Decentralized GNSS Network Architecture

• Decentralized GNSS networks are distributed systems that fuse consensus, optimization, and state estimation to eliminate single-point failures and enhance overall resilience.
• They employ peer-to-peer state sharing, dynamic graph modeling, and protocols like Raft and gradient tracking for efficient, low-latency and fault-tolerant operations.
• Experimental studies demonstrate these networks achieve high accuracy, e.g., orbit RMS errors of 0.06–0.12 m, while remaining robust against adversarial conditions and communication disruptions.

A decentralized GNSS (Global Navigation Satellite System) network is a networked positioning, navigation, and timing (PNT) architecture in which state estimation, consensus, corrections, and service delivery are performed collaboratively by a distributed collection of satellites, ground stations, or peer agents, eschewing dependence on centralized processing or fixed reference nodes. Such networks are motivated by the limitations of classical GNSS architectures, including single-point-of-failure risks, vulnerabilities to jamming/spoofing, limited scalability for corrections, and dependence on global infrastructure. Decentralized GNSS architectures incorporate protocols from distributed optimization, graph theory, consensus (e.g., Raft, momentum gradient tracking), and factor-graph inference, achieving robustness to partial outages, communication disruptions, and adversarial manipulation, and enabling new regimes of resilience, autonomy, and scalability across terrestrial, airborne, and spaceborne domains. Implementations range from real-world UAV swarms using consensus-based state fusion to large-scale networks of LEO satellites and continental networks of ground stations, each leveraging tailored protocols for efficient, low-latency, and fault-tolerant operation (Dev et al., 1 Aug 2025, Liu et al., 24 Dec 2025, Zheng et al., 21 Dec 2025).

1. Core Principles and Architectural Models

Decentralized GNSS architectures supplant the traditional hub-and-spoke paradigm with collaborative estimation and control, where network nodes (satellites, receivers, drones, ground stations) perform local inference and state propagation, followed by communication of succinct state summaries to selected neighbors.

Key operating regimes include:

• Peer-to-Peer State Sharing: Each agent maintains a local estimate of position/clock and exchanges updates with immediate neighbors. For instance, in LEO constellations, each satellite maintains a private state (orbit, clock, ambiguities, biases) and exchanges only low-dimensional state vectors, not raw measurements, with its physical neighbors, thus keeping communication bandwidth independent of constellation size (Liu et al., 24 Dec 2025).
• Consensus on Shared Products: Nodes jointly estimate global or regional corrections (e.g., satellite clocks, atmospheric delays) via distributed consensus/optimization such as gradient tracking, enabling users to retrieve these corrections from any reachable peer or by local fusion (Zheng et al., 21 Dec 2025).
• Dynamic Graph Modeling: The network is modeled as a time-varying graph $\mathcal G(t) = (V, E(t))$; communication strategies, mixing matrices, and inference protocols exploit the evolving connectivity and ensure convergence under partial synchrony and intermittent outages (Liu et al., 24 Dec 2025, Zheng et al., 21 Dec 2025).
• Generalization to Mobile Swarms: Mobile platforms (UAVs, ground robots) participate as first-class network nodes, fusing GNSS/INS/ranging data through consensus protocols (e.g., Raft), allowing the network to dynamically extend or reconfigure as agents enter or exit (Dev et al., 1 Aug 2025).

The following table summarizes representative architectural variants:

Regime	Key Node Role	Communication Pattern
LEO satellite constellation	GNSS receiver + peer	Inter-satellite, sparse
Ground station network	Reference + consensus	Regional mesh
Mobile peer swarm (UAV, WSN)	Agent + consensus	Fully/partially meshed

2. Distributed Estimation and Consensus Protocols

Distributed estimation in decentralized GNSS involves decomposing the inference problem into local and shared components, followed by iterative optimization and consensus over the shared variables.

Representative methodologies:

• Consensus Algorithms: In SwarnRaft (Dev et al., 1 Aug 2025), the Raft protocol is adapted so nodes agree on position/state updates per time step. States (position, heading, motion) are fused from all nodes, then broadcast and committed atomically after majority agreement. Extended to large-scale networks, consensus can propagate pseudorange corrections or state updates robustly, with known convergence and termination criteria.
• Gradient Tracking and Diffusion: In global-scale ground networks and spaceborne constellations, consensus is achieved by graph-aware gradient tracking and diffusion adaptation. Each node computes local gradients (w.r.t global corrections), updates local and tracker variables, and mixes these with neighbors via topology-aware mixing matrices (Liu et al., 24 Dec 2025, Zheng et al., 21 Dec 2025). Offline-learned mixing schedules accelerate convergence despite time-varying graph connectivity.
• Factor-Graph Message Passing: Cooperative 3D positioning in GNSS-denied environments is performed by distributed sum-product over a spatio-temporal factor graph, with message updates using scaled unscented transform numerical approximations for nonlinearity. Marginal distributions are iteratively improved via belief propagation, with local "pseudo-anchors" arising dynamically as nodes' state estimates converge (Cao et al., 2022).

Crucial elements include majority thresholds for agreement, mixing matrix regularity (block contraction), resilience to temporary link failures, and flexibility to support both synchronous and asynchronous operation (Dev et al., 1 Aug 2025, Zheng et al., 21 Dec 2025).

3. System Components and Measurement Fusion

Decentralized GNSS networks are inherently heterogeneous, integrating numerous sensor and communications modalities:

• GNSS Receivers: Provide code and carrier-phase measurements; may be on LEO spacecraft, UAVs, ground stations, or crowd-sensing devices. Standalone on-node solutions are augmented by network-wide state sharing, improving performance in degraded conditions (Liu et al., 24 Dec 2025, Dev et al., 1 Aug 2025).
• INS/IMU Subsystems: Facilitate dead-reckoning and motion increment estimation, allowing nodes to propagate state even during GNSS outages. These are combined with peer-consensus for accurate, robust localization under loss or malicious spoofing (Dev et al., 1 Aug 2025).
• Ranging Sensors (UWB, RSSI, TOA, AoA): Support direct inter-node distance or angle measurements, crucial for cooperation in GNSS-denied or urban environments (Cao et al., 2022, Dev et al., 1 Aug 2025).
• Communication Backbones: Rely on WiFi mesh, radio relays, fiber-optic rings (for terrestrial PNT), or secure spaceborne links; synchronization is supported by protocols such as White Rabbit/IEEE 1588 PTP for sub-nanosecond network timing (Koelemeij et al., 2023).
• Onboard Compute: ARM-class CPUs, FPGAs, or custom ASICs implement sensor fusion, local inference, and distributed update protocols (gradient tracking, consensus, etc.).

Fusion strategies typically employ probabilistic models—e.g., INS-propagated priors revised using peer-fused observations to reconstruct missing measurement epochs under GNSS loss. In participatory sensing, crowd-sourced AGC and C/N₀ outputs enable distributed detection/localization of jammers (Olsson et al., 2022).

4. Performance Evaluation, Scalability, and Trade-offs

Empirical and simulated results consistently demonstrate that decentralized GNSS networks can recover or exceed centralized performance at a fraction of the communication and latency cost under various operational regimes:

• Positioning Accuracy: For example, momentum GT over LEO networks achieves orbit RMS errors of 0.06–0.12 m and clock bias errors down to 0.11 ns, matching centralized weighted least-squares benchmarks (Liu et al., 24 Dec 2025). In UAV swarms, SwarnRaft maintains mean absolute error (MAE) under 1 m even with N/2 nodes spoofed, and recovery times of ≈50 ms on WiFi (Dev et al., 1 Aug 2025). Ground station diffusion matches centralized PPP (precise point positioning) accuracy to within 1E-8 m (Zheng et al., 21 Dec 2025).
• Consensus Latency and Overhead: Raft-based consensus cycles terminate in ≤Tₑ+Tₕ+3δ, independent of the swarm size in fully-connected topologies; GT and diffusion algorithms exploit multi-round neighbor mixing to accelerate convergence relative to standard Metropolis or Markov chain protocols (Dev et al., 1 Aug 2025, Liu et al., 24 Dec 2025, Zheng et al., 21 Dec 2025).
• Communication Cost: Communication per node is a function of neighbor count and fixed-length state vectors (not raw data), and remains scalable (e.g., 2Km scalars per consensus round) across hundreds or thousands of nodes (Liu et al., 24 Dec 2025).
• Resilience and Fault Tolerance: Networks tolerate up to f Byzantine nodes, loss of up to 5% of links per round, and intermittent outages. Protocols such as majority voting or median-of-honest-peers guarantee correctness under standard fault models (Dev et al., 1 Aug 2025). Block contraction/batching strategies enable operation under link churn or asynchronous schedules (Zheng et al., 21 Dec 2025).

A further trade-off emerges between accuracy and communication: full mesh connectivity optimizes consensus rate and dilution of precision but increases per-node overhead; region-based or hierarchical architectures can scale further with slight compromises in latency or convergence (Dev et al., 1 Aug 2025).

5. Security, Robustness, and Adversarial Mitigation

Decentralized GNSS networks integrate protocol-level safeguards against attacks such as GNSS spoofing, jamming, node compromise, and information poisoning:

• Byzantine Resilience: SwarnRaft uses residual-based voting and majority aggregation protocols, tolerating f out of N compromised GNSS sensors (N≥2f+1 required for safety), and recovers states by median of honest peer fused estimates (Dev et al., 1 Aug 2025).
• Participatory Sensing: Crowdsourced anomaly detection/localization for jamming leverages joint maximum-likelihood over AGC, C/N₀ measurements with robust estimation against outliers and NLOS links (Olsson et al., 2022). The decentralized estimator achieves <5 m error with only 5 receivers in favorable scenarios, and maintains robustness as noise increases.
• Graph-Aware Diffusion: Protocols leverage offline-learned mixing schedules to defensively accelerate consensus despite sparse or unreliable graphs; further extensions include Byzantine-robust mixing, quantization-resilient updates, and secure broadcast of consensus products for end users (Zheng et al., 21 Dec 2025).
• Synchronization and Timing Faults: Integration of fiber-optic timing networks, hardware timestamping, and real-time calibration suppresses clock drift and supports sub-nanosecond alignment, which is essential for high-precision user timing and resilience against time-shift attacks (Koelemeij et al., 2023).

6. Extensions, Generalizations, and Open Challenges

The decentralized GNSS model generalizes to multiple domains and is an active area for systems research:

• Hybrid Air-Ground-Space Architectures: Integration of LEO constellations, terrestrial fiber/radio networks, and mobile ground/airborne agents enables multi-domain coverage and resilience to partitioning or region-specific denial-of-service (Liu et al., 24 Dec 2025, Koelemeij et al., 2023).
• PPP and RTK Service Delivery: Consensus-based correction products can be disseminated as real-time corrections to user receivers, supporting PPP and PPP-RTK without reliance on centralized control (Zheng et al., 21 Dec 2025).
• Asynchrony and Scalability: Protocol extensions accommodate asynchrony, dynamic topology, and regionally variable trust levels through asynchronous consensus (e.g., Multi-Paxos, EPaxos), weighted voting, and dynamic cluster/partitioning (Dev et al., 1 Aug 2025).
• Cooperative Positioning in Sensor Networks: Distributed factor-graph methods realize fully-GNSS-like positioning in GNSS-denied or urban environments, leveraging TOA, RSS, or AOA ranging to mimetically replace satellite service with a mesh of peer anchors and adaptive message-passing (Cao et al., 2022).
• Autonomous Relativistic Reference Frames: Satelite constellations may achieve a primary self-referenced inertial frame using inter-satellite proper-time exchanges and emission coordinates, fully eschewing terrestrial reference frames and achieving millimetre-inertial stability (Kostić et al., 2014).
• Real-World and Large-Scale Experimentation: Research open problems include full streaming operation, integration with 5G MEC, edge-compute offloading, adaptive noise/multipath modeling, and empirical performance validation under dense urban, underground, and adversarial conditions (Zheng et al., 21 Dec 2025, Dev et al., 1 Aug 2025).

7. Representative Implementations and Experimental Results

Selected deployments and simulation results illustrate performance regimes:

Architecture	Nodes	Key Protocol	Accuracy	Consensus Latency	Scalability Features	Reference
SwarnRaft (UAV Swarm)	5–15 UAVs	Raft consensus	<1 m (MAE)	≈30 ms	Fault tolerance to N/2 Byzantine nodes	(Dev et al., 1 Aug 2025)
LEO GNSS Network	500 sats	Gradient tracking	0.06–0.12 m RMS	N/A	Communication scales with neighbor count, not N	(Liu et al., 24 Dec 2025)
Graph-aware GNSS PPP	100+ GSN	GT diffusion	cm-level PPP	Faster (per iter)	Offline-learned schedules, handles sparse graphs	(Zheng et al., 21 Dec 2025)
Participatory Sensing	5–10 nodes	Joint ML estimation	<5 m (jammer)	N/A	Linear per-node overhead	(Olsson et al., 2022)
Terrestrial PNT	N/A	Fiber + OFDM downlink	2–10 cm static	N/A	High robustness via hardware sync	(Koelemeij et al., 2023)
Cooperative 3D WSN	120 nodes	Factor-graph + SUT	1.8 m RMSE	<20 iter (EAU)	Iterative anchor-upgrading, lightweight comm	(Cao et al., 2022)

Quantitative results confirm robustness to node failure, adversarial activity, and scalability in node count and graph complexity.

---

Decentralized GNSS networks unify advanced consensus, distributed optimization, topology-aware inference, and practical sensing to realize scalable, resilient, and accurate network-based PNT across terrestrial, airborne, and spaceborne deployments (Dev et al., 1 Aug 2025, Liu et al., 24 Dec 2025, Zheng et al., 21 Dec 2025, Koelemeij et al., 2023, Kostić et al., 2014, Cao et al., 2022, Olsson et al., 2022).