2023MNRAS.525.3946C

We present a new method, called 'forced-spectrum fitting', for physically based spectral modelling of radio sources during deconvolution. This improves upon current common deconvolution fitting methods, which often produce inaccurate spectra. Our method uses any pre-existing spectral index map to assign spectral indices to each model component cleaned during the multifrequency deconvolution of WSCLEAN, where the pre-determined spectrum is fitted. The component magnitude is evaluated by performing a modified weighted linear least-squares fit. We test this method on a simulated LOFAR high-band antenna (HBA) observation of the 3C 196 QSO and a real LOFAR HBA observation of the 4C+55.16 FRI galaxy. We compare the results from the forced-spectrum fitting with traditional joined-channel deconvolution using polynomial fitting. Because no prior spectral information was available for 4C+55.16, we demonstrate a method for extracting spectral indices in the observed frequency band using 'clustering'. The models generated by the forced-spectrum fitting are used to improve the calibration of the data sets. The final residuals are comparable to existing multifrequency deconvolution methods, but the output model agrees with the provided spectral index map, embedding correct spectral information. While forced-spectrum fitting does not solve the determination of the spectral information itself, it enables the construction of accurate multifrequency models that can be used for wide-band calibration and subtraction.