行政院國家科學委員會補助專題研究計畫
□ 成 果 報 告
期中進度報告
具高效能及服務品質之合作式網路設計
子計畫二：合作式網路之頻寬管理機制
計畫類別：□ 個別型計畫
整合型計畫
計畫編號：NSC 98－2221－E－009－059－MY2
執行期間： 98 年 8 月 1 日至 100 年 7 月 31 日
計畫主持人： 林一平 (國立交通大學 資訊工程學系(所) 教授)
共同主持人： 楊舜仁 (國立清華大學 資訊工程學系(所) 助理教授)
計畫參與人員： 邱冠龍、高健淇、李易達、林沂珊、王佳韞、施宗欽、
莊雅茹、吳家瑋、邱靖捷、謝萬瀚
成果報告類型(依經費核定清單規定繳交)：
精簡報告 □完整報告
本成果報告包括以下應繳交之附件：
□赴國外出差或研習心得報告一份
□赴大陸地區出差或研習心得報告一份
出席國際學術會議心得報告及發表之論文各一份
□國際合作研究計畫國外研究報告書一份
處理方式：除產學合作研究計畫、提升產業技術及人才培育研究計畫、
列管計畫及下列情形者外，得立即公開查詢
□涉及專利或其他智慧財產權，□一年
二年後可公開查詢
執行單位：國立交通大學、國立清華大學
中 華 民 國 99 年 5 月 30 日
附件一
（一） 計畫中文摘要。
（五百字以內）
隨著使用者對於高速無線網際網路的需求與期待日益增加，未來的行動通訊系統被期
待能提供更高的資料傳輸率、更廣的訊號涵蓋範圍及更可靠的移動性。為了滿足無線網際
網路高頻寬的需求，基於基礎建設網路（Infrastructure-based network）之多躍中繼（Multihop
Relay）技術於近年內被提出，並迅速成為被廣泛討論的研究議題。多躍中繼網路（Multihop
Relay Network）是一種合作式通訊網路（Cooperative Network）
，其主要概念是增設中繼站
（Relay Station，RS）作為 BS 與 MS 之間的轉傳（relay）媒介。只要將 RS 佈放在適當的
位置，訊號將可避開不理想的傳播路徑並減少訊號衰減。由於在轉傳訊號時可再度提高其
傳輸功率及使用直視傳輸（Line-Of-Sight，LOS），因此 MS 所接收到的訊號品質可以大幅
改善。在以上這些優點同時作用下，網路系統最終能達到提升傳輸速率 （throughput
enhancement）
、增加系統容量（capacity enhancement）
、延伸服務範圍（coverage extension）、
解決遮蔽衰落效應影響等目的。此外，由於 RS 與 BS 之間採用無線傳輸，因此可以節省後
端固接網路（wire-line backhaul）的建置成本。
本計畫將針對 IEEE 802.16j Multihop Relay（MR）WiMAX 系統，深入研究多躍中繼網
路之 RS 分群演算法及空間再利用無線電頻寬排程演算法。利用多中繼網路技術，IEEE
802.16j 標準已被制定用以提升現有 IEEE 802.16e 網路的系統效能。然而，多躍中繼技術同
時也會造成頻繁換手的問題，進而降低 IEEE 802.16j 的系統效能。RS 分群機制是用以解決
此類問題的一項技術。RS 分群的概念是將相鄰的 RS 聚集成為一個 RS 群組(邏輯上可視為
一個覆蓋大範圍的 RS)。在這個計畫中，我們以系統流量(throughput)和換手頻率(handoff rate)
做為研究 RS 分群制制效能的依據。我們提出一個貪婪 RS 分群演算法用以最小化換手頻
率。在此演算法中，我們以相鄰 RS 間的換手頻率為依據，換手頻率較高的 RS 配對即較優
先被選為同一群組。我們的網路模擬結果顯示，我們提出的 RS 分群演算法可大量地減少
換手頻率。此外，我們提出兩種集中式空間再利用排程演算法。第一種為流量優先策略，
用以最大化系統總流量。第二種為延遲優先策略，用以最小化封包延遲時間。模擬結果指
出延遲優先策略不僅能最小化平均封包延遲時間，同時也能提供良好的公平性給不同流量
負載的使用者。
關鍵詞：多躍中繼網路、合作式通訊網路、RS、IEEE 802.16j 網路、RS 分群演算法、
空間再利用
（二） 計畫英文摘要。
（五百字以內）
As the user demand for high-speed wireless communication service increases gradually, the
future wireless mobile systems are expected to provide higher data transmission rate, wider signal
coverage and more reliable mobility. In order to meet the high-bandwidth requirement,
infrastructure-based multihop relay technologies have been proposed in the recent years. A
multihop relay network is a type of cooperative networks. The main concept of multihop relay
networks is to establish relay stations (RSs) as relay media between base stations (BSs) and
mobile stations (MSs). If the RSs are deployed in proper positions, the situation of choosing
unsatisfactory signal routes can be avoided and the signal attenuation can be reduced. In addition,
while relaying signals the transmission power can be increased and the line-of-sight transmission
may be used. In this case, the signal quality which MSs receive can be improved significantly.
From the above advantages, multihop relay networks can achieve the goals of throughput
enhancement, capacity enhancement and coverage extension. Moreover, based on the wireless
transmission between RSs and BSs, the installation cost of the wire-line backhaul can be
completely eliminated as well.
The IEEE 802.16j standard has been developed to provide performance enhancement to the
existing IEEE 802.16e network by incorporating the multihop relay (MR) technology. However,
frequent handoffs and low spectrum-utilization issues that were not encountered in IEEE 802.16e
may be incurred in IEEE 802.16j. The relay station (RS) grouping is one optional mechanism in
the IEEE 802.16j MR standard to overcome these problems. The concept of RS grouping is to
group neighboring RSs together to form an RS group, which can be regarded as a logical RS with
larger coverage. In this project, we investigate RS grouping performance enhancement in terms
of throughput and handoff frequency. We design an RS grouping algorithm to minimize handoffs
by utilizing a greedy grouping policy: RS pairs with higher handoff rates will have higher priority
for selection. The simulation results show that the handoff frequency of the considered MR
network can significantly be reduced, and suitable RS grouping patterns can be derived using our
grouping algorithm. In addition, we propose two centralized scheduling policies, i.e., the
throughput-first (TF) policy to maximize the system throughput and the delay-first (DF) policy to
minimize the average packet delay. By integrating our RS grouping algorithm and centralized
scheduling algorithms, the simulation results indicate that, for the case of fixed users, groupings
with smaller group sizes can result in better throughput performance. However, when user
mobility is considered, the throughput value increases as the group size increases. Furthermore,
we also show that the DF policy can both minimize the average packet delay and provide the
fairness property among users with different traffic loads.
Keywords: Grouping algorithm, IEEE 802.16j, multihop relay (MR), scheduling policy,
WiMAX.
（三）研究計畫之背景及目的
本研究計畫之背景
Incorporating the multihop relay (MR) technology [25], the IEEE 802.16j MR standard [14]
has been developed to provide throughput improvement, coverage extension, and capacity
enhancement to the existing IEEE 802.16e protocol [1]. By deploying relay stations (RSs), the
end-to-end communication quality between base stations (BSs) and mobile stations (MSs) can be
improved without high infrastructure deployment costs. In addition, spatial reuse [19] is another
promising approach that can be employed in IEEE 802.16j MR networks to improve spectral
efficiency. Based on the centralized scheduling, spatial diversity gain can be achieved if multiple
simultaneous transmissions using the same bandwidth resources are realized within the same BS
cell.Although IEEE 802.16j has the potential to provide substantial performance enhancements,
several issues that were not addressed in IEEE 802.16e may be encountered. For example,
frequent handoffs may occur during the movements of MSs since the RS cells are smaller than
the BS cells. To avoid the consequent performance degradation, RS grouping has been identified
as an optional mechanism in the IEEE 802.16j standard. The main idea of RS grouping is to put
adjacent RSs together to form an RS group, wherein the RS members are required to
simultaneously receive and transmit the same data. From the MSs’ viewpoint, the RS group can
be regarded as a logical RS with larger coverage. Therefore, the handoff frequency can be
reduced since no handoff procedure would be triggered, even when an RS crossing event within
the RS group occurs.
本研究計畫之目的
To the best of our knowledge, no RS grouping strategy has been proposed or discussed in
the literature. We argue that different grouping criteria may lead to various performance results.
Specifically, utilizing a smaller RS group size is advantageous to spatial reuse, because more RS
groups can perform simultaneous transmissions at the same time and frequency. Thus, improved
average system throughput can be expected. However, a smaller RS group size may also lead to
higher packet loss rate due to more frequent handoffs between RS groups. In conclusion, when
implementing the RS grouping mechanism in IEEE 802.16j MR networks, the performance
tradeoff between throughput and handoff frequency should seriously be considered. In this
project, we analyze grouping strategies for RS grouping-enabled IEEE 802.16j MR networks and
propose an efficient RS grouping algorithm to minimize the handoff frequency. As we have
pointed out, RS grouping strategies will also influence the throughput performance. To
investigate the impacts of RS grouping on the IEEE 802.16j system throughput, we design two
centralized downlink (DL) scheduling policies for RS-grouping-enabled IEEE 802.16j MR
networks. One of these two scheduling policies aims to maximize the system throughput, and the
other is to minimize the average packet delay.
（四）研究計畫之方法與成果
(1) 設計最少換手次數之 RS 分群演算法
我們針對 IEEE 802.16j 標準中的 RS 群組化機制，透過分析 RS 分群的一些特性，設計
出理想的 RS 分群演算法。我們發現只要控制合理的群組大小，且能找到群組內各 RS 間之
相交邊數最多的群組排列，便能使 IEEE 802.16j 網路的 MS 換手次數有效降低，且同時達
到有效無線頻寬分配及傳輸速率提高等優點，這些特性都是我們設計 RS 分群演算法時的
重要考量因素，我們稱之為最少換手次數之 RS 分群演算法(Handover-Minimizing RS
Grouping Algorithm)。
(2) 設計使用空間再利用的集中式下傳頻寬排程演算法
為了在 IEEE 802.16j 網路中能更有效地實現空間再利用及 RS 分群的好處，我們設計結
合 RS 分群及空間再利用特性之集中式下傳頻寬排程演算法。其中我們採用 throughput
optimal policy 的排程演算法，使用夏農定理(Shannon’s formula)來模擬傳輸速率的變化，透
過 selective relaying 的概念來模擬 RS 群組的合作式傳送 (cooperative transmission)，並利用
Rayleigh fading 的數學模型來模擬 RS 與 MS 之間的非直視無線傳輸(Non-Line-Of-Sight,
NLOS)。我們設計公式(1)
X t
g G t R t D gj K j g g( ; ) ( ; ) max ( ) , 1,..., } ,..., 1 { (1)
其中 D
g(t;ψ)代表群組 g 在第 t 個時槽內所需要的傳輸量比重，此比重值愈大表示該群組會
被挑選作為同時傳輸的機率愈高。最後，我們再利用公式(2)
g g t D t) arg max ; ((2)
決定出在第 t 個時槽內，在所有 activation set ψ 可能組合的總集合 Ω 中最適合被安排同時
傳輸資料的 RS 群組組合 ψ(t)。根據此演算法所得之結果，可以作出第 t 個時槽的頻寬分配。
(3) 實作群組化 IEEE 802.16j 多躍中繼網路系統之模擬環境
為了要評估上述 RS 分群演算法及頻寬排程演算法對 IEEE 802.16j 多躍中繼網路的效能
影響，我們實作出此 IEEE 802.16j 系統的模擬環境。透過模擬所得的結果，我們可以計算
系統傳輸量(system throughput)、總換手次數(the number of handovers)、封包遺失率(packet
loss rate)、封包傳送延遲(packet transmission delay)等系統效能指標。利用這些數據我們可以
分析出 RS 分群演算法及排程演算法在哪些條件下會讓系統達到最佳狀態，以驗證我們所
提出的演算法的確比原方法有更好的效率。
Simulation Results and Discussion
A. Effects of RS Group Sizes on Handoff Frequency
The following figure individually evaluates the handoff frequency of each grouping under
different user densities (specifically, 100, 500, and 1000 users in our experiments). The results
indicate that, using the greedy grouping policy of our RS grouping algorithm, the handoff
frequencies of the groupings gradually decline, regardless of the user density, as the group size
increases. The main reason is that larger group sizes generally lead to lower handoff probabilities.
It could be concluded that it is reasonable to choose larger group sizes if higher user mobility
speeds are observed.
B. Performance Analysis of Different Scheduling Policies
Based on the groupings, we estimate the throughput and delay performances of the TF, DF,
and Round-Robin centralized scheduling policies as functions of the group size. The following
figures present the average throughput and delay of the network under the situations with fixed
and mobile users, respectively. Note that the average throughput is estimated as the average
amount of packets actually received by the users within the simulation time. Moreover, the packet
delay indicates the time difference between when a packet arrives at the BS and when it is
received by the corresponding user. The figures show the average network throughput under the
three scheduling policies. For the case where users are stationary, the TF and DF policies achieve
similar throughput performance. However, for the case with user mobility, the average
throughput under the DF policy is higher than those under the other two scheduling policies. This
phenomenon implies that the mobility behavior of the MSs is a major factor affecting the
throughput performance under the TF policy, whereas the DF policy can provide relatively stable
throughput, regardless of the user-mobility effect. Specifically, since the DF policy is able to
balance the queue length of each user while the TF policy only prioritizes the user with the
longest queue length, the average buffered packet amounts for all users under the DF policy are
relatively small. Therefore, the number of dropped packets during the handoffs in DF can
reasonably be reduced, and the throughput can further be improved. The Round-Robin policy
inefficiently performs in both cases since its scheduling decisions are sequentially made, despite
the system state information. In addition, we also observe that, compared with groupings with
larger group sizes, groupings with smaller group sizes result in better throughput performance for
the case of fixed users. However, when user mobility is considered, the throughput value
increases as the group size increases. This is because, when users are stationary, the throughput
performance is mainly dominated by the spatial diversity gain. Under such circumstances, smaller
RS group sizes lead to higher throughput values. On the other hand, when the user mobility speed
is high enough, the larger packet loss rates due to more frequent handoffs of smaller group sizes
have more significant impact on the throughput performance. Consequently, larger RS group
sizes are more advantageous. The figures also show the packet delay performance for the two
scenarios without and with user mobility. We notice that groupings with larger group sizes result
in higher average packet delay. This is owing to the lower spectrum reuse of larger group sizes,
and the packets in the system may suffer from longer queueing delay. Moreover, as expected, the
DF policy can provide the lowest average delay for all group size cases, compared with the TF
and Round-Robin policies. The TF policy causes higher average delay since it considers only the
queue length states but no packet waiting time information. The Round-Robin policy, which
depends on neither the queue length nor the waiting time information, incurs the worst average
delay performance through almost all group size cases.
（五）結論
The IEEE 802.16j MR standard has been developed to provide performance enhancement to
the existing IEEE 802.16e network. However, issues such as frequent handoffs and low spectrum
utilization, which were not encountered in IEEE 802.16e, may occur in IEEE 802.16j. The RS
grouping is one optional mechanism in the IEEE 802.16j MR standard to overcome these
problems. This project has investigated the RS grouping performance enhancement in terms of
throughput and handoff frequency. An RS grouping algorithm was designed by utilizing a greedy
grouping policy: RS pairs with higher handoff rates will have higher priority to be selected. The
simulation results have shown that the handoff frequency of the considered MR network can
significantly be reduced, and suitable RS grouping patterns with determined activation set
assignments can be derived using our grouping algorithm. In addition, we have proposed the TF
and DF centralized scheduling policies to maximize the system throughput and to minimize the
average packet delay, respectively. By integrating our RS grouping algorithm and
centralized-scheduling algorithms, the simulation results have indicated that, for the case of fixed
users, groupings with smaller group sizes can result in better throughput performance. However,
when user mobility is considered, the throughput value increases as the group size increases.
Furthermore, we have also shown that the DF policy cannot only minimize the average packet
delay but can also provide fairness among different traffic-load users.
(六)成果自評
The main contribution of this project is to propose the first integrated algorithmic framework
that can be utilized to investigate the performance interaction between RS grouping and resource
scheduling for IEEE 802.16j MR networks. The results of this work have been published in the
international journal (IEEE Transactions on Vehicular Technology, 2010) [A1]. In addition, this
project also supports us to attend the conference in Korea for presenting our another work [A2].
1.完成之工作項目
設計 IEEE 802.16j 最少換手次數之 RS 分群演算法
設計 IEEE 802.16j 使用空間再利用的集中式下傳無線電頻寬排程演算法
實作 IEEE 802.16j 群組化多躍中繼網路系統之模擬環境
2. 對於學術研究、國家發展及其他應用方面之貢獻
本計畫的研究成果已發表於國際期刊[A1]
提出電腦模擬模型以研究 “合作式通訊網路之系統效能”
這個計畫的結果可以提供工業界做為參考，我們很樂意將研發的技術技轉給工業界
3.對於參與之工作人員，獲得以下訓練
在研究的過程中瞭解並學習 Cooperative Communication Networks 的相關知識和運作
方式，及其設計的內含意義，並熟識 IEEE 802.16e、IEEE 802.16j 等相關業界標準
學習 “Discrete-Event System Simulation” 之電腦模擬模型技巧
Appendix:
[A1] Shun-Ren Yang, Chien-Chi Kao, Wai-Chi Kan, and Tzung-Chin Shih, “Handoff
Minimization Through a Relay Station Grouping Algorithm With Efficient Radio-Resource
Scheduling Policies for IEEE 802.16j Multihop Relay Networks,” IEEE Transactions on
Vehicular Technology, Feb. 2010.
[A2] Shin-Hua Yang, Shun-Ren Yang, and Chien-Chi Kao, “Analyzing VoIP Capacity with
Delay Guarantee for Integrated HSPA Networks,” International Conference on Future Generation
Communication and Networking (FGCN), 2009.
（七）參考文獻
[1] Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems Amendment
2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation
in Licensed Bands, IEEE Std. 802.16e-2005, Feb. 2006.
[2] C. Castelluccia, “Extending mobile IP with adaptive individual paging: A performance
analysis,” in Proc. IEEE ISCC, 2000, pp. 113–118. 910
[3] C.-M. Weng and P.-W. Huang, “Modified group method for mobility management,” Comput.
Commun., vol. 23, no. 2, pp. 115–122, Jan. 2000.
[4] C.-Y. Hong and A.-C. Pang, “Link scheduling with QoS guarantee for wireless relay
networks,” in Proc. IEEE 28th INFOCOM, Apr. 2009, pp. 2806–2810.
[5] D. Ghosh, A. Gupta, and P. Mohapatra, “Admission control and interference-aware
scheduling in multi-hop WiMAX networks,” in Proc. IEEE MASS, Oct. 2007, pp. 1–9.
[6] D. J. A. Welsh and M. B. Powell, “An upper bound for the chromatic number of a graph and
its application to timetabling problems,” Comput. J., vol. 10, no. 1, pp. 85–86, 1967.
[7] D. Schultz, R. Pabst, and T. Irnich, “Multi-hop based radio network deployment for efficient
broadband radio coverage,” in Proc. WPMC, Yokosuka, Japan, 2003, pp. 377–381.
[8] G. Varsamopoulos and S. K. S. Gupta, “Dynamically adapting registration areas to user
mobility and call patterns for efficient location management in PCS networks,” IEEE/ACM
Trans. Netw., vol. 12, no. 5, pp. 837–850, Oct. 2004.
[9] G. Varsamopoulos and S. K. S. Gupta, “Optimal offline and online registration techniques for
location management with overlapping registration areas,” IEEE Trans. Mobile Comput., vol.
4, no. 5, pp. 474–488, Sep./Oct. 2005.
[10] G. Wan and E. Lin, “A dynamic paging scheme for wireless communication systems,” in
Proc. ACM/IEEE MobiCom, 1997, pp. 195–203.
[11] H. Viswanathan and S. Mukherjee, “Performance of cellular networks with relays and
centralized scheduling,” IEEE Trans. Wireless Commun., vol. 4, no. 5, pp. 2318–2328, Sep.
2005.
[12] H. Xie, S. Tabbane, and D. J. Goodman, “Dynamic location area management and
performance analysis,” in Proc. IEEE Veh. Technol. Conf., May 1993, pp. 536–539.
[13] Ranging Process for IEEE 802.16j, IEEE 802.16j-06/172, Nov. 2006.
[14] Part 16: Air Interface for Broadband Wireless Access Systems Amendment 1: Multiple
Relay Specification, IEEE Std. 802.16j-2009, Dec. 2009.
[15] J. Laneman, D. Tse, and G. Wornell, “Cooperative diversity in wireless networks: Efficient
protocols and outage behavior,” IEEE Trans. Inf. Theory, vol. 50, no. 12, pp. 3062–3080,
Dec. 2004.
[16] J. Nuevo, Mobility Generator for NS-22002. [Online]. Available:
http://www.math.keio.ac.jp/matumoto/emt.html
[17] J. Tang, G. Xue, and C. Chandler, “Interference-aware routing and bandwidth allocation for
QoS provisioning in multihop wireless networks,” Wireless Commun. Mobile Comput. J.,
vol. 5, no. 8, pp. 933–943, Dec. 2005.
[18] L. Kleinrock, Queueing Systems. Hoboken, NJ: Wiley, 1975.
[19] L. Kleinrock and J. Silvester, “Spatial reuse in multihop packet radio networks,” Proc. IEEE,
vol. 75, no. 1, pp. 156–167, Jan. 1987.
[20] L. Wang, Y. Ji, and F. Liu, “A novel centralized resource scheduling scheme in
OFDMA-based two-hop relay-enhanced cellular systems,” in Proc. IEEE WIMOB, Oct.
2008, pp. 113–118. 959
[21] M. K. Awad and X. Shen, “OFDMA based two-hop cooperative relay network resources
allocation,” in Proc. IEEE ICC, May 2008, pp. 4414–4418.
[22] M. Yang and P. H. J. Chong, “Time slot allocation schemes for multi964 hop TDD-CDMA
cellular system,” in Proc. IEEE WCNC, Mar. 2007, pp. 3099–3104.
[23] P. Djukic and S. Valaee, “Link scheduling for minimum delay in spatial re-use TDMA,” in
Proc. IEEE 26th INFOCOM, May 2007, pp. 28–36.
[24] R. Jain, W. Hawe, and D. Chiu, “A quantitative measure of fairness and discrimination for
resource allocation in shared computer systems,” Digital Equipment Corp., Maynard, MA,
Tech. Rep. DEC-TR-301, Sep. 1984.
[25] R. Pabst, B. H. Walke, and D. C. Schultz, “Relay-based deployment concepts for wireless
and mobile broadband radio,” IEEE Commun. Mag., vol. 42, no. 9, pp. 80–89, Sep. 2004.
[26] T.-W. Kim, T.-Y. Min, and C.-G. Kang, “Opportunistic packet scheduling algorithm for load
balancing in a multi-hop relay-enhanced cellular OFDMA-TDD system,” in Proc. 14th
APCC, Oct. 2008, pp. 1–5.
[27] T. Wu, G. Li, Y. Wang, J. Huang, X. Yu, and H. Tian, “Fairness-oriented scheduling with
equilibrium for multihop relaying networks based on OFDMA,” in Proc. IEEE 68th
VTC—Fall, Sep. 2008, pp. 1–5.
[28] W.-C. Kan and S.-R. Yang, The Supplements to “Handoff minimization through a relay
station grouping algorithm with efficient radio resource scheduling policies for IEEE
802.16j multihop relay networks,” Nat. Tsing Hua Univ., Hsinchu, Taiwan. [Online].
Available: http://www.cs nthu.edu.tw/~sryang/submission/TechHMRSGA.pdf
[29] X. Guo, W. Ma, Z. Guo, X. Shen, and Z. Hou, “Adaptive resource reuse scheduling for
multihop relay wireless network based on multicoloring,” IEEE Commun. Lett., vol. 12, no.
3, pp. 176–178, Mar. 2008.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 1
Handoff Minimization through a Relay Station
Grouping Algorithm with Efficient Radio
Resource Scheduling Policies for IEEE 802.16j
Multihop Relay Networks
Shun-Ren Yang, Member, IEEE, Chien-Chi Kao, Student Member, IEEE, Wai-Chi Kan, and
Tzung-Chin Shih
Abstract—The IEEE 802.16j standard has been devel-oped to provide performance enhancement to the existing IEEE 802.16e network by incorporating the multihop relay (MR) technology. However, frequent handoffs and low spectrum utilization issues that were not encountered in IEEE 802.16e may be incurred in IEEE 802.16j. The relay station (RS) grouping is one optional mechanism in the IEEE 802.16j MR standard to overcome these problems. The concept of RS grouping is to group neighboring RSs together to form an RS group, which can be regarded as a logical RS with larger coverage. In the paper, we investigate the RS grouping performance enhancement in terms of throughput and handoff frequency. This paper designs an RS grouping algorithm to minimize handoffs by utilizing a greedy grouping policy: RS pairs with higher handoff rates will have higher priority to be selected. The simulation results show that the handoff frequency of the considered MR network can be significantly reduced and suitable RS grouping patterns can be derived using our grouping algorithm. In addition, we propose two centralized scheduling policies, the throughput-first (TF) policy to maximize the system throughput and the
delay-Manuscript received March 19, 2009; revised August 18, 2009 and November 16, 2009. This work was supported in part by the National Science Council (NSC) of Taiwan under Contracts 96-2752-E-007-003-PAE, 96-2221-E-007-025-, 96-2221-E-007-027-, 96-2219-E-007-012- and 96-2219-E-007-011-, and Chunghwa Telecom. The review of this paper was coordinated by Prof. Vincent Wong.
S.-R. Yang is with the Department of Computer Science and Insti-tute of Communications Engineering, National Tsing Hua University, Hsinchu, Taiwan 300, R.O.C. (e-mail: sryang@cs.nthu.edu.tw).
C.-C. Kao is with the Department of Computer Science, Na-tional Tsing Hua University, Hsinchu, Taiwan 300, R.O.C. (e-mail: mickey@wmnet.cs.nthu.edu.tw).
W.-C. Kan was with the Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan 300, R.O.C. He is now with the Synology Incorporated, Taipei, Taiwan 10351, R.O.C. (e-mail: waichik@gmail.com).
T.-C. Shih is with the Institute of Communications Engineering, National Tsing Hua University, Hsinchu, Taiwan 300, R.O.C. (e-mail: stonc@wmnet.cs.nthu.edu.tw).
first (DF) policy to minimize the average packet delay. By integrating our RS grouping algorithm and centralized scheduling algorithms, the simulation results indicate that for the case of fixed users, the groupings with smaller group sizes can result in better throughput performance. However, when user mobility is considered, the throughput value increases as the group size increases. Furthermore, we also show that the DF policy can both minimize the average packet delay, and provide the fairness property among users with different traffic loads.
Index Terms—Grouping algorithm, IEEE 802.16j, mul-tihop relay, scheduling policy, WiMAX.
I. INTRODUCTION
Incorporating the multihop relay (MR) technology [25], the IEEE 802.16j MR standard [14] has been developed to provide throughput improvement, cover-age extension and capacity enhancement to the existing IEEE 802.16e protocol [1]. By deploying relay stations (RSs), the end-to-end communication quality between base stations (BSs) and mobile stations (MSs) can be im-proved without high infrastructure deployment costs. In particular, it becomes possible to forward data to an MS using a high transmission rate in Line-Of-Sight (LOS) conditions through an MR path to avoid the Non-Line-Of-Sight (NLOS) direct (i.e., single-hop) transmission with bad channel quality. In addition, spatial reuse [19] is another promising approach that can be employed in IEEE 802.16j MR networks to improve spectral efficiency. Based on the centralized scheduling, spatial diversity gain can be achieved if multiple simultaneous transmissions using the same bandwidth resources are realized within the same BS-cell.
Although IEEE 802.16j has the potential to provide substantial performance enhancements, several issues
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 2
that were not addressed in IEEE 802.16e may be en-countered. For example, frequent handoffs may occur during the movements of MSs since RS-cells are smaller than BS-cells. Moreover, the unbalanced resource allo-cation of different RSs may result in inefficient spectrum utilization. To avoid the consequent performance degra-dation, RS grouping has been identified as an optional mechanism in the IEEE 802.16j standard. The main idea of RS grouping is to put adjacent RSs together to form an RS group, wherein the RS members are required to receive and transmit the same data simultaneously. From MSs’ viewpoint, the RS group can be regarded as a logical RS with larger coverage. Therefore, the handoff frequency can be reduced since no handoff procedure would be triggered even when an RS crossing event within the RS group occurs. In addition, since the radio resources of the RS members are aggregated together, the resources can then be allocated more flexibly and effi-ciently to achieve higher spectrum utilization. However, to the best of our knowledge, no RS grouping strategy has been proposed or discussed in the literature. We argue that different grouping criteria may lead to various performance results. Specifically, utilizing a smaller RS group size is advantageous to spatial reuse because more RS groups can perform simultaneous transmissions at the same time and frequency. Thus, improved average sys-tem throughput can be expected. However, a smaller RS group size may also lead to higher packet loss rate due to more frequent handoffs between RS groups. In conclu-sion, when implementing the RS grouping mechanism in IEEE 802.16j MR networks, the performance tradeoff between throughput and handoff frequency should be seriously considered.
In this paper, we analyze grouping strategies for RS grouping-enabled IEEE 802.16j MR networks and propose an efficient RS grouping algorithm to minimize the handoff frequency. As we have pointed out, RS grouping strategies will also influence the throughput performance. To investigate the impacts of RS grouping on the IEEE 802.16j system throughput, we design two centralized downlink scheduling policies for RS grouping-enabled IEEE 802.16j MR networks. One of these two scheduling policies aims to maximize the system throughput and the other is to minimize average packet delay. The throughput estimation results under different grouping configurations can assist network ser-vice providers to choose the most appropriate settings of grouping factors (e.g., group size). Notice that the spatial reuse concept is considered during the scheduling
MR-BS RS 1 RS 2 RS 3 Relay link Access link MS 1 MS 2 MS 3 MS 4
Fig. 1. Sample topology of IEEE 802.16j MR networks
procedure to improve the spectral efficiency. By integrat-ing our proposed RS groupintegrat-ing algorithm and centralized scheduling policies, the simulation results show that the throughput and delay performance can be improved in addition to the significantly reduced handoff frequency. The main contribution of this paper is to propose the first integrated algorithmic framework that can be utilized to investigate the performance interaction between RS grouping and resource scheduling for IEEE 802.16j MR networks.
The remaining parts of this paper are organized as follows. In sections II and III, we give an overview of the IEEE 802.16j standard and a discussion of the IEEE 802.16j RS grouping mechanism, respectively. Section IV presents an RS grouping strategy analysis and details our proposed RS grouping algorithm. In section V, two efficient centralized downlink scheduling policies for IEEE 802.16j MR networks are addressed. The performance evaluations of our grouping algorithm and scheduling policies are presented in section VI. Section VII summarizes the related work. Finally, we conclude the paper in section VIII.
II. IEEE 802.16JMULTIHOPRELAYNETWORKS
A. Network architecture
The IEEE 802.16j standard is expected to enhance the system performance of IEEE 802.16-based networks through multihop relaying technologies. A typical topol-ogy example of IEEE 802.16j MR networks is illustrated in Fig. 1. In this network, an MS can access the MR-BS either through a multihop relaying path (e.g., MS1, MS3 and MS4) or directly (e.g., MS2). In addition, a station (BS or RS) is called an access station if it provides network attachment functionality to a given MS or RS. On the other hand, an RS is a subordinate RS of another station if that station serves as the access station for that RS. For instance, RS2 is the access station of RS3, and RS3 is a subordinate RS of RS2. The wireless links that directly connect access stations with their respective subordinate RSs are called relay links, while the links
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 3
between MSs and their corresponding access stations are known as access links. Since an RS can only be subordinated to one station, the MR-BS and the RSs in this MR network form a tree-based multihop relay topology. Note that it has been shown that the throughput decreases as the number of hop-counts increases [7], and thus we only investigate two-hop IEEE 802.16j networks in this paper.
B. Frame structure
The frame structure of IEEE 802.16j MR systems is extended from that of IEEE 802.16e networks, which also adopt orthogonal frequency division multiple access (OFDMA) as the primary channel access mechanism for NLOS communication. The basic unit of resource for allocation in OFDMA is a slot, which is comprised of a number of symbols in the time domain, and one sub-channel in the frequency domain. The timeline is divided into contiguous frames, each of which further consists of a downlink (DL) and a uplink (UL) subframes. In IEEE 802.16j, the DL and UL subframes shall include one access zone for MR-BS↔ RS and MR-BS ↔ MS
transmissions and may include one relay zone for RS↔
subordinate MS transmissions, respectively.
III. IEEE 802.16JRS GROUPINGMECHANISM
Although deploying RSs in IEEE 802.16 networks can provide significant throughput or coverage enhance-ments, several issues regarding the relaying architecture of IEEE 802.16j should be addressed. These issues in-clude frequent handoffs, redundant control overhead and low spectral efficiency. It is perceived that these issues will result in unpredictable performance reduction for IEEE 802.16j MR networks. Therefore, the IEEE 802.16j standard provides the optional RS grouping mechanism to reduce the impacts of these issues. The concept of an RS grouping is that adjacent RSs could be grouped together as an RS group which acts as a virtually regular RS to its associated MSs. The grouping criteria are decided by the controlling MR-BS, based on the targeted performance requirement. Note that the coverage of an RS group is larger than that of its regular member RSs, and no handoff event would be triggered even though an RS-cell crossing event within the same RS group occurs. Consequently, the MSs under the RS grouping mechanism will experience lower handoff probability. On the other hand, MR-BSs can manage RS groups using only one set of control header, and hence the control signal overheads are reduced. Finally, the radio
P re am b le F C H D L -M A P Access zone (Relay link transmission)
Relay zone (Access link transmission)
DL bursts [BS=>RS Group]
DL bursts [RS Group (one or more) =>MS]
DL Subframe
Fig. 2. Modified IEEE 802.16j downlink subframe structure for supporting RS grouping
resource (i.e., the relay zone) of each member RS can be aggregated and shared by all the MSs under the corresponding RS group, so that the spectral efficiency is improved.
For the downlink operation, the member RSs of a group should be configured to transmit equivalent data signals to the same MS. Thus the subordinated MSs can receive the best quality signal within the group, no matter where they are located. This operation is also called cooperative transmission because the member RSs will form a virtual antenna array to exploit macro-diversity. On the other hand, the diversity combining of the information received by the member RSs of an RS group can be performed in the uplink situation. Both the downlink and uplink diversity gains can be achieved under a centralized scheduling scheme by keeping the MS list of each RS group at the respective MR-BS. Therefore, the RS grouping mechanism is reasonable to improve the data transmission rate.
To support an RS grouping, the original IEEE 802.16j frame structure should be modified (see Fig. 2 for the modified DL subframe). Specifically, access zones should handle the transmissions between BSs and RS groups, while relay zones should handle the transmis-sions between RS groups and subordinate MSs. From the mobility management point of view, we note that implementing RS grouping will not incur extra costs to IEEE 802.16j. As aforementioned, for movement between RS-cells of an RS group, an MS will not initiate the handoff procedure. The MS CDMA periodic ranging process with aggregated ranging sub-channel allocation [13] can be employed to handle the RS reselection during intra-RS-group movement. On the other hand, when the MS roams from an RS group to another RS group, since an RS group can be seen as a legacy BS, the conventional MAC layer handoff procedure is applied directly for this scenario.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 4
IV. OURPROPOSEDRS GROUPINGALGORITHM FOR
IEEE 802.16JMR NETWORKS
This section proposes an IEEE 802.16j RS grouping algorithm to reduce the handoff frequency of mobile users under prescribed IEEE 802.16j MR network per-formance requirements. The system assumptions and the concept of our proposed algorithm are described in the first subsection. Then, the proposed algorithm is discussed as a three-phase procedure in the following subsections.
A. System Assumptions and the Concept of Our RS Grouping Algorithm
To accommodate general scenarios, our algorithm does not make any assumptions of the underlying IEEE 802.16j MR network topology, the user mobility behav-ior, and/or the packet traffic pattern. Specifically, within a considered BS, the IEEE 802.16j RSs can be deployed arbitrarily and the coverage area of each RS can be irreg-ular. In addition, the MSs within the considered BS can move randomly. In such an arbitrary environment, we only require that the handoff-rate information between each two RS-cells should be available. The handoff rate between two RS-cells, RS-cell 1 and RS-cell 2, is the total rate that the resident MSs hand off from RS-cell 1 to RS-cell 2 or from RS-cell 2 to RS-cell 1. Note that the handoff-rate information can be simply derived from the statistical data that are collected by the network service providers.
To design the RS grouping algorithm in our considered environment, we first specify the factors that may affect the grouping result. First, the group size is a factor which has the potential to influence the spatial diversity gain and the handoff frequency. It can be observed that utilizing a smaller group size will result in more RS groups. Such a grouping policy is beneficial to spatial reuse, though it causes higher handoff frequency. On the other hand, a larger group size has reverse effects on the spatial reuse and the handoff frequency, respectively. In addition to group size, the selection order of group members is also significant to the grouping strategy de-sign. Even with the same group size, different grouping orders may lead to different numbers of handoff events. Once an RS grouping layout is provided by applying the most desired group size and grouping order, determining which set of RS groups to transmit data simultaneously further impacts the spectrum efficiency critically. In this paper, we define an activation set as a particular set of RS groups which can transmit data to their respective
Grouping Activation Set Assignment
Scheduling Simulation
1. Handoff-rate information and a preferred group size
2. Group adjacency matrix for the preferred group size
3. Grouping result with determined activation set assignment
4. The respective performance gain
1
2 3
4
Fig. 3. The concept of the proposed RS grouping algorithm
resident MSs at the same time and frequency without interference. Clearly, the assignment of activation sets is another important factor that should be considered in the design of the RS grouping algorithm.
Taking all the above factors into account, we propose our RS grouping algorithm, which contains three phases as illustrated in Fig. 3: grouping phase,
activation-set-assignment phase and scheduling-simulation phase.
Based on the handoff-rate information, the grouping phase first constructs the handoff-minimizing RS ing result for a preferred group size. Then, the group-ing phase generates the adjacency matrix of the con-structed RS groups. Based on the adjacency matrix, the activation-set-assignment phase assigns each RS group to an activation set in order to enhance the spatial diversity gain. Afterward, in the scheduling-simulation phase, the downlink transmission simulation is executed for the given grouping result with determined activation set assignment. By comparing the performance gains of the scheduling results from different preferred group sizes, network service providers can finally choose the most favorable grouping result with the associated acti-vation set assignment as their IEEE 802.16j MR network configuration.
The operation of our RS grouping algorithm is de-picted in Algorithm 1, where two input parameters, the set α of all the RSs within a considered BS and the
handoff-rate matrixH of the RSs in α, should first be
provided. In H, the entry H[i][j] represents the total
handoff rate between RS-cell i and RS-cell j. After
variable initialization, the three algorithm phases at lines 14-16 are repeated for each preferred group size. Then, at line 17, the weighted function ω(Si) calculates the
weighted performance gain for a simulation result Si.
The highest performance gain is recorded in key at
line 18 and the best grouping result with determined activation set assignment is stored in Ω at line 19.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 5
Algorithm 1 THE PROPOSED RS GROUPING ALGO
-RITHM
1:Input:
2:α (the set of RSs within a considered BS), H (the handoff-rate matrix of the RSs in α). 3:
4:Output:
5:Ω (the desired grouping result with determined activation set assignment). 6: 7:Initialization: 8:Ω ← N IL; 9:key ← 0; 10:M ← rows[H]; 11: 12:Procedure: 13:fori = 1 to M do 14: Γi, Ai← Grouping(α, H, i); 15: Ωi← ActivationSet(Ai); 16: Si← Scheduling(Γi, Ωi); 17: ifω(Si) > key then 18: key ← ω(Si); 19: Ω ← Ωi; 20: end if 21:end for
When Algorithm 1 terminates, the desired grouping result can thus be obtained. In the following subsections, we present the details of each phase individually.
B. Grouping Phase
The operation of the grouping phase is shown in Algorithm 2 and is explained as follows. According to the functionalities, the RS grouping procedure is partitioned into four main portions:
Input parameters: At the beginning of the procedure
(line 2), three parameters,α, H and X, are required
as the inputs, whereX denotes the preferred group
size.
Output results: The outputs of the procedure (line 5)
are the final grouping resultΓ = {g1, g2, ..., gn, ...}
and the adjacency matrix A of the constructed RS
groups, wheregn denotes thenth RS group.
Initialization stage: In this stage (lines 8 to 10), some
variables must be initialized before executing the grouping procedure. We use T empαto denote the
set of the ungrouped RSs within the considered BS.
T empα is initialized as α. Then, we use M to
denote the total number of RSs in α. Moreover,
the variablen is initialized as 0.
Procedure: After initialization, the main grouping
pro-cedure is started from line 13 to line 47. Since the purpose of our algorithm is to minimize the handoff probability, we introduce a greedy grouping policy, under which RS pairs with higher handoff rates (provided from H) will have higher priority to be
Algorithm 2 Grouping(α, H, X) 1:Input:
2:α (the set of RSs within a considered BS), H (the handoff-rate matrix of the RSs in α), X (the preferred
group size). 3: 4:Output:
5:Γ (a set of constructed RS groups {g1, g2, ..., gn, ...}), A (the adjacency matrix of the constructed RS groups inΓ). 6: 7:Initialization: 8:T empα← α; 9:M ← rows[H]; 10:n ← 0; 11: 12:Procedure:
13:/* Each RS group is constructed based on the handoff-rate matrixH. */
14:while there exist pairs of ungrouped adjacent RSs inT empαdo
15: n ← n + 1;
16: Find the ungrouped adjacent RS pair(a, b) that has the highest handoff rate H[a][b];
17: ifa.T otHR(T empα) ≥ b.T otHR(T empα) then
18: Removea from T empαand add it into thenth group gn;
19: else
20: Removeb from T empαand add it into thenth group gn;
21: end if
22: while|gn| < X and group gnhas ungrouped neighboring RSs do
23: Find the neighboring RSc with the highest total handoff rate to/from group gn;
24: Removec from T empαand add it intogn;
25: end while
26: AddgnintoΓ;
27:end while
28:/* An isolated RS forms an RS group itself. */
29:while there exists an isolated ungrouped RSd in T empαdo
30: n ← n + 1;
31: Removed from T empαand add it into thenth group gn;
32: AddgnintoΓ;
33:end while
34:/* Initialize the RS-group adjacency matrixA. */
35:fori = 1 to n do
36: forj = 1 to n do
37: A[i][j] ← 0;
38: end for
39:end for
40:/* Compute the RS-group adjacency matrixA. */
41:fork = 1 to M do
42: forl = 1 to M do
43: if the handoff rateH[k][l] > 0, RS k ∈ giand RSl ∈ gjthen
44: A[i][j] ← 1;
45: end if
46: end for
47:end for
selected. A detailed description of the steps is as follows.
• Group-construction loop for non-single-RS
groups (lines 14 to 27): This loop constructs RS groups from the ungrouped adjacent RS pairs. The ungrouped adjacent RS pair with the highest handoff rate is first considered (line 16). Of this pair, the RS that has a higher total handoff rate to/from other RSs in T empα is
selected to be the starting member of the nth
groupgn(lines 17 to 21). Then, the ungrouped
neighboring RSs of gn are selected to join gn
in descending order of the total handoff rate to/from gn. This join procedure (lines 22 to
25) is repeated until the group size of gn is
equal to the preferred group size X or gn
has no ungrouped neighboring RS. This group-construction loop is iterated to form RS groups
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 6
until no pair of ungrouped adjacent RSs exists.
• Group-construction loop for isolated RSs (lines
29 to 33): This loop constructs RS groups for isolated ungrouped RSs. Each isolated un-grouped RS forms an RS group itself (line 31).
• Group adjacency matrix initialization (lines 35
to 39): Each entry of the group adjacency matrixA is initialized as 0.
• Group adjacency matrix computation (lines 41
to 47): After initialization, the adjacency matrix
A is computed by examining whether a member
of a groupgiis adjacent to a member of another
group gj (line 43). If yes, the entryA[i][j] is
set to 1 (line 44). After examining all the M
RSs, the adjacency matrixA of the RS groups
inΓ can be derived.
After the above grouping procedure, the group-ing result Γ and the group adjacency matrix A can
be obtained for use in the subsequent activation-set-assignment phase. We note that the time complexity of the grouping phase is mainly dominated by the two nested loops in lines 14-27 and in lines 41-47. For the nested loop in lines 14-27, the numbers of iterations for both the outer and inner loops are no larger than|α|. On
the other hand, for the nested loop in lines 41-47, the numbers of iterations for the outer and inner loops are both equal to|α|. Therefore, the overall time complexity
of the grouping phase isO(|α|2).
C. Activation-Set-Assignment Phase
Before elaborating on how our activation-set-assignment phase can generate appropriate activation-set-assignment results, we first point out that the number of ac-tivation sets significantly affects the data transmis-sion performance. Specifically, the fewer the activa-tion sets, the more the RS groups that can trans-mit data simultaneously. Consider the RS-group lay-out in Fig. 4(a), where its RS-group adjacency graph is shown in Fig. 4(b). Clearly, the assignment
{{g1, g3, g5}, {g2, g4, g6}} is better than another
assign-ment {{g1, g4}, {g2, g5}, {g3, g6}} since the former
as-signment contains fewer activation sets and thus has improved spatial diversity gain.
Given an RS grouping resultΓ and the corresponding
adjacency matrixA from Algorithm 2, we first derive the
adjacency graphG(V, E) where V represents the set of
the RS groups andE represents the interference relation
among the RS groups. To minimize the number of activation sets, we model the activation-set-assignment
BS RS group 2 RS group 1 RS group 6 RS group 3 RS group 4 RS group 5
(a) The RS-group layout (b) The RS-group adjacency graph
g3 g4 g5 g6 g1 g2
Fig. 4. An activation-set-assignment example
problem as the minimum coloring problem of G. The
rule to color all the vertices ofG is that no two adjacent
vertices can share the same color. Based on this rule, each set of the vertices that are assigned the same color is equivalent to an activation set. However, to color G
with the minimum number of colors is known as an NP-hard problem. To achieve a polynomial computation time, the well-known greedy coloring algorithm, Welsh Powell algorithm [6], is adopted in this paper. For the self-containedness purpose, we briefly summarize the key concept of the Welsh Powell algorithm below. First, all possible colors are numbered. Then following the descending order of vertex degree, each vertex in G
is sequentially selected to be assigned a color. When a vertex v attempts to be colored, the first color is
examined to check if it has already been occupied by any of v’s neighbors. If no, the first color is assigned
to v. Otherwise, the next unoccupied color (by v’s
neighbors) is used to colorv. This procedure is repeated
for the subsequent vertices until the vertices in G have
all been processed. The time complexity of the Welsh Powell algorithm is proven to be O(|V |2). This
algo-rithm guarantees that the number of required colors is at most one more than the maximum degree ∆(G) of G.
That is, the Welsh Powell algorithm determines at most
∆(G) + 1 activation sets. Our activation-set-assignment
phase applies the Welsh Powell algorithm and therefore can assign each constructed RS group (from the grouping phase) to an appropriate activation set.
D. Scheduling-Simulation Phase
In this subsection, we characterize the concept of the scheduling simulation. By adopting a specific scheduling policy, scheduling simulations are conducted for the different grouping results with associated activation set assignments provided from the grouping and
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 7
set-assignment phases. From these simulations, several output measures (e.g., system throughput and packet de-lay time) can be derived. The network service providers can evaluate the performance gains of these grouping results by assigning weighted values to the respective output measures following their preference rules. Then, by comparing the weighted indices from the different grouping results, the most desirable grouping result can be determined. In the following sections, the scheduling policies for RS grouping-enabled IEEE 802.16j MR networks will be elaborated on in section V and the representative simulation results will be presented in section VI.
V. MULTIHOPCENTRALIZEDDOWNLINK
SCHEDULINGPOLICIES
In the scheduling-simulation phase of our RS grouping algorithm described in section IV, centralized downlink scheduling policies are required to evaluate the per-formance gain of a given RS grouping layout. In this section, we first define the scheduling problem for RS grouping-enabled IEEE 802.16j MR networks. Then the system description of our considered multihop network is addressed. Finally, we propose two centralized down-link scheduling policies for IEEE 802.16j MR networks under RS grouping and spatial reuse assumptions, with the objectives of maximizing the system throughput and minimizing the downlink traffic delay, respectively.
A. Scheduling problem for RS grouping-enabled IEEE 802.16j MR networks
Consider the downlink transmission of an IEEE 802.16j MR network with a BS and a set of pre-configured RS groups, as shown in Fig. 5. Assume that a number of packets from the external networks are desired to be delivered to MS 1 residing in RS group
g. The BS, which acts as the network gateway for its
served MSs, will first buffer these packets for MS 1 in the corresponding packet queue. In an appropriate time frame, the BS will transmit these packets to RS group
g through the relay link transmission. These packets
received by RS groupg are also buffered in the packet
queue corresponding to MS 1 until the access link transmission between RS groupg and MS 1 is scheduled
in another following frame.
To realize this two-hop relaying operation, the scheduling procedure should be performed by the BS at the beginning of each frame. Specifically, the scheduling procedure consists of two parts: relay link scheduling
MR-BS RS group g MS 1 (Target) MS 2 MS K X1(t)
Relay link (cooperative link)Access link BS X2(t) BS XKBS(t) X1g(t) X2(t) g XK(t) g Ȝ1(t) BS Ȝ2(t) BS ȜK(t) BS The MSs residing in group g Ȝ1(t) Ȝ2(t) ȜK(t) g g g The MS residing in another group Queue state information
Fig. 5. System model of RS grouping-enabled IEEE 802.16j MR networks
and access link scheduling. In the relay link scheduling, the BS selects an RS group. The relay link between the BS and this RS group will be activated during the access zone for transmitting packets destined to the MSs residing in this RS group. These packets will be queued in the RS group for future relaying. On the other hand, in the access link scheduling, the BS selects an appropriate activation set of RS groups. The access links between these RS groups and their respective resident MSs will be activated during the relay zone for packet relaying. Note that since the BS has only one radio transmitter, only one relay link between the BS and a particular RS group can be activated at a time in the access zone of each time frame. However, based on the concept of spatial reuse, more than one access link transmissions can be activated simultaneously in the relay zone.
To resolve the scheduling problem for RS grouping-enabled IEEE 802.16j MR networks, scheduling policies should be employed to decide suitable relay link trans-missions in access zones and access link transtrans-missions in relay zones. Different scheduling policies may be designed based on different criteria to meet different system performance requirements.
B. Radio Resource Scheduling Policies
1) System model: Consider again the single-BS IEEE
802.16j MR network in Fig. 5. Using our RS grouping algorithm, the RSs in this network are partitioned into a setΓ of RS groups, and the RS groups are classified into
a set Ω of activation sets. The RS groups within each
activation set can be activated simultaneously.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. X, NO. X, XXXX 201X 8
For both the relay link and access link transmissions in the considered IEEE 802.16j MR network, the Rayleigh fading channel model is assumed to be applied in the physical layer. Letting src(l) be the transmitter and dst(l) the receiver of a radio link l, we can compute
the transmission rate over the radio linkl as Rl= B · log2(1 + |h
src(l) dst(l)|
2· P W
src(l)/σ2N). (1)
In (1), B denotes the bandwidth (in Hz) of the MR
network spectrum, and P Wsrc(l) andσ2N represent the
transmission power of src(l) and the variance of signal
noise, respectively. Moreover,hsrc(l)dst(l)in (1) is a circularly symmetric complex Gaussian random variable. Readers interested in the derivation of (1) are referred to [28] for a detailed explanation.
During the relay link transmission, the BS adopts the minimum transmission rate among the relay links between the BS and the selected RS group as the real transmission rate. This is due to the fact that the RSs within the selected RS group must correctly receive and decode the data from the BS at the same time and frequency. In this case, the effective transmission rate is dominated by the relay link with the minimum rate. Thus the transmission rate of the relay link between the BS and an RS groupg can be computed based on (1) as
RBS,g= B · log2(1 + min ∀r∈g|h BS r | 2· P W BS/σN2). (2)
For the access link transmission, although the RS members in the same RS group should serve a par-ticular MS at the same time, this signal-combining cooperative transmission with multiple sources is diffi-cult to implement. Therefore, we adopt a compromise cooperation scheme called selection relaying [15] to realize the access link transmission, whereby only the RS member with the highest signal quality is selected for data transmission towards a resident MS. Therefore, the transmission rate between the RS groupg and a resident
MSj can be derived similar to (2) as Rg,j= B · log2(1 + max ∀r∈g|h r j| 2· P W r/σN2). (3)
In the two proposed scheduling policies that will be described later, the network queue state is a common parameter utilized to make the scheduling decision. We denoteXBS
j (t) as the queue length of the queue
corre-sponding to MS j in the BS at the start of time frame t. In the RS group g of MS j, every RS member will
also maintain a queue for MSj. However, as mentioned
above, all the RS members serve MSj at the same time.
Thus from MS j’s viewpoint, the queues maintained in
the RSs for MS j can be regarded as a logical single
queue. We denote Xjg(t) as the length of the logical
queue for MSj in RS group g at the start of time frame t.
Note that the RS buffer synchronization for selection relaying can be easily achieved by utilizing the standard IEEE 802.16j ARQ mechanism and by arranging a new multicast information element in the DL-MAP. More specifically, when the BS receives an ACK from an MS indicating the sequence number of the latest correctly decoded packet, it will, in the next time frame, announce an additional DL-MAP multicast information element containing this sequence number to the corresponding RS group. The obsolete packets at each individual RS buffer can then be removed accordingly.
2) General scheduling procedure of the RS grouping-enabled IEEE 802.16j MR network: Recall that the
scheduling procedure should make the decision of the relay link and access link transmissions of a frame in the relay link scheduling and access link scheduling, respectively. Although the selection criteria of these two parts are actually the same under the given schedul-ing policy, some detailed processes are different due to the special characteristics of RS grouping such as minimum transmission rate constraints and selection relaying cooperative transmissions. Therefore, we detail the procedures of these two scheduling parts individually as follows.
Part 1: Relay link scheduling. In this part, the ob-jective of the scheduler is to activate the most favorable relay link between the BS and an RS group inΓ. For comparison, we claim that the major
task of a scheduling policy is to assign a weighted index DRL
g (t) for each RS group g to represent
the priority of RS group g to be selected in the
relay link scheduling of time frame t. Note that
the implementation ofDRL
g (t) is dependent on the
concerned performance metrics (e.g., throughput or delay time) of the scheduling policy. In order to make the scheduling decision in a channel-aware manner, the scheduling policy may require the transmission rateRBS,gto decide theDRLg (t) of RS
group g. Two different approaches are given later
in this section. After both the transmission rate and the weighted index computations of each RS group, the final target RS groupg(t) for time frame t isˆ
