Redundancy Estimation in Semantic and Task-Oriented V2X Communications: Role and Challenges

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE Redundancy Estimation in Semantic and Task-Oriented V2X Communications: Role and Challenges Luca Lusvarghi, Javier Gozalvez UWICORE Laboratory, Universidad Miguel Hernandez de Elche, Elche, Spain. Email: {llusvarghi, j.gozalvez}@umh.es Abstract—Semantic and task-oriented V2X communications have the potential to improve the scalability of future Vehicleto-Everything (V2X) networks by transmitting only the most relevant information to the intended receivers. The relevance of a piece of information depends on whether it is redundant or not for an intended receiver. However, redundancy estimation is inherently challenging and error-prone in broadcast V2X communications. Redundancy estimation errors can compromise the accuracy of the estimated relevance, ultimately affecting the content of the transmitted messages. This study presents the first analysis of the impact of redundancy estimation errors on semantic and task-oriented V2X communications. In particular, it characterizes the probability that a redundancy estimation error prevents the delivery of relevant information at the receiver – an event that can impair the receiver’s understanding of the driving context and, potentially, its driving tasks. Our analysis also outlines conditions under which redundancy estimation errors can prevent the delivery of relevant information at the receiver. These findings provide valuable insights to understand the operation of semantic and task-oriented V2X communication systems and to optimize their performance and design. Keywords—semantic communications, task-oriented communications, V2X, redundancy, relevance. I. INTRODUCTION The realization of the Connected and Autonomous Driving (CAD) vision of safer, more efficient, and more sustainable mobility relies on the scalable support of Vehicle-toEverything (V2X)-enabled CAD services. These services enable connected vehicles to broadcast information such as their position and speed (cooperative awareness), detected objects (cooperative perception or sensor sharing), or planned trajectories (cooperative maneuvering or driving) that is critical to enhance driving safety and efficiency. However, the large-scale adoption of V2X services is challenged by their bandwidth-demanding nature and the scarcity of V2X spectrum resources [1]. One approach to address these scalability challenges is redundancy mitigation [2] or control [3], which is used in the context of cooperative perception to regulate the amount of redundant information transmitted by each vehicle. In cooperative perception, information about a detected object is considered redundant if it has already been transmitted by other nearby vehicles. While redundant information has little to no value for receiving vehicles, it can overload the V2X channel, compromising the effectiveness of cooperative perception as the number of communicating vehicles increases [4]. Existing studies have shown that redundancy control can greatly enhance the scalability of cooperative perception [5]-[7]. Despite its benefits, redundancy control alone has proven insufficient to support multiple V2X services at scale [8]. Redundancy control aims to minimize redundancy, an information attribute that is estimated at the transmitter and does not consider the impact of the transmitted information on the receiving vehicles. This transmitter-centric approach can result in the dissemination of information that is not valuable at the receiver, thereby leading to an inefficient use of communication resources and limiting the scalability of V2X networks. Semantic and task-oriented communications is a novel and promising communication paradigm that aims to address the scalability challenges of future 6G and beyond networks by focusing on the relevance of the exchanged information. Relevance is a receiver-centric concept that captures the context-dependent impact that the meaning of a piece of information has on the receiver’s tasks. It allows to jointly address the semantic and task-oriented (or effectiveness) communication problems originally identified by Shannon and Weaver in [9]. The semantic problem is concerned with the identification and transmission of the essential information needed to convey the desired meaning, while the task-oriented problem focuses on transmitting the information needed by the receiver to execute its tasks. The authors propose in [10] a semantic and task-oriented V2X communication paradigm in which vehicles include in the transmitted messages only the information that is estimated to be relevant for the intended receivers. The application of this novel communication paradigm to the V2X domain is not exempt from challenges given the broadcast nature of V2X communications. However, preliminary results in [10] demonstrate that semantic and task-oriented V2X communications can significantly improve the scalability and efficiency of future V2X networks with respect to existing V2X communication approaches. The effectiveness of semantic and task-oriented V2X communications depends on the ability of the transmitting vehicles to accurately estimate the relevance of the information. The relevance of a piece of information is influenced by several factors, including its redundancy for the intended receivers. However, redundancy estimation is a complex and challenging process in broadcast communication scenarios such as the V2X domain [11]. Redundancy estimation can only be indirectly performed by monitoring the content of the messages exchanged on the V2X network. It involves probabilistically estimating whether information successfully decoded by the transmitter is also available at the intended receivers or not. Redundancy estimation errors occur, for example, when the transmitter incorrectly estimates that a piece of information is already available at the intended receiver. In semantic and task-oriented V2X communications, redundancy estimation errors can compromise the accuracy of the relevance estimated at the transmitter. This can potentially cause the transmitter to incorrectly omit information that is This work was supported by the European Union under the 2023 MSCA Postdoctoral Fellowship program (project no. 101153845) and by MCIN/AEI/10.13039/501100011033 (PID2023-150308OB-I00). relevant for the intended receivers from a transmitted message, compromising the effectiveness of semantic and task-oriented communications. This study analyses for the first time the impact of redundancy estimation errors on semantic and task-oriented V2X communications, particularly focusing on their impact on the selection of the transmitted messages content. It characterizes the probability that relevant information is not delivered to an intended receiver because it was incorrectly estimated as redundant - a critical event that can impair the receiver’s understanding of the driving context and, potentially, its driving tasks. Our analysis outlines conditions under which redundancy estimation errors can potentially prevent the delivery of relevant information at the receiver. These findings can offer valuable insights to understand the operation of semantic and task-oriented V2X communication systems and to optimize their performance and design. II. SEMANTIC AND TASK-ORIENTED V2X COMMUNICATIONS Semantic and task-oriented V2X communications is a promising communication paradigm that can improve the scalability and efficiency of V2X communications by selecting the information to transmit based on its estimated relevance for the intended receivers. By doing so, semantic and task-oriented V2X communications can minimize the consumption of communication resources while guaranteeing the delivery of all the relevant information needed at the receiver [10]. Relevance is defined as the context-dependent impact that the meaning of a piece of information has on the receiving vehicle’s understanding of the driving environment and, ultimately, on its driving tasks. For example, information about a pedestrian crossing an intersection (meaning) is relevant for a vehicle approaching the same intersection (context). In the vehicle’s context, it potentially represents a safety-critical situation that the vehicle must avoid (driving task impact). Conversely, the information about the pedestrian is irrelevant for a vehicle leaving the intersection (context). The relevance of a piece of information can greatly vary depending on the context. Context in the V2X domain refers to all the circumstances (e.g., driving environment characteristics, vehicles’ actions and intentions, exchanged V2X messages) under which a piece of information is exchanged or processed. In broadcast communication scenarios such as the V2X domain, context is specific and unique to each intended receiver. At each vehicle, the contextdependent relevance of the information is captured by a contextual relevance function. The contextual relevance function is a complex nonlinear multi-dimensional function that determines the relevance of each piece of available information by combining its meaning and attributes (quality, freshness, redundancy) with the vehicle’s context [10]. The transmitter-side operation of semantic and taskoriented V2X communications is illustrated by the block diagram in Fig. 1. The transmitting vehicle collects contextual information either locally, using its onboard sensors, or from other vehicles and roadside units via V2X communications. Contextual information includes any kind of information about the driving environment (e.g., traffic lights status, detected objects, road topology) and the communication context (e.g., messages exchanged on the V2X network). The transmitting vehicle processes contextual information to estimate the context of each intended receiver. Context estimation is a complex yet fundamental task to correctly estimate the information relevance. Then, the transmitting vehicle uses a semantic model to estimate the context-dependent relevance of locally collected information for each intended receiver and determine what information should be part of the transmitted message content. In V2X, vehicles can include only locally collected information in the transmitted messages to avoid the propagation of potential detection errors across the V2X network. The semantic model begins by performing redundancy estimation to determine which locally collected information is already available at the intended receivers. Information estimated as redundant is considered irrelevant in this work, as we assume that its transmission would not improve the intended receiver’s understanding of the driving environment. Next, the semantic model leverages the estimated context of each intended receiver to perform relevance estimation and quantify the context-dependent relevance that locally collected information has for the intended receivers. Due to the broadcast nature of V2X communications, a transmitting vehicle might have multiple intended receivers. Accordingly, the semantic model performs relevance estimation considering all possible intended receivers and outputs a multi-receiver relevance estimate for each piece of locally collected information. The transmitting vehicle leverages the relevance estimate provided by the semantic model to select the content of the transmitted messages, subject to the potential communication constraints imposed by congestion control protocols. Content selection aims to deliver the most relevant locally collected information to all intended receivers as efficiently as possible, avoiding the transmission of irrelevant information. For the sake of brevity, we will refer to semantic and task-oriented V2X communications as semantic V2X communications in the rest of this paper. III. IMPACT OF REDUNDANCY ESTIMATION ERRORS Redundancy estimation has traditionally been employed in V2X communications to control the transmission of redundant information. In semantic V2X communications, it is a key component of the semantic model used by transmitting vehicles to estimate the relevance of locally collected information for all their intended receivers. Hence, redundancy estimation directly influences the contentFig. 1. Semantic and task-oriented V2X communications: transmitter-side block diagram . selection process in semantic V2X communications. However, redundancy estimation is an inherently challenging and error-prone process in the V2X domain due to the broadcast nature and challenging propagation conditions of V2X communications. A redundancy estimation error is particularly significant when it wrongly estimates a piece of information as redundant. In this case, the semantic model can incorrectly estimate a piece of information as irrelevant, leading the transmitter to omit relevant information from the transmitted message content – an event that can potentially cause a semantic content-selection error. A semantic contentselection error occurs when a vehicle correctly receives a message, but the received message does not contain all relevant information needed by the receiving vehicle due to a redundancy estimation error at the transmitter1. A semantic content-selection error is particularly critical, as it involves relevant information that can impact the receivers’ contextual awareness. Fig. 2 depicts a cooperative perception example to analyse and illustrate the conditions that affect the occurrence probability of a semantic content-selection error. Fig. 2 represents a T-intersection with one object 𝑥𝑥1 and three communicating vehicles: 𝑉𝑉𝑇𝑇 (the transmitter), 𝑉𝑉𝑂𝑂 (the observer), and 𝑉𝑉𝑅𝑅 (the receiver). Let’s assume that both 𝑉𝑉𝑇𝑇 and 𝑉𝑉𝑂𝑂 detect object 𝑥𝑥1 (Condition A) and that 𝑥𝑥1 is relevant for 𝑉𝑉𝑅𝑅 (Condition B). If 𝑉𝑉𝑇𝑇 correctly estimates the relevance of 𝑥𝑥1 for 𝑉𝑉𝑅𝑅 using its semantic model, 𝑉𝑉𝑇𝑇 includes 𝑥𝑥1 in its transmitted message (Condition C). If 𝑉𝑉𝑂𝑂 does not correctly decode the message of 𝑉𝑉𝑇𝑇, it is not aware of having detected the same object as 𝑉𝑉𝑇𝑇 and will not consider 𝑥𝑥1 as redundant information for 𝑉𝑉𝑅𝑅. If 𝑉𝑉𝑂𝑂 correctly decodes the message transmitted by 𝑉𝑉𝑇𝑇, it becomes aware that 𝑉𝑉𝑇𝑇 has also detected 𝑥𝑥1 (Condition D). To determine if 𝑥𝑥1 is redundant for 𝑉𝑉𝑅𝑅, 𝑉𝑉𝑂𝑂 needs to estimate whether 𝑉𝑉𝑅𝑅 has also correctly decoded the message from 𝑉𝑉𝑇𝑇 or not. In the latter case, 𝑉𝑉𝑂𝑂 estimates that 𝑥𝑥1 is non-redundant for 𝑉𝑉𝑅𝑅, and uses its semantic model to estimate whether 𝑥𝑥1 is relevant for 𝑉𝑉𝑅𝑅. If it is, 𝑉𝑉𝑂𝑂 includes 𝑥𝑥1 in its next transmitted message. If 𝑉𝑉𝑂𝑂 estimates that 𝑉𝑉𝑅𝑅 has also correctly decoded the message from 𝑉𝑉𝑇𝑇, it estimates that 𝑥𝑥1 is redundant and, hence, irrelevant for 𝑉𝑉𝑅𝑅 (Condition E) and will not include it in its next message. A redundancy estimation error occurs if 𝑉𝑉𝑅𝑅 did not correctly decode the message from 𝑉𝑉𝑇𝑇 (Condition F) and, therefore, 𝑥𝑥1 is not actually redundant for 𝑉𝑉𝑅𝑅. In this case, 𝑥𝑥1 is incorrectly estimated as irrelevant for 𝑉𝑉𝑅𝑅 and is incorrectly omitted by 𝑉𝑉𝑂𝑂 from its message. If the message transmitted by 𝑉𝑉𝑂𝑂 is not correctly received by 𝑉𝑉𝑅𝑅, the redundancy (and relevance) estimation error does not have any impact on 𝑉𝑉𝑅𝑅. If the 1 It is worth highlighting that a semantic content-selection error can also occur if redundancy estimation is correct but the semantic model incorrectly estimates the relevance of a piece of information, e.g., due to an inaccurate message transmitted by 𝑉𝑉𝑂𝑂 is correctly received by 𝑉𝑉𝑅𝑅 (Condition G), a semantic content-selection error occurs. A semantic content-selection error directly affects the receiving vehicle, as it prevents 𝑉𝑉𝑅𝑅 from receiving any information about 𝑥𝑥1. This loss of information reduces 𝑉𝑉𝑅𝑅’s contextual awareness and can potentially compromise its driving safety and efficiency. For example, as illustrated in Fig. 2, not receiving any information about 𝑥𝑥1 can represent a safetycritical situation for 𝑉𝑉𝑅𝑅 if 𝑥𝑥1 is a non-connected vehicle crossing its trajectory. IV. EVALUATION FRAMEWORK We perform our analysis considering a driving scenario populated by 𝑁𝑁 vehicles and 𝐾𝐾 exogenous variables. Exogenous variables are randomly distributed according to a 2D Poisson Point Process (PPP) and represent all the driving environment information that is external to the ego-vehicle. For example, exogenous variables represent detected objects, e.g., vehicles and vulnerable road users. We denote the set of all exogenous variables in the driving scenario with 𝒦𝒦 and the 𝑘𝑘-th exogenous variable with 𝑥𝑥𝑘𝑘, 𝑘𝑘= 1, … , 𝐾𝐾. Each vehicle 𝑉𝑉 𝑛𝑛 in the scenario collects exogenous variables either locally, through its onboard sensors, or from other vehicles via V2X communications. We model the probability that a vehicle 𝑉𝑉 𝑛𝑛 can locally detect an exogenous variable 𝑥𝑥𝑘𝑘 using its onboard sensors with a logistic function of the distance between 𝑥𝑥𝑘𝑘 and 𝑉𝑉 𝑛𝑛. Vehicles periodically broadcast a new message every 𝑇𝑇 ms. Each transmitted message contains a selected subset of the exogenous variables locally detected by the transmitting vehicle through its onboard sensors. In semantic V2X communications, a transmitting vehicle includes in the transmitted message only the subset of exogeneous variables that it estimates to be relevant for its intended receivers. An intended receiver is a vehicle that can decode the transmitted message and should, therefore, be considered when curating the content of the transmitted message. We assume that a vehicle 𝑉𝑉 𝑛𝑛 is estimated as an intended receiver of the transmitter if the transmitter can decode at least one of the last two messages transmitted by 𝑉𝑉 𝑛𝑛. To estimate the relevance of locally detected exogeneous variable and curate the content of the transmitted messages, the transmitting vehicle first needs to perform redundancy estimation (see Fig. 1). If an exogenous variable 𝑥𝑥𝑘𝑘 is estimated as redundant for an intended receiver 𝑉𝑉𝑅𝑅, it is also deemed irrelevant for 𝑉𝑉𝑅𝑅 (i.e., its estimated relevance is set equal to zero). If 𝑥𝑥𝑘𝑘 is estimated as non-redundant, the transmitting vehicle leverages a simplified, yet realistic, implementation of a semantic model to estimate its relevance for 𝑉𝑉𝑅𝑅. At each intended receiver 𝑉𝑉𝑅𝑅, relevance is captured by a contextual relevance function 𝑓𝑓𝑅𝑅(𝒦𝒦) that assigns a nonuniform semantic value 𝑤𝑤𝑅𝑅,𝑘𝑘 to each exogenous variable 𝑥𝑥𝑘𝑘 that is included within a relevance range 𝑅𝑅𝑚𝑚𝑚𝑚𝑚𝑚. The semantic value 𝑤𝑤𝑅𝑅,𝑘𝑘 is a measure of the context-dependent relevance of each variable 𝑥𝑥𝑘𝑘 for 𝑉𝑉𝑅𝑅. The relevance range 𝑅𝑅𝑚𝑚𝑚𝑚𝑚𝑚 defines the maximum distance between 𝑉𝑉𝑅𝑅 and a relevant variable. Within 𝑅𝑅𝑚𝑚𝑚𝑚𝑚𝑚 , the fraction of relevant variables is ∆𝐻𝐻. The remaining 1− ∆𝐻𝐻 exogenous variables are considered irrelevant for 𝑉𝑉𝑅𝑅. Relevant variables are assigned a semantic value larger than zero (𝑤𝑤𝑅𝑅,𝑘𝑘> 0), whereas irrelevant variables are assigned a semantic value equal to zero (𝑤𝑤𝑅𝑅,𝑘𝑘= 0). All estimation of the intended receiver’s contextual relevance function. This type of errors is out of the scope of this study. Fig. 2. Cooperative perception : semantic content-selection error example. exogenous variables located outside the relevance range 𝑅𝑅𝑚𝑚𝑚𝑚𝑚𝑚 are also considered irrelevant for 𝑉𝑉𝑅𝑅 and assigned a semantic value of zero. The semantic model estimates the semantic value 𝑤𝑤�𝑅𝑅,𝑘𝑘, i.e., the context-dependent relevance assigned by 𝑓𝑓𝑅𝑅(𝒦𝒦) to 𝑥𝑥𝑘𝑘, subject to an estimation error 𝜀𝜀 that depends on the amount of contextual information available at the transmitter. After estimating the relevance of locally collected exogenous variables for all intended receivers, the transmitter leverages the estimated relevance to select the content of the transmitted messages (see Fig. 1). During the contentselection step, the transmitter first filters out those variables that are estimated as irrelevant for all intended receivers. Then, it ranks the remaining locally detected variables based on the number of intended receivers for which they are estimated as relevant. After ranking, the transmitter includes in the transmitted message only the top Γ variables. We assume that the number of exogenous variables that can be included in each message (a proxy of the message size) is constrained by a limit Γ which could be imposed by congestion control protocols (see Fig. 1). A. Redundancy Estimation Redundancy is estimated at the transmitter by monitoring the exogenous variables contained in the messages exchanged over the V2X channel. An exogenous variable 𝑥𝑥𝑘𝑘 correctly received via V2X by the transmitter is estimated to be redundant for an intended receiver 𝑉𝑉𝑅𝑅 with probability 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟, using one of the two following approaches. 1) Hard redundancy estimation: the transmitter assumes that any exogenous variable it receives via V2X is also available at all intended receivers. Consequently, 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟 is fixed and equal to 1. The hard redundancy estimation approach is generally employed in cooperative perception studies [7] and standards [3]. 2) Soft redundancy estimation: if 𝑉𝑉𝑅𝑅 has explicitly included 𝑥𝑥𝑘𝑘 in its last transmitted message, the transmitter assumes that 𝑥𝑥𝑘𝑘 has been locally detected by 𝑉𝑉𝑅𝑅 and, therefore, sets 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟 equal to 1. Otherwise, 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟 is set equal to the message reception probability that characterizes the Vehicle-to-Vehicle (V2V) link between 𝑉𝑉𝑅𝑅 and the vehicle 𝑉𝑉𝑇𝑇 that reported 𝑥𝑥𝑘𝑘. The reception probability depends on the channel load measured at 𝑉𝑉𝑅𝑅 and the distance between 𝑉𝑉𝑇𝑇 and 𝑉𝑉𝑅𝑅. It represents the probability that the message containing 𝑥𝑥𝑘𝑘 was correctly received by 𝑉𝑉𝑅𝑅. In case of multiple vehicles reporting 𝑥𝑥𝑘𝑘, 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟 is estimated by evaluating the complement of the joint probability that 𝑉𝑉𝑅𝑅 has not correctly received 𝑥𝑥𝑘𝑘 from any of the reporting vehicles, assuming each transmission is an independent event. B. V2X Communication Technology We assume that semantic V2X communications are performed employing the C-V2X sidelink communication technology, which allows vehicles to directly communicate without relying on the cellular infrastructure. We model the C-V2X sidelink performance using the analytical models released in [12]. These models accurately model the message reception probability taking into account the impact of cochannel interference (i.e., packet collisions), propagation impairments due to pathloss, shadowing, and fast-fading, and the half-duplex limitations of V2X transceivers. The models provide the message decoding probability as a function of the transmitter-receiver distance and the Channel Busy Ratio (CBR) measured at the receiver. The CBR is a channel load indicator that reports the fraction of channel resources (known as resource blocks in C-V2X sidelink) that are sensed as occupied by other vehicles over the last 𝑇𝑇 ms. In C-V2X sidelink, a message is sensed by a vehicle when its received signal power is above some pre-defined threshold, regardless of whether the message can be correctly decoded. Message sensing is modelled using the packet sensing probability analytical model also provided in [12]. V. NUMERICAL RESULTS This Section numerically characterizes the impact of redundancy estimation errors on semantic V2X communications. It quantifies the semantic content-selection error probability and numerically analyses the main conditions that influence its occurrence. To this aim, we consider a 2 km x 2 km driving scenario with an exogenous variables’ density of 350 vars/km2. The probability 𝑃𝑃(𝐷𝐷𝑘𝑘,𝑛𝑛) that a vehicle 𝑉𝑉 𝑛𝑛 locally detects an exogenous variable 𝑥𝑥𝑘𝑘 with its own onboard sensors is modelled with the following logistic function: 𝑃𝑃�𝐷𝐷𝑘𝑘,𝑛𝑛�= 1 1 + 0.08 ∗ 𝑒𝑒 −0.08(𝐷𝐷𝑘𝑘,𝑛𝑛−60) , (1) where 𝐷𝐷𝑘𝑘,𝑛𝑛 represents the distance between 𝑥𝑥𝑘𝑘 and 𝑉𝑉 𝑛𝑛. According to (1), a vehicle’s perception range is 150 m. The perception range is defined as the distance at which the probability of locally detecting an exogenous variable is equal to zero. We assume that vehicles transmit a new message every 𝑇𝑇 = 20 ms, a setting that is representative of a multiV2X CAD services scenario [8]. We consider a 10 MHz channel bandwidth and we assume that vehicles communicate employing the QPSK-0.7 Modulation and Coding Scheme (MCS). We assume that the size of an exogenous variable is 35 bytes, a setting that corresponds to the typical size of a detected object in cooperative perception [6]. We set the percentage of relevant variables within the relevance range to 30% (∆𝐻𝐻 = 0.3) and the relevance range 𝑅𝑅𝑚𝑚𝑚𝑚𝑚𝑚 to 400 m. The relevance range is selected according to the V2X service requirements defined by ETSI in [13]. (a) Hard (b) Soft Fig. 3. Semantic content-selection error probability. Hard and soft redundancy estimation. Fig. 3 reports the Semantic Content-Selection Error (SCSE) probability, P(SCSE), obtained when employing the hard (Fig. 3a) and soft (Fig. 3b) redundancy estimation approach. P(SCSE) quantifies the probability that at least one relevant exogenous variable is not delivered to an intended receiver due to a redundancy estimation error at the transmitter. Fig. 3 reports the semantic content-selection error probability as a function of the number of communicating vehicles 𝑁𝑁 and for varying communication constraints Γ. Fig. 3a shows that the semantic content-selection error probability can be as large as 0.2 when the communication constraint is more relaxed (Γ = 8) and the hard redundancy estimation approach is employed. This means that, on average, 1 every 5 transmitted messages do not contain relevant information for an intended receiver due to a redundancy estimation error. The comparison between Fig. 3a and Fig. 3b highlights that soft redundancy estimation can greatly reduce the semantic content-selection error probability. In Fig. 3b, P(SCSE) is confined below 0.05, a smaller yet non-negligible value. Fig. 3 also shows that the semantic content-selection error probability is sensitive to both the number of communicating vehicles 𝑁𝑁 and the communication constraint Γ. The semantic content-selection error probability reduces in less dense (smaller 𝑁𝑁) and more constrained (smaller Γ) scenarios. A larger number of communicating vehicles increases the probability that two or more vehicles locally detect the same exogenous variables (Condition A in Sec. III). This is a necessary condition to trigger the redundancy estimation process that influences the semantic content-selection error probability. The occurrence probability of Condition A, P(A), is reported in Fig. 4a as a function of 𝑁𝑁. In Fig. 4a, P(A) represents the probability that two or more vehicles commonly detect at least one exogenous variable. Fig. 4a shows that P(A) grows from 0.45 to 1 when moving from the least to the most densely populated driving scenario. Note that P(A) does not depend on either the employed redundancy estimation approach or the communication constraints Γ. More constrained communication scenarios (i.e., lower Γ values) do not allow the transmitting vehicles to transmit all locally detected exogenous variables that are estimated as relevant for the intended receivers. The transmission of exogenous variables estimated to be relevant is another key condition that affects the semantic content-selection error probability (Condition C in Sec. III). Fig. 4b reports the occurrence probability of Condition C, P(C), measured as the probability that an exogenous variable estimated as relevant is included in the transmitted message. The impact of varying communication constraints on P(C) is clear. Γ = 8 actually represents an unconstrained communication scenario in our analysis. When Γ = 8, vehicles can transmit all locally detected exogenous variables that are estimated as relevant for the intended receivers and P(C) is equal to 1. On the other hand, Γ = 2 represents a constrained communication scenario where only a restricted subset of relevant variables can be included in the transmitted messages and, therefore, P(C) drops to 0.35. The reduction of P(C) experienced under more stringent communication constraints Γ ultimately reduces the semantic content-selection error probability. However, it is worth highlighting that a reduction of P(C) inherently limits the capability of semantic V2X communications to provide the intended receivers with the most relevant information. The values of P(C) reported in Fig. 4b have been obtained employing the hard redundancy estimation approach. However, similar trends and values are observed when soft redundancy estimation is employed. The probabilistic approach employed by soft redundancy estimation reduces the probability that a piece of information is estimated as redundant (Condition E in Section III). This, in turn, reduces the semantic content-selection error probability (see Fig. 3), but it also increases the probability that an exogenous variable is estimated as non-redundant and is therefore re-transmitted by multiple vehicles. This is visible in Fig. 5, which reports the probability that the transmitter retransmits at least one exogenous variable that was already reported by another vehicle. The comparison between Fig. 5a and Fig. 5b shows that soft redundancy estimation significantly increases the re-transmission probability compared to hard redundancy estimation. Such increase in the re-transmission probability is reflected into the higher CBR (a) P(A) (b) P(C) Fig. 4. Occurrence probability of Condition A, P(A), and Condition C, P(C). (a) Hard (b) Soft Fig. 5. Re-transmission probability. Hard and soft redundancy estimation. values attained by the soft redundancy estimation technique in Fig. 6. Fig. 6 depicts the CBR corresponding to the semantic content-selection error probability levels reported in Fig. 3 as a function of 𝑁𝑁. The CBR differences between hard and soft redundancy are more visible as the communication constraint Γ is relaxed, since decreasing values of Γ inherently reduce the re-transmission probability. The comparative analysis of Fig. 3a and Fig. 6a shows that hard redundancy estimation can lead to non-negligible semantic content-selection error probability levels for CBR values ranging from medium to highly congested channel conditions. VI. CONCLUSIONS The limited availability of spectrum resources and the bandwidth-demanding nature of V2X-enabled CAD services are set to critically challenge the scalability of V2X networks as the number of connected vehicles increases. Semantic and task-oriented V2X communications have the potential to address these scalability challenges by focusing on the transmission of only the most relevant information to the intended receivers. The relevance of a piece of information strongly depends on whether it is redundant or not for an intended receiver. However, redundancy estimation is a challenging task due to the broadcast nature of V2X communications and may be subject to estimation errors. This study presents the first analysis of the impact of redundancy estimation errors on semantic and task-oriented V2X communications. In particular, it identifies conditions under which redundancy estimation errors can prevent the delivery of relevant information to the intended receivers, leading to a semantic content-selection error. The analysis numerically quantifies the occurrence probability of semantic content-selection errors when using two different redundancy estimation techniques under varying vehicular densities and communication constraints. The obtained results reveal that the semantic content-selection error probability augments as the number of communicating vehicles increases and less stringent communication constraints are enforced. Conversely, more stringent communication constraints reduce this probability but inherently limit the delivery of relevant information to the intended receivers. The obtained results also show that the hard redundancy estimation approach traditionally employed in existing V2X solutions can lead to a large semantic contentselection error probability, especially in unconstrained communication scenarios. A large semantic content-selection error probability can potentially reduce the amount of relevant information delivered to the intended receivers, reducing their contextual awareness. This study has also analysed a soft redundancy estimation approach, and showed that it can significantly reduce the semantic content-selection error probability while increasing the channel load. Future work includes analysing the impact of semantic content-selection errors and potential mitigation measures on the design and effectiveness of semantic and task-oriented V2X communications – a communication paradigm designed to maximize the amount of relevant information delivered to the intended receivers at the minimum communication cost. REFERENCES [1] C2C-CC, “Road Safety and Road Efficiency Spectrum Needs in the 5.9 GHz for C-ITS and Cooperative Automated Driving,” C2C TR 2050 V2.00, Feb. 2020. [2] ETSI, “Analysis of the Collective Perception Service (CPS); Release 2,” ETSI TR 103 562 V2.1.1, Dec. 2019. [3] ETSI, “Collective Perception Service; Release 2,” ETSI TS 103 324 V2.1.1, June 2023. [4] G. Thandavarayan et al., “Cooperative Perception for Connected and Automated Vehicles: Evaluation and Impact of Congestion Control,” IEEE Access, vol. 8, pp. 197665-197683, 2020. [5] G. Thandavarayan, M. Sepulcre and J. Gozalvez, “Redundancy Mitigation in Cooperative Perception for Connected and Automated Vehicles,” in Proc. VTC2020-Spring, Antwerp, Belgium, 2020,pp.1-5. [6] G. Thandavarayan et al., “Scalable cooperative perception for connected and automated driving,”, Elsevier Journal of Network and Computer Applications, vol. 216, 2023. [7] Q. Delooz et al., “Analysis and Evaluation of Information Redundancy Mitigation for V2X Collective Perception,” IEEE Access, vol. 10, pp. 47076-47093, 2022. [8] M. Sepulcre et al., “LTE-V2X Scalability and Spectrum Requirements to support Multiple V2X Services”, in Proc. VTC2024-Fall, 7-10 October 2023, Washington DC, USA. [9] C. Shannon and W. Weaver, “The Mathematical Theory of Communication,” University of Illinois Press, 1962. [10] L. Lusvarghi et al., “The Search for Relevance: A Context-Aware Paradigm Shift in Semantic and Task-Oriented V2X Communications,” under review. [Online]. Available: https://uwicore.umh.es/semanticV2X/publications.html. [11] H. Shen, S. He, L. Yu and A. Sarker, “Prediction-based redundant data elimination with content overhearing in wireless networks,” in Proc. 2017 IEEE PerCom conference, Kona, HI, USA, 2017, pp. 50-58. [12] M. Gonzalez-Martín et al., “Analytical Models of the Performance of C-V2X Mode 4 Vehicular Communications,” IEEE Transactions on Vehicular Technology, vol. 68, no. 2, pp. 1155-1166, Feb. 2019. [13] ETSI, “5G; Service requirements for enhanced V2X scenarios,” ETSI TS 122 186 V17.0.0, April 2024. (a) Hard (b) Soft Fig. 6. Channel Busy Ratio (CBR). Hard and soft redundancy estimation.