3320 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO. 11, NOVEMBER 2010
Evaluation of Scan and Association Process for
Real-Time Communication in Mobile WiMAX
Shiao-Li Tsao, You-Lin Chen, and Chia-Hsiang Chang
Abstract—One of the most important issues in offering real-time communication services in a mobile environment is sup-port for seamless handover between base stations during a communication session. In Mobile WiMAX, a mobile station may have to perform a scan and association process before handover, but unfortunately, this scan and association process introduces service disruption. In this letter, we investigate the scan and association latency in Mobile WiMAX, and evaluate its performance through analytic models and simulations.
Index Terms—Mobile WiMAX, IEEE 802.16e, handover, scan and association.
I. INTRODUCTION
T
HE development of WiMAX networks based on IEEE802.16 has grown rapidly over the past several years. Furthermore, Mobile WiMAX based on IEEE 802.16e sup-ports mobility, thus facilitating broadband communications in a mobile environment. A critical issue pertaining to Mobile WiMAX offering voice and video services is the provision of seamless handover during a real-time communication session. According to the IEEE 802.16e specification [1], handover consists of four phases of processes: network topology acqui-sition, scan and association process, handover procedures, and network re-entry procedures. During these processes, a mobile station (MS) may have to leave its serving base station (BS) and switch to other channels to perform certain operations. Therefore, the communication between the MS and the serving BS is temporarily suspended, and service disruption may occur. Such a service disruption influences the quality of service (QoS) of real-time communication in Mobile WiMAX. Many studies have investigated and proposed solutions to speed up the procedures of handover in Mobile WiMAX [2]. Most of these studies assume that the scan and association process can be done in the background and before handover without influencing the current communication. Although, with full network assistant, the scan and association process can be quickly finished, it requires support from all BSs with frame-level time synchronization. On the other hand, the scan and association process with contention-based ranging is mandated by the WiMAX Forum and should be supported by all BSs and MSs. Unfortunately, the contention-based ranging may introduce long service disruptions and packet delays if the scan and association parameters are not properly configured [3]. Rouil and Golmie [3] thus proposed a mechanism to schedule the scan and association process and maintain the QoS of a communication session. However, the loss of ranging messages was not evaluated in their study. In this letter, we Manuscript received April 30, 2009; revised October 3, 2009 and May 10, 2010; accepted August 14, 2010. The associate editor coordinating the review of this letter and approving it for publication was D. Zeghlache.
The authors are with the Department of Computer Science, Na-tional Chiao Tung University, Hsin-Chu, 30010, Taiwan (e-mail: {sltsao, youlin}@cs.nctu.edu.tw; nelsonchang1218@gmail.com).
Digital Object Identifier 10.1109/TWC.2010.091510.090619
Target BS MS Serving BS Other BSs Da ta tr af fic MOB_SCN-RE Q /R SP CDMA c o de RNG-R SP Da ta tr af fic CDMA c o de Da ta tr af fic MOB_MSHO -RE Q /R SP MOB_HO-IND
Trigger scan Trigger HO
Da
ta
tr
af
fic
Scan and association duration (T’a) Scanning interval (Fscn)
Interleaving interval (Fitvl)
CDMA c o de RNG-R SP X CDMA c o de RNG-R SP X in serving channel in neighboring channels (service disruption) HO procedures
Fig. 1. Scan and association process.
consider the scan and association process with contention-based ranging. The service disruption owing to the scan and association process is modeled and the performance is evaluated.
II. SCAN ANDASSOCIATIONPROCESS
An MS can perform association with neighboring BSs during scan to reduce the latency introduced by initial ranging before the actual handover. Fig. 1 illustrates an example of the scan and association process. An MS which performs the scan and association has to switch to neighboring channels for rang-ing procedures durrang-ing scannrang-ing intervals. The communication between the MS and the serving BS is temporarily suspended during these periods and service disruption may occur. To min-imize the service disruption, the MS may return to the serving channel periodically and continue packet exchanges with the serving BS during interleaving intervals. Unfortunately, the neighboring BS is not aware whether the MS is in a scanning interval or not, and it may respond to the ranging request when the MS stays in the serving channel, i.e., during interval intervals. Thus, the ranging response is missing, and the MS has to perform the ranging with the neighboring BS again. Obviously, the scan and association process introduces service disruption during a communication session, and the lengths of the scanning and interleaving intervals also influence the ranging process and the overall scan and association duration. Therefore, a scan and association process that introduces less service disruptions of a real-time communication and can minimize the duration of the process is important.
III. SCAN ANDASSOCIATIONSTRATEGIES
A. Scan and Association Without Interleaving (SA)
The first scan and association strategy is called Scan and Association without Interleaving (SA). An MS requests a long scanning interval, temporarily stops exchanging packets with the serving BS, and switches to neighboring channels. The 1536-1276/10$25.00 c⃝ 2010 IEEE
TSAO et al.: EVALUATION OF SCAN AND ASSOCIATION PROCESS FOR REAL-TIME COMMUNICATION IN MOBILE WIMAX 3321
ranging process start
Select a backoff counter from contention window
Listen to initial ranging slots and count down the
backoff counter backoff counter
is not zero
Transmit the CDMA ranging code and wait for
ranging response backoff counter
is zero
Adjust the transmission power and double the contention window if it is less than the maximal contention window size
(Wmax)
Ranging timeout (T3)
receive ranging response with status ‘success’
ranging process end
Exceed maximal retry (L)? No
Fig. 2. Flowchart of a contention-based ranging.
MS does not switch back to the serving channel until the scan is completed; in other words, there is no interleaving interval during the scan and association operations. This approach is simple and straightforward, and can minimize the scan and association duration since no interleaving interval is involved. However, this approach introduces a long service disruption.
Fig. 2 illustrates the procedures for a contention-based rang-ing based on the binary exponential back-off (BEB) algorithm in IEEE 802.16e. Kwak et al. [4] analyzed the performance of the exponential back-off protocol, and Peyre et al. [5] also evaluated the contention resolution scheme in a WiMAX network. The BEB algorithm, which is used in bandwidth request and initial ranging in WiMAX networks, was studied in [6] and [7]. In this letter, we model the service disruption and the duration of the scan and association process based on these previous studies.
The parameters advertised by the serving BS for applying scan include the window size of the ranging back-off start
𝐵𝑠, the window size of the ranging back-off end𝐵𝑒, and the ranging retry limit𝐿 for the neighboring BS. These parameters
indicate the minimum and maximal contention window sizes, i.e.,𝑊0 = 2𝐵𝑠 and𝑊𝑚𝑎𝑥 = 2𝐵𝑒, and the maximal retry of the ranging request. The initial ranging procedure is complete when the MS successfully receives the ranging response from the neighboring BS. If the MS cannot receive the ranging response within T3 ms, which is defined in IEEE 802.16e, the ranging fails and the MS doubles its contention window size to perform the next ranging. Therefore, the contention window size of the𝑖𝑡ℎ ranging request is denoted as:
𝑊𝑖= {
2𝑖× 𝑊
0, 𝑖 ≤ 𝑚
2𝑚× 𝑊0, 𝑚 < 𝑖 ≤ 𝐿 , 𝑚 = 𝐵𝑒− 𝐵𝑠 (1) In this letter, we assume that each frame has an average of𝑆 available slots for initial ranging and the MS picks up
the back-off counter from the contention window based on a uniform random process. The average delay for waiting back-off counter at the𝑖𝑡ℎ transmission, say 𝑑𝑖, is
𝑑𝑖= ⌈ 𝑊𝑖− 1 2𝑆 ⌉ × 𝑇𝑓, 0 ≤ 𝑖 ≤ 𝐿 (2)
where𝑇𝑓 denotes the frame duration in milliseconds.
Moreover, we assume that the loss of ranging requests comes only from collisions, not channel error. Then, the
probability of ranging failure, 𝑃𝑐, can be derived from the
number of MSs performing ranging, the size of the contention window, the ranging retry limit, and number of initial ranging CDMA codes, based on the studies in [4] and [5]. Also, we assume that the response delay is a uniform distribution and the neighboring BS responds to the MS in𝑇𝑤ms if it receives the ranging request. This implies that the MS may receive a
ranging response after one frame (𝑇𝑓 ms) to𝑇𝑤ms. However,
since the MS does not know whether the BS receives the ranging or not, the MS has to wait for the responses for T3 ms, according to the specification. Therefore, the average time of a successful ranging, say 𝑇𝑎, for performing the scan and association with a neighboring BS can be calculated as:
𝑇𝑎= 𝐿−1 ∑ 𝑘=0 𝑃𝑘 𝑐 ⋅(1−𝑃𝑐)⋅(𝐷𝑘+𝑘 ⋅T3+𝑇2𝑤), 𝐷𝑘= 𝑘 ∑ 𝑖=0 𝑑𝑖 (3) The above equation only models the scan and association time for one neighboring BS. The equation can be also applied to multiple neighboring BSs, which may have different
ranging parameters. Then, the summation of𝑇𝑎 for different
neighboring BSs becomes the average scan and association duration. The total average scan and association duration can be also considered as the service disruption time for the SA approach, because the communication session is temporarily paused during the scan and association process.
B. Scan and Association With Interleaving (SAI)
The SA strategy performs consecutive scan and association processes without interleaving and may introduce a long ser-vice disruption. For the Scan and Association with Interleaving (SAI) strategy, an MS stays in the neighboring channel for
𝐹𝑠𝑐𝑛frames for initial ranging and then returns to the serving channel for𝐹𝑖𝑡𝑣𝑙frames for communication. The total duration of the scan and association process increases, since extra time is introduced for the interleaving intervals. Moreover, the ranging response may be missing if the MS leaves the neighboring channel just when the neighboring BS responds to the ranging request. 𝑃𝑐 is the probability that the ranging request is lost due to collision, and𝑃𝑚is the probability that the ranging response is missing due to the MS leaving the neighboring channel. Thus, we can model the average scan and association duration in frames as:
𝑇𝑎′= 𝐿−1∑ 𝑘=0 ( (𝑃𝑐+ (1 − 𝑃𝑐) ⋅ 𝑃𝑚)𝑘⋅ (1 − 𝑃𝑐) ⋅ (1 − 𝑃𝑚) ⋅(𝐷𝑘′+ 𝑘 ⋅ (T3 + 𝐹𝑖𝑡𝑣𝑙2⋅ 𝑇𝑓) +𝑇2𝑤) ) (4) where 𝐷𝑘′ =∑𝑘𝑖=0𝑑𝑖′ and 𝑑𝑖′ is the average waiting time
for the 𝑖𝑡ℎ transmission of the ranging request. Due to the
involvement of interleaving intervals, the average delay for the back-off window increases to
𝑑𝑖′= ⌈ 𝑊𝑖− 1 2𝑆 ⌉ × 𝑇𝑓× (1 +𝐹𝐹𝑖𝑡𝑣𝑙 𝑠𝑐𝑛) (5)
In the above equation, we also consider the case that the timeout of the ranging response occurs when the MS is in
3322 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO. 11, NOVEMBER 2010 R When F ш Fit l R When F < Fit l R ec ei ve t h e Fw Fw R ec ei ve t h e
When Fscnш Fitvl When Fscn< Fitvl
e req u es t a w w F +F e req u es t a ȴ // ȴ// and s end ra Fscn+Fitvl Fscn+Fitvl and s end ra an gin g res p (x1,y1) Fitvl an gin g res p ȴ/ ȴ/ p onse af te r itvl Fitvl
Ranging response received
p onse af te r r y fra m es 0 Fscn 0 Fscn
Ranging response lost g g p
r y
fra
m
es
send CDMA ranging code at frame x send CDMA ranging code at frame x Fig. 3. Possible situations for ranging requests and responses.
an interleaving interval. The MS cannot immediately initiate the next ranging request when it is still in an interleaving interval. Therefore, the average time for waiting a ranging response timeout and initiating the next ranging request should be modified as T3+𝐹𝑖𝑡𝑣𝑙×𝑇𝑓
2 .
Next, we model 𝑃𝑚, as shown in Fig. 3. The x-axis is the
time that an MS sends the ranging request to the neighboring BS, and the y-axis presents the delay time of the BS response to the ranging request if the BS receives the request. Only
situations where 𝑦 is less than 𝐹𝑤 =
⌊ 𝑇𝑤
𝑇𝑓
⌋
are considered, since we assume that the BS can respond to the ranging
request in 𝑇𝑤 ms if it receives the request. As can be seen
from the figure,(𝑥1, 𝑦1) means that the MS sends the ranging
request at frame𝑥1, and the BS responds to the request after 𝑦1frame. When the value of(𝑥1+ 𝑦1) modulo (𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙)
is between 𝐹𝑠𝑐𝑛 and 𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙− 1, the neighboring BS
responds to the ranging request when the MS returns to the serving channel. The ranging response is missing. Otherwise, the MS can receive the ranging response, since the MS is in a scanning interval. Therefore, the area in which the
x-axis is less than 𝐹𝑠𝑐𝑛 and the y-axis is less than 𝐹𝑤 shows
all possible situations for ranging requests and responses. It is important to note that𝑥 and 𝑦 must be integers. For the shadow
area within the𝐹𝑠𝑐𝑛× 𝐹𝑤 rectangle, the ranging response is missing. Thus, we can derive𝑃𝑚= 𝐹𝑠𝑐𝑛Δ×𝐹𝑤, whereΔ is the size of the shadow area. If𝐹𝑤> 𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙, we can rewrite
𝐹𝑤 = 𝑟 × (𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙) + 𝐹𝑤′ where 𝑟 is an integer and
0 ≤ 𝐹𝑤′ < 𝐹
𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙. There are 𝑟 + 1 separated shadow
areas, i.e.,Δ = 𝑟 × Δ′+ Δ′′.Δ′= 𝐹
𝑠𝑐𝑛× 𝐹𝑖𝑡𝑣𝑙, and forΔ′′, three possible situations should be considered.
Δ′′= 𝐹𝑤′× (𝐹𝑤′− 1) 2 , if 0 ≤ 𝐹𝑤′< min(𝐹𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) (6a) Δ′′= 𝐹𝑤′× min(𝐹 𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) − min(𝐹𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) × (min(𝐹𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) − 1)/2, if min(𝐹𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) ≤ 𝐹𝑤′ < max(𝐹𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) (6b) Δ′′= 𝐹 𝑠𝑐𝑛× 𝐹𝑖𝑡𝑣𝑙 − (𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙− 𝐹𝑤′) × (𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙− 𝐹𝑤′− 1)/2, if max(𝐹𝑠𝑐𝑛, 𝐹𝑖𝑡𝑣𝑙) ≤ 𝐹𝑤′ < 𝐹𝑠𝑐𝑛+ 𝐹𝑖𝑡𝑣𝑙 (6c)
The service disruption time during the scan and association
process is reduced to𝐹𝑠𝑐𝑛×𝑇𝑓 because the MS returns to the serving channel and resumes communication during interleav-ing intervals. However, the packets which are buffered durinterleav-ing scanning intervals may suffer from additional delay. The average additional packet delay introduced by applying the SAI can be denoted as𝐹𝑠𝑐𝑛
2 ×𝑇𝑓×𝐹𝑠𝑐𝑛𝐹𝑠𝑐𝑛+𝐹𝑖𝑡𝑣𝑙 =
𝐹2
𝑠𝑐𝑛⋅𝑇𝑓 2(𝐹𝑠𝑐𝑛+𝐹𝑖𝑡𝑣𝑙),
which is no longer than a half of a scanning interval. From the above equations and discussion, we can learn that, for a long scanning interval and short interleaving interval, the duration of the scan and association process can be reduced. However, the QoS of communication is scarified because the packet delay and service disruption time during the scan and association process increase for a long scanning interval.
IV. ANALYTIC ANDSIMULATIONRESULTS
An IEEE 802.16e MAC-layer simulator written in C++ was developed to evaluate the performance of scan and association
process. The simulator used a 5 ms frame length (𝑇𝑓 = 5),
50 ms timeout (T3= 50), and a BPSK 1/2 modulation and
coding scheme. The low modulation and coding scheme—i.e.,
BPSK 1/2—was assumed because the scan and association
process was usually triggered when the signal strength of the serving BS became poor. It was assumed that ten MSs were performing ranging for handover to the neighboring BS. Moreover, the configuration of the neighboring BS was:
𝐵𝑠 = 5, 𝐵𝑒 = 10, 𝐿 = 16, 𝑇𝑤 = 50, two CDMA codes
for ranging, and two slots per ranging region every 50 ms on average.
The first simulation considers the ideal scenario, where the MS performs the scan and association with only one neighbor-ing BS and subscribes no real-time services. Fig. 4 shows both analytical and simulation results of the scan and association duration under different scanning and interleaving intervals. When comparing the simulation and analytical results, the differences are less than 5%. These results demonstrate the accuracy of the proposed models.
Without an interleaving interval (𝐹𝑖𝑡𝑣𝑙 = 0), the result
presents the scan and association duration by applying the SA strategy. Fig. 4 shows that the scan and association duration when employing the SA is about 122 frames, i.e., 610 ms, which is the service disruption time during the scan and association process. As can be seen in Fig. 4, short scanning intervals and long interleaving intervals result in long scan and association durations. To reduce the duration of the scan and association process, long scanning intervals and short interleaving intervals are preferred. However, a long scanning interval introduces more communication service disruptions.
In the second simulation, we assume that an MS establishes a G.711 voice communication, which is 64 kbps and 20ms frame length, and the serving BS suffers from different work-loads. When the serving BS has a high workload, the MS may have to stay in the serving channel for more frames—i.e., a longer interleaving interval— to maintain service. Two lines in Fig. 5 indicate feasible configurations when the serving BS workloads are 10% and 70%. A feasible configuration means that the configuration of scanning and interleaving intervals satisfies the bandwidth requirements for the real-time communication under the workload of the specific serving BS. A minimal interleaving interval can be determined for each given scanning interval to generate a feasible configuration.
TSAO et al.: EVALUATION OF SCAN AND ASSOCIATION PROCESS FOR REAL-TIME COMMUNICATION IN MOBILE WIMAX 3323 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400 3600 3800 4000 5 10 15 20 25 30 35 40 Scan and association duration (ms)
Scanning interval (frame)
Interleaving interval = 0 Interleaving interval = 2 Interleaving interval = 4 Interleaving interval = 8 (Analytic) Interleaving interval = 0 (Analytic) Interleaving interval = 2 (Analytic) Interleaving interval = 4 (Analytic) Interleaving interval = 8
Fig. 4. Scan and association duration under different scanning and inter-leaving intervals.
If we consider the delay constraint, some configurations in the figure become infeasible. For example, the SA strategy introduces a 610 ms service disruption time, which may be too long for a real-time communication. If we set the service disruption time as 50 ms, configurations where the scanning interval is less than 10 frames are feasible for the SAI strategy. If real-time commutation can tolerate a maximum of 200 ms service disruption time, the configurations can accommodate scanning intervals with 40 frames. The above examples demonstrate that the scan and association strategy and the scanning and interleaving interval should be carefully chosen to minimize the service disruption time and the scan and association duration. The proposed analytic model helps the MS and/or the BS to easily determine the parameters of the scan and association process.
The next simulation considers an H.261 video communica-tion, which is 256 kbps and 25 video frames per second. Video communication requires more radio resources, which implies a longer interleaving interval, when the scanning interval increases. In these cases, increasing the scanning interval may often increase the scan and association duration, especially when the serving BS suffers from a heavy workload. Fig. 6 demonstrates the scan and association duration when an MS establishes a video communication. In Fig. 5 and Fig. 6, each peak of the feasible configurations implies a change of the length of the interleaving interval owing to the bandwidth constraint of the real-time communication.
V. CONCLUSIONS
In this letter, we investigated the scan and association process and proposed analytical models of the scan and association duration for real-time communication in Mobile WiMAX. The accuracies of the proposed models were verified by simulations. With the proposed models, an MS or a serving BS can easily determine the parameters for performing the scan and association process without affecting the QoS of communication. The results demonstrate that with proper selection of the scanning and association interval, the service disruption time of real-time communication and the duration of the scan and association process could be both reduced.
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 5 10 15 20 25 30 35 40 Scan and association duration (ms)
Scanning interval (frame)
Feasible configurations when the serving BS workload = 10% Feasible configurations when the serving BS workload = 70% (Analytic)Feasible configurations when the serving BS workload = 10% (Analytic)Feasible configurations when the serving BS workload = 70% Interleaving interval = 0 Interleaving interval = 1 Interleaving interval = 2 Interleaving interval = 3
Fig. 5. Scan and association duration under different scanning and inter-leaving intervals and serving BS’s workload for a voice communication.
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 5 10 15 20 25 30 35 40 Scan and association duration (ms)
Scanning interval (frame)
Feasible configurations when the serving BS workload = 10% Feasible configurations when the serving BS workload = 70% (Analytic)Feasible configurations when the serving BS workload = 10% (Analytic)Feasible configurations when the serving BS workload = 70% Interleaving interval = 0 Interleaving interval = 2 Interleaving interval = 4 Interleaving interval = 8
Fig. 6. Scan and association duration under different scanning and inter-leaving intervals and serving BS’s workload for a video communication.
ACKNOWLEDGMENT
The authors would like to thank the National Science Council of the Republic of China for financially supporting this research under Contract No. NSC 98-2220-E-009-013-, NSC 99-2220-E-009-045-, and NSC 98-2219-E-009-019-.
REFERENCES
[1] IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Broadband Wireless Access Systems, IEEE Std. 802.16-2009, May 2009.
[2] J. Chen, C.-C. Wang, and J.-D. Lee, “Pre-coordination mechanism for fast handover in WiMAX networks,” in Proc. 2nd International Conf.
Wireless Broadband Ultra Wideband Commun., Aug. 2007.
[3] R. Rouil and N. Golmie, “Adaptive channel scanning for IEEE 802.16e,” in Proc. IEEE Military Commun. Conf., Oct. 2006, pp. 1–6.
[4] B. J. Kwak, N. O. Song, and L. E. Miller, “Performance analysis of exponential backoff,” IEEE/ACM Trans. Networking, vol. 13, no. 2, pp. 343–355, Apr. 2005.
[5] T. Peyre and R. ElAzouzi, “Performance analysis of single cell IEEE 802.16e wireless man,” in Proc. 32nd IEEE Conf. Local Computer
Networks, Oct. 2007, pp. 262–263.
[6] Y. P. Fallah, F. Agharebparast, M. R. Minhas, H. M. Alnuweiri, and V. C. M. Leung, “Analytical modeling of contention-based bandwidth request mechanism in IEEE 802.16 wireless networks,” IEEE Trans.
Veh. Technol., vol. 57, no. 5, pp. 3094–3107, Sep. 2008.
[7] L. Lin, W. Jia, B. Han, and L. Zhang, “Performance improvement using dynamic contention window adjustment for initial ranging in IEEE 802.16 P2MP networks,” in Proc. IEEE Wireless Commun. Networking
