Blanket WLAN: A Distributed Network Evolution Paper #: 53 ABSTRACT This paper proposes a novel WLAN configuration designed to address the needs of networks with wireless transmitters that can provide coverage to a very limited area. To expand such coverage over a much larger area we propose a network with a dense deployment of Network Access Points (NAPs) and where a station simultaneously connects to multiple NAPs. Unlike the existing virtual WLAN solutions, our approach heavily emphasizes distributed decision making. Further, our approach preserves much of the current CSMA protocol in WLANs while effectively utilizing the legacy backhaul over which our network may be deployed. In this work we discuss some basic issues associated with such a network. Specifically we define an architecture which can leverage both the backhaul communication and the backhaul topology. Additionally, we propose two simple, but effective solutions to manage End-Node (EN) mobility in a fully distributed fashion. Finally, we demonstrate effective data communication protocols that allow the network to take advantage of the spatial diversity offered by the multi-NAP communication system while preserving the essential distributed scheduling and coordination benefits of CSMA. Keywords AP, CSMA, BWLAN, NAP, EN 1. INTRODUCTION Consider a wireless network comprising Network Access Points (NAPs), most of which provide very limited spatial coverage. The network as a whole, is, however, required to provide full coverage of a relatively large area, compared with the coverage of a single NAP. We think of this network as a LAN network, as its coverage is typically extended over a single building or certain substantial area of a building. Moreover, the entire network is under the control of its user, rather than a 3rd party operator like cellular networks. The proposition here is, essentially, to take a common WLAN approach and replace a relatively high-power Access Point (AP) with a large number of low-power ones, which we call NAPs to distinguish them from APs. Several underlying trends in today’s technology motivate the need of creating such networks. Some trends are outlined below: Enable new spectra and access technologies As communication engineers begin to explore everhigher electromagnetic spectra in search of new bandwidth, the need for such a network might arise naturally. A simple example of high spectra is the Visible Light Communications (VLC). The recent status of VLC development can be found in , as well as other references. Consider each light-bulb in a room as a NAP, as it has a very limited cone of coverage. A full coverage of an area significantly larger than a single cone will require utilization of multiple light-bulb NAPs. Such challenges of covering a large area with limited-coverage devices may also arise from other spectra like the 60 GHz RF (see, e.g., ) and ultraviolet. Solve coverage issues Modern WLAN systems suffer from the inconsistent coverage problem. A usual way of closing a coverage hole is increasing the transmission power. This results in the increase of the overall coverage area. On the other hand, it follows from security requirements that wireless signals should not be propagated beyond a desired coverage area. To some extent these challenges are being addressed with careful wireless network planning. However, a considerate site planning is in general beyond the capabilities of many homeowners and even enterprises. Moreover, the extent to which this approach addresses the coverage and leakage issues is often less than perfect. Low-coverage NAPs provide an alternative solution to this problem. Rather than resorting to a distant high-power AP, the network operator simply places a low-coverage NAP where it is needed. Because the new NAP’s coverage is by definition highly-limited, signal leakage outside the desired area is minimized. Enable low-power, reliable applications A feature of the network with low-coverage NAPs is that any End-Node (EN) in the network is likely close to several NAPs. This means that the EN communicating with the network requires lower power than in a traditional network. Additionally, reliability can be improved if the media access protocol can take advantage of the multiple available paths. We note that the concept described here is not entirely novel to the wireless LAN industry. Several WLAN solution providers offer an enterprise solution designed to provide consistent coverage using a large number of simple APs. Such a solution is called virtual WLAN, and the solution providers include Meru , , Extricom , Cisco, Aruba, and others. More proposals about virtual WLAN can be found in —. A major feature of virtual WLAN appears to be1 the presence of a centralized decision making entity which coordinates the operations of the WLAN APs. While an effective solution, this limits the ability of such a network to scale easily, as it places severe bandwidth and latency restrictions on the backhaul. Though an enterprise deploying a state of the art WLAN may be able to supply an appropriate backhaul, its deployment at a typical home or small enterprise environment may be difficult. In this paper we propose an alternative approach – one where the control of the network is mostly in a distributed fashion. Specifically, we decentralize most of the operations in the virtual WLAN into NAPs. Our proposed network is aimed at restricting the backhaul load, and hence obtaining the scalability of such a network. On the other hand, our proposed network is able to provide consistent network coverage over certain areas, like the virtual WLAN solutions. In this sense, our proposed network is called the Blanket WLAN (BWLAN). modifications of the current CSMA protocol such that the new protocol fits in the BWLAN, which is illustrated through simulations. Section 6 concludes this paper. 2. BLANKET WLAN DEFINITION Figure 1 illustrates the structure of a BWLAN, which serves a number of ENs. These ENs communicate with a set of NAPs through the wireless medium. The NAPs are interconnected through some backhaul network which could be either wired or wireless. As mentioned above, the backhaul network is critical as it must be able to support all the protocols which enable the BWLAN operations. There are several issues resulting from the decentralization of the network. The first one is on the network control, especially when the EN moves. As noted, each NAP in the BWLAN has a small service area. This leads to frequent changes of serving NAPs for a particular EN, in the case of EN movement. Unlike the virtual WLAN approach where a centralized entity takes the responsibility of handling all these changes, the NAPs in the BWLAN need to have the functionality of efficiently dealing with such changes. The second issue is data transmission in the BWLAN. Because of the dense deployment of NAPs, each EN is connected with multiple NAPs in a BWLAN. Although these NAPs serve a common EN, the Distributed Coordination Function (DCF) in the CSMA protocol of traditional WLANs causes unnecessary contentions among these NAPs. These contentions may be severe when the number of participant NAPs becomes large. In this paper, we propose some modifications on the current CSMA protocol so as to minimize the contentions among NAPs. Furthermore, our approach could take advantage of the spatial diversity to increase the transmission reliability over the traditional WLANs. The rest of the paper is organized as follows. Section 2 provides the definition of BWLAN. The architecture of BWLAN is introduced in Section 3. Section 4 discusses the BWLAN control protocols for EN mobility. The performance of these protocols is also simulated and analyzed in Section 4. In Section 5, we propose two 1 We say “appear” as these are publically available industry white papers and the information about the details of their internal solutions is limited. Figure 1: BWLAN Structure The BWLAN structure also contains centralized services and databases, as certain network operations, e.g., admission control, are best handled via a centralized gateway. Still, the major functionality associated with maintaining communication with an EN in this network is implemented in a distributed fashion. The configurations of the BWLAN are described below. Single channel operation: All NAPs in a BWLAN shall operate on the same frequency channel. Note that this does not preclude operation on multiple channels – such operation would be treated as multiple BWLANs separated in frequency. However, unlike most existing WLANs, this approach has an effective frequency re-use factor of 1. The rough equivalent for most WLANs is 3, which means that this approach has the potential of improving the spectral efficiency of the network by a factor of 3. Transparency to an EN: Given the low-coverage area of NAPs, it is inconvenient to have an EN associate with a single NAP as such associations require frequent network management operations in the case of moving EN. Instead, an EN is associated with the whole network, which acts as a single virtual network entity. This concept is in common with existing Virtual WLAN solutions —. Virtual private network for each EN: Each EN communicates with a set of NAPs. These NAPs shall create a virtual network dedicated to serving the given EN. This group of NAPs will be referred to as the EN’s “blanket.” Figure 2 illustrates 3 different layers of NAPs with respect to an EN: i). a set of NAPs in active communication with the EN (which is the precise definition of a “blanket”); ii). a set of NAPs which are able to receive from an EN but do not actively transmit to it; and iii). a set of NAPs which are aware of the EN’s presence, but are unable to receive from it. Dynamic network separation: The notion of a blanket allows the network to locate each EN and transmit to the EN only from the NAPs that can reach it. This, combined with the likely effect of low-power transmission by the EN, reduces the overall contention for the wireless medium and permits spatial re-use within the network. That is, multiple devices can transmit on the same channel at the same time without a collision – if they are far enough away from each other. Design of a network with the above features presents certain challenges. In the rest of this paper we define the general architecture of a BWLAN. In addition, we propose two simple protocols for establishing and maintaining a blanket, mainly addressing EN mobility. Finally, we illustrate the way of efficiently utilizing the NAPs to serve a common EN. 3. BWLAN ARCHITECTURE We begin by defining the architecture of the BWLAN, which addresses both the EN-NAP communication over the wireless medium and the communication over the backhaul network (cf. Figure 1). The wireless medium is assumed to utilize the same structure as existing WLAN systems (e.g., , ). As mentioned, our approach is to modify the operations of the components in the wireless medium as little as possible. The backhaul, on the other hand, shall require much more attention. In particular, we define the notion of a backhaul neighborhood, whereby the latency of general backhaul communication may be significantly higher than the latency of localized backhaul communication within a neighborhood of NAPs. This assumption reflects the topologies of certain backhaul architectures such as powerline networks and hierarchical LANs (with individual routers supporting a neighborhood and interconnected to create the complete backhaul). Furthermore, we wish to produce a design which is maximally agnostic to the details of a particular wireless communication protocol and therefore can work with a multitude of underlying wireless communication technologies. The proposed architecture is then shown in Figure 3, with its major components described below. MAC (at both EN and NAP) We shall assume the IEEE 802.11 MAC (cf. , ) for the BWLAN. NAC (Network Access Function) The NAC is effectively the MAC protocol by which NAPs communicate over the backhaul. In BWLAN, this consists of two protocols: one for gateway and the other for NAP. More details on this are provided below. Inter-Working Function (IWF) Figure 2: Relationship between EN and NAP sets The IWF allows the NAP’s MAC to hide BWLANspecific details from the ENs, e.g., use of virtual cell ID/MAC address as provided from neighborhood NAC. It also provides the capability to extract relevant information from received MAC frames for potential use by the decision making functions in the NAC entities. MAC data frames will typically pass through the IWF transparently to the higher layer entities, including the gateway. The interface between the NAP and the gateway is referred to as the X2 interface. Hence, the IWF maintains a list of ENs served by this NAP, and their corresponding blanket (virtual) MAC addresses. Neighborhood NAC Each NAP contains a neighborhood NAC function. The neighborhood NAC maintains the contact information of a list of its neighbor NAPs. It is a local NAP function interfacing with the MAC IWF and supporting a distributed control plane protocol with other NAPs via the backhaul interface. The backhaul interface between all neighborhood NACs is referred to as X1 interface. The neighborhood NAC function can be topologyspecific and adapt to the restrictions and take advantage of opportunities offered by each specific backhaul network. Gateway NAC Each EN is associated with a gateway node that maintains a forwarding table to track the set of NAPs that are aware of the EN. Hence, the gateway NAC maintains global information about the BWLAN. The gateway relays information between the set of NAPs communicating with the EN and the core network, through the X3 interface. The gateway node can be a centralized external entity as shown in Figure 3. Alternately, one of the NAPs could take the role of the gateway, acting as an anchor for forwarding data frames to/from the appropriate NAPs serving specific ENs. Figure 3: BWLAN Architecture to it. The separation of the IR state and the AT state follows from the data plane protocols, and they may be merged into a single state under certain conditions. PHYs (not shown in Figure 3) The PHY layers of both the wireless access technology and the backhaul communication technology are assumed to be unaffected by our design. 4. BWLAN CONTROL The blanket mobility function is one of the main control functionalities of the BWLAN. This function is responsible for the maintenance of a blanket, as well as the support of EN mobility and data communication between the EN and the blanket. The function is primarily located within the IWF and the neighborhood NAC and relies heavily on the X1 interface. 4.1 The State Machine For any EN in the BWLAN, the IWF within each NAP stays in one of the following states. Not Aware (NA): The NAP is unaware of the EN, i.e., it has no information about it. Active Transmit (AT): The NAP is actively transmitting to the EN. In Range (IR): The NAP is aware of the EN and is receiving from the EN, but is not actively transmitting Neighbor Range (NR): (optional) The NAP has a neighbor NAP which is in the AT state or the IR state for the EN. A blanket associated with an EN is composed of the set of NAPs for which the given EN is in the AT state. There is a unique virtual MAC address for each blanket in the BWLAN. Note that if a NAP is in the IR state with respect to an EN, it belongs to the middle set of the NAPs in Figure 2. Similarly, if a NAP is in the NR state with respect to an EN, it belongs to the outer set of the NAPs in Figure 2. Hence, the states defined above directly reflect the BWLAN nature as illustrated in Figure 2. The state transition diagram depends on the mobility protocol utilized. We propose two types of EN mobility protocols: a poll mobility protocol and a push mobility protocol. Both protocols require each NAP to maintain a list of neighbor NAPs, which are in the vicinity of the given NAP. The notion of vicinity should be tied to the physical distance, but can, in fact be left up to the system designer. Next, we propose the protocols for supporting mobility of the EN through the BWLAN. The protocols assume that the EN has some mechanism for discovery, initial access and admission into BWLAN and that these have been accomplished. Thus, these protocols deal with the steady state behavior of an EN within the BWLAN. 4.2 Poll Mobility Protocol The first protocol does not have the NR state. It operates as follows: Upon receiving a burst from an unaware EN, the IWF of a NAP uses the neighborhood NAC to poll the NAPs on its neighbor list. o o If at least one of the neighbors is aware of the EN, it responds to the NAP of interest with all the relevant information about the EN, e.g., access control parameters, QoS parameters, active connections, security parameters, etc. Then, the NAP adds the EN to the IR set or the AT set, depending on the data plane protocol. If none of the neighbor NAPs respond with the EN information, the EN is flagged as “new” and a network admission protocol needs to be utilized. Transition between the AT state and the IR state depends on data plane operation and may also be forced by the network. Figure 4: State transition diagram for EN within a NAP under the poll mobility protocol. However, our analysis shows that this protocol suffers from a drawback. The latency of the polling process is required to be low as the NAP may need to transit to the AT state rather quickly in order to provide instantaneous services. To address this we propose an alternative protocol which proactively makes the EN information available at the NAPs that might serve the EN in the near future. 4.3 Push Mobility Protocol The second protocol operates as follows: A NAP may drop an EN, i.e., transiting to the NA state, if instructed by the network. Otherwise, each transmission from the EN or to the EN successfully received by the NAP resets a timer for that EN. The timer value may depend on whether the EN is in power-saving mode. If the timer expires, the NAP sends a Range Probe to the EN to check if the EN is still within range of the NAP. If no response is received, the EN is dropped. Figure 4 shows the state transition diagram for an EN within a NAP with the poll mobility protocol. It follows from the figure that the network can exercise control over whether a NAP supports and EN or not. This protocol has several advantages. It provides for a natural way to change the set of serving NAPs. NAPs begin to serve an EN when necessary, i.e. when the EN can communicate with them, and stop service when no longer necessary. Furthermore, a significant portion of the information exchange can be limited to neighboring NAPs (using the Neighborhood NAC), which can reduce the overall backhaul load in some types of backhaul networks. Each NAP informs its neighbor NAPs of the ENs for which it is in the AT state or the IR state. The IWF provides a list of such ENs to the neighborhood NAC which manages such updates. We shall refer to this message as an Update Report from Neighbor (URN). The URN either carries all access information for each EN, including access rights and cryptography keys, or the NAP has a means of obtaining these from the network. o The URN may be issued periodically and/or it may be triggered by a status change from one of the ENs. o The contents of the URN may include the complete EN information, or only relevant changes, or mixture. Upon receiving an URN, a NAP checks the list of ENs in the message against the list of its aware ENs (i.e. those in the AT, IR or NR state). If the NAP is already aware of all the ENs in the URN, a NAP does nothing. If the URN contains an EN that it is unaware of, the NAP adds the EN to its NR set and places it in the NR state. Upon first receiving a transmission from an EN that a NAP has not previously communicated with, the NAP’s IWF checks its NR set for this EN. o If the EN is in the NR set, the IWF informs the gateway that it is ready to communicate with the EN. Based on the response from the gateway, the EN may be moved into the IR state or the AT state. o If the new EN is not in the NAP’s NR list, a fault occurs. The NAP can revert to the poll mobility protocol – it polls its neighbors before resorting to the gateway. Transition between the AT state and the IR state depends on data plane operation and may also be forced by the network. A NAP maintains a timer for each EN in the AT state or the IR state. If the timer expires, the EN is transited to the NR state from either the AT state or the IR state. A NAP may drop an EN if instructed by the network For each EN in the NR state, the NAP maintains a timer, different from the timer for the AT and IR states. If this timer expires without a transmission from the EN or to the EN, the NAP sends a Range Probe to the EN to check if the EN is still within range of the NAP. If no response is received, the EN is dropped. Figure 5: State transition diagram for EN within a NAP under the push mobility protocol The state transition diagram for the push protocol is shown in Figure 5 which also shows how the network can exercise control over whether a NAP supports and EN or not. 4.4 Performance Evaluation 4.4.1 Test Bench Description We set up a test bench to examine the performance of our proposed mobility protocols. One of the purposes is to illustrate that the BWLAN is able to handle EN movement by providing continuous connections, especially when an EN is entering or exiting the coverage area of a NAP. This feature indicates the advantage of BWLAN over traditional WLANs. As mentioned above, one advantage of our proposed distributed approach over the centralized approach (cf. -) is the relief of the backhaul traffic. Hence, our test bench shall focus on the backhaul load of BWLAN. Our test bench is built on OPNET. In the test bench, we set up a BWLAN composed of 63 NAPs uniformly distributed in a square space of 70 meters by 70 meters. The distribution of the NAPs is shown in Figure 6. The distance between two neighbor NAPs is 10 meters. A Figure 6: Floor plan of the test bench blanket gateway is resided at one of the NAPs. All these NAPs are connected with the gateway, which in turn is connected with an outside network. The connection between NAPs and gateway is wired, and all the inter-NAP communication is through the gateway. In our simulations, each inter-NAP packet is assumed of size 32 bytes, which contains the information of blanket ID, EN MAC address, etc. In the poll mobility protocol, the time-out for the transition from the AT state (or the IR state) to the NA state is set as 25 seconds. In the push mobility protocol, the time-out for the transition from the AT state (or the IR state) to the NR state is 5 seconds, while the time-out for the transition from the NR state to the NA state is 20 seconds. Each NAP in the BWLAN regards 8 closest NAPs as its neighbors. The connection between NAPs and ENs follows the IEEE 802.11g protocol, and all the NAPs operate on a common channel. The maximum transmission rate between NAPs and ENs is chosen as 24 Mbps. Since in our simulations, the receiver sensitivity is -86 dBm for data packet and -95 dBm for beacon packet, the transmission power of a NAP increases 9 dBm when it joins a blanket or begins to serve an EN. This scheme guarantees the successful data transmission when a NAP is serving an EN. On the other hand, it does not affect the blanket maintenance, as an EN could still detect beacons from NAPs beyond its serving blanket. In our simulations, the transmission power of NAPs ranges from -36.5 dBm to -29.4 dBm when it is not in a blanket, while the transmission power ranges from -27.5 dBm to -20.4 dBm otherwise. depends not only on EN’s moving speed but also on the message rate from EN to NAPs. If an EN sends messages to NAPs at a higher rate, then the receiving NAPs which are not in the blanket will send polling messages more frequently, resulting in an increased backhaul load. In our simulations, the message rate from the EN to NAPs is 26 messages/sec. The backhaul load for the poll protocol will decrease with the message rate. On the other hand, the backhaul load of the push protocol does not rely on the message rate from the EN to NAPs. Overall, the backhaul load of either protocol is small (on the magnitude of kbps), compared with the capabilities of most current networks. This implies that a BWLAN with the mobility protocols could be set up on top of most existing backhaul networks. Note that transmission power of NAPs is closely correlated with blanket size, as the larger transmission power, the larger blanket size. 4.4.2 Performance Results We first examine the case of a single EN. Suppose an EN moves within the coverage of the BWLAN, at a pedestrian speed. Its moving trajectory is shown as the line in Figure 6. Figure 7 shows the backhaul loads of our proposed mobility protocols, in terms of the moving speed of EN. In plotting this figure, the transmission power of serving NAPs is -24 dBm. This results in an average of about 12 NAPs in a blanket during the whole simulation period. To reduce the backhaul load in the poll protocol while still guaranteeing the connectivity of EN, the interval between two polling messages from a NAP is restricted to be no less than 50 ms. It is seen from the figure that the backhaul loads almost linearly increase with the moving speed of EN. This is because with an acceleration of EN, the frequency that NAPs join and exit the blanket is linearly increased, and hence the communication amount between NAPs and the gateway is linearly increased. Also, it is observed that the poll protocol results in a little bit higher backhaul load than the push protocol at high EN speeds. According to the poll protocol, the backhaul load Figure 7: The backhaul load vs. EN moving speed Figure 8 shows the effects of blanket size on the backhaul load. Here, the blanket size is configured by adjusting the transmission power of NAPs. In the simulations, the moving speed of EN is set as 0.7 m/s. As expected, the backhaul load is shown to increase with the blanket size. Next, we illustrate the backhaul load when the BWLAN is serving more than one EN simultaneously. For comparison, the speed of all ENs is fixed at 0.7 m/s and the trajectory of these ENs is predetermined. The transmission power of serving NAPs is chosen as -25.5 dBm. Figure 9 shows the backhaul loads under different number of ENs. It is seen from the figure that the backhaul load increases with the number of ENs almost linearly. This establishes the scalability of our mobility protocols. Figure 10: The connection status when the regular APs are used 5. MODIFICATIONS OF CSMA PROTOCOL FOR BWLAN Figure 8: The backhaul load vs. blanket size. Having illustrated simple protocols can solve the mobility problem in the BWLAN with minimal impact to the underlying MAC/PHY and with small load on the backhaul, we turn our attention to the actual data transmission in this network. As discussed, each EN in a BWLAN is connected with multiple NAPs. These NAPs compose a blanket for the EN. One assumption is that all these NAPs are provided with identical data that needs to be sent to the EN. Figure 9: The backhaul load vs. number of blankets We examine the simulations above and find there is not a single connection loss when ENs move inside the BWLAN. In other words, our BWLAN approach overcomes the handoff problem in traditional WLANs. Figure 10 shows the real-time transmission rate if the NAPs in Figure 6 are replaced by the regular APs defined in WLANs. It is calculated that the EN experiences connection loss in almost 53% of time, if the transmission power of NAPs is -25.5 dBm and the moving speed of EN is 0.7 m/s. Some recent work — proposed schemes to reduce the handoff period, which could relieve the amount of disconnection time. It follows from the BWLAN architecture that the wireless medium between the NAPs and the EN is WLAN. According to the DCF in the CSMA protocol of current WLANs, all the NAPs compete for the channel resource for their transmissions, even though they aim to deliver the same data to a common EN. This unnecessary contention among the NAPs results in a huge waste of channel resources and a large amount of repeated packets. Our simulations show that such a contention could significantly reduce the network throughput, when the number of NAPs in a blanket is large. Therefore, it is desirable to have the NAPs in a blanket coordinate on their transmissions. The coordination among the NAPs is easy to implement in a centralized network. A simple example is the TDMA system, where the NAPs are assigned different time slots for their transmissions. In a more sophisticated network, the NAPs could transmit at the same time in a cooperative way. Specifically, these NAPs perform as a distributed MIMO system (cf. e.g., ). The accurate synchronization among the NAPs is required to support the distributed MIMO system. Further, much instantaneous channel state information needs to be exchanged to optimize the MIMO operations. These necessitate a low latency and high date rate backhaul. However, such a requirement is not satisfied in the BWLAN, and it is even against the design purpose of BWLAN. Our approach to addressing the contention issue is by means of some modifications on the current CSMA protocol. With the modified CSMA protocol, the NAPs in a blanket could have coordinated data transmissions, rather than contentious data transmissions. Comparing with the traditional WLANs, the BWLAN with the modified CSMA protocol can take the advantage of the spatial diversity to provide more reliable transmissions. transmission and is halved after each successful transmission. It is clear that if the BWLAN applies the EIED back-off algorithm, more NAPs in a blanket will contribute to the real transmissions to the EN, in the benefit of the gentle decrease of the contention window size. This will increase the spatial diversity, and hence increment the throughput. Our modifications include a changed back-off protocol and a coordinated acknowledgement mechanism, which are discussed in the following subsections. Note that these modifications are on the NAPs only. The EN and the gateway are (almost) remained intact. With the assumption that all the NAPs in a blanket are provided with identical data to be sent to an EN, each NAP in a blanket tries to send all the data to the EN in the traditional CSMA protocol. This causes lots of redundant transmissions, and hence reduces the throughput. It should be mentioned that the contention among NAPs does not occur in the virtual WLANs, as the centralized entity of a virtual WLAN controls the actual transmissions of AP. This guarantees an EN is served by a single AP at any time. 5.1 Back-off Protocol We begin with the back-off protocol. The present IEEE 802.11 DCF utilizes a Binary Exponential Back-off (BEB) algorithm to adjust contention window size. Specifically, at the first transmission attempt, the contention window CW is set to the minimum contention window CWmin. After each unsuccessful transmission, (CW+1) is doubled until it reaches the maximum contention window CWmax. After each successful transmission, CW is reset to CWmin. However, such BEB back-off algorithm is not suitable in the BWLAN. Suppose one NAP seizes the channel and makes a successful transmission at the beginning. Other NAPs overhear the channel is busy and hence, they double their contention window size. On the other hand, the NAP with a successful transmission resets its contention window to CWmin. Subsequently, this NAP will win the next contention with a high probability due to its small contention window size. Finally, only a small portion of the NAPs in a blanket contribute to the actual transmissions to the EN. This deviates from the purpose of utilizing multiple NAPs to serve a single EN, and it reduces the diversity of BWLAN. The main reason why the BEB back-off algorithm will result in only a small portion of the NAPs in a blanket serving the EN is that the contention window is reset to CWmin on each successful transmission. This sharp decrease of contention window leads to the current transmitting NAP more competitive for the next transmission. Many alternative back-off algorithms have been proposed to achieve a better performance over BEB algorithm. One of the proposed back-off algorithms is called the Exponential Increase Exponential Decrease (EIED). According to EIED (cf. e.g., , ), the contention window size is doubled after every unsuccessful 5.2 Coordinated Acknowledgement Mechanism We propose a mechanism with very little modification on the underlying CSMA protocol, such that the throughput of EN is largely increased. This mechanism is to make use of the ACK/NACK feedback from the EN. Specifically, each NAP in a blanket pays attention to the ACK feedback from the EN, though the feedback does not correspond to a transmission from this NAP. This information helps the NAP know which packet has been received by the EN. With the availability of such information, the NAP could delete that packet from its data buffer. The redundant transmissions within a blanket are largely reduced via this way. To implement such mechanism, we should require the packet ID (e.g. sequence number) be contained in the ACK message, since otherwise the NAP cannot identify the received packet at EN. This poses a small modification on the existing CSMA protocol. 5.3 Performance Evaluation In this subsection, we shall examine the performance gain from the two modifications stated above. We apply the same simulation settings as described in Section 4.4.1, and focus on the downlink throughput. Further, the push mobility protocol is applied in the following simulations. Figure 11 shows the downlink throughput in terms of the transmission rate at NAPs. In plotting this figure, the moving speed of EN is 0.7 m/s and the transmission power of serving NAPs is -24 dBm. It is shown from the figure that the downlink throughput decreases with the transmission rate in the original CSMA protocol. This is because the increased number of redundant packets and transmission collisions crushes the whole system. On the other hand, the downlink throughput linearly increases with the transmission rate, provided that the two modifications have been applied. The actual throughput is only slightly less than the transmission rate. The downlink throughput with the modification of ACK mechanism only is also plotted in the figure. It is seen that the performance is good at low transmission rates. In the high transmission rate region, it does not perform as good Figure 11: Downlink throughput vs. transmission rate Figure 12: Downlink throughput vs. blanket size as that when the EIED back-off algorithm is applied. Figure 12 shows the downlink throughput in terms of the blanket size. The transmission rate of NAPs is set as 5.12 Mbps in the simulations. As expected, the downlink throughput decreases with the blanket size in the original CSMA protocol, while the modified CSMA protocol provides steady downlink throughput, no matter of the blanket size. Finally, we examine the downlink throughput of the whole network when there is more than one EN. In the simulations, the transmission power of serving NAPs is 25.5 dBm and the moving speed of ENs is 0.7 m/s. The transmission rate from all NAPs is fixed as 1.024 Mbps. Figure 13 shows the overall network downlink throughput, in terms of the EN number in the BWLAN. For comparison, we also plot the upper bound of the network downlink throughput in the same figure. It is seen from the figure that our BWLAN is able to achieve the throughput upper bound when it simultaneously serves a small number of ENs. Though there is a gap between our achieved network throughput and the upper bound in the condition of a large number of ENs, our achieved throughput still increases with the EN numbers. Furthermore, this gap can be reduced with an incremented density of NAPs in the BWLAN. 6. CONCLUSIONS In this paper we presented a novel approach to constructing a wireless network out of low power/low coverage access points. We noted that such a network can be effectively managed using simple distributed algorithms Figure 13: Network downlink throughput vs. number of blankets while supporting seamless mobility, taking advantage of the spatial diversity created by such a performance and retaining the key features of the CSMA protocol which forms the backbone of most modern WLANs and WPANs. Furthermore, we argues that all of this can be done while placing reasonable demand on the backhaul supporting the multiple access points. Admittedly, the results presented in this paper do not fully address the problems associated with deploying such a network. While many such problems can be addressed by the same means as those which exist today (e.g. admission control), others require further work, which we hope to present in follow-up papers. Among the latter is the challenge of addressing the overlapping blankets, and, with mobility, addressing colliding blankets. The advantages that can be derived from the potential for frequency re-use inherent in BWLAN should be quantified. And finally, and critically, more spectrally efficient cooperation techniques should be investigated. 7. REFERENCES  S. Rajagopal, et. al. “Samsung, Intel, ETRI and CSUS merged proposal text” IEEE document P802.15-090786-01-0007, Nov. 2009, available at http://mentor.ieee.org/802.15/dcn/09/15-09-0786-010007-15-7-merged-draft-text-etri-samsung-csusintel.pdf  R. Murty, J. Padhye, A. Wolman and M. Welsh, “An architecture for extensible wireless LANs,” Proc. Workshop on Hot Topics in Networks, Oct., 2008.  V. Shrivastava, N. Ahmed, S. Rayanchu, S. Banerjee, S. Keshav, K. Papagiannaki and A. Mishra, “CENTAUR: Realizing the full potential of centralized WLANs through a hybrid data path,” Proc. MobiCom, 2009.  IEEE 802.11 standard, Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, 2007.  B. Bing, Emerging technologies in wireless LANs: Theory, design, and deployment, Cambridge, 2008.  A. Mishra, M. Shin and W. Arbaugh, “An empirical analysis of the IEEE 802.11 MAC layer handoff process,” Proc. ACM SIGCOMM, pp. 93-102, 2003.  A. Maltsev, et. al. “Channel Models for 60 GHz WLAN Systems,” IEEE document 802.11-09/0334r06, Jan. 2010, available at http://mentor.ieee.org/802.11/dcn/09/11-09-0334-0600ad-channel-models-for-60-ghz-wlan-systems.doc  T. Takayama and K. Sato, “High-speed roaming method of wireless LAN,” U. S. patent 6,917,804, July 2005.  Meru Networks, “Wireless LAN Virtualization: Twice the Network at half the Cost,” available at http: //www.merunetworks.com  G. Athanasiou, T. Korakis and L. Tassiulas, “Cooperative handoff in wireless networks,” Proc. IEEE PIMRC, 2008.  Meru Networks, “From centralization to virtualization: The evolution of wireless infrastructure” available at http://www.merunetworks.com  S. Venkatesan, A. Lozano, R. Valenzuela, “Network MIMO: Overcoming Intercell Interference in Indoor Wireless Systems,” Proc. Asilomar Conf. on Signals, Systems and Computers, 2007, pp.83-87, 2007.  Extricom, “Enterprise wireless LAN and the broadband myth: The challenge of 802.11 coverage and capacity,” available at http://www.extricom.com  R. Murty, J. Padhye, R. Chandra, A. Wolman and B. Zill, “Designing high performance enterprise wi-fi networks,” Networked Systems Design & Implementation, 2008.  H. Wu, S. Cheng, Y. Peng, K. Long and J. Ma, “IEEE 802.11 distributed coordination function (DCF): Analysis and enhancement,” Proc. International Conference Communications, pp. 605—609, 2002.  N. Song, B. Kwak and L. Miller, “Enhancement of IEEE 802.11 distributed coordination function with exponential increase exponential decrease backoff algorithm,” Proc. Veh. Tech. Conf., 2003.