Paper deep dive

Resilient Consensus-Based Target Tracking under False Data Injection Attacks in Multi-Agent Networks

Amir Ahmad Ghods, Mohammadreza Doostmohammadian

Year: 2026Venue: arXiv preprintArea: eess.SYType: PreprintEmbeddings: 58

Intelligence

Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 94%

Last extracted: 8/4/2026, 3:30:21 AM

Summary

This paper proposes a resilient consensus-based target tracking algorithm for multi-agent networks that mitigates false data injection attacks (FDIA). The method integrates a nearly-constant-velocity (NCV) model with saturation-based filtering to suppress impulsive measurement variations and employs a dynamic innovation threshold mechanism to detect and isolate compromised agents. Simulations demonstrate that the approach significantly improves estimation accuracy and convergence speed while maintaining robustness against localized, coordinated, and transient attacks.

Entities (9)

Consensus-Based Estimation Algorithm · algorithm · 98%False Data Injection Attacks · threat · 98%False Data Injection Detection Mechanism · mechanism · 97%Multi-Agent Networks · system · 96%Nearly-Constant-Velocity Model · model · 95%Saturation-Based Filtering · technique · 95%Mean Squared Estimation Error · metric · 93%Innovation Thresholds · parameter · 92%Time-of-Arrival Multilateration · method · 90%

Relation Signals (7)

Consensus-Based Estimation Algorithm → mitigates → False Data Injection Attacks

confidence 97% · enhance the resilience of decentralized target tracking algorithms against ... false data injection attacks

False Data Injection Detection Mechanism → detects → False Data Injection Attacks

confidence 96% · dynamic false data injection detection and isolation mechanism ... identify and disregard suspicious measurements

Consensus-Based Estimation Algorithm → uses → Saturation-Based Filtering

confidence 95% · integrates a nearly-constant-velocity model with saturation-based filtering

Consensus-Based Estimation Algorithm → uses → Nearly-Constant-Velocity Model

confidence 95% · integrates a nearly-constant-velocity model with saturation-based filtering

Consensus-Based Estimation Algorithm → optimizes → Mean Squared Estimation Error

confidence 93% · reduces the mean squared estimation error (MSEE) by about 65%

False Data Injection Detection Mechanism → uses → Innovation Thresholds

confidence 92% · uses innovation thresholds to identify and disregard suspicious measurements

Multi-Agent Networks → employs → Time-of-Arrival Multilateration

confidence 90% · measurement model is based on Time-of-Arrival (TOA) multilateration

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:Distributed target tracking in multi-agent networks plays a critical role in cooperative sensing and autonomous navigation. However, it faces significant challenges in highly dynamic and adversarial setups. This study aims to enhance the resilience of decentralized target tracking algorithms against measurement faults and cyber-physical threats, especially false data injection attacks. We propose a consensus-based estimation algorithm that integrates a nearly-constant-velocity model with saturation-based filtering to suppress impulsive measurement variations and promote robust, distributed state estimation. To counteract adversarial conditions, we incorporate a dynamic false data injection detection and isolation mechanism that uses innovation thresholds to identify and disregard suspicious measurements before they can degrade the global estimate. The effectiveness of the proposed algorithms is demonstrated through a series of simulation-based case studies under both benign and adversarial conditions. The results show that increased network connectivity and higher consensus iteration rates improve estimation accuracy and convergence speed, while properly tuned saturation filters achieve a practical balance between fault suppression and accurate estimation. Furthermore, under localized, coordinated, and transient false data injection attacks, the detection mechanism successfully identifies compromised agents and prevents their data from corrupting the distributed global estimate. Overall, this study illustrates that the proposed algorithm provides a simplified fault-tolerant solution that significantly enhances the accuracy and resilience of distributed target tracking without imposing excessive communication or computational burdens.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

56,616 characters extracted from source content.

Expand or collapse full text

Signals 2025, 6, x hĴps://doi.org/10.3390/x Article Resilient Consensus-Based Target Tracking under False Data Injection Attacks in Multi-Agent Networks Amir Ahmad Ghods 1 and Mohammadreza Doostmohammadian 1, * 1 Faculty of Mechanical Engineering, Semnan University, Iran; amir-ghods@semnan.ac.ir (A Ghods); doost@semnan.ac.ir (M Doostmohammadian) * Correspondence: doost@semnan.ac.ir Abstract: Distributed target tracking in multi-agent networks plays a critical role in coop- erative sensing and autonomous navigation. However, it faces significant challenges in highly dynamic and adversarial setups. This study aims to enhance the resilience of de- centralized target tracking algorithms against measurement faults and cyber-physical threats, especially false data injection attacks. We propose a consensus-based estimation algorithm that integrates a nearly-constant-velocity model with saturation-based filtering to suppress impulsive measurement variations and promote robust, distributed state es- timation. To counteract adversarial conditions, we incorporate a dynamic false data injec- tion detection and isolation mechanism that uses innovation thresholds to identify and disregard suspicious measurements before they can degrade the global estimate. The ef- fectiveness of the proposed algorithms is demonstrated through a series of simulation- based case studies under both benign and adversarial conditions. The results show that increased network connectivity and higher consensus iteration rates improve estimation accuracy and convergence speed, while properly tuned saturation filters achieve a practi- cal balance between fault suppression and accurate estimation. Furthermore, under local- ized, coordinated, and transient false data injection attacks, the detection mechanism suc- cessfully identifies compromised agents and prevents their data from corrupting the dis- tributed global estimate. Overall, this study illustrates that the proposed algorithm pro- vides a simplified fault-tolerant solution that significantly enhances the accuracy and re- silience of distributed target tracking without imposing excessive communication or com- putational burdens. Keywords: Distributed target tracking; Consensus algorithms; Cyber-physical security; False data injection; Multi-agent systems; Decentralized estimation; Fault detection 1. Introduction In recent years, distributed target tracking has played a crucial role in various appli- cations, including autonomous surveillance, environmental monitoring, and cooperative robotics. These systems rely on distributed estimation and local communication among agents to track moving targets in a scalable, robust, and real-time manner. However, maintaining estimation accuracy and network resilience in distributed systems is signifi- cantly challenged by faulty measurement, environmental disturbances, and the increas- ing risk of cyber-physical threats. To address these challenges, researchers have examined various architectural strate- gies for estimation and communication in multi-agent systems. Estimation architectures Academic Editor: Firstname Last- name Received: date Revised: date Accepted: date Published: date Citation: To be added by editorial staff during production. Copyright: © 2025 by the authors. Submitted for possible open access publication under the terms and conditions of the Creative Commons Attribution (C BY) license (https://creativecommons.org/license s/by/4.0/). Signals 2025, 6, x FOR PEER REVIEW 2 of 21 are generally divided into two primary categories based on how information is shared. The first category is a centralized platform, where a single central node collects data from all agents in the communication network, conducts the necessary computations, and sub- sequently shares the results with those agents [1-4]. While this approach can achieve high estimation accuracy, it suffers from several critical drawbacks. The reliance on a single node introduces a potential point of failure, requires high computational capacity at that node, and leads to poor scalability and security concerns. As a result, many researchers have shifted their attention toward decentralized frameworks, in which the central node is eliminated and each agent processes data locally while communicating with neighbor- ing agents. This structure improves system scalability and fault tolerance, and signifi- cantly reduces the communication overheads. Nonetheless, decentralized systems come with their own set of challenges, including increased network complexity and the need for reliable communication and coordination to ensure accurate and consistent perfor- mance across all agents. Several estimation strategies have been explored to address these challenges. Among the most widely studied are variants of the Kalman filter which are adapted for distrib- uted use, commonly known as Distributed Kalman Filters (DKFs) [5-10]. While these ap- proaches allow each agent to maintain a local estimate, they still face limitations in prac- tice. DKFs typically require high communication bandwidth due to the need for covari- ance information exchange and are not well-suited to adversarial environments due to their limited robustness and dependence on accurate model assumptions. These factors reduce their scalability and make them vulnerable to both faults and intentional data cor- ruption in dynamic settings. To address the shortcomings of DKFs, researchers have turned to adaptive and learn- ing-based approaches that aim to improve resilience. For example, model-free adaptive control allows agents to follow reference signals using only local information, making it effective even in the presence of communication delays or denial-of-service attacks [11]. Neural network–based techniques have been employed to approximate unknown nonlin- ear dynamics in time-varying systems, enabling faster adaptation and more accurate tracking [12]. Likewise, adaptive fuzzy controllers provide a way to estimate unmeasura- ble states and reduce the impact of cyber-physical attacks, keeping tracking errors within acceptable bounds in large-scale networks [13]. More recently, data-driven and federated learning strategies have been explored to counter more severe disruptions such as Byzan- tine attacks or actuator faults, often by combining robust aggregation rules with online learning of uncertain system dynamics [14,15]. Despite these advances, such methods of- ten entail trade-offs, including intensive computation, significant communication over- head, or complex agent coordination. These drawbacks limit their practicality in large, resource-constrained networks and leave open the need for simpler, more scalable, and inherently robust alternatives. To overcome these limitations, researchers have increasingly turned to consensus- based estimation algorithms [16-25]. These approaches use simple local rules and infor- mation exchange between neighboring agents to gradually reach a shared understanding across the entire network. Unlike traditional distributed Kalman filters, consensus-based algorithms require less communication, scale more efficiently with larger systems, and are more resilient to issues like unstable connections and noisy measurements. Their simple design makes them particularly suitable for real-world deployments, especially in mobile or resource-constrained multi-agent systems [26]. Despite the advantages offered by decentralized estimation, a significant challenge remains in ensuring the resilience of such systems against cyber-physical threats, particu- larly false data injection attacks (FDIA). In such situations, an attacker deliberately feeds false data into the system, often by taking control of certain agents or by tampering with Signals 2025, 6, x FOR PEER REVIEW 3 of 21 their communication channels. When it comes to tracking moving targets, these types of attacks can cause serious errors in estimating position and velocity, which are essential for maintaining accurate and reliable system behavior. As a result, improving the resili- ence of distributed estimation methods against false data injection has become a key focus in recent research. Numerous studies have explored ways to enhance both the communi- cation and estimation processes within distributed networks. A central approach in this area involves developing mechanisms that can detect and isolate suspicious data inputs quickly, preventing them from affecting the accuracy of the overall network estimation [27-37]. This paper builds upon our previous study in [38], which introduced a decentralized, consensus-based target tracking algorithm employing a nearly-constant-velocity (NCV) model and saturation-based filtering to mitigate impulsive measurement variations. While that earlier work established the baseline framework, it did not incorporate mech- anisms to identify or isolate malicious data. In the present study, we extend the frame- work by introducing a dynamic false data injection (FDI) detection and isolation mecha- nism that allows the algorithm not only to achieve accurate estimates under nominal con- ditions but also to remain resilient against adversarial attacks in multi-agent networks. This mechanism enables each agent to evaluate incoming measurements in real time, de- tect anomalies, and disregard suspicious data before it can corrupt the global estimate. To assess the effectiveness of this extension, we conduct a series of baseline and adversarial case studies, including localized, coordinated, and transient attack scenarios. The results demonstrate substantial improvements compared to [38]; for instance, under widespread coordinated aĴacks with 50% of agents compromised, the proposed method reduces the mean squared estimation error (MSEE) by about 65% relative to the baseline algorithm. Overall, this paper goes beyond our earlier contribution by providing a more comprehen- sive solution that improves both the accuracy and the security of distributed target track- ing, while keeping communication and computational costs low. The remainder of this article is organized as follows. Section 2 reviews the system modeling setup, including the dynamic model of the moving target and the derivation of the observation matrix as introduced in our earlier work, along with a brief overview of graph theory and network representation. Section 3 is divided into two subsections. The first subsection outlines the consensus-based estimation algorithm with saturation filter- ing, as proposed in our previous study, and the second subsection introduces the enhance- ments made through a false data injection detection and isolation mechanism, including the detection thresholds and agent isolation process. Section 4 presents the simulation setup, baseline and robustness case studies, and the associated results, illustrating the al- gorithm’s tracking accuracy and resilience under different network and attack scenarios, while also summarizing the key findings of the article. Section 5 discusses the resilience, limitations, and potential improvements of the proposed methods, and Section 6 provides the overall conclusion of this work. 2. System Modeling This section introduces the mathematical models used to describe the systems. The target's motion is described using a Nearly-Constant-Velocity (NCV) model, which is a kinematic-based model that incorporates process noise to account for the acceleration un- certainties. The measurement model is based on Time-of-Arrival (TOA) multilateration, enabling agents to estimate distances from the target and build local observation matrices. Finally, we used randomly generated networks to present the multi-agent system. These components together form the foundation for the distributed estimation and fault detec- tion methods which will be presented later on in this study. Signals 2025, 6, x FOR PEER REVIEW 4 of 21 2.1. Target Dynamics with NCV Model The motion of the moving target is modeled using a Nearly-Constant-Velocity frame- work. This approach employs a kinematic transition model and accounts for system un- certainty by incorporating acceleration terms as process noise inputs. Thus, the unmod- eled accelerations are represented as disturbances in the following state-space formula- tion: 푥 ( 푘+1 ) =퐴.푥 ( 푘 ) +퐵.푤 ( 푘 ) , (1) where 푥 ( 푘 ) and 푥 ( 푘+1 ) denote the state vector at the current and next time steps, re- spectively. 퐴 and 퐵 are the state transition and input matrices, and 푤 ( 푘 ) is the process noise vector. Using the NCV model with sampling interval 훥, the matrices 퐴 and 퐵 are given by: 퐴= ⎣ ⎢ ⎢ ⎢ ⎢ ⎡ 1 0 0훥0 0 0 1 0 0훥0 0 0 1 0 0훥 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 ⎦ ⎥ ⎥ ⎥ ⎥ ⎤ 푎푛푑 퐵= ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡ 훥 ଶ 2 00 0 훥 ଶ 2 0 0 0 훥 ଶ 2 훥0 0 0훥0 00훥 ⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ , (2) in the next section, we present the measurement model used by the agents to complete the mathematical representation of the system. 2.2. Measurement Model Using TOA Multilateration In the context of distributed state estimation, each agent independently collects meas- urement data which is linked to the global state of the system. The process of capturing these local observations can be mathematically described by the following measurement model: 푦 ௜ ( 푘 ) =퐻 ௜ .푥 ( 푘 ) +푣 ௜ ( 푘 ) , (3) Where푦 ௜ ( 푘 ) denotes the observation vector obtained by agent 푖 at time step 푘. 푣 ௜ ( 푘 ) , models the measurement noise in the data collected by the agents. 퐻 ௜ is the obser- vation matrix that shows how the system state is projected onto the measurements acces- sible to agent 푖. The accurate formulation of the observation matrix 퐻 ௜ is essential to ensure that each agent can accurately interpret and reconstruct the state of the system from its own measurements. In this study, we construct 퐻 ௜ using a multilateration approach based on the Time-of-Arrival (TOA) principle, as introduced in [39]. In this setup, the target contin- uously sends out a signal that moves through the environment at a known speed. When agents detect this signal, they log the time at which it arrives, enabling the calculation of their respective distances from the source. These distance estimates form the basis for both localization and the mathematical formulation of the measurement model by capturing the differences in signal arrival times at various agents. This derivation is visually illustrated in Figure 1, which shows a scenario where three agents work together to determine the location of a moving target. The figure demon- strates how sending and receiving the beacon signal helps the agents understand their positions relative to the target. Signals 2025, 6, x FOR PEER REVIEW 5 of 21 Figure 1. Illustration of the multilateration approach used to construct the observation matrix. For practical implementation, we assume that each agent exchanges data only with its set of immediate neighbors, reflecting a realistic scenario in localized and scalable multi-agent networks. This neighbor-limited information flow allows for the observation matrix to be defined in a linearized format as below: ൞ 푝 ௝,௜ ௫ =푝 ௝ ௫ −푝 ௜ ௫ 푝 ௝,௜ ௬ =푝 ௝ ௬ −푝 ௜ ௬ 푝 ௝,௜ ௭ =푝 ௝ ௭ −푝 ௜ ௭ , 푎푛푑 퐻 ௜ =቎ 푝 ௝ భ ,௜ ் 0 ଷ ் ⋮ 푝 ௝ ห ಿ ೔ ห ,௜ ் 0 ଷ ் ቏, (4) where 푝 ௡,௜ =(푝 ௡,௜ ௫ ,푝 ௡,௜ ௬ ,푝 ௡,௜ ௭ ) expresses the differences of positions between neighbor 푛 and agent 푖. The notation | 푁 ௜ | denotes the immediate neighboring set for the agent 푖. This setup ensures that each agent’s observation matrix is customized to its local sur- roundings, using only data that can be directly measured or shared within a distributed system. 2.3. Graph Theory and Network Representation In this section, we provide an overview of graph theory and describe the specific network configuration employed in this study. Multi-agent systems are modeled within the framework of graph theory as 풢 =풩, ℒ, 풲, where 풩 =1,...,푛 denotes the set of nodes, ℒ = ( 푖,푗 ) | 푖 →푗;푖,푗∈ 풩 represents the set of links (edges) among nodes, and 풲 =[퓌 ௜௝ ] corresponds to the weight matrix quantifying relevant attributes of the con- nections, such as distance. For the purposes of this work, all network topologies are generated using the Erdős– Rényi random graph model. In This model 퐺(푛,푝), a graph is formed from a fixed set of nodes 푛, where each possible edge between node pairs is included independently with probability 푝. This approach offers analytical simplicity, scalability, and serves as a stand- ard benchmark for examining the effects of random connectivity and network structure in multi-agent systems. In the following section, we provide a detailed demonstration of the estimation and fault detection algorithms. 3. Proposed Estimation and Fault Detection Algorithms In this section, we first introduce the Consensus-Based Estimation with Saturation Filtering, which was comprehensively explained in our previous work in [38]. We then enhance the performance of this method by incorporating a False Data Injection Detection mechanism. This addition aims to improve the robustness of our approach in adversarial environments and against cyber-physical threats, particularly False Data Injection Attacks on agents, which typically appear as sharp anomalies in the innovation during the Signals 2025, 6, x FOR PEER REVIEW 6 of 21 tracking of a moving target. For clarity and ease of reference, Table 1 summarizes all major variables, parameters, and symbols used throughout the manuscript. Table 1. Summary of key parameters and variables used throughout the manuscript. Symbol Description 푥 ( 푘 ) State vector of the target at time step 푘 푥ො ௜ ( 푘 ) Estimated state of the target by agent 푖 at time step 푘 |푁 ௜ | Set of immediate neighbors of agent 푖 퐴 State transition matrix 퐵 Input matrix 퐻 ௜ Observation matrix for agent 푖 푦 ௜ ( 푘 ) Measurement/Observation vector of agent 푖 at time step 푘 푣 ௜ ( 푘 ) Measurement noise of agent 푖 푤 ( 푘 ) Process noise 푔 ௜ ( 푘 ) Saturation gain agent 푖 at time step 푘 휉 Observation confidence parameter 퐿 Number of consensus iterations 휀 Consensus step size 훾 ௜ (푘) Innovation of agent 푖 at time step 푘 휑 ௜ ( 푘 ) Dynamic detection threshold for agent 푖 at time step 푘 푏 ௪ Upper bounds on the process noise 푏 ௩ Upper bounds on the measurement noise 푓 ( 푘 ) Global consensus mismatch at time step 푘 휌̅ ௞,௜ Estimation error bound of agent 푖 at time step 푘 훿 ௞,௜ Local consensus mismatch of agent 푖 at time step 푘 푒 ௜ Mean Squared Estimation Error (MSEE) of agent 푖 푒 ஺௩௚ Average Mean Squared Estimation Error of all agents 3.1. Consensus-Based Estimation with Saturation Filtering To provide context, we briefly summarize the distributed filtering framework in our earlier work. This framework operates in two steps: updating the state estimate using local observations (observation update) and then refining estimates by exchanging information between neighboring agents (estimation consensus). During the observation update, each agent integrates its own measurement 푦 ௜ ( 푘 ) through a saturation-based filtering for the purpose of limiting the effect of corrupted data. The update rule for agent 푖 at time step 푘 is computed as: 푥ො ௜ ( 푘 ) =퐴푥ො ௜ ( 푘−1 ) +푔 ௜ ( 푘 ) 퐻 ௜ ் ൫푦 ௜ ( 푘 ) −퐻 ௜ 퐴푥ො ௜ ( 푘−1 ) ൯, (5) where 푥ො ௜ ( 푘 ) represents the estimated state, 퐴 is the known state transition matrix ob- tained from equation (2), 퐻 ௜ is the observation matrix for agent 푖 obtained from equation (4), and 푔 ௜ ( 푘 ) is a saturation gain that addresses the innovation magnitude which is de- fined by the rule: 푔 ௜ ( 푘 ) =ቐ 1,푖푓 | 푦 ௜ ( 푘 ) −퐻 ௜ 퐴푥ො ௜ ( 푘−1 )| ≤휉 휉 | 푦 ௜ ( 푘 ) −퐻 ௜ 퐴푥ො ௜ ( 푘−1 )| ,표푡ℎ푒푟푤푖푠푒. (6) in this format, 휉>0 is the observation confidence parameter. This parameter restricts the influence of large innovations, which may be caused by measurement noise or adversarial interference, thereby enhancing the filter’s robustness against such disturbances. Following the observation update, agents engage in an iterative consensus process to improve estimation accuracy by exchanging information with neighboring agents. This Signals 2025, 6, x FOR PEER REVIEW 7 of 21 consensus step is performed multiple times within a single time iteration 푘. At consensus iteration 푙, agent 푖’s estimate is updated by combining its own estimate with those of its immediate neighbors 푁 ௜ : 푥ො ௜,௟ ( 푘 ) =푥ො ௜,௟ିଵ ( 푘 ) −휀 ∑ ቀ푥ො ௜,௟ିଵ ( 푘 ) −푥ො ௃,௟ିଵ ( 푘 ) ቁ ௃∈ே ೔ , (7) in this expression, 푥ො ௜,௟ ( 푘 ) is the updated estimate, 휀 is a small positive which controls the step size of the consensus update, and the summation adjusts the estimate toward the average of neighbors’ estimates. Performing multiple consensus steps promotes agree- ment across the network and improves accuracy compared to using local information alone. By integrating the saturation-based observation filter with the consensus-based esti- mation step, the proposed method achieves robustness against corrupted measurements while effectively fusing information across agents. This combined approach forms the foundation of our decentralized mobile target tracking algorithm based on the NCV model. The detailed steps of this method are presented in Algorithm (1). Algorithm 1. Decentralized mobile target tracking using the Consensus-Based Estimation Filter al- gorithm. Initialize variables = (푥ො ௜ ( 0 ) ,휉,퐿,휀) for 푘=1,2,... do for 푖=1,...,푁 do Observation Matrix Calculation: for 푗=1,...,|푁 ௜ | do Compute 푯 풊 “from Equation (4)” End Measurement Update with Saturation Filtering: 푔 ௜ ( 푘 ) =푚푖푛ቀ1, క | ௬ ೔ ( ௞ ) ିு ೔ ஺௫ ො ೔ ( ௞ିଵ ) | ቁ 푥ො ௜ ( 푘 ) =퐴푥ො ௜ ( 푘−1 ) +푔 ௜ ( 푘 ) 퐻 ௜ ் ൫푦 ௜ ( 푘 ) −퐻 ௜ 퐴푥ො ௜ ( 푘−1 ) ൯ Estimate Consensus: Let 푥ො ௜,଴ ( 푘 ) =푥ො ௜ ( 푘 ) for 푙=1,...,퐿 do 푥ො ௜,௟ ( 푘 ) =푥ො ௜,௟ିଵ ( 푘 ) −휀 ∑ ቀ푥ො ௜,௟ିଵ ( 푘 ) −푥ො ௃,௟ିଵ ( 푘 ) ቁ ௃∈ே ೔ End Let 푥ො ௜ ( 푘 ) =푥ො ௜,௅ ( 푘 ) End End 3.2. False Data Injection Detection and Isolation To improve the resilience of Algorithm (1) against cyber-physical threats, particu- larly false data injection attacks (FDIA), we introduce a detection and isolation mecha- nism. The core idea is to verify whether each agent’s local measurement is consistent with its predicted observation before it is incorporated into the distributed estimation process. If a measurement deviates significantly from what the model expects, it is considered po- tentially compromised and excluded from the update step. At each time step 푘, agent 푖 computes an innovation value, which represents the difference between the current measurement and the predicted output based on the agent’s previous state estimate. This innovation is given by: 훾 ௜ (푘)=푦 ௜ ( 푘 ) −퐻 ௜ 퐴푥ො ௜ ( 푘−1 ) , (8) Signals 2025, 6, x FOR PEER REVIEW 8 of 21 where 푦 ௜ ( 푘 ) is the measurement received by agent 푖, 퐻 ௜ is the observation matrix, 퐴 is the state transition matrix defined by the Nearly-Constant-velocity model, and 푥ො ௜ ( 푘−1 ) is the predicted state estimate from the previous step. To assess whether the innovation 훾 ௜ (푘) is acceptable, it is compared against a dy- namic detection threshold 휑 ௜ (푘). The threshold is defined as: 휑 ௜ ( 푘 ) =ห | 퐴 | ห.(휌̅ ௞ିଵ,௜ +푓 ( 푘−1 ) +푏 ௪ +푏 ௩ ), (9) where ห | 퐴 | ห is the norm of the state transition matrix 퐴. The term 휌̅ ௞ିଵ,௜ denotes the esti- mation error bound at the previous time step for agent 푖, and 푓 ( 푘−1 ) represents the consensus mismatch across the network, which measures how far apart the agents’ esti- mates are at time 푘−1. The constants 푏 ௪ and 푏 ௩ represent predefined upper bounds on the process and measurement noise, respectively. The consensus mismatch term 푓 ( 푘−1 ) is computed as the maximum deviation be- tween any two agents’ state estimates at the previous time step and is calculated as: 푓 ( 푘−1 ) =max ௜,௝∈ே ቚห푥ො ௜ ( 푘−1 ) −푥ො ௝ ( 푘−1 ) หቚ, (10) where 푁 denotes the set of all agents in the network. This term reflects the global estima- tion disagreement and captures the worst-case divergence among the agents. The estimation error bound 휌̅ ௞,௜ is updated recursively at each time step using the fol- lowing relation: 휌̅ ௞,௜ =ห | 퐴 | ห.휌̅ ௞ିଵ,௜ +휀.훿 ௞ିଵ,௜ +푏 ௪ , (11) here, 휀 is the consensus step size and 훿 ௞ିଵ,௜ is the local consensus disagreement for agent 푖, representing the maximum deviation between its own estimate and those of its imme- diate neighbors of 푁 ௜ . This local mismatch is calculated as: 훿 ௞ିଵ,௜ =max ௜,௝∈ே ೔ ቚห푥ො ௜ ( 푘−1 ) −푥ො ௝ ( 푘−1 ) หቚ, (12) if the magnitude of the innovation exceeds the threshold | 훾 ௜ ( 푘 )| >휑 ௜ ( 푘 ) ), the agent is flagged as compromised. In this case, the agent does not incorporate its current measure- ment into the state update. Instead, it uses the best prediction of the dynamic model: 푥ො ௜ (푘)=퐴푥ො ௜ ( 푘−1 ) , (13) this action effectively isolates the compromised measurement, preventing it from corrupt- ing the overall consensus process. If the innovation falls within the acceptable threshold, the agent continues with the normal measurement update and consensus-based fusion steps as outlined in the original Algorithm (1). By applying this detection and isolation process at every iteration, the esti- mation process becomes capable of automatically identifying and isolating faulty meas- urements. The complete procedure, incorporating this fault-resilient mechanism, is sum- marized in Algorithm (2). In the next section, we take a closer look at how Algorithms (1) and (2) perform in a range of testing scenarios. These fall into two main categories: tests under normal condi- tions with varying network setups, and tests where agents are subjected to false data in- jection attacks. This approach allows us to evaluate how effectively the algorithms handle disruptions, adapt to changing conditions, and detect faulty information. The analysis aims to assess estimation accuracy, convergence behavior, and overall resilience in a de- centralized framework. Signals 2025, 6, x FOR PEER REVIEW 9 of 21 Algorithm 2. Decentralized Mobile Target Tracking Using Consensus-Based Estimation Algorithm with Fault Detection. Initialize variables = (푥ො ௜ ( 0 ) ,휉,퐿,휀) for 푘=1,2,... do for 푖=1,...,푁 do Observation Matrix Calculation: for 푗=1,...,|푁 ௜ | do Compute 푯 풊 “from Equation (4)” End Innovation and Fault Detection: Compute 훾 ௜ ( 푘 ) and 휑 ௜ ( 푘 ) “from Equations (8) and (9)” if |훾 ௜ ( 푘 ) | > 휑 ௜ ( 푘 ) then 푥ො ௜ ( 푘 ) =퐴푥ො ௜ ( 푘−1 ) Else Measurement Update with Saturation Filtering: 푔 ௜ ( 푘 ) =푚푖푛ቀ1, క | ௬ ೔ ( ௞ ) ିு ೔ ஺௫ ො ೔ ( ௞ିଵ ) | ቁ 푥ො ௜ ( 푘 ) =퐴푥ො ௜ ( 푘−1 ) +푔 ௜ ( 푘 ) 퐻 ௜ ் ൫푦 ௜ ( 푘 ) −퐻 ௜ 퐴푥ො ௜ ( 푘−1 ) ൯ Estimate Consensus: Let 푥ො ௜,଴ ( 푘 ) =푥ො ௜ ( 푘 ) for 푙=1,...,퐿 푥ො ௜,௟ ( 푘 ) =푥ො ௜,௟ିଵ ( 푘 ) −휀 ∑ ቀ푥ො ௜,௟ିଵ ( 푘 ) −푥ො ௃,௟ିଵ ( 푘 ) ቁ ௃∈ே ೔ end if Let 푥ො ௜ ( 푘 ) =푥ො ௜,௅ ( 푘 ) End End 4. Case Studies and Results This section presents a comprehensive evaluation of the proposed decentralized, con- sensus-based target tracking algorithm under a range of operating conditions. The evalu- ation is divided into two main categories. The first category, referred to as the Baseline Case Studies, investigates the performance of Algorithm (1) in attack-free environments. This part examines the effects of key design parameters, such as graph connectivity, the number of consensus iterations, and the saturation filter settings, on the accuracy of target tracking and the convergence behavior of the distributed estimates. The second category, referred to as the Robustness Case Studies, evaluates the per- formance of both Algorithms (1) and (2) under various false data injection attacks. This part emphasizes the resilience and adaptability of the proposed framework when facing cyber-physical threats and highlights its effectiveness in mitigating the impact of faulty measurements. Performance is assessed using the Mean Squared Estimation Error (MSEE), defined in Equation (14), the Average Mean Squared Estimation Error given in Equation (15), and the innovation norm as defined in Equation (8): 푒 ௜ = ଵ ௞ ∑ ൫푥ො ௜ ( 푘 ) −푥 ௜ ( 푘 ) ൯ ଶ ௞ୀଵ,ଶ,... , (14) 푒 ஺௩௚ = 1 | 푁 | ෍푒 ௜ | ே | ௜ୀଵ , (15) all simulations were implemented and executed in the MATLAB environment. The simu- lation setup reflects realistic network conditions and includes multiple agent configura- tions. Finally, a Summary and Discussion of the Results is provided to show clear com- parative analysis. Signals 2025, 6, x FOR PEER REVIEW 10 of 21 4.1. Baseline Case Studies (1–3): Without FDIA This set of experiments evaluates the performance of Algorithm (1) under benign conditions, where no false data injection attacks occur. The purpose of these baseline stud- ies is to examine the influence of key design parameters, including the communication topology, the number of consensus iterations, and the configuration of the saturation filter on the estimation accuracy and convergence behavior. These tests provide a foundation for understanding the algorithm’s behavior under ideal conditions, before introducing adversarial scenarios. 4.1.1. Case Study 1 – Effect of Graph Connectivity This case study investigates how the density of the communication graph affects the tracking performance. Two random graphs with 20 agents are generated: one with a con- nection probability of 푝 = 0.7, representing a highly connected network (Fig. 2-b), and one with a low connectivity of 푝 = 0.3 (Fig. 2-a). The corresponding average MSEE re- sults for these configurations are shown in Fig. 2-c and Fig. 2-d. The results indicate that a higher level of connectivity improves estimation accuracy and accelerates convergence by allowing more information to be exchanged between agents. However, this enhanced performance requires greater communication bandwidth and computational resources. Therefore, designers must carefully balance the trade-off between estimation performance and the communication costs when choosing the net- work topology. (a) (b) (c) (d) Figure 2. Effect of network connectivity on tracking accuracy: (a) Graph with 푝 = 0.3; (b) Graph with 푝 = 0.7; (c) Average MSEE for the low-density network (shown in blue lines); (d) Average MSEE for the high-density network (shown in blue lines). Signals 2025, 6, x FOR PEER REVIEW 11 of 21 4.1.2. Case Study 2 – Number of Consensus Iterations This experiment examines the effect of the number of consensus iterations, denoted by 퐿, which controls how often each agent exchanges information with its immediate neighbors between successive time steps. Two different settings are tested: a lower com- munication rate (퐿 = 10) and a higher rate (퐿 = 100). The results for a random communication network in Fig. 3-a, indicate that using fewer iterations slows the convergence process and yields higher estimation errors (Fig. 3-b and Fig. 3-c), whereas increasing the number of iterations improves the convergence rate and reduces estimation errors (Fig. 3-d and Fig. 3-e). However, this performance gain requires increased communication and computation. Therefore, the number of iterations must balance accuracy with system constraints. (a) (b) (c) (d) (e) Signals 2025, 6, x FOR PEER REVIEW 12 of 21 Figure 3. Impact of the number of consensus iterations on estimation performance: (a) Communica- tion graph; (b–c) MSEE of individual agents and the average MSEE of all agents for 퐿 = 10, shown in colored and blue curves, respectively; (d–e) MSEE of individual agents and the average MSEE of all agents for 퐿 = 100, shown in colored and blue curves, respectively. 4.1.3. Case Study 3 – Saturation Filter Parameters In this final baseline case, the influence of the saturation filter’s configuration on tracking robustness and sensitivity is investigated. Two filter settings are tested: a permis- sive (soft) threshold with (휉=4) and a conservative (harsh) threshold with (휉=2). These settings control how much deviation is tolerated in received data before it is attenuated or discarded. The results are shown for a random communication network in Fig. 4-a. Under the soft threshold setting (휉=4), the system achieves higher estimation errors, as seen in the innovation norm and average MSEE plots (Fig. 4-b and Fig. 4-c), and it is more vulnerable to faulty measurements. By contrast, the harsh threshold setting (휉=2) yields smoother innovation norms and more stable behavior (Fig. 4-d and Fig. 4-e), at the cost of less sen- sitivity to measurement innovations. This case highlights the critical trade-off between filter sensitivity and robustness. (a) (b) (c) Signals 2025, 6, x FOR PEER REVIEW 13 of 21 (d) (e) Figure 4. Effect of saturation filter parameters on tracking performance: (a) Network graph; (b–c) Normalized innovation of individual agents and average MSEE of all agents for the soft filter (휉= 4), shown in colored and blue curves, respectively; (d–e) Normalized innovation of individual agents and average MSEE of all agents for the harsh filter (휉=2), shown in colored and blue curves, respectively. 4.2. Robustness Case Studies (4–6): With FDIA In this section, we extend our experiments to evaluate the performance of the Algo- rithm (2) alongside the baseline Algorithm (1), under adversarial conditions. Our focus is on cyber-physical threats posed by false data injection attacks from an external attacker. FDIA can cause substantial disturbances in measurements, creating anomalies or sharp spikes in the data that degrade the accuracy of the state estimates. In the first case, we introduce a localized false data injection attack of high magnitude to show how the detection and isolation mechanism protects the estimates and improves the reliability of the communication system under cyber-physical threats. The second case considers a widespread coordinated attack where half of the agents receive false data at varying magnitudes, allowing us to assess the algorithm’s resilience when a significant portion of the network is compromised. Finally, the third case investigates the algorithm’s behavior under transient attacks that last for only a short period of time. In each case, we evaluate and compare the performance of both algorithms to highlight the effectiveness of the proposed detection mechanism under different adversarial scenarios. 4.2.1. Case Study 4 – Localized False Data Injection In this case study, we assume that two randomly selected agents are subjected to a high-magnitude false data injection attack after a specified time step (푘>40). The objec- tive is to demonstrate how the detection and isolation mechanism can identify and isolate the compromised agents so that their corrupted data does not propagate through the net- work or degrade the overall consensus-based state estimation. The detection process on a randomly generated network is illustrated in Fig. 5-a. The innovation norm for each agent is compared against the dynamic threshold explained in Equation (9), and any value exceeding this threshold is flagged as compromised, as shown in Fig. 5-b. Figures 5-c and 5-d present the MSEE over time for the algorithms without and with FDIA detection, respectively. The results clearly show that the proposed detection and isolation method effectively mitigates the impact of the attacks, achieving lower er- rors than the baseline algorithm without detection. Figure 5-e compares the average MSEE across all agents for both algorithms, further confirming the significant performance improvements provided by the detection Signals 2025, 6, x FOR PEER REVIEW 14 of 21 mechanism under cyber-physical threats. Finally, Fig. 5-f illustrates the detection heat map, which correctly identifies agents 5 and 17 as under aĴack after 푘>40. (a) (b) (c) (d) (e) (f) Figure 5. Effect of localized false data injection on the proposed algorithms: (a) Random network topology; (b) Normalized innovation values for individual agents compared against the detection threshold 휑, where spikes above the threshold (red dashed line) indicate agents under false data injection attack (colored curves), while values below the threshold (gray curves) correspond to be- nign agents; (c) Mean squared estimation error (MSEE) over time for individual agents for the base- line algorithm without FDIA detection (shown in colored curves); (d) MSEE over time for individual agents for the proposed algorithm with FDIA detection enabled (shown in colored curves); (e) Com- parison of average MSEE of all agents for both algorithms, where the blue line denotes the FDIA- Signals 2025, 6, x FOR PEER REVIEW 15 of 21 enabled method and the red dashed line denotes the baseline; (f) Attack detection map (“heat map”), where red regions indicate malicious agents and green regions indicate benign agents. 4.2.2. Case Study 5 – Widespread Coordinated Attacks In this case study, half of the agents in the communication network (Fig. 6-a) are ran- domly selected and subjected to false data injection attacks of varying magnitudes (low, medium, and high). The objective is to evaluate the algorithms under severe conditions where a large number of agents are compromised. Particular attention is given to as- sessing the performance of the proposed detection algorithm when the attack magnitudes approach the detection threshold. Figure 6-b displays the innovation norm measurements across all agents, showing the presence of attacks with different magnitudes. The MSEE results for both algorithms are shown in Figures 6-c and 6-d, and a comparison of the average MSEE is provided in Fig. 6-e. In the case of low-magnitude attacks, which are close to the detection threshold, some of the compromised agents are successfully detected and isolated. In contrast, others remain undetected and continue to influence the estimates. Even under these challenging conditions, the algorithm with the detection mechanism outperforms the baseline version that lacks detection, achieving significantly lower estimation errors. Finally, Fig. 6-f pre- sents a heat map indicating all detected attacks over time, illustrating the effectiveness of the proposed algorithm in mitigating widespread coordinated false data injections. (a) (b) (c) (d) Signals 2025, 6, x FOR PEER REVIEW 16 of 21 (e) (f) Figure 6. Effect of widespread coordinated attacks on the proposed algorithms: (a) Random network topology; (b) Normalized innovation values for individual agents compared against the detection threshold 휑, where spikes above the threshold (red dashed line) indicate agents under false data injection attack (colored curves), while values below the threshold (gray curves) correspond to be- nign agents; (c) Mean squared estimation error (MSEE) over time for individual agents for the base- line algorithm without FDIA detection (shown in colored curves); (d) MSEE over time for individual agents for the proposed algorithm with FDIA detection enabled (shown in colored curves); (e) Com- parison of average MSEE of all agents for both algorithms, where the blue line denotes the FDIA- enabled method and the red dashed line denotes the baseline; (f) Attack detection map (“heat map”), where red regions indicate malicious agents and green regions indicate benign agents. 4.2.3. Case Study 6 – Transient Attacks In this final robustness case study, we evaluate the performance of the FDIA detec- tion algorithm under transient attacks. Unlike the previous case studies, where the at- tacker’s influence continuously throughout the entire process, this scenario considers a situation where a subset of agents in the network (Fig. 7-a) experiences short-duration false data injection attacks with medium to high magnitudes. The corresponding innova- tion norm is shown in Fig. 7-b. Figures 7-c and 7-d present the MSEE for the algorithms without and with FDIA de- tection, respectively. A comparison of the average MSEE for both algorithms is displayed in Fig. 8-e. The results demonstrate that the algorithm with detection enabled successfully prevents transient attacks from significantly affecting the estimation process and is able to quickly detect the attacked agents, as indicated in Fig. 7-f. (a) (b) Signals 2025, 6, x FOR PEER REVIEW 17 of 21 (c) (d) (e) (f) Figure 7. Effect of transient attacks on the proposed algorithms: (a) Random network topology; (b) Normalized innovation values for individual agents compared against the detection threshold 휑, where spikes above the threshold (red dashed line) indicate agents under false data injection attack (colored curves), while values below the threshold (gray curves) correspond to benign agents; (c) Mean squared estimation error (MSEE) over time for individual agents for the baseline algorithm without FDIA detection (shown in colored curves); (d) MSEE over time for individual agents for the proposed algorithm with FDIA detection enabled (shown in colored curves); (e) Comparison of av- erage MSEE of all agents for both algorithms, where the blue line denotes the FDIA-enabled method and the red dashed line denotes the baseline; (f) Attack detection map (“heat map”), where red regions indicate malicious agents and green regions indicate benign agents. 4.3. Summary and Discussion of Results This section summarizes the key findings from the baseline and robustness case stud- ies presented earlier. The baseline case studies assessed the performance of Algorithm (1) under attack-free conditions. In the first baseline case, increased network connectivity led to faster convergence and lower mean squared state estimation error, although this also increased communication overhead. In the second baseline case, the impact of the number of consensus iterations was examined. The results showed that increasing the number of iterations accelerated agreement among agents and reduced estimation errors, but re- quired greater communication bandwidth and power consumption. In the third baseline case, the influence of the saturation filter parameters was investigated. A more conserva- tive threshold achieved higher resilience to measurement variations but also reduced re- sponsiveness to legitimate changes. Signals 2025, 6, x FOR PEER REVIEW 18 of 21 The second set of case studies focused on the performance of Algorithm (1) alongside its enhanced version with FDIA detection presented in Algorithm (2), under cyber-phys- ical threats. The fourth case introduced localized high-magnitude false data injection at- tacks on two agents starting at time step 푘>40. Algorithm (2) successfully detected and isolated the compromised agents and maintained accurate tracking estimates. The fifth case involved widespread coordinated aĴacks on 50% of the agents with varying aĴack magnitudes. The results showed that the medium and high-magnitude attacks were reli- ably detected and mitigated, while a few low-magnitude attacks near the detection thresh- old occasionally evaded detection. Even under these challenging conditions, Algo- rithm (2) achieved substantially better average MSEE and tracking accuracy than Algo- rithm (1). Finally, the sixth case investigated the algorithms performance under transient attacks. The enhanced algorithm responded quickly to these brief anomalies and pre- served estimation performance with minimal delay. Table 2 concisely summarizes the different scenarios investigated in this study, in- cluding both baseline and adversarial conditions explored through simulation. This table highlights the key experimental setups, such as graph connectivity, consensus steps, and attack types, along with the primary outcomes and performance observations for each case. It is important to note that all scenarios are evaluated under the assumption that a majority of agents in the network are benign, which is a standard requirement for consen- sus-based resilience. Table 2. Summary of Baseline and Robustness Case Studies. Scenario FDA Used Attack Type Key Parameters Main Outcome 1. Graph Connectivity No None p=0.3 vs p=0.7 Dense graphs im- proved conver- gence and lowered MSEE 2. Consensus Itera- tions No None L=10 vs L=100 More iterations im- proved accuracy at increased commu- nication cost 3. Saturation Filter Parameters No None 휉=4 vs 휉=2 Conservative fil- ters enhanced ro- bustness but re- duced responsive- ness 4. Localized Fault In- jection Yes Targeted high-magnitude FDI on 2 agents Fault at k = 30 Algorithm (2) de- tected and sup- pressed the faulty data 5. Widespread AĴack Yes Randomized FDI (mixed magni- tude) on 50% of agents FDI levels varied (low/med/high) Algorithm (2) de- tected most at- tacks; low-magni- tude injections sometimes evaded detection but accu- racy was preserved 6. Transient AĴack Yes Random short-duration FDI Limited attack du- ration Algorithm (2) sup- pressed transient anomalies with minimal delay Signals 2025, 6, x FOR PEER REVIEW 19 of 21 5. Discussion While the proposed algorithm demonstrates strong resilience against various false data injection attack scenarios, several limitations should be acknowledged to provide a balanced assessment. First, the method assumes that a majority of agents in the network are benign, a common condition for consensus-based resilience, which may not hold in highly compromised environments. Second, although computational and communication overheads remain modest for small- to medium-scale networks, scaling to very large net- works may require additional optimization to prevent excessive processing and band- width usage. Third, the approach relies on a Nearly-Constant-Velocity (NCV) motion model, which can reduce estimation accuracy for targets exhibiting highly nonlinear or rapidly changing dynamics; replacing or augmenting the NCV model with a Nearly-Con- stant-Acceleration (NCA) model or adaptive/learning-based models could improve track- ing performance in such scenarios. Finally, the detection mechanism’s performance may degrade when attack magnitudes are close to the detection threshold, occasionally result- ing in missed detections, as observed in the widespread attack case study. Addressing these limitations could guide future research toward enhancing robustness, scalability, and adaptability of decentralized target tracking algorithms under adversarial conditions. 6. Conclusions In This paper, we developed and analyzed a resilient decentralized target tracking algorithm that operates effectively under both benign and adversarial conditions. By com- bining a saturation-based consensus filter with a false data injection detection and isola- tion mechanism, the proposed algorithm can achieve accurate estimates across multi- agent networks even in the presence of localized, coordinated, and transient false data injection attacks. The results from the case studies showed that increased graph connec- tivity and more consensus iterations improve estimation accuracy, while carefully tuned saturation filters balance estimation precision and sensitivity towards faulty measure- ments. Under cyber-physical attacks, the dynamic detection threshold successfully iden- tified and isolated compromised agents, allowing the system to maintain stable tracking performance and significantly reduce the mean squared estimation errors. Author Contributions: Conceptualization, A.A.G. and M.D.; methodology, A.A.G. and M.D.; soft- ware, A.A.G.; validation, A.A.G. and M.D.; formal analysis, A.A.G. and M.D.; investigation, A.A.G. and M.D.; resources, M.D.; data curation, A.A.G.; writing—original draft preparation, A.A.G.; writ- ing—review and editing, A.A.G. and M.D.; visualization, A.A.G.; supervision, M.D.; project admin- istration, M.D. All authors have read and agreed to the published version of the manuscript. Funding: This research received no external funding. Data Availability Statement: No new data were created or analyzed in this study. The MATLAB simulation code and relevant configuration files used to generate the results are available from the authors upon request. Conflicts of Interest: The authors declare no conflicts of interest. References 1. N. Geng, Q. Meng, D. Gong and P. Chung. "How Good are Distributed Allocation Algorithms for Solving Urban Search and Rescue Problems? A Comparative Study With Centralized Algorithms." IEEE Transactions on Automation Science and Engineering, 16 (2019): 478-485. hĴps://doi.org/10.1109/TASE.2018.2866395. 2. M. Rahmani and M. Abolhasani, "State estimation for stochastic time-varying multisensor systems with multiplicative noises: Centralized and decentralized data fusion," Asian Journal of Control, vol. 21, no. 4, p. 1547-1555, 2019. Signals 2025, 6, x FOR PEER REVIEW 20 of 21 3. A. Rajoriya, R. Budhiraja, and L. Hanzo, "Centralized and decentralized channel estimation in FDD multi-user massive MIMO systems," IEEE Transactions on Vehicular Technology, vol. 71, no. 7, p. 7325-7342, 2022. 4. S. Kang, A. Eryilmaz, and C. Joo, "Comparison of Decentralized and Centralized Update Paradigms for Distributed Remote Estimation," IEEE/ACM Transactions on Networking, vol. 32, no. 2, p. 1839-1853, 2023. 5. R. Olfati-Saber, "Distributed Kalman filter with embedded consensus filters," in Proceedings of the 44th IEEE Conference on Decision and Control, 2005: IEEE, p. 8179-8184. 6. R. Olfati-Saber and J. S. Shamma, "Consensus filters for sensor networks and distributed sensor fusion," in Proceedings of the 44th IEEE Conference on Decision and Control, 2005: IEEE, p. 6698-6703. 7. R. Olfati-Saber, "Distributed Kalman filtering for sensor networks," in 2007 46th IEEE conference on decision and control, 2007: IEEE, p. 5492-5498. 8. R. Carli, A. Chiuso, L. Schenato, and S. Zampieri, "Distributed Kalman filtering based on consensus strategies," IEEE Journal on Selected Areas in communications, vol. 26, no. 4, p. 622-633, 2008. 9. M. S. Mahmoud and H. M. Khalid, "Distributed Kalman filtering: a bibliographic review," IET Control Theory & Applications, vol. 7, no. 4, p. 483-501, 2013. 10. M. B. Khalkhali, A. Vahedian, and H. S. Yazdi, "Multi-target state estimation using interactive Kalman filter for multi-vehicle tracking," IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 3, p. 1131-1144, 2019. 11. C. Deng, X. -Z. Jin, Z. -G. Wu and W. -W. Che, "Data-Driven-Based Cooperative Resilient Learning Method for Nonlinear MASs Under DoS Attacks," in IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 9, p. 12107-12116, Sept. 2024, doi: 10.1109/TNNLS.2023.3252080. 12. S. Liu and X. Jiang, "Adaptive neural network decentralized control for nonlinear interconnected systems with time-varying constraints," International Journal of Robust and Nonlinear Control, vol. 32, no. 9, p. 5520-5533, 2022. 13. Z. Zhao, T. Wang, F. Peng and J. Yu, "Adaptive Fuzzy Resilient Decentralized Control for Nonlinear Large-Scale CPSs Under DoS Attacks," in IEEE Transactions on Fuzzy Systems, vol. 32, no. 10, p. 5899-5909, Oct. 2024, doi: 10.1109/TFUZZ.2024.3434726. 14. C. Deng, D. Yue, W. -W. Che and X. Xie, "Cooperative Fault-Tolerant Control for a Class of Nonlinear MASs by Resilient Learn- ing Approach," in IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 1, p. 670-679, Jan. 2024, doi: 10.1109/TNNLS.2022.3176392. 15. Y. Yang and L. Wang, "Byzantine-resilient decentralized network learning," Journal of the Korean Statistical Society, vol. 53, no. 2, p. 349-380, 2024. 16. S. S. Stanković, M. S. Stankovic, and D. M. Stipanovic, "Decentralized parameter estimation by consensus based stochastic ap- proximation," IEEE Transactions on Automatic Control, vol. 56, no. 3, p. 531-543, 2010. 17. A. Petitti, D. Di Paola, A. Rizzo, and G. Cicirelli, "Consensus-based distributed estimation for target tracking in heterogeneous sensor networks," in 2011 50th IEEE Conference on Decision and Control and European Control Conference, 2011: IEEE, p. 6648-6653. 18. N. Ilić, M. S. Stanković, and S. S. Stanković, "Adaptive consensus-based distributed target tracking in sensor networks with limited sensing range," IEEE Transactions on Control Systems Technology, vol. 22, no. 2, p. 778-785, 2013. 19. Z. Zhou, H. Fang, and Y. Hong, "Distributed estimation for moving target based on state-consensus strategy," IEEE Transactions on Automatic Control, vol. 58, no. 8, p. 2096-2101, 2013. 20. A. T. Kamal, J. H. Bappy, J. A. Farrell, and A. K. Roy-Chowdhury, "Distributed multi-target tracking and data association in vision networks," IEEE transactions on paĴern analysis and machine intelligence, vol. 38, no. 7, p. 1397-1410, 2015. 21. J. Yan, Z. Xu, Y. Wan, C. Chen, and X. Luo, "Consensus estimation-based target localization in underwater acoustic sensor networks," International Journal of Robust and Nonlinear Control, vol. 27, no. 9, p. 1607-1627, 2017. 22. Z. Ding et al., "Adaptive interacting multiple model algorithm based on information-weighted consensus for maneuvering target tracking," Sensors, vol. 18, no. 7, p. 2012, 2018. 23. M. Doostmohammadian, U. A. Khan, M. Pirani, and T. Charalambous, "Consensus-based distributed estimation in the presence of heterogeneous, time-invariant delays," IEEE Control Systems LeĴers, vol. 6, p. 1598-1603, 2021. 24. J. K. Verma, J. K. Chhabra, and V. Ranga, "Track consensus-based labeled multi-target tracking in mobile distributed sensor network," IEEE Transactions on Mobile Computing, vol. 23, no. 6, p. 7351-7362, 2023. 25. Y. Sun, X. Yang, P. Shi, and H. Su, "Consensus tracking of switched heterogeneous nonlinear systems with uncertain target," IEEE Transactions on Circuits and Systems I: Regular Papers, 2024. 26. M. Doostmohammadian et al., "Survey of distributed algorithms for resource allocation over multi-agent systems," Annual Reviews in Control, vol. 59, p. 100983, 2025. Signals 2025, 6, x FOR PEER REVIEW 21 of 21 27. A. Teixeira, D. Pérez, H. Sandberg, and K. H. Johansson, "Attack models and scenarios for networked control systems," in Proceedings of the 1st international conference on High Confidence Networked Systems, 2012, p. 55-64. 28. J. Giraldo et al., "A survey of physics-based attack detection in cyber-physical systems," ACM Computing Surveys (CSUR), vol. 51, no. 4, p. 1-36, 2018. 29. X. He, X. Ren, H. Sandberg, and K. H. Johansson, "Design of secure filters under attacked measurements: A saturation method," ed, 2020. 30. Q. Jiang, H. Chen, L. Xie, and K. Wang, "Learning-based cooperative false data injection attack and its mitigation techniques in consensus-based distributed estimation," IEEE Access, vol. 8, p. 166852-166869, 2020. 31. H. Liu, B. Niu, and Y. Li, "False-data-injection attacks on remote distributed consensus estimation," IEEE Transactions on Cy- bernetics, vol. 52, no. 1, p. 433-443, 2020. 32. X. He, X. Ren, H. Sandberg, and K. H. Johansson, "How to secure distributed filters under sensor attacks," IEEE Transactions on Automatic Control, vol. 67, no. 6, p. 2843-2856, 2021. 33. Y. Shui, Y. Wang, Y. Li, Y. Shan, N. Cui, and B. Pang, "Consensus-based distributed target tracking with false data injection aĴacks over radar network," Applied Sciences, vol. 11, no. 10, p. 4564, 2021. 34. X. He, X. Ren, H. Sandberg, and K. H. Johansson, "Secured Filters Based on Saturated Innovations," in Security and Resilience of Control Systems: Theory and Applications: Springer, 2022, p. 3-29. 35. X. Luo, J. He, X. Wang, Y. Zhang, and X. Guan, "Resilient defense of false data injection attacks in smart grids via virtual hidden networks," IEEE Internet of Things Journal, vol. 10, no. 7, p. 6474-6490, 2022. 36. Y. Yang, X. Gao, S. Song, and Z. Li, "Distributed consensus-based filtering against false data injection attacks," IEEE Transactions on Network Science and Engineering, 2024. 37. Z. Cheng, H. Ren, J. Qin, and R. Lu, "Distributed Secure Consensus Estimation for Power Systems Against False Data Injection AĴacks," International Journal of Robust and Nonlinear Control, 2025. 38. A. A. Ghods and M. Doostmohammadian, "Decentralized Mobile Target Tracking Using Consensus-Based Estimation with Nearly-Constant-Velocity Modeling," 2024 12th RSI International Conference on Robotics and Mechatronics (ICRoM), Tehran, Iran, Islamic Republic of, 2024, p. 70-75, doi: 10.1109/ICRoM64545.2024.10903532. 39. M. Doostmohammadian, A. Taghieh and H. Zarrabi, "Distributed Estimation Approach for Tracking a Mobile Target via For- mation of UAVs," in IEEE Transactions on Automation Science and Engineering, vol. 19, no. 4, p. 3765-3776, Oct. 2022, doi: 10.1109/TASE.2021.3135834. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual au- thor(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Instant research discovery