Title: Rotation measure synthesis revisited

Abstract: We re-formulate rotation measure (RM) synthesis for data sets with discrete frequency channels and an arbitrary channel response function. The most commonly used version of the formalism by Brentjens & De Bruyn assumes a top-hat response function in wavelength squared, while real data sets can often be approximated better with a top-hat in frequency. We simulate mock data sets for various source geometries, using a top-hat response function in frequency, and we compare the quality of the RM spectra that are found with both formalisms. We include the response function of the simulated data to calculate exact RM spectra using our formalism. We show that the formalism by Brentjens & De Bruyn produces accurate results even if depolarization at the lowest frequencies in the observing band is severe. If RMs are large, our formalism reconstructs the emitted signal more accurately, with a higher amplitude and (in most cases) a narrower RM spread function. Our formalism can also detect sources with larger (absolute) RMs for a given sensitivity level of the observations.