Restoration of multi-channel signal loss using autoencoder with recursive input strategy

Multi-channel sensor data often suffer from missing or corrupted values due to sensor failures, communication disruptions, or environmental interference. These issues severely limit the accuracy of intelligent systems relying on sensor data integration. Existing data restoration techniques often fail to capture complex correlations among sensor channels, especially when data losses occur randomly and continuously. To overcome these limitations, we propose an autoencoder-based data recovery algorithm that recursively feeds reconstructed outputs back into the model to progressively refine estimates. A dynamic termination criterion monitors reconstruction improvements, automatically stopping iterations when further refinements become negligible. This recursive input strategy significantly enhances restoration accuracy and computational efficiency compared to conventional single-step methods. Experiments on multivariate sensor datasets show that the proposed method significantly outperforms the one-time autoencoder restoration method and maintains robust performance across diverse datasets and missing data scenarios. This approach provides a scalable and adaptable solution to ensure data integrity in complex sensor networks, enabling improved reliability and operational efficiency in industrial and technological applications.