WA-GPSR: Weight-Aware GPSR-Based Routing Protocol for VANET

—The extremely fast topology has created new requirements for the geographic routing protocol, which has been the most efficient solution for Vehicular Ad-hoc Networks (VANETs). The frequent disconnection of links makes the choice of the next routing node extremely difficult. Hence, an effi - cient routing algorithm needs to deliver the appropriate path to transfer the data packets with the most relevant quality of service (QoS). In this work, the weight-aware greedy perimeter stateless (WA-GPSR) routing protocol is presented. The enhanced GPSR protocol computes the reliable communication area and selects the next forwarding vehicle based on several routing criteria. The proposal has been evaluated and compared to Maxduration-Minangle GPSR (MM-GPSR) and traditional GPSR using strict metric analysis. Our experimental results using NS-2 and VanetMobiSim, have demonstrated that WA-GPSR has the ability to enhance network performance.


Introduction
Over the recent decade, Vehicular Ad hoc Network (VANET) underwent tremendous technological advances. It is an integral part of the Intelligent Transportation System (ITS) [1]. VANETs provide many applications [2], such as board active safety, commercial services. Due to the fast mobility, the routing path is often happens reconstructed or broken when a vehicle transmits packets. Hence, the communication performance is affected in many situation according to speed, link lifetime, location of vehicles [3], etc. Thus, the research on VANET becomes crucial. In consequence, routing in VANETs has been a challenge that has led designers to try various strategies to solve it. To this purpose, routing design requires vigilance and development in VANET. Where, routing focuses on transmitting information from the source to the destination through an adequate path to enhance the travel convenience applications.
The classification of the routing protocols is according to two types: routing based on topology and geographic routing. The literature attests that geographical routing tends to be the most forceful and attractive one [4] [5]. Researchers in [6] attests that Geo-Networking has many merits and features, such as high scalability, good performance when velocity of vehcles is high and various information can be included in the status information like geographic location, time stamp, speed and exchange it via location based service. Greedy Perimeter StatelessRouting(GPSR) [7] is the most widely recognized geographical protocol, which exploits the geographic information of vehicles according to VANET characteristics. GPSR uses two routing mechanisms, the primary one is greedy routing which takes the closest vehicle to the destination as the next forwarding vehicle. The second is the perimeter transferwhich is invoked only when the greedy process is unsuccessful.
The proposed algorithm selects the most convenient vehicle for the next jump based on additional criteria and not only on the closest distance to the destination. The WA-GPSR considers various routing measures and assigns weights through additional features in order to offer a superior quality of service (QoS). In this work, we evaluate and compare our improved WA-GPSR protocol with the original GPSR protocol and the Maxduration-Minangle (MM-GPSR) GPSR protocol [8] using exhaustive QoS metrics. From the experimental results, WA-GPSR demonstrates improved performance in urban area.
The remaining section of this work is divided to the following structure: Section 2 discusses some related work. Section 3 presents the concept of a routing protocol based on the GPSR. Section 4 presents the proposed protocol. We present the experimental results with a discussion in section 5. The conclusion is given in section 6.

Related work
In VANET, the greater mobility and higher speed of vehicles lead to frequent update of routing table and irregular distribution of vehicles. Some of the studies conducted on the typical GPSR protocol are described below: work in [14] presented as Routing Based on Greedy Forwarding (GFR) selects the nexthop on the basis of the link quality and the distance between neighboring nodes. The authors define the quantity of neighbor nodes, which we cannot know the exact number of nodes from the source to the destination node.
Despite decades of research, we did not find a significant number of works that meet several routing metrics based on the reliable communication area, weight of all neighboring vehicles to improve the accident controlling applications with best QoS in VANETs. There is a necessity for more generalized algorithm to handle any network scenario. In this work, we propose an improved routing protocol called Weight-Aware Geographic Perimeter Stateless Routing protocol (WA-GPSR) based on additional features. We focus our work on urban scenarios where routing information is more complex due to the presence of intersections, building and traffic lights. The improved WA-GPSR protocol takes into account different routing measures and assigns weights obtained from a large number of parameter configurations. We have compared the performance of WA-GPSR with a recent proposal called MM-GPSR and the original GPSR protocol. Analyzing performance metrics, our results show that our improved WA-GPSR improves packets delivery ratio, reduces end-to-end delay and reached lowers routing cost as well as improves network efficiency.

3
Location based routing protocol

Routing strategy
GPSR is one of the most robust protocol to validate the concept of location-based routing strategy. Depending on the situation of communicating nodes, GPSR protocol uses two forwarding schemes for forwarding packets (see Figure 1(a)). It is assumed that every vehicle obtains its own position coordinates information by using positioning system such as GPS, GNSS [15]. The greedy forwarding (GF) strategy is used to elect the best neighbor that is nearest to the destination. Nevertheless, if there is not a better neighbor nearest toward the destination. The packet will be routed through the perimeter routing process as shown in Figure 1(b) [7], [16], [17].

The problem description
In order to select the best neighbor in the GF strategy, the geographical location of the destination and neighbors is taken into account.
Generally, due to the fast changes of nodes, the node under selection as the best nexthop may be out of communication range. And therefore, the node may not receive the packet like Figure 2. In GF strategy, the current node (source node) selects node B as the next-hop node, however, the position of B, S, A moves to B', S', A', respectively. and then the node could not receive the packet because it is located not within the transmission range. Furthermore, the communication link is broken, which degrades the performance in the network.

Efficient communication area
As depicted on Figure 3, S, A, B, H represent the nodes of the vehicle and the destination node is represented by D. S will choose the closest node to the destination node from the neighboring nodes of the vehicle. The closest node to D is B (xB, yB). The following is the distance between B and D (1), and between B and S (2): The area where two circles overlap with the source S as center, R as radius, dmax as radius and is defined as the effective communication area. The blue part represents this area called ECA. The distance dmax is calculated by the subsequent formula (3) [8].
In the above equation (3), λ Є [0, 1]. It appears that λ affects the size of the area ECA. When λ approaches 0, ECA becomes smaller, subsequently the next-hop in the ECA area is easily selected as the node close to D. When λ approaches 1, ECA becomes longer and the next hop in ECA is selected as the node near S, however the node D may have an increase in the number of hops. When λ is equal to 0.3, it has a satisfactory performance in the greedy forwarding strategy.

Exchanging HELLO message and creating neighbor table
In WA-GPSR, Hello messages are extended (as shown in Table 1) and used to exchange the required information by broadcasting it to neighboring nodes by a onehop communication. After a Hello message is received (Figure 4), a routing record is created in the neighbor table for every vehicles. Upon reception of the beacon packets, each source computes the average movement speed, mobility, link lifetime and cumulative time of the communication and stores them with the remaining neighbor information. Finally, when it receives Hello messages, the neighbor tables are updated and the routing measurements is used to calculate the weight of nodes. Thereafter, the packet is transmitted to the vehicle with highest weight.

Routing metrics
The link lifetime. The smallest duration between nodes defines the link lifetime that remain in communication to transfer packets. Because of the variation in speed and different obstacles, the topology of the link changes and the nodes frequently break.
Consider (xi, yi), (xs, ys) as the coordinates of neighbor i and source node S and the corresponding speeds given by vi, and vs (vs < vi). Then, consider R as the communication range. Subsequently, the lifetime is solved according to Eq (4) [12]: The cumulative communication time. The cumulative communication time T i builds a stable path and avoids frequent rebuilding of routing due to changing node positions in greedy mode. T i is calculated as follows [8]: Where ∆t=t t i i 1 − − In equation (5), T i -1 is the last cumulative time of the reception of Hello message, ti is the current reception time of the Hello message. The cumulative time is initialized firstly at T 1 = 0, by comparing T i in the area ECA (Figure 3), the node with maximum T i is chosen as the next-hop node of the source S (the node is steady to S and close to D).
The traffic node density. Traffic is complex in many roads and the density of nodes is an important metric to determine the efficient routing path. The forward node having highest traffic density is chosen is computed as follow [18]: Where N neighb designates the number of neighbours. R denotes the transmission range of the vehicle. The source path in the GPSR protocol is interrupted when there is no neighbor in the vicinity of the sending vehicle to transmit the packet. While our proposal protocol WA-GPSR establishes a reliable routing path by regarding the density of the traffic node.
The node's mobility. With a minimum mobility value, the mobility of vehicles s and v will be close to each other and they will stay longer in each other's communication area. Node mobility is computed using the follows formula [19]: Where, speeds and speed v are the speeds of nodes s and v, respectively. By choosing a minimum value of Mobility v , the mobility of vehicles s and v becomes closer and thus they will stay longer in the communication area of each other.

Calculating of the weight of neighboring nodes
The computation of the weight of the neighbor node proposed in the WA-GPSR routing protocol is calculated with the formula (8): Weight v = w 1 × LLT s,i + w 2 × T i + w 3 × Density v + w 4 × (1/Mobility v ) Where |w 1 | + |w 2 | + |w 3 | + |w 4 | = 1 In the above equation, Weight v denotes the weight of vehicle v, LLT s,i denotes the link lifetime of connections between the sender node s and the neighbor node i, Ti denotes the cumulative communication time of node, Density v denotes the traffic density of node, Mobility v denotes the mobility of node v and wi are constant coefficient.
The pseudo code of the WA-GPSR routing protocol is described in algorithm 1.

The simulation setup
The vehicle simulation has been conducted on an area of 1000 m² and implemented in NS-2.35 [20]. Starting position of the vehicles used was random distribution. Table 2 presents the microscopic mobility model and Table 3 summarizes all additional parameters in our simulations.   End-to-End Delay (E2ED): The amount of time that a packet spends on average moving from its source to its target. The subsequent equation is applied to compute the delay:

E ED
Time of transmission Total number of received data packet 2 = � Routing cost: is the ratio of the transmitted routing packets to number of data packet received at the destination. It is also a measure of the total number of control packets in the network. Its calculation is made by the following equation:

Routing cost
Total number of transmitted routing packet Total numb = e er of received data packet Network efficiency: defined as a ratio between the total number of data sent and the sum of the number of packets transmitted with the data sent. Its calculation is made by the following equation:

Network efficiency
Number of sent data packet

Packets delivery ratio
The packet delivery rate (PDR) of WA-GPSR, GPSR and MM-GPSR depicted in Figure 5(a) illustrates an increase in the number of CBR connections, especially, for WA-GPSR and MM-GPSR. This is because as more vehicles are connected to the network, the probability of having void problems is less. The simulation results illustrates that the proposal WA-GPSR protocol is more performant than MM-GPSR and GPSR, with a higher PDR for all results. Except for the results with 35 connections where MM-GPSR reaches its destination by the use of stability parameters leads to different route and then, higher PDR. For high CBR data traffic, MM-GPSR can help to balance the load on the network and then increase the PDR. Nevertheless, in that situation, the average difference between WA-GPSR and MM-GPSR is small (< 8%). Compared to MM-GPSR and GPSR, WA-GPSR has a higher PDR on the overall results and aims to avoid communication disruptions. Figure 5(b) illustrates the comparison of PDRs involving WA-GPSR, GPSR and MM-GPSR at different numbers of vehicles. WA-GPSR surpasses GPSR and MM-GPSR due to having more source-destinations available in the urban scenario; the geographic routing protocols make it easy to locate the sending node. With a higher number of 70 vehicles, the WA-GPSR, MM-GPSR and GPSR protocols achieve a high PDR due to the short distance between nodes when the number of vehicles increases, resulting in a meaningful improvement in PDR. Overall, the WA-GPSR protocol has a higher PDR because it considers the stability of neighbouring nodes, link lifetime and density measurements to identify the optimum path to the destination.

5.3
End to end delay Figure 6(a) displays the latency comparison of WA-GPSR, MM-GPSR and GPSR while making a different number of CBR connections. The latency increases of three implemented routing protocols. When transmitting packets, the neighbourhood relationship in urban environments is unreliable and unstable, resulting in higher path redundancy and therefore higher delay. On the whole, WA-GPSR has a clearly lower latency than MM-GPR and GPSR because of optimal routing path selection and a more stable next-hop node, resulting in a robust link of communication and hence a lower latency at the end. The packet will find the optimal path to its destination, resulting in a reduced delivery delay. In Figure 6(b), E2ED under our proposal appears to be smaller in comparison to MM-GPSR and GPSR, since our algorithm selects a more stable next sending node that will reduce the chance of missing vehicles in the communication range, resulting to a smaller E2ED.

Routing cost
The routing cost in Figure 7(a) decrease with the increase of CBR in the network. In some number of CBR connections. we have obtained a lower number of control packets of WA-GPSR which leads to a decrease in the routing cost. Nevertheless, the routing cost of WA-GPSR at 15, 30 and 35 connections is higher than MM-GPSR protocol by 2.9%, 7.4% and 3.3%, respectively. Due to additional fields in Hello packets, leading the nodes to continuously update their one-hop neighbors with more information parameters. However, in low data traffic of about 5 to 10 connections, the routing cost of the GPSR is greater than the other two routing protocols. In most cases, the routing cost of WA-GPSR is lower than that of MM-GPSR and GPSR, mainly because WA-GPSR establishes a reasonable link that improves communication performance. Figure 7(b) depicts the routing cost of WA-GPSR, MM-GPSR and GPSR as a function of the number of vehicles. The routing cost of simulated protocols becomes higher in most situations. In every case, the routing cost of proposed WA-GPSR is less as compared to MM-GPSR and GPSR. MM-GPSR and GPSR improves, especially when the network is highly connected. The WA-GPSR protocol has a better performance in comparison to the other two routing protocols for almost all numbers of CBR connections. Indeed, as they are very appropriate in this scenario, they will make the packet to look for the most appropriate route toward the destination and subsequently increase the efficiency of the network. The exception is when the network is well connected at 30 and 35 connections; MM-GPSR has a higher efficiency. This can be explained by the recovery mode of MM-GPSR, which uses both the minimum angle and the right hand rule, which leads to the best path option in this situation. Figure 8(b) represents the network efficiency of WA-GPSR, MM-GPSR and GPSR according to the number of vehicles. The network efficiency declines as the number of vehicles increases. This is a consequence of the high rate of communication link failures as vehicle density increases. In general, the network efficiency of the WA-GPSR network is higher than that of MM-GPSR and GPSR, as shown in Figures 6(b) and 7(b), our WA-GPSR algorithm reduces the packet delivery time and controls the packets to achieve their destination, which also increases network efficiency.

Conclusion
The extremely dynamic topology and unpredictable behavior of VANET make routing design very difficult. In the presently work, we propose the WA-GPSR routing protocol which is based on the existing GPSR protocol, by discovering and exploiting neighbor information to obtain a steadier routing path. Our improved WA-GPSR optimizes and improves the greedy routing strategy, which is established in a reliable communication area and takes into account various routing measures to adapt the situation of link instability and find a better routing path. The experiments show that the performance of the enhanced protocol WA-GPSR outperforms that of MM-GPSR and GPSR in regards to packet delivery ratio, end-to-end delay, network efficiency and routing cost.