Spectral Deconvolution — Linear + Nonlinear

Drop CSV files (two columns: x, y). Raw spectrum first, then one or more basis spectra. Linear mode solves y ≈ Σ cᵢ · bᵢ(x) by least squares. Nonlinear mode fits parametric peaks (Gaussian / Lorentzian / pseudo‑Voigt) and optional baseline.

1) Raw Spectrum

Click or Drop a single raw spectrum CSV here

Expected CSV: two numeric columns (x, y). Header lines are ignored.

2) Basis Spectra (1…∞)

Click or Drop one or more basis spectra CSV files

Each basis file should be two numeric columns (x, y).

Toggle which bases to include in the fit.

3) Options

Ridge λ: Constrain cᵢ ≥ 0 Use a tiny λ (e.g., 1e-8 to 1e-4) if the system is ill‑conditioned; enable positivity to enforce NNLS.

Results & Plots (Linear)

Basis chosen

R²

—

RMSE

—

Overlay: Raw (dots) vs Reconstructed (line)

Residuals: Raw − Reconstructed

Coefficients

Linear combination: y_fit(x) = Σ cᵢ · bᵢ(x)

#	Basis name	Coefficient cᵢ

4) Nonlinear Deconvolution (Peaks)

Peak shape: Add baseline (c₀ + c₁·(x−x̄)) Constrain A ≥ 0

Iter max: λ₀ (damping): Tol Δp:

Quick add: center at mid‑x, FWHM ≈ 2% of range, A ≈ 20% of (max−min)

Auto‑guess top N peaks:

#	μ (center)	FWHM w	A (amp)	η (Voigt)	On	✕

Results & Plots (Nonlinear)

Peaks used

R²

—

RMSE

—

Overlay: Raw (dots) vs NL fit (line) + individual peaks (thin)

Residuals (Nonlinear)

Peak Parameters

#	μ	FWHM w	A	η