JASMINE · Jamming and Spoofing Resilient Deep Learning Based Software-Defined Multi-Antenna Multi-GNSS Receiver (JASMINE)

Jamming and Spoofing Resilient Deep Learning Based Software-Defined Multi-Antenna Multi-GNSS Receiver (JASMINE)

JASMINE tackles a critical PNT challenge: ensuring GNSS integrity and resilience in complex environments facing deliberate threats such as jamming and spoofing, requiring robust, intelligent, and scalable protection. The project’s overarching goal is to advance GNSS integrity monitoring and interference resilience by leveraging state-of-the-art machine learning, reinforcement learning, and high-performance signal processing techniques. By combining physics-based models with data-driven intelligence, JASMINE aims to deliver next-generation GNSS receivers and analytical frameworks capable of adapting to dynamic threats, predicting orbital states beyond nominal validity, and ensuring trust in navigation solutions. The pathway to impact is structured around four core research pillars, each contributing to scientific progress and practical applicability: 1. GPU-Based GNSS SDR Development To enable real-time, high-throughput GNSS signal processing, JASMINE implemented a scalable Software-Defined Radio (SDR) architecture optimized for NVIDIA RTX A6000 GPUs. This design achieves reproducible real-time performance (~1.18× at 2 MHz across 8 channels) and supports multi-GNSS, multi-frequency, and multi-antenna configurations. Key innovations include in-loop decimation and channel-vectorized correlators, which significantly enhance computational efficiency. Impact: This work demonstrates practical hardware-software co-design for GNSS SDRs, providing a blueprint for future GPU-accelerated navigation systems and contributing to the scientific community through open, reproducible performance benchmarks. 2.GNSS Orbit Prediction Using Deep Learning The project developed long-horizon orbit forecasting models using advanced neural architectures such as N-HiTS and BiLSTM, extending broadcast ephemerides beyond their nominal validity. By integrating physics-based constraints (e.g., two-body + J2 perturbation models) with residual learning, these models achieve error reductions of 95–99%, significantly improving prediction accuracy and reliability. Impact: This approach enhances GNSS orbit integrity, enabling better continuity of service during ephemeris outages and supporting mission-critical applications that require extended prediction horizons. 3. GNSS Jamming and Spoofing Detection via Deep Reinforcement Learning JASMINE reframed interference detection as a reinforcement learning task, enabling autonomous adaptation to dynamic threat environments. DRL agents such as DQN, PPO, and QR-DQN were trained to learn interference patterns and classify jamming/spoofing events with 98% accuracy. Impact: This represents a paradigm shift in GNSS security, moving from static detection algorithms to intelligent, learning-based systems that can evolve with emerging threats, strengthening the resilience of GNSS infrastructure. 4. Receiver Autonomous Integrity Monitoring (RAIM) with DRL A novel RAIM framework based on Deep Reinforcement Learning (PPO) was developed to enhance fault detection and integrity assurance in multi-GNSS receivers. By leveraging constellation-aware features—such as pseudorange residuals, C/N₀, elevation, azimuth, LOS vectors, and satellite geometry, the system achieves ~91% one-shot decision accuracy, even under degraded conditions like ionospheric disturbances and satellite faults. Integrated protection-level computation and fault detection & exclusion (FDE) make this solution suitable for safety-critical applications. Impact: This work advances integrity monitoring beyond traditional RAIM, introducing adaptive, learning-based mechanisms that ensure trustworthy navigation solutions under challenging scenarios.

Текст от CORDIS, на английски · Данни: CORDIS, © Европейски съюз