Spectrometer

Reconstructing low-resolution spectra from Osram sensors

Introduction

The Osram AS7341 IC is an 11 channel spectral sensor, with sensitivity from 350nm to 1000nm. Sensitivity is provided by various filters that correspond to 11 spectral bins that are read out by an ADC. These bins however are not evenly spaced in wavelength, bands overlap, and their spectral response is non-uniform over their wavelength band. Therefore with knowledge of the frequency response of each filter, we can deconvolve the signal according to the frequency response of each channel and reconstruct the original spectrum.

From the spectral response listed in the AS7341 datasheet, the FWHM resolution across the wavelength range can be seen in the plot below:

This is currently set up for the AS7341, but in future will also support TCS3448 (on its own and in combination with AS7341). Very preliminary prototype hardware designs for a TCS3448 interface board are available in the hardware/TCS3448_board directory.

Installation

This project uses Poetry for dependency management. To install the package along with its dependencies, run:

poetry install

Usage

Most of the heavy lifting is done by the SpectrumSolver, which takes a set of reference points and a wavelength grid as input. This object can then be used to 'solve' for the original spectrum when provided ADC counts, using Tikhonov regularised least squares. The reference points should correspond to a NumPy array of shape (n_points, n_bins) (seeextract_reference.py), describing the frequency response of the wavelength bin, and therefore y_meas is expected to be a NumPy array of shape (n_bins,).

from spectrum.solver import SpectrumSolver

solver = SpectrumSolver.from_reference_points(reference, wavelength_grid)

y_out = solver.solve(y_meas, lam_smooth=1e-5, mu=1e-8)

The scripts/ directory contains helper programs:

• extract_reference.py – extract response curves from a PDF and save them to an .npz file.

python scripts/extract_reference.py sensor.pdf curves.npz

• fwhm_scan.py – scan a narrow Gaussian across wavelengths and plot the recovered FWHM resolution.

python scripts/fwhm_scan.py curves.npz out.png