arXiv CSopen access

Repeated Contention Scheduling: A Novel Resource Allocation Algorithm Toward 6G Vehicular Networks

arXiv:2607.06103v1 [cs.NI] 7 Jul 2026

© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. 2026 Mediterranean Artificial Intelligence and Networking Conference (MAIN 2026)

Repeated Contention Scheduling: A Novel Resource Allocation Algorithm Toward 6G Vehicular Networks Alexey Rolich∗ , Marco Tricco∗ , Simone Paroli∗ , Mert Yildiz∗ , and Andrea Baiocchi∗ ∗ University of Rome Sapienza, Italy

{alexey.rolich, mert.yildiz, andrea.baiocchi}@uniroma1.it {paroli.1853547, tricco.1894220}@studenti.uniroma1.it

Abstract—Efficient decentralized resource allocation remains a fundamental challenge in NR-V2X sidelink communications, where conventional Semi-Persistent Scheduling (SPS) and Dynamic Scheduling (DS) suffer from persistent collisions and limited adaptability under dynamic and dense conditions. This paper proposes Repeated Contention Scheduling (RCS), a novel resource allocation algorithm based on multi-round, feedbackdriven contention that eliminates long-term reservations and enables fully distributed operation. Simulations demonstrate that RCS outperforms SPS and DS in terms of success probability, collision and loss reduction, and timeliness metrics such as Packet Inter-Reception Delay and Age of Information, particularly under high load. The practical feasibility of the approach is validated through an SDR-based experimental testbed, which confirms robust operation under realistic hardware impairments and closely matches theoretical and simulation results. These findings establish RCS as a viable and scalable solution for resource allocation in 5G NR sidelink and a promising candidate for future 6G vehicular communication systems. Index Terms—5G, 6G, V2X, Semi-Persistent Scheduling, Sidelink, Age of Information, Dynamic Scheduling, Persistence, Vehicular Networks, Repeated Contention, Vehicular Communication, Resource Allocation.

I. I NTRODUCTION Intelligent Transportation Systems (ITS) constitute a key component of next-generation transportation systems, enabling improved safety, traffic efficiency, and automation through pervasive connectivity. Vehicular networks provide the communication backbone of ITS, supporting real-time information exchange among vehicles, infrastructure, and vulnerable road users. In this context, 5G New Radio (NR)Vehicle-to-Everything (V2X) has emerged as a technological solution offering high-reliability, low-latency communication tailored to advanced vehicular applications [1]. A central feature of NR-V2X is sidelink communication, which enables direct and decentralized interactions in both in-coverage (Mode 1) and out-of-coverage (Mode 2) scenarios [1, 2]. Efficient sidelink resource allocation remains a fundamental challenge in dynamic vehicular environments characterized by high mobility, heterogeneous traffic density, and rapidly varying channel conditions. The 5G NR-V2X sidelink framework defines two operational modes. In Mode 1, the gNB performs centralized scheduling through dynamic grants for per-transmission allocation and configured grants for semipersistent reservations, supporting both periodic and aperiodic traffic under network coverage. In Mode 2, user equipments autonomously select resources based on sensing mechanisms. Two standardized allocation schemes are defined for this mode: Semi-Persistent Scheduling (SPS) and Dynamic Scheduling (DS) [3, 4]. SPS adopts a reservation-based strategy in which resources are selected and retained over multiple transmission intervals

according to a predefined Resource Reservation Interval (RRI), controlled by parameters such as the Reselection Counter (RC) and persistence probability Ppers . In contrast, DS follows a non-persistent approach, where resources are selected independently for each packet, making it more suitable for aperiodic traffic [5]. Although both schemes are standardized, no explicit guideline is provided for their selection, and their effectiveness depends on traffic characteristics and network conditions. SPS is widely adopted due to its efficiency for periodic traffic and reduced control overhead in decentralized operation. However, its persistence-based design introduces inherent limitations. Repeated reuse of reserved resources increases the probability of persistent packet collisions when multiple vehicles select overlapping resources, particularly in the presence of hidden terminals and sensing inaccuracies [6]. These collisions may persist over consecutive transmissions [6, 7, 8], leading to inefficient spectrum utilization [5] and degraded quality of service under high mobility or network congestion [9]. To mitigate these issues, numerous enhancements have been proposed. At the protocol level, 3GPP introduces mechanisms such as short-term sensing [10], resource re-evaluation [11, 12, 13], pre-emption [14], and inter-UE coordination [15], along with feedback-based reliability features including Hybrid Automatic Repeat Request (HARQ) and Physical sidelink Feedback Channel (PSFCH) [16, 17, 18]. From an algorithmic perspective, optimization-based approaches, including convex optimization, game theory, and graph-based formulations, have been applied to improve resource selection [19]. In parallel, machine learning approaches, particularly reinforcement learning [20, 21, 22, 23] and multi-agent systems [24, 25], enable adaptive and context-aware decision-making [26, 27, 28, 29, 30, 31, 32] in decentralized environments. Additional directions include the integration of Non-Orthogonal Multiple Access (NOMA) for enhanced spectral efficiency and contextaware strategies such as RSU-assisted or position-based allocation [33, 34, 35, 36]. Despite these advances, most approaches remain evolutionary and retain the semi-persistent allocation paradigm, thereby remaining susceptible to recurring collisions. Consequently, robust distributed resource allocation remains an open problem. Survey and tutorial studies on resource allocation in 5G NRV2X and beyond [37, 38, 39, 40] confirm that recent progress has largely focused on incremental refinements of SPS/DSbased mechanisms rather than fundamentally new designs. As a result, persistent collisions caused by hidden terminals, sensing inaccuracies, and limited coordination remain unresolved. Moreover, non-persistent allocation strategies have received comparatively limited attention, particularly those

II. D ESCRIPTION OF RCS In this section, we describe the system model and the configuration of 5G NR multiple access to support RCS. Then, the RCS algorithm is presented in Section II-B. Then we provide an analysis of RCS (Section II-C). A. System model Let us consider n nodes, each one generating update messages. We assume a message fits into a single Sub-Channel (SC). The average generation time, i.e., the mean time elapsing

Activity phase

Contention phase

SCj

sub-carriers

Frequency

incorporating effective feedback and coordination without relying on long-term reservations. This highlights that it is worth investigating alternative paradigms that overcome the inherent limitations of reservation-based schemes. Motivated by these limitations, this paper proposes a new sidelink resource allocation approach, termed Repeated Contention Scheduling (RCS), for NR-V2X and future 6GV2X systems. Unlike SPS, the proposed method is not based on semi-persistent reservations. Instead, it employs repeated contention rounds [41, 42, 43], where vehicles contend for resources for each message they have to send. Contention exploits sub-carrier signaling with simultaneous transmission and reception. The feasibility of this key function is proved via an experimental test-bed. This design eliminates longterm reservations and replaces them with a dynamically reevaluated contention process. RCS is fully compatible with 5G NR-V2X sidelink while improving reliability and timeliness in dense scenarios, as highlighted by numerical results. RCS provides a scalable solution for current vehicular networks and a foundation for future intelligent resource allocation mechanisms in 6G vehicular systems. Main contribution of the paper: • We propose a novel resource allocation algorithm, termed RCS, based on a repeated contention mechanism for NRV2X sidelink communication. Unlike conventional DS and SPS schemes, the proposed approach eliminates semipersistent reservations and instead relies on feedbackdriven contention rounds, enabling dynamic and collisionresilient resource selection. • We demonstrate through extensive simulations that RCS outperforms traditional DS and SPS in terms of reliability, collision mitigation, information timeliness, particularly in dense vehicular scenarios. • We validate the practical applicability of RCS through a real-world experimental setup, providing a proof-ofconcept implementation that confirms its feasibility and effectiveness under realistic deployment conditions. The rest of this paper is organized as follows. Section II presents the proposed RCS algorithm, detailing its repeated contention mechanism and design principles. Section III describes the simulation framework, including system assumptions, parameter configuration, and evaluation metrics relevant to reliability and timeliness. Also, this section reports the simulation results and provides a comparative performance analysis against baseline schemes. Section IV presents the experimental proof-of-concept implementation and discusses practical considerations and observed performance in a realworld setup, as illustrated in the prototype system. Finally, Section V summarizes the main findings and outlines directions for future research.

Time Ts

Tf

Figure 1. Configuration of the channel using RCS. Table I U SED SYMBOLS AND CORRESPONDING DEFINITIONS . Parameter n Tg Ts Tf nf nSC K a M J w m δ TDB

Definition Number of nodes Message generation interval Slot duration Frame duration, Tf = nf Ts Number of slots per frame Number of SCs per slot Number of SCs per frame, K = (nf − 1)nSC System load, given by a = (n/Tg )/(K/Tf ) Total number of sub-carriers used during contention Number of contention groups (divisor of K) Number of SCs associated with one contention group Sub-carriers per group, m = ⌊M/J⌋ Duration of one contention round Packet delay budget

for a node to generate a new update message, is denoted with Tg . The multiple access structure is shown in Figure 1. It consists of frames, each frame comprising nf 5G NR subframes, hence lasting Tf = nf Ts , where Ts is the sub-frame time (referred to as slot time in the following). Each frame comprises a contention time, lasting one or more time slots, and SCs in all remaining time slots, where SC definition and structure are as in sidelink. The number of slot times devoted to contention time depends on the selected numerology. In the case of µ = 0 (reference numerology), it is assumed that contention time reduces to its minimum, one time slot. Hence, the remaining nf − 1 slots of a frame are used to carry SCs. If nSC SCs are configured in each time slot, the overall number of SCs in one frame is given by K = (nf − 1)nSC . It is then possible to define a load factor a as follows: a=

n Tf K Tg

(1)

Note that, if a > 1, the system is structurally overloaded, i.e., there are not enough SCs for all generated messages. A delay budget is associated to each message. If the time elapsing since when the message was generated exceeds the delay budget and the message has not been transmitted yet, the message is discarded, without being transmitted. In the following, we assume that the delay budget is the same for all messages, denoted with TDB . To guarantee that at least one contention is possible irrespective of the generation time of the message, it must be TDB ≥ 2Tf . On the other hand, there is no point in setting TDB > Tg , given that a new update message is generated every time period Tg . Hence, we assume 2Tf ≤ TDB ≤ Tg (holding for Tg > 2Tf ).

B. RCS algorithm

a divisor of K. Each resource group comprises w = K/J distinct SCs of a frame. The basic procedure defined in the previous subsection is the special case where J = K and hence w = 1. Since multiple SCs are associated with each group, the contention in a group defines multiple winners. More in depth, exactly w winners should be ideally identified. The number of actual winners in a given contention may fall short of w, because too few nodes joined that group. It may also exceed w, in case multiple nodes select the same winning sub-carriers (then, a collision event will occur for at least one of the SCs belonging to the group). To define the generalized procedure, the whole set of M sub-carriers used for contention is split into J subsets, each comprising m = ⌊M/J⌋ sub-carriers. Assume a contending node selects a sub-carrier with frequency fx . The contending node wins if it detects less than w sub-carriers with frequency fy < fx . On the contrary, if at least w sub-carriers with frequency fy < fx are detected, the node leaves the contention. This generalized contention is referred to as “multi-win”. 3) Implementation of the RCS algorithm: The critical points to implement the RCS algorithm in 5G NR and towards 6G sidelink are as follows. • Simultaneous transmission and reception of sub-carriers must be feasible. • Nodes must be synchronized so that boundaries of contention rounds are aligned. As for the first point, note that RCS does not require implementing a true full-duplex radio. The requirement to implement RCS is only that a node be able to transmit a tone at frequency fx and at the same time detect tones in a frequency band between f1 and fx−1 , where fx is the tone frequency selected by the node. The feasibility of this mild form of duplex is proved in experiments described in Section IV. As for the second point, the duration δ of a contention round should be one or a few symbol times (with numerology µ = 0, a sub-frame lasts 1 ms and contains 14 OFDM symbols, hence one symbol lasts ≈ 71 µs). We will see in Section IV that a round time of 3 symbol times is enough to guarantee reliable simultaneous sub-carrier transmission and detection. Tight synchronization among nodes operating in sidelink should be guaranteed anyway, independently of RCS, to guarantee correct detection of SC control information and carried data.

Let us follow the operation of a backlogged node, first introducing the baseline contention procedure (single-win). Generalizations to multiple-win are then introduced. Used notation is listed in Table I. 1) Basic RCS algorithm: As soon as a new message is available, say at time t0 , the backlogged node waits until the first occurrence of the contention time after t0 . Then it starts the contention procedure for that message. Since there is one contention time per frame, the node must wait at most for a time Tf − Ts , where we account for a contention time equal to one slot time. A set of sub-carriers is designated out of all sub-carriers in the sidelink channel.1 Let M be the overall number of sub-carriers used in the contention procedure. Since there are K SCs to be assigned in one frame, sub-carriers are organized into K subsets, each comprising m sub-carriers. Sub-carrier subset j is associated to the j-th SC of the frame, j = 1, . . . , K.2 A backlogged node selects one sub-carrier subset at random and contends for the SC associated with that subset. The contention time is organized into r consecutive rounds. At the beginning of the first round, the node selects at random one out of the m sub-carriers of the chosen subset. Let f1 , f2 , . . . , fm be the frequencies of the m sub-carriers, labeled in increasing order, i.e., so that fi < fj if i < j. Let fx be the frequency picked at random by the considered node. Each contending node transmits its selected sub-carrier for the whole round and simultaneously the node listens to the frequency band comprising the selected sub-carrier group, to identify sub-carriers transmitted by other nodes. If the considered node detects a frequency fy < fx , then the node deems itself as losing the contention and leaves the current contention. It will take part in the next contention (after one frame time), if the packet delay budget allows this additional delay. If instead the considered node does not detect any frequency less than its own selected frequency fx , then it will deem itself as a winner, and it will move to the next round. This contention procedure is repeated the same in every round, each time independently selecting a sub-carrier at random from the considered subset. Nodes surviving up to the last round included (the r-th round) are the final winners. Winners of sub-carrier subset j are entitled to use SC j in the frame where they have won the contention. A collision C. Analysis of RCS collision probability will occur if and only if there is more than one winner. The A model is presented to evaluate the collision probability analysis in Section II-C will show that the probability of of the basic RCS contention procedure [43]. The model holds collision decays exponentially fast with the number of rounds. under the assumption that each contending node detects subAfter having won the contention and having transmitted carriers transmitted by all other contending nodes. in the SC associated with the selected sub-carrier subset, the Let n be the number of nodes backlogged at the beginning of node goes back to idle if it has no other pending message. the contention phase. Let qi denote the probability that a node Otherwise, it starts the whole procedure all over again for the picks frequency fi , i = 1, . . . , m. Let also Gi = Pm qj be j=i next scheduled message. the corresponding Complementary Cumulative Distribution 2) Generalized RCS algorithm: Splitting sub-carriers into Function (CCDF). K groups and letting contending nodes joining groups at We define a Discrete Time Markov Chain (DTMC) X(t), random may lead to some of the groups being void, i.e., with where X(t) denotes the number of nodes contending in round no contending node. Then, the associated SC would go unused t = 1, . . . , r. X(0) = n is the initial number of contending in that frame. To mitigate this potential waste of resource, we nodes. The DTMC evolves over the state space {1, . . . , n}. generalize the scheme of the previous sub-section as follows. The one-step transition probability matrix of the DTMC is Let us define J ≤ K resource groups, where J is chosen as denoted with P, with entries given by: Pm−1
k h k−h 1We will see in Section IV that reliable detection of sub-carriers requires  Pi=1 h qi Gi+1 h = 1, . . . , k − 1 a frequency separation between sub-carriers used in the contention procedure. m k Pk,h = (2) h = k, 2 SCs belonging to one frame can always be numbered in a non-ambiguous i=1 qi   order. 0 h = k + 1, . . . , n,

for k = 1, . . . , n. communication channel. Then, packet reception is successful Let p(t) denote the state probability vector at time t, the except in the case of collision (more than one node transmitting k-th component of which is pk (t) = P(X(t) = k), k = on the same SC). Hence, under RCS possible outcomes of 1, . . . , n. At the outset of the contention we have pn (0) = 1 packet P generated by node A are as follows: (i) node A is and pk (0) = 0, k = 1, . . . , n − 1. In the following, we choose the unique winner of an SC resource for packet P; then P a uniform probability distribution for sub-carrier selection in will be received successfully by all other nodes; (ii) multiple all rounds, i.e., we let qi = 1/m, i = 1, . . . , m, for every nodes win the same SC as A; then packet P incurs a collision round. The corresponding matrix having entries as given event that disrupts its reception; (iii) packet P is dropped (after in Equation (2) is denoted with Pu , to emphasize that a having lost all available contentions), because it exceeds its uniform probability distribution has been adopted for sub- delay budget. The considered reliability metrics are related to carrier selection. Therefore, the probability distribution of outcomes of a packet: Packet Reception Ratio (PRR), collision the state of the DTMC at time t is p(t) = p(0)Ptu , for probability (Pcoll (sim)), and packet drop probability (Pdrop ), t = 1, . . . , r. The probability of a successful contention, i.e., with P RR + Pcoll + Pdrop = 1. The PRR is defined as the that a single node wins the last round, is Psucc = P(X(r) = fraction of successfully delivered packets: 1) = p1 (r). Correspondingly, the collision probability is Nsinglewin PRR = (6) Pcoll = P(X(r) > 1) = 1 − p1 (r). Let Qu denote the square Ntotal matrix obtained by taking the last n − 1 rows and columns The collision probability is defined as: of Pu . The collision probability can be written as Nmultiplewin Pcoll = y Qru e (3) Pcoll (sim) = (7) Ntotal where y is a row vector containing the last n − 1 components The packet drop probability accounts for delay constraints of p(0), i.e., yn−1 = 1, yk = 0, k = 1, . . . , n − 2, and e is and is defined as: a column vector of 1’s. The matrix Qu is lower triangular, Ndrop Pdrop = , (8) with diagonal elements given by Pu,kk in Equation (2) for Ntotal k = 2, . .P . , n. Hence, its dominant eigenvalue is η = Qu,11 = m Pu,22 = i=1 qi2 = 1/m, where we have used the fact that where a packet is discarded if its delay exceeds TDB . A qi = 1/m, ∀i. Since Qu is also a non-negative matrix, the packet gets delayed when it loses a contention, and it has right eigenvector u associated to η is positive. It is possible to go to the next contention. Since each packet is associated to find the closed form of u, namely uT = [2 3 . . . n]/2, with a deadline, if the elapsed time from generation exceeds where the superscript T denotes transposition. Since it is this limit, the packet is considered obsolete and is discarded, eT = [1 1 . . . 1] ≤ [2 3 . . . n]/2 = uT and all vectors and giving up on running more contention. The timeliness of the system is assessed through Packet matrices are non-negative, we have: Inter-Reception Delay (PIR) [44], mean Age of Information 1 n Pcoll = yQru e ≤ yQru u = r yu = (4) (AoI), and the AoI violation probability. The PIR [44] m 2mr quantifies the time interval between two consecutive successful In the end, we find the following upper bound on the collision receptions of packets belonging to the same application flow, probability for any value of the number of contending nodes, thus capturing the effective update timing at the receiver. n: n Consider the local dynamic map maintained at node j, which n o Pcoll ≤ min 1, (5) aggregates information received from neighboring nodes. r 2m Let tij (g) denote the reception time at node j of the g-th The key feature of the upper bound is that it reveals that the successfully received packet from node i, with g ≥ 2. AoI at collision probability decays exponentially as the number of node j for updates from node i is defined as rounds r grows. It also highlights the role of the number m of sub-carriers used in the contention subset. The analysis Aij (t) = t − tij (g − 1), t ∈ [tij (g − 1), tij (g)) (9) developed in this Section and the resulting upper bound hold as long as nodes can hear each other and sub-carrier detection The peak AoI, which coincides with PIR, corresponding to the maximum age just before a new reception, is therefore is error-free. Yij (g) = tij (g) − tij (g − 1) (10) III. S IMULATIONS RESULTS This section presents the simulation-based evaluation of the Let zij denote the number of successfully received updates proposed RCS scheme. Section III-A defines the metrics used from node i to node j. The average PIR at node j, accounting to assess both reliability and timeliness of packet delivery. The for all incoming flows (i, j), is computed as a weighted simulation setup is described in Section III-B. Section III-C average: Pzij X Yij (g) analyzes the impact of the number of contention groups on Pg=1 E[PIRj ] = (11) reliability and timeliness over a wide range of load conditions. g∈Nj zgj i∈Nj Finally, Section III-D compares the performance of RCS with the standardized sidelink multiple access schemes SPS and where Nj is the set of nodes from which at least Ω packets have been successfully received by node j (with Ω = 2 in DS. simulations). The network-level PIR is obtained by averaging A. Key performance metrics for simulation of RCS E[PIRj ] over all nodes. It is assumed that all nodes can hear one another. As AoI measures the freshness of the most recently received a consequence, sub-carrier signaling works as described information and applies to a wide range of communication in Section II-C. An all-or-nothing model is used for the scenarios, including cooperative awareness, perception, and

coordination. The average AoI for updates from node i to node j is defined as Pzij 1 2 Z 1 T g=1 Yij (g) Aij (t) dt ≈ Pzij 2 (12) E[AoIij ] = lim T →∞ T 0 g=1 Yij (g) The relation between AoI and PIR can be expressed as E[AoI] =

E[PIR2 ] , 2 · E[PIR]

(13)

showing that AoI depends on both the mean and the variability of inter-reception intervals. Based on inter-reception intervals, the probability of AoI violation with respect to a threshold Ath is defined for each node pair (i, j) as Pzij g=2 max {0, Yij (g) − Ath } Pmij Vij (Ath ) = (14) g=2 Yij (g) The global AoI violation probability is obtained by averaging Vij over node pairs (i, j). B. Simulation setup

Table II S IMULATION PARAMETERS UNDER DIFFERENT NUMEROLOGIES Unified parameter µ=0 µ=1 Sub-Carrier Spacing (SCS) 15 kHz 30 kHz Slot duration (Ts ) 1 ms 0.5 ms Total RBs 106 51 Used sub-carriers (M ) 400 200 SCs per slot (nSC ) 4 2 Message generation time (Tg ) 100 ms Frame duration (Tf ) 10 ms Bandwidth (BW P ) 20 MHz RBs per SC 25 OFDM symbols per slot 14 Time Delay Budget (TDB ) 60 ms Load coefficient (a) [0.3; 1.1] Simulation time (Tsim ) 50 s AoI violation threshold (Ath ) 100, 200, 500 ms DS parameter µ=0 µ=1 Number of SCs for allocation in frame (K) 40 Persistence probability (Ppers ) 0 RC 1 SPS parameter µ=0 µ=1 RRI 100 ms Number of SCs for allocation in frame (K) 40 Persistence probability (Ppers ) 0, 0.8 RC [5; 15] RCS parameter µ=0 µ=1 Max. number of contention rounds (r) 3 Number of SCs for allocation in frame (K) 36 Number of contention groups (J) 1, 2, 3, 6, 9, 18, 36 SCs per contention group (w) 36, 18, 12, 6, 4, 2, 1 Sub-carriers per group (mJ ) 400,200,133, 200,100,66, 66,44,22,11 33,22,11,5

Simulations consider two numerologies, µ = 0 and µ = 1, corresponding to sub-carrier spacings of 15 kHz and 30 kHz, respectively. The only frame parameters kept fixed across numerologies are the bandwidth part, BW P = 20 MHz, and the frame duration, Tf = 10 ms. The time slot duration depends on the numerology, namely Ts = 1 ms for µ = 0 and Ts = 0.5 ms for µ = 1. The frame partition into contention Section IV). In contrast to DS and SPS, only 36 SCs are time and SCs is adapted accordingly. For µ = 0, the contention time occupies one slot, cor- available for allocation per frame, since one 1 ms portion responding to 1 ms, over the full 20 MHz bandwidth. The of the frame is reserved for contention. The number of remaining nine slots per frame host SCs, i.e., the resource to contention groups takes values in the set {1, 2, 3, 6, 9, 18, 36}, be allocated. Each SC is composed of 25 Resource Blocks which corresponds to {36, 18, 12, 6, 4, 2, 1} SCs associated (RBs). With 4 SCs per slot, 36 SCs per frame are available with each group. Accordingly, the number of sub-carriers per group is {400, 200, 133, 66, 44, 22, 11} for µ = 0 and for allocation. For µ = 1, the full available bandwidth for two slot times {200, 100, 66, 33, 22, 11, 5} for µ = 1. As for SPS configuration, the resource reservation interval is devoted to contention, so that contention time still amounts to 1 ms. The data part of the frame, therefore, consists is set to RRI = 100 ms, the persistence probability is set of the remaining 18 slot times. Assigning still 25 RBs to to two values Ppers = 0 and Ppers = 0.8, and the RC is one SC, each slot accommodates 2 SCs. Then, one frame uniformly drawn from the range [5, 15]. In this case, all 40 provides 36 SCs for allocation, as in the case of µ = 0. This SCs in each frame are available for allocation. For a more design ensures that the two numerologies use the same frame detailed description of the operation of DS and SPS, the reader duration, with the same contention time and the same amount is referred to [1, 3, 4, 5, 9]. DS and SPS are implemented in accordance with the model of assignable SCs, thus enabling a fair comparison in the presented in [7, 8]. In the SPS implementation, we additionally subsequent experiments. Given the numerology dependent SCS and the overall incorporate the delay budget. TDB . If, during the sensing and bandwidth, 1200 and 600 sub-carriers are available for µ = 0 resource selection phase, insufficient resources are available and µ = 1, respectively. As we will see in Section IV, reliable for reservation, the packet is considered dropped. This does sub-carrier detection is attained by using one sub-carrier out not occur in the DS case, as transmission is performed in a of three consecutive ones. Then, 400 and 200 sub-carriers one-shot manner. Another important modification of the model are available for contention in case of µ = 0 and µ = 1, is that the channel load can exceed one, i.e., N > K. It is also assumed that all nodes are within mutual communication range, respectively. Message generation time is set to Tg = 100 ms, delay and that each message requires a single SC for successful budget to TDB = 60 ms, and AoI violation threshold to transmission, meaning it fits within one transport block. Table II summarizes the value of all relevant parameters. three different value: 100 ms, 200 ms and 500 ms. There are 14 Orthogonal Frequency-Division Multiple Access (OFDM) symbols per slot. The channel load coefficient a (see Equa- C. Tuning of RCS parameters In this section, we compare different settings of RCS tion (1)) is used as the independent variable. Simulation time parameters, focusing on different numbers of contention is Tsim = 50 s. As for RCS configuration, the number of contention rounds groups J under various numerologies. All metrics presented is fixed to r = 3, thus assigning three OFDM symbol times are functions of the load a. to each round (about 213 µs), with one symbol time gap Figure 2 shows the PRR for different values of J. As between subsequent rounds (according to results obtained in the load increases, the PRR decreases for both values of

0.7

0.4

J=6 J=9

J=18 J=36

0.6 0.8 Load, a

J=1 J=2 J=3

0.8 0.7

1.0

0.4

(a) µ = 0

J=6 J=9

J=18 J=36

0.6 0.8 Load, a

J=18 J=36

0.1 0.0

0.4

0.6 0.8 Load, a

1.0

J=6 J=9

J=18 J=36

0.1

(a) µ = 0

0.4

0.6 0.8 Load, a

1.0

0.4

0.6 0.8 Load, a

1.0

(b) µ = 1

70 J=1 J=2 J=3

65

J=6 J=9

J=18 J=36

60 55 50

0.4

0.6 0.8 Load, a

(a) µ = 0

(b) µ = 1

Figure 3. Pcoll (sim) vs. load for different numbers of contention groups J.

J=18 J=36

105 100

1.0

70 J=1 J=2 J=3

0.2

0.0

0.6 0.8 Load, a

J=6 J=9

110

(a) µ = 0

E[AoI], [ms]

J=6 J=9

Pcoll (sim)

Pcoll (sim)

0.2

0.4

J=1 J=2 J=3

115

Figure 5. E[P IR] vs. load for different numbers of contention groups J.

0.3 J=1 J=2 J=3

J=18 J=36

105

(b) µ = 1

Figure 2. PRR vs. load for different numbers of contention groups J. 0.3

J=6 J=9

110

100

1.0

120 J=1 J=2 J=3

115

E[AoI], [ms]

J=1 J=2 J=3

0.8

PRR

0.9

PRR

0.9

120 E[PIR], [ms]

1.0 E[PIR], [ms]

1.0

1.0

J=1 J=2 J=3

65

J=6 J=9

J=18 J=36

60 55 50

0.4

0.6 0.8 Load, a

1.0

(b) µ = 1

Figure 6. E[AoI] vs. load for different numbers of contention groups J.

transmissions. For µ = 0 and µ = 1, J = 1 yields the best performance. The absolute differences in performance remain moderate, on the order of 15 ms under high load. 10 10 These results reveal two representative operating regimes 10 10 for RCS. At low load, J = 36 achieves the highest probability 10 10 0.4 0.6 0.8 1.0 0.4 0.6 0.8 1.0 Load, a Load, a of successful reception and the lowest collision probability, as losses are dominated by collisions and a larger number of (a) µ = 0 (b) µ = 1 groups reduces contention effectively. As the load increases, Figure 4. Pdrop vs. load for different numbers of contention groups J. this advantage diminishes, and performance across group µ. For µ = 0, at a load of a = 0.9, J = 36 is no configurations converges. In the high-load regime (a ≥ 0.9), longer optimal, and J = 1 becomes the best configuration in J = 1 becomes preferable, offering better timeliness and the terms of successful packet delivery. In contrast, for µ = 1, highest successful transmission probability, while J = 36 J = 36 consistently achieves the highest performance, while becomes slightly less effective. Overall, RCS maintains J = 1 yields the lowest. This occurs because the number satisfactory performance, with collision probabilities on the of sub-carriers within a group for µ = 1 is half that of the order of 10−1 at the maximum considered load. In this regime, corresponding configuration at µ = 0, leading to increased packet loss is increasingly influenced by drop events, as packet loss due to collisions or drops. For all other metrics, packets cannot always be delivered within the available time J = 1 and J = 36 represent boundary conditions, with budget. intermediate values of J yielding performance between these Therefore, in Section III-D, we consider only the boundary extremes. cases J = 1 and J = 36 for RCS and restrict the analysis Figure 3 illustrates the collision probability, i.e., the prob- to µ = 1, as DS and SPS performance is independent of µ ability that multiple winners emerge during the contention with given channel configuration, while RCS is evaluated in phase for a given SC, causing simultaneous use of the same its worst-case PRR configuration. resources by multiple nodes. For µ = 0, this probability is lower than for µ = 1. For J = 36, performance remains D. Comparison with standardized sidelink multiple access nearly invariant with numerology, while for a smaller number algorithms of contention groups under µ = 1, performance deteriorates. In this section, we compare the configurations RCS with The collision probability increases with load in all cases. J = 1 and J = 36, DS, and the standardized persistence Figure 4 presents the packet drop probability. Here, J = 36 probability values 0 and 0.8 for SPS. We analyze the exhibits a substantially higher drop probability than J = 1. performance of the standardized resource selection algorithms For J = 1 under low load, the drop probability is near zero and the proposed RCS using the PRR, mean PIR, and mean and increases sharply only when a > 0.9. The figure uses AoI. Additionally, we examine the AoI violation probability a logarithmic scale, with maximum values remaining in the for three thresholds: 100 ms, 200 ms, and 500 ms. All metrics range [0.01, 0.1] even at a = 1. The drop probability is presented are functions of the load a. largely insensitive to numerology. The higher drop probability Figure 7 presents three metrics: PRR (left), mean PIR for J = 36 is due to the presence of only a single SC in (center), and mean AoI (right, plotted on a logarithmic scale). the group, resulting in more packets being discarded after The results show that the proposed RCS can outperform the TDB during repeated contention, where competition is the standardized SPS and DS under periodic network traffic. more intense. Moreover, with J = 36 the probability that From the perspective of PRR, the advantage of RCS over some group is deserted is not negligible, thus reducing the SPS is moderate, as the curves for different RCS and effectiveness of resource assignment. SPS configurations intersect. Nevertheless, RCS consistently Figure 5 and Figure 6 show the mean PIR and mean AoI, outperforms DS and SPS with a persistence probability of 0 respectively. Both metrics increase with load due to more across all load conditions and matches the PRR of SPS only frequent collisions and longer intervals between successful when the persistence probability is 0.8. 100

−2

100

J=1 J=2 J=3

J=6 J=9

J=18 J=36

10−1

Pdrop

Pdrop

10−1

−2

−3

−3

−4

−4

J=1 J=2 J=3

J=6 J=9

J=18 J=36

RCS, J=36

0.8

250

0.6 0.4 0.2 0.0 0.3

0.5

0.7 Load, a

0.9

1.1

SPS, P=0

SPS, P=0.8

200 150 100 0.3

DS

103 E[AoI], [ms]

300 E[PIR], [ms]

PRR

RCS, J=1 1.0

0.5

0.7 Load, a

0.9

1.1

102

0.3

0.5

0.7 Load, a

0.9

1.1

Figure 7. Comparison of DS, SPS, and RCS performance in terms of PRR (left), E[PIR] (center), and E[AoI] (right).

0.8 0.6 0.4 0.2 0.0 0.3

0.5

0.7 Load, a

0.9

1.1

1.0

SPS, P=0

SPS, P=0.8

Ath = 200 ms AoI violation prob.

1.0

RCS, J=36

AoI violation prob.

AoI violation prob.

RCS, J=1

Ath = 100 ms

0.8 0.6 0.4 0.2 0.0 0.3

0.5

0.7 Load, a

0.9

1.1

1.0

DS

Ath = 500 ms

0.8 0.6 0.4 0.2 0.0 0.3

0.5

0.7 Load, a

0.9

1.1

Figure 8. Comparison of DS, SPS, and RCS in terms of AoI violation probability for different thresholds Ath : left (Ath = 100ms), center (Ath = 200ms), and right (Ath = 500ms).

The analysis of the mean PIR reveals a clear trend: as duration of persistent collisions. The obtained results also indicate that a reliability metric, the load increases, the average interval between successfully delivered packets grows. This increase is steeper for DS and such as PRR, for RCS is fully correlated with timeliness SPS, particularly at high channel loads around 0.8–0.9. At metrics, such as AoI. In contrast, for SPS, this correlation low loads, SPS with a persistence configuration of 0.8 is is nontrivial and not immediately evident, due to the hidden comparable to RCS, but under higher loads, RCS clearly issue of persistent collisions, including those that may last for extended periods. outperforms SPS. Considering the mean AoI as a function of load, the IV. E XPERIMENTAL PROOF - OF - CONCEPT proposed RCS significantly surpasses all standardized resource This section presents the experimental validation of the selection algorithms. In DS, collisions are frequent but ranRCS scheme. Section IV-A defines metrics for reliability dom and uncorrelated, whereas SPS experiences consecutive and feasibility. Section IV-B details the hardware, PHY, and persistent collisions and burst losses, leading to a sharper synchronization setup. Section IV-C examines key PHY and increase in mean AoI. protocol parameters, and Section IV-D compares experimental This behavior is further illustrated in Figure 8, which results with simulations, highlighting real-world non-idealities. shows the AoI violation probability. RCS outperforms SPS and DS across all thresholds, including the stringent 100 A. Key performance metrics for experimental evaluation of ms threshold, demonstrating that RCS effectively mitigates RCS persistent collisions and bursty losses. The considered reliability metrics are: probability of success For SPS, the violation probability changes little across (Psucc ), collision probability (Pcoll (exp)), no-winner probabilthresholds, indicating that many AoI samples exceed 500 ms, ity (Pnowin ). Unlike simulations, the experiment yields three reflecting persistent collisions that block communication for possible outcomes: (i) success with a single winner; (ii) colliextended periods. In contrast, for DS and RCS, the violation sion with multiple winners; and (iii) contention terminating probability decreases with increasing thresholds, indicating without any winner. Hence it is Psucc +Pcoll (exp)+Pnowin = 1 fewer extreme AoI values, consistent with observations in [9, and the probability of packet loss is Ploss = Pcoll (exp) + 27]. This demonstrates more regular and reliable packet Pnowin . delivery. Notably, for the 200 ms threshold, RCS exhibits In the following calculations, Psucc is equivalent to the extremely low violation probabilities, not exceeding 0.1 even PRR (Equation (6)), while the collision probability defined in under high load. For the 500 ms threshold, the probability simulations, Pcoll (sim) (Equation (7)), corresponds directly approaches zero, showing that messages are delivered consis- to its experimental counterpart, Pcoll (exp). tently even in the presence of collisions and drops, ensuring The no-winner probability, observed only in experiments, is stable communication. defined as the fraction of contentions that end without selecting These results indicate that RCS outperforms the standard- any winner, relative to the total number of contentions. It is ized DS and SPS under the worst-case configuration (µ = 1), given by: Nnowin providing high packet delivery reliability and regular, stable Pnowin = (15) Ntotal updates between nodes. This improvement arises from the complete elimination of the sensing and resource reservation To identify a suitable configuration, we evaluated the false procedure, which significantly reduces the probability and positive probability (Pf p ), false alarm probability (Pf a ), and

Table III E XPERIMENTAL PARAMETERS

C. Experimental parameters tuning

The sub-carrier selection (see Table IV) mitigates hardware non-idealities by introducing a spacing of 2 sub-carriers between contention levels, reducing inter-carrier interference, and excluding the DC and mirror frequencies to avoid sensing errors due to local oscillator leakage and image components. This spacing also limits false positives caused by spectral leakage into adjacent bins, which may lead nodes to prematurely withdraw from contention. Evaluation of the false positive probability for different guard spacings shows that using 2 guard sub-carriers achieves Pf p of approximately the detection probability (Pd ). A false alarm occurs when a 10−2 (see Figure 9a), which is acceptable for the experimental node detects a non-existent tone, potentially causing all nodes setup. to withdraw and leaving the resource unassigned. A missed The round duration determines the trade-off between detection occurs when a transmitted tone is not detected, detection reliability and the number of rounds per slot: which may result in multiple nodes claiming the resource and short durations degrade detection performance, while longer causing a collision. The false positive probability is defined durations reduce the number of available rounds. Given the 14as the probability that a node detects a tone on a sub-carrier symbol slot structure, with the first and last symbols reserved that immediately precedes the transmitted one, typically due for PHY functions [1] and a guard symbol required between to spectral leakage caused by the signal’s shape. consecutive rounds, four time-slot configurations are feasible: • 6 rounds of 1 symbol each, with 6 guard symbols; B. Experimental setup • 4 rounds of 2 symbols each, with 4 guard symbols; • 3 rounds of 3 symbols each, with 3 guard symbols; The experimental setup implements a wireless node proto• 2 rounds of 5 symbols each, with 2 guard symbols. type capable of simultaneous tone transmission and channel sensing, as required for frequency-domain RCS operation The false alarm (Pf a ) and detection (Pd ) probabilities were [43]. The system is built on a Software Defined Radio evaluated for each round duration (see Figures 9b and 9c). (SDR) platform based on Ettus USRP B200 devices, enabling A duration of three OFDM symbols achieves acceptable flexible physical layer control. Signal processing and protocol performance (Pf a ≈ 10−4 , Pd ≈ 0.999), whereas shorter logic are implemented in GNU Radio, allowing modular durations lead to significantly higher error rates. Specifically, one-symbol rounds provide insufficient enintegration of the contention mechanism with the radio frontend. Synchronization across nodes is achieved via an Ettus ergy integration, resulting in poor detection performance CDA-2990 OctoClock, which provides a common frequency (Pd ≈ 0.5). Two-symbol rounds improve detection but exhibit elevated false alarm rates due to limited robustness against reference and aligned sampling time. The SDR implementation introduces practical impairments, noise and spectral leakage. A duration of three OFDM symbols including noise, interference, limited dynamic range, and provides the best trade-off, ensuring reliable detection while oscillator instability. These effects require careful parameter allowing multiple contention rounds within a single slot. This tuning, particularly for guard band allocation and sub-carrier configuration is adopted as the baseline for the hardware selection, to ensure reliable operation under FR1 sidelink implementation of the RCS algorithm. specifications [45]. The adopted PHY configuration, summaD. Experimental results rized in Table III, balances standard compliance with real-time The experimental success probability, compared with simuprocessing constraints. The system operates with numerology µ = 0, corresponding to a SCS of 15 kHz and a slot duration lation results, is shown in Figure 10. Please note that the simulation parameters shown in Figure 10 are consistent of 1 ms. with the experimental parameters reported in Table III. The Following the RCS principle, contention levels are mapped comparative analysis reveals that the experimental Psucc onto OFDM sub-carriers. Due to computational constraints, measured on the USRP nodes is slightly lower than the the FFT size is fixed to NF F T = 32, from which a subset of values obtained through simulations. This performance gap 10 sub-carriers is selected to represent contention levels. is directly attributable to the emergence of Pnowin within Each node transmits a single OFDM tone by activating the hardware based setup. While the simulation environment one randomly selected sub-carrier while sensing the channel assumes ideal conditions where every non-colliding contention simultaneously. If a tone is detected at a lower frequency, the results in a success, the physical implementation introduces node withdraws; otherwise, it proceeds to the next contention real-world impairments. Specifically, a small fraction of these round. A tone is considered valid if its power exceeds a Parameter Value Central frequency (fc ) 5 GHz FFT size (NF F T ) 32 samples Numerology (µ) 0 SCS 15 kHz Slot duration (Ts ) 1 ms Bandwidth (BW P ) 480 kHz OFDM symbols per slot 14 Symbol duration 71.428 ms Cycle Prefix length 3 samples

threshold determined by the noise floor. To emulate a contention with multiple stations, one of the available USRP devices was programmed to represent more than a single node, allowing the total number of competing nodes n to range between 3 and 10. At each round, it generates simultaneous OFDM tones over m sub-carriers and executes the RCS procedure. The evaluation is based on an extensive experimental campaign comprising more than 105 contention cycles per configuration.

Table IV M APPING OF CONTENTION SUB - CARRIERS INDICES AND FREQUENCY OFFSETS WITH RESPECT TO THE CENTER FREQUENCY. Index

Offset (kHz)

Index

Offset (kHz)

-15 -12 -9 -6 -3

-225 -180 -135 -90 -45

+1 +4 +7 +10 +13

+15 +60 +105 +150 +195

1.0

10

10

−1

0.8

10

−2

10

−2

0.6

10−3 10

−4

10

−5

0

1 2 Number of Guard Carriers

3

Pd

100

−1

Pfa

Pfp

100

10−3

0.4

10

−4

0.2

10

−5

(a) Pf p as a function of guard carriers number

1

2 3 5 Round duration, [OFDM symbols]

(b) Pf a as a function of round duration

0.0

1

2 3 5 Round duration, [OFDM symbols]

(c) Pd as a function of round duration

Figure 9. Results of experimental parameters tuning.

Probability

1.000 0.998 0.996 0.994

Simulation Experiment

0.992 0.990

4

6 8 Number of nodes, n

10

Figure 10. Comparison between experimental and simulation-based success probability. Pcoll (exp) Pnowin

Pcoll (sim) Pcoll

Probability

10−2

10−3

contending nodes increases. This trend can be attributed to the fact that a higher density of participants increases the likelihood that false detections caused by noise are masked by the actual tones transmitted by the stations. In a scenario with a limited number of available sub-carriers, the event in which a station erroneously withdraws due to a low index false tone becomes increasingly rare as the spectrum becomes more populated. Conversely, the collision probability (Pcoll ) scales proportionally with the number of contending nodes, as the statistical likelihood of multiple stations selecting an identical sub-carrier in the final round increases. This inverse relationship between Pnowin and Pcoll is depicted in Figure 11, which shows how the growth of n shifts the system’s primary impairment from sensing-induced deferrals to physical collisions. The experimental data reveal a clear transition in the dominant impairment factor as the network density increases. V. C ONCLUSION

10−4

4

6 8 Number of nodes, n

10

Figure 11. Comparison of experimental and simulation-based collision probabilities, the theoretical upper bound on collision probability, and the no-winner probability.

This paper introduced RCS, a novel resource allocation algorithm for NR-V2X and towards 6G sidelink communications that departs from DS and SPS. By replacing reservationbased mechanisms with a multi-round contention process, RCS enables fully distributed, feedback-driven resource selection, mitigating persistent collisions and adapting to dynamic vehicular environments. Simulation results demonstrate that RCS consistently outperforms DS and SPS in terms of reliability and timeliness, achieving higher PRR, lower loss rates, and improved PIR and AoI, particularly under high load conditions. The feasibility of RCS is further validated through an SDR-based proof-of-concept. Despite hardware impairments, experimental results align with theoretical and simulation trends, confirming that RCS operates reliably in practical settings and represents a viable solution for decentralized resource allocation in 5G NR and beyond. Future work will focus on detailed modeling of the proposed RCS algorithm under more realistic conditions, including aperiodic traffic, hidden nodes, signal propagation and fading models, as well as experimental investigations of scenarios involving multiple winners in the contention phase.

contention slots is lost due to false alarms during the sensing phase. In such instances, environmental impulse noise or interference is erroneously interpreted as an active tone from a competitor, leading nodes to defer and resulting in an inconclusive contention cycle even in the absence of actual collisions. The no-winner probability is illustrated in Figure 11, which compares simulation and experimental results. The figure reports the theoretical upper bound on collision probability from Equation (5) (black curve), the simulated collision probability (red markers), and the experimentally measured collision (blue curve) and no-winner (green curve) probabilities. First, the presence of no-winner events (Pnowin ) reduces the number of contention phases that can result in collisions, as some rounds terminate prematurely due to sensing errors. Second, physical-layer impairments may induce a collisionACKNOWLEDGMENTS to-success transition. When two nodes select the same subThis work was partially funded by Sapienza University of carrier in the final round, a collision is expected; however, sensing errors or noise fluctuations may cause one node to Rome under the “Progetti per Avvio alla Ricerca – Tipo 2” detect a spurious lower-frequency tone and withdraw. As a program (2025) for the project 5GSL-RTA-VRUP “Advancing result, the other node gains access to the resource, effectively 5G Vehicular Communication for Real-Time Awareness and converting a potential collision into a successful transmission. VRU Protection” (prot. AR225199B968B6C6) Although infrequent, this effect further reduces the observed R EFERENCES Pcoll compared to the ideal model. [1] M. H. C. Garcia et al., “A tutorial on 5G NR V2X communications,” The experimental results indicate that the no-winner conIEEE Communications Surveys & Tutorials, vol. 23, no. 3, pp. 1972– tention probability (Pnowin ) decreases as the number of 2026, 2021.

[2] Z. Ali, S. Lagén, L. Giupponi, and R. Rouil, “3GPP NR V2X mode 2: overview, models and system-level evaluation,” IEEE Access, vol. 9, pp. 89 554–89 579, 2021. [3] 3GPP, “NR; Medium Access Control (MAC) protocol specification (Release 19),” 3GPP, TS 38.321 V19.2.0, Dec. 2025. [4] 3GPP, “Overall description of Radio Access Network (RAN) aspects for Vehicle-to-everything (V2X) based on LTE and NR (Release 19),” 3GPP, TR 37.985 V19.0.0, Oct. 2025. [5] L. Lusvarghi, A. Molina-Galan, B. Coll-Perales, J. Gozalvez, and M. L. Merani, “A Comparative Analysis of the Semi-Persistent and Dynamic Scheduling Schemes in NR-V2X Mode 2,” Vehicular Communications, p. 100 628, 2023. [6] A. Bazzi, C. Campolo, A. Molinaro, A. O. Berthet, B. M. Masini, and A. Zanella, “On Wireless Blind Spots in the C-V2X Sidelink,” IEEE Transactions on Vehicular Technology, vol. 69, no. 8, pp. 9239–9243, 2020. [7] A. Rolich, I. Turcanu, A. Vinel, and A. Baiocchi, “Understanding the impact of persistence and propagation on the Age of Information of broadcast traffic in 5G NR-V2X sidelink communications,” Computer Networks, vol. 248, p. 110 503, 2024. [8] A. Rolich, I. Turcanu, A. Vinel, and A. Baiocchi, “Impact of Persistence on the Age of Information in 5G NR-V2X Sidelink Communications,” in 2023 21st Mediterranean Communication and Computer Networking Conference (MedComNet), 2023, pp. 15–24. [9] A. Rolich, M. Yildiz, I. Turcanu, A. Vinel, and A. Baiocchi, “Rethinking Persistent Scheduling in 5G New Radio Vehicle-to-Everything Sidelink Communications,” IEEE Access, vol. 13, pp. 164 065–164 083, 2025. [10] S. Ha, W. Yoo, H. Kim, and J.-M. Chung, “C-V2X Adaptive Short-Term Sensing Scheme for Enhanced DENM and CAM Communication,” IEEE Wireless Communications Letters, vol. 11, no. 3, pp. 593–597, 2022. [11] V. Todisco, Z. Wu, and A. Bazzi, “Improving NR-V2X Autonomous Mode Through Resource Re-Evaluation,” in 2024 IEEE Vehicular Networking Conference (VNC), 2024, pp. 125–131. [12] A. Molina-Galan, L. Lusvarghi, B. Coll-Perales, J. Gozalvez, and M. L. Merani, “On the Impact of Re-Evaluation in 5G NR V2X Mode 2,” IEEE Transactions on Vehicular Technology, vol. 73, no. 2, pp. 2669–2683, 2024. [13] A. Molina-Galan, J. Gozalvez, and B. Coll-Perales, “A Selective ReEvaluation Mechanism for 5G NR V2X Mode 2 Communications,” IEEE Transactions on Vehicular Technology, vol. 75, no. 2, pp. 2939– 2951, 2026. [14] Q. Han, M. Zhou, L. Zeng, L. Ye, and L. Yang, “NR-V2X Mode2 Resource Selection Window Size Adjustment Based on Resource Priority Index,” in 2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), 2024, pp. 411–416. [15] M. Shehata, C. Ciochina, and J.-C. Sibel, “On the Performance of Inter User Coordination for Resource Allocation Enhancement in NR V2X Mode 2,” in Communication Technologies for Vehicles, Cham: Springer International Publishing, 2021, pp. 3–14. [16] S. An and K. Chang, “Enhancing Reliability in 5G NR V2V Communications Through Priority-Based Groupcasting and IR-HARQ,” IEEE Access, vol. 11, pp. 72 717–72 731, 2023. [17] C. Mahabal and T. Shimizu, “Impact of HARQ Retransmissions on Information Age in 5G NR Sidelink,” in 2024 IEEE 100th Vehicular Technology Conference (VTC2024-Fall), 2024, pp. 1–6. [18] Q. Fu, J. Liu, J. Wang, and J. Li, “SecHARQ: A Secure Scheme for HARQ-Assisted NR V2X Communications,” IEEE Transactions on Network and Service Management, vol. 22, no. 2, pp. 2199–2213, 2025. [19] M. Parvini, P. Schulz, and G. Fettweis, “Resource Allocation in V2X Networks: From Classical Optimization to Machine Learning-Based Solutions,” IEEE Open Journal of the Communications Society, vol. 5, pp. 1958–1974, 2024. [20] 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, 2026. [21] A. Hegde, R. Song, and A. Festag, “Radio Resource Allocation in 5G-NR V2X: A Multi-Agent Actor-Critic Based Approach,” IEEE Access, vol. 11, pp. 87 225–87 244, 2023. [22] Z. Li, P. Wang, Y. Shen, and S. Li, “Reinforcement LearningBased Resource Allocation Scheme of NR-V2X Sidelink for Joint Communication and Sensing,” Sensors, vol. 25, no. 2, 2025. [23] M. Montaño, J. Gomez-Ponce, M. Antonieta-Alvarez, F. Novillo, and R. Cajo, “Adaptive MCS Optimization with Feature Selection and Machine Learning for C-V2X Sidelink,” in Advanced Research in Technologies, Information, Innovation and Sustainability, Springer Nature, 2026, pp. 311–323. [24] M. M. Saad, M. A. Tariq, M. Ajmal, D. Kim, and G. Srivastava, “Federated Multiagent Reinforcement Learning for Resource Allocation in NR-V2X Mode 2,” IEEE Internet of Things Journal, vol. 12, no. 13, pp. 23 402–23 417, 2025.

[25] T. Xiaolu, S. Yan, X. Yaqi, C. Shanzhi, and G. Yuming, “AoI and TTC based resource allocation in C-V2X sidelink via multi-agent reinforcement learning,” China Communications, vol. 22, no. 8, pp. 281– 297, 2025. [26] A. Rolich, M. Yildiz, and A. Baiocchi, “Safety-Critical Delays in Vehicular Networks Toward 6G: A Novel Metric for Disconnection Time Assessment,” IEEE Communications Letters, vol. 30, pp. 1860–1864, 2026. [27] A. Rolich, M. Yildiz, I. Turcanu, A. Vinel, and A. Baiocchi, “On the Trade-off Between AoI Performance and Resource Reuse Efficiency in 5G NR V2X Sidelink,” in IEEE Vehicular Networking Conference (VNC), Porto, Portugal: IEEE, Jun. 2025. [28] W.-D. Shen and H.-Y. Wei, “Age-of-Information Performance Analysis in Power-Efficient Sidelink Communications,” IEEE Internet of Things Journal, vol. 12, no. 20, pp. 42 116–42 132, 2025. [29] F. Peng, Z. Jiang, S. Zhang, and S. Xu, “Age of Information Optimized MAC in V2X Sidelink via Piggyback-Based Collaboration,” IEEE Transactions on Wireless Communications, vol. 20, no. 1, pp. 607–622, 2021. [30] L. Cao, H. Yin, R. Wei, and L. Zhang, “Optimize Semi-Persistent Scheduling in NR-V2X: An Age-of-Information Perspective,” in 2022 IEEE Wireless Communications and Networking Conference (WCNC), 2022, pp. 2053–2058. [31] A. Rolich, I. Turcanu, and A. Baiocchi, “AoI-Aware and PersistenceDriven Congestion Control in 5G NR - V2X Sidelink Communications,” in 2024 22nd Mediterranean Communication and Computer Networking Conference (MedComNet), 2024, pp. 1–4. [32] A. Rolich, M. Yildiz, I. Turcanu, A. Vinel, and A. Baiocchi, “From Latency to Value of Information: A Review of Timeliness Metrics for Safe Transportation Systems in the 6G Era,” IEEE Vehicular Technology Magazine, pp. 2–11, 2026. [33] A. Bazzi et al., “Toward 6G Vehicle-to-Everything Sidelink: Nonorthogonal Multiple Access in the Autonomous Mode,” IEEE Vehicular Technology Magazine, vol. 18, no. 2, pp. 50–59, 2023. [34] V. Todisco, C. Campolo, A. Molinaro, A. O. Berthet, R. A. StirlingGallacher, and A. Bazzi, “On the Performance of SIC-based NOMA in the C-V2X Sidelink Autonomous Mode,” in 2023 IEEE Conference on Standards for Communications and Networking (CSCN), 2023, pp. 66–72. [35] T. Hirai, T. Kimura, and N. Wakamiya, “Spatial Performance Analysis of Autonomous Sidelink Cellular-V2X with NOMA,” in 2022 IEEE Global Communications Conference (GLOBECOM), 2022, pp. 1205–1210. [36] T. Fuchikami, T. Hirai, and N. Wakamiya, “Spatial Performance Analysis of V2X with Sidelink NOMA with Multiple Interfering Vehicles,” in 2024 International Conference on Consumer Electronics Taiwan (ICCE-Taiwan), 2024, pp. 605–606. [37] C. Shin, E. Farag, H. Ryu, M. Zhou, and Y. Kim, “Vehicle-to-Everything (V2X) Evolution From 4G to 5G in 3GPP: Focusing on Resource Allocation Aspects,” IEEE Access, vol. 11, pp. 18 689–18 703, 2023. [38] A. N. Al-Najjar, M. F. A. Rasid, F. Hashim, F. A. Ahmad, and A. Jamalipour, “A systematic literature review in distributed resource allocation for C-V2X,” Ingénierie des Systèmes d’Information, vol. 29, no. 3, pp. 771–808, 2024. [39] Annu and P. Rajalakshmi, “Towards 6G V2X Sidelink: Survey of Resource Allocation—Mathematical Formulations, Challenges, and Proposed Solutions,” IEEE Open Journal of Vehicular Technology, vol. 5, pp. 344–383, 2024. [40] C. Bin Ali Wael, E. Hadj Dogheche, N. Armi, A. Subekti, and I. Dayoub, “Leveraging 3GPP Features and Optimization Techniques for 5G NRV2X Resource Allocation: A Survey,” IEEE Open Journal of Intelligent Transportation Systems, vol. 6, pp. 967–994, 2025. [41] W. Zame, J. Xu, and M. van der Schaar, “Winning the Lottery: Learning Perfect Coordination With Minimal Feedback,” IEEE Journal of Selected Topics in Signal Processing, vol. 7, no. 5, pp. 846–857, 2013. [42] A. Baiocchi, I. Tinnirello, D. Garlisi, and A. L. Valvo, “Random access with repeated contentions for emerging wireless technologies,” in IEEE INFOCOM 2017 - IEEE Conference on Computer Communications, 2017, pp. 1–9. [43] A. Baiocchi, D. Garlisi, A. L. Valvo, G. Santaromita, and I. Tinnirello, “‘Good to repeat’: Making random access near-optimal with repeated contentions,” IEEE Transactions on Wireless Communications, vol. 19, no. 1, pp. 712–726, 2019. [44] 3GPP, “Study on evaluation methodology of new Vehicle-to-Everything (V2X) use cases for LTE and NR; (Release 15),” 3GPP, TR 37.885 V15.3.0, Jun. 2019. [45] 3GPP, “NR; Physical channels and modulation (Release 19),” 3GPP, TS 38.311 V19.3.0, Mar. 2026.