Gaussian and FlexKnot signal model comparison in |$\Delta \log (\mathcal {Z})$| and the signal RMSE between the recovered signal and the simulated signal. The true signal is a globalemu signal whose minimum point is at 73 MHz with an amplitude of about 150 mK. The parameters used to generate the signal are listed in the first entry G73 of Table 1. The results yielded by the FlexKnot model is shown are solid lines, and the Gaussian cases are shown in dashed lines as reference. The left panel shows that |$\Delta \log (\mathcal {Z})$| is sufficiently high in all cases, indicating that the model with a signal is strongly preferred. The decreasing |$\Delta \log (\mathcal {Z})$| shows that a higher number of knots does not improve the confidence in signal recovery. The y-axis is inverted in the right panel. The right panel shows that in these cases, by the lower signal RMSEs, the FlexKnot signal model yields better signal recoveries than the Gaussian signal model, and 4 to 6 knots is optimal and sufficient to recover the signal. Alike to the Fig. 6, it shows that the presence of systematic decreases the quality of fit progressively with the level of its amplitude in all cases, with the case of 50 and 100 mK having high values, indicating the unreliability of the signal recovery.
