A Noise Separation Method Based on Spatial Sound Suppression and Equivalent Field Transformation

The traditional equivalent source method suffers from ill-posed inverse problems in sound field separation, which lead to unstable solutions and high sensitivity to measurement noise and array errors. Moreover, the separation accuracy strongly depends on prior assumptions regarding the number and spatial distribution of equivalent sources. In complex multi-source environments or under strong background noise, this method is prone to generating spurious sources and severe distortions. To address these limitations, this paper proposes an equivalent field transformation theory based on a virtual sound-insulating space. By exploiting the spatial distinguishability between interfering sources and target sound sources, a physically constrained regularized optimization model is constructed. The proposed method establishes the acoustic pressure transfer relationships among the target space, observation space, and virtual sound-insulating space through analytical Green’s functions. The sound field is mathematically represented by a distributed equivalent source strength model, and the sound pressure amplitude within the sound-insulating region is constrained to zero to solve for the weighted coefficients of the observation array. Based on the acoustic reciprocity principle, these coefficients enable effective equivalent blocking of interfering sources and accurate extraction of the target sound field. Simulation results demonstrate that, under sparse sampling conditions, the average separation accuracy improvement reaches 69.97%, whereas under strong noise interference, the average separation accuracy improvement reaches 89.91%. In the 2.7–6 kHz frequency band, the reconstruction error of this method remains consistently below 3%. These results verify the correctness of the proposed method, its robustness against noise, and its strong potential for practical engineering applications.