An Estimated Distance-Based Routing Protocol for Mobile Ad hoc ...

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

3473

An Estimated Distance-Based Routing Protocol for Mobile Ad hoc Networks Xin Ming Zhang, Member, IEEE, En Bo Wang, Jing Jing Xia, and Dan Keun Sung, Senior Member, IEEE

Abstract—In mobile ad hoc networks (MANETs), the network topology changes frequently and unpredictably due to the arbitrary mobility of nodes. This feature leads to frequent path failures and route reconstructions, which causes an increase in the routing control overhead. Thus, it is imperative to reduce the overhead of route discovery in the design of routing protocols of MANETs. In this paper, we propose an estimated distance (EstD)-based routing protocol (EDRP) to steer a route discovery in the general direction of a destination, which can restrict the propagation range of route request (RREQ) and reduce the routing overhead. In the EDRP, the change regularity of the received signal strength (RSS) is exploited to estimate the geometrical distance between a pair of nodes, which is called the estimated geometrical distance (EGD). Simulation experiments based on a random waypoint (RWP) model show that the EGD can effectively reflect the actual distance when it is less than the expected value of the distance [which is called the estimation radius (E-Radius)] between any node pairs. We also propose an estimated topological distance (ETD), which is a topology-based EstD, as an aid to the EGD, which can mitigate the effect of inaccurate EGD. The EstD is a combination of EGD and ETD. In the protocol, every node evaluates the link quality through the computational process of the EGD to eliminate the weak links and then uses the EstD (EGD and ETD) to steer the RREQ packets toward the general direction of the destination. Simulation results show that the proposed protocol can significantly reduce the routing overhead and improve the routing performance in dense or high-mobility networks. Index Terms—Distance estimation, mobile ad hoc networks (MANETs), route discovery, routing overhead.

I. I NTRODUCTION

M

OBILE ad hoc networks (MANETs) are multihop wireless networks with mobile nodes that can move freely. Due to a dynamic topology and limited resources, developing a dynamic routing protocol that can efficiently find a routing path Manuscript received April 20, 2010; revised November 26, 2010, February 27, 2011, and May 23, 2011; accepted May 27, 2011. Date of publication June 7, 2011; date of current version September 19, 2011. This work was supported in part by the National Natural Science Foundation of China under Grant 61073185 and Grant 60673171, by the National Grand Fundamental Research 973 Program of China under Grant 2006CB303006, by the open research fund of the National Mobile Communications Research Laboratory, Southeast University, under Grant 2010D15, and by the Anhui Provincial Natural Science Foundation under Grant 11040606M139. The review of this paper was coordinated by Prof. Y. Cheng. X. M. Zhang is with the School of Computer Science and Technology, University of Science and Technology of China, Hefei 230027, China, and also with the National Mobile Communications Research Laboratory, Southeast University, Nanjing 210096, China (e-mail: [email protected]). E. B. Wang and J. J. Xia are with the School of Computer Science and Technology, University of Science and Technology of China, Hefei 230027, China (e-mail: [email protected]; [email protected]). D. K. Sung is with the Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon 305-701, Korea (e-mail: [email protected]). Digital Object Identifier 10.1109/TVT.2011.2158865

with low control overhead is crucial to MANETs. In MANETs, since the nodes arbitrarily move, the network topology changes frequently and unpredictably. The frequent link breakages lead to frequent path failures and route reconstructions, which increase the overhead of the routing protocol and result in lower packet delivery ratio and longer end-to-end delay. It is very important to reduce the routing overhead in route discovery and maintenance. In view of this, we propose a novel route discovery mechanism based on the estimated distance (EstD) to reduce the control overhead of routing protocols in MANETs. Many routing protocols have been proposed for MANETs in the past few years. According to whether they depend on physical position knowledge, these protocols can be divided into topology- and position-based protocols. The topologybased routing protocols use link information to establish a path. When a node needs to discover a route, it broadcasts a route request (RREQ) packet to its neighbors. Due to a lack of position information, each node blindly rebroadcasts the received RREQ until the route is established. Although flooding is an effective mechanism for route discovery, it propagates the RREQ through the entire network, which is an unnecessary operation. The initial motivation of our protocol is as follows: if a source node or an intermediate node possesses distance or direction information about a destination node, then it can be used to steer the route discovery toward the general direction of the destination. The directed route discovery may restrict the propagation of RREQ packets within a narrow region, which includes the destination, and avoid the region that is far away from the destination. Thus, the number of RREQ packets can significantly be reduced. Position-based routing protocols that know the physical position of the nodes have a feature to restrict the propagation of RREQ packets within a narrow region. However, the geographic knowledge is not available in many scenarios. In the absence of positioning service, we need a method to estimate the distance or direction to the destination. Thus, we combine the position-based routing features into on-demand routing protocols and propose an EstD-based routing protocol (EDRP) in the absence of positioning service to improve the route discovery. The contributions of this paper are as follows. 1) We propose an algorithm to estimate the distance of two nodes without positioning service. The EstD includes two parts: a) the estimated geometrical distance (EGD), which is based on the change regularity of the received signal strength (RSS) at the contact time of two nodes to estimate the future geometrical distance after the nodes have

0018-9545/$26.00 © 2011 IEEE

3474


parted from each other, and b) the estimated topological distance (ETD), which is a topology-based EstD to refine the inaccurate estimation of the EGD when it grows large. By using the EstD, we divide the entire network area into three zones: a) src-Zone; b) dst-Zone; and c) other-Zone. In each different zone, we adopt a different strategy to forward RREQ packets. 2) We propose a method utilizing the computational process of the EGD to evaluate the quality of link between neighbors and then exclude the weak links. This is very important for routing protocols because it can reduce the frequency of path failures and route discoveries. 3) By the combination of exclusion of weak links and utilizing the EstD (EGD and ETD) to steer the propagation direction of RREQ packets to the general direction of the destination, the protocol can significantly reduce the routing overhead and improve the routing performance in dense or high-mobility networks. The rest of this paper is organized as follows: In Section II, we introduce the related previous work. In Section III, we describe the RSS-based algorithm to estimate the future geometrical distance and characterize the properties of EGD, E-Radius, and estimation zone (E-Zone). In Section IV, we propose a routing protocol using EGD and ETD. In Section V, we evaluate the performance of the proposed protocol in terms of node density and mobility speed through simulations. Finally, we conclude this paper in Section VI. II. R ELATED W ORK Ad hoc on-demand distance vector routing (AODV) [1] and dynamic source routing (DSR) [2] are topology-based and on-demand routing protocols. Both protocols use flooding to discover routes only when needed, which incurs a small routing overhead. The gossip-based ad hoc routing [3] is an efficient optimization scheme for flooding, where each node forwards a packet with a probability. This scheme can significantly reduce the number of redundant packets while keeping network connectivity. However, the gossiping in [3] still propagates many unnecessary packets to areas that are far away from the region between source and destination nodes [4]. After all, the objective of flooding RREQ packets is to find the destination, not to disseminate data to the entire network. Therefore, even if one flooding scheme can minimize the number of rebroadcast packets to the theoretical best-case bound provided by the minimum connected dominating set [5], which is the smallest set of rebroadcasting nodes, this best-case bound is not the lowest bound for route discovery. Thus, we need to do optimization for routing (not just for broadcasting) to restrict the propagation range of RREQ packets. The expanding ring search (ERS) [1] mechanism, which has been used in AODV, is an efficient method to restrict the propagation range of RREQ packets. However, in the ERS mechanism, the source node does not exploit any information about where the destination is; therefore, the flooding is omnidirectional and not targeted at any particular direction [6]. In position-based protocols, each node determines its own position through Global Positioning System (GPS) or some

other positioning services, and these protocols attempt to use some position information to restrict the flooding region. The distance routing effect algorithm for mobility (DREAM) [7] is a typical directed flooding routing protocol. In DREAM, every node maintains the location information of other nodes in routing tables and sends packets in the direction that is computed based on these location tables. The location-aided routing [8] protocol uses the information about the positions of the other nodes to limit the flooding within a certain area, and thus, it can reduce the overhead in the route discovery of ondemand routing protocols. Compared with the classic gossipbased routing, Li et al. [4] proposed a regional gossip approach. This approach defines some regions that are ellipses using a source and a destination as foci. In this approach, only the nodes inside the ellipse execute the gossip-based protocol to forward packets, and the nodes outside the ellipse do not participate in the gossiping at all. The position-based routing protocols can efficiently control the propagation range of packets, but there are some networks where positioning service is not available, and then, these protocols do not work in this situation. In addition, these protocols need to know the position of the destination. To maintain the location table accurately, each node needs to periodically broadcast its own coordinates to the network [8], which incurs additional overhead. Since conventional ondemand routing protocols do not have any positioning service, it is difficult to exactly control the propagation of RREQ packets. Many routing protocols use historical information to restrict the RREQ flooding within a limited region of the network. Castaneda et al. [9] proposed a query localization technique to reduce the range of network flooding on subsequent route discoveries after a route was previously known. Costa et al. [10] proposed a controlled flooding mechanism to avoid networkwide broadcast in route discovery when a previous path is existed. This mechanism is based on an assumption that if some nodes have recently contacted with a node, then these nodes may know the approximate position of this node in a short interval later. Kim et al. [11] proposed a selective rebroadcast suppression scheme to reduce unreachable nodes. This scheme uses a distributive manner to determine a group of relay nodes, which reduces the unnecessary flooding overhead. Gomez et al. [12] proposed a neighbor-assisted route discovery (NARD) protocol. In NARD, when a source node floods the RREQ packet to a limited region of the network, it looks not only for the destination node but for some neighbors that have been near the destination node recently as well. The second limited flooding that searches for the destination node can use the neighbor nodes as anchor points and then reduces the control overhead of route discovery. Bai and Singhal [13] proposed a bilateral route discovery (BRD) scheme. In BRD, besides the source node, the destination node also actively participates in the route discovery, which has a potential to reduce the routing control overhead by one half. Dubois-Ferriere et al. [6] proposed a FResher Encounter SearcH (FRESH) scheme that steers a flooding-based search in the general direction of the destination by using encounter ages, and then, it can reduce the number of packet transmissions required to find the destination. Therefore, the efficiency of the FRESH is better than the omnidirectional

ZHANG et al.: ESTIMATED DISTANCE-BASED ROUTING PROTOCOL FOR MOBILE AD HOC NETWORKS

search. The authors of the FRESH scheme argued that “For most mobility processes, the distance traveled during a time interval of duration t is positively correlated with t.” This basic principle is valid when the mobility processes are homogeneous, but if the mobility processes are very heterogeneous, this principle becomes unclear, which has been pointed out by the authors. Beraldi et al. [14] proposed a hint-based probabilistic protocol that exploits their proposed metric hints to discover a route to the destination. The computation of hints combines the elapsed and contact times between two nodes. The contact time provides an estimation of the relative speed between two nodes, but this estimation has some deficiency. For example, nodes i and j traverse across the transmission range of node d, with the contact time of node i traversing along the diameter is different from the time of node j traversing along a chord that is close to the perimeter, even if they have the same relative speed with d. The polarized gossip protocol (PGP) proposed in [15] is an enhancement of [14]. In the PGP protocol, the gossiping probability of a node is determined by the difference between the distance to the destination of itself and the distance to the destination of its previous node. In addition, the distance is also estimated by hints, as in [14]. III. E STIMATING G EOMETRICAL D ISTANCE BASED ON C HANGE R EGULARITY OF R ECEIVED S IGNAL S TRENGTH IN C ONTACT T IME This section describes the computation of EGD and analyzes the properties of the EGD. By using the properties, we design an efficient routing protocol, which will be described in the next section. A. Computation of EGD To estimate the future geometrical distance of a two-node pair after the two nodes left each other’s transmission range, we summarize the change regularity of the distance when they are in contact time. The change regularity of distance in contact time can be used to estimate the future distance because the mobility process has a locality feature [6]. The regularity of the relative motion of two nodes in contact time is expected to continue for a while, and the future distance called EGD has some relation with its change in contact time. A change in distance can be reflected by a change in the RSS. Since the RSS measurement is available for almost all wireless devices, we can use the measured RSS and radio propagation model [16] to calculate the distance. This idea is somehow related to the signal-stability-based adaptive (SSA) routing protocol [17], and the distinctive feature of the SSA is to use the signal strength to find and maintain stable routes. However, our idea is to use the change of signal strength to obtain the correlation of distance and time. The computation of the EGD is similar to the connection lifetime prediction algorithm in [18], but our main objective is to estimate the distance when the link breaks and to obtain the link quality (LQ) in a simple method. A change in RSS can be detected by overhearing either data packets or control packets received from neighbor nodes at network layer or link layer, and this assumption is the same as that in [6]. In this paper, we assume that the RSS can be used to

3475

Fig. 1. Calculation of EGD.

compute the distance between two nodes. The researchers in positioning have taken into account the error of RSS and have obtained many valuable results that can be used in our routing mechanism. The EGD is computed as follows. We assume that nodes Ni and Nj move at velocities of vi and vj . If we consider node Ni as a reference frame, then node Nj moves at a relative velocity of v = vj − vi . According to the locality feature, node Nj keeps this relative velocity in some distance. Fig. 1 shows the calculation of the EGD. When node Nj is in node Ni ’s transmission range, assuming at times T0 , T1 , and T2 , node Ni receives packets from Nj with signal strengths P0 , P1 , and P2 , then Ni can use a radio propagation model [16] to calculate the distances D0 , D1 , and D2 from Nj . Using the six values of Ti and Di , i = 0, 1, 2, node Ni can obtain the relative velocity v of Nj with itself. We assume that at time T (T > T2 ), node Nj still keeps the relative velocity v with Ni , and the distances between nodes Nj and Ni are represented by the following equations: ⎧ ⎨ D12 = D02 + (vt1 )2 − 2D0 (vt1 ) cos θ (1) D2 = D2 + (vt )2 − 2D0 (vt2 ) cos θ ⎩ 2 2 0 2 2 2 D(t) = D0 + (vt) − 2D0 (vt) cos θ where t1 = T1 − T0 , t2 = T2 − T0 , and t = T − T0 . According to the first two equations, we can solve v and θ, and then, we derive D(t) = At2 + Bt + C where ⎧ 1 1 1 2 2 2 ⎪ ⎨ A = t1 t2 D0 − t1(t2 −t1 ) D1 + t2 (t2 −t1 ) D2 B = − t11 + t12 D02 + t1 (tt22−t1 ) D12 − t2 (tt21−t1 ) D22 . (2) ⎪ ⎩ C = D02 Note that D0 , D1 , and D2 , as well as T0 , T1 , and T2 , are iteratively calculated. That means when node Ni receives the kth (k > 2) packet from Nj , we do the following iterative calculation: T0 ← Tk−2 , T1 ← Tk−1 , and T2 ← Tk ; and D0 ← Dk−2 , D1 ← Dk−1 , and D2 ← Dk . Thus, Ti and Di , i = 0, 1, 2, are always the last three packets’ transmission times and distances. Then, each node needs a table to store the encounter information such as Ti and Di , i = 0, 1, 2, and the size of the table is O(n), where n is the √ number of nodes. Now, EGD(t) = D(t) = At2 + Bt + C is represented as a function of time t, and t is the difference between the current

3476


Fig. 2. Empirical mean of distance conditional on the EGD.

time and the time of the third to the last packet received from node Nj . B. Properties of EGD To observe the relationship between the EGD and the actual distance, we plot the empirical conditional mean of the actual distance between node pairs, conditional on their EGD. Fig. 2 shows this empirical mean of the distance for the RWP model, over a rectangle surface, whose length is 1600 m and its width is set from 600 to 1600 m. Each point in the graph is computed by considering all the node pairs whose EGD is within a certain interval and averaging over the distance between these node pairs. We observe that the expected distance converges to a constant value as the EGD increases. This property is the same as in [6] and [14], and Dubois-Ferriere et al. [6] claimed that this constant value is on the order of a half side of the square surface. In our research, we find that the turning point of the EGD converging to a constant value happens near the expected value E{L} of distance L between any two independent random points sampled from a uniform distribution. For a rectangular network area a × b (a ≥ b), letting b = ka (0 < k ≤ 1), the expected value E{L} of L is represented in [20, Eq. 20] (which was derived by Ghosh [19]), i.e.,

1 1 1 3 2 2 3− + k + 1 + k − k E{L} = 15 k 2 k2 √ 1 2 1 1 + k2 + k arcosh + arcosh 1 + k2 ×a 6 k k (3) √ with arcosh(x) = ln(x + x2 − 1). The use of (3) is further explained in detail as follows. From Fig. 2, the EGD carries no information about the relative positions of two nodes when it is greater than E{L}. That implies that only when the EGD is less than E{L} can it be used to estimate the actual distance of node pairs. Thus, we use E{L} as an E-Radius and a circle, whose center is node Ni and radius is set to E-Radius, as node Ni ’s E-Zone. From [20], E{L} is monotonically increasing with respect to k. For k → 0, limk→0 E{L} = (1/3)a, and for k = 1, E{L} = 0.5214a [20]. Then, we have the following property.

Fig. 3.

Maximum length/expected length.

Property 1: For any network area a × b (a ≥ b), the effective estimate range of the EGD is between (0.3333a, 0.5214a], and the effective estimate range increases as b/a increases. For any node pairs in a rectangle of a × b (a √ ≥ b), the a2 + b2 = maximum distance among them is M {L} = √ 1 + k 2 × a (b = ka). To inspect the capacity of estimation using the EGD, we define H{L} = M {L}/E{L} to describe how many E-Zones can cover the whole distance between any node pairs. When k increases from 0 to 1, the H{L} is shown in Fig. 3. From Fig. 3, H{L} is monotonically decreasing with respect to k, and the maximum H{L} is 3 (when k → 0). Then, we have the following property. Property 2: For any network area of a × b (a ≥ b), at most three E-Zones can cover any node pairs in this area. According to Property 2, the distance of any source and destination nodes cannot be greater than three E-Radius. That implies that a path from a source to a destination at most passes through three E-Zones: 1) the zone that takes the source node as a center; 2) the zone that takes the destination node as another center; and 3) a mid-zone, if it exists, which is in the middle of the preceding two zones. C. Dividing Network Area Into Zones Based on E-Radius As described in Section III-A, a node can effectively estimate the distance to other nodes that are in its E-Zone. Then, we define the E-Zone of the source node as src-Zone and the E-Zone of the destination node as dst-Zone. Except for src-Zone and dst-Zone, according to Property 2, at most, three E-Zones can cover any node pairs, and thus, the rest area otherZone is only one of the three situations: 1) mid-Zone, which is in the middle of src-Zone and dst-Zone; 2) src-Bound, which is the bound zone in the side of src-Zone; and 3) dst-Bound, which is the bound zone in the side of dst-Zone. One example of the divided zones is shown in Fig. 4. To investigate the relationship between L and E-Radius, we obtain the probabilities of L ≤ E-Radius, L ≤ 2E-Radius, and L ≤ 3E-Radius, which are shown in Fig. 5. Bettstetter et al. [20] presented the probability density function (pdf) of L (refer to [20, Eqs. (19) and (20)] ) and the expected value E{L} of L (refer to [20, Eq. (21)] ). We have mentioned that E-Radius


3477

A. Evaluating the LQ Using EGD The calculation method of the EGD naturally has the ability to evaluate the LQ. Each node knows the EGDs of its neighbor nodes, and thus, from the first two relations of (1), the node can solve the relative velocity v with its neighbor nodes. Then, we √ obtain v = A, where A is shown in (2). We use D2 and D1 to determine whether a neighbor node is coming nearer or going away. If the neighbor is coming nearer, then this link may be strong; otherwise, we use a simple method to estimate the LQ. While receiving a packet from a neighbor node, the relative velocity v and the distances D2 and D1 can be calculated in the EGD calculation part. In addition, the distance can be calculated by the RSS when receiving the packet. Then, the algorithm is shown in Algorithm 1. Fig. 4.

Dividing network area into zones based on E-Radius.

Algorithm 1 LinkQuality(pkt) 1: if D2 ≤ D1 then 2:

return Strong;

3: else 4: Dis = getDistanceByRSS(pkt); LQ = (Tx_Radius − Dis)/v; 5: if LQ < STRONG_LINK_THRESH then 6: return Weak; 7: else 8: return Strong; 9: end if 10: end if Fig. 5.

Probability of L ≤ E-Radius, two E-Radius, and three E-Radius.

is equal to E{L}. According to the pdf of L and its expected value E{L}, we can obtain formulas of the probability of L ≤ E-Radius, L ≤ 2E-Radius, and L ≤ 3E-Radius about a and b, where a and b are the length and width of a network area, respectively. We can use these formulas to plot Fig. 5. We find that, for most of the network area, the probability of L ≤ 2E-Radius is greater than 90%, which implies that for any source and destination nodes, the probability of the intersection of src-Zone and dst-Zone is greater than 90%, and the probability of existing mid-Zone is lower than 10%. Now, we can obtain the basic property of the propagation of RREQ packets in route discovery. For most route discoveries, the RREQ packet is either in src-Zone or in dst-Zone, and the situation that needs to pass through the mid-Zone is less than 10%. src-Bound and dst-Bound should be avoided. Then, according to the preceding property, we design a new routing protocol that is shown in the next section. IV. E STIMATED D ISTANCE -BASED ROUTING P ROTOCOL In this section, first, we introduce one of the applications of the EGD, and second, we introduce a method to amend the EGD. The amended EGD using the ETD is the EstD, which is used as a metric to decide whether a node is a good nexthop candidate. In the third part of this section, we propose the EDRP.

In function LinkQuality(pkt), Tx_Radius is the transmission radius of a node, and STRONG_LINK_THRESH is a threshold that is used to determine whether the link is strong or weak. This function can be used by the node to evaluate the LQ with its neighbor node when receiving a pkt. This LQ is used to determine whether this node is a valuable candidate to forward an RREQ packet, because selecting a weak link may lead to a path disconnection a short time later. B. Amending the EGD Using ETD Using the EGD to estimate the actual distance is effective when it is less than the E-Radius. As time goes by, if the two nodes do not encounter for the second time, then the EGD may become very large, which cannot correctly estimate the actual distance. For a large EGD, we need a criterion to check its effectiveness. Using RSS, we can estimate the distance between neighbors, and in multihop networks, the sum of the distance of every hop can be used as a criterion. This distance is called the ETD because it is on the topology of the network and is not a geometrical distance. The ETD is computed as follows. We modify the RREQ and route reply (RREP) headers and add a new field ETD. When a source node sends an RREQ packet, it sets to ETD = 0. The intermediate node receives this packet and can calculate the distance d from the previous hop node and then sets to ETD = ETD + d. Obviously, the sum of the distance of every hop must be greater than or equal to the geometrical distance. To reduce the error of EGD and ETD, if

3478


EGD < ETD, then we determine that the EGD is effective, and then we use EGD to substitute the ETD. On the other hand, if ETD < EGD, we determine that the EGD is ineffective, and then we set to EGD = ∞. The following nodes (including the destination node) do a similar procedure when receiving an RREQ packet. Thus, the destination node can estimate the topological distance to the source node. When the destination node replies an RREP packet to the source node, it does a similar procedure as the source node, and so do the intermediate nodes and the source node when receiving an RREP packet. After that, the source node can also estimate the topological distance to the destination node. As historical information, the ETD has a lifetime ETD_LIFETIME. Thus, the ETD is also a function of time. If the ETD_LIFETIME expires, then we set ETD = ∞. Note that the ETD is either a topological distance in recent time or ∞. If the ETD of a node to the destination is less than ∞, then it must have a route to the destination in recent time. Based on the notion of spatial locality, “a mobile node cannot move too far too soon”, which is described in [9]; this node may be a good candidate for route discovery. From the computation of the ETD, we know that if EGD < ETD, then the EGD is effective, and if ETD < EGD, then ETD is effective. Therefore, the EstD should be the minimum of EGD and ETD, which is defined as follows: EstD = min{EGD, ETD}.

(4)

C. Protocol Description In the route discovery, when an intermediate node receives an RREQ packet, first, if the node has seen this packet, then it directly discards this packet. Then the node determines the LQ for the previous hop node; if the link is weak, the node also discards the packet. After doing that, the node determines its forwarding strategy for the RREQ packet according to its own EstD metric, and it is divided into three situations (we use EstD(u) and ETD(u) to denote EstD and ETD for node u, respectively): 1) If a node is in the dst-Zone, then the EstD(dst) used to estimate the actual distance is credible. Therefore, this node can execute greedy forwarding according to EstD(dst) to steer the RREQ packet toward the destination. When the EstD(dst) of the current node is less than the EstD(dst) carried in the RREQ packet, the RREQ packet is forwarded; otherwise, it is discarded. 2) If a node is not in the dst-Zone but is in the src-Zone, the node cannot use the EstD(dst) because it may be incredible. Thus, it uses ETD(dst) to determine whether it has some historical information. If ETD(dst) is less than ∞, that means in recent time, this node has a route to the destination. Therefore, when the ETD(dst) of the source node is less than ∞, based on the spatial locality, if the intermediate node whose ETD(dst) is also less than ∞, then it may be in the same side with the destination node, relative to the source node. Thus, this node should use a high probability to forward the RREQ packet. Otherwise, if the intermediate node whose ETD(dst) is ∞, then it may be in the reverse side with the destination

node, relative to the source node. Thus, this node should use a low probability to forward the RREQ packet. When the source has no information about the destination, then the node uses an omnidirectional gossip protocol. 3) Except for the preceding two situations, only when the node in the mid-Zone between src-Zone and dst-Zone, it needs to forward the RREQ packet. For the source node, if EstD(dst) ≤ E-Radius, then we use IN_1_ZONE to denote the position of the source node; if E-Radius < EstD(dst) ≤ 2E-Radius, then we use IN_2_ZONE to denote the position of the source node; if EstD(dst) > 2E-Radius, then that means there exists mid-Zone, and we use IN_3_ZONE to denote the position of the source node. Therefore, if the position of the source node is IN_3_ZONE, the intermediate node uses an omnidirectional gossip protocol to forward the RREQ packet; otherwise, it discards the packet. The proposed algorithm is shown in Algorithm 2. Note that to use the value of the EstD metric, we add an EstD field in the RREQ header (which denoted as pkt.EstD). When a node sends or forwards an RREQ packet, it calculates the EstD for RREQ destination and updates the EstD field in the RREQ packet. However, we do not need to add a field to carry the ETD(dst) of the source node (denoted as pkt.src.ETD), because we do not need the actual value of the ETD(dst), and we just need one bit in the Reserved field of the RREQ packet (refer to RFC 3561) to determine whether the ETD(dst) of the source node is ∞. In addition, the position of the source node (denoted as pkt.src.Pos) only has three values, i.e., IN_1_ZONE, IN_2_ZONE, and IN_3_ZONE, and thus, it just needs two bits in the Reserved field. Algorithm 2 Forward(pkt) Definitions: Gossip(pkt): A gossip algorithm for optimization of broadcasting Random(0, 1): Return a random number between [0, 1) Pmax , Pmin : The high and low probabilities used to forward the RREQ packet pthresh : The probability calculated for forwarding the RREQ packet 1: if pkt.id already seen then 2: discard(pkt); return; 3: end if 4: 5: if LinkQuality(pkt) == Weak then 6: discard(pkt); return; 7: end if 8: 9: if EstD(dst) ≤ E-Radius then 10: {The RREQ has entered the dst-Zone, and in this zone the estimation for the dst is credible.} 11: if EstD(dst) < pkt.EstD then 12: pkt.EstD = EstD(dst); 13: broadcast(pkt); return; 14: else


15: discard(pkt); return; 16: end if 17: else if EstD(src) ≤ E-Radius 18: {The RREQ is still in the src-Zone.} 19: if pkt.src.ETD < ∞ then 20: if ETD(dst) < ∞ then 21: pthresh = Pmax ; 22: else 23: pthresh = Pmin ; 24: end if 25: else 26: pthresh = Gossip(pkt); 27: end if 28: else 29: {The RREQ has left the src-Zone, but it does not enter the dst-Zone.} 30: if pkt.src.Pos == IN_3_ZONE then 31: pthresh = Gossip(pkt); 32: else 33: discard(pkt); return; 34: end if 35: end if 36: 37: if Random(0, 1) < pthresh then 38: pkt.EstD = EstD(dst); 39: broadcast(pkt); return; 40: else 41: discard(pkt); return; 42: end if Compared with geographic greedy routing, our protocol does not need any positioning service, and our protocol is not a pure greedy forwarding protocol. Our protocol integrates the elimination of weak links, EstD information, historical distance information, and gossip protocol. Our protocol combines the advantages of on-demand routing and geographic routing protocols. We mainly employ the distance information to steer RREQ packets toward the direction of destination, which helps reduce the number of RREQ packets greatly. Pure geographic routing protocols select the forwarding path based on the local geographic location, which is the main reason of the local maximum problem. In addition, these geographic protocols may suffer performance degradation due to weak links. Our protocol can avoid weak links through on-demand routing discovery mechanisms, whereas we also use distance information to decrease the overhead of routing discovery (i.e., the number of RREQ packets). The elimination of weak links is used to eliminate the weak routes, and the EstD information (in the dst-Zone) and the historical distance information (in the srcZone) are used to steer the RREQ packet to the general direction of the destination node. When there is no information to use, the omnidirectional gossip protocol is used to pass through the src-Zone and the mid-Zone. In the dst-Zone, the basic RREQ forwarding strategy is greedy forwarding like greedy perimeter stateless routing (GPSR) [21], which always selects a node that is the closest to the destination. In addition, in the src-Zone, the basic RREQ forwarding strategy is steering the RREQ to the dst-Zone side.

3479

The greedy forwarding strategy used in dst-Zone has a problem called dead ends, where greedy forwarding fails when a node does not have a neighbor closer to the destination. That implies that there is a hole in the geographic distribution of nodes. Fortunately, many solutions have been proposed to solve the problem, such as perimeter forwarding adopted by GPSR: when no neighbor is closer, the node marks the packet into perimeter mode; the geographical routing algorithm [22] begins to use depth first search. In recent years, overcoming the local maximum problems in location-aided routing has been an interesting area of research. For example, Li et al. [23] proposed a mobility-based adaptive greedy forwarding approach, and this approach takes advantage of the motion potential that combines the node mobility patterns with the node position to help make forwarding decisions, and Na et al. [24] proposed a potential field-based scheme yet another greedy routing to eliminate the local maximum condition. For our proposed protocol, we use a flag in the RREQ packet to indicate the retry times of route discovery. If the flag is set to 1, then it is the first retry to route discovery, then the nodes use our strategy to forward the RREQ packet. In addition, if the flag is set to 2, which implies that it may have some problems in the first route discovery, then the nodes can use perimeter forwarding or omnidirectional flooding. For simplicity, we do not implement these methods in our implementation of the EDRP. We just use a simple method: If the initial route discovery fails, then it needs to carry out the second round. Li et al. [25] analyzed the effect of node density on “geographic forwarding works best when nodes are dense enough so that geographic forwarding works best when nodes are dense enough so that dead ends are not common”. Therefore, the greedy forwarding used in the EDRP has the same behavior. In addition, the gossip protocol used in the situation that there is no information to use could also affect the performance of the EDRP. Before using the EDRP in practice, we need to know the effect of node density on the EDRP. In Section V-B, we analyze the effect of node density on our protocol. In addition, in Section V-C, we evaluate the routing performance at different mobility speeds. V. P ROTOCOL A NALYSIS AND P ERFORMANCE E VALUATION We compare the performance of our proposed protocol with that of other protocols using NS-2 simulator. We implement our proposed protocol by modifying the current AODV implementation in NS-2. We also compare the routing performance of the EDRP with that of the PGP in [15], which is a similar protocol in recent literature, and the conventional AODV. The PGP uses hints that combine the elapsed and contact times between two nodes to estimate the distance of a node to the destination and uses it to steer the RREQ to the destination. The simulation parameters are as follows: The MAC layer protocol is a distributed coordination function of the IEEE 802.11 protocol. The radio model follows a Lucent’s WaveLAN with a bit rate of 2 Mb/s, and the transmission range is set to 250 m. We consider constant bit rate (CBR) traffic and randomly choose 20 source–destination flows. Every source sends

3480


TABLE I SIMULATION PARAMETERS

all possible delays caused by buffering during routing discovery, queuing at the interface queue, retransmission at the MAC layer, propagation, and transfer time. A. Protocol Implementation

two CBR packets per second, and the size of CBR packets is 512 bytes. The mobility model is based on an RWP model in a field of 1600 m × 1000 m. In this mobility model, each node chooses a random destination and a randomly selected speed from a uniform distribution [1, max-speed]. After the node reaches its destination, it stops for a pause time interval and chooses a new destination and speed. The simulation time for each simulation scenario is set to 900 s. In the results, each data point represents the average of 30 trials of experiments. The confidence level is 95%, and the confidence interval is shown as a vertical bar in the figures. To build a highly mobile scenario, we set the pause time to 0. There are some parameters in our protocol. We set STRONG_LINK_THRESH = 5, ETD_LIFETIME = 10, Pmax = 1.0, and Pmin = 0.3, which all are experiential values. Table I summarizes the detailed simulation parameter values. We evaluate the protocols using the following performance metrics. 1) Normalized number of RREQ packets: It is defined as the ratio of the total number of RREQ packets to the total number of data packets delivered to the destinations. For the RREQ packets sent over multiple hops. each single hop is counted as one transmission. 2) Normalized routing overhead: It is defined as the ratio of the total packet size (bytes) of control packets [including RREQ, RREP, route error (RERR), and Hello] to the total packet size (bytes) of data packets delivered to the destinations. For the control packets sent over multiple hops, each single hop is counted as one transmission. To preserve fairness, we use the size (bytes) of control packets instead of the number of control packets, because the EDRP protocol adds two extra fields EstD and ETD in the RREQ packet and one extra field ETD in the RREP packet, and both sizes are bigger than the size of the original AODV. 3) Packet delivery ratio: It is defined as the ratio of the number of data packets successfully delivered to the total number of data packets generated by the CBR sources. 4) Average end-to-end delay: It is defined as the average delay of a successfully delivered CBR packet from the source node to the destination node. This delay includes

In our simulation, we modify the source code of AODV in NS-2 (v2.30) to implement our proposed protocol, which is described as follows. 1) Each node maintains an encounter table, and the entries of the table are encounter_info, etd, and etd_time for the other nodes. The encounter_info includes the RSS (recv_power) and the received time (recv_time) of any packet received from other nodes in the networks. 2) To calculate the EGD, every node needs to record the last three encounter_infos for the other nodes. When a node receives any packet from the other nodes, it updates the encounter_info in its encounter table for the corresponding node. 3) To calculate the ETD, when a node receives an RREQ or RREP packet, it can calculate the distance from the previous node according to the RSS and the radio propagation model used in NS-2. If EGD < ETD, it uses the EGD to substitute the ETD, and if ETD < EGD, it is set to EGD = ∞. The node updates the etd field in the RREQ or RREP packet and then updates the etd and etd_time fields in its encounter table for the corresponding node. 4) Before an intermediate node forwards an RREQ packet, it first calculates the LQ from the previous node according to Algorithm 1 and eliminates the weak links. In addition, it then calculates the EGD according to (2) and obtains the EstD according to (4). After that, the node according to Algorithm 2 determines whether the RREQ packet will be forwarded. 5) To reduce the overhead of Hello packets, we do not use the periodical Hello mechanism. Since a node sending any broadcasting packet can inform its neighbors of its existence, the broadcasting packets, such as RREQ and RERR, can play the role of Hello packets. We use the following mechanism to reduce the overhead of Hello packets: Only when the time elapsed from the last broadcasting packet (RREQ, RERR, or some other broadcasting packets) is greater than the value of HelloInterval does the node to send a Hello packet. The value of HelloInterval is equal to that of the original AODV. B. Protocol Analysis In the protocol analysis, we pay attention to the routing performance of the PGP and EDRP, compared with an omnidirectional protocol, i.e., the conventional AODV. As mentioned in [25], node density is an important factor that affects the performance of geographical forwarding; in the protocol analysis, we fix the maximum mobility speed to 15 m/s but vary the number of nodes from 50 to 200 to represent different node densities. The analysis results are shown as follows. The normalized number of RREQ packets is shown in Fig. 6. This metric reflects the direct effect of the optimization method


Fig. 6.

Normalized number of RREQ packets.

Fig. 7.

Normalized routing overhead.

for route discovery. As the node density increases, when a node forwards an RREQ packet, there are more nodes receiving and forwarding the RREQ packet. If the protocol does not restrict the propagation range of RREQ packets, then the RREQ packets may be propagated to some areas which are far away from the destination nodes. Thus, the normalized number of RREQ packets of the conventional AODV significantly increases as the node density increases. Both PGP and EDRP protocols restrict the propagation range of RREQ, and thus, the number of RREQ packets is limited. On average, the normalized number of RREQ packets decreases by about 77.3% in the EDRP when compared with that of the conventional AODV. Under the same network conditions, the number decreases by about 54.9% when the number of PGPs is compared with that of the AODV. From the results, we find that both PGP and EDRP can significantly reduce the number of RREQ packets in various node density environments. The normalized routing overhead is shown in Fig. 7. The routing control overhead includes RREQ, RREP, RERR, and Hello packets. The figure shows a similar trend to Fig. 6. Although both PGP and EDRP induce an extra overhead of Hello packets, they also significantly reduce the number of RREQ packets. Thus, they can reduce the normalized routing overhead. On average, the routing overhead is reduced by about 56.3% in the EDRP when compared with that of the conventional AODV. Under the same network conditions, the PGP also can reduce the routing overhead by 35.6% when compared with that of the AODV.

3481

Fig. 8. Packet delivery ratio.

Fig. 9. Average end-to-end delay.

Fig. 8 shows the packet delivery ratio of all the protocols for varying number of nodes in a fixed network area. The result shows that both the PGP and EDRP have a negative effect when the node distribution is sparse. When the number of nodes is less than 75, the EDRP has a lower packet delivery ratio than the conventional AODV. The reason is that when the node distribution is sparse, the situation that there is a hole in the network is more common. In addition, in our implementation, we do not process the dead ends problem, which induces a negative effect. As the node density increases, when the number of nodes is greater than 100 (about 62 nodes per square kilometer), the packet delivery ratio of the EDRP is greater than for the AODV, and it can reach about 95%. The result of the EDRP (without location service) is very similar to the simulation result in [25], whose result shows that geographical forwarding (with a perfect location service) works well for more than 50 nodes per square kilometer. In addition, the gossip protocol could also affect the packet delivery ratio when the node distribution is sparse. Fig. 9 shows the average end-to-end delay of CBR packets that have been received at the destinations in different node densities. The end-to-end delay shows a similar trend to the packet delivery ratio in the sense that when the node distribution is sparse, the EDRP has poor performance than the AODV, and so does the PGP. However, there is a slight difference. When the number of nodes is greater than 125 (about 78 nodes per square kilometer), the EDRP has a lower delay than the AODV.

3482


Fig. 10. Normalized number of RREQ packets.

Fig. 11.

Normalized routing overhead.

Fig. 12.

Packet delivery ratio.

When the network is sparse, because of the hole in the network, the number of retries of route discovery of the EDRP increases, and thus, the delay is greater than for the AODV. From Figs. 8 and 9, we can infer that the proposed protocol EDRP should be used in dense networks, for instance, with more than 78 nodes per square kilometer, which is an empirical value obtained from our simulation experiments. C. Performance Evaluation After analyzing the protocol, we evaluate the performance of the EDRP for varying mobile speeds. In mobile scenarios, the node speed is an important metric that affects the network topology. In this section, we fix the number of nodes to 150 (about 94 nodes per square kilometer) and vary the maximum speed from 5 to 25m/s. The evaluation results are described as follows. The normalized number of RREQ packets is shown in Fig. 10. The main idea of this paper is to use EGD and ETD to steer the route discovery toward the general direction of the destination and restrict the propagation of RREQ packets within a limited region, which includes the destination, to decrease the number of RREQ packets. On average, the normalized number of RREQ packets decreases by about 80.2% in the EDRP when compared with that of the conventional AODV. Under the same network conditions, the number decreases by about 55.7% when the EDRP is compared with the PGP. The normalized number of RREQ packets of all the three protocols increases as the maximum speed increases, but the one of the EDRP increases very slowly. This is unlike AODV and PGP, whose number of RREQ packets increases as the speed increases. As the mobility speed increases, frequent link breakages lead to frequent path failures and route reconstructions, and thus, the number of RREQ packets in the AODV increases as the node speed increases. Although the PGP can decrease the number of RREQ packets by using a polarized gossip strategy based on hints, 1) it does not avoid weak links, and route reconstructions may happen in a little time later, to bring the RREQ packets in route discovery; and 2) it does not use the historical information, and thus, it cannot obtain profit from the previous route. Our proposed EDRP uses a combination of EGD and ETD, which can significantly reduce the number of RREQ packets: 1) Since the EGD naturally has the ability to estimate the LQ, it can avoid selecting a weak link; and

2) since the ETD inherently uses the historical information of the previous route, it can further avoid the propagation of RREQ packets to the zone that does not include the destination node. The normalized routing overhead is shown in Fig. 11. In the calculation of the EGD, each node needs Hello packets to advertise its existence and sense the existence of other nodes. Therefore, the EDRP protocol incurs the overhead of Hello packets, and so does the PGP protocol. To reduce the negative effect of Hello packets, we do not use a periodical Hello mechanism. In our implementation of EDRP and PGP, only when the time elapsed from the last broadcasting packet (RREQ, RERR, or some other broadcasting packets) is greater than the value of the HelloInterval does the node need to send a Hello packet. Even so, it increases the control overhead inevitably. However, the AODV can work very well by using link layer detection instead of using exchanges of hello messages to detect link failures. To preserve fairness, the routing overhead includes all of the routing packets: RREQ, RREP, RERR, and Hello packets. On average, the overhead is reduced by about 57.0% in the EDRP when compared with that of the conventional AODV. Obviously, the percentage of reduction of this metric is less than the normalized number of RREQ packets. Under the same network conditions, the overhead is reduced by about 35.4% when the EDRP is compared with that of the PGP. Fig. 12 shows the packet delivery ratio of all three protocols. It decreases as the maximum speed increases because the network becomes less stable. In this node density, although the


3483

that ETD ≥ EGD and 2) to determine whether a node without geometrical distance information to the destination is a good next-hop candidate by using historical information. Thus, this protocol can estimate the distance of two nodes more accurately without positioning service to more efficiently steer the RREQ packet to the destination node and avoid RREQ packet to the entire network. The simulation results show that this protocol generates less routing control traffic than the AODV and a recent optimized method PGP [15] for various node density and speed cases. The packet delivery ratio and the average end-toend delay give some negative effect when the node distribution is very sparse. R EFERENCES Fig. 13. Average end-to-end delay.

proposed EDRP protocol restricts the RREQ propagation range, it does not incur a negative impact. On the contrary, the EDRP improves the packet delivery ratio since it greatly enhances the stability of the routing path. In addition, the EDRP reduces the number of RREQ packets, which helps decrease the probability of packet collisions and increase the packet delivery ratio. On average, the packet delivery ratio is improved by 3.2% in the EDRP when compared with the conventional AODV. In addition, in the same situation, our approach improves the packet delivery ratio by about 2.1% when compared with the PGP. Fig. 13 shows the average end-to-end delay of CBR packets that have been received at the destinations. When the speed increases, the frequency of link breakage increases. The frequent route reconstructions incur more routing control overhead so as to increase the probability of packet collisions and channel contention. Then, the end-to-end delay increases as the speed increases. First, the EDRP can alleviate the collision and contention problem by decreasing the routing control traffic. Next, the EDRP can avoid weak paths that result in many retransmissions. Therefore, the EDRP also performs better in terms of end-to-end delay. On average, the end-to-end delay is reduced by about 37.2% in the EDRP when compared with that of the conventional AODV. Under the same network conditions, the delay decreases by about 24.5% when the EDRP is compared with that of the PGP. VI. C ONCLUSION In this paper, we have proposed EDRP to reduce the routing control overhead by restricting the propagation range of RREQ packets. The EstD is a combination of EGD and ETD. We use the EstD to divide the network area to 3 zones, and in each different zone we adopt a different strategy to forward RREQ packets. The EGD uses the change regularity of RSS to estimate the future distance after a node leaves the transmission range of another node, and thus, it can reflect the future information to a certain extent. The EGD has two functions: 1) to estimate the future distance after two nodes leave each other and 2) to evaluate the LQ when the two nodes are close to each other. The ETD uses the sum of every hop distance of the previous route to estimate the distance of node pairs, which can benefit from historical information. The ETD also has two functions: 1) to determine whether the EGD is valid according to the fact

[1] Internet Society C. Perkins, E. Belding-Royer, and S. Das, “Ad hoc ondemand distance vector (AODV) routing,” RFC 3561, Internet Society, Jul. 2003. [2] IETF Trust D. Johnson, Y. Hu, and D. Maltz, “The dynamic source routing protocol for mobile ad hoc networks (DSR) for IPv4,” RFC 4728, IETF Trust, Feb. 2007. [3] Z. Haas, J. Y. Halpern, and L. Li, “Gossip-based ad hoc routing,” in Proc. IEEE INFOCOM, 2002, vol. 21, pp. 1707–1716. [4] X. Li, K. Moaveninejad, and O. Frieder, “Regional gossip routing for wireless ad hoc networks,” Mobile Netw. Appl., vol. 10, no. 1/2, pp. 61– 77, Feb. 2005. [5] B. Williams and T. Camp, “Comparison of broadcasting techniques for mobile ad hoc networks,” in Proc. ACM MobiHoc, 2002, pp. 194–205. [6] H. Dubois-Ferriere, M. Grossglauser, and M. Vetterli, “Age matters: Efficient route discovery in mobile ad hoc networks using encounter ages,” in Proc. ACM MobiHoc, 2003, pp. 257–266. [7] S. Basagni, I. Chlamtac, V. Syrotiuk, and B. Woodward, “A distance routing effect algorithm for mobility (DREAM),” in Proc. ACM/IEEE MobiCom, 1998, pp. 76–84. [8] Y. Ko and N. H. Vaidya, “Location-aided routing (LAR) in mobile ad hoc networks,” in Proc. ACM/IEEE MobiCom, 1998, pp. 307–321. [9] R. Castaneda and S. Das, “Query localization techniques for on-demand routing protocols in ad hoc networks,” in Proc. ACM/IEEE MobiCom, 1999, pp. 186–194. [10] L. H. M. K. Costa, M. D. de Amorim, and S. Fdida, “Avoiding networkwide broadcasting with controlled flooding for on-demand ad hoc routing protocols,” in Proc. IFIP Med-Hoc-Net, 2002, pp. 324–328. [11] I. W. Kim, M. S. Jeong, and C. G. Kang, “Selective rebroadcast suppression (SRS) scheme for directional border flooding in mobile ad hoc networks,” in Proc. IEEE ISWPS, 2006, pp. 1–5. [12] J. Gomez, J. M. Cervantes, V. Rangel, R. Atahualpa, and M. Lopez-Guerrero, “NARD: Neighbor-assisted route discovery in wireless ad hoc networks,” in Proc. IEEE MASS, 2007, pp. 1–9. [13] R. Bai and M. Singhal, “Route discovery in mobile ad hoc networks: From unilaterality to bilaterality,” in Proc. MobiQuitous, 2007, pp. 1–8. [14] R. Beraldi, L. Querzoni, and R. Baldoni, “A hint-based probabilistic protocol for unicast communications in MANETs,” Ad Hoc Netw., vol. 4, no. 5, pp. 547–566, Sep. 2006. [15] R. Beraldi, “The polarized gossip protocol for path discovery in MANETs,” Ad Hoc Netw., vol. 6, no. 1, pp. 79–91, Jan. 2008. [16] T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, “A survey of various propagation models for mobile communication,” IEEE Antennas Propag. Mag., vol. 45, no. 3, pp. 51–82, Jun. 2003. [17] R. Dube, C. D. Rais, K. Y. Wang, and S. K. Tipathi, “Signal stability based adaptive routing (SSA) for ad-hoc mobile networks,” IEEE Pers. Commun., vol. 4, no. 1, pp. 36–45, Feb. 1997. [18] X. M. Zhang, F. F. Zou, E. B. Wang, and D. K. Sung, “Exploring the dynamic nature of mobile nodes for predicting route lifetime in mobile ad hoc networks,” IEEE Trans. Veh. Technol., vol. 59, no. 3, pp. 1567–1572, Mar. 2010. [19] B. Ghosh, “Random distances within a rectangle and between two rectangles,” Bull. Calcutta Math. Soc., vol. 43, pp. 17–24, 1951. [20] C. Bettstetter, H. Hartenstein, and X. Perez-Costa, “Stochastic properties of the random waypoint mobility model,” Wireless Netw., vol. 10, no. 5, pp. 555–567, Sep. 2004. [21] B. Karp and H. Kung, “GPSR: Greedy perimeter stateless routing for wireless networks,” in Proc. ACM/IEEE MobiCom, 2000, pp. 243–254.

3484


[22] R. Jain, A. Puri, and R. Sengupta, “Geographical routing using partial information in wireless ad hoc networks,” IEEE Pers. Commun., vol. 8, no. 1, pp. 48–57, Feb. 2001. [23] J. Li and S. M. Shatz, “Toward using node mobility to enhance greedyforwarding in geographic routing for mobile ad hoc networks,” in Proc. MODUS, 2008, pp. 1–8. [24] J. Na, D. Soroker, and C. Kim, “Greedy geographic routing using dynamic potential field for wireless ad hoc networks,” IEEE Commun. Lett., vol. 11, no. 3, pp. 243–245, Mar. 2007. [25] J. Li, J. Jannotti, D. S. J. De Couto, D. R. Karger, and R. Morris, “A scalable location service for geographic ad hoc routing,” in Proc. ACM/IEEE MobiCom, 2000, pp. 120–130.

Jing Jing Xia received the B.E. degree in computer science and technology from Heifei University of Technology, Hefei, China, in 2008. She is currently working toward the Master’s degree with the School of Computer Science and Technology, University of Science and Technology of China, Hefei.

Xin Ming Zhang (M’08) received the B.E. and M.E. degrees in electrical engineering from the China University of Mining and Technology, Xuzhou, China, in 1985 and 1988, respectively, and the Ph.D. degree in computer science and technology from the University of Science and Technology of China, Hefei, China, in 2001. Since 2002, he has been with the faculty of the University of Science and Technology of China, where he is currently an Associate Professor with the School of Computer Science and Technology. He is also with the National Mobile Communications Research Laboratory, Southeast University, Nanjing, China. From September 2005 to August 2006, he was a Visiting Professor with the Department of Electrical Engineering and Computer Science, Korea Advanced Institute of Science and Technology, Daejeon, Korea.

Dan Keun Sung (S’80–M’86–SM’00) received the B.S. degree in electronics engineering from Seoul National University, Seoul, Korea, in 1975 and the M.S. and Ph.D. degrees in electrical and computer engineering from the University of Texas at Austin in 1982 and 1986, respectively. Since 1986, he has been with the faculty of the Korea Advanced Institute of Science and Technology (KAIST), Daejon, Korea, where he is currently a Professor with the Department of Electrical Engineering. From 1996 to 1999, he was the Director of the Satellite Technology Research Center, KAIST. He was the Division Editor of the Journal of Communications and Networks. His research interests include mobile communication systems and networks, with special interest in resource management, machine-to-machine communications, smart grid networks, wireless local area networks, wireless personal area networks, high-speed networks, next-generation Internet protocol-based networks, traffic control in wireless and wireline networks, signaling networks, intelligent networks, performance and reliability of communication systems, and microsatellites. Dr. Sung is a member of the National Academy of Engineering of Korea. He is the Editor of the IEEE COMMUNICATIONS MAGAZINE. He was the recipient of the 1992 National Order of Merits, the Dongbaek Medal, the 1997 Research Achievement Award, the 1997 MoMuc Paper Award, the 2000 Academic Excellence Award, the Best Paper Award at the 2000 Asia-Pacific Conference on Communications, the 2004 This Month’s Scientist Award from the Ministry of Science and Technology and the Korea Science and Engineering Foundation, and the Best Paper Award at the 2005 Next-Generation PC International Conference.

En Bo Wang received the B.S. degree in computer science and technology from Jilin University, Changchun, China, in 2007 and the M.E. degree in computer science and technology from the University of Science and Technology of China, Hefei, China, in 2010. He is currently with the School of Computer Science and Technology, University of Science and Technology of China.