sensors Article

A Fairness Oriented Neighbor-Channel-Aware MAC Protocol for Airborne Sensor Networks Xiaolin Gao 1,2 , Jian Yan 3, * and Jianhua Lu 1 1 2 3

*

Department of Electronic Engineering, Tsinghua University, Beijing 100084, China; [email protected] (X.G.); [email protected] (J.L.) Beijing Aerospace Control Center, Beijing 100094, China Tsinghua Space Center, Tsinghua University, Beijing 100084, China Correspondence: [email protected]; Tel.: +86-010-6279-7642

Academic Editor: Mohamed F. Younis Received: 7 March 2017; Accepted: 10 May 2017; Published: 16 May 2017

Abstract: In airborne sensor networks (ASNs), the media access control (MAC) protocol faces a serious unfairness problem due to the traditional protection mechanism of air-to-air communications among aircraft. Actually, by using the binary exponential back-off algorithm at high traffic loads to minimize collisions among users, the latest successful node can always benefit from this kind of MAC to obtain channel resources. Moreover, when taking the existence of the hidden nodes in ASNs into account, the inaccurate traffic load information will further aggravate the system’s unfairness. In this paper, a neighbor-channel-aware (NCA) protocol is proposed to improve the fairness of MAC protocol in ASNs. In the proposal, the NCA frame is firstly added and exchanged between neighbor nodes periodically, which helps to resolve the inaccurate traffic load information, so as to avoid reducing the probability of successful message transmission. Then a traffic-loading based back-off algorithm is involved to make the neighbor nodes cooperatively adjust the inter-frame space (IFS) interval to further reduce the unfairness. The simulation results show that, the proposed MAC protocol can guarantee the satisfied fairness, simultaneously avoiding heavy network overloads to protect key messages’ successful transmissions in ASNs. Keywords: airborne sensor networks; media access control; fairness; neighbor-channel-aware; back-off

1. Introduction With the in depth research on wireless sensor network (WSN) applications, in recent years the application scene has gradually extended to the three-dimensional scene, and produced some new research areas, such as airborne sensor networks for air pollution monitoring [1] or wildfire detection and monitoring [2], underwater sensor networks for marine environment monitoring [3] etc. In airborne sensor networks (ASNs), for several decades now information has been shared among manned and unmanned aircraft, as well as surface and ground platforms, to provide protected air-to-air and air-to-ground communications for civil services [4]. Different from traditional WSNs, ASNs have several unique features, including large coverage areas, highly dynamic network topologies, unstable communication channels in the air, etc. One of the biggest challenges of designing ASNs is how to ensure a large number of geologically distributed nodes can simultaneously and efficiently access the shared communication channel with some notion of fairness among aircraft. With the fairness gradually becoming one of the important performance indicators for the network system, its related research has become a hot spot. In [5], a token-based media access control (MAC) protocol as well as a time division multiple access-based (TDMA-based) protocol called Token-MAC for passive radio frequency identification (RFID) systems was proposed to solve the unfairness and low throughput of the current Sensors 2017, 17, 1130; doi:10.3390/s17051130

www.mdpi.com/journal/sensors

Sensors 2017, 17, 1130

2 of 14

contention-based MAC protocols. Reference [6] evaluated the performance of several secondary network (SN) coexistence schemes in terms of fairness among the coexisting SNs. The proposed coexistence scheme could achieve throughput and energy efficiency gains, while maintaining fairness among the coexisting SNs. In Reference [7], the authors discussed the present uplink limitations due to the inherent fairness design of IEEE 802.11 distributed coordination function (DCF) by employing the carrier sense multiple access with collision avoidance (CSMA/CA) scheme with a binary exponential back-off (BEB) algorithm, and proposed the solution of offloading the uplink traffic using IP Flow Mobility (IFOM). The proposed uplink access scheme for IFOM combined weighted proportional fairness in the Wi-Fi access and price-based resource allocation in the long term evolution (LTE) upload, which improved the energy efficiency of the user equipments (UEs) and increased the offloaded data volume under the concurrent use of access technologies that IFOM allows. Moreover, due to the strict demands of timeliness and reliability of the time-sensitive information, and the existence of propagation delay in ASNs, the MAC protocol for ASNs must be optimized to meet the requirements of low latency and the end-to-end success ratio of the time-sensitive information. In previous studies on MAC protocols for WSNs, there were two categories: the scheduled MAC protocols and the random access protocols. In the scheduled MAC protocols, a node is explicitly assigned transmission time slots to ensure its Quality of Service (QoS), which can precisely control active nodes or links for transmission at any time by scheduling information [8,9]. Therefore, with the scheduled MAC each active link can obtain higher throughput. On the contrary, in the random access protocols, the WSNs can enable lower latency operations with less node coordination [10]. However, due to the lack of precise coordination, the random access MAC suffers from excessive collisions, and accordingly a kind of adaptive random access MAC protocols for traffic loads in ASNs is proposed [11–14]. The basic idea of adaptive random access MAC is to collect traffic loads in channel and wait for a random time by employing the binary exponential back-off (BEB) algorithm at high traffic loads, which helps to reduce the traffic load and minimize collisions while maintaining system stability. The BEB algorithm always favors the latest successful nodes, while keeping for other nodes under starvation, which obviously results in the short-term unfairness of the MAC [15]. Furthermore, the hidden nodes would aggravate the above unfairness [15,16]. Specially, in [14] a receiver-based time constraint model is built to help calculate precise network loadings to adaptively adjust the access threshold, but the fairness among all nodes cannot be guaranteed. In order to improve fairness among all nodes, the size of contention window of MAC layer is adjusted by collision statistics [17–21] or handshake mechanisms [22–25]. Specifically, by accurately adjusting the inter-frame space (IFS) interval in the MAC layer, a better fairness performance could be achieved [21]. Unfortunately, due to the significantly increased propagation delay caused by retransmissions under collisions, the above modified MAC cannot be directly applied in ASNs. In this paper, a neighbor-channel-aware (NCA) MAC for ASNs is proposed to improve the MAC fairness with low overhead. The highlights of the proposed protocol lie in two aspects. Firstly, by adding NCA frame and exchanging NCA between neighbor nodes periodically, all nodes could receive their neighbors’ actual channel load information (CLI). Simultaneously, the problem of the inaccurate traffic load caused by the hidden nodes also can be solved. Secondly, a novel back-off algorithm facilitates the neighbor nodes to cooperatively adjust the MAC layer IFS interval based on the traffic load. Thus the unfairness problem caused by the BEB algorithm and the hidden nodes is tentatively resolved. Furthermore, by avoiding network overloading, the end-to-end success ratio of the packets transmission among active nodes can be sufficiently improved, with decreased delays in the process of the retransmission. In our proposal, the added NCA frames and exchanging operations show an asymptotically decreasing property under a normalized protocol overhead, which is quite different from the collision statistics and handshake which are effective. In the following parts of this paper, the network model and the unfairness problem in ASNs will be firstly presented in Section 2, based on which a new MAC protocol will be proposed in detail

Sensors 2017, 17, 1130

3 of 14

Sensors 2017, 17, 1130

3 of 14

in Section 3. Then comes the simulation results in Section 4. Finally, conclusions are given in the last section. in Section 3. Then comes the simulation results in Section 4. Finally, conclusions are given in the Sensors 2017, 17, 1130 3 of 14 section. 2.last The Unfairness Problem in ASNs in Section 3. Then comes the simulation results in Section 4. Finally, conclusions are given in the last

2. The Problem in ASNs section. 2.1. The Unfairness Network Model

2.1. The Network 2. TheModel Unfairness Problem ASNs In this paper, we consider aninaeronautical ad hoc network model as described in [11], where frequency/time-hopping technology techniques are used in the physical layer to reduce the effects of In this paper, we consider an aeronautical ad hoc network model as described in [11], where 2.1. The Network Model interference and increase the overall throughput and a MAC protocol adaptive to traffic loads is frequency/time-hopping technology arehocused in the physical layer to reduce In this paper, we consider antechniques aeronautical ad network model as described in [11], where the effects applied to provide the QoS guarantee. The network topology is shown in Figure 1a. frequency/time-hopping technology techniques are used in the physical layer to reduce the effects of of interference and increase the overall throughput and a MAC protocol adaptive to traffic loads is interference and increase the overall throughput and a MAC protocol adaptive to traffic loads is applied to provide the QoS guarantee. The network topology is shown in Figure 1a. applied to provide the QoS guarantee. The network topology is shown in Figure 1a.

0

(a)

1

0

(a)

0

2 1

0

2

1

(b)(b)

(c)

1

2 2

(c)

Figure 1. Three topology scenarios: (a) High contention topology, (b) The hidden nodes topology, Figure scenarios: Figure1.1.Three Three topology scenarios: (a) (a) High High contention contention topology, topology,(b) (b)The Thehidden hiddennodes nodestopology, topology,(c) (c) topology The string topology.

The string topology. (c) The string topology.

All nodes in this topology scenario have the same function and are assumed to be randomly distributed geometrically. In each node, two kinds of packets are generated: the small number of All in topology scenario have same and are be low-latency and the larger number ofthe normal-latency packets. Then are pushedto into All nodes nodes in this thispackets topology scenario have the same function function andpackets are assumed assumed to be randomly randomly two different queues adhere to the First-In-First-Out (FIFO) principle. The packets are split into number distributed In each node, two kinds are generated: the small distributedgeometrically. geometrically. Inand each node, two kinds of of packets packets The small numberof of several bursts before being transmitted, and the transmission time and frequency (Time-Hopping low-latency packets and the larger number of normal-latency packets. Then packets are pushed into low-latency packets and the larger normal-latency Then terminal, packetswhen are pushed into and Frequency-Hopping) fornumber each burstofare randomly chosen. packets. At the receiving two queues and adhere to First-In-First-Out packets coming different transmission nodes overlap(FIFO) in termsprinciple. of time andThe frequency, burstare twodifferent differentbursts queues andfrom adhere to the the First-In-First-Out (FIFO) principle. The packets aresplit splitinto into collision is thought to take place. Figure 2 is the schematic diagram of the collision among three several bursts before being transmitted, and the transmission time and frequency (Time-Hopping several bursts before being transmitted, and the transmission time and frequency (Time-Hopping packets when they are transmitted into the channel simultaneously. Though a portion of the bursts and for eachtechnology, burst randomly chosen. At receiving thefrequency receiving terminal, when and Frequency-Hopping) for each burst areare randomly chosen. At the terminal, collide, by channel coding burst technology and asynchronous hopping, when bursts bursts fromwould different nodes overlap in ofistime andthefrequency, burstis still be transmission receivednodes correctly if the number of colliding bursts less than error collision comingcoming frompackets different transmission overlap in terms of terms time and frequency, burst correcting capability of the channel coding. collision is thought to take place. Figure 2 is the schematic diagram of the collision among thought to take place. Figure 2 is the schematic diagram of the collision among three packets whenthree they

packets when they are transmitted into the channel simultaneously. Though a portion of by thechannel bursts are transmitted into the channel simultaneously. Though a portion of the bursts collide, packet  collide, by channel burst coding technology, technology and asynchronous frequency hopping, fburst f f f f f coding technology, technology and asynchronous frequency hopping, packets would still be  t nodes m packets would still be received correctly if the number of colliding bursts is less than the error f f f f isf less than f received correctly if the numbermof colliding bursts error correcting capability of the  the t correcting capability of the channel coding. f f f f f channel coding.  1

1

2

3

1

1

2

4

5

6

n-1

5

3

3

5

3

16

4

16

n

12

8

5

1

6

8

10

2

m3

f9

2

packet nodes m1

f1 1

1

3

16

2

3

4

12

5

6



3

t

f : frequency n-1 mn: The number of bursts

fFigure f2. f5 of burst f12 collision.f8 3 16 Illustration 2

3

4

5

6



t

 t m2 set the access threshold to maintain the channel traffic loads in a stable range. On the one hand, the f3 f2 f12 f3 f9 f16 low-latency packets hand,t before transmitting the m3 should be transmitted immediately. On the other  normal-latency packets, the nodes collect bursts of channel to calculate the traffic loads and compare f : frequency it with the setting access threshold. If the value of traffic loads is lower than the access threshold, of bursts m : The number which means the current channel is not overloaded, then the normal-latency packet is transmitted immediately. Otherwise, the back-off mechanism is launched. In our proposal in this paper, we still Figure Illustration ofup burst collision. Figure 2.2.Illustration of burst collision. use the receiver-based time constraint model to build a mapping relation between channel traffic f3 ratiof16of the packets f8 f1 f10transmission f5 success In order to increase the in such an ASN, nodes would

In Inorder orderto toincrease increasethe thesuccess successratio ratioof of the the packets packets transmission transmissionin insuch suchan anASN, ASN,nodes nodeswould would set the access threshold to maintain the channel traffic loads in a stable range. On the one hand, the set the access threshold to maintain the channel traffic loads in a stable range. On the one hand, low-latency packets should be transmitted immediately. On the other hand, before transmitting the the low-latency packets should be transmitted immediately. On the other hand, before transmitting the normal-latency normal-latencypackets, packets,the thenodes nodescollect collectbursts burstsof ofchannel channelto tocalculate calculatethe thetraffic trafficloads loadsand andcompare compare itit with the setting access threshold. If the value of traffic loads is lower than the access threshold, with the setting access threshold. If the value of traffic loads is lower than the access threshold, which which means means the the current current channel channel is is not not overloaded, overloaded, then then the the normal-latency normal-latency packet packet isis transmitted transmitted immediately. immediately. Otherwise, Otherwise,the theback-off back-offmechanism mechanismisislaunched. launched.In Inour ourproposal proposalin inthis thispaper, paper,we westill still use the receiver-based time constraint model to build up a mapping relation between channel traffic use the receiver-based time constraint model to build up a mapping relation between channel traffic

Sensors 2017, 17, 1130

4 of 14

Sensors 2017, 17, 1130

4 of 14

loads and packet transmission success ratio, the calculation method for the access threshold is as described in [9]. It should be noted that the topology shown in Figure 1a is an ideal topology, in loads success ratio,transmission the calculation method for thethe access threshold is as whichand all packet nodes transmission are within each other’s range. Actually, ASNs are typically described in [9]. It should be noted that the topology shown in Figure 1a is an ideal topology, in which multi-hop networks. In this paper, the topology with the hidden nodes and the string topology all nodes are within other’s the In ASNs are typically scenarios (shown in each Figures 1b,c, transmission respectively) range. are alsoActually, considered. the dashed circle,multi-hop the nodes networks. this paper, the topology with the hidden and the string topology scenarios (shown are in theIntransmission range of each other, and arenodes called neighbor nodes. Obviously, the traffic in Figure 1b,c, respectively) are also considered. In the dashed circle, the nodes are in the transmission loads in each node’s channel is different due to the different numbers of neighbor nodes. Due to the range of each other, calledloads neighbor nodes.the Obviously, the traffic loads each node’s channel high contention in and highare offered condition, nodes have to adopt theinback-off algorithm to isadjust different dueinterval to the different neighbor nodes. Due to the high contention in high offered the IFS to retrynumbers accessingofchannels. loadsTherefore, condition, in thethe nodes have phase, to adopt back-off algorithm adjust the intervala to retry back-off thethe node adopts the BEB to algorithm toIFS compute random accessing channels. back-off time to retry an access channel, which is defined as the overload avoidance process. In cw −1) Therefore, the back-off phase, the node adopts the BEB algorithm to compute general, in theinoverload avoidance process the back-off counter is chosen in thea random range (0,back-off cw time to retryThus, an access is channel, defined as the overload avoidance process.on In general, in the randomly. definedwhich as theis contention window sizes and it depends the number of overload avoidance process the back-off counter is chosencw in the range (0,cwcw − 1) randomly. Thus, min , which is defined as overload detection of the networks. At the initial phase, is equal to cw is defined as the contention window sizes and it depends on the number of overload detection the initial contention window. Then as the overload avoidance process is performed once, cw is of the networks. At the initial phase, cw is equal to cwmin , which is defined as the initial contention n max , which is cw=2 cwmin . The packet will be doubled Then untilasitthe reaches theavoidance maximumprocess value iscw window. overload performed once, cw is doubled until it reaches the dropped after number of retry)n times of access failures. The node will access a certain maximum valuencw(the max , which is cw = 2 cwmin . The packet will be dropped after n (the number of channel and transmit the packet the back-off time counter zero and channel is the not retry) times of access failures. The when node will access a certain channelreaches and transmit thethe packet when overloaded. back-off time counter reaches zero and the channel is not overloaded. 2.2.The TheUnfairness UnfairnessProblem Problem 2.2. Weshow showthe theunfairness unfairnessproblem problemininthis thissubsection. subsection.An Anactually actuallyfair fairMAC MACprotocol protocolin inan anASN ASN We shouldhave havethe thefollowing followingproperties: properties: When loads offered by the nodes are much lower should (1)(1) When thethe loads offered by the nodes are much lower thanthan the the system capacity, which is defined the access threshold this paper, the transmission request system capacity, which is defined as theasaccess threshold in thisinpaper, the transmission request from fromnode eachshould node should (2) nodes’ When nodes’ loads exceed the system capacity, each node each be met;be(2)met; When offeredoffered loads exceed the system capacity, each node would would be able to get its fair share of the channels, i.e., the same throughput of each node in the be able to get its fair share of the channels, i.e., the same throughput of each node in the condition that condition that the loads are equally That of is, the the MAC fairness of the enables MAC protocol enablesmedia using the loads are equally offered. That is,offered. the fairness protocol using shared shared media of the evenly. this weindex use the defined resources of theresources node evenly. In node this paper, weInuse thepaper, fairness (FI)fairness definedindex in [26](FI) given by: in [26] given by: ( ∑ f i )2 2 (1) FI = (∑i 𝑖 𝑓 )
(1) 𝐹𝐼 =n ∑i f i2𝑖

𝑛(∑𝑖 𝑓𝑖2 )

where f i denotes the throughput of the flow i, and n is the total number of the flows. As the above where 𝑓𝑖 denotes the throughput of the flow i, and n is the total number of the flows. As the above definition of fairness, absolute fairness is achieved when FI = 1, while absolute unfairness is achieved definition of fairness, absolute fairness is achieved when FI = 1, while absolute unfairness is achieved when FI = 1/n. Also as in [19], the index has been averaged over all back-off time of the frames, which when FI = 1/n. Also as in [19], the index has been averaged over all back-off time of the frames, which will be considered in our simulations. In order to show the fairness effects of different MAC protocols will be considered in our simulations. In order to show the fairness effects of different MAC in ASNs, we will give here some initial numerical simulation results in the scenario shown in Figure 1. protocols in ASNs, we will give here some initial numerical simulation results in the scenario shown The results are presented in Figure 3, in which the fairness indexes are shown to decrease with the in Figure 1. The results are presented in Figure 3, in which the fairness indexes are shown to decrease increase of flow numbers. with the increase of flow numbers. 1

1

0.95

0.95

0.9

Fairness Index

Fairness Index

0.9

0.85

0.8

0.85

0.8

0.75

0.75

0.7

0.7

0.65

0.65 5

10

15

20

25

4

6

8

10

12

Hidden node numbers

Flow numbers

(a)

(b) Figure 3. Cont.

14

16

Sensors Sensors2017, 2017,17, 17,1130 1130

55ofof14 14

2

Node individual Throughput/Mbps

1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 Node0 Node2

0.2 0

4

5

6

7

8

9

10

Offered loads /Mbps

(c) Figure3.3.(a)(a) Fairness index of high contention topology, (b) Fairness hidden nodes Figure Fairness index of high contention topology, (b) Fairness index of index hiddenofnodes topology, topology, (c) Starvation phenomenon in string topology. (c) Starvation phenomenon in string topology.

Firstly, the node topology as Figure 1a is considered, with the results shown in Figure 3a. In Firstly, the node topology as Figure 1a is considered, with the results shown in Figure 3a. In fact, fact, when the offered loads of the nodes in the dashed circle (in Figure 1a) reach a higher level, the when the offered loads of the nodes in the dashed circle (in Figure 1a) reach a higher level, the nodes nodes have to use the BEB algorithm to contend the channel resources to transmit. When nodes do have to use the BEB algorithm to contend the channel resources to transmit. When nodes do not get the not get the chance to access the channel, they would be in overload avoidance process. In this paper, chance to access the channel, they would be in overload avoidance process. In this paper, P is defined 𝑃𝑡𝑟 is defined as the transmission probability of nodes, where 𝑃𝑡𝑟 = 1/(2𝑛 ∙ 𝐶𝑊𝑚𝑖𝑛 +tr 1). 𝑃𝑡𝑟 will n as the transmission probability of nodes, where Ptr = 1/(2 ·CWmin + 1). Ptr will decrease exponentially decrease exponentially along with n (the number of back-offs). In the overload avoidance process, along with n (the number of back-offs). In the overload avoidance process, the contention window the contention window (𝐶𝑊) is adjusted through the BEB algorithm to resolve the overload problem. (CW) is adjusted through the BEB algorithm to resolve the overload problem. However, because of However, because of randomness, the statistical traffic loads of the nodes may reach the access randomness, the statistical traffic loads of the nodes may reach the access threshold consecutively threshold consecutively in a short-time. The above inaccurate evaluation of traffic loads will lead to in a short-time. The above inaccurate evaluation of traffic loads will lead to the overload avoidance the overload avoidance process consecutively, thus encounter severe degradation in throughput. process consecutively, thus encounter severe degradation in throughput. Therefore, the BEB algorithm Therefore, the BEB algorithm is ineffective in resolving the overload problem in ASNs, by which the is ineffective in resolving the overload problem in ASNs, by which the short-term unfairness still keeps short-term unfairness still keeps as randomness runs through. as randomness runs through. Then, the hidden nodes topology in Figure 1b is considered. The node 0 is hidden from node 2. Then, the hidden nodes topology in Figure 1b is considered. The node 0 is hidden from node 2. Node 0 and 2 would access the channel differently in different traffic loads. For instance, node 0 Node 0 and 2 would access the channel differently in different traffic loads. For instance, node 0 would would access the channel constantly in the condition of lower traffic loads. Conversely, node 2 has to access the channel constantly in the condition of lower traffic loads. Conversely, node 2 has to contend contend to access the channel with its large number of neighbor nodes and encounter different to access the channel with its large number of neighbor nodes and encounter different degradation in degradation in throughput because of the different 𝑃𝑡𝑟 . Even if the traffic loads of node 0 and 2 are of throughput because of the different Ptr . Even if the traffic loads of node 0 and 2 are of the same level, the same level, the 𝑃𝑡𝑟 is different due to the randomness as analyzed above. When node 0 and 2 the Ptr is different due to the randomness as analyzed above. When node 0 and 2 transmit to node 1 transmit to node 1 simultaneously, the success ratio of the packets deceases intensely due to the simultaneously, the success ratio of the packets deceases intensely due to the amount of collisions in amount of collisions in node1. The related results are presented in Figure 3b with different numbers node1. The related results are presented in Figure 3b with different numbers of hidden nodes, which is of hidden nodes, which is also far from 1. also far from 1. Finally, the string topology as shown in Figure 1c is also considered. As the loads increase to a Finally, the string topology as shown in Figure 1c is also considered. As the loads increase to a certain degree, even if two dashed circles are in lower traffic loads, node 1 would suffer from certain degree, even if two dashed circles are in lower traffic loads, node 1 would suffer from starvation. starvation. As the nodes in the two dashed circles are their respective neighbor nodes, it results in As the nodes in the two dashed circles are their respective neighbor nodes, it results in the sum of the sum of statistical loads in the two dashed circles, which will always exceed the access threshold. statistical loads in the two dashed circles, which will always exceed the access threshold. This reflects This reflects the starvation phenomenon presented in Figure 3c. When the offered traffic load the starvation phenomenon presented in Figure 3c. When the offered traffic load increases to a certain increases to a certain degree, node 0 will “capture” the channel and node 1 will suffer severe degree, node 0 will “capture” the channel and node 1 will suffer severe degradation in throughput. degradation in throughput. This starvation problem would further lead to congestion when node 1 This starvation problem would further lead to congestion when node 1 is the intermediate of flow is the intermediate of flow from node 0 to 2, which might reduce system stability. So the key problem from node 0 to 2, which might reduce system stability. So the key problem in the above scenarios is in the above scenarios is that the transmitters could not receive the accurate traffic loads information that the transmitters could not receive the accurate traffic loads information in multi-hop share channel in multi-hop share channel with hidden nodes. Then the transmit nodes can only access the channel with hidden nodes. Then the transmit nodes can only access the channel blindly with local statistical blindly with local statistical loads information, due to the lack of accurate channel information loads information, due to the lack of accurate channel information recorded by the receivers. recorded by the receivers. So far, as can be seen in Figure 3, by traditional MAC protocols the FIs are always less than and So far, as can be seen in Figure 3, by traditional MAC protocols the FIs are always less than and far from 1. The problems (i.e., randomness, and inaccurate traffic loads) that cause unfairness in ASNs far from 1. The problems (i.e., randomness, and inaccurate traffic loads) that cause unfairness in have been explicitly identified. ASNs have been explicitly identified.

Sensors 2017, 17, 1130

3. The2017, Proposed Sensors 17, 1130 MAC

6 of 14

Protocol

6 of 14

In general, all nodes in WLAN use control frames by handshake mechanism to relieve the hidden nodes problem. Based on this idea, we add a neighbor-channel-aware (NCA) frame in the 3. The Proposed MAC Protocol MAC layer (which belongs to the small number of low-latency packets and could be transmitted In general, all nodes in WLANsharing use control framesthe by neighbor handshakenodes, mechanism relieve relieve the hidden directly) to realize information between whichto could the nodes problem. Based on this idea, we add a neighbor-channel-aware (NCA) frame in the MAC layer inaccurateness of the traffic loads caused by the hidden nodes. Different from the handshake (which belongs the smallbased number low-latency and frame could be directly) mechanism in toWLANs onofdata driven, packets the NCA intransmitted our proposal will toberealize sent information sharing between the neighbor nodes, which could relieve the inaccurateness of the traffic periodically. In this section, the NCA-based MAC protocol is firstly presented in detail. As loads caused by the hidden nodes. Differenthappens from the mechanism in WLANs based on mentioned before, the unfairness frequently inhandshake overload avoidance process, so we are going data driven, thetoNCA frame our proposal willavoidance be sent periodically. In NCA this section, thethe NCA-based to explain how ensure the in networks overload by using the frame in following MAC protocol is firstly presented in detail. As mentioned before, the unfairness frequently happens subsection. in overload avoidance process, so we are going to explain how to ensure the networks overload avoidance byOverload using the NCA frame in the following subsection. 3.1 Networks Avoidance The NCA-based frame contains the node ID and its local channel traffic loads information 3.1. Networks OverloadMAC Avoidance (CLI). The CLI is defined as the number of bursts which received by the nodes from the frequency The NCA-based MAC frame contains the node ID and its local channel traffic loads information hopping channel in a cycle. The whole NCA frame structure is shown in Figure 4, for instance, (CLI). The CLI is defined as the number of bursts which received by the nodes from the frequency occupying 4 bytes in this paper. hopping channel in a cycle. The whole NCA frame structure is shown in Figure 4, for instance, occupying 4 bytes in this paper. CLI

Node ID

byte 1

byte 2

byte 3

byte 4

Figure The NCA NCA frame frame structure. structure. Figure 4. 4. The

neighbornodes nodesexchange exchange NCA frames each other periodically. a node The neighbor thethe NCA frames withwith each other periodically. When aWhen node receives receives NCA frames from itsnodes, neighbor nodes, will theNCA CLI frames of the NCA frames andwhich reset the NCAthe frames from its neighbor it will read it the CLIread of the and reset its CLI its CLI which is equal to the maximum value of the CLIs (its own CLI and its neighbor’s CLI) to avoid is equal to the maximum value of the CLIs (its own CLI and its neighbor’s CLI) to avoid the inaccurate the inaccurate the hidden nodes problem. When local the node’s CLIthan is higher than the CLI caused byCLI thecaused hiddenby nodes problem. When the node’s CLI islocal higher the Max_NL Max_NL (i.e., the access threshold), it would the BEBmechanism back-off mechanism to avoid networks (i.e., the access threshold), it would launch thelaunch BEB back-off to avoid networks overload. overload. the However, the randomness would still lead to of thethe unfairness of the obtained However, randomness would still lead to the unfairness obtained channel resources.channel In this resources. In this paper, further define A_Threshold as a new access threshold: paper, we further define we A_Threshold as a new access threshold: (2) 𝐴_𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 = 𝛼 ∙ 𝑀𝑎𝑥_𝑁𝐿 A_Threshold = α· Max_NL (2) in which 𝛼 is defined as the percentage of the Max_NL. in which α is defined as the percentage the Max_NL. The MAC protocol in ASNs used of Max_NL as the only criterion for back-off, resulting in the The MAC protocol ASNs used Max_NL as the only criterion with for back-off, in the unfairness of the randominBEB algorithm. We introduce A_Threshold, the aim resulting of allowing unfairness of the random BEB algorithm. We introduce A_Threshold, with the aim of allowing the back-off time dynamically varying with the network loads among the interval [A_Threshold, back-off time dynamically with theofnetwork amongcan thebe interval [A_Threshold, Max_NL]. Max_NL]. In such a case, varying the unfairness the BEBloads algorithm meliorated by adjusting the In such value ofaα.case, the unfairness of the BEB algorithm can be meliorated by adjusting the value of α. Each node checks if the CLI exceeds the new access threshold before the transmission. If the CLI exceeds the A_Threshold, which means that there would potentially be overload condition, then the node would have to enable the back-off algorithm to increase the IFS interval as shown in the following sub-section. 3.2. TheBack-off Back-offAlgorithm Algorithm 3.2 The The The proposed proposed back-off back-off algorithm algorithm can can be be described described as as follows: follows: The The node node collects collects the the traffic traffic loads loads and and compares compares the the CLI CLI to to the the A_Threshold A_Threshold before before transmission. transmission. Case I: If CLI < A_Threshold, the node would not use a back-off algorithm. Case I: If CLI < A_Threshold, the node would not use a back-off algorithm. Case A_Threshold, the the node node would Case II: II: If If CLI CLI > > A_Threshold, would use use aa back-off back-off algorithm algorithm by by counting counting an an accurate accurate IFS interval. The transmission of the frame would not begin until the back-off counter equals IFS interval. The transmission of the frame would not begin until the back-off counter equals zero. zero.

Sensors 2017, 17, 1130 Sensors 2017, 17, 1130

7 of 14 7 of 14

Next, we define some parameters as follows: Next, we define parameters as follows: R is defined as some the average number of bursts per second sent by the nodes. R min is defined as the minimum of R, and Rmax defined R is defined as the average number of bursts perwas second sentas bythe themaximum nodes. of R. is defined as the exponential decrease of the R, and R changes in the rang [Rmin, Rmax] with the R𝜌 min is defined as the minimum of R, and Rmax was defined as the maximum of R. CLIρchanging [A_Threshold, Max_NL], is definedinasthe therang exponential decrease of the R, and R changes in the rang [Rmin , Rmax ] with the CLI changing in the rang [A_Threshold, Max_NL], So, 𝜌 can be given by: So, ρ can be given by: 𝑙𝑛𝑅𝑚𝑖𝑛 − 𝑙𝑛𝑅𝑚𝑎𝑥 𝜌 = lnRmin − lnRmax ρ= 𝐴_𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 𝑀𝑎𝑥_𝑁𝐿 A_Threshold −− Max_NL

(3) (3)

and: and: −CLI ) R = Rmax ·eρ( A_Threshold 𝑅 = 𝑅𝑚𝑎𝑥 ∙ 𝑒 𝜌(𝐴_𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑−𝐶𝐿𝐼)

(4) (4)

Thus ThusRRwould woulddecrease decreasewith withthe theincrease increaseofofCLI, CLI,which whichmeans meansthe theinterval intervalfor forsending sendingtwo two consecutive packets would increase. consecutive packets would increase. T𝑇 IFS isisdefined definedasasthe theinterval intervalfor forsending sendingtwo twoconsecutive consecutivepackets packets(also (alsobe beknown knownasasthe theIFS IFS 𝐼𝐹𝑆 interval), given by: interval), given by: (  0,0，CLI ≤ A_Threshold (5) TIFS = 1𝐶𝐿𝐼 ≤ 𝐴_𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 1 − 1 , CLI > A_Threshold T 𝑇𝐼𝐹𝑆 = { f R1 Rmax (5) 𝑇 ( − ) ， 𝐶𝐿𝐼 > 𝐴_𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 𝑓

𝑅

𝑅

T f is defined as the transmission time of a𝑚𝑎𝑥 frame, which is: 𝑇𝑓 is defined as the transmission time of a frame, which is: T f = L f /R0 ′ 𝑇𝑓 = 𝐿𝑓 /𝑅

(6) (6)

T𝑇 defined asas thethe length of of thethe frame in bits, andand R0 is𝑅′defined as the rate (bit/s), so byso f 𝑓isis defined length frame in bits, is defined astransmit the transmit rate (bit/s), controlling the T , node could control the average number of transmission bursts per second. IFS 𝑇 , node could control the average number of transmission bursts per second. by controlling the 𝐼𝐹𝑆 Figure 5 shows Figure 5 showsthe theexponential exponentialincrease increaseofofthe theIFS IFSinterval intervalalong alongwith withthe thetraffic trafficload loadincrease, increase, under the back-off parameters shown in Table 1. The IFS interval (the time of waiting to access under the back-off parameters shown in Table 1. The IFS interval (the time of waiting channel) to access would be longer the larger of α in the same loadstraffic condition. channel) wouldwith be longer withvalue the larger value of α intraffic the same loads condition. 0.1 alpha=80% alpha=70% alpha=60% alpha=50%

0.09 0.08

IFS interval /s

0.07 0.06 0.05 0.04 0.03 0.02 0.01 0

2

2.2

2.4

2.6

2.8

Traffic load / pulse/s

3

3.2 5

x 10

Figure5.5.IFS IFSinterval intervalvs. vs.the thetraffic trafficload. load. Figure Table1.1.NCA-based NCA-based Back-off Parameters. Table Back-off Parameters. Parameters Parameters R_max R_max R_min R_min Max_NL Max_NL α α

Value Value 64,000 bursts/s 64,000 bursts/s 60 bursts/s 60 bursts/s 320,000 bursts/s 320,000 bursts/s 80%, 70%, 60%, 50% 80%, 70%, 60%, 50%

Sensors 2017, 17, 1130 Sensors 2017, 17, 1130

8 of 14 8 of 14

The nodes nodes will will access access the the frequency-hopping frequency-hopping channels channels distributively distributively by by the the above above real-time real-time The back-off algorithm, thereby avoiding the short-time unfairness of the BEB algorithm caused by the back-off algorithm, thereby avoiding the short-time unfairness of the BEB algorithm caused by randomness. the randomness. 3.3. Analysis Analysis of of NCA’s NCA’sOverhead Overheadand andFairness Fairness 3.3. 3.3.1. Protocol Overhead Analysis 3.3.1 Protocol In thethe useuse of NCA frames obviously indicates extra overhead increases in the original Inour ourproposal, proposal, of NCA frames obviously indicates extra overhead increases in the MAC protocol. Since theSince NCA the frame is only periodically sent among neighbor thenodes, overhead original MAC protocol. NCA frame is only periodically sent among nodes, neighbor the size can besize decided bydecided the nodeby number andnumber the periodic time (cycle) time of the(cycle) NCA frame, but without overhead can be the node and the periodic of the NCA frame, relation to therelation traffic load. theory, theIn protocol is proportional the node number and but without to theIn traffic load. theory,overhead the protocol overhead is to proportional to the node Overhead   N , T ) the periodic the NCA frame, i.e., Overhead ( Nnode , Tcycle ). Thenodefunctional relationship NCA cycle NCA ∝ number andtime the ofperiodic time of the NCA frame, i.e., . The functional between the normalized protocol overhead (overhead/traffic load) and the traffic load can be shown relationship between the normalized protocol overhead (overhead/traffic load) and the traffic load in Figure 6 by numerical results. can be shown in Figure 6 by numerical results. Cycle1:10ms Cycle2:5ms Cycle3:2.5ms

0.8

Normalize overhead

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

1000

2000

3000

4000

5000

6000

7000

8000

9000 10000

traffic loads /kbps

Figure 6. 6. Normalized Normalized overhead overhead of of NCA-based NCA-based MAC. MAC. Figure

In Figure 6, it can be seen that, under the same traffic load conditions, the greater the cycle, the In Figure 6, it can be seen that, under the same traffic load conditions, the greater the cycle, less normalized overhead. When the traffic load increases, the corresponding normalized overhead the less normalized overhead. When the traffic load increases, the corresponding normalized overhead decreases. Take the 10 ms cycle as an example, as the traffic load is increased to 2 Mbps, the decreases. Take the 10 ms cycle as an example, as the traffic load is increased to 2 Mbps, the normalized normalized protocol overhead will be reduced less than 9%, which would even drop further along protocol overhead will be reduced less than 9%, which would even drop further along with the increase with the increase of traffic load. Therefore, we can conclude that, by the NCA-based MAC, the of traffic load. Therefore, we can conclude that, by the NCA-based MAC, the protocol overhead shows protocol overhead shows asymptotically decreasing property in ASNs. asymptotically decreasing property in ASNs. 3.3.2. Fairness Fairness Analysis Analysis 3.3.2. We can can also alsoanalyze analyze the thefairness fairnessprovided providedby bythe theNCA-based NCA-basedMAC MAChere. here. Define Define the the channel channel We T frame T f rame utilization of the nodes as T + TIFS , where T f rame is the actual sending time. According to the NCA f rame T frame  TIFS T frame utilization of the nodes as , where is the actual According the. frame for the accurate channel traffic load information, the node will sending calculate time. the precise IFS bytoTIFS NCA for the accurate channel3.2, traffic the traffic preciseload. IFS By the frame calculation process in Section eachload nodeinformation, could obtainthe thenode samewill TIFScalculate in the same T T When the frames are in the same length, the channel utilization of each node is equal, which means IFS IFS by . By the calculation process in Section 3.2, each node could obtain the same in the same that theload. fairness of the enables media resources among allofthe nodes uniformly. traffic When theprotocol frames are in theusing sameshared length, the channel utilization each node is equal, which means that the fairness of the protocol enables using shared media resources among all the 4. Simulations nodes uniformly. In this section, we are going to present the simulation experiments by using opnet14.5 network 4. Simulations simulation software in which the seed parameter in each simulation scenario is set to 1000 (that is, the number of samples is 1000), we compare the protocol performance of theopnet14.5 NCA-based and In this section, we are goingand to present the simulation experiments by using network the traditional BEB-based MAC schemes. The receiving range of the nodes is set simulation software in which the seed parameter in each simulation scenario is as setits to transmission 1000 (that is,

the number of samples is 1000), and we compare the protocol performance of the NCA-based and the traditional BEB-based MAC schemes. The receiving range of the nodes is set as its transmission

Sensors 2017, 17, 1130

9 of 14

Sensors 2017, 17, 1130

9 of 14

range. The physical parameter is based on those of the frequency-hopping channel simulations in [14] range. The physical parameter is based the frequency-hopping simulations in throughout this section, which are shownon in those Table of 2. The NCA-based back-offchannel parameters are set as [14] throughout section, in Table 2. The NCA-based back-off parameters are Table 1 in Sectionthis 3. The cyclewhich time isare setshown as 10 ms. The BEB algorithm parameters refer to the IEEE set as Table 1 in Section 3. The cycle time is set as 10 ms. The BEB algorithm parameters refer to the 802.11 DCF [19,20]. The CWmin is chosen as 15 and it has been retried for 5 times. IEEE 802.11 DCF [19,20]. The CWmin is chosen as 15 and it has been retried for 5 times. Table 2. Physical simulation parameters. Table 2 Physical simulation parameters. Parameters Parameters Value Value Node transmission rate rate 2 Mbps/64 000 bursts/s bursts/s Node transmission 2 Mbps/64,000 Packet length 1024 bits1024 bits Packet length Number of bursts per packet 32 Number of bursts per packet 32 Number of frequency 16 Number of frequency 16 Error correction coding 1/3Turbo code If the number of number bursts received per received packet is no than 18, it can be Error correction coding 1/3Turbo code If the of bursts perless packet is no less successfully coding. recovered by turbo coding. thanrecovered 18, it canby beturbo successfully Access threshold 10 Mbps/320 000 bursts/s bursts/s Access threshold 10 Mbps/320,000

11

10

10

9

Average Throughput / Mbps

Average Throughput / Mbps

Before discussing the fairness performance of the protocol, we present the simulation Before discussing the fairness performance of the protocol, we present the simulation experiments experiments to compare the average throughput of NCA-based with different value of α and to compare the average throughput of NCA-based with different value of α and BEB-based MAC BEB-based MAC schemes. In this simulation scenario, nodes are randomly distributed in a 150 km × schemes. In this simulation scenario, nodes are randomly distributed in a 150 km × 150 km plane area 150 km plane area and the number of nodes is 100. The NCA frame’s cycle equals to 10 ms, so the and the number of nodes is 100. The NCA frame’s cycle equals to 10 ms, so the protocol overhead could protocol overhead could be calculated, which equals to 0.32 Mbps. Figure 7 shows the average be calculated, which equals to 0.32 Mbps. Figure 7 shows the average throughput of the NCA-based throughput of the NCA-based with different value of α and BEB-based MAC schemes. with different value of α and BEB-based MAC schemes.

9 8 7 6 5 BEB-based NCA-based

4 3

0

20

40

60

80

Offered loads / Mbps

(a)

100

8

7

6

5 NCA:alpha=80% NCA:alpha=70% NCA:alpha=60% NCA:alpha=50%

4

120

3 0

20

40

60

80

100

120

Offered loads / Mbps

(b)

Figure7.7.(a)(a)Average Average throughput of protocols, two protocols, (b) Average throughput of NCA-bases with Figure throughput of two (b) Average throughput of NCA-bases with different α. different α.

In of of thethe NCA-based MAC with the the α = 80% is lower than than that InFigure Figure7a, 7a,the theaverage averagethroughput throughput NCA-based MAC with α = 80% is lower of the BEB-based due to the protocol overhead. Nevertheless, the difference approximately equals that of the BEB-based due to the protocol overhead. Nevertheless, the difference approximately to the corresponding overhead, which iswhich far less thethan maximum channel channel loads of loads the system. equals to the corresponding overhead, is than far less the maximum of the In Figure 7b, the different of the average throughput of NCA-based with different value of α is very system. In Figure 7b, the different of the average throughput of NCA-based with different value of α small, so the value of α would not affect throughput performance, especially in networks overload is very small, so the value of α would not affect throughput performance, especially in networks conditions. In following simulation experiments, the valuethe of αvalue is setof asα80%. FIuse as overload conditions. In following simulation experiments, is setWe as still 80%.use Wethe still (1) given in Section 2.2. The indicators of measuring the performance of the protocol include the FI, the FI as (1) given in Section 2.2. The indicators of measuring the performance of the protocol include the and the end-to-end success ratio.success To be specific, thebeend-to-end success ratio is theaverage FI, the throughput average throughput and the end-to-end ratio. To specific, the end-to-end defined as the total number of transmitted packets divided by the number of successfully received success ratio is defined as the total number of transmitted packets divided by the number of packets, which is one of the most important in ASNs. Then indexes we check MAC performance successfully received packets, which is one indexes of the most important inthe ASNs. Then we check under different topologies, including fully-connected topology, hidden nodes topology and the MAC performance under different topologies, including fully-connected topology, hidden string nodes topology in Figure 1. shown in Figure 1. topology shown and string topology

4.1. Fully-Connected Topology In the dashed circle shown in Figure 1a, nodes are in the transmission range of each other. Five individual experiments with different numbers of nodes were carried out. In the simulation

Sensors 2017, 17, 1130

10 of 14

Sensors 2017, 17, 1130

10 of 14

4.1. Fully-Connected Topology

scenario, arecircle randomly in anodes 150 km plane area.range For each node, the In thenodes dashed showndistributed in Figure 1a, are ×in150 thekm transmission of each other. Poisson traffic is adopted and the mean rate could achieve the maximum value (2 Mbps). The Five individual experiments with different numbers of nodes were carried out. In the simulation unfairness occurs only whendistributed there are in more than transmissions in For process scenario, nodes are randomly a 150 km five × 150 km plane area. eachsimultaneously, node, the Poissonin which the maximum networks load is 10 Mbps. The results of average aggregated throughput and traffic is adopted and the mean rate could achieve the maximum value (2 Mbps). The unfairness occurs end-to-end success ratio are given in Table 3, and the FIs under different MAC protocols are shown only when there are more than five transmissions in process simultaneously, in which the maximum in Figureload 8. is 10 Mbps. The results of average aggregated throughput and end-to-end success ratio networks are given in Table 3, and the FIs under different MAC protocols are shown in Figure 8.

Table 3 Average Aggregated Throughput and end-to-end success ratio for Fully-connected Topology. Table 3. Average Aggregated Throughput and end-to-end success ratio for Fully-connected Topology. Average Throughput (Mbps) End to End Success Ratio End to End Success Ratio ScenarioAverage Throughput (Mbps) NCA-Based BEB-Based NCA-Based BEB-Based Scenario NCA-Based BEB-Based NCA-Based BEB-Based 10 nodes 9.836 10.105 0.99545 0.99461 10 nodes 9.836 10.105 0.99545 0.99461 20 nodes 9.821 10.089 0.99434 0.99121 20 nodes 9.821 10.089 0.99434 0.99121 30 nodes 9.838 9.838 10.018 0.99501 0.99190 30 nodes 10.018 0.99501 0.99190 40 nodes 9.830 10.102 0.99383 0.99201 40 nodes 9.830 10.102 0.99383 0.99201 50 nodes 9.818 10.055 0.99377 0.99097 50 nodes 9.818 10.055 0.99377 0.99097 1 0.9 0.8

Fairness Index

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

NCA-based BEB-based 5

10

15

20

25

Node numbers

Figure8.8.Fairness FairnessIndex Indexfor forthe thefully-connected fully-connectedtopology. topology. Figure

In Figure 8, the BEB-based MAC protocol results in serious unfairness problem in all cases In Figure 8, the BEB-based MAC protocol results in serious unfairness problem in all cases except except the first one as the channel is not filled up (i.e., when the node number is 5, all the nodes can the first one as the channel is not filled up (i.e., when the node number is 5, all the nodes can access access channels with sufficient fairness by both the BEB-based and NCA-based protocols). This is channels with sufficient fairness by both the BEB-based and NCA-based protocols). This is consistent consistent with our previous analysis. In all cases of different node numbers, the FI of NCA-based with our previous analysis. In all cases of different node numbers, the FI of NCA-based MAC protocol MAC protocol can always approach 1 with tiny gaps, which means that most nodes in the network can always approach 1 with tiny gaps, which means that most nodes in the network can share the can share the channel resources and chances to access for successful transmissions. It makes all channel resources and chances to access for successful transmissions. It makes all nodes share the nodes share the wireless channel with enough fairness and stability. From Table 3, by the BEB-based wireless channel with enough fairness and stability. From Table 3, by the BEB-based and NCA-based and NCA-based protocols, the end-to-end success ratios can both reach up to 99%. We can also protocols, the end-to-end success ratios can both reach up to 99%. We can also notice that the average notice that the average throughput of the NCA-based MAC is a little lower than that of the throughput of the NCA-based MAC is a little lower than that of the BEB-based, but is still close to the BEB-based, but is still close to the maximum networks loads (10 Mbps in theory), which tells us the maximum networks loads (10 Mbps in theory), which tells us the NCA-based MAC can guarantee the NCA-based MAC can guarantee the system throughput close to the ideal level. system throughput close to the ideal level. 4.2.Hidden HiddenNodes NodesTopology Topology 4.2. Wethen thencheck checkthe theperformance performanceunder underthe thehidden hiddennodes nodestopology topologyasasshown shownininFigure Figure1b. 1b.We Weset set We different number of nodes to generate different offered loads in the two dashed circles. For each different number of nodes to generate different offered loads in the two dashed circles. For each node, node, the Poisson traffic is and adopted andofthe of its meanenough rate istohigh enough to achieve the Poisson traffic is adopted the value its value mean rate is high achieve maximum rate maximum rate (2 Mbps). The experiment parameters are shown in Table 4 (node amount (left) (2 Mbps). The experiment parameters are shown in Table 4 (node amount (left) includes node 0 and includes node 0 and node amount (right) includes node 1). Scenario 1 shows that the channel of node 0 and node 2 are both in a low traffic loads condition. Scenario 2 shows that the channel of node 0 is

Sensors 2017, 17, 1130

11 of 14

node amount (right) includes node 1). Scenario 1 shows that the channel of node 0 and node 2 are both inSensors a low2017, traffic loads condition. Scenario 2 shows that the channel of node 0 is in a low traffic loads 17, 1130 11 of 14 condition while the channel of node 2 is in a high traffic loads condition. And scenario 3 shows that in a of low traffic condition while condition. the channel of node 2 is in a high traffic loads condition. And both them are loads in a high traffic loads scenario 3 shows that both of them are in a high traffic loads condition. Table 4. Experimental parameters for hidden nodes topology. Table 4. Experimental parameters for hidden nodes topology. Scenario Node Amount (Left) Node Amount (Right) Scenario Node Amount (Left) Node Amount (Right) 1 1 2 2 2 2 2 2 2 2 6 6 3 6 6 3 6 6

The inin Table 5 and Figure 9. The BEB-based MAC protocol results in Thesimulation simulationresults resultsare aregiven given Table 5 and Figure 9. The BEB-based MAC protocol results ain serious unfairness problem and and an obvious decrease of theofend-to-end success ratio ratio in scenario 2 and2 a serious unfairness problem an obvious decrease the end-to-end success in scenario scenario 3. Due3.to theto unequal traffictraffic loads loads in channels, node 0node will 0get much higherhigher throughput than and scenario Due the unequal in channels, will get much throughput node In the the FI ofthe NCA-based MAC protocol close tois1 in scenarios and 3, indicating than1.node 1. contrary, In the contrary, FI of NCA-based MAC is protocol close to 1 in2scenarios 2 and 3, that nodes would get a better the channel and resources more chances access the channel to indicating that nodes wouldshare get aof better share ofresources the channel andto more chances to access transmit evento though theyeven are hidden other. Itfrom makes nodes share multi-hop wireless the channel transmit thoughfrom theyeach are hidden each other. It the makes nodes share the channels fairly and stably by receiving NCA frames and adjusting back-off accurately. multi-hop wireless channels fairly and stably by receiving NCA the frames and time adjusting the back-off time accurately. Table 5. Node individual throughput and end-to-end success ratio for the hidden nodes topology.

Table 5. Node individual throughput and end-to-end success ratio for the hidden nodes topology. Scenario

Node 0’s Throughput (Mbps)

Node 0’s Throughput (Mbps) NCA-Based Scenario BEB-Based BEB-Based NCA-Based 1 1 1.988 1.965 1.988 1.965 2 1.763 1.290 2 1.763 1.290 3 0.831 1.213 3 0.831 1.213

Node 2’s Throughput (Mbps)

Node 2’s Throughput (Mbps) BEB-Based NCA-Based BEB-Based NCA-Based 1.991 1.953 1.991 1.953 1.061 1.290 1.061 1.290 0.056 1.190 0.056 1.190

End to End Success Ratio

End to End Success Ratio BEB-Based NCA-Based BEB-Based NCA-Based 0.99787 0.99683 0.99787 0.99683 0.90563 0.99348 0.90563 0.99348 0.83294 0.99298 0.83294 0.99298

1 0.9 0.8

Fairness Index

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

NCA-based BEB-based 1

2

3

Scenario

Figure9.9.Fairness FairnessIndex Indexfor forthe thehidden hiddennode nodetopology. topology. Figure

4.3.String StringTopology Topology 4.3. Finally,wewe present the experiments of evaluation performance of the NCA-based and Finally, present the experiments of evaluation performance of the NCA-based and BEB-based BEB-based schemes under the string as shown 1c. one In this case, oneinnode is set in schemes under the string topology, as topology, shown in Figure 1c. in InFigure this case, node is set left dashed left dashed circle, and it generates Poisson traffic with the same mean rate as node 0 and 2 of 2 Mbps. circle, and it generates Poisson traffic with the same mean rate as node 0 and 2 of 2 Mbps. Five nodes Five nodes are set in the right dashed circle, which generate Poisson traffic at the same mean rate. are set in the right dashed circle, which generate Poisson traffic at the same mean rate. Figure10 10shows showsthat thatthe theBEB-based BEB-basedscheme schemewill willcause causeaaserious seriousstarvation starvationproblem problemfor fornode node11 Figure and an obvious decrease of the end-to-end success ratio when the loads offered by the five nodes and an obvious decrease of the end-to-end success ratio when the loads offered by the five nodes inin theright rightdashed dashedcircle circlereach reach66Mbps. Mbps.These Theseresults resultsare areconsistent consistentwith withthe thefollowing followingfacts: facts:when whenall all the nodes in the two dashed circles are node-1’s neighbor nodes, node 1 has to content with the other nodes for channel resources. As the neighbor nodes are hidden from each other, they could transmit without back-off due to the lower loads in each dashed circle, thus their transmission could fill up the node 1’s channel as the loads offered by the five nodes in the right dashed circle increase to 6 Mbps (meanwhile, the traffic load in the channel of node 1 reaches 10 Mbps). This causes node 1 to

Sensors 2017, 17, 1130

12 of 14

2

1

1.8 0.98

1.6

End-to-End successful ratio

Node0 & Node2 individual Throughput/Mbps

nodes in the two dashed circles are node-1’s neighbor nodes, node 1 has to content with the other nodes for channel resources. As the neighbor nodes are hidden from each other, they could transmit Sensors 2017, 17, 1130due to the lower loads in each dashed circle, thus their transmission could fill up 12 the of 14 without back-off node 1’s channel as the loads offered by the five nodes in the right dashed circle increase to 6 Mbps always be in the back-off phase to avoid its local channel from overloading. On the contrary, in our (meanwhile, the traffic load in the channel of node 1 reaches 10 Mbps). This causes node 1 to always be NCA-based scheme, if nodes in two dashed circles receive the NCA frame transmitted from node 1, in the back-off phase to avoid its local channel from overloading. On the contrary, in our NCA-based its CLI will indicates its neighbor nodes with a higher proportion of channel share than what they scheme, if nodes in two dashed circles receive the NCA frame transmitted from node 1, its CLI will should obtain. Its neighbor nodes would increase their back-off time accordingly. Along with the indicates its neighbor nodes with a higher proportion of channel share than what they should obtain. increase of nodes’ inaccurate back-off time, the node 0’s throughput can be balanced with the Its neighbor nodes would increase their back-off time accordingly. Along with the increase of nodes’ neighbor nodes, showing better fairness than the BEB-based one. At the same time, the end-to-end inaccurate back-off time, the node 0’s throughput can be balanced with the neighbor nodes, showing success ratio in node 1 could be maintained above 99% even under the condition of high offered better fairness than the BEB-based one. At the same time, the end-to-end success ratio in node 1 could loads as Figure 9b shows. be maintained above 99% even under the condition of high offered loads as Figure 9b shows.

1.4 1.2 1 0.8 Node0 BEB-based Node2 BEB-based Node0 NCA-based Node2 NCA-based

0.6 0.4

0.96

0.94

0.92

0.9 BEB-based NCA-based

0.2 0

4

5

6

7

8

9

10

0.88

4

Offered loads /Mbps

(a)

5

6

7

8

9

10

Offered loads /Mbps

(b)

Figure10. 10. (a) (a)Node Nodeindividual individualthroughput, throughput,(b) (b)end-toend-to-end endsuccess successratio ratiofor forthe thestring stringtopology. topology. Figure

5. Conclusions 5. Conclusions In this paper, the performance degradation of the MAC protocol is recognized as a In this paper, the performance degradation of the MAC protocol is recognized as a compounded compounded result of the randomness of the BEB algorithm and inaccurate traffic loads due to the result of the randomness of the BEB algorithm and inaccurate traffic loads due to the hidden nodes hidden nodes in multi-hop ASNs. In order to solve those two problems of traditional MAC protocols in multi-hop ASNs. In order to solve those two problems of traditional MAC protocols of ASNs, of ASNs, a proposal of exchanging CLI periodically between neighbor nodes in the MAC layer is a proposal of exchanging CLI periodically between neighbor nodes in the MAC layer is developed to developed to overcome the inaccurate traffic loads caused by the hidden nodes. Then a modified overcome the inaccurate traffic loads caused by the hidden nodes. Then a modified back-off algorithm back-off algorithm adaptive to traffic loads is used to calculate the accurate IFS interval, which helps adaptive to traffic loads is used to calculate the accurate IFS interval, which helps to reduce the to reduce the randomness of the BEB algorithm. In order to evaluate the MAC performance of the randomness of the BEB algorithm. In order to evaluate the MAC performance of the NCA-based NCA-based protocol, some simulation experiments have been carried out to compare its protocol, some simulation experiments have been carried out to compare its performance with that performance with that of the BEB-based scheme. Simulation results show that the average of the BEB-based scheme. Simulation results show that the average throughput of NCA is lower throughput of NCA is lower than that of BEB. Nevertheless, the difference is far less than the than that of BEB. Nevertheless, the difference is far less than the maximum channel loads of the maximum channel loads of the system. Simulation results also indicate that the NCA-based MAC system. Simulation results also indicate that the NCA-based MAC protocol achieves better fairness protocol achieves better fairness and higher end-to-end success ratio, which is almost independent and higher end-to-end success ratio, which is almost independent of the traffic loads. In addition, of the traffic loads. In addition, the NCA-based scheme could always achieve stable and high the NCA-based scheme could always achieve stable and high throughput under different topologies, throughput under different topologies, while the BEB-based scheme performs poorly in terms of while the BEB-based scheme performs poorly in terms of metrics for multi-hop topologies. Thus, metrics for multi-hop topologies. Thus, the NCA-based MAC protocol provides a very stable link the NCA-based MAC protocol provides a very stable link layer by the loss of a small acceptable layer by the loss of a small acceptable amount of throughput, which is more adaptive and suitable amount of throughput, which is more adaptive and suitable for future applications in multi-hop ASNs. for future applications in multi-hop ASNs. Acknowledgments: This work is partially supported by Natural Science Foundation of China under Grant Acknowledgments: This work is partially supported by Natural Science Foundation of China under Grant No. 91338108, 91438206. No. 91338108, 91438206. Author Contributions: Xiaolin Gao, Jian Yan and Jianhua Lu conceived and designed the NCA-based MAC protocol; Xiaolin Gao performed the experiments; Gao Jian Yanand analyzed thethe data; Xiaolin Gao wrote Author Contributions: Xiaolin Gao, Jian Yan andXiaolin Jianhua Luand conceived designed NCA-based MAC the paper. protocol; Xiaolin Gao performed the experiments; Xiaolin Gao and Jian Yan analyzed the data; Xiaolin Gao wrote the paper.

Conflicts of Interest: The authors declare no conflict of interest.

Sensors 2017, 17, 1130

13 of 14

Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3. 4.

5. 6.

7. 8. 9.

10. 11. 12.

13. 14. 15.

16.

17. 18.

19. 20.

Yi, W.Y.; Lo, K.M.; Mak, T.; Leung, K.S.; Leung, Y.; Meng, M.L. A Survey of Wireless Sensor Network Based Air Pollution Monitoring Systems. Sensors 2015, 15, 31392–31427. [CrossRef] [PubMed] Allison, R.S.; Johnston, J.M.; Craig, G.; Jennings, S. Airborne Optical and Thermal Remote Sensing for Wildfire Detection and Monitoring. Sensors 2016, 16, 1310. [CrossRef] [PubMed] Xu, G.; Shen, W.; Wang, X. Applications of Wireless Sensor Networks in Marine Environment Monitoring: A Survey. Sensors 2014, 14, 16932–16954. [CrossRef] [PubMed] Cheng, B.-N.; Block, F.J.; Hamilton, B.R.; Ripplinger, D.; Timmerman, C.; Veytser, L.; Narula-Tam, A. Design Considerations for Next-Generation Airborne Tactical Networks. IEEE Commun. Mag. 2014, 52, 138–145. [CrossRef] Chen, L.; Demirkol, I.; Heinzelman, W. Token-MAC: A Fair MAC Protocol for Passive RFID Systems. IEEE Trans. Mob. Comput. 2014, 13, 1352–1365. Mesodiakaki, A.; Adelantado, F.; Antonopoulos, A.; Alonso, L.; Verikoukis, C. Fairness evaluation of a secondary network coexistence scheme. In Proceedings of the 2013 IEEE 18th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks, Berlin, Germany, 25–27 September 2013. Miliotis, V.; Alonso, L.; Verikoukis, C. Weighted proportional fairness and pricing based resource allocation for uplink offloading using IP flow mobility. Ad Hoc Netw. 2016, 49, 17–28. [CrossRef] Hamid, M.A.; Abdullah-Al-Wadud, M.; Chong, I. A Schedule-based Multi-channel MAC Protocol for Wireless Sensor Networks. Sensors 2010, 10, 9466–9480. [CrossRef] [PubMed] Oh, H.; Vinh, P.V. Design and Implementation of a MAC Protocol for Timely and Reliable Delivery of Command and Data in Dynamic Wireless Sensor Networks. Sensors 2013, 13, 13228–13257. [CrossRef] [PubMed] Campbell, C.E.-A.; Loo, K.-K.; Gemikonakli, O.; Khan, S.; Singh, D. Multi-Channel Distributed Coordinated Function over Single Radio in Wireless Sensor Networks. Sensors 2011, 11, 964–991. [CrossRef] [PubMed] Stephen, M.C. Statistical Priority-Based Multiple Access and Method. U.S. Patent 7680077 B1, 16 March 2010. Zhang, H.; Peng, S.; Zhao, Y.; Bian, D. An Improved Algorithm of Slotted-ALOHA based on Multichannel Statistics. In Proceedings of the 2012 Fifth International Symposium on Computational Intelligence and Design, Hangzhou, China, 28–29 October 2012. Bian, D.; Zhang, H.; Peng, S. An Improved Protocol of Ad Hoc Based on Multichannel Statistics on MAC Layer. J. Air Force Eng. Univ. 2013, 14, 80–84. Gao, X.; Yan, J.; Lu, J. A Collision model Providing QoS Guarantee for the feedback-free MAC in Aeronautical Ad Hoc Networks. J. Beijing Univ. Aeronaut. Astronaut. 2016, 42, 1169–1175. Koksal, C.E.; Kassab, H.; Balakrishnan, H. An Analysis of Short-Term Fairness in Wireless Media Access Protocols. In Proceedings of the 2000 ACM SIGMETRICS international conference on Measurement and modeling of computer systems, Santa Clara, CA, USA, 18–21 June 2000. Berger-Sabbatel, G.; Duta, A.; Gaudoin, O.; Heusse, M.; Rousseau, F. Fairness and its impact on delay in 802.11 networks. In Proceedings of the Global Telecommunications Conference, Dallas, TX, USA, 29 November–3 December 2004. Lei, C.; Bie, H.; Fang, G.; Zhang, X. An Adaptive Channel Access Method for Dynamic Super Dense Wireless Sensor Networks. Sensors 2015, 15, 30221–30239. [CrossRef] [PubMed] Ware, C.; Judge, J.; Chicharo, J.; Dutkiewicz, E. Unfairness and capture behaviour in 802.11 ad hoc networks. In Proceedings of the 2000 IEEE International Conference on Communications, New Orleans, LA, USA, 18–22 June 2000. Jancza, T.; Wozniak, J. Fair access to shared radio channel in IEEE 802.11 networks. Syst. Sci. 2002, 28, 131–140. Barret, C.L.; Engelhart, D.D. Analyzing the short-term fairness of IEEE 802.11 in wireless multi-hop radio networks. In Proceedings of the 10th IEEE International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunications Systems, Fort Worth, TX, USA, 16 October 2002.

Sensors 2017, 17, 1130

21. 22.

23.

24. 25.

26.

14 of 14

Kim, Y.; Hwang, G. Design and Analysis of Medium Access Protocol: Throughput and Short Term Fairness Perspective. IEEE/ACM Trans Netw. 2015, 23, 959–972. [CrossRef] Bensaou, B.; Wang, Y.; KO, C.C. Fair Medium Access in 802.11 based Wireless Ad-Hoc Networks. In Proceedings of the 1st ACM International Symposium on Mobile Ad Hoc Networking & Computing, Boston, MA, USA, 11 August 2000. Li, Z.; Nandi, S.; Gupta, A.K. Achieving MAC fairness in wireless ad-hoc networks using adaptive transmission control. In Proceedings of the Ninth International Symposium on Computers and Communications, Alexandria, Egypt, 28 June–1 July 2004. Pan, L.; Cao, X.; Wu, H. Design and analysis of a distributed and fair access (DFA) MAC protocol for multihop wireless networks. IEEE Trans. Wirel. Commun. 2009, 8, 2434–2442. [CrossRef] Yu, Q.; Zhuang, Y.; Ma, L. Dynamic Contention Window Adjustment Scheme for Improving Throughput and Fairness in IEEE 802.11 Wireless LANs. In Proceedings of the 2012 IEEE Global Communications Conference, Anaheim, CA, USA, 3–7 December 2012. Jain, R.; Chiu, D.; Hawe, H. A Quantitative Measure of Fairness and Discrimination for Resource Allocation in Shared Systems; Digital Equipment Corporation: Hudson, MA, USA, 1984. © 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).