pupperemeritus’s GSoC Blog

Analyzing unevenly sampled data cannot be done with methods of the fourier domain. We use the lomb scargle domain to account for the unevenness of the data. I will be implementing the crossspectrum from scratch and power spectrum using scipy implementation or from scratch. Then I will be creating wrapper classes and helper functions that will provide a sensible API to the user. The Lomb Scargle cross spectrum implementation can be optimized by using JAX. Reason for choosing the scipy implementation is that it allows us to get an unnormalised PSD to which we can apply stingray’s methods of normalisation and astropy’s implementation does not have that option, and their normalisations are different from the ones used in stingray. There would also be a need of helper functions to create objects of these classes automatically from various input types similar to the existing PSDs. Extra parameter of maximum frequency is to be given as it cannot be inferred from the data as equal to 1/time between two data points. Further the frequency resolution cannot be inferred due to the same reason. The crosspectrum class must be compatible with statistic functions already existing in stingray such as the time lag, phase lag and coherence (which exist only for crossspectra) (if this turns out to not be possible then create new similar statistic functions).