Topology-Age-Aware Cooperative Awareness in Vehicular Ad-Hoc Networks

1

Topology-Age-Aware Cooperative Awareness in Vehicular Ad-Hoc Networks Mostafa Lotfi∗ and Masoumeh Moradian∗ ∗ School of Computer Engineering, K. N. Toosi University of Technology, Iran

arXiv:2608.01026v1 [cs.NI] 2 Aug 2026

Emails: mostafa.lotfi@email.kntu.ac.ir, mmoradian@kntu.ac.ir

Abstract—In vehicular ad-hoc networks (VANETs), maintaining the freshness of status information among vehicles is critical for enabling timely and reliable safety-related decisions. In this paper, we consider the problem of scheduling cooperative status broadcasting in VANETs, where vehicles are allowed to broadcast not only their own updates but also the updates of other vehicles. We propose the age-aware cooperative broadcasting (A-CB) policy, in which a road-side unit (RSU) schedules target-sender pairs based on the total achievable age-gain over the sender’s network links. Moreover, through exploiting the network topology, we introduce two topology-age-aware cooperative broadcasting (TA-CB) policies; the Local TA-CB that weights links according to the number of second-hop neighbors, and the Global TA-CB that leverages the betweenness centrality of vehicles to prioritize broadcasting decisions. The proposed policies are evaluated under both random and clustered network topologies. Numerical results demonstrate that cooperative policies significantly outperform non-cooperative broadcasting. Notably, in clustered network structures, the Local and the Global TA-CB policies achieve substantial improvements over A-CB, with each outperforming the other under different network conditions. Index Terms—Cooperative awareness, age of information, network topology, scheduling, vehicular Ad-Hoc networks, betweenness centrality.

I. I NTRODUCTION The freshness of information has emerged as a critical performance metric in internet-of-things (IoT)-driven technologies such as smart cities, industrial monitoring and intelligent transportation systems (ITS). Specifically, in ITS, the network structure is dynamic and the network entities, such as vehicles, should be continuously updated about the status of surrounding vehicles to make timely and appropriate safetyrelated decisions [1]. This awareness feature is provisioned in cellular vehicle-to-everything (C-V2X) standard, where the vehicles can broadcast their status updates through cooperative aware messages (CAMs), and also issue event-triggered warnings through decentralized environmental notification messages (DENMs) [2], [3], [4]. However, preserving the freshness of the broadcast information across the network remains challenging due to the dynamic environment and relies substantially on applied broadcasting policies. The information freshness is characterized by the age of information (AoI), which is defined as the time elapsed since the generation time of the most recent update at a monitoring node [5]. The AoI-aware broadcasting policies have been studied in different network scenarios including centralized [6], [7] and distributed broadcasting [8] as well as networks with fixed [6], and dynamic structures [9], [8], [10]. In centralized scenarios, where a central unit is responsible

for broadcasting the status updates to a set of clients, either fixed [6], or mobile ones such as vehicles [7], the primary objective is to schedule the transmission and power resources in the central unit such that the expected sum AoI (ESAoI) of the clients is minimized. In the case of a one-hop downlink network with fixed clients and unreliable channels, it is shown that the greedy policy which schedules the client with the highest AoI for transmission, referred to as max-age scheduling, minimizes the ESAoI under symmetric channel conditions. On the other hand, the max-weight and Whittle’s index policies outperform in nonsymmetric conditions [6]. When the clients are moving, such as vehicles in VANETs, the age-optimal policies incorporate the vehicle positions as decision variables as well, thus, leading to more complex strategies that are not necessarily index-based, such as deep reinforcement learning (DRL)-based solutions [7]. In contrast to centralized broadcasting, distributed broadcasting involves multiple sources in the network transmitting their information to nearby entities to keep their surroundings aware of their status [11]. A typical example is cooperative awareness (CA) in modes 1 and 3 of the C-V2X standard [12], [13], [14], where each vehicle, when scheduled by an road-side unit (RSU), broadcasts its own updates (e.g., speed and position) to surrounding vehicles. In such a case, when the RSU and all vehicles are assumed to be within the same transmission range, every entity effectively obtains the same AoI for any given vehicle [15]. In this setting, a max-age scheduling policy, which prioritizes the node with the highest AoI perceived at the RSU, has been proposed to minimize ESAoI [10]. However, when vehicles are moving and not within each other’s transmission range, the AoI of a specific vehicle may differ across other vehicles in the network, as it can only update a limited and varying set of neighbors whenever it is allowed to broadcast. In this case, it is proposed to prioritize those vehicles with higher ESAoI at their neighbors [8]. Nevertheless, because vehicles in a VANET are not in transmission range of each other frequently, cooperative broadcasting, where vehicles also broadcast the updates of other vehicles, can significantly improve the freshness of information across the network compared to the non-cooperative case. This cooperative approach is the main focus of this paper. In this paper, we study centralized AoI-aware scheduling for cooperative distributed broadcasting in VANETs. We consider a scenario in which an RSU schedules vehicles for broadcasting their own information as well as the updates of other vehicles. We propose the AoI-aware cooperative

2

broadcasting (A-CB) algorithm, in which scheduling decisions are taken based on the age-gains, i.e., the reductions in AoIs, achievable over the links of a broadcasting vehicle when it transmits the update of a specific target vehicle. Furthermore, for the first time, we introduce topology-ageaware cooperative broadcasting (TA-CB) policies that explicitly incorporate the network graph topology into scheduling decisions. Specifically, we propose the Local TA-CB and the Global TA-CB policies. Compared to A-CB, the Local TA-CB policy takes into account the second-hop neighborhood of each broadcasting vehicle to assign weight to the vehicles. It is shown that the Local TA-CB, not only decides on instantaneous achieved age-gains but also creates additional opportunities for future AoI reduction. On the other hand, the Global TA-CB policy assigns weights based on the betweenness centrality of vehicles, a global network measure, to prioritize broadcasting decisions. We implement our proposed policies under two network models, including random and clustered. Our results show that the cooperative policies significantly outperform the non-cooperative ones across network models. In particular, the topology-aware policies, Local TA-CB and Global TA-CB, outperform the A-CB in the clustered model, where topological diversity across different parts of the network can be exploited. Our contributions are summarized as follows: We consider a scenario in which a road-side-unit (RSU) schedules vehicles for cooperative awareness at the beginning of each frame. For such a cooperative scenario, where the vehicles cooperate in broadcasting each other’s updates, we propose the A-CB algorithm that assigns indices to the pairs of broadcasting vehicles and the broadcast information based on the age-gain achieved over the outgoing links of the sender vehicle. We demonstrate that A-CB policy significantly outperforms the non-cooperative policy. • We propose the Local TA-CB policy, which assigns weights to broadcasting vehicles based on their potential to create opportunities for reducing the total ESAoI in upcoming frames. In particular, the number of secondhop neighbors of a broadcasting vehicle that are reachable form one of its links is exploited to weigh that link. We also propose the Global TA-CB policy which leverages the betweenness centrality measure to weigh the links, aiming to transmit updates to the distant parts of the network. • Through extensive simulations, we show that all cooperative policies outperform the non-cooperative policy across various network settings. Furthermore, in clustered network structures, both the Local and the Global TA-CB achieve significant improvements over A-CB, with each policy excelling under different conditions, i.e., the Local TA-CB performing better in highly connected regimes, while the Global TA-CB excels in sparse or intermediate connectivity scenarios. These results highlight the importance of incorporating topological information into AoI-aware scheduling for VANETs. We show that, with appropriate parameter settings, the Global TA-CB can outperform the local version, albeit at the cost of higher computational com•

plexity. The rest of this paper is organized as follows. In Section II, we introduce the proposed scenario and the optimization goal. In Section III and IV, we introduce the A-CB and the TA-CB policies, respectively. The numerical results are discussed in Section V. Finally, conclusions are drawn in Section VI along with potential future directions.

A. Literature review The problem of update broadcasting in downlink one-hop networks is studied in [6], [7]. In [6], the authors study a scenario in which a BTS, having the updates of a set of fixed clients at the onset of time frames, is supposed to select a subset of clients for updating. The authors show that the max-age scheduling minimizes ESAoI for symmetric clients, while the max-weight and Whittle’s index policies outperform in nonsymmetric channel conditions. In [7], the RSU is responsible for broadcasting the updates of a set of physical processes to the clients, where clients are vehicles on a road. A DRL-based algorithm is proposed to optimize the transmission mode and power of the RSU while minimizing the ESAoI of vehicles. On the other hand, age-aware broadcasting in VANETs has also been studied in several works [9], [8], [10], [14]. In [9], CA is performed in a distributed manner, where each vehicles broadcast its own updates periodically according to the IEEE802.11p standard, a random access protocol for direct short-range communications in V2X. The authors in [10] propose both centralized and decentralized AoI-based broadcast protocols for VANETs. In the centralized version, the RSU selects the vehicles with maximum perceived AoIs at RSU for broadcasting over their selected channels, assuming that all vehicles and RSU are in transmission range of each other. In [8], the vehicles are not in the same transmission range and the RSU selects those vehicles that the weighted ESAoIs in their neighbors are larger for broadcasting, where the weights are the transmission success probabilities over links. The randomized CA in VANETs can also be considered as a special case of gossip networks. In gossip networks, the nodes transmit their available information randomly to their neighbours with the goal of maintaining the freshest information. In [16], the authors study the version AoI in a gossip network with one information source and n nodes, where the update transmission process between the source and other nodes and also among the neighbouring nodes happens according to independent Poisson processes. The information freshness in gossip networks has also been studied under various scenarios, such as the presence of unreliable sources [17] and transmission mutation [18], albeit assuming age-independent spreading processes at the nodes. However, in [19], due to the limited updating rate in the network, each node waits for a random time proportional to its age before updating its neighbors, thus, giving the nodes with lower AoI more opportunities to disseminate their information. In [20], the average and variance of the AoI have been derived in gossip networks with linear, ring, and star structures. Here, the nodes transmit the information of other nodes

3

in addition to their own information according to a timeinvariant distribution. In the aforementioned works except [20], the nodes broadcast only their own information and do not cooperate in transmitting the updates of other nodes. Furthermore, in [20], the cooperation is done in a distributed manner, and the cooperation process is irrespective to the current AoI of the nodes. Finally, None of the above works consider centralized scheduling in gossip networks.

computing the target-sender pairs at the beginning of each frame. Finally, it is assumed that the update of any vehicle is broadcast only once within each frame. For ease of notation, hereafter, the indices k, i, and j are used for the target, sender, and receiver vehicles, respectively. The AoI of vehicle k at vehicle j at the beginning of frame t, denoted by Xkj (t), is written as

II. S YSTEM M ODEL

where t is the index of the current frame and ukj (t) is the frame index at which the most recent update of vehicle k available at vehicle j is generated. Furthermore, Xkj (t) evolves as follows:   Xki (t) + 1; ∃l : (k, i) ∈ Bl (t), j ∈ Ni (t), Xkj (t + 1) = Xki (t) < Xkj (t),   Xkj (t) + 1; otherwise. (2) The first case of (2) indicates that vehicle j receives an update of target vehicle k in frame t only if (k, i) is a target-sender pair in this frame, j is a neighbor of i, and the broadcast update by vehicle i is fresher than the available one at vehicle j, i.e., Xki (t) < Xkj (t). Otherwise, Xkj (t) only increases by one, as stated in the second case of (2), where one accounts for one frame duration. It is worth noting that (2) is written based on the assumption that the information of any target vehicle k is updated only once in a frame. ¯ is written as: Using Xkj (t), the ESAoI, denoted by ∆, PT PN E[Xkj (t)] ¯ = limT →∞ 1 t=1 k,j=1 . (3) ∆ 2 T N In the above equation, it is assumed that the vehicles remain in the coverage area of the RSU for a sufficiently long time such that the time-average AoI can be derived regarding (3). Moreover, the expectation is taken over the random movement of the vehicles, i.e., random G(t). Regarding that the RSU has access to G(t), it is capable of tracking the changes in Xkj (t), ∀k, j, according to (2), and thus, has the updated values of AoIs at the beginning of each frame. Then, it chooses the target-sender pairs in each frame, such that the ESAoI is minimized. Consequently, the optimization problem is written as

We consider a scenario in which an RSU is responsible for scheduling the broadcast transmissions of the vehicles through its coverage area. The set of vehicles in this area is assumed to be fixed, and is indicated by V = {1, 2, · · · , N }. Moreover, when allowed by RSU, each vehicle can broadcast either its own information or the information of other vehicles. The broadcast content includes vehicles’ status updates such as location and speed updates as well as emergency alarms. We assume a slotted time structure, where all vehicles are time-synchronous. Each slot duration is equal to the time required for broadcasting one single update, which is identical for all vehicles. Moreover, every L slots constitute a frame. It is assumed that the relative positions of the vehicles do not change within a frame. Furthermore, due to limited transmission power, each vehicle has a limited communication range. As a result, the vehicular network during frame t is represented by the graph G(t) = (V, E(t)), where E(t) indicates the edge set and (i, j) ∈ E(t) implies that vehicles i and j are within each other’s communication area during frame t and can successfully transmit updates to each other. Moreover, A(t) denotes the adjacency matrix of G(t), where Aij (t) = 1 if (i, j) ∈ E(t) and zero otherwise. At the beginning of each frame t, the RSU receives the position information of all vehicles through an uplink control channel and thus, is able to construct G(t). Based on G(t) and obtained AoI information of the vehicles (discussed later in this section), RSU broadcasts the vector B(t) = (B1 (t), B2 (t), · · · , BL (t)) to all vehicles, where the entry Bl (t) specifies the list of broadcasting vehicles in the l-th slot of frame t as well as their broadcasting information. In particular, we refer to a broadcasting vehicle as the sender and the vehicle whose information is being broadcast by the sender as the target. Then, Bl (t) is written as Bl (t) = {(T1 , S1 ), (T2 , S2 ), ..., (Tm , Sm )}, where the ordered pair (Tx , Sx ), x ∈ {1, · · · , m} is an authorized target-sender pair in slot l, i.e., vehicle Sx broadcasts the update of vehicle Tx in that slot. Whenever Sx broadcasts the update of Tx , all neighbors of Sx , denoted by NSx (t), receive the update of vehicle Tx and replace any stale information they have for vehicle Tx with the newly received one. We also consider the following assumptions. First, we assume that any two senders scheduled to transmit in the same slot should not be neighbors or have any neighbors in common, otherwise their transmission will interfere with each other or at their common neighbors, respectively. Second, the number of broadcasting vehicles per slot is assumed to be limited, i.e., |Bl (t)| ≤ M, ∀t, l, in order to control the signaling overhead at RSU as well as the complexity of

Xkj (t) = t − ukj (t).

(1)

¯ min ∆

s.t.

B(t),∀t ′

k ̸= k , ′ ′ ∀t, ∀(k, i), (k ′ , i′ ) ∈ ∪L l=1 Bl (t), (k, i) ̸= (k , i ), (4a)

Ni (t) ∩ Ni′ (t) = ∅, i ∈ / Ni′ (t) ∀l, t, ∀(k, i), (k ′ , i′ ) ∈ Bl (t), i ̸= i′ , |Bl (t)| ≤ M,

∀l, t,

(4b) (4c)

where Ni (t) denotes the set of neighbors of vehicle i in frame t. Constraint (10a) guarantees that each target vehicle update is broadcast at most once within a frame. Constraint (10b) prohibits interfering senders, i.e., those with common neighbors, to transmit in the same slot. Finally (10c) bounds the number of senders in each slot to M . The above optimization problem can be formulated within an infinite-horizon Markov decision process (MDP) frame-

4

work. More specifically, according to standard MDP terminology, the MDP problem is characterized by the following components: State space: The state is denoted by S(t) = (G(t), X(t) = [Xij (t)]). Unlike X(t), which has a fixed dimension of N 2 , G(t) can be represented by its edge set E(t). Consequently, its dimension is of order O(N d), where d is the average degree of the network. As d grows to O(N ) in dense scenarios, the state dimension can reach O(N 2 ), further exacerbating the curse of dimensionality. Action space: The action in each slot consists of selecting at most M non-interfering senders as well as their corresponding targets. It is worth noting that for two vehicles to be non-interfering, their distance in graph G(t) must be at least three so that they are neither adjacent nor share a common neighbor. Thus, the set of non-interfering senders in each slot corresponds to a set of vertices with pairwise distance of at least three, where such a set in graph theory is referred to as a 2-independent set of G(t). Hence, the action space comprises all subsets of size at most M from any 2-independent set of G(t), paired with M arbitrary target vehicles. However, enumerating all 2-independent sets is generally an NP-hard problem, rendering the action space prohibitively large. Transition probabilities: The transition probability between states is determined by the mobility pattern of the vehicles, i.e., evolution of G(t) across time, as well as the changes in X(t), which happens according to (2). Reward: The reward obtained in frame t is the average P Xkj (t) . AoI of all vehicles, i.e., k,iN 2 As can be seen, the MDP formulation suffers from the curse of dimensionality, arising from both the state-space size, which is of order O(lN 3 ), and the combinatorially large action space. To address this, we propose two lightweight ESAoI-minimizing algorithms, namely, AoI-aware cooperative broadcasting (A-CB), and two Topology-AoIaware cooperative broadcasting (TA-CB) that take advantage of age-gain minimization as well as the underlying vehicular network topology. III. AO I- AWARE C OOPERATIVE B ROADCASTING (A-CB) In this section, we first introduce the age-gain of a target over a specific link as well as the age-gain of a target-sender pair. Then, we propose the A-CB algorithm. The age-gain of the target vehicle k over the link (i, j) ∈ (k) E(t), denoted by Qij (t), is defined as the AoI reduction of vehicle k at vehicle j, given that vehicle i broadcasts the (k) update of vehicle k. Thus, Qij (t) is written as: (k)

Qij (t) = {Xkj (t) − Xki (t)}+ ,

(5)

where {x}+ = max{0, x}, and is applied in (5) since the gain is achieved only if the update of the broadcasting vehicle i is fresher than that of receiving vehicle j, i.e., Xki (t) ≤ Xkj (t). Also, the age-gain of the target-sender pair (k, i) at the beginning of frame t, denoted by Q(k,i) (t), is defined as the sum of age-gains over vehicle i’s links. Thus, Q(k,i) (t) is computed as: X (k) Q(k,i) (t) = Qij (t). (6) j∈Ni (t)

Now, we derive the sum AoI (SAoI) reduction in frame t, i.e., 1 X Xij (t) − Xij (t + 1), (7) N 2 ij in terms of the introduced age-gains. Let us define P(t) as P(t) = {(k, j)|∃i : (k, i) ∈ ∪L l=1 Bl (t), i ∈ Nj (t)}.

(8)

In fact, P(t) indicates all pairs of vehicles (k, j), where vehicle j is being updated about target vehicle k during frame t. Then, the SAoI in frame t is derived as (the fixed term 1 N 2 is omitted): X

(a)

Xkj (t) − Xkj (t + 1) =

k,j

X (k,j)∈P(t) /

(b)

X

−1 +

Xkj (t) − Xkj (t + 1) =

(k,j)∈P(t)

|P(t)| − N 2 +

L X X X

(c)

{Xkj (t) − Xki (t)}+ − 1 =

j l=1 (k,i) ∈Bl ∈Ni (t)

− N2 +

L X X

(k)

X

(d)

Qij (t) =

l=1 (k,i)∈Bl j∈Ni (t)

− N2 +

L X X

Q(k,i) (t),

l=1 (k,i)∈Bl

(9) where equality (a) is written regarding the fact that Xkj (t) for a vehicle j that is not updated about vehicle k in frame t, i.e., (k, j) ∈ / P(t), increases by one, i.e., Xkj (t) − Xkj (t + 1) = −1. Also, equality (b) is written since for (k, i) ∈ P(t), we have Xkj (t + 1) − Xkj (t) = {Xki (t) − Xkj (t)}+ − 1, according to (2). Finally, equalities (c) and (d) are written using (5) and (6), respectively. In the A-CB algorithm, the RSU exploits a greedy approach, i.e., based on the information about G(t) and X(t), it selects those target-sender pairs that maximize the in (9), or equivalently maximize PL SAoI P reduction (k,i) Q (t). Suppose that y (k,i) (l) is equal to l=1 (k,i)∈Bl one if the target-sender pair (k, i) is allowed to transmit in the l-th slot in frame t, and zero otherwise. Then, the optimization problem of A-CB in frame t is written as L X X

max y (k,i) (l)

s.t.

y (k,i) (l) Q(k,i) (t)

l=1 (k,i)

L X N X

y (k,i) (l) ≤ 1,

l=1 i=1 X (k,i)

(10a)

∀l

(10b)

y (k,i) (l)y (k ,i ) (l) = 0,

∀l, (k, i) ̸= (k ′ , i′ ), di,i′ ≤ 2 (10c)

y

(l) ≤ M,

∀k

(k,i) ′

′

where di,i′ is the distance between vehicles i and i′ in frame t. The first constraint in (10) indicates that each vehicle status is updated at most once in a frame, while the second constraint states that the maximum number of transmissions

5

inside a slot is M . Also, the last constraint ensures selection of non-interfering senders inside a slot. It is worth noting that if we set M = 1, then (10c) becomes irrelevant and the answer to (10) is to assign the L sender-target pairs with maximum age-gains and distinct targets to L slots in each frame. However, if M > 1, then (10c) comes to the picture and the optimization (10) becomes a quadratically constrained integer program, which requires the information of the graph structure G(t) to establish the quadratic constraints in (10c). These constraints increase the complexity of solving the problem, especially in the case of connected and dense graphs. In this regard, we propose A-CB policy in Algorithm 1 to solve (10). In the A-CB policy, a pool of eligible target-sender pairs is first initialized for each slot l, denoted as P ooll , ∀l, containing all possible sender-target pairs (line 1). Additionally, a list S is constructed, which sorts all possible target-sender pairs in descending order of their Q(k,i) values (line 2). Then, in each iteration, the first element of S, say (k, i), is selected (line 5) and assigned to the first slot l that satisfies two conditions. First, (k, i) is eligible in slot l, i.e., (k, i) ∈ P ooll ; and second, the capacity M of slot l has not yet been reached (lines 8,9). If such a slot exists, all targetsender pairs involving target k are removed from all pools, because each target is allowed to be updated only once per frame (line 10). Moreover, pairs with interfering senders, i.e., ∀i′ : di,i′ ≤ 2 are removed from P ooll (line 11). If the pair (k, i) cannot be assigned to any slot, it is removed from the head of S (lines 16-18). This process continues until every slot either has reached its capacity or has an empty pool, or until the list S becomes empty (i.e., the conditions in line 4 are met). IV. T OPOLOGY-AO I-AWARE C OOPERATIVE B ROADCASTING (TA-CB) In calculating the age-gain of a target-sender pair, the A-CB algorithm assigns equal weights to the age-gains associated with a sender’s link, as in (6). However, the neighbors of a sender have different structural positions in the network and thus, can affect the dissemination of the information differently in consequent frames. In this regard, the main idea in the TA-CB algorithm is to take advantage of the topological position of the vehicles in the network to create more opportunities for fresh information dissemination in the upcoming frames. We propose two versions of TACB, namely the Local TA-CB and the Global TA-CB, where the former relies on local topological information of each vehicle, while the latter exploits the global network properties. A. Local TA-CB In order to clarify the idea of the Local TA-CB, we bring an example. Consider Fig.s 1(a) and (b) that illustrate two different neighborhoods of sender vehicle 0, respectively. In this figure, the link age-gains are shown for a specific (k) target vehicle k. Consequently, Qij is denoted by xij and is shown over the link (i, j). It is also assumed that vehicle 0 has fresher update than its neighbour vehicles 1 and 2,

Algorithm 1 A-CB Algorithm Require: ∀(k, i) : Q(k,i) (t), G(t), M , L Ensure: Set of target-sender pairs assigned to each slot 1: Initialize P ooll = V × V for 1 ≤ l ≤ L {eligible pairs per slot} 2: S ← sorted list of V × V in descending order of Q(i,k) 3: n(l) ← 0 for 1 ≤ l ≤ L {number of assigned pairs in slot l}
4: while ∃l : P ooll ̸= ∅ ∧ n(l) < M ∧ (S ̸= ∅) do 5: (k, i) ← S[1] {head of sorted list} 6: f lag ← 0 7: for l = 1 to L do 8: if (k, i) ∈ P ooll then 9: Assign (k, i) to slot l 10: n(l) ← n(l) + 1 11: Remove all (k, i′ ) from P ooll′ for every l′ {no other sender can send update of vehicle k} 12: Remove all (k ′ , i′ ) with di,i′ ≤ 2 from P ooll {interference constraint} 13: f lag ← 1 14: break 15: end if 16: end for 17: if f lag = 0 then 18: Remove (k, i) from the head of S 19: end if 20: end while

i.e., x01 , x02 > 0. Moreover, G(t) is assumed to remian unchanged in two consecutive frames. In both scenarios in Fig.s 1(a) and (b), when vehicle ν0 broadcasts the update of vehicle k, the associated age-gains over its own links as well as the links between its neighbors are reset to zero, as all neighbors obtain the same update of vehicle k. However, in Figure 1(b), where vehicle 0 has second-hop neighbors, i.e., vehicles {3, 4, 5, 6}, the age-gains of the second-hop links also change. As such, the age-gain of the link (1, 3) is incremented by x01 , while the agegains of links (2, 4), (2, 5), and (2, 6) are all incremented by x02 . Thus, when a sender broadcasts an update, the agegain of each of its links are added to the corresponding second-hop links, thus, incrementing the age-gains over these links. This phenomenon of age-gain spreading provides the opportunity of having higher target-sender age-gains in frequent frames, and thus, increasing the chance of obtaining larger AoI reductions, given that the network does not change significantly in near future. Note that, the A-CB algorithm does not differentiate between the structures in Fig.s 1 (a) and (b) when computing the age-gain of the target-sender pair (k, 0) since in both cases we have Q(k,0) = x01 + x02 . In this regard, we propose the Local TA-CB algorithm which takes into account the second hop links of each sender to determine the broadcasting target-sender pairs in each frame. More precisely, according to the Local TA-CB algorithm, the age-gain of target-sender pair (k, i) is redefined as: X (k) Q(k,i) = wij (t)Qij (t), (11) j∈Ni

6

TABLE I D EFAULT SIMULATION PARAMETERS . Parameter Number of vehicles, N Number of slots per frame, L Maximum senders per slot, M Number of clusters, Nc Graph update interval, Ng α in (16)

Default Value 20 4 2 3 10 1

of vehicle j in frame t, denoted by Bj (t), is defined as [21]: X nsp (j, (x, y)) , (15) Bj (t) = nsp (x, y) 2 (x,y)∈V

Fig. 1. In graphs of (a) and (b), ν0 broadcasts the update of target vehicle (k) k to its neighbors, where xij = Qij is the age-gain of link (i, j) with respect to target k. In (a), vehicle ν0 has no second-hop neighbors. In (b), the age-gains are added to the gains of the second-hop links.

(k)

where Qij (t) is defined as in (5). Also, the weight wij (t) associated to link (i, j) ∈ E(t) is defined as (2)

wij (t) = 1 + Ni→j (t)

(12)

(2) where Ni→j (t) is the number of second-hop neighbors of

vehicle i that are neighbors of vehicle j as well. In other (2) (2) (2) words, Ni→j (t) = |Ni→j (t)|, where Ni→j (t) is the set of second-hop neighbors of vehicle i via vehicle j: (2)

Ni→j (t) = {m ∈ V|m ∈ Nj (t), m ∈ / Ni (t)}.

(13)

It is worth noting that regarding (11) and (12), we have X (k) X (2) (k) Q(k,i) = Qij (t) + Ni→j (t)Qij (t) (14) j∈Ni

where nsp (x, y) is the total number of shortest paths between vehicles x and y and nsp (j, (x, y)) is the number of those paths that pass through node j. In fact, the more shortest paths that traverse node j, the more critical its role in connecting different parts of the network. Consequently, we define wij (t), as follows: wij (t) = 1 + αBj (t).

(16)

In the above equation, each link weight is inflated by αBj (t) compared to A-CB, thereby giving more weight to the links that terminate at centrally located vehicles. Also, as α increases the importance of transferring information to other parts of the network becomes more than reducing the age over local link (i, j). Unlike the Local TA-CB method, the Global TA-CB method requires the entire graph structure to compute Bj (t) and thus, is computationally more expensive. In Section V, we show that each of these two policies outperforms the other under certain circumstances.

j∈Ni

where the first term is the age-gain achieved for target vehicle k in the current frame as a result of having vehicle i as the (2) (k) sender. Furthermore, Ni→j (t)Qij (t) in the second term is the improvement in Q(k,j) , since as illustrated in Fig. 1, the (k) age-gain Qij is added to the age-gain of all links between (2) j and the vehicles in Ni→j (t). Thus, the Local TA-CB prioritizes vehicles that not only achieve higher age-gains in current frame but also lead to higher age-gain improvements in subsequent frames. Albeit, this assumption is justified by the fact that the graph G(t) remains unchanged over multiple frames, which is realistic in practical vehicular scenarios. The optimization of the Local TA-CB is written similarly to (10) except that Q(i,k) (t) is defined according to (11). Moreover, the algorithm of the Local TA-CB is exactly as Algorithm 1, except that in line 2, the list S is sorted according to Q(i,k) (t) in (11).

B. Global TA-CB In the Global TA-CB algorithm, rather than weighting a sender’s links based on their second-hop neighbors as in (12), we consider the capability of vehicles to effectively transfer the information to more distant parts of the network. In network analysis terminology, this property is captured by the betweenness centrality measure. Accordingly, the centrality

V. N UMERICAL R ESULTS A. Simulation setup In this section, we evaluate our proposed algorithms A-CB, Local TA-CB, and Global TA-CB against each other and an age-aware non-cooperative broadcasting policy, referred to as N-CB [8], under different network settings. N-CB behaves similarly to A-CB, except that the vehicles only broadcast their own information. Moreover, the graph G(t) is generated according to two models; First, the standard Erdos-Renyi (ER) model, in which each link is independently created according to a fixed probability pl . Second, a clustered model in which N vehicles are partitioned into Nc equalsized non-overlapping consecutive clusters along the road, with intra-cluster connections constructed according to the ER model. Additionally, each cluster is connected to the subsequent cluster via a single inter-cluster link, simulating the platoon-like structures. Also, in order to model realistic mobility patterns, the graph topology G(t) is updated every Ng frames. The default values of parameters are considered as in Table I, unless otherwise stated. In our simulations, we mainly focus on the clustered structure since, as will be seen later, the TA-CB policies outperform A-CB under such topologies. Fig. 2 illustrates an example of a clustered graph with three clusters, where the vehicles that are serving as the endpoints of the intercluster links are referred to as connecting vehicles. In the

7

Fig. 2. An example of clustered graph with three clusters. The patterned circles show the connecting vehicles.

following, we analyze the effect of pl , i.e., the probability of intra-cluster link construction, on the weights wij (t) in the Local and the Global TA-CB policies, respectively, as this analysis is useful for interpreting the simulation results presented later. We denote the size of c-th cluster by nc . Consider a link (i, j) ∈ E(t), where vehicle i is the sender. Also, suppose that neither i nor its neighbor j is a connecting vehicle. Then, the number of second-hop neighbors of i (2) via j, i.e., Ni→j (t) in (12), follows a binomial distribution Bin(nc − 2, pl (pl − 1)). This is due to the fact that any other vehicle in the same cluster is a neighbor of j, or equivalently a second neighbor of i, with probability pl , and is not a neighbor of i, with probability 1 − pl . Thus, the expected value of wij (t), defined in (12), is E[wij (t)] = 1 + (nc − 2)pl (1 − pl ).

(18)

which increase monotonically with pl . Finally, if i is not a (2) connecting vehicle but j is, we have Ni→j (t) = 1 and thus: E[wij (t)] = wij (t) = 2.

more likely available. In particular, in the extreme case of pl = 1, all vehicles except the connecting ones have zero betweenness centrality, since the graph inside each cluster is complete and the two connecting vehicles are directly connected.

(17)

It is concluded from (17) that E[wij (t)] increases with pl initially but eventually decreases since the cluster becomes densely connected, approaching to a complete graph, as pl increases. Now suppose that link (i, j) is an inter-cluster link, i.e., vehicle i belongs to cluster c and vehicle j belongs (2) to cluster c + 1. In this case, Ni→j (t) follows binomial distribution Bin(nc+1 − 1, pl ) since any vehicle in the next cluster can be a second-hop neighbor of vehicle i if it is a neighbor of j (with probability pl ). Thus, the expected weight becomes E[wij (t)] = 1 + (nc − 1)pl ,

Fig. 3. The ESAoI versus the link construction probability, pl , in random model (L = 4, M = 2, Ng = 10).

(19)

Hence, it can be concluded that in the Local TA-CB policy, for equal age-gains, the connecting vehicles are assigned higher weights due to their inter-cluster links. In particular, when pl is larger, this weight disparity becomes more noticeable. On the other hand, in the Global TA-CB policy, the connecting vehicles possess the largest betweenness centrality values since any path from their cluster to the subsequent clusters must pass through them. Moreover, when pl is small, the clusters are sparse, and the non-connecting vehicles inside intermediate clusters exhibit higher betweenness centrality compared to those in peripheral clusters. This is because they lie on paths between the two connecting vehicles of their cluster. However, as pl increases, the clusters become more densely connected, and thus, the betweenness centralities of vehicles inside the intermediate clusters decrease as the direct link between two connecting vehicles of their clusters is

B. Simulation results Fig. 3 depicts the ESAoI versus the link probability pl for the random topology, under the N-CB, A-CB, Local TACB, and Global TA-CB policies. As observed, in cooperative policies A-CB and TA-CB, the ESAoI is significantly lower than non-cooperative policy N-CB since cooperation enables vehicles to transmit each other’s updates, thereby disseminating status information more broadly across the network and reducing the ESAoI. Moreover, under all policies, ESAoI decreases with pl since higher connectivity increases the likelihood that vehicles lie within each other’s communication range, and thus, receive fresher updates of each other. Furthermore, the Local and Global TA-CB do not perform better than the A-CB in the random graph since the TA-CB policies rely on the topological differences of vehicles to act, while in the random graph all vehicles have the same topological position. In Fig. 4, the percentage of ESAoI improvement in the Local and Global TA-CB over A-CB policy are shown versus the intra-cluster link probability pl , for different numbers of clusters Nc . It is observed in Fig. 4 that in general, the Local and Global TA-CB policies outperform A-CB at all values of pl . However, the improvement achieved by the Local TACB remains almost low, i.e., less than 3 percent, when pl is less than a certain threshold and increases with pl beyond that point. In particular, this threshold is approximately 0.3, 0.5, and 0.6, for Nc = 2, 3, and 4, respectively. In fact, when pl is small, as concluded in Section V-A, the inter-cluster links are not significantly prioritized over other links. Besides, the network remains sparse within clusters. Thus, even if an update is forwarded to other clusters via inter-cluster links, those clusters cannot effectively propagate information further due to their sparse internal connectivity.

8

Fig. 4. The improvement in ESAoI versus the intra-cluster link construction probability, pl , at different values of Nc (L = 4, M = 2, Ng = 10).

Fig. 5. The improvement in ESAoI versus the intra-cluster link construction probability, pl , at different values of M (L = 4, Nc = 3, Ng = 10).

However, when pl increases beyond the threshold, the intercluster links receive higher weights and the clusters become sufficiently connected to spread the information effectively. Consequently, the improvement increases with pl . On the other hand, the Global TA-CB policy exhibits a steady increase in improvement with pl , even at low values. This is because in the Global TA-CB, when pl is small, the vehicles in intermediate clusters have higher betweenness centrality than those in peripheral clusters, as noted in Section V-A. Thus, they are prioritized for update broadcasting, leading to a steady increase in improvement with pl . Also, as pl becomes larger, though the priority of intermediate vehicles diminishes, the increased connectivity of the network compensates for this effect and the outperformance continues to grow. It is also observed in Fig. 4 that for an intermediate range of link probabilities pl , the Global TA-CB outperforms the Local TA-CB. This is due to the reason that as explained earlier, the Global TA-CB policy assigns higher betweenness centrality to intermediate vehicles at low pl , thereby prioritizing them for status updates. At higher values of pl , however, the Local TA-CB significantly outperforms the global one. In fact, at these values, both policies essentially treat intracluster links across different clusters similarly, i.e., give lower weight to them compared to inter-cluster links, which is due to the similar topological structure of the vehicles within clusters in the Local TA-CB, and low betweenness centrality in the Global TA-CB. Nevertheless, the inter-cluster links in the Local TA-CB policy are weighted significantly higher than in the Global TA-CB since the weight include the number of second-hop neighbors which is large for intercluster links in connected regime of the network. Consequently, the connecting links are selected more frequently for broadcasting, leading to greater improvement in the Local TA-CB. As an example the improvement is 9% and 18% higher compared to the Global TA-CB at pl = 0.9 and pl = 1, respectively. Another key observation in Fig. 4 is that as the number of clusters Nc increases, the improvement of both the Local

and Global TA-CB over A-CB degrades. This is due to the fact that, given a fixed N , the number of vehicles in each cluster decreases, thus, the topological distinctiveness between vehicles within clusters and the connecting ones diminishes, making the weights wij (t) less useful in the Local TA-CB. Also, in the Global TA-CB, the betweenness centrality of the connecting vehicles approaches that of the internal cluster vehicles, leading to degradation of the performance. Fig. 5 depicts the improvement in the Local and Global TA-CB over A-CB in the clustered model, for different numbers of transmissions per slot, M . In general, the ESAoI decreases with M under all policies and across all values of pl since more broadcasting opportunities are provided in each slot, though marginal improvement diminishes as M increases due to the interfering senders constraint in (10c). However, the amount of improvement of the Global TACB over A-CB, as well as that of the Local TA-CB for pl ≥ 0.7, decreases with M . This is because as M increases, the base-line policy A-CB also benefits from the additional broadcasting opportunities, thus narrowing the performance gap that TA-CB policies can achieve. The opposite behaviour is observed for the Local TA-CB policy when pl ≤ 0.7, where the improvement grows with M , which is justified as follows. At low pl , the network is sparse and the number of second-hop neighbors is very limited. As a result, any broadcasting primarily benefits its first-hop neighbors, without increasing the age-gains of other links (please see Fig. 1) that could otherwise be exploited later to further reduce ESAoI as M increases. In this regime, the Local TA-CB can achieve greater improvement over A-CB as M increases since it prioritizes vehicles with more second-hop neighbors. In Fig. 6 the improvement in the Local and Global TACB over A-CB is shown versus pl for different number of frame lengths, L. The behaviour of curves is similar to those observed in Fig. 5. Here, increasing L, similar to increasing M , provides more broadcasting opportunities. Thus, all behaviours can be justified similar to those in Fig. 5. The only difference is that here, the improvements

9

Fig. 6. The improvement in ESAoI versus the intra-cluster link construction probability, pl , at different values of L (Nc = 3, M = 2, Ng = 10).

Fig. 7. The improvement in ESAoI versus the number of vehicles, N , in the clustered model, at different values of M (L = 4, Nc = 3, Ng = 10, pl = 0.8).

at different values of L are closer to each other, e.g., at pl = 0.9, all percentages lie within the range of 11% and 12.5%. The reason is that, while increasing M introduces additional senders within the same slot, which must satisfy the interfering-sender constraint, increasing L adds new slots that are not subject to this constraint. As a result, the additional opportunities provided by larger L can be more effectively exploited, leading to sustained improvement and closer performance across different frame lengths. Fig. 7 illustrates the percentage of improvement of the Local and Global TA-CB policies over A-CB versus the number of vehicles, N , for different number of clusters, Nc , at pl = 0.8. When N is small (less than 10), the size of clusters is too small to exhibit a meaningful clustered structure, and thus, the TA-CB policies make little or no improvement over A-CB. However, as N increases, the clusters become more crowded, enhancing the importance of inter-cluster links in disseminating information. These connecting links gain more second-hop neighbors in local

Fig. 8. The improvement in ESAoI versus coefficient α in the clustered model, at different values of M (L = 3, Ng = 10, pl = 0.8).

TA-CB and larger betweenness centrality measures in the Global TA-CB, both of which play a decisive role in their respective policies. Also, since pl = 0.8 corresponds to a densely connected regime, the local TA-CB policy transfers updates more effectively across clusters, consistent with the observations made earlier. In Fig. 8, the effect of α in (16) is examined on the improvement of the Global TA-CB policy over A-CB, at pl = 0.8. The improvement of local TA-CB, which is independent of α is also shown for the sake of comparison. It is observed in this figure that as α increases from one, the Global TA-CB improvement surpasses that of local TACB. This is due to the fact that the links ending at highcentrality vehicles are weighted more, resulting in faster information dissemination across clusters. However, as α exceeds a threshold, the improvement degrades. This is because transferring information to distant parts of the network are so highlighted that the current age-gain reduction over firsthop links in the current broadcast, i.e., equivalent to the first term in (16), are overlooked. This leads to not effectively transmitting the fresher information. Also, it is worth noting that at the optimal value α = 10 at Nc = 2, the Global TA-CB outperforms the Local TA-CB by approximately 4%, albeit at the cost of higher computational complexity for centrality computation. VI. C ONCLUSION In this paper, we studied the problem of AoI-aware scheduling for cooperative distributed broadcasting in VANETs, where vehicles cooperate by broadcasting not only their own status updates but also those of other vehicles. We considered a centralized framework in which an RSU schedules target-sender pairs at the beginning of each frame based on the network topology and the current AoI information of the vehicles. We proposed three scheduling policies. First, the A-CB algorithm, which prioritizes target-sender pairs based on the total age-gain achieved over the links of each sender. Second, the local TA-CB policy, which additionally accounts for the second-hop neighborhood of each sender to weigh

10

its links, thereby creating more opportunities for future AoI reduction. Third, the Global TA-CB policy, which leverages the betweenness centrality of vehicles to prioritize links that terminate at structurally important nodes, facilitating faster information dissemination across the network. Through extensive simulations under random and clustered topologies, we demonstrated that cooperative policies significantly outperform non-cooperative N-CB, and that TA-CB policies achieve substantial gains over A-CB in clustered networks, with local TA-CB excelling in densely connected regimes and the Global TA-CB in sparse or intermediate connectivity scenarios, though at the cost of higher computational complexity for centrality computation. This highlights the importance of topology-aware scheduling in VANETs and the trade-off between performance and complexity. Future research directions include distributed scheduling, learningbased adaptations, and extensions to other AoI metrics. R EFERENCES [1] R. Das, L. Zhou, S. Bi, T. Wang, and T. Hou, “Towards the safety of intelligent transportation: A survey on the security challenges and mitigations in internet of vehicles (iov),” ACM Transactions on Sensor Networks, vol. 22, no. 3, pp. 1–44, 2026. [2] S. Chen, J. Hu, Y. Shi, and L. Zhao, “Lte-v: A td-lte-based v2x solution for future vehicular network,” IEEE Internet of Things journal, vol. 3, no. 6, pp. 997–1005, 2016. [3] G. P. W. NBA, J. Haapola, and T. Samarasinghe, “A discrete-time markov chain based comparison of the mac layer performance of c-v2x mode 4 and ieee 802.11 p,” IEEE Transactions on Communications, vol. 69, no. 4, pp. 2505–2517, 2020. [4] G. Kueppers, J.-P. Busch, L. Reiher, and L. Eckstein, “V2aix: A multi-modal real-world dataset of etsi its v2x messages in public road traffic,” in 2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2024, pp. 392–398. [5] S. Kaul, R. Yates, and M. Gruteser, “Real-time status: How often should one update?” in 2012 Proceedings IEEE INFOCOM. IEEE, 2012, pp. 2731–2735. [6] I. Kadota, A. Sinha, E. Uysal-Biyikoglu, R. Singh, and E. Modiano, “Scheduling policies for minimizing age of information in broadcast wireless networks,” IEEE/ACM Transactions on Networking, vol. 26, no. 6, pp. 2637–2650, 2018. [7] A. Al-Habob, H. Tabassum, and O. Waqar, “Dynamic unicast-multicast scheduling for age-optimal information dissemination in vehicular networks,” in 2022 IEEE Globecom Workshops (GC Wkshps). IEEE, 2022, pp. 1218–1223. [8] Y. Ni, L. Cai, and Y. Bo, “Vehicular beacon broadcast scheduling based on age of information (aoi),” China Communications, vol. 15, no. 7, pp. 67–76, 2018. [9] A. Baiocchi and I. Turcanu, “A model for the optimization of beacon message age-of-information in a vanet,” in 2017 29th International Teletraffic Congress (ITC 29), vol. 1. IEEE, 2017, pp. 108–116. [10] H.-C. Jang and C.-Y. Huang, “Age-of-information-based transmission protocol in vehicular network,” Engineering Proceedings, vol. 55, no. 1, p. 87, 2024. [11] Y. Meng, M. A. Naeem, R. Ali, Y. B. Zikria, and S. W. Kim, “Dcs: Distributed caching strategy at the edge of vehicular sensor networks in information-centric networking,” Sensors, vol. 19, no. 20, p. 4407, 2019. [12] V. Vukadinovic, K. Bakowski, P. Marsch, I. D. Garcia, H. Xu, M. Sybis, P. Sroka, K. Wesolowski, D. Lister, and I. Thibault, “3gpp c-v2x and ieee 802.11 p for vehicle-to-vehicle communications in highway platooning scenarios,” Ad Hoc Networks, vol. 74, pp. 17– 29, 2018. [13] M. A. Ali, S. A. Khan, S. Aldirmaz Colak, S. Kosunalp, and T. Iliev, “Towards 6g c-v2x networks: A comprehensive survey on mobility management, multi-rat coexistence, and machine learning (3m) framework for c-its,” Electronics, vol. 15, no. 5, p. 1042, 2026. [14] X. Xu, Q. Wu, P. Fan, K. Wang, N. Cheng, W. Chen, and K. B. Letaief, “Velocity-adaptive access scheme for semantic-aware vehicular networks: Joint fairness and aoi optimization,” IEEE Transactions on Mobile Computing, 2026.

[15] M. M. Elsayed, K. M. Hosny, M. M. Fouda, and M. M. Khashaba, “Vehicles communications handover in 5g: A survey,” ICT Express, vol. 9, no. 3, pp. 366–378, 2023. [16] P. Kaswan, P. Mitra, A. Srivastava, and S. Ulukus, “Age of information in gossip networks: A friendly introduction and literature survey invited paper,” IEEE Transactions on Communications, 2025. [17] P. Kaswan and S. Ulukus, “Sacrificing freshness for reliable information in age-based gossiping,” IEEE Transactions on Information Theory, 2025. [18] ——, “Misinformation spread in gossip networks: The influence of transmission mutations,” IEEE Transactions on Communications, 2025. [19] P. Mitra and S. Ulukus, “Age-aware gossiping in network topologies,” IEEE Transactions on Communications, vol. 72, no. 11, pp. 7032– 7046, 2024. [20] J. Selen, Y. Nazarathy, L. L. Andrew, and H. L. Vu, “The age of information in gossip networks,” in International Conference on Analytical and Stochastic Modeling Techniques and Applications. Springer, 2013, pp. 364–379. [21] M. E. J. Newman, Networks: An Introduction. Oxford, U.K.: Oxford University Press, 2010. [22] J. M. I. Parella, O. T. Ajayi, and Y. Cheng, “Adaptive messaging based on the age of information in vanets,” in GLOBECOM 2022-2022 IEEE Global Communications Conference. IEEE, 2022, pp. 1235–1240.

Related documents

• How We Use Together: Coordinating Individual Preferences for Using Shared Devices at Home#384988
• 47 U.S.C. § 336 — Broadcast spectrum flexibility#597993
• Interference-Driven Clustered Optimisation for FM Spectrum Coordination#1035844
• A multiple core mode for protocol independent multicast#445169
• Multi-network broadcasting within the Internet#208903