Aligned Beacon Transmissions to Increase IEEE s Light Sleep Mode Scalability

Transcription

1 Aligned Beacon ransmissions to Increase IEEE s Light Slee Mode Scalability Marco Porsch and homas Bauschert echnische Universität Chemnitz, Reichenhainer Str. 70, Chemnitz, Germany {marco.orsch Abstract. In wireless mesh networks battery-owered devices rely on a sohisticated ower save scheme to rovide enduring network connectivity. herefore, IEEE s defines different ower save modes to reduce the energy consumtion of Wi-Fi mesh nodes. he most common of these modes, light slee mode, aims at conserving energy while maintaining mesh link erformance. o achieve this, it relies on eriodic receits of corresonding eer nodes beacons. his aroach makes it unsuitable for dense toologies where an increasing number of eer links leads to frequent wakeus and higher energy consumtion. o overcome this issue we introduce a novel synchronization algorithm to minimize the overhead of scheduled beacon receits. We evaluate the effectiveness of the algorithm using testbed measurements and a state model-based aroach, in which we comare its erformance with resect to energy efficiency with that of the standard s synchronization algorithm. Keywords: wireless mesh networks, IEEE s, ower save, synchronization, green networking, energy efficiency 1 INRODUCION Wireless mesh networks are well-known in industry and academia but have not yet made their breakthrough in consumer electronics. Actually, wireless mesh networking would be a romising addition to today s Wi-Fi devices as it rovides increased flexibility and robustness. For examle in a smart home there is no need for additional access oints to increase the network coverage if e.g. the Wi-Fi-equied V serves as a mesh access oint to other mesh and legacy devices. Another examle are smarthone users who exchange multimedia content in a Wi-Fi hotsot; in case the hotsot and hones are mesh-enabled, the users may freely roam out of the access oint coverage without losing their connection. In its 2012 revision the IEEE standard includes the former s amendments. Note that s mesh uses regular Wi-Fi hardware as commonly found in almost any of today s smart consumer electronics. Of course, for use on battery-driven user devices there are high requirements concerning energy efficiency. herefore, s introduces different ower modes which, in the meshed environment, are set er neighbor link. Light slee ow- A. Gravey, Y. Kermarrec (Eds.) Eunice 2014 Sringer-Verlag Berlin Heidelberg 2011

2 er mode romises high energy savings on soradically used links while still maintaining roer connectivity with the corresonding neighbors. As we have shown in [4], light slee mode scales oorly with the number of eer nodes since the energy consumtion can even increase linearly in the worst-case scenario. herefore, this aer rooses an standard-comliant synchronization algorithm that reduces the effects of an increasing number of light slee eers on a node s energy consumtion. We focus on the case of an idle network with almost no user data being sent or received. For tyical user devices this case is relevant as, for examle, a smarthone rests in standby state most of the time and just eriodically checks for incoming mails. Similarly, a lato or tablet does not send or receive data while the user is reading a web age. Conserving energy in these situations is most crucial for roer battery runtime. his aer refers to the rules and definitions of the IEEE standard for wireless LAN in its 2012 revision [1]. Previously, in [2] we focused on the effects of ower save on user data forwarding on a s mesh link. Similar research has been conducted by Alam et al. in [3] using ns-2 simulations. In [4] we used testbed measurements to quantify the scalability roerties of s light slee mode. he authors of [5] highlight the imortance of information disseminated in s beacon frames and why legacy beaconing, as in ad hoc networks, is not suitable here. For synchronization numerous algorithms have been roosed in the literature for mobile ad hoc networks and wireless mesh networks. For examle, in [6] yrrell et al. derive an algorithm insired from nature where fireflies synchronize their flashing. Key contributions of this aer are testbed measurements and modelling of IEEE s ower save mode with resect to an increasing number of eers and the roosal of a standard-comliant synchronization algorithm for imroved scalability. he aer is structured as follows: section 2 gives an introduction to IEEE s synchronization and ower save mode. Section 3 exlains our novel beacon alignment synchronization algorithm before we assess its erformance in section 4, where we use results from our mesh testbed to create a model to rove the algorithm s erformance. Section 5 concludes and summarizes the aer. 2 IEEE s Synchronization and Power Save All family rotocols have a common aroach to ower save: for saving energy the radio is susended in times of no activity. With the comlex receiver baseband rocessing disabled, the device ower consumtion is reduced drastically. his makes it more useful than sole transmission ower control as it esecially allows idle nodes to conserve energy. Also, without any receiver interruts, the CPU may send more time in low ower slee states. he radio will be reactivated for frame transmission or for a scheduled receit, e.g. of a neighbor beacon. o reliably schedule doze and wakeus in any of the family ower save schemes, a tight synchronization is required. he time reference is maintained by a local SF (timing synchronization function) and by sending out beacon frames with a fixed beacon interval. he SF timer is a 64bit counter incrementing in one micro-

3 beacon interval second intervals. It determines the B (target beacon transmission time) when the counter value is a multile of the beacon interval. At each B the node schedules the beacon transmission and subsequently transmits according to the general CSMA/CA channel access rules. A beacon frame contains a timestam of the transmitter s SF at the time of transmission. It should be mentioned that, although the beacon frames are transmitted using CSMA/CA, the transmitter sets the SF timestam just once the transmission actually begins. hus, the channel access delay does not influence the accuracy of the synchronization algorithms. Aart from these similarities, the synchronization and beaconing algorithm of IEEE s mesh networks differs from that of IEEE managed mode or ad-hoc networks. First of all, in mesh mode each node has to send its own beacon. his is necessary, since in mesh mode the beacon frame carries many information elements with time-critical content that makes the mesh beacon highly dynamic comared to that of ad-hoc mode [7]. In mesh mode the algorithm used in synchronization is not strictly defined. o adat to the different use cases of wireless mesh networks, IEEE s defines an extensible synchronization framework that allows imlementing alication-secific synchronization algorithms s networks announce the use of such algorithm in beacon and robe resonse frames and only nodes suorting this method may join the MBSS (mesh basic service set). As a common fallback IEEE s defines a default synchronization method in [1] section , called neighbor offset synchronization. his method does not shift the B to a common time; instead, it focuses on keeing the arbitrary offsets between the SF and B of neighboring nodes constant. Uon receit of a beacon or robe resonse frame nodes will evaluate the contained eer s SF at transmission time t and their own SF at the time of receit r to calculate the SF offset to the resective eer. By eriodically monitoring these values, nodes can determine the maximum clock drift within the MBSS. Subsequently, all eers adjust their own SF increments to the slowest clock to mitigate further clock drift. IEEE s defines that this adjustment may only be erformed with a maximum seed of 0.04% of the beacon interval er beacon eriod to avoid that ower-saving nodes lose sync with the nodes shifting their B. With the time references synchronized, a doze/wakeu schedule can be set u according to the Bs of the neighbors towards which a node is in light slee mode. Again, the IEEE s schemes differ from those of legacy Wi-Fi, as there are multile wakeu events determined by multile eer Bs and the own beacon transawake window node A time node B time Fig. 1. Doze/wakeu scheduling of two eers in light slee mode towards each other

4 mission. Furthermore, in IEEE s the ower mode is set for each eer link individually for each direction by the resective arty. Furthermore, this ower mode is not just a binary value reresenting ower save on or off; instead, three ower modes are defined: active, light slee and dee slee mode. A link in active mode may receive ackets anytime and therefore the radio cannot be susended at all. A eer link in light slee mode will only receive ackets at certain times. his allows susending the radio while there are no receits scheduled on any link. A link in dee slee mode behaves similarly but the amount of wakeus and also the ossible erformance is further reduced. Fig. 1 illustrates the behavior of links in light slee mode; here both eers are in light slee mode towards each other. In this examle no further links are established and the doze/wakeu schedule is solely determined by the own and the single eer s B. After waking u to transmit its own beacon, a node stays awake for the awake window duration, which is advertised in an information element within the beacon. During this time interval it can receive frames from other eers. After the awake window assed, the node resumes its doze. Mesh nodes wake u just before the B of their light slee eers to receive their beacon frames and resume doze afterwards. his rocess is reeated eriodically with the beacon interval. With a link in dee slee mode on the other hand, a node will not wake u for the B of the resective eer. his increases delay on that link further and requires secial routines to maintain synchronization and for the forwarding of broadcast frames which are, for examle, required for ath discovery. As the standard leaves many arts of the behavior in dee slee mode undefined and as the erformance on these links might be severely degraded, we will focus solely on light slee mode in the remainder of this aer. 3 BEACON ALIGNMEN SYNCHRONIZAION ALGORIHM An issue of the IEEE s ower save algorithm is its scalability concerning the number of eers; with more eers more wakeus are necessary when links are in light slee mode [4]. heoretically, this should not be a big issue, as the time needed to receive a mesh beacon frame is tyically very low comared to the beacon interval. But besides the actual receive time there are additional delays due to the CSMA/CA channel access mechanism and the time for owering u the receiver s radio, rocessing the beacon and resuming doze. o minimize these wakeu overheads we roose an algorithm subsequently called beacon alignment synchronization. Via this synchronization algorithm we align the B of all MBSS members in a narrow time range to benefit from overlaing awake hases, hence keeing the number of wakeus er beacon interval minimal. In the case that all MBSS members are in direct range of one another and use a common beacon interval, subsequently only a single wakeu is required for all light slee eers. Our synchronization algorithm fits into the extensible synchronization framework that IEEE s rovides as exlained in section 2. hus, it is fully standard-comliant.

5 max min offset before A B C D A time after A B C D time Fig. 2. Behavior of beacon alignment synchronization for common beacon interval values he algorithm consists of three stes; the first ste is to determine the B offsets between all eers. With this information then the steady node can be determined, which will furthermore not shift its B. In the third ste all but the steady node will gradually shift their Bs until they are aligned with their eers. he algorithm s behavior is illustrated for an examle MBSS in Fig. 2. Here in the uer art the B timing of the MBSS is given before the beacon alignment algorithm starts, while the lower art shows the situation after the alignment rocedure is finished. he beacon transmissions at B are indicated in black while the awake hases are marked in grey. In the following we will look at the individual stes in more detail. he first ste for each node is to calculate the B offsets among all eers. o determine these, only two values are required, which can both be determined on receit of a beacon or robe resonse frame. his makes the aroach equally alicable to active or assive scanning before joining the MBSS. he first value is the beacon interval of the neighbor in microsecond units calculated from the corresonding frame field. he second value is the SF offset offset of the local node towards the eer node in microsecond units as described in the IEEE standard [1] for neighbor offset synchronization: offset is calculated from the SF timestam set by the transmitter t and the SF timestam r set by the receiver. he offset determined by a node A towards a node B, offset,ab = t,b r,a, reresents the absolute SF difference between the two nodes and allows to determine the eer s SF from the local SF. Based on and offset, node A is able to determine the B offset towards node B as follows: Boffset,AB= offset,ab mod min,a,,b As we look for offsets towards later B, which generally corresond to negative values, it is sometimes necessary to get the comlementary offset value to the next later B: min * Boffset,AB= Boffset,AB,A,,B o determine the B offset between two eers B and C, node A alies the following equation: mod min Boffset,BC= offset,ac offset,ab,b,,c

6 In the revious ste we calculated all eer s B offsets towards all other eers. Out of these only the offsets between Bs that directly follow one another are of interest, e.g. in the examle of Fig. 2 Boffset,AB and Boffset,BC. hese are the er-eer minimum B offsets. Notice that offsets to B later than the own are negative values. Hence, when comaring offsets we seak in absolute terms. With this information gathered, the node is aware of its eer s B timing as deicted in the uer illustration of Fig. 2. In the second ste the individual nodes determine the node with the largest minimum B offset among all eers this node is subsequently called the steady node. In the examle of Fig. 2 node D is the steady node, due to its large offset towards A. All but the steady node will gradually delay their B, thus slowly shifting towards the right in Fig. 2. he result of the algorithm in the given examle is shown in the lower art of Fig. 2: all MBSS members wake u only once er beacon interval, thus minimizing the wakeu overhead and energy consumtion. A minimum remaining B offset is re-configured to avoid increased contention between the aligned beacon transmissions. he key criterion for our algorithm is to minimize the time until all eer Bs are aligned. Our algorithm avoids moving the longest B offset while all other node s B move concurrently. hus the time until all beacons are aligned is minimal. For neighbor offset synchronization, IEEE mandates shifting B with a maximum of 0.04% of the beacon interval er beacon eriod. For our algorithm the maximum time to termination would be exerienced in the case of an infinite number of eers with common beacon interval and Bs distributed evenly. In this case the maximum distance shifted would be one full beacon interval. In this worst case scenario the beacon alignment synchronization would require 2500 beacon eriods until termination which translates to 512s for a beacon interval of 200U (time units of 1024µs). his may not be suitable for a short-lived mesh network of mobile nodes. o avoid this issue, our algorithm might as well be imlemented as art of the mesh joining rocedure. In this case all required inut values are acquired through active or assive scanning before the node starts beaconing itself. Based on this information the node may then choose a suitable B. he algorithm erforms similar in an MBSS of eers with different beacon interval values. IEEE mandates that the beacon intervals of MBSS members do not have to be equal, but only values to the ower of two are allowed. In this case the rules formulated in this section form multile wakeu grous deending on the minimum beacon interval. o coe with clock drift among eer nodes, the schemes from neighbor offset synchronization may be used once the alignment rocedure is finished. Node movement or limited visibility may result in beacon losses and a differing determination of the steady node. Under most circumstances this will not cause further issues excet for creating multile wakeu grous. Under rare circumstances a circular deendency may lead to ever-following B shifts. For this to haen, a quasi-circular toology is required in which the initial B distribution of the neighboring eers follows one another. his issue may be avoided by imlementing a timeout, where the node stos shifting its B after the exected maximum algorithm runtime of 2500 beacon eriods.

7 Fig. 3. Beacon alignment synchronization of seven eers with common beacon interval ongoing (50mV/div vert., 200ms/div hori.) Fig. 4. Beacon alignment synchronization of seven eers with common beacon interval terminated (50mV/div vert., 200ms/div hori.) 4 ALGORIHM EVALUAION In this section we evaluate the new synchronization algorithm described in section 3. First, we resent testbed measurements to qualitatively visualize the algorithm and to obtain real-life arameters. Second, we feed the measured values into models of the owersave algorithms to quantitatively evaluate and comare their erformance. 4.1 estbed Imlementation In order to evaluate the synchronization and ower saving algorithms of IEEE s and to obtain arameters from real-life Wi-Fi mesh equiment, we set u a testbed. We imlemented the necessary routines in the Linux kernel s Wi-Fi stack [8] in cooeration with the original authors of the oen Wi-Fi mesh imlementation, cozybit Inc. [9]. Most of the code is made oen source and included in current Linux distributions. On hardware, which emloys a soft-mac architecture where most MAC layer management functions are erformed in software, the full range of the s ower saving and synchronization schemes is available. Furthermore, we imlemented the beacon alignment algorithm using the extensible synchronization framework, which allows us to switch between our algorithm and the standard s neighbor offset synchronization. As embedded mobile mesh devices, e.g. smarthones, are not yet available, we used off-the-shelf Netgear WNDR3800 Wi-Fi routers that rovide a suitable soft-mac rotocol stack. In our setu we measured the overall device current consumtion that also includes ower sinks that are not Wi-Fi related, like CPU or Ethernet controller. he measurements were rocessed using a ektronix LS-216 digital oscilloscoe and logic analyzer. he oscilloscoe lots were taken with a samling rate of 1kS/s and a scale of 50mV/div vertical and 200ms/div horizontal. o reduce transients and noise effects an averaging factor of 10 was used. Fig. 3 shows a measurement lot taken in a setu with seven close-by eer nodes where the beacon alignment synchronization is enabled and has just started. he highest sike in current consumtion marks the device s beacon frame transmission followed by the awake window eriod configured to 100U. he smaller consumtion

8 able 1. Model Parameters Model Parameter Configured beacon interval Configured awake window AW Measured awake window duration AWO Measured eer wakeu duration P Configured residual B offset RO Value 200U 10U 16.24ms 12.12ms 4.9ms eaks are the wakeus for the eer Bs. he vertical bars serve as a reference to the beacon interval of 1000U. he area under the grah between the drawn reference bars corresonds to the current consumtion er beacon interval. Given that the network is idle, this curve reeats eriodically for each beacon interval. he mesh router of Fig. 3 has seven established eer links and all eer nodes use an equal beacon interval of 1000U. As all nodes are direct eers of each other they can equally determine the steady node. In the situation of Fig. 3 the steady node is the node whose B is marked by the vertical reference bars. All other nodes start shifting their B towards the right. he node with the highest to-shift distance is the node on the right side of the ga; it has aroximately 60% of the beacon interval to shift. hus, in the situation of Fig. 3 the algorithm will take about 1500s until it terminates with all beacons aligned. Because of the gradual movement of all but the steady node s B and the alied averaging, the current consumtion eaks are slightly deformed to a triangular shae. Fig. 4 shows the outcome of the situation of Fig. 3 after the synchronization algorithm has terminated. As all nodes use a common beacon interval, one single wakeu grou is formed. he final current consumtion deends on the osition within the grou of aligned Bs. his osition determines whether the own awake window overlas with the wakeus for eer Bs. In Fig. 4 the electrical current consumtion er beacon interval is 354,69mAs, while it had been mAs in the situation of Fig. 3. Instead of the actual seven eers of this MBSS, this corresonds to the consumtion in a scenario with only three eers. So the overall wakeu duration and energy consumtion is successfully reduced by the roosed beacon alignment synchronization algorithm. 4.2 Model-based Evaluation In this section we evaluate our beacon alignment algorithm using a model-based aroach. herefore, we use a state model consisting of the states seen in Fig. 5: doze state, awake window and eer wakeu. he doze state corresonds to the state when both the receiver and transmitter hardware are owered down and the ower consumtion is minimal. he awake window state refers to the awake duration after the beacon transmission and the corresonding wakeu overhead. he eer wakeu state corresonds to the awake duration including overhead required for waking u to receive eer beacons in light slee mode. Fig. 5 illustrates the general case for IEEE s light slee mode. Deending on the lacement of the own B and eer s

9 S 0 : doze S 1 : awake window S 2 : eer wakeu S 0 : doze S 1 : awake window + eer wakeus Fig. 5. General State model of mesh light slee mode without traffic Fig. 6. State model for the best case outcome after beacon alignment beacon transmissions the state transitions occur. i corresonds to the time share of the sojourn time in the resective state i. As we assume the system to be modelled comrehensively within these three states, the sum of their time shares is one: For this general case it is non-trivial to determine the time shares of the individual states. Hence, we chose to evaluate only the following three case scenarios: the worst case scenario of unaligned Bs without any overla, the best and the worst case after beacon alignment. We will subsequently use the arameters given in able 1. he arameters for beacon interval and awake window are the ones recommended for moderate ower save in [1] Annex W.3.4, while the measured values have been obtained from our lab setu of section 4.1. he arameter RO describes the minimum offset that is reserved between aligned Bs to avoid contention between the beacon transmissions. he value has been configured according to the value given as the average beacon frame delay in [10]. We furthermore assume that all eers are configured equally and have common characteristics. Using these models and arameters we will derive erformance bounds to comare the erformance of our beacon alignment algorithm to that of the standard s neighbor offset synchronization. Model for unaligned Bs, worst case. his model case refers to the situation that, due to the arbitrary B lacement, there are no overlaing awake hases for the current node, i.e. a eer wakeu never haens during an awake window or during another eer wakeu. For the model deicted in Fig. 5 this means that there are no state transitions between S 1 and S 2. Deending on the arameters and the number of eers N we can immediately determine the time shares of the awake window state, 1, and eer wakeu state, 2 : 2 1 N AWO N P 1

10 he model can easily be solved to determine the time share of the doze state: 0 N 1 AWO N Model for aligned Bs, worst case. his model case refers to the situation that the Bs have been aligned by our beacon alignment synchronization algorithm and we are looking at the current consumtion of the steady node, i.e. the node at the end of the wakeu grou. As this node has its awake window trailing the eer wakeus, it has the highest energy consumtion of the grou. For the model deicted in Fig. 5 this means that there is a strict order of events from S 0 to S 2 to S 1 and to S 0 again, while the transitions S 0 to S 1, S 1 to S 2 and S 2 to S 0 do not occur here. he state time shares of this model are given below. For ease of notation we use a variable time PA, which is the duration of the aligned eer wakeu. In case of a single eer, the aligned wakeu duration PA is equal to P, while all further aligned eers extend its duration by RO. 2 PA N N 1 PA AWO P 0 N 1, for N 0 RO P, for N 0 he model can be solved to determine the time share of the doze state: 0 N 1 AWO Model for aligned Bs, best case. For the case that the Bs have been aligned and we are looking at the current consumtion of first node of the grou we use a different model from the one used before. his is necessary, as the aligned eer wakeu state may be omitted in case it is comletely enclosed by the awake window. he corresonding model is illustrated in Fig. 6. Again, the state time shares sum u to one: 0 1 In this model the number of eers influences whether the eer wakeus fit into the own awake window or if the wakeu eriod is extended due to the trailing eer wakeus. he time share of the awake state in this model is as follows: max 1N 1 AWO, PA PA

11 doze state time share aligned, best case aligned, worst case unaligned, worst case number of light slee eer links Fig. 7. Doze state time share of the different models he model can be solved to determine the time share of the doze state 0 : 0 N 1 max Model comarison. With the doze state time share determined for the three different model cases, we can now comare the erformance of the beacon alignment algorithm with the standard s neighbor offset synchronization. Fig. 7 shows the doze state time share of the different models. It can be seen that the unaligned Bs of neighbor offset synchronization have the lowest doze state time share of the three models. he grah shows only the worst case with a linear decline of doze state time share. Since in neighbor offset synchronization the B lacement is comletely arbitrary, there actually is a high sread of ossible outcomes; with a small robability this scheme may erform equally well to the uer bound of beacon alignment. he worst and best case results for beacon alignment on the other hand show a higher doze state time share. Also they dislay a much smaller sread due to the imlied ordering rocess of the algorithm. he best case grah starts with a constant doze state time share until the duration of aligned eer wakeus exceeds the modelled node s awake window. he grah for the worst case results of beacon alignment exhibits an initial strong decline in doze state time share for the first eer, while the sloe is smaller for any further node. he erformance bounds of beacon alignment may be further imroved by reducing the remaining B offset RO. But this in turn may increase the contention between eer s beacon and oll frame transmissions. Still, the doze state time share dros linearly with increasing number of eers. his issue can only be avoided comletely by using dee slee mode, where nodes will not wake u for eer s Bs anymore. his in turn reduces network erformance and requires secial treatment, e.g. for broadcast forwarding. AWO, PA

12 5 Conclusion In this aer we roose a standard-comliant synchronization scheme to overcome the scalability issues of IEEE s light slee mode. Our algorithm aligns all eer Bs in a narrow time range, which allows nodes in ower save mode to wake u for minimum time to receive all eer beacons. hus, we minimize the energy consumtion overhead of waking u the radio for each eer individually. o evaluate our roosed algorithm we use both testbed measurements and a model-based aroach. Using the testbed we rovide a roof of concet and extract real-life arameters of the s ower save algorithm on off-the-shelf hardware. In the models of the ower save algorithms we use these arameters to get an estimation of their efficiency with resect to increasing number of eer nodes in dense scenarios. We comare the erformance of our new synchronization algorithm to that of the one standardized by IEEE with resect to its effect on the ratio of doze and awake state, which directly translates to energy consumtion. he results show that, although a constant doze/awake ratio with increasing number of light slee eer links cannot be achieved, the roosed algorithm successfully increases the ower save algorithm s scalability. In our future work we will address oen issues of links in dee slee state and design heuristics to control the link ower modes deending on the network conditions. 6 References 1. IEEE Std , Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Secifications. IEEE Comuter Society (May 2012) 2. M. Porsch and. Bauschert, A estbed Analysis of the Effects of IEEE s Power Save on Mesh Link Performance, EUNICE 2012, Lecture Notes in Comuter Science volume 7479,. 1 11, Sringer, (2012) 3. M.N. Alam, R. Jäntti., J. Kneckt and J. Nieminen, Performance Study of IEEE s PSM in FP-CP, Vehicular echnology Conference (VC Fall), 2012 IEEE, M. Porsch and. Bauschert, A estbed Evaluation of the Scalability of IEEE s Light Slee Mode, EUNICE 2013, Lecture Notes in Comuter Science volume 8115, , Sringer, (2013) 5. V. Vishnevsky, A. Lyakhov and A. Safonov, New Asect of Beaconing in IEEE 802.1ls Mesh Networks, Comuters and Communications, ISCC th IEEE Symosium on, A. yrrell, G. Auer and C. Bettstetter, Synchronization Insired from Nature for Wireless Meshed Networks, International Conference on Wireless Communications, Networking and Mobile Comuting WiCOM 2006, A. Safonov, A. Lyakhov and S. Sharov, Synchronization and Beaconing in IEEE s Mesh Networks. International Conference on elecommunications 2008, IC 2008, June 16-19, Official Linux Kernel, mac80211 kernel module source code htt://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/net/mac cozybit Inc., S. Mangold, S. Choi, P. May, O. Klein, G. Hiertz and L. Stibor, IEEE e Wireless LAN for Quality of Service, Proceedings of the Euroean Wireless, , 2002