written 1986 – on the heyday of the pre-collapse NSA (of the early 1990s) – you can hear the self-confident tone. One gets an interesting data point. Note how the writer contrasts decoding of the code generators underlying spread-spectrum with the then state of the art in general cryptanalysis.
Note specifically how CA itself has moved beyond (as one perhaps might expect) “exhaustive hypothesis testing” in the Colossus/Tunny/Turingismus sense of the term.
By this we can presume folks are referring to non-linear detectors based on properties of certain chaotic systems – known to be harnessing useful non-linear approximation process.