Vehicular Traffic Optimization in VANETs: a Proposal for Nodes Re-routing and Congestion Reduction

Recently, vehicular networking has grown up in terms of interest and transmission capability, due to the possibility of exploiting the distributed communication paradigm in a mobile scenario, where moving nodes are represented by vehicles. The different existing standards for vehicular ad-hoc networks, such as DSRC, WAVE/IEEE 802.11p, have given to the research community the possibility of developing new MAC and routing schemes, in order to enhance the quality and the comfort of mobile users who are driving their vehicles. In this paper, we focus our attention on the optimization of vehicular traffic flowing, where the vehicle-2-roadside device is available. As shown in the next sections, the proposed idea exploits the information that is gathered by road-side units with the main aim of redirecting traffic flows (in terms of vehicles) to less congested roads, with an overall system optimization, also in terms of Carbon Dioxide emissions reduction. Several campaigns of simulations have been carried out to give more effectiveness to our proposal.


Introduction
Vehicular Ad-hoc NETworks (VANETs) represent a new and modern paradigm of communication, where the nodes are able to communicate in a distributed manner, based on the Ad-hoc paradigm [1].Each node is equipped with a wireless device, the On-Board Unit (OBU), which is able to interact with the mobile user, especially for comfort/security applications, trade and infotainment services.The OBU devices are able to realize the pure Ad-hoc communication networking in VANETs, indicated with Vehicle-2-Vehicle (V2V) communication.The complete architecture also provides the, so-called, Road-Side Units (RSUs) which can also be any equipment-certified packet forwarding, such as GSM, WLAN, and WiMAX towers.These devices realized the, so-called, Vehicle-2-Infrastructure (V2I) paradigm.The RSUs are very useful for guaranteeing the complete coverage of an area when some distributed nodes are disconnected, giving the driver the possibility of still being able to receive the needed information.
In this way, the road safety is improved, also because emergency vehicles can act more speedily; VANETs are able to broadcast real-time alerts to drivers about the risks of their planned journey and their immediate surroundings [2].In addition, if a danger situation is created or, at a particular place, an emergency vehicle is needed to come quickly, VANETs give the chance to improve the effectiveness of the needed operations, by exploiting the effects of dedicated protocols and algorithms [3], [4].For instance, if the cars involved in accidents can advise the event instantly to the emergency services, a ready and timely intervention can be immediately scheduled.If also the near cars can receive the update, they would reduce inconveniences: platooning would be really helpful in order to leave the right space on the roads for the emergency vehicles, without time wastages.V2V communication allows the development of new applications and one of the main desires of drivers is also to avoid congested roads during their journeys and traveling.In this paper, we focus our attention on the optimization of traffic flowing in a vehicular environment with V2I capability.The proposed idea enables the considered vehicular network to re-route all the vehicles on new paths toward destinations, avoiding useless time wastages and reducing the creation of harmful Carbon Dioxide (CO 2 ) emis-sions.As shown later, in the next sections, the proposed algorithm, called Congestion Avoidance in Vehicular Environments (CAVE), exploits the information that is gathered by RSUs, with the main aim of redirecting traffic flows (in terms of vehicles) to less congested roads, with an overall system optimization, also in terms of CO 2 emissions reduction.Our proposal is based on system modelling by a weighted oriented graph, able to capture all the real-time system values.Each vertex and edge of the graph participates to the evaluation of new paths, giving to mobile users the possibility of following different itineraries to their destinations more quickly.As for the majority of the vehicular applications, the routing protocol covers a crucial importance for the whole architecture [5], [6].
It is important to consider scalability properties of protocols and architectures such as in [7], [8], [9], [10], or optimization techniques such as proposed in [11] in order to improve multiple metrics in the defined problem.The paper is structured as follows: gives an in-depth overview on the related work about optimization schemes in VANETs and points out the main contributions given in this paper.
• Section 3. briefly introduces the standards and, then, describes in a detailed way the CAVE algorithm.
• Section 4. illustrates the obtained results, which confirm our expectations.
• Conclusions are resumed in section 5.

Related Works and Main Contributions
The aim of this work is to exploit the potential of the VANETs, building intelligent mechanisms that can lead to a better management of vehicular traffic, reducing the time spent in the city, the emissions of CO 2 and the fuel consumption.Predictive approaches are appreciated in telecommunication systems, not only in vehicular environments [12], [13], [14].In fact, in literature there are several works that try to get that benefit, using nodes mobility prediction policies and vehicular traffic re-routing approaches [15].In particular, in [16] the authors proposed some traffic re-routing strategies designed to be incorporated in a cost-effective and easily deployable vehicular traffic guidance system, which reduces travel time.These strategies proactively compute tailor-made re-routing guidance to be pushed to vehicles when signs of congestion are observed on their route.They also allow tuning the system to different levels of the trade-off between re-routing effectiveness and computational efficiency.In [17], the authors designed a mechanism for reducing/avoiding traffic waves by integrating Artificial Intelligence and VANET, to create a driver aid that helps in combating traffic congestion as well as embedding safety awareness by dynamically re-routing traffic depending on road conditions.In [18], the authors developed an Intelligent Transportation System (ITS) based on multi-mobile agent systems and VANETs.This approach enables individual vehicle drivers to make quick responses to the road congestion.In particular the drivers, around the congestion area, can also make the appropriate decision before they reach the congested road.In [19], the authors proposed a system able to reduce the travel times and the fuel consumption in different European cities.They also designed a Red Swarm architecture based on an evolutionary algorithm and on smart WiFi spots, located near traffic lights, which are used to suggest alternative routes for vehicles.In [20], the authors proposed two green driving suggestion models: Throughput Maximization Model and a model that aims to reduce the effects of acceleration and deceleration.The aim of the proposal is to minimize the CO 2 emissions, considering real-time traffic information nearby intersections.In [21], authors developed and implemented an instantaneous statistical model of emissions (CO 2 , CO, HC, NO x ) and fuel consumption for light-duty vehicles, which is derived from the physical load-based approaches.The model is tested for a restricted set of some vehicles models, used with standard and aggressive driving cycles.It is implemented in Veins Framework (also used for our simulations).
The main contribution of this paper consists in the proposal of a new traffic re-routing algorithm, able to manage the mobility patterns of vehicles for evaluating new routes on the roads with a lower traffic density.In particular: • The vehicular network is modeled by an oriented and weighted graph, for which the weights are dynamically updated based on the number of vehicles on the different streets.
• New paths are evaluated by taking into account the average congestion level on the paths, so the CAVE algorithm is needed for reducing the average size of the queues and the CO 2 emissions.

VANETs Introduction and the CAVE Algorithm
This section gives a detailed description of the proposed CAVE algorithm, after a brief introduction of vehicular environments and its different standards.

Vehicular Communications Through VANETs
Different proposed and accepted standards have contributed to the rapid growth of vehicular architectures.
VANETs are able to provide wireless networking capability in situations where the communication among nodes can be either direct or made via relaying nodes, as in classical Ad-hoc networks.The IEEE 802.11p, also called Wireless Access in Vehicular Environments (WAVE) [22], is an extension of the IEEE 802.11 standards family for vehicular communications.It aims at providing the standard specifications to ensure the interoperability between wireless mobile nodes of a network with rapidly changing topology (that is to say, a set of vehicles in an urban or sub-urban environment).The MAC layer in WAVE is equivalent to the IEEE 802.11eEnhanced Distributed Channel Access (EDCA) QoS extension.Therefore, application messages are categorized into different ACs (Access Classes), where AC0 has the lowest and AC3 the highest priority.Within the MAC layer, a packet queue exists for each AC. Figure 1 shows a typical VANET scenario, in which OBUs and RSUs can communicate in the distributed environment.An important issue in VANET is the choice of an appropriate transmission channel, not only considering the type of traffic (emergency, security, platooning, etc.) but, mainly, focusing on the reduction of the inter-node interference.The Dedicated Short Range Communication (DSRC) [23] spectrum is divided into 7 channels, each one with a 10 MHz bandwidth; it is allocated in the upper 5 GHz range.A mobile/stationary station switches its channel between the control channel and a service channel every channel interval.The default value for the control/service channel interval is set to 50 ms in the standard.The PHY layer employs 64-subcarrier OFDM.52 out of the 64 subcarriers are used for actual transmission consisting of 48 data subcarriers and 4 pilot subcarriers.Possible modulation schemes are BPSK, QPSK, 16-QAM and 64-QAM, with coding rates equal to 1/2, 1/3, 3/4 and an OFDM symbol duration of 8 µs.The WAVE standard relies on a multi-channel concept, which can be used for both safety-related and entertainment messages.The standard accounts for the priority of the packets using different ACs, having different channel access settings.This shall ensure that highly relevant safety packets can be exchanged timely and reliably even when operating in a dense urban scenario.Each station continuously alternates between the Control Channel (CCH) and one of the Service Channels (SCHs) or the safety channels.

The Congestion Avoidance in Vehicular Environments (CAVE)
Now the proposed idea is deeply illustrated.First, some basic definitions are given.Then the graph model is introduced, as well as the main steps of the algorithm.

1) Geographical Map Representation
The predictive forwarding scheme applies to a generic map (a square, rectangular or circular area).It is composed of a set of RoaDs RD = {r 1 , . . ., r m } (considered, traditionally, as hard flat surfaces for vehicles, people, and animals to travel on) modeled as lines, a set of Road Side Units (primary nodes) RSU = {p 1 , . . ., p n } modeled as points belonging to one or more lines (e.g. if their coverage range contains more than one road, as at intersections or if there are near parallel roads), and a dynamic set of Mobile Hosts (secondary nodes) M H(t) = s 1 , . . ., s q(t) (vehicular nodes enter and exit the map dynamically during time).Each primary node p k on the map is considered as a point with coordinates (x 0k , y 0k ) and a coverage radius R k .We have RD = m, RSD = n and M H(t) = q(t).A road segment is defined as a portion of the road that interconnects two primary nodes p i and p j , starting from p i = (x 0i , y 0i ) and ending in p j = (x 0j , y 0j ).So, the set Road Segments (RS) can be defined as: Clearly, a road segment rs ij ∈ RS may coincide with a whole road r l ∈ RD. 2) The Weighted Oriented Graph Associated to the Map Given the definitions above, the whole system topology can be modeled by a Weighted Oriented Graph W OG = V, E, W , where V is the set of vertices and each vertex is associated to a single primary node, so V = RSU = n, E is the set of edges, and W is the set of weights associated with each element of E. A couple of nodes v i and v j in W OG are neighbors if there exists r si ∈ RS such as vehicles can flow from p i to p j .
In our abstraction, we are not caring if an RSU is deployed at the road-side or its center.So, differently from the classical approaches based on the electromagnetic coverage, two nodes in the graph are directly connected by roads, disregarding the coverage radius of the associated RSU s.This is because we are caring about vehicles traffic (not data traffic) and the roads physical parameters need to be taken into account.In addition, if two primary nodes p i and p j are reciprocally covered (classical adjacency), then they are considered as one RSU node p k .That is to say, if ∃p i , p j ∈ RSU such that p i ∈ A|p j and p j ∈ A|p i , where A|p l represents the coverage area of node p l , belonging to (x − x 01 ) • 2 + (y − y 01 ) • 2 = R 12 , then the nodes p i , p j are removed from RSU and the new node p k with coordinates and coverage radius: is added to RSU (and, consequently, to the set V of W OG). The term Overlap i,j represents the average diameter of the coverage area shared among p i and p j .Figure 2 illustrates the overlapping for two primary nodes p i , p j .In this way, we are assuming that, in our W OG, there are no nodes with overlapping coverage area.Moreover, we assume that W OG is not disconnected (there are no isolated primary nodes into the system).So, under these assumptions, the considered traffic map can be completely modeled by a W OG, as illustrated in Fig. 3.If there are more than one road segments that interconnect two primary nodes, then the set E will contain some so-called multiarcs (W OG will not be a simple graph).In Fig. 3 RSU = {p 1 , . . ., p 11 }, so n=11, RD = {r 1 , . . ., r 10 }, so m = 10, V = RSU ; for sake of simplicity, the coverage radius has been represented to be the same for each primary node (R k =65 meters, ∀p k ∈ RSU ).
Fig. 2: The concept of primary nodes coverage overlapping and the new logical primary node p k .

3) How to Define the Weights of the Oriented Graph and Assumptions
At this point, some assumptions have to be made: • We assume that each vehicle knows exactly the best path to arrive to destination: the OBU of each secondary node s l ∈ M H(t) is integrated with a GPS device, on which the driver has set the itinerary before starting the trip.
• When the secondary node s l arrives under the coverage of the first road-side unit of the system, it points-out the itinerary that will be followed, by sending to the primary node the sequence of roadsegments IT sl = r sl1 , r sl2 , . . ., r slm and the related moving directions; in this way the system is aware about the trajectory that the mobile node wants to follow and the node s l will be inserted in the set of nodes covered by the local road-side unit.
• If the secondary node s l ∈ M H(t) signals to the system that it is going to be parked on a particular road-segment, the first road-side unit that receives the message will temporary remove node s l from M H(t); the node s l will belong to M H(t) again when it decides to move on the road; if s l is not covered by any primary node, the message will be forwarded on the basis of the V2V paradigm.
• After the initial communication (as described in the previous point), there is also a periodical communication: each primary node broadcasts a polling message to all the covered secondary nodes; this approach is needed for giving knowledge to the primary node of the presence of each vehicle in the covered area.
• Finally, each secondary node s l ∈ M H(t) puts into polling answers its GPS coordinates; this information is necessary to the primary node that is covering s l , in order to know the last position of the node before leaving the coverage area and, then, the road to which s l is flowing out.Now, the way the weights are determined is illustrated.All the nodes of the W OG store the weights of the edges in a data-structure (we do not care if it is a data-base or something different), associated to the adjacencies matrix of W OG.
It is a shared structure, so each primary node can send an update message to a dedicated server.We can distinguish among update/increase and update/decrease messages.In fact: • Update/increase: each time a secondary node s l ∈ M H(t) leaves a primary node coverage area Fig. 3: An example of RSU placement in a map of 1250×690 square meters (on the left) and the related W OG (on the right).
A|p i , the node p i knows exactly the itinerary of s l , so its destination is known, as well as its next serving road-side unit p j .At this point, there will be another mobile node traveling on the road segment rs ij , and node p i can signal an increment of one unit of the weight w i,j ∈ W . Based on the definition given before for the set W , we recall that the term w i,j represents the number of vehicles on the edge from v i to v j .
• Update/decrease: each time a secondary node s l ∈ M H(t) enters a primary node coverage area A|p j , the node p j , aware about the road segment from which s l arrived, knows the primary node p i that was serving s l before.At this point, node p j can signal to the system that the number of mobile nodes traveling on segment rs ij is decreased by 1, as well as the weight w i,j ∈ W .

4) The CAVE Steps
The core of our proposal is now illustrated.We resume the main steps of the CAVE algorithm in a pseudo-code and, then, we explain them.
This step simply consists in the construction of the main structure, the W OG, shared in the whole system and able to store the main parameters needed for CAVE.
• Step 1 (Departure of a mobile node s l from the coverage of v i ): Increase w i,j ; This step simply consists in the increasing of the w i,j weight, because of a departure of a mobile secondary node s l from A|p i towards A|p j .
• Step 2 (Arrival of a mobile node s l in the coverage of v j ): This step represents the core of the CAVE algorithm Alg.(1).When a secondary node arrives into the coverage of a road-side unit p j , if it is the first primary node, the itinerary IT sl is acquired, otherwise, if the node arrives from the road-side unit p i , the weight w i,j is decreased (the number of vehicles on the road-segment rs ij has decreased by one).The following pseudo-code explains the main operations that are carried out in the case of mobile node arrival.
Algorithm 1 The CAVE pseudo-code in the case of MH arrival.
In every case (in the sense that p j may not be the first visited primary node), the itinerary of the secondary node s l is checked.Each primary node checks if IT sl contains any congested road segment.In particular, the is_congested(.)function is based on the following observations.Based on the definitions and studies in [24], [25], we know that the capacity can be defined as "the maximum sustainable flow rate at which vehicles or persons reasonably can be expected to traverse a point or uniform segment of a lane or roadway during a specified time period under given roadway, geometric, traffic, environmental, and control conditions".Following the theory in [24] and the notations used in our paper, we can express some basic relationships and diagrams as follows: where k i,j is defined as the traffic density on the road segment rs i,j , given by the ratio of w i,j (the number of vehicles in the road segment rs i,j ) and the length of rs i,j , while I i,j is the traffic intensity, defined as the product of k i,j and the average speed v i,j of rs i,j .The term td i,j represents the trip delay of the roadsegment rs i,j , that is to say, the time needed to travel across the road-segment.Clearly, k i,j and I i,j are functions of time; the only constant term is rs i,j .There are many fundamental diagrams in the traffic flow theory, as the one depicted in Fig. 4, representing the relationship between the density and the speed of a road segment.From the previous figure, we can observe how the average speed on a road segment decreases when the density increases, until the value k jam , for which the mobility on the road is completely blocked.From [25]; the value k crit brings the road segment to be in the ideal situation, with the maximum traffic volume (measured in vehicles/time).In our work, we are not considering bigger roads with more lanes on the same direction.Considering ideal conditions, the maximum capacity can be numerically obtained by fitting the curves, but the more complex analytical analysis should be carried out for real cases.In particular, from [26] it can be written that, for a motorway, the capacity c i,j of the road-segment rs i,j is: where C i,j is the ideal capacity, N i,j is the number of lanes, F W i,j is a factor related to the width of rs i,j , F HW i,j is related to the probability of having heavy vehicles and F P i,j is a factor that derives from the driver population.In our proposal, the algorithm should guarantee that k i,j ∼ = k criti,j for each rs i,j ∈ RS.So, after these considerations, the is_congested(.)function returns true if and only if the following condition is satisfied: where k criti,j is the desired average density for rs i,j and α i,j is a near-to-zero value representing the maximum tolerable deviation from the desired value.With the term α i,j we want to analyze what happens to the system when we consider a different maximum capacity on a given road.When Eq. ( 12) is satisfied, the algorithm will find an alternative itinerary for s l , if it exists, which involves different/alternative roadsegments, with lower weights (in terms of density and trip delay).So, each road-side unit p k ∈ RSU is able to evaluate the alternative paths since it exactly knows the W OG structure.
In the paper, we are not caring about the signaling protocol needed to carry out the proposed idea.Generally, a Modified Adjacencies Matrix (M AM ) can represent the entire W OG. It is an n×n matrix, with each element equals to: M AM (i, j) = (w i,j rs i,j I i,j , td i,j ), We assume that the CAVE module of the covering node p k can apply the Dijkstra algorithm to evaluate the best path from v k to v D , where v D is the destination node of the vehicle s l using, in general, the Weighted Cost Term (W CT ) associated with each edge (v i , v j ) of path P : where Θ i are weighting terms and k max P , I max P , td max P are the maximum terms evaluated on path P .In this way, more effectiveness to different factors can be given, when choosing the metric that has to be evaluated.The relation 3 i=1 Θ i = 1 should always be satisfied.From Eq. ( 14), we can write the expression of the average W CT for a whole path P : where P represents the length of P expressed in number of edges.So, once the Θ i terms are set, we can evaluate all the possible paths from v k to v d as the The CAVE algorithm obtains a better result in terms of emissions reduction, because it uses the mechanism of the re-routing strategy for the vehicles, in order to spread the mobile nodes in the various available road segments, mitigating the vehicular congestion.Figure 8 shows that the CAVE algorithm reduces the residence time of the vehicles on the map, with a consequent reduction of fuel consuption, emissions and road congestion.This is due to the re-routing of the new itinerary, which brings the drivers to choose less con-gested paths.Figure 9 shows that using the mechanism of vehicles re-routing, the proposed algorithm outperforms the other approaches in terms of an average number of vehicles present on the considered road segments.This is possible because the CAVE algorithm can address the vehicles on the less congested available road segments.

Low Density
Low

Conclusions
In this paper, we proposed a new traffic optimization algorithm, very suitable for vehicular environments.We focused our attention on the reduction of roads density, with the proposal of a map model, based on a weighted oriented graph.The core idea consists in the evaluation of a re-routing strategy, based on the analysis of the roads structure, able to reduce also the CO 2 emissions.We investigated about the effectiveness of the proposed idea, obtaining very satisfactory results in terms of emissions and travel time reduction.

Fig. 4 :
Fig. 4: The classical trend of the function relating density and speed.

Fig. 9 :
Fig. 9: Average number of vehicles on the considered road segment (rs ij ).
Vehicles travel time versus nodal density.