Doubt they're using the dot size to measure distance. This looks quite similar to the primesense technology in the original Kinect. Basically, the theprocessing engine has the same dot pattern stored as an array of coordinates. The algorithm cross-correlates groups of dots from the captured image of the projected dots with the pattern stored. Since the pattern is pseudo-random, each group of dots will have a unique fingerprint. Depth/distance is then calculated from the deviation in position of each dot in the sub-group related to its position in the stored pattern.