Smoothing field trims

An issue with the field predictions from the TFT when calculating the $\Delta B$ to trim as $\Delta B=B_{\text{ref}}-B_{\text{pred}}$ and $B_{\text{ref}}=B_{\text{pred},0}$, is that the field predictions are not perfectly smooth. For the MBIs, the field is predicted at 50 Hz, but pointwise differences are common, and results in a noisy $\Delta B$ that is sent to LSA.

Zooming in on the flat top we see

Where the $\Delta B$ is jagged.

For smoothing the delta, we divide $\Delta B$ by its standard deviation to normalize it, and then apply different smoothing algorithms.

Univariate splines

We can fit smoothing splines with different s values. For larger s-values we see large overshoots at sharp areas (especially at the flat top here), which is undesirable. This is seen even for small s-valeus like 0.1, even though it follows well elsewhere.