Physics-informed spectral parameterization for tractable tolerance sampling in broadband imaging

Modeling wavelength-dependent aberrations in broadband imaging systems requires high-dimensional parameter spaces that render Monte Carlo (MC) tolerance analysis intractable. We propose a low-dimensional spectral parameterization of Zernike coefficients that compresses broadband aberration modeling from 288 to 36 parameters and enables tractable MC tolerance analysis for manufactured systems. For a dispersion-dominated DD-CASSI system, we further introduce a Cauchy-derived Hybrid basis whose 1/λ² dependence captures defocus-driven dispersion. Compared with nominal-PSF training, MC-augmented training with the proposed parameterization improves reconstruction PSNR by 1.06 dB, of which 0.48 dB is attributable to the Hybrid basis over a polynomial baseline.