This study presents an innovative mathematical framework operating entirely in the time–space (t-x) domain for pre-stack seismic noise attenuation and wavefield separation, establishing a modified Ladjadj method tailored for automated, high-fidelity wavelet extraction. Driven by a non-destructive matrix optimization algorithm pairing spatial trace windows with an empirical comparison rate consensus, the technique successfully addresses severe multi-modal noise fields and complex wavefield interferences. Methodological validation on highly contaminated synthetic datasets demonstrates the framework’s capability to isolate overlapping primary reflections and suppress complex noise without distorting the desired signal. When applied to conventional real production data, this empirical comparison rate-conditioned gating mask suppresses random and coherent noise while preserving structural horizon integrity and relative amplitude signatures, thereby eliminating the operational reliance on user-dependent manual top muting, structural trace re-sorting, or cumulative multi-domain transformation errors. Crucially, by demonstrating its robustness in handling overlapping signals, this framework establishes a fully automated approach for seismic wavefield deblending. While validated on conventional surveys, this framework provides a potential pathway for future seismic processing challenges, including multi-source high-productivity vibroseis acquisitions, spatial acquisition footprint reduction, and complex three-dimensional azimuthal imaging. By preserving subtle anisotropic variations in amplitude-versus-offset and amplitude-versus-azimuth responses that are relevant to fractured reservoir characterization, this research connects applied mathematics and advanced exploration geophysics, offering a potential tool for next-generation industrial seismic processing.