運用虛擬分割之階層式多服務無線網路之資源分配

行政院國家科學委員會補助專題研究計畫成果報告

運用虛擬分割之階層式多服務無線網路之資源分配

Resource Allocation of Hierarchical Multi-Service Wireless Networks Using Virtual Partitioning

計畫編號：NSC 96－2221－E－011－018 執行期限：96 年 8 月 1 日至 97 年 7 月 31 日

主持人：鍾順平 國立台灣科技大學電機系

計畫參與人員：張益禎、徐士茵、洪裕鈞、王建文、翁士洋 國立台灣科技大學電機系

一、中文摘要

由於連結允入控制(CAC)決定如何分配資源 給新連結與交遞連結以滿足服務品質(QoS)需 求，所以 CAC 是資源分配之一重要環節。我 們考慮支援多種用戶之階層式細胞網路，其中 不同用戶可具有不同之頻寬需求、平均服務時 間與移動性，且快速移動用戶是指配給上層之 巨細胞，而慢速移動用戶則是指配給下層之微 細胞。我們針對所研究系統提出一種運用虛擬 分割之連結允入控制(CAC-VP)。對於數學分 析，我們採用多維馬可夫鏈來描述微細胞與巨 細胞。我們並自行以 C 語言撰寫相關電腦模擬 程式以驗證解析結果之準確度。最後我們採用 連結位準之服務品質做為效能量度。

關鍵詞: 連結允入控制，虛擬分割，階層式網 路，阻塞機率

Abstract

Call admission control (CAC) is a very important component of resource allocation since CAC determines how to allocate resource to new calls and handoff calls in order to satisfy the quality of service (QoS) constraints. We consider hierarchical cellular networks with multiple classes of users with different bandwidth requirements, average call holding times, and mobility, where fast users are assigned to macrocells in the upper layer, whereas slow users are assigned to the microcells in the lower layer. We propose and study CAC with virtual partitioning (CAC-VP) for the system considered. For the mathematical analysis, we use multi-dimensional Markov chains to describe the microcell and macrocell.

Simulation program is written in C language to verify the accuracy of analytical results.

Call-level grade of service (GoS) is used as performance measures.

Key words: call admission control, virtual partitioning, hierarchical networks, blocking probability

二、緣由與目的

On one hand, the number of mobile users is growing up rapidly in recent years, since in addition to voice service using mobile phones, different kinds of equipments, e.g., PDAs and notebook computers, are being used over wireless links. On the other hand, radio channels are well known to be scarce resource. Therefore, it is important in the wireless cellular networks to make efficient use of scarce wireless resource while keeping Quality of Service (QoS) at an acceptable level. Thus, it is necessary to design a suitable call admission control (CAC) scheme applied to wireless networks in order to achieve the aforesaid goals.

In general, two commonly used QoS measures in wireless cellular networks are new call blocking probability and forced termination probability. In fact, blocking of new calls is more desirable than forced termination of calls in progress. The impact of forced termination probability is greater than that of new call blocking probability in terms of the QoS experienced by users. Thus, handoff calls are commonly given priority over new calls in order to reduce forced termination probability. There are basically two ways to prioritize handoff calls:

guard channel and handoff queuing.

Research on CAC has acquired much

attention in recent years. A complete sharing (CS) policy in which all users can be allowed without restriction to access a common resource is proposed in [1][2]. On the other hand, a complete partitioning (CP) divide available resource into separate sets with corresponding nominal capacity based on different QoS requirements, and each class of traffic is allocated a set of resources that can only be used by that class [2]. The concept of Virtual partitioning (VP) is first proposed by Mitra and Ziedins in [3]. The concept of VP is that each individual traffic class is allocated a nominal capacity according to expected offered load as well as required Qos. When overall traffic is light (heavy), VP behaves like unrestricted sharing (complete isolation). Therefore, VP manages to combine characteristics of CS and CP under different loads. VP with preemption for the scenarios with K classes of users in

single-layer cellular networks are studied in [4].

In this work, we consider the performance of two-tier cellular networks for a multiple of classes of users with CAC-VP policy, where each class of users may have different bandwidth requirements, average call holding times, and mobility We also implement a guard channel scheme to further reduce forced termination probability. Besides, for comparison purposes, we also consider CAC-CS with guard channels in hierarchical cellular networks. Mobile users with slow mobility are assigned to the associated microcell, whereas users with fast mobility are assigned to the associated macrocell. If users with slow mobility cannot be admitted into the associated microcell, it can be overflowed into the associated macrocell. In microcells, each underloaded group of slow mobility user can preempt another overloaded group of slow mobility users if necessary, but only fast mobility users can preempt the users of the other group in macrocells. An analytical method is derived to calculate the performance metrics of interest, e.g., new call blocking probability, and forced termination probability.

三、結果與討論

3.1. System model

We consider a two-tier hierarchical cellular

network which consists of many microcells in the lower layer and many macrocells in the upper layer, where each macrocell overlays

N

( )m

CN

ov

CN N

CS

CN

MS can move in one of four directions at a constant speed. The trajectory of MS is restricted on the street. The MS at every junction turn to right or left with a given probability of

p , and

p

It is assumed that there are

C

C

b channel(s) for transmission. Groups

/ 2 1

N + toN.

To simplify our analysis, hard handoff scheme is adopted, i.e., there are no overlapping regions with corresponding neighbor cells. As is well known, maintaining call continuity is more important than admitting a new call. To assign higher priority to handoff calls over new calls, a guard channel scheme is employed in the system.

3.1.1 Traffic model

New call arrivals for class- k slow (fast) users at a microcell (macrocell) are assumed to follow a Poisson process with average arrival rate

λ λ

k

N

arrivals are also assumed to be Poisson.

Overflow call arrivals, if exist, are also assumed to be Poisson. It is noted that both handoff and overflow call arrival rates are derived iteratively.

The arriving location of any call is selected at random and the moving direction is randomly generated.

We assume that call holding times for slow (fast) users are exponentially distributed with mean

μ

μ

( , )s m

μ

μ

μ

3.1.2 CAC-CS

Using the CAC-CS scheme, all users are admitted as long as enough free channels are available. On one hand, CAC-CS can lead to efficient usage of scare wireless resource and achieve maximum channel utilization. On the other hand, users with lower bandwidth requirement are more likely to be admitted.

3.1.3 CAC-VP

Using CAC-VP, we divide N classes of users into two groups: group-1(

GR

GR

3.2 Analytical Method

In this section, an analytical method for computing the performance measures of interest in a two-tier cellular system is derived. Since we assume that the studied system is homogeneous, we can focus on one particular cell at each layer for analysis purposes. It is noted that slow (fast)

class-k handoff arrival rate

λ λ

k

N

handoff rate or cell dwell rate is assumed to be exponentially distributed with mean _hk

v

μ

s

v

s is the size of the square cell in question.

3.3. Numerical Results

We consider only homogeneous systems, i.e.

all cells in the same layer in the network considered are statistically identical. The simulation program is written in C language and is used to verify the accuracy of the analytical results. The widely used Manhattan Model is employed in the simulation. There are four classes of calls in the system considered.

Furthermore, when a new call is generated, it can move in any one of the four directions: east, west, south and north. The mobile users travel in one of the directions at a constant speed, either a high speed or a low speed. Therefore, for each class of users they can be further divided into fast users and slow users. The movement of the mobile users is restricted along the streets and the originating location of new calls is randomly selected. The cell dwell time is determined by the originating location of the new calls, the moving directions, and the speed of the mobiles.

The performance measures of interest are new call blocking probability, forced termination probability and preemption probability. The performance measures of CAC-CS and CAC-VP are compared.

3.3.1 Accuracy of Analytical Results

It is observed that the analytical results are in good agreement with the simulation results.

The difference between analytical and simulation results seems to appear in the heavy traffic conditions, where the analytical results under- estimate simulation results. One of the possible reasons for the difference is that we assume the overflow traffic to be Poisson in our analysis. The other reason for the difference is that although the mobiles are assumed to have a constant speed, we approximate the cell dwell time distribution by an exponential distribution.

When slow new call arrival rate increases, the difference in analytical results and simulation results also increase, since the overflow traffic

increases. It is worth mentioning that when fast new call arrival rate increases, the difference in analytical results and simulation results become smaller, since the overflow traffic decreases.

3.3.2 Comparison of CAC-CS and CAC-VP

Next, recall that CAC-VP approaches CP as the load increases. This implies that CAC-VP can achieve fairness among different group of users when the system load is heavy. To demonstrate this, we compare the performance of CAC-VP and CAC-CS with one group of users exceeding its nominal offered load. As an example, we increase the load of slow calls of group-1. The new call blocking probability and forced termination probability of all classes of slow and fast calls versus the total group-1 slow new call arrival rate using CAC-CS and CAC-VP with

λ

Group-2, in the system, the new call blocking probability and forced termination probability of users of the other group (group-1) increase significantly with CAC-CS. On the other hand, with CAC- VP, it is observed that the new call blocking probability and forced termination probability of users of group-1 increase slowly.

This is due to the fact that if users of one group occupy capacity nominally allocated to the other group, users of the other group can still access the nominal capacity assigned to them by the preemption process. Therefore, CAC-VP has the function of protecting behaving users against excess resource utilization of misbehaving users.

In other words, CAC-VP achieve a higher level of fairness for different groups in heavy traffic condition and maintains better QoS for behaving users.

四、成果自評

In this work, we study CAC-VP for two-tier cellular networks supporting multiple classes of users, where each class of users may have different bandwidth requirement, average call holding time, and mobility. We use multi-dimensional Markov chains to describe the considered system and derive an analytical method to compute the performance measures of interest iteratively. Although we did not prove the convergence of our iterative algorithm, for all the cases studied, convergence is always observed. The performance measures of interest are call-level grade of service (GoS), such as new call blocking probability and forced termination probability. The simulation program is written in C language not only to verify the accuracy the analytical results but also to understand the details of CAC. It is shown that for most of the scenarios considered the analytical results are in reasonable agreement with the simulation results. It is demonstrated the new call blocking probability and forced termination probability with CAC-VP are smaller than those with CAC-CS. Furthermore, it is shown that CAC-VP can protect behaving traffic against the excess resource usage of misbehaving users and maintain better GoS for behaving users.

五、參考文獻

[1] S. T. Yang and A. Ephremides, “On the optimality of complete sharing policies of resource allocation,” Proc. IEEE CDC’96, pp 299-300, 1996.

[2] C. C. Beard and V. S. Frost,” Prioritized resource allocation for stressed networks,” IEEE

Trans. on Networking, 2001 pp. 618-633.

[3] D. Mitra and I. Ziedins,” Virtual partitioning by dynamic priorities: Fair and efficient sharing,” Broadband comm.., B. Plattner, Springer-Verlag, 1996, pp. 173-185.

[4] J. Yao, J. W. Mark, and T. C. Wong, “Virtual partitioning resource allocation for multiclass traffic in cellular systems with Qos constraints,”

IEEE Trans. on Vehicular Technology, 2004, pp.

847-864.