Searching for Unknown Shapes

[mwilensky768]

(1) Some shapes may be very rare, but no shape should go unnoticed.

(2) Searching for every possible contiguous shape in the band goes as the square of the number of frequencies, which is computationally intractable.

(3) However, an algorithm could be implemented where the shapes consist of successive divisions of the band by powers of 2 (or whatever combinations arise from its prime factorization). This search would be linear in the number of frequencies (upper bound of 2N for only searching powers of 2), and we already search each frequency channel for single-sample outliers, so expanding this by a number of order 1 would not be totally intractable.