Modelling the Behavior of a Beacon-Based Link Sensing Mechanism with Variable Sensing Range Geir Egeland Department of Electrical and Computer Engineering University of Stavanger, Norway geir.egeland@gmail.com Abstract The links in an ad hoc or wireless mesh network are normally kept alive by the exchange of beacon-messages between neighboring nodes. These beacons are prone to collisions with traffic from hidden nodes. If several beacons are lost due to overlapping transmissions, the node expecting the beacons erroneously assumes that the link is down. This is called an apparent link-failure. This paper provides an analytical model of apparent link-failures in a mesh network. The paper also extends the model with an algorithm for finding the upper and lower bound for apparent linkfailures in an arbitrary mesh topology. The model is verified using simulations. The validity of the model is investigated using different link models. Since the radio sensing and transmission range of a node influence its ability to detect ongoing transmissions, the paper also analyze how the sensing range affects the apparent link-failure model. 1. Introduction With the advent of the IEEE 802.11s standard [1] and implementation in enabled devices [2], ubiquitous computing is becoming a reality, promising connection of devices and users in a truly ad hoc fashion. The usability of such wireless ad hoc networks will depend heavily on the routing protocol’s ability to preserve and detect failures of links between neighboring nodes. In the majority of cases, link-failures are present in an ad hoc network regardless of the use of link-maintenance schemes. Some link-failures are unavoidable, such as when a wireless node deliberately leaves a network or is subject to a power failure. Also, a link will cease to be operative when two nodes move outside each others radio transmission range. In addition to these, a set of link-failures exists which are referred to as apparent link-failures. They are primarily caused by wireless links being vulnerable to radio induced interference, 978-1-4244-9328-9/10/$26.00 ©2010 IEEE Paal E. Engelstad University of Oslo, Simula and Telenor GBD&R, Norway paal.engelstad@telenor.com but also when a link-maintenance mechanism erroneously assumes a link to be inoperable due to loss of beaconmessages. The purpose of beacons is to detect and keep links alive. Beacons are normally broadcasted and are thus not acknowledged, i.e. they are unreliable and vulnerable to overlapping transmissions from hidden nodes [16]. Apparent link-failures are often a significant cause to performance degradation of ad hoc networks since erroneous routing information is spread in the network. Also, it might lead to a disconnected topology or less optimal routes to a destination. Analysis of a real life network [12] has demonstrated that it takes a significant amount of time to restore failed links [8]. An example of the effect of these failures is illustrated in Figure 1. Using ns2 [13] we have measured the throughput from node d8 →d7 for the topology shown in Figure 5a. As the load from the hidden nodes [16] increases, the link-maintenance mechanism of the routing protocol [5] starts to loose beacons and assume that the link between node 7 and 8 is unavailable, i.e. an apparent linkfailure. Thus, the throughput from node d8 →d7 is reduced, either because the routing protocol forces the data packets to traverse a longer path or simply because node 7 drops packets when buffers are filled as a result of having no operational route to node 8. The throughput would remain relatively stable if the apparent link-failures were eliminated, as seen from the “No apparent link failure” graph in Figure 1. The main contribution of this paper is an analytical model for the link-failure probability in an ad hoc network, where link-failures are a result of lost beacons caused by overlapping transmissions from hidden nodes. With a model in place it is possible to detect and avoid undesirable topologies that might lead to a high frequency of apparent failures, thus improving the performance and general user experience. Being able to provide stable and reliable links is prerequisite for the success of ubiquitous computing. The model is based on prior work [7] and makes use of the probability pe , that a beacon is lost as a result of a collision with search on ad hoc networks (e.g. cf. [11] [10]), the existence of a link i,j depends on the Euclidean distance between the two nodes vi and vj . If we assume a log-normal shadowing radio propagation model [4] where the logarithmic value of the mean power at different locations is normally distributed with standard deviation σ around the logarithmic value of the area mean power, the link-existence probability [10] is: 0.06 Fixed rate (d8 → d7 ) with no apparent link-failures. Throughput (d8 → d7 ) 0.05 0.04 0.03 Fixed rate (d8 → d7 ) with apparent link-failures. 0.02 pd (r) = 1 − Apparent link-failures No apparent link-failures 0.01 0.00 0.05 0.10 0.15 0.20 Load from hidden nodes {d4 , d10 , d12 } (λc T ) 1 2 h 1 + erf 10log(r/r0 ) √ 2log(10)ψ i 4 ,ψ = σ η (1) 0.25 Figure 1: Throughput from node d8 → d7 with and without apparent link-failures for the topology in Figure 5a. By setting an infinite time-out value, link-failures are artificially removed. transmissions from hidden nodes, to model a beacon-based link-maintenance mechanism [5]. This paper also extends this model with an algorithm for calculating pe in an arbitrary topology, thus improving the bounded pe found in [7]. The radio sensing range (rcp ) of a wireless node determines its ability to detect whether the radio channel is busy and will affect the loss of beacons and the apparent link-failure probability. Consequently, the paper examines the impact of the radio sensing range on the apparent link-failure probability model. The rest of the paper is organized as follows: In Section 2 the link-models and the beacon-based maintenance scheme is introduced. In Section 3 assumptions and the analytical model is presented. In section 4 an algorithm is developed in order to apply the model for arbitrary topologies. In Section 5 the analytical model is studied when the radio sensing range is increased in addition to studying the effect of using a distance-dependent link model. Related work is discussed in Section 6 and finally conclusions are drawn in Section 7. 2. Network Link Model 2.1. Radio-link models In radio communications, the average value of the received signal power on a link s,d decreases with an increasing geometric distance between the two nodes vs and vd . This is often referred to as pathloss, and the area mean power, Pa is often modeled by a power law Pa ∝r−η where η, depends on the terrain and the environment. In the analyses we first use a pathloss model and assume that a signal transmitted from vs is received correctly at vd if the received signal power exceeds a minimum required threshold. Thus, each transmitting node has a circular coverage area of radius r0 . In reality however, the received power levels might vary significantly in space and time around the area mean power value of the pathloss model. For the well-known link failure model that is distance-dependent and is being used in re- where r0 is given by the same threshold as for the area mean power model. Empirically ψ may vary in the range 0 to 6 [10]. In the analyses presented later, ψ will be varied and set to either of the values {0.25,2.5}. These two values alone gives good insight into the effect of radio propagation, e.g. as seen in Figure 10c later in the paper. 2.2. Beacon-based link maintenance In wireless network for ubiquitous computing, links are usually established and maintained proactively by the use of one-hop beacons, which are exchanged between neighboring nodes. Beacons are broadcasted in order to conserve network resources. Thus, every link on which a beacon is received effectively obtains maintenance from only one transmission. Broadcast packets are not acknowledged and beacons are therefore inherently unreliable. A node anticipates to receive a beacon from a neighbor node within a given time-interval and can accept that some beacons occasionally will be missing due to various error conditions. However, if a node fails to receive a number (θ+1) of consecutive beacons, it will assume that the link is inoperable. The value of θ is typically a tradeoff between providing the routing protocol with stable and reliable links (a large θ), and the ability to detect link-failures in a timely and fast manner (a small θ). Since beacons are broadcasted, they are unable to take advantage of RTS/CTS signaling, which adds protection against hidden nodes for the IEEE 802.11 MAC protocol’s unicast data transmissions. Although some beacon loss is avoided using RTS/CTS for the unicast data traffic in the network, it will only help the links of the node that issues the CTS. Examples of routing protocols that make use of beacons are the proactive protocol Optimized Link State Routing (OLSR) [5] and an optional mode of operation for the reactive Ad hoc On-Demand Distance Vector (AODV) routing protocol [15]. The typical value of θ is 2, which we will use later in the analysis. Various beacon-based schemes may differ in the method used to determine whether or not a failed link is operational again. Link-stability is desirable and introducing a link too early can lead to a situation where a link oscillates between an operational and a non-operational state. This can be avoided by measuring the signal-to-noise ratio (SNR) of the failed link and define the link as operational when beacons are being received and the SNR is above a defined threshold [3]. However, if SNR measurement is not available or not practical, a simple solution is to introduce hysteresis by requiring a number of consecutive beacons to be received (θh +1) before the state of the link again is set to be operational. This is the solution chosen in this analysis. s0 B (2) where p0 is the probability that the packet queue of the hidden node is empty and ωb represents the average transmission time of the beacon packet. p0 = I:Isolated, node s1 λc = 0 1 − mρ 1 − mρ(m+1) m+1−ρ III:Connected, node s1 λc = 0 1−ρps1 −ρ II:Isolated, node s1 λc > 0 D D s2 s7 s0 B s1 s5 s3 D s6 D s4 (a) Isolated hidden nodes pe The work in [7] provides an upper and a lower bound for beacon loss probability for IEEE 802.11 based networks. We will now briefly describe the beacon loss model using the topologies in Figure 2. The model depends on three main assumptions. First, it is assumed that a beacon sent by a node has a negligible probability of colliding with a beacon from any of the neighboring nodes since beacons are short packets that are transmitted periodically and at a random instant at a relatively low rate. Second, it is assumed that the probability of a beacon colliding with a data transmission from any of the (non-hidden) neighboring nodes also is negligible, i.e pe pcoll because the IEEE802.11 has mechanisms that reduce such collisions to a minimum. Third, the packet buffers of a node is modeled as an M/M/1 queue with packet arrival rate λc and the channel access and data packet transmission times are exponential distributed with parameter 1/µ. Even though traffic in a real network will follow other distributions, results presented later in the paper based on a large number of random independent traffic scenarios suggest that these assumptions are fair. The probability of a collision between a beacon from node s1 and a data packet from the hidden node s2 in Figure 2 can be shown to be: 1−ρ 1−ρ D s6 s4 3.1. Beacon-loss probability pe pe = 1 − p0 · e s1 s3 s7 s5 (b) Connected hidden nodes Figure 2: Sample topologies where the hidden nodes {s2 , s4 , s6 } are isolated or connected. When the hidden nodes send data (D), this may collide with the beacons (B) sent by node s0 . 3. Link-Failures Caused by Beacon-loss −λc ωb /µ D s2 (3) IV:Connected, node s1 λc > 0 Eq. (3) is an expression for p0 where m is the number of hidden nodes. The hidden nodes are either isolated (Figure 2a) or connected (Figure 2b). The probability that node s1 gains access to the channel is given by ps1 and an approximate expression is found in [7]. If the hidden nodes are isolated (Figure 2a), i.e. outside the transmission range of each other, the probability that one of them sends a data 1 − pe pe pe s0,0 s1,0 s2,0 s2,1 1 − pe s2,2 pe 1 − pe (1 − pe ) Figure 3: A Markov model of a link-sensing mechanism. Table 1: Simulation parameters. IP/MAC layer Values Beacon/ 30/ Data 1500 bytes MAC 802.11 protocol Queue Length 50 Physical layer Values Propagation Free Space model Data rate 11Mbps Simulation Values Simulation/ 900s/25s transient time Traffic/ Poisson Distribution 250m µs Replications #50 rcp , rrx packet which overlaps with a beacon is binomial distributed and given by: pe = Pm k=1 m k k pe (1 − pe )m−k (4) The connected case represents an upper bound for the beacon loss probability while a lower bound will be represented by the isolated case [7]. 3.2. A model for apparent link-failures If we assume that the event of losing a beacon is random and independent, apparent link-failures can be analyzed using a Markov model as shown in Figure 3 where the state variable si,j describes the number of i∈[0, θ] beacons lost and j∈[0, θh ] beacons received in the hysteresis state. Solving the state equation, the probability of apparent link-failure (pf ) is found as the sum of the state probabiliPθh ties j=1 pi,j given by: pf (pe ) = p3e (2 − pe )/(p3e − pe + 1) (5) where pe is the probability of losing a single beacon and is given by Eq. (2)–(4). In order to test the model’s accuracy, we used the ns2 simulation model [13] with OLSR [5] as the ad hoc routing protocol. The sensing range (rcp ) of the physical layer was set to be equal to the transmission range (rrx ). This is not the case in a real network, but simplifies our analysis and provides some topology control. In section 5 we will study ~ Algorithm 1: H(G, S) Data: An undirected graph G(V, E), a directed graph S ⊆ G. u , H l , of the vertice n that Result: The number of neighbors, Hn,i n,i are hidden from vertice i and a traffic indicator Λn . (a) Isolated hidden nodes. (b) Connected hidden nodes. Figure 4: The probability of link-failure (pf ) for the sample topologies where the hidden nodes {s2 , s4 , s6 } hidden nodes send data (D) which may collide with the beacons (B) sent by node s0 . the effect of the sensing range on the beacon loss probability, i.e. for the more realistic case of rcp >rrx . As Figure 4 shows, the model matches well with the simulation results. 3.3. Apparent link-failure probability for complex traffic patterns In order to calculate the link-failure probability for links in a given topology with an arbitrary traffic pattern, an algorithm is needed to determine the number of hidden nodes that have an impact. A topology can be described as a directed graph G=(V, E), where the nodes in the network serve as the vertices vj ∈V (G). Any two distinct nodes vj and vi create an edge i,j ∈E(G) if there is a link between them. A random traffic pattern where a set of nodes transmit data over a link i,j ∈E(G) will form a directed graph S(V, E) that is a subset of G. It is assumed that every node vj ∈S generates data packets at the same rate. Alg. 1 calculates the number of neighbor nodes of n which are hidden from a vertice i∈V (G):i,n ∈E(G) where hu =|{j, ∀j:j∈V (G) ∧ n,j ∈E(G) ∧ ∃j→k∈V (S) ∈E(S)}|. Applying Eq. (5) on these parameters will give the upper bound link-failure probability pf for the link n→i. For the calculation of the lower bound, an average value for the number of hidden nodes is used, which is denoted hl in Alg. 1. The justification for this is that for a set of nodes R⊆V (S) hidden from node i, the carrier sense nature of the MAC protocol will in the cases where two nodes k, z∈R where ∃z6=k:z,k ∈E(G) result in that only a subset of the nodes in R will transmit data at any given time. The parameter hl ≤hu is the average number of nodes in R that transmit data at a given time. The upper bound for the average apparent link-failure probability for a graph (G, E) begin H l ← ∅; H u ← ∅;Λ ← ∅ for i ∈ V (G) do J ← {j, ∀j:i,j ∈E(G)}; for n ∈ J do R ← {r, ∀r6=i:n,r ∈E(G)}; for k ∈ R do if |{j, ∀j:k,j ∈E(S)}|>0∧k∈G / i then hu ← hu + 1 N ← ∅; for k = 0 to 2|R| do ni ← 0; ca ← ∅; for p = 0 to |R|] do if k>>p & 1 ∧ n,Rp ∈ E(S) then ca ← ca ∪ n,Rp ; ni ← ni + 1 if not [∃z:n,z ∈ca ∧ ∃w6=z:n,w ∈ca:z,w ∈E(G)] then N ← N ∪ ni P|N | l 1 Hn,i ← |N k=0 Nk | u Hn,i ← hu Λn ← |{j, ∀j:n,j ∈E(S)}|?0:1} can now be calculated as: puf 1 = |E(G)| X ∀i,n ∈E(G) 3 u u pe (Hn,i ) 2 − pe (Hn,i ) 3 u u )+1 ) − pe (Hn,i pe (Hn,i (6) where H u is provided by algorithm 1. The expression for pe is given by Eq. (2) and p0 is calculated according to Eq. (3) (III:Connected for Λn = 0 and IV:Connected if Λn > 0). The calculation for the lower bound for the average apparent link-failure probability (plf ) is analogous to Eq. (6) using H l and Eq. (4). For a node n, by measuring the traffic load from neighbor nodes j:n,j ∈E(G) and combining this with the 2-hop horizon routing information gathered from the beacons, it can estimate the probability of which neighbor nodes it is likely to receive beacons in a timely manner. This can serve as additional information for the link-maintenance mechanism when deciding on the status of a link. 4. Link-failures in Arbitrary Topologies We now want to investigate how the analyses of the topologies in Figure 2 can be applied to more complex mesh topologies. Without loss of generality, we now focus on the two topologies in Figure 5 as examples, observing that the analysis can easily be generalized for any arbitrary mesh topology. The topologies in Figure 5 do not resemble the topologies in Figure 2, but equations Eq. (2)–(5) will together with Alg. 1 be able provide an upper and lower bound for the link-failure probability pf . The simplest approach to analyzing a bursty traffic pattern is to generate a snapshot of the traffic in the topology. d10 d6 1.0 d10 d8 d12 d4 d8 d2 d7 d9 d11 d2 d7 d0 d1 d3 d5 (a) Topology A d0 d9 d1 d11 d3 d5 (b) Topology B Probability of apparent link failure (pf ) d4 Upper bound Lower bound 1.0 Upper bound Lower bound λ = 0.9 λ = 0.5 Probability of apparent link failure (pf ) d6 λ = 0.4 0.8 λ = 0.3 0.6 0.4 λ = 0.2 0.2 λ = 0.9 0.8 λ = 0.5 0.6 λ = 0.4 0.4 λ = 0.3 0.2 λ = 0.2 Figure 5: The distribution of nodes in two example topologies. λ = 0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Probability of traffic on a link (ptx ) (a) Results for topology A 1.0 0.0 0.1 λ = 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Probability of traffic on a link (ptx ) 1.0 (b) Results for topology B Figure 7: Analytical results for the upper/lower bound of the apparent link-failure probability for the topologies in Figure 5. (a) Results for topology A. (b) Results for topology B. Figure 6: Apparent link-failure probability for Figure 5 (λc = 0.2). Simulation results are shown with a 95% confidence interval. We assume that the time between each snapshot is sufficiently long for the traffic patterns of each snapshot to be considered independent and that for each link in the topologies in Figure 5, a burst of data packets is transmitted with the probability ptx /2. Each node generates data packets within a burst according to a Poisson process with the rate parameter λc . This snapshot will represent a possible data transmission pattern. By generating a large number of random snapshots for a given ptx , the overall average apparent link-failure probability for a given λc can be found. Figure 6 shows the analytical average upper and lower bound for the apparent link-failure probability for λc =0.2 for Topology A and B (Figure 5) using Alg. 1 and Eq. (2)– (5) on the randomly generated traffic patterns. The figure also shows simulation results for the average apparent linkfailure. As the simulation results show, the analytical upper and lower bound provide a good indicator of the average link-failure probability even though it can be seen that the gap between the upper and lower bound increases as ptx →1. This is a result of complex traffic patterns and interaction between the nodes that the simple model does not incorporate. At low values for ptx , the model’s upper and lower bound is as expected, more accurate. In Figure 7 the upper and lower bound link-failure prob- ability for different values of λc is shown. As can be seen from the figure, for small and large values of λc , the gap between lower and upper bound is negligible. The reason for this is that when λc '0, the sum of the packets awaiting transmission in the buffers of the hidden nodes is almost zero in both the isolated and the connected case. Therefore, the apparent link-failure probabilities are almost identical. For the case when λc /1, the sum of packets awaiting transmission in the buffers of the hidden nodes is always greater that zero, i.e. there is always a packets waiting to be transmitted. Hence, the difference in apparent link-failure probability is almost negligible. For 0.2<λc <0.6, the various combinations of empty and non-empty buffers for the isolated and the connected case is large, thus it is expected that there will be a difference in the upper and lower bound. 5. Factors That Impact the Number of Hidden Nodes 5.1. The radio sensing range Until now it has been assumed that the radio transmission range (rrx ) and the radio sensing range (rcp ) are equal. However, in a real network, the sensing range is normally larger than the transmission range, i.e. rcp >rrx . In this section we investigate in more detail how the sensing range affects the apparent-link failure probability. Before doing so, we note that at modest traffic intensities in Eq. (2), it is primarily the number of actively transmitting hidden nodes (given by the term p0 ) that determines the apparent link-failure probability. At high traffic intensities, on the contrary, p0 does not play a role, since the p0 -term is suppressed in Eq. (2), and pe =1 for links with hidden nodes. In this stage, it is the links with no hidden nodes that contributes to an apparent link-failure probability below unity. r cp rrx s0 B s1 s2 D s3 (a) Sample topology with rcp >rrx . (a) rcp /rrx =1.8. (b) rcp /rrx =1. Figure 9: P Analytical and simulation pf =1/|E(G)| ∀ij ∈E(G) pf (i, j). results for Table 2: Simulation results for λc = 0.9. rcp >rrx (b) Number of hidden nodes. (c) Links with no hidden nodes. hu hl |E(S)| h0 pf 4.3 1.3 500 97 0.806 hu hl |E(S)| h0 pf 2.8 1.1 500 94 0.812 rcp =rrx Figure 8: The effect of increasing the sensing range for rcp >rrx . Thus, when investigating the role of the sensing range, we are interested in how the sensing range affects the number of hidden nodes (for the low λc case) and how it affects the number of links with no hidden nodes (for the high λc case). Obviously, the sensing range influences the probability of a beacon loss, as shown for the sample topology in Figure 8a. When the sensing range is larger than the transmission range, it is possible for a beacon from s0 to overlap/collide with a data packet from s3 at node s1 (Fig 8a). Hence, the transmission interferes with a larger part of the area covered by the network topology and is thus subject to a potentially higher number of hidden nodes. Thus, the number of hidden nodes is expected to increase as the sensing range increases for large networks. However, this only take place until a certain point: As the sensing range is in the same order of magnitude as the size of the network, a node can detect ongoing transmissions from a larger part of the topology, thus the number of hidden nodes decreases again. This effect is shown in Figure 8b as an average over 10000 different random topologies, each with N =70 nodes that are randomly and uniformly distributed on the area Ω=1250×1250m2 . The graph shows how the average number of hidden nodes (hu ) changes when the sensing range is gradually increased in the interval [250m, 1250m]. In summary, for modest traffic intensities, whether an increased sensing range has a positive or a negative effect on the apparent link-failure probability depends on the actual network topology, and especially on the size of the topology relative to the sensing range. For high traffic intensities, on the contrary, it is the number of links (h0 ) with no hidden nodes that determines how much lower the average apparent link-failure probability is below unity. Intuitively, as the sensing range increase, a node can detect ongoing transmissions from a larger part of the topology, thus the number of links (h0 ) with no hidden nodes must increase. This is also illustrated in Figure 8c which shows the fraction of the links with no hidden nodes, i.e. h0 /|E(G)|, for the same topologies as used in Figure 8b. Thus, as the sensing range is increased and the traffic intensity is high, the apparent link-failure probability will decrease. In order to verify this, we analyze a random topology G(E, V ) where every node transmits data with a probability ptx =1 and the traffic is in the range λc ∈[0, 1]. The results in Figure 9a–9b show the analytical upper and lower bound for the average apparent link-failure probability for rcp /rrx ∈{1, 1.8}. Simulation results are also shown and it can be seen that the simulation is bounded by the analytical upper/lower bound. The results illustrated in Figure 9a–9b show indeed that the apparent link-failure probability converges towards a limit as the load is increased. This limit can now be calculated as pf =(|E(S)|−h0 )/|E(S)|, where |E(S)| is the number of links formed by the sensing range. From Tab.2 we find pf =0.806 for rcp /rrx =1.8 and pf =0.812 for rcp /rrx =1, which match well with Figure 9a–9b. For low value of λc , it is the number of active, i.e. with data to transmit, hidden nodes that determines the apparent link-failure. From Figure 8c we see that when rcp /rrx =1.8, more hidden nodes are generated than for rcp /rrx =1, which is also verified by the measurement in Tab.2. The result is that the apparent link-failure for rcp /rrx =1.8 is higher than for rcp /rrx =1. This is also verified in Figure 9a–9b. Table 3: Average number of hidden nodes. Hidden nodes hu hl ψ = 0.25 Long links (4%) 4.7 1.5 Short links (96%) 2.8 1.1 Hidden nodes hu hl ψ = 2.5 Long links (53%) 8.7 2.5 Short links (47%) 6.6 2.3 5.2. A distance-dependent link model In the previous sections we have assumed that the links follow a distance-independent model. In the following we make use of a distance-dependent link model and investigate the effect such a model has on the apparent link-failure probability. The link-existence probability, pd , is shown in Figure 10c for ψ∈{0.25, 2.5}. The figure illustrates that a value of ψ=0.25 gives a sharp threshold for r/r0 =1.0, which is similar to a transmission range of rrx = 250m used in previous sub-section. However, a value of ψ=2.5 changes the link-existence probability considerably. The result is that the link-existence probability decreases for shorter links (r/r0 < 1.0) while the link-existence probability for longer links (r/r0 >1.0) increases. In Figure 10d the average apparent link-failure probability for different values of ψ is calculated using Eqs.(1)–(5). This is a snapshot of the linkexistence probability and we assume that this does not vary with time for when pf is calculated. The results are based on the topology in Figure 10b, where rcp =rrx and is calculated according to Eq. (1). Every node in the topology transmits data with probability ptx =1. Figure 10d establish that a considerably higher apparent link-failure probability is generated by the high value of ψ. For λc T <0.6, the explanation for this can be found in the number of hidden nodes that different values for ψ generate. Tables 3 shows the average number of hidden nodes used in the calculation for pf , both for Long links, i.e. r/r0 >1.0 and Short links where r/r0 <1.0. Tab. 3 demonstrates that almost every link is defined as a short link for ψ=0.25 while for ψ=2.5 the fraction of long links and short links are approximately equal. The average number of hidden nodes for ψ=2.5 is also almost twice as large as for ψ=0.25. This explains why the average apparent link-failure probability is much higher for ψ=2.5. For λc T ≥0.6, it is as in previous sub-section the value of h0 that determines the apparent link-failure probability. For ψ=0.25 we find h0 =82 and |E(G)|=493, which gives pf =(493−82)/493=0.83. For ψ=2.5, we find that h0 =0, thus pf =1. This match well with the results in Figure 10d. Implicitly ψ is also an expression for the radio receiver’s ability to correctly receive data. As the bitrate on a link increase, the probability of correctly receiving data ap- r0 na nc nb nd (a) Distance dependent model (c) Link-existence probability (pd ). (b) Random topology (d) Link-failure probability (pf ). Figure 10: Average apparent link-failure probability for different link-existence probability. proaches a step function, which is similar to ψ approaching zero. 6. Related Work The paper is based on prior work in [7] where a model for beacon loss was developed. This work however did not encompass a link-maintenance mechanism nor was it applicable for a general topology. The work in [14] aims to improve performance by introducing a mobility detection mechanism in order to differentiate between node mobility and congestion. This detector is similar to the linkmaintenance mechanism in [5], however, the work does not provide any probabilistic measures of the correctness of the detector. In [9] introduces a mechanism for detecting when a link is about to be broken instead of waiting for the break to happen. This approach the signal strength from neighbor nodes in order to estimate whether an active route to a neighboring node will fail. The disadvantage with this approach is that the detector needs a constant flow of packet on which it can measure the signal strength. In [6] the reliability and availability of a set of topologies are studied using both a distance-dependent and a distance-independent linkexistence model, but the effects of beacon-based link maintenance and hidden nodes are ignored. Here it is assumed that apparent link-failures are a result of radio-induced interference only. In [17] the performance of neighbor sensing in ad hoc networks is studied, but only parameters such as the transmission frequency of the beacons and the linklayer feedback are covered. 7. Conclusion and Future Work This paper introduces an approximate model for the probability of apparent link-failure in beacon based link maintenance schemes. The model can provide a rough upper and lower bound for the apparent link-failure probability. Using a simulation tool, it was observed that the model provides acceptable accuracy for simple topologies. Also, arbitrary mesh topologies with random traffic patterns and bursty traffic have been studied, where the model was extended with an algorithm in order to provide an average upper and lower bound for the link-failure probability. The number of hidden nodes is a key factor for the calculation of the apparent link-failure probability. Since the number of hidden nodes in an arbitrary mesh topology will depend on parameters such as radio sensing range and different link-existence models, the apparent link-failure model has been thoroughly studied varying these parameters. The analyses show that our model is also valid for networks where the sensing range is larger than the transmission range. Also, the geometry of the topology determines the number of hidden nodes, so an increase in the sensing range can either increase or decrease the apparent link-failure probability. However, above a certain sensing range, the number of hidden nodes will decrease as the sensing range increases. The paper shows that the link-existence model has a considerable impact on the number of hidden nodes in a topology, thus affecting the apparent link-failure probability. However, for future work it would be interesting to include the time-aspect of the link-existence model in the apparent link-failure calculations. Last, but not least, we have analyzed how a distance-dependent model affects apparent link-failure probability, showing that the radio transceiver’s ability to sense/receive data influences the existence of hidden nodes. In future work it will be interesting to study how the apparent link-failure model can be applied as an additional parameter upon which a link-maintenance mechanism can base its link-status decision. References [1] Lan/man specific requirements - part 11: Wireless medium access control (mac) and physical layer (phy) specifications: Amendment: Ess mesh networking, 2008. [2] One laptop per child project., July 2010. [3] H. M. Ali, A. M. Naimi, A. Busson, and V. Vèque. Signal strength based link sensing for mobile ad-hoc networks. Telecommunication Systems, 42(3-4):201–212, 2009. [4] H. L. Bertoni. Radio Propagation for Modern Wireless Systems. Prentice Hall Professional Technical Reference, 1999. [5] T. Clausen and P. Jacquet. Optimized link state routing protocol (olsr), October 2003. [6] G. Egeland and P. E. Engelstad. The availability and reliability of wireless multi-hop networks with stochastic link failures. IEEE J.Sel. A. Commun., 27(7), 2009. [7] G. Egeland and P. E. Engelstad. A model for the loss of Hello-Messages in a wireless mesh network. In IEEE ICC 2010 - Ad-hoc, Sensor and Mesh Networking Symposium, Cape Town, South Africa, 2010. [8] G. Egeland and F. Li, Y. Prompt route recovery via link break detection for proactive routing in wireless ad hoc networks. In 10th International Symposium Wireless Personal Multimedia Communications (WPMC), Jaipur, India, Dec. 3-6 2007. [9] T. Goff, N. Abu-ghazaleh, D. Phatak, and R. K. B. Preemptive routing in ad hoc networks. In In Proc. ACM/IEEE MobiCom, 2001. [10] R. Hekmat and P. V. Mieghem. Study of connectivity in wireless ad–hoc networks with an improved radio model. In in Proc. of WiOpt, 2004. [11] R. Hekmat and P. Van Mieghem. Degree distribution and hopcount in wireless ad-hoc networks. Networks, 2003. ICON2003. The 11th IEEE International Conference on, pages 603–609, Sept.-1 Oct. 2003. [12] F. Li, Y, L. Vandoni, and Z. S. et al. Deploying and experimenting wireless ad hoc networks in mountainous regions for broadband access. In Broadband Europe Conference, Antwerpen, Belgium, Dec. 3-6 2007. [13] The Network Simulator NS-2. http://www.isi.edu/ nsnam/ns/. [14] M. Pandey, R. Pack, L. Wang, Q. Duan, and D. Zappala. To repair or not to repair: Helping ad-hoc routing protocols to distinguish mobility from congestion. pages 2311 –2315, may. 2007. [15] C. Perkins, E. Belding-Royer, and S. Das. Ad hoc ondemand distance vector (aodv) routing, July 2003. [16] F. Tobagi and L. Kleinrock. Packet switching in radio channels: Part ii–the hidden terminal problem in carrier sense multiple-access and the busy-tone solution. Communications, IEEE Transactions on, 23(12):1417–1433, Dec 1975. [17] M. Voorhaen and C. Blondia. Analyzing the impact of neighbor sensing on the performance of the olsr protocol. In Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, 2006 4th International Symposium on, pages 1–6, April 2006.