A source of uncertainty in our proposed PSD analysis is the addition of Poisson noise. Poisson noise is extra variability added to light curve due to measurement errors. Since measurement error does not depend on temporal frequency it is manifested as a flattening of the PSD at high frequencies. The frequency at which Poisson noise dominates is then dependent on the variability strength of the AGN as well as the average count rate (i.e. S/N).
To determine the effect of Poisson noise, we simulated 4 light curves, each generated from a different PSD shape. The PSD shapes only differ in their break frequencies from \(10^{-6}\) to \(10^{-3}\) Hz. The spectral indices are fixed at -1 and -2 at low and high frequencies respectively. The normalization at \(10^{-6}\) Hz is fixed at \(10^{4.6}\) Hz\(^{-1}\), the average normalization of the PSD fits from \citet{Shimizu_2013}. The mean count rates for the simulate light curves are all 1 count s\(^{-1}\), a conservative estimate for our NuSTAR light curves. After generating the simulated light curve, Poisson noise is added. The light curves are then re-binned into 500 s bins.
Figure \ref{fig:sims} shows the resulting simulated PSDs (colored points) on top of “true” PSD. The only light curves where Poisson noise is evident are the ones with a break frequency at \(10^{-6}\) and \(10^{-5}\). Further Poisson noise does not dominate until frequencies much higher than the break frequency.