<p>Active noise control (ANC) systems for headphones suppress environmental noise by acquiring reference and residual noise signals to generate anti-noise through adaptive control. In conventional ANC approaches, the left and right channels are typically processed independently, and binaural reference signals are treated as isolated inputs. Consequently, the inherent temporal correlation, complementary spectral characteristics, and spatial information between the two channels are not fully exploited, which limits noise reduction performance in complex acoustic environments. To address these limitations and better conform to the binaural processing mechanism of the human auditory system, this paper proposes a dual-filter feedforward binaural ANC algorithm for headphones. Unlike conventional bilateral ANC and existing binaural ANC methods that rely on single-filter cross-channel processing, the proposed algorithm employs two independently updated adaptive filters in each ear, enabling the simultaneous utilization of ipsilateral and contralateral reference signals. This structure effectively exploits contralateral time-advance information of the primary noise while avoiding inter-band interference between reference signals with different spectral characteristics, thereby enhancing adaptation stability and robustness under multi-directional and broadband noise conditions. Furthermore, considering the pronounced frequency-dependent sensitivity of human hearing, an A-weighting-based perceptual optimization mechanism is embedded directly into the adaptive filtering framework. By incorporating perceptual weighting into the filter coefficient update process rather than as a post-processing stage, the proposed method reformulates the adaptive cost function from physical error minimization to perceptually weighted error minimization. As a result, the adaptive filters allocate control effort preferentially to perceptually critical frequency bands, yielding residual noise that better matches human auditory perception. To comprehensively evaluate the performance of the proposed algorithm, both objective noise reduction and perceptual sound quality are assessed. A set of psychoacoustic indicators—including loudness, roughness, sharpness, and tonality—is employed, and the subjective quality of the residual noise is further evaluated using a predictive pleasantness model. Experimental results demonstrate that the proposed algorithm achieves higher noise reduction. performance and significantly improved perceptual sound quality compared with conventional bilateral ANC and existing binaural ANC methods.</p>