PaperlayerView exact revision on arXiv

Source-linked AI summary

Federated Learning in Intelligent Transportation Systems: Recent Applications and Open Problems

Shiying Zhang, Jun Li, Long Shi, Ming Ding, Dinh C. Nguyen, Wuzheng Tan, Jian Weng, Zhu Han

arXiv:2309.11039v1cs.LGcs.AIcs.DC

TL;DR

Dynamic vehicle networks make timely and accurate decisions difficult, while wireless connectivity exposes vehicle information to privacy and security risks. This paper surveys FL applications in ITS by scenario and synthesizes their challenges and supporting mechanisms. It concludes that FL addresses some ITS problems but leaves challenges, including uneven data and limited resources, for further work.

  • Problem

    Dynamic IoV networks face difficult timeliness and accuracy demands, and their wireless information exchange is vulnerable to privacy and security risks.

    Source: 2) Privacy and Security Issues · p. 2

  • Method

    The paper surveys FL-enabled ITS applications across four scenarios and organizes ITS challenges and supplementary mechanisms for addressing FL deficiencies.

    Source: A. Current State of Art and Our… · p. 4, §III · p. 7

  • Results

    The survey identifies four application scenarios and concludes that FL can address key ITS challenges while open challenges remain.

    Source: §III · p. 7, §VI · p. 21

  • Takeaways & Limitations

    FL can alleviate some ITS problems, but dynamic deployment retains communication overhead and latency and may require additional mechanisms.

    Source: §IV · p. 12

    FL does not eliminate privacy and security concerns: model parameters or gradients can reveal vehicle data, while defenses can compromise accuracy or add overhead.

    Source: §V · p. 19

Abstract

from arXiv · show

Intelligent transportation systems (ITSs) have been fueled by the rapid development of communication technologies, sensor technologies, and the Internet of Things (IoT). Nonetheless, due to the dynamic characteristics of the vehicle networks, it is rather challenging to make timely and accurate decisions of vehicle behaviors. Moreover, in the presence of mobile wireless communications, the privacy and security of vehicle information are at constant risk. In this context, a new paradigm is urgently needed for various applications in dynamic vehicle environments. As a distributed machine learning technology, federated learning (FL) has received extensive attention due to its outstanding privacy protection properties and easy scalability. We conduct a comprehensive survey of the latest developments in FL for ITS. Specifically, we initially research the prevalent challenges in ITS and elucidate the motivations for applying FL from various perspectives. Subsequently, we review existing deployments of FL in ITS across various scenarios, and discuss specific potential issues in object recognition, traffic management, and service providing scenarios. Furthermore, we conduct a further analysis of the new challenges introduced by FL deployment and the inherent limitations that FL alone cannot fully address, including uneven data distribution, limited storage and computing power, and potential privacy and security concerns. We then examine the existing collaborative technologies that can help mitigate these challenges. Lastly, we discuss the open challenges that remain to be addressed in applying FL in ITS and propose several future research directions.

I. INTRODUCTION

ITS must make timely, accurate decisions in dynamic vehicle networks, yet latency, privacy and security risks, and data and sensor limitations complicate that task. The introduction presents federated learning (FL) as a distributed approach and describes how it may address some ITS challenges while retaining limitations of its own.

I. INTRODUCTION

1) High Response Latency:

  • Unstable links and uneven vehicle resources make training and decision-making slow: mobility disrupts routes, while heterogeneous data and computing capacity produce widely varying training durations.Routing alternatives may mitigate link failures but their alternating process can add latency; frequent terminal-server interactions can also congest channels.Source: 1) High Response Latency · pp. 1–2

2) Privacy and Security Issues:

  • Wireless vehicle-to-vehicle, vehicle-to-RSU, and RSU-to-server links expose training to attacks, while low-latency requirements constrain security processing.Existing edge access-control and rule-based approaches are described as complex and poorly scalable.Source: 2) Privacy and Security Issues · p. 2

3) Low Decision-making Accuracy:

  • Data islands, sparse labels and rare events, and insufficient real-time sensor fusion constrain decision accuracy and model generalization.The introduction identifies poor generalization from difficult real-world data collection and says real-time fusion demands substantial computing and communication performance.Source: 3) Low Decision-making Accuracy · p. 2
  • FL coordinates local training and central model averaging without direct data exchange, and selective or asynchronous aggregation can mitigate information imbalance.The introduction contrasts this with centralized collection, which consumes bandwidth and adds latency; it also notes FL alone retains communication overhead and latency in dynamic ITS.Source: 3) Low Decision-making Accuracy · pp. 2–3

A. Current State of Art and Our Contributions

The paper positions its survey against reviews with limited scenario coverage or emphasis on particular FL aspects. It organizes FL-enabled ITS research by application scenarios, synthesizes ITS challenges and supplementary mechanisms, and identifies future research directions.

A. Current State of Art and Our Contributions

  • Earlier reviews vary in scope: some focus on robustness, resource management, or blockchain-assisted FL, while many do not organize work by ITS task scenarios.The paper notes that scenario organization can support more intuitive understanding of FL-enabled ITS.Source: A. Current State of Art and Our… · p. 4
  • The survey classifies FL-enabled ITS work into traffic management, object recognition, service providing, and traffic status identification, then describes technologies and architectures in each.Its scenario framing follows task phases including perception, prediction, and decision.Source: A. Current State of Art and Our… · p. 4
  • The survey summarizes ITS resource, transmission, security, privacy, and distribution challenges, and discusses supplementary mechanisms to compensate for FL deficiencies.It describes this systematization of supplementary mechanisms as absent from existing surveys at the time of writing.Source: A. Current State of Art and Our… · p. 4
  • The paper concludes by identifying remaining open challenges and proposing future research directions for FL-enabled ITS.Its stated aims include helping researchers track developments and identify directions for future work.Source: Table I · p. 4, A. Current State of Art and Our… · p. 5

B. Structure of The Survey

The survey is organized around ITS application scenarios, FL architectures, deployment challenges, and future directions. Its FL-enabled ITS overview describes a three-layer architecture linking local data collection and training to RSU relays and central aggregation.

B. Structure of The Survey

  • The survey maps application scenarios and FL challenges before discussing future directions, with its overall organization outlined in Fig. 1.It covers application results by scenario, architecture enhancements, open challenges, and a conclusion.Source: B. Structure of The Survey · p. 5, Figure 1 · p. 6

II. FL-ENABLED ITS

  • A hierarchical FL architecture uses vehicle-to-server links alongside top, middle, and bottom layers to accommodate dense networks and short-range connectivity.The central server coordinates aggregation, RSUs relay models, and vehicles and roadside sensors collect data and train locally.Source: §II · pp. 5–6
  • RSUs aggregate and store parameters, then relay server-updated models to vehicles by broadcast.RSUs have limited computing and storage capacity and may retain limited contextual information.Source: §II · p. 6
  • Vehicles cache sensor data and train locally before sending models to RSUs; overloaded vehicles may use flexible upload deadlines or offload training and storage to trusted edge resources.The framework also treats vehicular, edge, mobile IoT, and infrastructure nodes as possible FL clients.Source: §II · pp. 6–7

III. APPLICATION SCENARIOS OF FL IN ITS

The survey classifies FL-enabled ITS applications into four scenarios according to ITS task phases. It links recognition and prediction tasks to distinct data types and places management and service delivery in the decision phase.

III. APPLICATION SCENARIOS OF FL IN ITS

  • The survey groups FL-enabled ITS applications into object recognition, traffic status identification, traffic management, and service providing.Examples include license-plate recognition, autonomous controller design, congestion monitoring, knowledge sharing, and route planning.Source: §III · p. 7
  • The scenario division follows ITS task phases: image-based recognition, time-series traffic status analysis, and decision-stage management or service delivery.Traffic status examples include travel-mode recognition and traffic-flow prediction.Source: §III · p. 7
  • Parking control illustrates decision-stage service: vehicles share information with a control center while seeking parking, and decisions follow training.Source: §III · p. 7

A. Object Recognition

FL combines distributed vehicle and roadside-sensor data for ITS object recognition, addressing individual-view obstructions and dynamic-scene ambiguity. The survey covers hierarchical, collaborative-inference, edge-oriented, and semi-supervised approaches while noting real-time multimodal analysis as an ongoing challenge.

A. Object Recognition

  • Object-recognition models face a trade-off between detection accuracy and inference speed: two-stage R-CNN is more accurate but slower and more complex than one-stage detectors.Source: A. Object Recognition · p. 8
  • FL aggregation enabled detection of trucks and pedestrians missed by a single vehicle's local training in a four-client, five-epoch YOLO simulation.The cited follow-up proposes two-layer CNN-based FL, with RSUs and the cloud assessing model quality for hierarchical selection.Source: A. Object Recognition · p. 8
  • Collaborative inference-based online FL, CIOFL, uses a large high-quality model to label real-world UAV images and expand training data when online samples are scarce.UAV mobility makes it difficult to add samples during online training.Source: A. Object Recognition · p. 8
  • Edge-deployable license-plate recognition addresses blur and orientation anomalies, while a receptive-field SNN encoding scheme is reported to outperform FL using CNN in noise immunity and accuracy.The cited discussion identifies multidirectional plates and motion blur as bottlenecks.Source: A. Object Recognition · p. 8
  • Semi-supervised FL frameworks target scarce labeled images; one UAV recognition approach combines a dynamic hyperparameter term with two active-learning strategies.The survey states that these methods help address label scarcity and improve recognition accuracy and effectiveness.Source: A. Object Recognition · pp. 8–9
  • FL can combine distributed vehicle and road-sensor data to address single-sensor occlusion and ambiguity from vehicle movement, but real-time multimodal analysis under low-latency requirements remains difficult.The survey also identifies road-aware sensor layout and high-fidelity image simulation as needs for reducing perceptual errors.Source: A. Object Recognition · p. 9

B. Traffic Status Identification

Traffic-status tasks require varied sensing and collaborative data, with FL applications spanning positioning, travel-mode identification, trajectory work, and traffic-flow prediction. The survey also notes detector scarcity as a constraint that FL collaboration can mitigate.

1) Vehicle Position:

  • Vehicle positioning supports navigation and lane keeping, but GNSS-dependent positioning is costly, and sensor-rich vehicles require high implementation and bandwidth resources.The passage describes multi-system data fusion with onboard and visual positioning sensors as a development direction.Source: 1) Vehicle Position · p. 9
  • FL-based cooperative perception and remote sensing have been used for GPS error correction and wider congestion-detection coverage, while smartphone inertial data can support positioning in closed areas without GPS.The cited closed-area approach uses GPS data only for training in open environments and is described as alleviating heterogeneity from different driving and device hardware.Source: 1) Vehicle Position · p. 9, 2) Travel Mode Identification · p. 9

2) Travel Mode Identification:

  • Travel-mode work addresses missing labels with semi-supervised FL and pseudo-labeling, and uses sequential one-class SVM learning for abnormal-trajectory detection at intersections.Other cited work explores personalized parameter updates and aggregation weighted toward comparable observations.Source: 2) Travel Mode Identification · pp. 9–10

3) Traffic Flow Prediction:

  • Traffic-flow prediction combines spatial and temporal inputs, including location, time, camera, speed, and weather, while FL-based collaboration can mitigate gaps caused by missing arterial-road detectors.One cited FL-GRU approach uses random participant subsampling to reduce communication overhead.Source: 3) Traffic Flow Prediction · p. 10

C. Traffic Management

  • Traffic management uses object recognition and traffic-state identification to inform decisions such as parking and intersection signal control.Source: C. Traffic Management · p. 10

1) Parking Management:

The surveyed ITS applications span parking management, traffic signal control, and service delivery, including charging and route-planning services. Across these settings, FL-related work addresses operational constraints while leaving challenges around communication, mobility, and resource use.

1) Parking Management:

  • Parking-management work uses FL incentives to harness parked vehicles for workloads, while parking trajectory planning must handle narrow spaces, nondifferentiable points, and stricter obstacle avoidance.The incentive-based approach models parking-lot and vehicle interaction as a non-cooperative game and analyzes a unique Stackelberg equilibrium.Source: 1) Parking Management · p. 10

2) Traffic Signal Control:

  • Signal-control studies use evolutionary and metaheuristic optimization and reinforcement learning, while FL simulations report faster convergence; centralized RL is infeasible for large-scale control.Optimization targets include network, intersection, roundabout, and timing-cycle control; the cited RL process uses traffic state, policy actions, and rewards.Source: 2) Traffic Signal Control · pp. 10–11
  • FL-based edge cooperative caching supports dynamic caching, but high-precision map caching remains largely theoretical and dynamic topologies and high-speed mobility remain ongoing challenges.Traffic management can require frequent inter-organization exchanges and substantial model-aggregation communication overhead.Source: 2) Traffic Signal Control · p. 11

1) Charging Service:

  • Charging-service work applies FL to load prediction, communication-efficient data clustering, cross-platform recommendations, and energy-demand forecasting and dispatch.Examples combine FL with random forests and CNNs, hash and RSA encryption, probabilistic prediction, or HFL with LSTM and stochastic games.Source: 1) Charging Service · p. 11

D. Service Providing

  • Charging design must balance operator, aggregator, and EV objectives amid fluctuating electricity demand, uncertain heat loads, cost control, and limited charging-station utilization.The section also notes that a mature public-service system has not yet emerged; centralized charging optimization can face system-crash risks and added complexity as constraints increase.Source: 1) Charging Service · pp. 11–12, D. Service Providing · p. 11

2) Route Planning:

Route-planning services must respond to changing road networks, traffic rules, and dynamic obstacles, while FL-enabled ITS faces broader mobility, resource, privacy, and communication challenges. The paper presents FL as a partial response and points to complementary mechanisms for remaining problems.

2) Route Planning:

  • Vehicle route planning must account for road layout and traffic rules, refresh routes as networks change, and manage latency and memory constraints.One cited approach places routing at RSU network edges with intermittent cloud access and compresses the search space; another uses a time-dependent graph and enhanced A*.Source: 2) Route Planning · p. 12
  • FL-enabled route planning still faces high computational costs, potential deadlocks as vehicle counts grow, and substantial memory and bandwidth demands for RL-based dynamic planning.UAV planning adds flight obstacles and limited power-storage constraints; cited work combines federated DRL or joint power distribution and UAV scheduling.Source: 2) Route Planning · p. 12

IV. CHALLENGES IN ITS BASED FL

  • FL-enabled ITS can alleviate privacy and computational-overhead problems, but may add communication overhead and latency; blockchain, UAVs, and edge caching are proposed as complementary mechanisms.The discussion frames these mechanisms as ways to mitigate or potentially solve existing problems in highly dynamic ITS settings.Source: §IV · p. 12

1) Non-uniform Samples and Equipment:

The section examines FL adaptations to uneven data, mobile participants, and heterogeneous resources, including changes to aggregation, client selection, and model updates. It also notes privacy and efficiency limits in some proposed strategies.

1) Non-uniform Samples and Equipment:

  • Vehicle-selection methods use indicators such as vehicle speed, image quality, link stability, model contribution, execution time, and update similarity, with some approaches jointly optimizing resource allocation.The cited joint optimization considers latency, transmission energy, mobility, image quality, local accuracy, CPU frequency, and transmission power.Source: 1) Non-uniform Samples and Equipment · pp. 13–14
  • Cloud-side selection based on vehicle indicators raises a permissions issue for information such as location, while anonymization or obscuring can reduce privacy risks but adds information exchange.A server-held, ground-truth validation dataset is proposed to assess selected client models without being sufficient to train a global model.Source: 1) Non-uniform Samples and Equipment · p. 14
  • Mobility-aware aggregation can use a capable vehicle as a server or combine synchronous edge aggregation with cloud aggregation of heterogeneous models.The section identifies nonuniform data and vehicles left behind as consequences of IoV’s dynamic topology, and describes participant selection and aggregation changes as possible responses.Source: 1) Non-uniform Samples and Equipment · pp. 12–14
  • Adaptive local training steps and model quantization can reduce training time and communication costs, but direct transfer of some standard-FL strategies to ITS can reduce performance.Sharing client data to improve Non-IID learning risks serious privacy breaches, while local-loss regularization is described as inefficient; coding redundancy is another proposed response to stragglers.Source: 1) Non-uniform Samples and Equipment · pp. 14–15

2) Unstable Transmission:

Vehicle-density variation and limited bandwidth contribute to communication latency in ITS, although FL can distribute computation and avoid transmitting original data. The section surveys communication protocols, edge and aerial resources, and redesigned network links as responses to these constraints.

2) Unstable Transmission:

  • Changing vehicle density destabilizes information transmission and raises communication latency, while FL reduces cloud-computing overhead and avoids original-data transmission but still incurs frequent communication costs.The discussion distinguishes communication latency from computing latency, which it attributes to limited computing capability.Source: 2) Unstable Transmission · p. 15
  • DSRC and cellular-based IoV have coverage, rate, security, or congestion limitations, motivating heterogeneous protocols and FL-based management of dynamic on-vehicle WiFi links.One cited design combines DSRC for frequent control messages and small data with millimeter-wave communication for mass communication; another estimates packet loss and latency under link switching and retransmission.Source: 2) Unstable Transmission · p. 15
  • FL-based routing and resource allocation can estimate MEC storage and energy costs under storage and latency constraints, while UAV relays and UAV-assisted FL address connectivity and energy optimization.The UAV-assisted work considers placement, power control, model accuracy, and bandwidth allocation in an energy-minimization problem.Source: 2) Unstable Transmission · p. 15
  • Multi-hop clustering and end-to-end FL are used to reduce redundant communication, while blockchain-assisted approaches address privacy-conscious communication and model propagation.The cited blockchain methods analyze communication and consensus latency or use Proof-of-FL competition in smart contracts for message propagation.Source: 2) Unstable Transmission · p. 15
  • Fog communication designs include proxy-broker, publish-subscribe, and localized peer-to-peer approaches, while SDN can coordinate routes and support service-specific QoS configurations.The section describes fog resources as being placed at the network edge and SDN as combining distributed fog deployment with programmable network control.Source: 2) Unstable Transmission · pp. 15–16

B. Limited Storage and Computing Capability

The section surveys ways FL-enabled ITS can address constrained computation and storage, including offloading, caching, and auxiliary infrastructure. These approaches still face resource, latency, trust, and deployment trade-offs.

Limited Storage and Computing Capability

  • Vehicles have substantial energy reserves relative to traditional MEC settings, yet their storage and computing limits remain relevant to FL deployment.Real-time data collection and storage remain significant concerns, and existing approaches commonly optimize resource scheduling or architecture design.Source: B. Limited Storage and Computing… · p. 16

Auxiliary Mechanism

  • UAVs and edge servers can act as auxiliary resources, but UAV-FL work often overlooks transmission distance, delay, energy use, and distributed deployment.Proposed approaches include FL-based multi-agent offloading, power-allocation scheduling, and UAV cognitive radio networks. Third-party edge storage also raises trust concerns.Source: B. Limited Storage and Computing… · p. 16

Edge Cache Optimization

  • Edge caches must balance hit rates, storage pressure, and offloading delay; FL can allocate aggregation weights by vehicle storage resources.Passive caching risks obsolescence and requires backup storage, whereas active caching uses content prediction. FL-based offloading can improve decisions but incurs additional latency.Source: B. Limited Storage and Computing… · pp. 16–17
  • Dynamic on-board storage can transfer data from departing to arriving vehicles over a one-hop link, with erasure-code redundancy used to maintain integrity.The proposed transfer occurs at designated entrance and exit areas to reduce data loss when vehicles leave the management area.Source: B. Limited Storage and Computing… · p. 17

Auxiliary Mechanism

  • SDN-assisted FL coordinates task migration and model aggregation across changing network topologies, while adding network complexity and computational redundancy.Local SDN controllers gather vehicle state information and manage task migration; distributed controllers can select capable execution nodes and suitable FL aggregation servers.Source: B. Limited Storage and Computing… · p. 17

Edge Cache Optimization

  • FL can reduce centralized training-data collection in edge caching, but lightweight edge vehicles often cannot independently support a reinforcement-learning framework.FL is described primarily as a privacy-preserving mechanism in edge caching; RL-based resource allocation is reported as effective for energy efficiency, subject to device capacity.Source: B. Limited Storage and Computing… · p. 17

C. High Information Sensitivity

Vehicle connectivity and frequent location-based communication expose ITS to privacy and security threats, while FL's local-data approach offers only partial protection. The section reviews blockchain, encryption, and network-security approaches alongside their remaining costs and vulnerabilities.

C. High Information Sensitivity

  • Frequent vehicle-RSU communication and movement constraints heighten tracking risks, while uploaded gradients can still reveal original data and malicious participants can disrupt training.The described threats include eavesdropping, location attacks, false model parameters, and attackers manipulating congestion signals.Source: C. High Information Sensitivity · pp. 17–18

1) Blockchain:

  • Blockchain-supported FL can decentralize trust and address untrusted storage or tampering, but direct blockchain coupling can reduce efficiency in ITS P2P networks.Blockchain-based identity authentication preserves identity privacy, but does not protect model parameters from poisoning or address transmission security and location privacy.Source: 1) Blockchain · p. 18

2) Encryption Mechanism:

  • Differential privacy perturbs local outputs or model information, but its privacy factor requires tuning and added noise can compromise accuracy in safety-critical ITS settings.Four proposed LDP mechanisms have lower worst-case noise variance over segmented privacy-budget ranges [177]; DP-based methods also address topology information and membership inference.Source: 2) Encryption Mechanism · p. 18
  • ITS encryption research includes lightweight gradient encryption, data-source-specific FL with CPC, and identity anonymity and traceability schemes.The cited identity-anonymity scheme reports nearly 99% privacy protection in a single full authentication using an 8-bit key.Source: 2) Encryption Mechanism · pp. 18–19
  • Homomorphic encryption is considered impractical for large-scale ITS because of its substantial computing requirements, while SDN-based security must account for uneven channels, resource constraints, and varied topologies.SDN can segment networks and enforce policies dynamically, but those operational conditions remain necessary design considerations.Source: 2) Encryption Mechanism · p. 19

3) Detection Mechanism:

The survey describes FL-based detection mechanisms and identifies open research directions spanning privacy, security, resources, architecture, and deployment. It emphasizes that specialized ITS solutions and real-world evaluation remain incomplete.

3) Detection Mechanism:

  • FL-based detection work includes passive-attacker detection using semi-supervised self-labeling and collaborative leakage detection using data transformation and randomized, DP-distorted updates.The latter aims to prevent model differences from revealing provider information.Source: 3) Detection Mechanism · p. 19

V. CHALLENGES AND FUTURE RESEARCH DIRECTIONS

  • ITS needs lightweight encryption designed for its constraints, alongside real-scenario testing against varied attacks such as backdoor and symbol-flipping attacks.The survey notes that many existing encryption algorithms add computational overhead and that few works test defenses across distinct attack types.Source: §V · p. 19
  • FL privacy and security remain limited: model or gradient information can expose vehicle data, DP can reduce decision accuracy, and crafted training samples can manipulate data distributions.Authentication and other defenses also add communication overhead and system-management costs.Source: §V · p. 19
  • FL incentives and participant selection lack standardization, while labeled samples remain scarce across differing vehicle sensors and environments.Data augmentation may add noise or require human intervention, and transfer learning must address inter-domain discrepancies and adaptation challenges.Source: §V · pp. 19, 21
  • Many FL-RL models for signal control and route planning are not yet applicable to real-world environments, and the survey calls for stronger simulator performance and human-intervention modeling.It also proposes studying generalized scenarios that use FL as a shared underlying service and more efficient sensor configurations across tasks.Source: §V · p. 21
  • Future work should balance privacy protection against model precision and develop lightweight verification for edge results, since third-party RSUs cannot be fully trusted.Static RSUs may expose information to newly entering vehicles, while resource-limited vehicles may offload computing tasks to edge nodes.Source: §V · p. 21
  • Reducing transmission overhead requires designing suitable compressed models and combining communication standards to improve network reliability.The survey identifies pruning and quantization as existing compression approaches in FL.Source: §V · p. 21
  • Few FL architectures are tailored to distinct ITS data types and task settings, and the survey calls for deeper integration of sensors, network computing, and vehicle networks.Mature commercial platforms such as OpenDaylight and Cisco Kinetic are also identified as possible ITS deployment options.Source: §V · p. 21

VI. CONCLUSION

The conclusion summarizes the survey of FL integration in ITS, including its applications, deployment challenges, security mechanisms, and future research directions. It notes that several enabling ITS technologies remain outside the paper's detailed discussion.

VI. CONCLUSION

  • The survey reviews FL applications in ITS and discusses how FL may address uneven data, resource limits, transmission, privacy, and security challenges.It also covers blockchain security mechanisms, UAV-aided resource scheduling, open problems, and future directions.Source: §VI · p. 21
  • Integrated sensing, mobile fog computing, and cluster routing are among the ITS enabling technologies left for further discussion.Source: §VI · p. 22
Loading 2309.11039v1…