LightTR is a lightweight framework for federated trajectory recovery, designed to address the challenges of recovering high-sampled trajectories from low-sampled ones in decentralized environments. With the proliferation of GPS-equipped edge devices, vast amounts of trajectory data are generated, but low sampling rates can degrade the effectiveness of urban applications. LightTR aims to recover missing points in incomplete trajectories without relying on road network information, enhancing data usability. Traditional methods assume centralized data, which is impractical for decentralized settings and poses privacy risks. LightTR employs a client-server architecture to maintain data decentralization and privacy, featuring a lightweight local trajectory embedding module and a meta-knowledge enhanced local-global training scheme to reduce communication costs and improve efficiency. The lightweight ST-operator in LightTR reduces computational complexity while preserving feature extraction capabilities. The meta-knowledge module uses knowledge distillation to guide local model training, accelerating convergence and improving accuracy. Extensive experiments on real datasets demonstrate LightTR's effectiveness and efficiency, outperforming baselines in accuracy and reducing communication costs. LightTR is scalable, efficient, and suitable for decentralized trajectory recovery in real-world applications.LightTR is a lightweight framework for federated trajectory recovery, designed to address the challenges of recovering high-sampled trajectories from low-sampled ones in decentralized environments. With the proliferation of GPS-equipped edge devices, vast amounts of trajectory data are generated, but low sampling rates can degrade the effectiveness of urban applications. LightTR aims to recover missing points in incomplete trajectories without relying on road network information, enhancing data usability. Traditional methods assume centralized data, which is impractical for decentralized settings and poses privacy risks. LightTR employs a client-server architecture to maintain data decentralization and privacy, featuring a lightweight local trajectory embedding module and a meta-knowledge enhanced local-global training scheme to reduce communication costs and improve efficiency. The lightweight ST-operator in LightTR reduces computational complexity while preserving feature extraction capabilities. The meta-knowledge module uses knowledge distillation to guide local model training, accelerating convergence and improving accuracy. Extensive experiments on real datasets demonstrate LightTR's effectiveness and efficiency, outperforming baselines in accuracy and reducing communication costs. LightTR is scalable, efficient, and suitable for decentralized trajectory recovery in real-world applications.