Sparse Representation Methods for Effective Data Recovery in Wireless Sensor Networks: A Review

Missing and corrupted data remain persistent challenges in Wireless Sensor Networks (WSNs), particularly under energy constraints, harsh operating environments, and structured loss patterns. This review systematically examines sparse-representation based data recovery methods by categorizing them into two methodological classes: low-order sparsity and high-order sparsity frameworks. The analysis emphasizes how these approaches exploit spatial, temporal, and multi-attribute correlations, and critically discusses the implications of their modeling assumptions, rank selection strategies, and computational complexity. While compressive sensing and matrix completion offer favorable computational efficiency under predominantly random missingness, tensor-based methods including CP, Tucker, Tensor Train, Tensor Ring, and t-SVD often deliver improved reconstruction accuracy for multi-attribute sensing data and structured missing patterns, with increased computational and memory requirements. By clarifying recoverability conditions and performance trade-offs across these frameworks, this review establishes a unified methodological foundation for principled, comparable, and deployment-oriented advances in WSNs data recovery.