An Integrated WiMAX/WiFi Architecture with QoS Consistency over Broadband Wireless Networks

An Integrated WiMAX/WiFi Architecture with QoS Consistency over Broadband Wireless Networks Hui-Tang Lin 1,2,#, Ying-You Lin 2,#, Wang-Rong Chang 1,#, and Rung-Shiang Cheng 1,* 1 Department of Electrical Engineering, National Cheng Kung University, Taiwan (R.O.C.) 2 Institute of Computer and Communication Engineering, National Cheng Kung University, Taiwan (R.O.C.) E-mails: # {htlin, n2892138, q3695119}@mail.ncku.edu.tw; * chengrs@nsda.ee.ncku.edu.tw. Abstract WiMAX and WiFi have emerged as promising broadband access solutions for the latest generation of wireless MANs and LANs, respectively. Their complementary features enable the use of WiMAX as a backhaul service to connect multiple dispersed WiFi hotspots to the Internet. This study proposes an integrated architecture utilizing a novel WiMAX/WiFi Access Point (W 2 -AP) device to effectively combine the WiMAX and WiFi technologies. In the proposed architecture, the protocol operation of the WiFi hotspots is the same as that of the WiMAX system. As a result, the WiFi network can support connection-oriented transmissions and QoS in a similar fashion to the WiMAX system, and thus a considerable improvement in the delay performance is obtained. The numerical results obtained using a Qualnet simulator confirm both the effectiveness and the efficiency of the proposed architecture. Keywords WLAN; WiFi; WiMAX; Wireless Networks; I. INTRODUCTION Wireless access techniques are continuously expanding their transmission bandwidth, coverage, and Quality of Service (QoS) support in recent years. With the huge market success of Wireless Local Area Networks (WLANs) (governed by the IEEE 802.11 standard), the new-generation wireless technique, WiMAX (i.e., IEEE 802.16), has now been standardized and deployed [1-3]. Traditionally, WiFi hotspots are connected to the Internet through a wired connection (e.g., Ethernet), and therefore have high deployment costs, particularly in remote rural or suburban areas with low population densities. Therefore, it is necessary to develop new schemes capable of providing sufficient bandwidth to meet the enormous access requirements of WiFi nodes while simultaneously reducing the backhaul cost. It has been suggested that the evolving family of WiMAX-based Wireless Metropolitan Area Network (WMAN) technologies represent a promising solution for providing WLAN hotspots with backhaul support [4]. Using a WiMAX-based backbone network to connect WiFi hotspots to the Internet not only avoids the requirement for a costly wired infrastructure, but also makes possible the provision of mobile hotspot services to realize Intelligent Transportation System (ITS) applications. Generally speaking, when constructing integrated WiMAX/WiFi networks, one of the most challenging issues facing network designers is that of designing efficient links and Medium Access Control (MAC) layer protocols to optimize the QoS between the WiMAX and the WiFi components of the architecture [5]. Several researchers have recently proposed QoS provisioning mechanisms for integrated WiMAX/WiFi systems [6][7]. For example, in [6], the authors proposed a QoS framework for 802.16/802.11 internetworking applications designed to map the QoS requirements of an application originating in an IEEE 802.11e network to an IEEE 802.16 network. Similarly, in [7], a QoS control protocol was presented to support an integrated QoS for converged networks comprising WiMAX and WiFi systems. However, in [6], the mechanisms required to satisfy the QoS requirements (e.g., bandwidth assignment, scheduling, admission control, and so forth) were not considered, while in [7], implementing the proposed QoS provisioning mechanism required a major rework of the WiMAX and WiFi control protocols. Accordingly, the current study proposes an efficient and unified connection-oriented architecture for integrating WiMAX and WiFi technologies in broadband wireless networks. In the proposed approach, a new wireless Access Point (AP) device, designated as WiMAX/WiFi AP (W 2 -AP), is developed to manage the WiMAX/WiFi interface. In addition, a modified convergence MAC layer of WiFi interfaces is designed by embedding the 802.16 subscriber MAC function within the original 802.11 MAC. Collectively, the W 2 -AP device and the extended MAC functionality enable each WiFi hotspot to support connection-oriented transmissions and differentiated services. This study also develops a Two-level Hierarchical Bandwidth Allocation approach, designated as THBA, for controlling and allocating the available bandwidth within the WiMAX/WiFi network. Importantly, the protocol used by the W 2 -AP device in communicating with its WiFi nodes is based upon the same IEEE 802.16 protocol used by the WiMAX Base Station (BS) in communicating with the W 2 -AP device, and thus QoS continuity and bandwidth management consistency can be obtained throughout the integrated network. The remainder of this paper is organized as follows. Section II describes the proposed integrated network architecture, while Section III introduces the THBA scheme. Section IV presents the results of a series of simulations designed to evaluate the performance of the proposed architecture. Finally, Section V draws some brief conclusions and indicates the intended direction of future research. II. DESCRIPTION OF NETWORK MODEL This section commences by introducing the underlying WiMAX/WiFi system and then provides a detailed description of the proposed MAC layer module. A. Integrated WiMAX/WiFi System Figure 1 presents a typical example of the integrated WiMAX/WiFi network architecture considered in this study. As shown, a single WiMAX BS, operating in a licensed band, serves both multiple WiMAX Subscriber Stations (SSs) and multiple W 2 -APs within its coverage area. In other words, the WiMAX system provides broadband wireless access to multiple W 2 -AP devices in a point-to-multipoint (PMP) topology. As shown, every WiFi network is connected to the WiMAX BS through a W 2 -AP. The connection between the BS 978-1-4244-2309-5/09/$25.00 2009 IEEE

Fig. 1. Integrated WiMAX/WiFi Network architecture. and a SS is dedicated to a single user. However, the connection between the BS and each W 2 -AP is shared amongst all the nodes within the WLAN served by the W 2 -AP. As a result, the WiMAX network provides a backhaul service connecting multiple dispersed WiFi hotspots to the Internet. B. Proposed MAC Layer Module Figure 2 illustrates the MAC layer module developed in this study to support the integrated operations of the WiMAX/WiFi network. As described in the following, the MAC layer module is realized using two adaptive MAC frameworks, namely MultiMAC and SoftMAC. MultiMAC: As in [8], the MultiMAC framework acts as a mediating MAC layer between the physical (PHY) device and the network layer, and performs a dynamic switching between different MAC protocols. Therefore, individual MAC variants can be used by MultiMAC for decoding their respective incoming frames and encoding outgoing frames with the MAC best suited to the current network conditions. When a frame arrives, the appropriate MAC layer claims the frame and then decodes it. As discussed in [8], claiming a frame can be achieved in a variety of ways, e.g., by marking identifiers of the existing frame, adding a byte to the header of the frame for identification purposes, and so on. SoftMAC: The SoftMAC framework is a convergence sub-layer located beneath the WiMAX frame layer. The objective of SoftMAC is to encapsulate a WiMAX Packet Data Unit (PDU) into a single WiFi PDU over 802.11a OFDM PHY or to decapsulate a single WiFi PDU into its WiMAX PDU. (Note that the technique of embedding an 802.16 PDU over 802.11a has been investigated and validated in [9] and is therefore not discussed here.) As shown in Fig. 2(a), the MAC protocol module proposed in this study comprises three major elements; namely the W 2 -AP device, the WiFi node, and the WiMAX BS. MAC Module in W 2 -AP Device The W 2 -AP element comprises MultiMAC, the proposed convergence 802.16 MAC, and 802.16 OFDM PHY embedded within a conventional 802.11 AP device. In the upstream service flow, when MultiMAC receives a packet from the 802.11a PHY device, it decides which MAC (i.e., 802.11 MAC or the convergence 802.16 MAC) is best suited to claiming the packet by examining the current connection state of the network. If the network is currently connecting a WiFi node to the WiMAX network, MultiMAC switches the received packets to the convergence 802.16 MAC. As shown in Fig. 2(b), the uplink (UL) traffic is classified as either UL data packets or Bandwidth Requests (BW-REQs) of WiFi nodes via a specific Separator module. The received UL data packets are forwarded by the Separator module, through the Common Part Sub-layer (CPS) and the Convergence Sub-layer (CS) of Layer 2 BS-MAC, and then to the UL Aggregator/Classifier of the proposed Layer 2.5 MAC. For the BW-REQs, which are typically processed by the 802.16 MAC in a conventional 802.16 system, instead, they are forwarded to the UL Aggregator/Classifier and the UL Scheduler for further processing. The UL Aggregator/Classifier performs the following two functions in the upstream service flow: Packet Classification: Similar to the packet classification in the conventional 802.16 SS-MAC, the packet classification function is to temporarily buffer the UL data packets based on their corresponding priority. Since each W 2 -AP acts as a relay node between a WiFi and a WiMAX network, all these backlogged packets are directly forwarded to CPS plane of 802.16 SS-MAC, and are relayed to the WiMAX BS during the granted time interval defined in the UL-MAP from the WiMAX BS. BW-REQ Aggregation: This function is to aggregate all bandwidth demands at the same QoS level from WiFi nodes into a single bandwidth request. Subsequently, these aggregated bandwidth demands at various QoS levels (the gray arrow in Fig. 2(b)) are forwarded to the UL BW Request Generator in the CPS plane which then transmits these aggregated BW-REQs to the WiMAX BS via the 802.16 PHY. In the downstream service flow, if WLAN hotspots currently use WiMAX as their backhaul services, the downlink (DL) traffic is separated by a Separator module into DL data packets and the DL/UL-MAP. As shown in Fig. 2(c), the DL data packets and the original the DL/UL-MAP are forwarded to the DL Traffic Processor of CPS plane in 802.16 SS-MAC and the Layer 2.5 MAC (the UL Aggregator/Classifier and the UL Scheduler). Upon receiving the UL-MAP, the UL Aggregator/Classifier can determine when to forward the buffered data packets from WiFi nodes in the upstream service flow. When the UL Scheduler receives the UL-MAP, it then finely reallocates the UL granted bandwidth to the WiFi nodes has sent the BW-REQs. The UL and DL bandwidth information is passed down to the DL/UL-MAP Generator (the gray arrow in Fig. 2(c)) to construct the corresponding DL/UL-MAP, which subsequently broadcasts to its associated WiFi nodes. The DL data packets from WiMAX BS are temporarily buffered at the corresponding priority queues in the CPS plane in 802.16 BS-MAC. They are relayed to their corresponding WiFi nodes following the schedule defined in the DL-MAP within the DL subframe. Through its use of MultiMAC and SoftMAC embedded in the proposed convergence 802.16 MAC, the W 2 -AP acts either as a bridge in translating frames between a WiFi and an Ethernet interface or as a relay node in transferring frames between a WiFi and a WiMAX network. Moreover, since the W 2 -AP incorporates a convergence 802.16 MAC comprising both SS and BS functions, it can also serve as a WiMAX sub-bs capable of sending BW-REQs (from the WiFi nodes) to the BS and allocating the granted bandwidth (from the BS) to the WiFi nodes. From the preceding discussions, it is clear that by upgrading existing 802.11 AP devices, the resulting W 2 -AP devices provide the means for WiFi users to connect to the Internet via a wired Ethernet infrastructure or a wireless broadband access mechanism (see Fig. 1).

Fig. 2. (a) MAC layer module. (b) Upstream service flow in the proposed convergence 802.16 MAC. (c) Downstream service flow in the proposed convergence 802.16 MAC. MAC Module in WiFi Node In the WiFi node element of the MAC protocol module, the conventional Network Interface Card (NIC) is modified by adding MultiMAC, SoftMAC and 802.16 SS-MAC functions to the 802.11 MAC layer using a software upgrade technique [8][9]. The lower MultiMAC module in the WiFi node assigns the received frames to the appropriate MAC layer depending upon the current connection state of the network, while the upper MultiMAC module decides which MAC is best suited to transmitting the individual packets received from the upper layer. In the proposed protocol module, WiFi nodes are permitted to transmit 802.11 and 802.16 frames to a wired Ethernet interface and a wireless WiMAX network, respectively. Therefore, the principal advantage of the upgraded 802.11 NIC card proposed in this study is that users located within the coverage area of a W 2 -AP can access the backhaul service provided by the local WiMAX network. In other words, no additional WiMAX interface is required for the users with traditional 802.11 devices to access the services provided by the WiMAX network, and thus the hardware and Internet access costs are significantly reduced. MAC Module in WiMAX BS In the WiMAX BS element of the MAC protocol module, the MAC layer is the same as that defined in the IEEE 802.16 standard. This MAC layer enables the WiMAX BS to assign available bandwidth to both SSs and W 2 -APs, and thus enables the coexistence and inter-working of WiMAX and WiFi technologies within a single, integrated network. The proposed combination of WiFi and WiMAX may be an attractive solution to wireless broadband access, which enables two techniques to inter-work each other in many aspects. One of the most significant factors to motivate such integration is supporting personal hotspot services. For example, consider the scenario where a group of users are currently within a building, but will shortly leave the premises and travel to another destination with low population density, e.g., a rural or a suburban area. Imagine also, that the users are currently using the Internet and wish to remain connected to the network when they reach to their destination. To support this personal hotspot requirement, each user can utilize a terminal device equipped with an upgraded NIC card (i.e., the proposed MAC module in the WiFi node shown in Fig. 2(a)) to access a nearby W 2 -AP device. When the users leave the building, they can simply take the W 2 -AP with them. Once they reach to the destination, they can install the W 2 -AP such that they are able to access the Internet through the backhaul service provided by the local WiMAX networks within the regions of their destination. Note that the demand for personal hotspot services such as that described in this fictitious scenario is expected to grow significantly in the near future to accommodate the requirement for ubiquitous network environments. III. BANDWIDTH ALLOCATION AND MANAGEMENT This section commences by introducing the proposed bandwidth allocation scheme and then describes the transmission frame format. Since the data communication between WiFi and Ethernet technologies is common knowledge, the following discussions on packet forwarding and bandwidth allocation focus on data communications between WiFi and WiMAX in the proposed integrated system. Note that in the discussions, an assumption is made that the WiFi and WiMAX networks use different frequency bands, and thus the problem of interference is neglected.

Fig. 3. Two-hierarchy bandwidth allocation scheme. A. Two-Level Hierarchical Bandwidth Allocation (THBA) WiMAX is a connection-oriented transmission technique in which each service flow is allocated a unique Connection ID (CID) [2][3], and BW-REQs and QoS support are processed in a connection-oriented manner. In satisfying the users bandwidth demands, the WiMAX BS allocates an aggregated bandwidth to each SS, and the SS then allocates this bandwidth to its various service connections. However, existing WiFi technology does not support this type of bandwidth allocation. As the mandatory access mechanism in 802.11e [10], Enhanced Distributed Channel Access (EDCA) defines the prioritized Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) mechanism. Differentiated services can be provided by introducing different Arbitration Inter-Frame Spaces (AIFSs) and Contention Windows (CWs) for different Access Categories (ACs) as well as different virtual collisions for priority queues in the same station. In 802.11e, each WiFi station maintains up to four priority queues, with each queue corresponding to one AC. Obviously, the overall operational principles of WiMAX and WiFi are quite different, particularly as regards their bandwidth access and QoS provisioning mechanisms. WiMAX systems generally utilize bandwidth more finely than WiFi systems. Furthermore, connection-oriented bandwidth allocation approaches tend to provide a more predictable QoS than CSMA/CA-based bandwidth contention schemes. Therefore, it is reasonable to expect 802.16 technologies to provide a better QoS than their 802.11e counterparts. Since WiFi and WiMAX use different operational protocols in their bandwidth access mechanisms, it is necessary to embed additional functionality into the MAC layer of the WiFi NIC card and AP device (described in Section II) to enable WiFi hotspots to support connection-oriented services in the same way as in WiMAX systems. By implementing this modification,, a common bandwidth access mechanism can be deployed throughout the entire WiMAX/WiFi network. The current study develops a Two-level Hierarchical Bandwidth Allocation (THBA) scheme to request and grant bandwidth within the proposed integrated network. The following discussions consider the illustrative example shown in Fig. 3. As shown, two WiFi nodes (Nodes #3 and #4) send BW-REQs (CIDs #A1 and #A2) to the W 2 -AP. The W 2 -AP intercepts and aggregates these requests via the UL Aggregator/Classifier module (described in Section II-B), and then sends the aggregated BW-REQs (Request CID #B3) to the BS. It is recalled that when aggregating the BW-REQs from the WiFi nodes within a WLAN, the BW aggregation function performs the aggregation process in accordance with the QoS requirements of the individual requests. In other words, the W 2 -AP attempts to aggregate bandwidth demands having the same level of QoS. Once the BS receives the aggregated request, it grants an aggregated bandwidth (Grant for W 2 -AP #2) to the W 2 -AP, which then finely reallocates this bandwidth to the two WiFi nodes (Grants for Nodes #3 and #4) through its UL Scheduler. In other words, the protocol operation comprises two levels, namely Level A between the WiFi nodes and the W 2 -AP and Level B between the W 2 -AP and the BS. The BS is unaware of the detailed bandwidth requirements of the WiFi nodes in Level A since this information is summarized by the W 2 -AP prior to its transmission to the BS. Note that the CIDs assigned by the W 2 -AP and the BS are independent of one another and can therefore exist simultaneously in the two different levels of the bandwidth allocation hierarchy without affecting the operation of the system. The two-level bandwidth allocation approach described above enables the WiMAX and WiFi systems within the integrated network to be controlled via a single protocol based on the IEEE 802.16 standard used by the WiMAX system. Thus, the proposed bandwidth allocation scheme has two major advantages. First, the MAC protocols laid down in the WiFi and WiMAX standards require only slight modification, and therefore the time and expense incurred in implementing the integrated network are reduced. Second, the WLAN can support a fine level of QoS as in WiMAX without the need for any form of QoS mapping mechanism (e.g., in [6]). Therefore, the implementation complexity of the integrated network is reduced, and both QoS continuity and bandwidth management consistency can be obtained throughout the network. B. MAC Frame Structure Figure 4 illustrates the MAC frame structure to realize the THBA scheme for the illustrative example shown in Fig. 3. Note that to simplify the following discussions, the propagation and the W 2 -AP processing delay are deliberately neglected. Each frame broadcast by the BS or the two W 2 -AP devices consists of a DL subframe and an UL subframe, respectively. The length of both subframes is dynamically determined in accordance with the varying traffic load received from the upper layer. The DL subframes start with the periodic broadcast of various items of control information, including the DL preamble, the frame control header (not shown in Fig. 4), and the first DL burst (i.e., DL Burst #1). As shown, the first DL burst includes both the DL-MAP and the UL-MAP to define the access time interval to the DL and UL channels, respectively. The UL subframe commence with a BW-REQ interval and then comprise a series of UL bursts. The BW-REQ interval in Level B (or in Level A) is reserved for the BS (or for a W 2 -AP) to poll the W 2 -AP devices (or the WiFi nodes). The BS (or W 2 -AP) defines the duration carried by the UL-MAP during which a given W 2 -AP (or a given WiFi node) is permitted to issue the BW-REQ to the BS (or the W 2 -AP). Based on such contention-free polling approach [11][12], the BS (or W 2 -AP) allocates burst transmissions of the UL subframe to different W 2 -APs (or to different WiFi nodes). The W 2 -APs (or WiFi nodes) use the periodic DL preamble issued by the BS (or by a W 2 -AP) for synchronization purposes and assume the current frame duration to be equivalent to the time interval between the broadcast of two consecutive DL preambles. Note that the frame duration has a value in the range of 2.5~20 ms, and is fixed during normal operation.

Fig. 4. MAC frame structure. To achieve a unified frame structure in the WiMAX and WiFi networks, it is assumed that the frame durations broadcasted by the BS and the W 2 -AP devices are identical. Frame Operation in Level B From the perspective of the BS in Fig. 3, W 2 -AP #1 and W 2 -AP #2 in Level B act as conventional SSs. Thus, the DL-MAP and UL-MAP messages advertised by the BS at the beginning of each frame contain the time boundaries of the DL and UL grants assigned to W 2 -APs #1 and #2. More specifically, a DL grant in the DL-MAP announces the transmission by the BS of a burst of PDUs addressed to a given W 2 -AP (or SS) at a certain time interval within the DL Subframe K. Similarly, a UL grant in the UL-MAP indicates a time interval within the UL Subframe K during which a particular W 2 -AP (or SS) is allowed to transmit the burst of PDUs requested in the previous MAC frame, i.e., Frame K 1. In DL Subframe K, after W 2 -APs #1 and #2 receive the first DL burst, they listen to the following bursts transmitted by the BS and receive and process the particular PDU bursts addressed to them. In UL Subframe K, each W 2 -AP relays a burst of MAC PDUs from its WiFi nodes to the BS in a Time Division Multiple Access (TDMA) manner. Frame Operation in Level A In Level A, the WiFi nodes appear to be ordinary SSs from the perspective of the W 2 -AP devices, while the W 2 -AP devices appear as conventional BSs to the WiFi nodes. As in the protocol operation described above for Level B, each W 2 -AP device periodically advertises transmission frames to its WiFi nodes. Each W 2 -AP adopts the THBA scheme to further make scheduling decisions for all its WiFi nodes when it obtains a transmission grant for the aggregated bandwidth (requested in the previous Frame K 1) from the BS in the UL-MAP of Frame K. As a result, the associated Frame K can be immediately advertised to the WiFi nodes through the WiFi interface as soon as the UL Scheduler of the W 2 -AP has received the UL-MAP information in Frame K via the WiMAX interface. Therefore, as shown in Fig. 4, a gap (referred to as the Second-Hop Frame Gap (SFG)) is introduced between Frame K and Frame K. Typically, the length of the SFG takes into account: The time required to send the control message in Frame K. The propagation delay. The DL/UL-MAP processing time. Since in the present discussions, the propagation delay and processing time are neglected, the SFG shown in Fig. 4 consists only of the control message transmission time. When the UL Scheduler receives the DL-MAP of Frame K from the WiMAX BS, it is able to derive the length of DL Subframe K, which is equal to T K SFG T BS, where T K is the time duration of DL Subframe K and T BS is the length of a pre-determined BW-REQ interval. This information is passing down to the DL/UL-MAP Generator to construct the DL/UL-MAP of Frame K, which is immediately broadcasted to the associated WiFi nodes to indicate the start of Frame K. Therefore, the end of the BW-REQ interval of UL Subframe K can align with the beginning of the BW-REQ interval of UL Subframe K as shown in Fig. 4. When a W 2 -AP and its WiFi nodes use the BW-REQ slots to make BW-REQs, this back-to-back arrangement of the BW-REQ intervals ensures that the W 2 -AP is able first to receive and aggregate the BW-REQs from its WiFi nodes and then to immediately access the BW-REQ slots of UL Subframe K in order to transmit this aggregated BW-REQ to the BS. (Note that to achieve a unified frame length in the WiMAX and WiFi networks, a time interval equivalent to the sum of SFG and T BS is appended to UL Subframe K to increase its length.) In DL Subframe K, the W 2 -AP uses its own scheduler to determine the particular time intervals at which it will issue the PDUs to the various WiFi nodes. Note that these MAC PDUs are PDUs previously transmitted by the BS in DL Subframe K 1, received and buffered at the W 2 -AP queue (i.e., buffered at the corresponding priority queues maintained by 802.16 BS-MAC). After an interval T BS in UL Subframe K, the following UL bursts are immediately scheduled for the UL transmissions of the individual WiFi nodes. For each W 2 -AP, the total amount of bandwidth granted to the WiFi nodes is equal to the aggregated bandwidth granted to the W 2 -AP by the BS. As a result, each W 2 -AP afterward relays the burst sent from its WiFi nodes in UL Subframe K to BS during its granted time interval in UL Subframe K. IV. PERFORMANCE EVALUATION The performance of the proposed system in providing an integrated QoS management capability across the WLAN (i.e., Level A) and WMAN (i.e., Level B) network segments was evaluated by performing a series of numerical simulations using the QualNet simulator [13]. A. Simulation Model A simulation model was developed comprising an integrated network with one BS, five stationary SSs, and two W 2 -APs

Mean End-to-End Delay (ms) 200 180 160 140 120 100 80 60 40 20 0 VoIP (THBA) Video (THBA) Web (THBA) VoIP (Indep.) Video (Indep.) Web (Indep.) VoIP (EDCA) Video (EDCA) Web (EDCA) 1.28 2.56 3.84 5.12 6.4 7.68 8.96 10.24 Network Offered Loads (Mbps) Mean End-to-End Delay (ms) 200 180 160 140 120 100 80 60 40 20 0 VoIP (THBA) Video (THBA) Web (THBA) VoIP (Indep.) Video (Indep.) Web (Indep.) VoIP (EDCA) Video (EDCA) Web (EDCA) 10 20 30 40 50 60 70 Number of VoIP Connections (a) (b) Fig. 5. Mean end-to-end delay of various traffic classes for: (a) various network offered loads, and (b) the number of VoIP connections within the network. (i.e., two WiFi hotspots). The IEEE 802.16 protocol was implemented using the WirelessMAN-OFDM air interface in TDD mode. The total bandwidth of the BS was assumed to be 64 Mbps. The transmission frame size was specified as 10 ms in every case, and the ratio of the DL subframe duration to the UL subframe duration was set to 50:50. The SSs and WiFi nodes were assumed to generate three different types of traffic, each with a different priority, namely (i) VoIP (UGS) with a 64 kbps Constant Bit Rate (CBR) and stringent delay requirements, (ii) Video (rtps) with a 128 kbps CBR, and (iii) Web (BE) modeled for simplicity with a 64 kbps CBR. The packet intervals of VoIP, Video, and Web were assumed to be 20 ms, 10 ms, and 12.5 ms, respectively. For the WLAN segment of the integrated network, an assumption was made that the upstream channel rate was 12 Mbps and the cell radius was 50 meters. The key evaluation metric is the mean uplink end-to-end traffic delay. The end-to-end delay is defined here as the elapsed time between the arrival of a data packet at the WiFi node and the instant at which this packet was received by the BS receiver. The simulation runs achieve a confidence interval of 5% for a 95% confidence level. Note that since the confidence intervals of all numerical results are very small, these confidence intervals are not plotted along with the simulation results shown in this section. The following discussions compare the performance results obtained for three specific bandwidth allocation protocols in the WiMAX/WiFi system, namely: THBA scheme: In this case, the proposed two-level hierarchical scheme is employed. In other words, each W 2 -AP aggregates all the bandwidth demands with the same QoS level received from its WiFi nodes and then requests the BS to allocate the total amount of bandwidth required. Having received this aggregated bandwidth from the BS in the UL-MAP of a particular Frame K, the W 2 -AP establishes the corresponding Frame K to perform a fine allocation of the bandwidth amongst its WiFi nodes in accordance with their original BW-REQs. As a result, each W 2 -AP has a guaranteed ability to relay all of the WiMAX PDUs transmitted by its WiFi nodes to the BS during its granted time interval in UL Subframe K. Independent bandwidth allocation control (Independent): The bandwidth allocation approach in this case is based on the integrated system described above. However, in the WLAN and WMAN segments, the BW-REQ and allocation procedures are performed independently using their respective IEEE 802.16 protocols without tightly coupling the two segments in terms of time. In other words, in a WLAN segment, the W 2 -AP directly allocates its own bandwidth to its WiFi nodes based on their bandwidth requirements and buffers the packets sent from the WiFi nodes. Meanwhile, in the WMAN segment, the W 2 -AP sends BW-REQs to the BS in accordance with the length of its queues and transmits the buffered packets from the WiFi segment to the BS during the granted time interval within the UL frame. WiMAX with EDCA control (EDCA): Under this procedure, the WiFi nodes now are equipped with IEEE 802.11e NIC card to employ a priority-based EDCA bandwidth contention protocol for accessing the channel. The simulation parameters of EDCA are shown in Table I. In addition, in each W 2 -AP device, the convergence MAC (shown in Fig. 1) comprising SoftMAC and 802.16 BS-MAC now is replaced with 802.11e MAC as well as no Multi-MAC functionality is required. Therefore, each W 2 -AP device in this case only acts as a bridge in translating frames between the WiFi and WiMAX networks. Like the Independent scheme, the W 2 -APs transmit BW-REQs to the BS in accordance with the length of their local queues. Note that even though 802.11e also defines an alternative protocol, designated as HCCA (HCF Control Channel Access), which is an enhanced version of PCF (Point Coordination Function), most of commodity WLAN interface products do not support the operation of HCCA or PCF. Therefore, the performance comparison between the proposed THBA scheme and HCCA is excluded in the simulation. TABLE I. SIMULATION PARAMETERS FOR EDCA. Voice Video Best Effort Transport Protocol UDP (CBR) UDP (CBR) UDP (CBR) AC VO VI BE CW min 3 7 15 CW max 7 15 1023 AIFSN 2 2 3 B. Numerical Results Figure 5 shows the variation in the end-to-end delay for various network offered loads (Fig. 5(a)) and the number of VoIP connections within the network (Fig. 5(b)), respectively.

Mean End-to-End Delay (ms) 200 180 160 140 120 100 80 60 40 20 0 VoIP (THBA) Video (THBA) Web (THBA) VoIP (Indep.) Video (Indep.) Web (Indep.) VoIP (EDCA) Video (EDCA) Web (EDCA) 5 10 15 20 25 30 35 Number of WiFi Nodes per WiFi Hotspot Fig. 6. Mean end-to-end delay versus number of WiFi nodes per WiFi hotspot. Note that each WLAN segment is assumed to have 5 WiFi nodes. It is observed that the proposed THBA scheme yields a significant improvement in the end-to-end delay thanks to its use of a unified connection-oriented bandwidth allocation approach, which enables each W 2 -AP to sends the aggregated BW-REQ of its WiFi nodes to the BS in the current frame and then relays these packets sent from its WiFi nodes to the BS at the next frame immediately. In other words, the elapsed time between the bandwidth requirements of the backlogged packets issued from WiFi nodes until the instant at which these packets have been received by the BS is equivalent to the duration of one frame. The results therefore confirm the ability of the proposed THBA scheme to yield efficient use of the available bandwidth to deliver various traffic classes. It can also be seen that in the proposed scheme, the VoIP flow achieves its committed QoS performance in terms of the end-to-end delay and is not significantly affected as the network traffic load or the number of VoIP connections is progressively increased. Figure 6 illustrates the variation of the average end-to-end delay of each type of traffic as the number of WiFi nodes within each WLAN segment is increased from 5 to 35. It can be seen that the end-to-end delay curves increase only slightly when the system operates under light traffic loads (i.e., the number of WiFi nodes is less than or equal to 10), since under this condition, all uplink queues of WiFi nodes are almost empty. However, when the system operates under high traffic loads (i.e., the number of WiFi nodes exceeds 10), the end-to-end delays of the various traffic classes increase. It can be seen that the delay incurred when using the EDCA scheme is significantly higher than that incurred under either the THBA scheme or the Independent scheme. This result is to be expected since when the EDCA protocol is applied, the number of contentions between traffic flows within the same priority class increases, and hence the average end-to-end delay also increases. Furthermore, it is clear that the THBA scheme proposed in this study consistently achieves a lower delay than the Independent method. Intuitively, this performance enhancement again confirms that when a WiFi node requests the bandwidth of the backlogged packets to its attached W 2 -AP, the W 2 -AP is capable of relaying these packets to the BS with a delay time of one frame duration, and thus the average end-to-end delay of each traffic priority is reduced. V. CONCLUSIONS AND FUTURE WORKS This study has proposed a unified connection-oriented architecture to support the integration of WiFi and WiMAX technologies in broadband wireless networks. In the proposed architecture, a W 2 -AP device is employed to interconnect WiFi hotspots with a WiMAX backhaul service. By using the W 2 -AP and a terminal device configured with the software-upgraded 802.11 NIC card, the integrated architecture enables the use of a common protocol operation between the WiMAX and WiFi components of the network. A THBA mechanism has been proposed for arbitrating the DL and UL transmissions and for ensuring QoS continuity and bandwidth management consistency throughout the network. The numerical results have confirmed that the integrated architecture enables the provision of differentiated services between WiMAX and WiFi interfaces, while significantly reduces the end-to-end delays of high priority traffic. In the future study, the authors intend to examine how to extend the integrated network architecture to OFDMA mode for supporting mobile hotspot services and mobile wireless telemedicine services. In the first case, W 2 -AP devices with 802.16e OFDMA PHY can be installed in public transport vehicles (e.g., trains, busses, and so forth), such that any traveler with an upgraded NIC interface can access the Internet via the local WiMAX network. In the second case, the W 2 -AP device can be installed in an ambulance, for example, to enable the paramedics to inform the hospital or health care center of the medical condition and requirements of the incoming patient. ACKNOWLEDGEMENT The authors would like to thank the National Science Council, Taiwan, R.O.C., for supporting this research under grant NSC 97-2221-E-006-175 -MY3. REFERENCES [1] IEEE Std 802.16-2004, IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems, Oct. 2004. [2] C. Eklund, et al, IEEE Standard 802.16: A Technical Overview of the Wireless MAN Air Interface for Broadband Wireless Access, IEEE Commun. Mag., vol. 40, no. 6, pp. 98-107, June 2002. [3] A. Ghosh, et al, Broadband Wireless Access with WiMax/802.16: Current Performance Benchmarks and Future Potential, IEEE Commun. Mag., vol. 43, no. 2, pp. 129-136, Feb. 2005. [4] Dusit Niyato and Ekram Hossain, Integration of WiMAX and WiFi: Optimal Pricing for Bandwidth Sharing, IEEE Commun. Mag., vol. 45, no 5, pp. 140-146, May 2007. [5] Cavalcanti D, et al, Issues in Integrating Cellular Networks WLANs, and MANETs: a Futuristic Heterogeneous Wireless Network, IEEE Wireless Commun. Mag., vol. 12, no. 3, pp. 30-41, June 2005. [6] Kamal Gakhar, Annie Gravey and Alain Leroy, IROISE: A New QoS Architecture for IEEE 802.16 and IEEE 802.11e Interworking, in Proc. of IEEE International Conference on Broadband Networks, pp. 607-612, Oct. 2005. [7] Pedro Neves, Susana SargentoRui, L. Aguiar, Support of Real-Time Services over Integrated 802.16 Metropolitan and Local Area Networks, in Proc. of IEEE ISCC, pp. 15-22, June 2006. [8] C. Doerr, et al, MultiMAC - An Adaptive MAC Framework for Dynamic Radio Networking, in Proc. IEEE DySPAN, pp. 548-555, Nov. 2005. [9] Djukic, Petar, Valaee, Shahrokh, Towards Guaranteed QoS in Mesh Networks: Emulating WiMAX Mesh over WiFi Hardware, in Proc. of International Conference on Distributed Computing Systems Workshops, pp. 15-15, June 2007. [10] IEEE draft for Wireless Medium Access Control (MAC) and Physical Layer (PHY) specifications, Medium Access Control Enhancements for Quality of Service (QoS), IEEE Std. 802.11e, Oct. 2004. [11] Rajagopal Iyengar, Prakash lyer, Biplab Sikdar, Delay Analysis of 802.16 based Last Mile Wireless Networks, in Proc. of IEEE GLOBECOM, vol. 6, pp. 5, Dec. 2005. [12] Qiang Ni, et al, Investigation of Bandwidth Request Mechanisms under Point-to-Multipoint Mode of WiMAX Networks, IEEE Commun. Mag., vol. 45, no 5, pp. 132-138, May 2007. [13] Qualnet Simulator Home. http://www.scalable-networks.com/