ARCHIVES

One previous study analyzed a chaotic logistic map optimized with a Genetic Algorithm (GA) (x0* = 0.487214, r* = 3.998721), then applied it to generate AES-256 keys on an ESP32 microcontroller. In hardware samples, that study measured Shannon entropy of 7.6307 bits per byte and an autocorrelation of -0.004526, against a 7.98 bits per byte baseline from 64-bit floating-point simulation. However, comparing the hardware analysis against the earlier simulation work it is based on turns up several things worth cleaning up before anyone builds further on either paper. There is a discrepancy in the reported GA search range, and a drop from partial NIST SP 800-22 testing in simulation to no NIST testing at all on hardware, along with the same autocorrelation value appearing with opposite signs inside the same paper. The precision experiment meant to explain the simulation-to-hardware entropy drop was not only too small. It pointed in the wrong direction entirely. Worse, the script used to generate independent key samples reused the same fixed seed every time, so those so-called “key-generation cycles” were not actually independent. This paper takes on two of these problems directly. First, a new continuous-trajectory export method resolves the repeated-seed-reset issue. Second, the precision gap is recomputed over a window matched in size to the published hardware dataset (2,560,000 bits). That makes the gap 0.4930 bits per byte, matching the observed hardware drop (0.37) and, this time, in the correct direction. The remaining problems (the search-range mismatch, the missing NIST testing, and the autocorrelation sign discrepancy) are documented here as open items. These two new results are kept narrow and self-contained, a targeted correction rather than a rewrite of the original work. Full NIST SP 800-22 testing and new hardware runs of the corrected export method are left for a future paper.