk minimization (SRM) approach, as opposed to the empirical risk minimization (ERM) approach commonly used in statistical learning. The SRM tries to minimize an upper bound on the generalization rather than minimize the training error, and is expected to perform better than the traditional ERM approach. Moreover, the SVM is a convex optimization, which ensures that the local minimization is the unique minimization. To solve a nonlinear regression or functional approximation problem, the SVM nonlinearly map the input space into a high-dimensional feature space via a suitable kernel representation, such as polynomials and radial basis functions with Gaussian kernels. This approach is expected to construct a linear regression hyperplane in the feature space, which is nonlinear in the original input space.
Then the parameters can be found by solving a
quadratic programming problem with linear equality and inequality constraints [32].
{
}
It is assumed that a training data set D = (x i , yi ) ∈ ℜ n × ℜ, i = 1,..., l , which consists of l pair training data (x i , yi ), i = 1,...l , is given.
The inputs xi’s are n-dimensional vectors, and
the system responses yi’s are continuous values.
Based on the knowledge of data set D, the
SVM attempts to approximate the following function: N
f (x, w ) = ∑ wi ⋅ ϕ i (x ) + b ,
(5)
i =1
where b is the bias term, and wi’s are the subjects of learning. Moreover, a mapping z = Φ(x ) is chosen in advance to map input vectors x into a higher-dimensional feature space
F, which is spanned by a set of fixed functions ϕ i (x ) ’s. By defining a linear loss function with the following ε-insensitivity zone as shown in 13
Fig. 6, ⎧ 0 yi − f (x i , w ) ε = ⎨ ⎩ yi − f (x i , w ) − ε
if yi − f (x i , w ) ≤ ε , otherwise
(6)
e
ε
ε
y−f(x,w) Fig. 6 ε-insensitivity loss function wi’s in Eq. (5) can be estimated by minimizing the risk: R=
1 ⎛ l ⎞ 2 w + C ⎜ ∑ yi − f (x i , w ) ε ⎟ , 2 ⎝ i =1 ⎠
(7)
where C is a user-chosen penalty parameter that determines the trade-off between the training error and VC dimension of the SVM model.
Note that the VC dimension is a scalar value
that measures the capacity of a set of functions [32]. Eq. (7) can be further derived into the following constrained optimization problem: R(w, ξ , ς ) =
l 1 ⎛ l ⎞ 2 w + C⎜ ∑ξ + ∑ξ * ⎟ , 2 i =1 ⎝ i =1 ⎠
subject to constraints 14
(8)
⎧ yi − w T x i − b ≤ ε + ξ ⎪ T * ⎨w x i + b − y i ≤ ε + ξ , ⎪ ξ ,ξ * ≥ 0 ⎩
(9)
where ξ and ξ * represent the measurements above and below the zone with the radius ε in Vapnik’s loss function as given in Eq. (6), respectively. It can be shown [32] that the above constrained optimization problem is solved by applying the Karush-Kuhn-Tucker (KKT) conditions [36] for regression, and maximizing the following Lagrangian: L(α ) = −0.5α T Hα + f T α ,
(10)
under constraints l l ⎧ * = α αi ∑ ∑ i ⎪ i =1 i =1 ⎪ ⎨ 0 ≤ α i ≤ C , i = 1,..., l , ⎪0 ≤ α * ≤ C , i = 1,..., l i ⎪ ⎩
(11)
where f = [ε − y1 ε − y 2 ... ε − y N ε + y1 ε + y 2 ... ε + y N ] , (α i ,α i* ) denotes one of l Lagrange multiplier pairs, and the Hessian matrix H is given as ⎡ G − G⎤ H=⎢ ⎥. ⎣− G G ⎦
(12)
G denotes the corresponding kernel matrix.
The best nonlinear regression hyperfunction is then given by f (x, w ) = Gw o + bo ,
(13)
where wo and bo denote the optimal desired weights vector and the optimal bias, respectively. wo and bo can be derived by
w o = α* − α ,
(14) 15
bo =
1 l ∑ ( yi − g i ) , l i =1
(15)
where g=G wo.
5. Simulation Results A series of simulations are conducted to compare the proposed support vector machine reservation scheme (SVMR) with the fuzzy logic reservation scheme (FLR), the fixed reservation scheme (FR), and the scheme without bandwidth reservation (NR). Meanwhile, the rate-based borrowing scheme (RBB) [15] is also compared in this study because it was reported in [15] that the RBB scheme achieves better performance than other representative bandwidth allocation and reservation schemes in the literature, such as the scheme presented in [8].
The Gaussian basis function is selected as the kernel function in the proposed
SVMR scheme. In the NR scheme, no bandwidth is reserved for handoff connections in each cell. If there is no bandwidth available when the MH moves to the new coverage area, the handoff call is disconnected and a forced termination occurs. The CDP is expected to be higher in the BS sectorized layout where the handoff events occur at a much higher rate. As for the FR approach, a set of channels called guard channels are preserved in each cell to provide a way of prioritizing handing off calls on new call originations by setting aside a fixed bandwidth to support handing off users. New call originations cannot be assigned bandwidth from the guard channel pool. The guard channels are set to 20% of the whole bandwidth for the FR scheme in our simulations. The connections are divided into two classes. A Class I traffic, which is a multimedia connection, is allowed to move to a neighboring cell only when the unallocated bandwidth in the target cell exceeds the requested bandwidth. A data connection of Class II traffic can be granted to switch to a neighboring cell as long as the target cell possesses any unused bandwidth.
Additionally, a new
connection of Class I real-time traffic is allowed to borrow bandwidth from Class II non
16
real-time connections in the same cell if the unallocated bandwidth in the current cell is smaller than the minimum bandwidth that the new Class I traffic requests in the proposed SVMR scheme. Similar approach was taken in [14] to effectively reduce the new call blocking probability of real-time traffic. There are 36 cells included in the simulation environment as shown in Fig. 7. A total of 30 Mbps bandwidth is allocated in each cell. Both classes of the connections are listed in Table 2. The bandwidth requirement for each connection is randomly selected within the range of the maximum and the minimum bandwidth requirement listed in Table 2. Both the class and the location of each MH are randomly selected at the initial state. Each MH is given a speed characteristic, which decides the time spent in a cell, in order to simulate handoffs. If a hot cell neighbors with the cell that a MH is located at, then the MH has a probability of 0.5 to move to the neighboring hot cell, and a probability of 0.1 to one of other neighboring cells. On the other hand, each MH will move to one of the six neighboring cells with equal probability if no neighboring hot cell exists. Figs. 8 through 9 show the comparison of the call blocking probability and the call dropping probability for the five schemes in the simulation, respectively. The bandwidth utilization of the five schemes is given in Fig. 10. Note that the bandwidth utilization is defined as:
Bandwidth Utilization =
∑ Used bandwidth of each cell . ∑ Maximum bandwidth of each cell for each cell
(16)
for each cell
As expected, the simulation results illustrate that the proposed SVMR scheme is better than the other four schemes if the objective is to improve perceived quality of cellular service by decreasing call dropping probability. The performance of the FLR scheme is inferior to the SVRM owing to its lack of generalization ability to the varied bandwidth requirement in the cellular networks. The call blocking probability is also improved by means of the
17
channel borrowing technique. However, the bandwidth utilization of the SVMR scheme is worse than the NR scheme when traffic load is heavy. It is believed that this behavior is caused by a waste of the excessive bandwidth reservation for the considerable amounts of the new calls and the handoff calls in the heavy traffic load.
Fig. 7. Cellular topology with 36 cells in the simulations.
Table 2. Multimedia Traffic Used in the simulations
Class I
Bandwidth Requirement 30 Kbps
Average Call Duration 3 minutes
Voice Service
Class I
256 Kbps
5 minutes
Video-Phone
Class I
1~4 Mbps
10 minutes
Video Service
Class II
5-20 Kbps
0.5 Seconds
E-mail, Paging
Class II
64~512 Kbps
3 minutes
Class II
1~5 Mbps
2 minutes
Traffic Class
18
Example
Remote Login & Data on Demand Ftp
NR
FR
RBB
FLR
SVMR
Call blocking probability 0.7
0.6 0.5 0.4 0.3 0.2 0.1 0 0.02
0.05
0.1
0.2
0.5
1
New call arrival rate Fig 8. Call blocking probability for the SVMR, FLR, NR, and FR schemes. Call dropping
NR
probability
FR
RBB
FLR
SVMR
0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0.02
0.05
0.1
0.2
0.5
1
New call arrival rate
Fig. 9. Call dropping probability for the SVMR, FLR, NR, and FR schemes.
19
Bandwidth utilization
NR
FR
RBB
FLR
SVMR
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0.02
0.05
0.1
0.2
0.5
1
New call arrival rate
Fig. 10. Bandwidth utilization for the SVMR, FLR, NR, and FR schemes.
6. Conclusion In this paper, a bandwidth reservation scheme is proposed to reduce forced termination or QoS degradation in the sectored cellular networks.
A support vector machine is
employed to compute the amount of reserved bandwidth reserved in the target cell. This work also tries to decrease the new call blocking probability of real-time multimedia traffic by using a channel borrowing technique. The simulation results show that the proposed scheme (SVMR) is better then the fuzzy logic reservation scheme (FLR), the fixed reservation (FR) scheme, the scheme without reservation (NR), and the rate-based borrowing scheme (RBB) when the new call blocking probability and the call dropping probability (forced termination probability) are compared.
The bandwidth utilization is also kept in an
acceptable level, which is merely worse than the scheme without bandwidth reservation when 20
the traffic load becomes heavy. Therefore, the proposed scheme is proved to be a good candidate for the bandwidth reservation scheme in sectorized cellular networks, which are typically deployed in the dense urban areas where the handoff events often occur at a much higher rate. Meanwhile, it is well known that the no reservation scheme can provide small blocking rate but high drop rate. However, the evidences show that the proposed scheme can effectively improve handoff call dropping probability than other reservation-based schemes, and also achieve lower blocking rate than the no-reservation scheme does. Although the three-sectored cellular networks are assumed in the simulations, the scheme can be also applied to the six-sectored cellular networks with a slight modification.
Subsequent
research will investigate the feasibility of applying other intelligent tools such as neuro-fuzzy and genetic algorithms into the proposed scheme to improve the accuracy of the motion prediction for the MH.
21
References [1] Y.-B. Lin, and I. Chlamtac, “Wireless and Mobile Network Architectures,” John Wiley & Sons, 2001. [2] N. D. Tripathi, J. H. Reed, and H. F. Vanlandingham, “Handoff in Cellular Systems,” IEEE Personal Communications, pp. 26-37, 1998. [3] M. V. Clark, V. Erceg, and L. J. Greenstein, “Reuse Efficiency in Urban Microcellular Networks,” IEEE Trans. Vehicular Technology, vol. 46, no. 2, pp. 279-288, 1997. [4] H. Jiang, and S. S. Rappaport, “Channel Borrowing Without Locking for Sectorized Cellular Communications,” IEEE Trans. Vehicular Technology, vol. 43, no. 4, pp. 1067-1077, 1994. [5] T. Liu, P. Bahl, and I. Chlamtac, “Mobility Modeling, Location Tracking, and Trajectory Prediction in Wireless ATM Networks,” IEEE J. Selected Areas in Comm., vol. 16, no. 6, pp. 922-936, 1998. [6] G.-C. Lee, T.-P. Wang, and C.-C. Tseng, “Resource Reservation with Pointer Forwarding Schemes for Mobile RSVP,” IEEE Comm. Letters, vol. 5, no. 7, pp. 298-300, 2001. [7] S. Boumerdassi, and A.-L. Beylot, “Adaptive Channel Allocation for Wireless PCN,” Mobile Networks and Applications, vol. 4, pp. 111-116, 1999. [8] C. Oliveira, J. B. Kim, and T. Suda, “An Adaptive Bandwidth Reservation Scheme for High-Speed Multimedia Wireless Networks,” IEEE J. Selected Areas in Comm., vol. 16, no. 6, pp. 858-874, 1998. [9] D. A. Levine, I.F. Akyildiz, and M. Naghshineh, “A Resource Estimation and Call Admission Algorithm for Wireless Multimedia Networks Using the Shadow Cluster Concept,” IEEE/ACM Trans. Networking, vol. 5, no. 1, pp. 1-12, 1997. [10] A. Malla, M. El-Kadi, and P. Todorova, “A Fair Resource Allocation Protocol for Multimedia Wireless Networks,” IEEE Inter. Conf. Parallel Processing, pp. 437-443, 2001. [11] J.-H. Lee, T.-H. Jung, S.-U. Yoon, S.-K. Youm, and C.-H. Kang, “An Adaptive Resource Allocation Mechanism Including Fast and Reliable Handoff in IP-Based 3Gwireless Networks,” IEEE Personal Communications, pp. 42-47, 2000. [12] M. R. Sherif, I. W. Habib, M. Nagshineh, and P. Kermani, “Adaptive Allocation of Resources and Call Admission Control for Wireless ATM Using Genetic Algorithms,” IEEE J. Selected Areas in Comm., vol. 18, no. 2, pp. 268-282, 2000. [13] G.-S. Kuo, P.-C. Ko, and M.-L. Kuo, “A Probabilistic Resource Estimation and Semi-Reservation Scheme for Flow-Oriented Multimedia Wireless Networks,” IEEE Communications Magazine, pp. 135-141, 2001. [14] X. Wu, K. L. Yeung, and J. Hu, “Efficient Channel Borrowing Strategy for Real-Time Services in Multimedia Wireless Networks,” IEEE Trans. Vehicular Technology, vol. 49, no. 4, pp. 1273-1284, 2000. [15] M. Ei-Kadi, S. Olariu, and H. Abdel-Wahab, “Rate-Based Borrowing Scheme for QoS Provisioning in Multimedia Wireless Networks,” IEEE Trans. Parallel and Distributed Systems, vol. 13, no. 2, pp. 156–166, 2002. [16] L. J. Cao, and F. E. H., Tay, “Support Vector Machine with Adaptive Parameters in 22
Financial Time Series Forecasting,” IEEE Transactions on Neural Networks, vol. 14, no. 6, pp. 1506 – 1518, 2003. [17] T. Van Gestel, J.A.K. Suykens, D.-E. Baestaens, A. Lambrechts, G. Lanckriet, B. Vandaele, B. De Moor, and J. Vandewalle, ”Financial Time Series Prediction Using Least Squares Support Vector Machines Within the Evidence Framework,” IEEE Transactions on Neural Networks, vol. 12, no. 4, pp. 809 – 821, 2001. [18] J.-Y. Zhu, B. Ren, H.-X. Zhang, and Z.-T. Deng, “Time Series Prediction via New Support Vector Machines,” 2002 International Conference on Machine Learning and Cybernetics, vol. 1, pp. 364 – 366, Nov. 2002. [19] T. Raicharoen, C. Lursinsap, and P. Sanguanbhokai, “Application of Critical Support Vector Machine to Time Series Prediction,” 2003 International Symposium on Circuits and Systems, vol. 5, pp. V-741 - V-744, May 2003 [20] X. Gong; and A. Kuh, “Support Vector Machine for Multiuser Detection in CDMA Communications,” The Thirty-Third Asilomar Conference on Signals, Systems, and Computers, vol. 1, pp. 680 – 684, Oct. 1999. [21] Chen, S.; Samingan, A.K.; Hanzo, L. “Support Vector Machine Multiuser Receiver for DS-CDMA Signals in Multipath Channels,” IEEE Transactions on Neural Networks, vol. 12, no. 3, pp. 604 – 611, May 2001. [22] P. Haffner, G. Tur,, J. H. Wright, “Optimizing SVMs for Complex Call Classification,” 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 1, pp. I-632 − I-635, April 2003. [23] A. Kuh, “Adaptive Kernel Methods for CDMA Systems,” 2001 IEEE International Joint Conference on Neural Networks, vol. 4, pp. 2404 – 2409, July 2001. [24] M. Hasegawa, G. Wu, and M. Mizuno, “Applications of Nonlinear Prediction Methods to the Internet Traffic,” The 2001 IEEE International Symposium on Circuits and Systems, vol. 2, pp. III-169−III-172, May 2001. [25] D. Anguita, and A. Boni, “Towards Analog and Digital Hardware for Support Vector Machines,” 2001 International Joint Conference on Neural Networks, vol. 4, pp. 2422 – 2426, July 2001. [26] D. Anguita, A. Boni, and S. Ridella, “Learning Algorithm for Nonlinear Support Vector Machines Suited for Digital VLSI,” Electronics Letters, vol. 35, no. 16, pp. 1349 – 1350, 1999. [27] D. Anguita, A. Boni, and S. Ridella, “A VLSI Friendly Algorithm for Support Vector Machines,” 1999 International Joint Conference on Neural Networks, vol. 2, pp. 939 – 942, July 1999. [28] J. Buckley, and Eslami, ‘An Introduction to Fuzzy Logic and Fuzzy Sets (Advances in Soft Computing),’ Physica Verlag, 2002. [29] Q. Ren, and G. Ramamurthy, “A Real-Time Dynamic Connection Admission Controller Based on Traffic Modeling, Measurement, and Fuzzy Logic Control,” IEEE J. Selected Areas in Comm., vol. 18, no. 2, pp. 184-196, 2000. [30] B. Bensaou, S. T. C. Lam, H.-W., Chu, and D. H. K. Tsang, “Estimation of the Cell Loss Ratio in ATM Networks with a Fuzzy System and Application to Measurement-Based Call Admission Control,” IEEE/ACM Trans. Networking, vol. 5, no. 4, pp. 572-584, 1997. 23
[31] J. Ye, X. Shen, and J. W. Mark, “Call Admission Control in Wideband CDMA Cellular Networks by Using Fuzzy Logic,” 2003 IEEE Wireless Communications and Networking, vol. 3, pp. 1538 –1543, March 2003. [32] V. Vapnik, The Nature of Statistical Learning Theory. New York: Springer-Verlag, 1995. [33] C. J. C. Burges, “A Tutorial on Support Vector Machines for Pattern Recognition,” Data Mining and Knowledge Discovery, vol. 2, no. 2, pp. 121–167, 1998. [34] V. Vapnik, “The Support Vector Method of Function Estimation,” in Non-linear Modeling: Advanced Black-Box Techniques, J. A. K. Suykens and J. Vandewalle, Eds. Boston, MA: Kluwer, 1998, pp. 55–85. [35] “Special Issue on VC Learning Theory and Its Applications,” IEEE Transactions on Neural Networks, vol. 10, no. 5, Sept. 1999. [36] H. A. Taha, Operations Research: An Introduction. Prentice-Hall, Inc, 1997.
24
