Controlling Unfairness due to Physical Layer Capture and Channel Bonding in 802.11n+s Wireless Mesh Networks Sandip Chakraborty Department of Computer Science and Engineering Indian Institute of Information Technology Guwahati Guwahati 781001 India [email protected]Sukumar Nandi Department of Computer Science and Engineering Indian Institute of Technology Guwahati Guwahati 781039 India [email protected]ABSTRACT This paper analyzes the effect of physical layer capture (PLC) and channel bonding over high throughput wireless mesh backbone network built over IEEE 802.11n+s standard. With the help of PLC, a signal can be recovered from channel noise and interference if the difference between the received sig- nal strengths is more than a predefined threshold. It is well known that PLC can affect fairness during channel access by biasing towards the link with better signal strength. In this paper, we show that in the presence of channel bond- ing, high throughput links always suffer due to PLC, that ensues severe unfairness in a mesh backbone network. As a consequence, the performance of the network drops signifi- cantly from its actual capacity. This paper presents an adap- tive bonding opportunity and channel reservation scheme to mitigate from network unfairness in an IEEE 802.11n+s mesh network. The performance of the proposed scheme is analyzed and evaluated using a 13 node mesh networking testbed. It has been observed that the adaptive bonding opportunity and channel reservation performs significantly better in terms of fairness compared to the standard, and re- sults in a notable performance improvement for end-to-end flow parameters. Categories and Subject Descriptors C.2.1 [Network Architecture and Design]: Wireless Com- munication General Terms Performance, Evaluation Keywords Wireless Mesh Network, Fairness, Channel Bonding, Physi- cal Layer Capture Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ICDCN ’15 Jan. 4-7, Goa, India Copyright 2015 ACM 978-1-4503-2928-6 ...$15.00. 1. INTRODUCTION High throughput wireless mesh network [8, 25] is gaining significant attention among the wireless researchers and sys- tem developers because of its capacity to replace the wired backbone infrastructure through a wireless one. The recent standardization of wireless mesh network, termed as IEEE 802.11s [3] amendment, and the commercial products for mesh networking supports, like Cisco 1500 series mesh ac- cess points [9], have advent the commercial and commodity uses of wireless mesh networks from small-scale ad-hoc net- works [7] to enterprise backbone network [5]. IEEE 802.11n high throughput physical layer technology [26] has further exaggerated the development of wireless mesh network for commercial and enterprise use. The IEEE 802.11n high throughput physical layer supports channel bonding [11] to improve physical data rate in wireless medium, upto 600 Mbps with the combination of advanced modulation and coding schemes (MCS). Channel bonding [11, 13, 18] com- bines two consecutive 20 MHz bands to a 40 MHz band of wider channels, so that more data bits can be transmitted through spatial multiplexing. Though 40 MHz band theo- retically doubles up the physical data rate compared to a 20 MHz band, it is more susceptible to channel noise. There- fore, minimum receiver sensitivity of a 40 MHz channel (-79 dBm) is comparatively more than that of a 20 MHz channel (-82 dBm) [13]. A few recent works [4, 22, 26] have designed MCS adaptation strategies to avoid data rate versus noise trade-off in IEEE 802.11n single hop networks. However, these schemes neither solve the problem completely, nor can avoid it for multi-hop communications. In spite of its problems over multi-hop wireless networks [14, 15], IEEE 802.11n is attaining popularity for the design of high throughput wireless mesh backbone networks [17, 23, 24] due to its elevated capacity. IEEE 802.11s amend- ment [3] over wireless networking standard gives the proto- col elements for medium access control (MAC) specifications for a mesh network with IEEE 802.11 supported physical layer compatibility. A wireless station (STA) that supports mesh functionality is termed as a mesh STA. A set of mesh STAs that share a common network profile is termed as a mesh basic service set (MBSS). In a MBSS, the mesh STAs form a multi-hop communication architecture, where one or more mesh STAs, called mesh gates, connect the network to the outside Internet. The intermediate mesh STAs work as a relay to forward the traffic towards or from the mesh gates. For a mesh backbone network, the mesh STAs also
Transcript
Controlling Unfairness due to Physical Layer Capture andChannel Bonding in 802.11n+s Wireless Mesh Networks
Sandip ChakrabortyDepartment of Computer Science and
EngineeringIndian Institute of Information Technology
ABSTRACTThis paper analyzes the effect of physical layer capture (PLC)and channel bonding over high throughput wireless meshbackbone network built over IEEE 802.11n+s standard. Withthe help of PLC, a signal can be recovered from channel noiseand interference if the difference between the received sig-nal strengths is more than a predefined threshold. It is wellknown that PLC can affect fairness during channel accessby biasing towards the link with better signal strength. Inthis paper, we show that in the presence of channel bond-ing, high throughput links always suffer due to PLC, thatensues severe unfairness in a mesh backbone network. As aconsequence, the performance of the network drops signifi-cantly from its actual capacity. This paper presents an adap-tive bonding opportunity and channel reservation schemeto mitigate from network unfairness in an IEEE 802.11n+smesh network. The performance of the proposed scheme isanalyzed and evaluated using a 13 node mesh networkingtestbed. It has been observed that the adaptive bondingopportunity and channel reservation performs significantlybetter in terms of fairness compared to the standard, and re-sults in a notable performance improvement for end-to-endflow parameters.
Categories and Subject DescriptorsC.2.1 [Network Architecture and Design]: Wireless Com-munication
Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.ICDCN ’15 Jan. 4-7, Goa, IndiaCopyright 2015 ACM 978-1-4503-2928-6 ...$15.00.
1. INTRODUCTIONHigh throughput wireless mesh network [8, 25] is gaining
significant attention among the wireless researchers and sys-tem developers because of its capacity to replace the wiredbackbone infrastructure through a wireless one. The recentstandardization of wireless mesh network, termed as IEEE802.11s [3] amendment, and the commercial products formesh networking supports, like Cisco 1500 series mesh ac-cess points [9], have advent the commercial and commodityuses of wireless mesh networks from small-scale ad-hoc net-works [7] to enterprise backbone network [5]. IEEE 802.11nhigh throughput physical layer technology [26] has furtherexaggerated the development of wireless mesh network forcommercial and enterprise use. The IEEE 802.11n highthroughput physical layer supports channel bonding [11] toimprove physical data rate in wireless medium, upto 600Mbps with the combination of advanced modulation andcoding schemes (MCS). Channel bonding [11, 13, 18] com-bines two consecutive 20 MHz bands to a 40 MHz band ofwider channels, so that more data bits can be transmittedthrough spatial multiplexing. Though 40 MHz band theo-retically doubles up the physical data rate compared to a 20MHz band, it is more susceptible to channel noise. There-fore, minimum receiver sensitivity of a 40 MHz channel (−79dBm) is comparatively more than that of a 20 MHz channel(−82 dBm) [13]. A few recent works [4, 22, 26] have designedMCS adaptation strategies to avoid data rate versus noisetrade-off in IEEE 802.11n single hop networks. However,these schemes neither solve the problem completely, nor canavoid it for multi-hop communications.
In spite of its problems over multi-hop wireless networks [14,15], IEEE 802.11n is attaining popularity for the designof high throughput wireless mesh backbone networks [17,23, 24] due to its elevated capacity. IEEE 802.11s amend-ment [3] over wireless networking standard gives the proto-col elements for medium access control (MAC) specificationsfor a mesh network with IEEE 802.11 supported physicallayer compatibility. A wireless station (STA) that supportsmesh functionality is termed as a mesh STA. A set of meshSTAs that share a common network profile is termed as amesh basic service set (MBSS). In a MBSS, the mesh STAsform a multi-hop communication architecture, where one ormore mesh STAs, called mesh gates, connect the networkto the outside Internet. The intermediate mesh STAs workas a relay to forward the traffic towards or from the meshgates. For a mesh backbone network, the mesh STAs also
act as the access point (AP) for the client STAs, and for-wards the traffic from the client STAs to the mesh gatesor vice versa. The IEEE 802.11s provides a set of protocolelements, namely mesh peer management protocol (MPM),mesh access function (MAF) and mesh path selection pro-tocol for MAC layer service managements. MPM is used formesh discovery and peer establishment between two meshSTAs. MAF supports two types of channel access proto-col, the mandatory IEEE 802.11 enhanced distributed chan-nel access (EDCA) and the optional MAF controlled chan-nel access (MCCA). EDCA is a contention based protocol,where a binary exponential back-off procedure is used forcontention resolution. On the contrary, MCCA is a con-tention free reservation based channel access, where a meshSTA reserves transmission opportunities within a contentionfree access interval, called a delivery traffic indication mes-sage (DTIM) interval. The maximum amount of channelreservation in a DTIM interval is bounded by a parametertermed as MAF limit. The standard supports a hybrid pathselection protocol for mesh networks, called hybrid wirelessmesh protocol (HWMP). HWMP has a proactive as well asa reactive path selection protocol element to determine thebest path between a pair of mesh STAs.
As mentioned earlier, the set of protocols for IEEE 802.11n+sphysical and MAC layer has the ability to exaggerate the de-sign and development of a high throughput wireless meshbackbone network. This paper investigates the effect ofIEEE 802.11n physical layer protocol elements over the per-formance of IEEE 802.11s MAC layer mesh functionality.The performance is analyzed and observed from practicalindoor testbed scenarios, designed and developed at IITGuwahati research laboratories. As mesh networking is knownto be susceptible on channel interference [10, 20], the testbedenvironment employs PLC [19, 28] to restore data from in-terfered signals. In case of PLC enabled devices, if the dif-ference between the strengths of the interfered signals atthe receiver exceeds a certain threshold, the receiver cancorrectly decode and capture the packet with higher signalstrength. Lee et al [21] have shown that an one dB differenceat the receive signal strength is sufficient for successful cap-ture in Atheros adapters. However, it is well known that fora multi-hop networks, PLC can result in unfairness [16, 27].However, the existing studies have not analyzed the effectof PLC over IEEE 802.11n+s with channel bonding. Thispaper investigates inter-effect of PLC and channel bondingover a high throughput mesh network. The major contribu-tions of this paper can be summarized as follows,
1. Using practical indoor mesh testbed setups, this paperanalyzes the impact of PLC and channel bonding overthe network performance metrics. Though PLC canimprove overall network capacity, it affects networkfairness. Network unfairness shows negative impactover the end-to-end performance parameters. Further,the analysis shows that PLC always biases the capturetowards the 20 MHz channel in case of adjacent chan-nel interference, while the signal transmitted using the40 MHz channel gets dropped. As a consequence, theexisting schemes [12, 13, 22, 27] fail to recover fromunfairness in case of a high throughput mesh networkin the presence of both 20 MHz and 40 MHz channels.It can be noted that the present research, like [13], aswell as the commercial IEEE 802.11n products sup-port intelligent and dynamic channel bonding, where
the bonding opportunities are determined on the fly.Therefore in a MBSS, a mixture of 20 MHz and 40MHz supported mesh STAs is common, and thereforethis problem can not be avoided.
2. Based on the observations and analysis of the resultsfrom the testbed, this paper proposes an adaptive chan-nel reservation algorithm with channel bonding syn-chronization, so that unfairness due to PLC can bemitigated in a best effort basis. It can be noted thatseveral factors play interdependent role in the designof an optimum physical and MAC performance cri-teria, like channel bonding opportunities, selection ofa particular MCS level that determines the physicaldata rate and the MAC layer channel reservation. Anadaptive tuning mechanism is used to adjust these pa-rameters for improved network performance in termsof fairness and throughput. This adaptive mechanismis termed as high throughput fair mesh (HT-fair Mesh).
3. The performance of HT-fair Mesh has been analyzedand evaluated using a general 13 node mesh network-ing testbed. The testbed result shows that HT-fairMesh significantly improves fairness and average net-work throughput. At the same time, a significantimprovement in the end-to-end flow performance hasbeen observed for HT-fair Mesh architecture, com-pared to the standard IEEE 802.11n+s mesh, and otherexisting schemes.
The rest of the paper is organized as follows. Section 2describes the general testbed setup and the results from thetestbed to analyze the effect of interference over the perfor-mance of PLC enabled IEEE 802.11n+s networks. Section 3gives the design and architecture of the protocol elementsand rules for supporting fairness in an IEEE 802.11n+s highthroughput wireless mesh network. The performance of theproposed protocol has been evaluated in Section 4, throughthe results from the testbed. Finally, Section 5 concludesthe paper.
2. EFFECT OF INTERFERENCE OVER PLCENABLED IEEE 802.11n+s
This section gives the detailed analysis of the effect of PLCand channel bonding in an IEEE 802.11n+s network. Forthis purpose, two simple topologies are used, similar to theconventional hidden terminal problem, as shown in Figure 1.Scenario-1 shows a direct interference scenario where STA-1 and STA-3 both simultaneously transmits to STA-2, andtherefore, STA-2 is the common receiver. In Scenario-2, thecommunication from STA-3 to STA-4 creates interference atSTA-2. This scenario is termed as the indirect interferencescenario. It can be noted that these simple interference sce-narios are used in this section for analyzing the behaviorof the network under different control parameter settings.Later, we have shown the evaluation and analysis of theprotocol elements from a more generalized scenario.
2.1 Hardware SetupIn these experiments, IEEE 802.11n supported Ralink RT-
3352 router-on-chip [2] has been used as the core of the meshSTAs. The RT-3352 router-on-chip combines 802.11n draftcompliant 2T2R MAC along with BBP/PA/RF MIMO, a
STA-1 STA-2 STA-3
STA-1 STA-2STA-3 STA-4
Scenario 1: Direct Interference
Scenario 2: Indirect Interference
Link-11 Link-12
Link-21Link-22
Figure 1: Interference Scenario
high performance 400MHz MIPS24KEc CPU core, a Gi-gabit Ethernet MAC, 5-ports integrated 10/100 EthernetSwtich/PHY, 64 MB of SDRAM and 32 MB of Flash. Thischip can support up to 300 Mbps data rate with the maxi-mum transmission power of 16dB.
2.2 Software SetupEvery mesh STA is equipped with Linux kernel version
3.12.24 with IEEE 802.11 protocol stack. The mac80211submodule of the net module in the Linux kernel networkprotocol stack is patched with open source IEEE 802.11simplementation, open80211s [1]. We have used the intelli-gent channel bonding mechanism proposed by Deek et al [13]along with the IEEE 802.11n rate adaptation mechanism [22].These schemes are implemented as a loadable kernel module(LKM), where the intelligent channel bonding mechanismfirst determines the bonding opportunity based on through-put measurement. Subsequently, the rate adaptation mech-anism uses the rate upgrade or rate downgrade mechanismfor the selected 20 MHz or 40 MHz channels. It can benoted that the capture threshold for the RT-3352 chipset isapproximately 1.6 dB. The open80211s control parametersetup is given in Table 1. In the testbed, Seagull Multi-protocol Traffic Generator is used to generate traffic andmeasure performance. It can be noted that to keep thenetwork saturated and to find out the maximum networkthroughput, we have used constant bit rate traffic with hightraffic generation rate (200 Mbps).
IEEE 802.11s uses EDCA for control message commu-nication at the contention phase, that are used to reserveMCCA transmission slots, also called transmission opportu-nities (TxOpp), within a DTIM interval. EDCA employs abinary exponential back-off mechanism, where a contentionwindow is used to avoid interference during communication.For every unsuccessful communication attempt, contentionwindow is doubled up, and that many slots are waited beforeattempting for another communication. Based on the stan-dard, this contention based back-off mechanism is utilizedin the experiments for MCCA control message communi-cations. For details of EDCA and MCCA, the readers arerefereed to the IEEE 802.11 standard [3].
For all the experiments described in this section, a singleexperimental setup is executed five times, with the meanduration of execution being 2 hours, and the average is usedto plot the graphs. The deviations in the results of differentexecutions, known as confidence intervals, are also shown inthe graphs as a vertical line at every point instance.
Table 1: open80211s Control ParametersParameter Default Value
Figure 2: Throughput Performance in Direct andIndirect Interference Scenario
In the first experiment, we have evaluated the through-put performance for the direct and the indirect interferencescenarios as shown in Figure 1, in the presence of PLC, dy-namic channel bonding [13] and dynamic MCS selection [22].Figure 2 shows the throughput performance for the links inthe direct and the indirect interference scenarios. It can beseen from the figure that for the direct interference scenario,the throughput for link-11 (STA-1 → STA-2) and link-12(STA-3 → STA-2) drops several times, and results in signif-icant throughput unfairness. Further, this throughput un-fairness is not specific to a particular link. For the indirectinterference scenario, the throughput for link-22 (STA-3 →STA-4) remains almost stable, whereas the throughput forlink-21 (STA-1 → STA-2) drops several times, resulting inthroughput unfairness. It can be noted that in the indirectinterference scenario, only the link-21 gets affected due tothe interference from link-22.
In the next experiments, we have analyzed this phenomenonin a case-by-case basis to find out the probable reasons be-hind this unfairness. These cases are described in the fol-lowing subsections.
2.4 Experiment 2: Effect of PLC over channelbonding
In this experiment, the effect of PLC over throughput per-formance is evaluated in the presence and absence of channelbonding. For this purpose, we have used the direct and in-direct interference scenarios as shown in Figure 1 with staticbonding opportunities and static MCS. Two MCS are used
0
20
40
60
80
100
120
140
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Disabled, MCS 7
Link-11: STA-1 -> STA-2Link-12: STA-3 -> STA-2
0
20
40
60
80
100
120
140
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Disabled, MCS 15
Link-11: STA-1 -> STA-2Link-12: STA-3 -> STA-2
0
20
40
60
80
100
120
140
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Enabled, MCS 7
Link-11: STA-1 -> STA-2Link-12: STA-3 -> STA-2
0
20
40
60
80
100
120
140
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Enabled, MCS 15
Link-11: STA-1 -> STA-2Link-12: STA-3 -> STA-2
Figure 3: Effect of PLC over Channel Bonding: Di-rect Interference Scenario
from the available MCS set, MCS 7 and MCS 14. MCS7 uses one spatial multiple input multiple output (MIMO)stream with 64-quadratic amplitude modulation (64-QAM)at a coding rate of 5/6 that provides 72.20 Mbps at 20 MHzchannel and 150 Mbps at 40 MHz channel. On the otherhand, MCS 15 uses two MIMO spatial streams with 64-QAM at the same coding rate, that provides 144.40 Mbpsat 20 MHz and 300 Mbps at 40 MHz. It can be noted thatall the links operate at overlapping channels, and thereforeco-channel interference is common during communication 1.
For direct interference scenario as shown in Figure 1, wehave changed the channel boding configurations for the twolinks, STA-1 → STA-2 (termed as link-11) and STA-3 →STA-2 (termed as link-12), from the configuration set {(20MHz, 20 MHz),(20 MHz, 40 MHz),(40 MHz, 20 MHz),(40MHz, 40 MHz)}. At every instance, the link throughput2
is calculated which is plotted in the graphs shown in Fig-ure 3. In the figure, these four instances are shown alongthe x-axis as 20/20, 20/40, 40/20 and 40/40 respectively. Itcan be noted that IEEE 802.11s supports peer link specificconfigurations, and therefore such link specific parameterconfigurations is directly available with IEEE 802.11s MPMprotocol for peer establishment. Figure 3 shows the resultsfor four instances, with all the links operating at MCS 7and MCS 15, and with enabling and disabling PLC. It can beseen from the figure that when PLC is disabled, the through-put for the links is at per the physical data rate, that is the40 MHz link attains higher throughput compared to the 20MHz link. However, when PLC is enabled, the throughputfor 20 MHz increases significantly, but the throughput for
1In 2.4 GHz ISM band, IEEE 802.11n supports 14 primarychannels, called blocks, which are 5 MHz apart. Therefore a20 MHz channel requires 4 such consecutive blocks, whereasa 40 MHz requires 8 such consecutive blocks. As a conse-quence, co-channel interference can not be avoided in IEEE802.11n with present industry standard ISM band.2Link throughput is measured as the amount of data re-ceived per second at the MAC layer of the destination meshSTA.
40 MHz increases only when both the links use the samechannel. In a mixed scenario of 20 MHz and 40 MHz chan-nels, the link throughput for 40 MHz is almost stalled. Asdiscussed earlier, IEEE 802.11s uses EDCA along with bi-nary exponential back-off for communicating MCCA controlpackets that are used for TxOpp reservations within a DTIMinterval. In the presence of PLC, the control packets trans-mitted using 20 MHz channel gets decoded correctly at thereceiver as they require lower receiver sensitivity (−82 dBm,compared to −79 dBm for 40 MHz), and TXOpps are re-served within the DTIM interval. In every such cases, themesh STA that communicates using the 40 MHz channelsuffers from consecutive back-off due to continuous packetloss, and gradually attains a large contention window. As aresult, the mesh STA with 40 MHz channel seldom gets achance to reserve TxOpps within the DTIM interval. There-fore, we observe severe unfairness in a mixed environmentof 40 MHz and 20 MHz channels.
0
50
100
150
200
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Disabled, MCS 7
Link-21: STA-1 -> STA-2Link-22: STA-3 -> STA-4
0
50
100
150
200
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Disabled, MCS 15
Link-21: STA-1 -> STA-2Link-22: STA-3 -> STA-4
0
50
100
150
200
20/20 20/40 40/20 40/40
Th
rou
gh
pu
t (M
bp
s)
Capture Enabled, MCS 7
Link-21: STA-1 -> STA-2Link-22: STA-3 -> STA-4
0
50
100
150
200
20/20 20/40 40/20 40/40 T
hro
ug
hp
ut
(Mb
ps)
Capture Enabled, MCS 15
Link-21: STA-1 -> STA-2Link-22: STA-3 -> STA-4
Figure 4: Effect of PLC over Channel Bonding: In-direct Interference Scenario
Figure 4 shows the results from the indirect interferencescenario. The main difference of this scenario from the di-rect interference scenario is that the link STA-3 → STA-4(termed as link-22) can transmit data without any interfer-ence, although the communication through the link STA-1→ STA-2 (termed as link-21) suffers from interference. Theexperimental results from this scenario give some importantand interesting observations which are summarized next.
1. When PLC is disabled, link-22 attains better through-put compared to link-21. This is natural as link-21experiences interference whereas link-22 can transmitdata with maximum capacity. Consequently, in thescenarios when both links use different channels, the20 MHz throughput for 40/20 scenario is more com-pared to the 20 MHz throughput for 20/40 scenario,and vice-versa for the 40 MHz throughput.
2. For the 20/40 scenario, enabling PLC improves thethroughput for both 20 MHz and 40 MHz links. How-ever in case of the 40/20 scenario, enabling PLC signif-icantly improves the throughput only for the 20 MHz
link (link-22), whereas the throughput for the 40 MHzlink (link-21) drops drastically. From the traces of themesh STAs, it has been observed that, in case of inter-ference between 20 MHz and 40 MHz channels (we callthis as asymmetric channel interference) at STA-2, thepackets transmitted through the 40 MHz channel getssuffered. Consequently, STA-2 can decode the over-heard control packets from STA-3, however fails to de-code the packets from STA-1 in case of an asymmetricchannel interference.
3. For symmetric channel interference (both the inter-fering links use either 20 MHz or 40 MHz channels),the decode probability solely depends on the signalstrength. In the testbed setup, the effect of external in-terference noise is almost equal for both the links, andtherefore they achieve long term throughput fairnessaccording to the IEEE 802.11 EDCA air-time fairnessprinciple [6].
The experiments discussed in this subsection reveal that inan environment where MCS is kept fixed, PLC significantlyhampers the performance of a 40 MHz link in the presenceof asymmetric channel interference. In the next set of exper-iments, we evaluate the effect of MCS setup over throughputperformance along with channel bonding and PLC.
2.5 Experiment 3: Effect of MCS selection overPLC and channel bonding
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 20 MHz, link-12: 20 MHz
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 20 MHz, link-12: 40 MHz
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 40 MHz, link-12: 20 MHz
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 40 MHz, link-12: 40 MHz
Figure 5: Effect of MCS Selection over PLC andChannel Bonding: Direct Interference Scenario withGood Signal Quality (> −50 dBm), Maximum Ag-gregate Network Throughput is 195 Mbps (MCS 15and 40 MHz at both the links)
In this set of experiments, we have evaluated the effect ofMCS selection along with PLC and channel bonding over thethroughput performance in the network scenario shown inFigure 1. To avoid the state space explosion, the effect of sixselected MCS levels have been considered based on MIMOstreaming. For single stream MIMO, we have taken MCS 5,
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 20 MHz, link-12: 20 MHz
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 20 MHz, link-12: 40 MHz
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 40 MHz, link-12: 20 MHz
5
6
7
13
14
15
5 6 7 13 14 15
link-11: 40 MHz, link-12: 40 MHz
Figure 6: Effect of MCS Selection over PLC andChannel Bonding: Direct Interference Scenario withPoor Signal Quality ( ≤ −50 dBm), Maximum Ag-gregate Network Throughput is 113 Mbps (MCS 13and 20 MHz at both the links)
6, 7; whereas for dual stream MIMO, MCS 13, 14 and 15are used in this evaluation. These six MCS levels supportphysical data rate of 57.80 Mbps, 65.00 Mbps, 72.20 Mbps,115.60 Mbps, 130 Mbps and 144.40 Mbps, respectively, in 20MHz channel; and 120.00 Mbps, 135.00 Mbps, 150.00 Mbps,240.00 Mbps, 270.00 Mbps and 300 Mbps, respectively, in40 MHz channel. The effect of MCS selections are evaluatedfor two channel conditions - with a good signal quality (sig-nal strength is more than −50 dBm), and with poor signalquality (signal strength is less than −50 dBm). The thresh-old of −50 dBm is chosen based on the receiver sensitivity ofMCS 7 and MCS 15, which are approximately −59.4 dBmand −56.5 dBm, respectively, for RT-3352 chipset. It canbe noted that the receiver sensitivity is higher for the higherMCS levels.
Figure 5 and Figure 6 show the results of MCS selectionalong with PLC and channel bonding for the direct inter-ference scenario depicted in Figure 1. In these figures, thex-axis represents the MCS for link-11 (STA-1→ STA-2); andthe y-axis represents the MCS for link-12 (STA-3→ STA-2).Every pie in the graphs represents the aggregate throughput,where the black arc denotes the throughput share for link-11 and the gray arc shows the throughput share for link-12.The total size of the pie indicates the percentage of aggre-gate throughput with respect to the maximum throughput(195 Mbps for good signal quality and 113 Mbps for poorsignal quality). The major observations from these figuresare summarized next.
1. The maximum throughput is achieved with good sig-nal quality, when both the links use 40 MHz channeland MCS 15. However, in the case of poor signal qual-ity, maximum throughput is obtained when both thelinks use 20 MHz channel and MCS 13. As discussedearlier, 40 MHz channel is more sensitive to interfer-
ence noise. Further, higher MCS levels require higherreceiver sensitivity. As a consequence, MCS 13 at 20MHz channel performs best with poor signal quality.
2. When both the links use same channel, fairness is af-fected by MCS selection. However, higher MCS doesnot indicate higher data rate always. For instance,in Figure 6 when both the links use 20 MHz chan-nel, whereas link-11 and link-12 are configured withMCS 13 and MCS 15 respectively, link-11 attains morethroughput compared to link-12. Similar observationscan be made for Figure 5 also, even with good signalquality. As higher MCS levels require higher receiversensitivity, PLC is more biased towards lower MCSlevels in case of an interference from the higher MCSlevels. The existing works [13, 22] are built on thefact that the throughput increases or decreases unilat-erally with MCS levels for single hop 802.11n withoutPLC. However, this fact does not hold true as PLCbiases towards lower MCS levels and 20 MHz channel.Therefore, the existing schemes fail to provide fairnessin case of PLC enabled system.
3. Irrespective of signal quality and MCS selection, the 40MHz link always suffers in case of asymmetric channelinterference, as evident from Figure 5 and Figure 6.
Similar observations are made for the indirect interferencescenario, except that only the link STA-1 → STA-2 getsaffected due to the interference from the link STA-3→ STA-4. However, the results are not shown in the paper due tospace constraints.
2.6 Should PLC be avoided in high through-put mesh networks?
From the testbed results, we can conclude that PLC sig-nificantly affects network fairness in a network with mixedmode 20/40 MHz channel bonding in the presence of asym-metric channel interference. Network unfairness significantlyaffects end-to-end performance, where some of the flows mayget stalled due to unfairness problem in one of the links atthe end-to-end flow path. Automatically a question mayarise: should PLC be avoided in high throughput mesh net-works to mitigate network unfairness? The answer of thisquestion is based on two observations:
• It is evident from Figure 5 and Figure 6 that a chan-nel contention among the 20 MHz and the 40 MHzchannels always causes the wider channel to suffer,when PLC is enabled. Even during the contentionbetween packets transmitted through different MCSlevels, packets with higher MCS level may droppedduring collision, if the required signal strength is notmet for PLC.
• However, from Figure 3 and Figure 4, we can see that ifthe interfering links use same bonding configurations,then PLC significantly improves link throughput. Insome cases, we can even witness more than 100% im-provement in the network throughput with PLC.
Although PLC may reduce average network performancein a high throughput mesh network, there is an option tointelligently use channel bonding configurations at the pres-ence of PLC, such that the advantages both these technolo-gies can be utilized simultaneously. However, unfairness can
not be avoided completely only through intelligent channelbonding configurations and MCS selections. From Figure 5and Figure 6, it can be seen that high throughput links aregenerally able to transmit more packets compared to lowthroughput links, in case of symmetric channel interference.Therefore we also need to tune the channel reservation pro-tocol, so that network fairness can be ensured without af-fecting network throughput.
3. HT-Fair Mesh: A HIGH THROUGHPUTFAIR MESH PROTOCOL
In this section, the design and implementation details ofHT-Fair Mesh, a fair high throughput mesh protocol, hasbeen discussed, that intelligently selects bonding opportu-nities and channel reservations within a DTIM interval, toavoid network level unfairness. Figure 7 gives a high levelblock diagram description of the proposed protocol elementsfor the HT-Fair Mesh, according to the IEEE 802.11n+sprotocol stack. In the proposed architecture, IEEE 802.11nis used for the physical layer as well as logical link control(LLC) sublayer of the data link layer, whereas IEEE 802.11sis used for MAC layer mesh functionality supports. As dis-cussed in previous section, the mesh STAs are PLC enabled.Both the transmitter and the receiver modules maintain adatabase called information engine (InEn) which is used tostore neighbor information. We exploit the IEEE 802.11nand IEEE 802.11s control messages, like MCS feedback andMCCA setup broadcast, to collect neighbor information.
In the proposed architecture, the IEEE 802.11n modulecontains a ‘Rate/Bonding Opportunities Selection Engine’(RBOSen) that determines the bonding opportunities, datastream and MCS levels for communication. The RBOSenworks in a close loop, where both the transmitter and thereceiver determine the data rate parameters and the bond-ing opportunities collectively. For this purpose, we haveused the standard IEEE 802.11n MCS Request and MCSFeedback messages, without incorporating any extra mes-sage overhead to the system. It can be noted that, a num-ber of works [12, 22] have mentioned that close loop rateadaptation is more effective for IEEE 802.11n, comparedto open loop rate adaptation 3. At the receiver side, theRBOSen coordinates with two more submodules, called theframe monitor and channel quality estimator. The framemonitor intercepts the incoming data and control frames tocollect neighbor information, and to store it at the InEn.The channel quality estimator determines the channel qual-ity from the signal strength measurements of the incomingpackets. In the implementation, we have used diffSNR asthe channel quality estimation metric, as proposed by [12],which is the difference between the best and the worst sig-nal strength measured at the receiver interface. As shownin [12], diffSNR gives a good prediction of the best MCSlevel along with MIMO data streaming, though it can notcapture fairness requirements when PLC is enabled. The ex-perimental results from the previous section indicate that,to achieve fairness along with PLC, all the neighboring meshSTAs should use either 20 MHz or 40 MHz, but not the both.For this purpose, we use a different method to determine the
3In case of an open loop rate adaptation, the transmitteralone decides the data rate, whereas for a close loop rateadaptation, the data rate is determined cooperatively bythe transmitter and the receiver.
Rate/Bonding Opportunities
Selection Engine (RBOSen)
Rate/Bonding Opportunities
Selection Engine (RBOSen)
Frame MonitorChannel Quality
Estimator
MCCA Reservation
Engine
MCCA Reservation
Engine
MCCA Setup Request
MCCA Setup Reply
MCS Request
MCS Feedback
802.11n
802.11s
Transmitter Receiver
InEn
InEn
Figure 7: Block Diagram for the Proposed High Throughput Fair Mesh Protocol
bonding opportunities, as discussed next.In IEEE 802.11s mesh mode with MCCA as the channel
access protocol, a mesh STA reserves slot within a DTIMinterval. The time is divided into periodic intervals of con-tention interval and DTIM interval. In the contention in-terval, mesh STAs exchange control frames to set up com-munication and reserve channel with the DTIM interval.These control frames are used to exchange and negotiatecontrol parameters between peer STAs, before they start ac-tual communication. In HT-Fair Mesh architecture, similarperiodic intervals, like contention interval followed by DTIMinterval, are used for setting up mesh communications.
In this subsection, we describe the protocol rules for theMCS, data stream and bonding opportunities selection. Asmentioned earlier, this is a cooperative decision between thetransmitter and the receiver. The 802.11n parameters arechosen based on following rules;
1. In IEEE 802.11n MCS Request frame, a bit is reservedto indicate bonding opportunities, where ‘zero’ (‘0’)denotes no bonding (20 MHz), and ‘one’ (‘1’) denoteschannel bonding (40 MHz). At the beginning of acontention interval, mesh STAs first exchange IEEE802.11n control frames, followed by IEEE 802.11s con-trol frames.
2. If a mesh STA has data packets to transmit, it firstwaits for a random amount of time to overhear anyMCS Feedback frame from its neighbors. After that,it sends a MCS Request frame to the peer mesh STAwhich acts as the receiver. A transmitter mesh STAincludes a ‘zero’ in the bonding opportunity field ofthe MCS request frame (such that, it is configured touse the 20 MHz channel), if and only if all of its active
neighbors4 are configured with 20 MHz channel. Oth-erwise it includes an ‘one’ in the bonding opportunityfield of the MCS Request frame.
3. Whenever a peer STA receives a MCS Request frame,it executes one of the following steps based on the bitvalue received with the bonding opportunity field;
(a) If the bonding opportunity field carries an ‘one’,it uses the transmitter reported setting, that is 40MHz, as the bonding opportunity.
(b) If the bonding opportunity field carries a ‘zero’, itcomputes the optimal bonding opportunity basedon [13].
After that, the receiver finds out the optimal MCSlevel and MIMO streaming mode for the correspond-ing bonding opportunity, according to the diffSNR re-ported by the channel quality estimator. The compu-tation of the optimal setting for MCS level and MIMOstreaming mode is based on the scheme proposed in [12].Finally, the receiver broadcasts a MCS Feedback framewith the selected bonding opportunity along with theMCS level and MIMO streaming mode.
In the design of HT-Fair Mesh, the receiver has to agreewith a 40 MHz communication, whenever the transmitter in-cludes an ‘one’ in the bonding opportunity field of the MCSRequest frame. The reason behind this is that, a trans-mitter includes an ‘one’ in the bonding opportunity field ofthe MCS Request frame, only when at least on of its peermesh STA is configured with 40 MHz channel. Therefore,the transmitter should not transmit using 20 MHz, as itmay affect the 40 MHz communication. However, the trans-mitter is free to transmit either by 40 MHz or by 20 MHz,when all of its active peer STAs are configured with 20 MHz.In that case, the receiver decides the best bonding oppor-tunity, along with MCS level and MIMO streaming mode,
4A neighbor mesh STA is called an active neighbor if it hasalready transmitted a MCS Feedback frame.
and broadcast the information through the MCS Feedbackframe.
3.2 Tuning 802.11s parameters to support finegrained fairness
Controlling channel bonding opportunities alone is notsufficient to avoid unfairness in the network. As discussedin the previous section, unfairness is still possible because ofthe data rate differences (that again depends on the MCSlevel, MIMO streaming mode and bonding opportunity) amongthe peer STAs. To avoid such unfairness, we further tune theIEEE 802.11s channel access protocol elements. Accordingto the MCCA protocol features, a mesh STA is allowed toreserve a maximum numbers of TxOpps within a DTIM in-terval, which is bounded by a parameter called ‘MAF Limit’.However, the standard does not provide any mechanism todetermine the value of the MAF limit. In this architecture,we have dynamically tuned MAF Limit to provide fair chan-nel share to all the competing mesh STAs. After the MCS se-lection is complete, every mesh STA is aware of the selecteddata rate (this can be directly computed from the bond-ing opportunity, MIMO streaming mode and MCS levelsreceived through the MCS Feedback frame). Consequently,every mesh STA computes MAF limits dynamically basedon the physical data rate (according to selected MCS levels)and traffic load, as discussed next.
Let DS denotes the differentiated physical data rate formesh STA S, which is calculated as follows:
DS = Maximum supported physical data rate−Physical data rate determined by RBOSen + 1
Every mesh STA also maintains a parameter called TrafficLoad Estimator (TLE), which is defined as the amount ofbacklogged traffic at every mesh STA. This information canbe obtained directly from the backlogged interface queue sizeand propagated through the traffic indication map (TIM) toone-hop neighbors. It can be noted that every mesh STAperiodically broadcast beacon which contains the TIM forthat mesh STA. Let, TS denotes the TLE for mesh STA .Then,
MAF Limit for STA S =DS × TS∑
∀Si∈NS
(DSi × TSi)
where NS is the set of two-hop peer mesh STAs for STAS. The dynamic MAF limit proposed in this paper allowedevery mesh STA to reserve channel according to its datarate and traffic load. Therefore, a mesh STA configured withhigher data rate is allowed to reserve inversely proportionatenumber of TxOpps in a DTIM interval, if it contend with amesh STA configured with lower data rate.
4. HT-Fair Mesh: PERFORMANCE ANAL-YSIS AND COMPARISON
In this section, we have analyzed the performance of HT-Fair Mesh using the results from the testbed. First, we haveevaluated the performance using the direct and the indi-rect interference scenario, as shown in Figure 1. After that,the performance is evaluated over a general 13-node meshtestbed, with similar software and hardware configurationsas discussed earlier.
4.1 Performance for the direct and indirect in-terference scenario
0
20
40
60
80
100
120
2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Th
rou
gh
pu
t (M
bp
s)
Time (Hours)
Direct Interference Scenario
Link-11 (STA-1 -> STA-2)Link-12 (STA-3 -> STA-2)
0
20
40
60
80
100
120
2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Th
rou
gh
pu
t (M
bp
s)
Time (Hours)
Indirect Interference Scenario
Link-21 (STA-1 -> STA-2)Link-22 (STA-3 -> STA-4)
Figure 8: Throughput Performance for the Pro-posed Protocol for Direct and Indirect InterferenceScenarios
Figure 8 shows the throughput performance for HT-FairMesh architecture, in the direct and the indirect interfer-ence scenario, respectively. In case of direct interference,both the links attain almost similar throughput over time,and results in a long-term throughput fairness. Though thethroughputs for the two links are different in case of theindirect interference scenario, it is because of the interfer-ence effect. In indirect interference scenario, communica-tion through link-22 (STA-3 → STA-4) causes interferenceto the communication through link-21 (STA-1 → STA-3),however, link-22 experiences an interference-free communi-cation. For this reason, link-22 attains higher throughputcompared to link-21. Although, the throughput for link-21 remains steady with respect to time, and does not fadeout like the earlier scenario, as shown in Figure 2. The pro-posed scheme avoids asymmetric channel interference, whichis the main reason behind throughput fading in the 40 MHzchannel. Further, the proposed scheme improves fairnessthrough a proportional allocation of channel share based ontheir physical data rate.
STA-1
STA-2
STA-3
STA-4
STA-5
STA-6
STA-7
STA-10
STA-11
STA-8 STA-13/Gate
STA-9 STA-12
Internet
Figure 9: 802.11n+s Indoor Mesh Testbed used forEvaluation
Next, we evaluate the performance of the proposed schemefrom a 13 node 802.11n+s indoor mesh networking testbedas shown in Figure 9. The hardware and software config-urations for every node is similar to that described earlierin Section 2. The traffic configuration is also similar exceptthat in this scenario, we gradually increase the applicationlayer data generation rate to evaluate the performance bothat the saturation as well as the unsaturated traffic scenar-
ios. Every mesh STA generates a traffic flow destined forthe mesh gate (upload traffic), and the vice versa (downloadtraffic from the mesh gate to the mesh STAs).
0
20
40
60
80
100
1 10 100 1000
Av
era
ge
Ne
two
rk T
hro
ug
hp
ut
(Mb
ps
)
Traffic Generation Rate (Mbps), log scale
(a) Average Network Throughput
IEEE 802.11n+s + MinstrelDynamic MCS Selection
HT-Fair Mesh
0
0.2
0.4
0.6
0.8
1
1 10 100 1000
Fa
irn
ess
In
dex
Traffic Generation Rate (Mbps), log scale
(b) Jain Fairness Index
IEEE 802.11n+s + MinstrelDynamic MCS Selection
HT-Fair Mesh
Figure 10: Network Performance Metrics from theMesh Testbed
Figure 10 shows the average network throughput and fair-ness index calculated from the testbed traces. We have com-pared the performance of HT-Fair Mesh with the standardIEEE 802.11n+s along with Minstrel rate adaptation (thedefault rate adaptation for Linux kernel protocol stack) andthe dynamic channel bonding and MCS selection [13, 22](denoted as Dynamic MCS Selection in the graphs). Fair-ness is measured in terms of Jain Fairness Index which isdefined as;
Jain Fairness Index =(∑Fi)
2
n∑Fi
2
where Fi is the performance parameter for ith elements (incase of throughput fairness among the mesh STAs, Fi is thethroughput for ith mesh STA), and n is the total numberof such elements. Figure 10 shows that HT-Fair Mesh sup-ports significantly higher fairness compared to the standardas well as the dynamic MCS selection plus channel bondingschemes, without any loss in the average network through-put. Furthermore, the average saturation network through-put improves around 20 percent compared to the dynamicMCS selection plus channel bonding schemes.
Figure 11 compares the performance of three protocols forend-to-end performance parameters; HT-Fair Mesh, the dy-namic channel bonding and MCS selection scheme [13, 22]and the standard IEEE 802.11n+s along with Minstrel rate
adaptation. Network unfairness affect the end-to-end per-formance parameters significantly by stalling some of theintermediate links in an end-to-end flow path. As a con-sequence, after the network gets saturated, the end-to-endthroughput drops, and the delay increases, as seen from Fig-ure 11. HT-Fair Mesh improves network fairness, which inturn improves the end-to-end throughput even more than120 percent in some cases, and reduces the end-to-end de-lay. The proposed scheme overcomes the unfairness causedby PLC in case of a high throughput mesh network, and re-sults in a noteworthy performance improvement, as evidentfrom the testbed results.
5. CONCLUSIONPLC can improve network capacity by reducing the effect
of inter-channel and co-channel interference, and is widelyused in modern commodity and commercial wireless devices.n this paper, we have analyzed the impact of PLC over highthroughput wireless mesh networks. This paper shows thatPLC may results in a negative impact over high throughputmesh networks in case of asymmetric channel interference.Whenever two communication with different channel bond-ing opportunities interfere, the data transmitted through the40 MHz channel gets lost. This in turn ramify severe un-fairness in the network. We have analyzed the effect of PLCover channel bonding using testbed analysis, and reportedthe probable causes for such unfairness behaviors. Follow-ing the findings from the testbed analysis, we have designedand developed a high throughput fair mesh protocol, by aug-menting the standard IEEE 802.11n+s protocol stack, thatemploys an adaptive selection of channel bonding opportuni-ties, along with dynamic MCS and MIMO streaming modeselections. The performance of the proposed protocol hasbeen analyzed through the testbed results, and comparedwith other existing protocols from the literature.
References[1] Open80211s: Open source implementation of IEEE
802.11s for Linux kernel. URL http://www.
open80211s.org.
[2] RaLink RT3352 series IEEE 802.11n routers-on-chip.URL http://www.mediatek.com/_en/01_products/
04_pro.php?sn=1006.
[3] IEEE standard for information technology–telecommunications and information exchange betweensystems local and metropolitan area networks–specificrequirements part 11: Wireless LAN medium accesscontrol (MAC) and physical layer (PHY) specifica-tions. IEEE Std 802.11-2012 (Revision of IEEE Std802.11-2007), pages 1–2793, March 2012.
[4] M. Y. Arslan, K. Pelechrinis, I. Broustis, S. Singh,S. V. Krishnamurthy, S. Addepalli, and K. Papagian-naki. ACORN: An auto-configuration framework for802.11n WLANs. IEEE/ACM Trans. Netw., 21(3):896–909, June 2013.
[5] T. Bakıcı, E. Almirall, and J. Wareham. A smart cityinitiative: The case of Barcelona. Journal of the Knowl-edge Economy, 4(2):135–148, 2013.
[6] M. Bredel and M. Fidler. Understanding fairness andits impact on quality of service in IEEE 802.11. In Pro-ceedings of IEEE INFOCOM, pages 1098–1106. IEEE,2009.
[7] R. Bruno, M. Conti, and E. Gregori. Mesh networks:Commodity multihop ad hoc networks. IEEE Commu-nications Magazine, 43(3):123–131, 2005.
[8] L. Cerda-Alabern, A. Neumann, and P. Escrich. Ex-perimental evaluation of a wireless community meshnetwork. In Proceedings of the 16th ACM Interna-tional Conference on Modeling, Analysis & Simulationof Wireless and Mobile Systems, pages 23–30, 2013.
[10] S. M. Das, D. Koutsonikolas, Y. C. Hu, and D. Per-oulis. Characterizing multi-way interference in wirelessmesh networks. In Proceedings of the 1st InternationalWorkshop on Wireless Network Testbeds, ExperimentalEvaluation & Characterization, pages 57–64, 2006.
[11] L. Deek, E. Garcia-Villegas, E. Belding, S.-J. Lee,and K. Almeroth. The impact of channel bonding on802.11n network management. In Proceedings of theSeventh COnference on Emerging Networking EXperi-ments and Technologies, pages 11:1–11:12, 2011.
[12] L. Deek, E. Garcia-Villegas, E. Belding, S.-J. Lee, andK. Almeroth. Joint rate and channel width adaptationfor 802.11 MIMO wireless networks. In Proceedings ofthe 10th Annual IEEE Communications Society Con-ference on Sensor, Mesh and Ad Hoc Communicationsand Networks, pages 167–175, 2013.
[13] L. Deek, E. Garcia-Villegas, E. Belding, S. Lee, andK. Almeroth. Intelligent channel bonding in 802.11nwlans. IEEE Transactions on Mobile Computing, 13(6):1242–1255, June 2014.
[14] J. Friedrich, S. Frohn, S. Gubner, and C. Lindemann.Understanding IEEE 802.11n multi-hop communicationin wireless networks. In Proceedings of the IEEE Inter-national Symposium on Modeling and Optimization inMobile, Ad Hoc and Wireless Networks, pages 321–326,May 2011.
[15] S. Frohn, S. GAijbner, and C. Lindemann. Analyz-ing the effective throughput in multi-hop IEEE 802.11nnetworks. Computer Communications, 34(16):1912 –1921, 2011.
[16] S. Ganu, K. Ramachandran, M. Gruteser, I. Seskar, andJ. Deng. Methods for restoring MAC layer fairness inIEEE 802.11 networks with physical layer capture. InProceedings of the Second International Workshop onMulti-hop Ad Hoc Networks: From Theory to Reality,pages 7–14, 2006.
[17] I. W.-H. Ho, S. C. Liew, P. P. Lam, and P. H. J. Chong.Harnessing the high bandwidth of multi-radio multi-channel 802.11n mesh networks. IEEE Transactions onMobile Computing, 99:1, 2014.
[18] P. Huang, X. Yang, and L. Xiao. Adaptive channelbonding in multicarrier wireless networks. In Proceed-ings of the Fourteenth ACM International Symposiumon Mobile Ad Hoc Networking and Computing, pages297–300, 2013.
[19] J. Jeong, S. Choi, J. Yoo, S. Lee, and C.-K. Kim. Physi-cal layer capture aware MAC for WLANs. Wirel. Netw.,19(4):533–546, May 2013.
[20] T. Kim, H. Lim, and C. Lim. Exploiting multi-flowdiversity for mitigating intra-flow interference in wire-less mesh networks. In Proceedings of the 2008 ACMCoNEXT Conference, pages 57:1–57:2, 2008.
[21] J. Lee, W. Kim, S.-J. Lee, D. Jo, J. Ryu, T. Kwon, andY. Choi. An experimental study on the capture effectin 802.11a networks. In Proceedings of the Second ACMInternational Workshop on Wireless Network Testbeds,Experimental Evaluation and Characterization, pages19–26, 2007.
[22] I. Pefkianakis, S.-B. Lee, and S. Lu. Towards MIMO-aware 802.11n rate adaptation. IEEE/ACM Trans.Netw., 21(3):692–705, June 2013. ISSN 1063-6692.
[23] R. Sheshadri and D. Koutsonikolas. An experimentalstudy of routing metrics in 802.11n wireless mesh net-works. IEEE Transactions on Mobile Computing, PP(99):1–1, 2014.
[24] R. K. Sheshadri and D. Koutsonikolas. Comparisonof routing metrics in 802.11n wireless mesh networks.In Proceedings of IEEE INFOCOM, pages 1869–1877,2013.
[25] K. S. Vijayalayan, A. Harwood, and S. Karunasekera.Distributed scheduling schemes for wireless mesh net-works: A survey. ACM Comput. Surv., 46(1):14:1–14:34, July 2013.
[26] C.-Y. Wang and H.-Y. Wei. IEEE 802.11n MAC en-hancement and performance evaluation. Mob. Netw.Appl., 14(6):760–771, Dec. 2009.
[27] W. Wang, W. Ooi, and B. Leong. Mitigating unfairnessdue to physical layer capture in practical 802.11 meshnetworks. IEEE Transactions on Mobile Computing, 99(PrePrints):1, 2014.
[28] K. Whitehouse, A. Woo, F. Jiang, J. Polastre, andD. Culler. Exploiting the capture effect for collision de-tection and recovery. In Proceedings of the Second IEEEWorkshop on Embedded Networked Sensors, pages 45–52, May 2005.
