Change cutoff for arctan function to fix TR-120 resp phase issue

Value should produce reasonably stable results while still respecting the need for a cutoff point due to variability close to zero