Measuring the spectral responsivity of photometric detectors via a monochromator-based system is inevitably affected by the bandwidth effect， which induces prominent systematic errors in spectral regions with steep responsivity variations， such as the blue rising edge and red falling edge. The conventional ratio method calculates responsivity by dividing the photocurrent of the device under test by that of the standard detector and multiplying the result by the standard detector’s responsivity. This method neglects the convolution effect of the instrument bandpass function， thereby introducing large calculation deviations in sharply varying spectral regions. The classical Stearns-Stearns method relies on a triangular bandpass profile and strict matching between scanning step and bandwidth， leading to harsh application constraints. To address the above limitations， this paper proposes a matrix-based bandwidth effect correction method， and systematically evaluates its robustness under practical non-ideal measurement conditi

ons.The proposed method characterizes the measurement process as a convolution between the true spectral responsivity and the instrument bandpass function. A linear equation set is established based on multi-wavelength scanning data， and the detector’s spectral responsivity is directly solved by adopting the LSQR least-squares algorithm. In the simulation， CIE standard illuminant A is selected as the incident spectrum， the intrinsic responsivity of the device under test adopts the CIE standard photopic spectral luminous efficiency function

V

（

λ

）， and the standard detector employs a trap detector with known and flat responsivity across the visible band. The monochromator bandwidth is set to 10 nm with a triangular bandpass function. Under ideal disturbance-free conditions， the responsivity curve retrieved by the matrix method is in great agreement with the true

V

（

λ

）， and its correction performance is obviously superior to that of the ratio method.Three typical interference factors in practical measurement are investigated through numerical simulation. In terms of bandwidth setting deviation， when the actual bandwidth deviates from the nominal value by 1 nm， the matrix method presents a much smaller average absolute deviation over the full spectral range and maintains stable correction capability compared with the ratio method. When the deviation increases to 2 nm， the matrix method still achieves satisfactory correction， though its superiority is weakened. It is suggested to limit the bandwidth setting deviation within ±1 nm. For wavelength calibration offset， a central wavelength shift of 0.1 nm for the bandpass function brings milder curve distortion and smaller absolute deviation using the matrix method than the ratio method. When the offset reaches 0.2 nm， the two methods perform similarly， and the matrix method no longer shows obvious advantages. Thus， the wavelength calibration deviation is recommended to be controlled within ±0.1 nm. Regardi

ng random measurement noise， the absolute noise amplitude is set to 0.000001 times the maximum signal and the relative noise level to 0.00001 in the simulation. Correspondingly， the LSQR convergence tolerance is set to 0.00002， matching the overall noise magnitude. Under such typical noisy conditions， the matrix method can still acquire stable and reliable inversion results， with better correction performance than the conventional ratio method， except for a slightly larger deviation in the low signal-to-noise blue spectral region.In practice， the actual bandpass function of a monochromator commonly presents Gaussian， rectangular and trapezoidal profiles. With the full width at half maximum kept constant， this paper compares the correction results of the three typical bandpass shapes. The results indicate that the Gaussian bandpass has the strongest spectral smoothing effect， the rectangular bandpass easily causes spectral aliasing， and the trapezoidal bandpass produces the slightest spectral distortion. The traditional ratio method shows distinct deviation differences under various bandpass shapes， with the maximum deviation appearing in the Gaussian case. In contrast， by constructing a complete bandpass convolution matrix and precisely modeling the energy distribution of different bandpass profiles， the matrix method maintains a low inversion deviation level for all shapes. Its deviation amplitude is far lower than that of the ratio method with gentle fluctuation， which verifies that the proposed method possesses excellent universality to bandpass shapes and is not limited to triangular profiles.In conclusion， the proposed matrix correction method outperforms the ratio method significantly under ideal 10 nm bandwidth simulation conditions. For engineering practice， it is feasible to restrict the bandwidth deviation within ±1 nm and wavelength calibration offset within ±0.1 nm， where the matrix method retains prominent correction superiority. The LSQR convergence tolerance should be matched to the actual system n

oise level： an overly small tolerance causes solution oscillation， while an overlarge tolerance results in the loss of spectral detail. In practical application， the system noise level should be evaluated first， and the optimal tolerance can be determined by simulation. Moreover， the matrix method exhibits good adaptability to diverse bandpass shapes. The application scope and parameter selection suggestions provided in this paper can offer practical guidance for high-precision spectral responsivity calibration of photometric detectors.