Underdetermined Time-Frequency Aliasing Signal Reconstruction Based on Modified Subspace Projection

Addressing the limitations of traditional underdetermined blind separation methods for time-frequency aliasing signal reconstruction, which often require high signal sparsity and yield low reconstruction accuracy, this paper introduces a modified subspace projection approach. This method involves calculating the actual number of source signals at each time-frequency point and their corresponding mixed column vector matrix. As long as the number of dominant source signals is fewer than the number of observed signals at any given time-frequency point, the time-frequency values of the reconstructed signals can be determined using the modified subspace projection measure. Simulation results demonstrate that the proposed method achieves higher reconstruction accuracy compared to several classical approaches in blind separation scenarios characterized by severe underdetermined time-frequency aliasing.