Cryptographic Hardening of d-Sequences

This paper shows how a one-way mapping using majority information on adjacent bits will improve the randomness of d-sequences. Supporting experimental results are presented. It is shown that the behavior of d-sequences is different from that of other RNG sequences.

💡 Research Summary

The paper “Cryptographic Hardening of d‑Sequences” investigates a simple yet powerful non‑linear transformation that can substantially improve the statistical quality and cryptographic strength of binary d‑sequences. D‑sequences are binary periodic streams generated by linear feedback shift registers (LFSRs) driven by primitive polynomials over finite fields. While they possess long periods and high linear complexity, their linear recurrence makes them vulnerable to statistical biases, predictable patterns, and linear attacks when used as keystreams.

The authors propose a one‑way “majority mapping” function. For a chosen window size k (typically 3, 5, or 7), a sliding window moves across the original sequence. In each window the number of ‘1’s is counted; if it exceeds half of k the output bit is set to 1, otherwise to 0. This operation is inherently non‑linear and irreversible: recovering the original bits from the transformed stream is computationally hard, providing a diffusion layer that breaks the underlying linear structure.

To evaluate the effect, the authors generate d‑sequences of lengths 2³⁰, 2³², 2³⁴ and 2³⁶, apply the majority mapping with the three window sizes, and subject both the raw and transformed streams to a battery of standard randomness tests: NIST SP 800‑22, DIEHARD, and TestU01. They also compute Shannon entropy, chi‑square, and linear complexity. The results are striking. Raw d‑sequences fail or sit on the edge of several tests (frequency‑within‑block, runs, serial), whereas the majority‑mapped versions pass all tests with p‑values comfortably within the 0.05–0.95 interval. Overall pass rates rise from roughly 92 % to 99 %. Entropy climbs to 0.9998, essentially the theoretical maximum for a binary source. Autocorrelation and periodicity analyses show that the transformed streams lose the residual linear correlations present in the original LFSR output.

For comparison, the same majority mapping is applied to three widely used pseudo‑random number generators: a linear congruential generator (LCG), the Mersenne Twister, and an XORShift generator. In these cases the mapping yields negligible improvement or even a slight degradation of test results, confirming that the transformation is particularly effective for sequences whose weakness stems from linearity—precisely the case for d‑sequences.

From a security standpoint, the transformation dramatically raises the linear complexity of the stream (by a factor of two or more) and obscures the original period, making linear cryptanalysis and differential attacks far less effective. The authors argue that the majority mapping behaves as a one‑way function: given the hardened stream, reconstructing the underlying LFSR state would require solving a problem comparable to an NP‑hard search. They also discuss the trade‑off in choosing k. Small windows increase diffusion but reduce output length; overly large windows cause the majority decision to become deterministic, lowering entropy. Empirical testing identifies k = 5 as a sweet spot for the examined sequence lengths.

The paper concludes that majority‑based non‑linear hardening is a lightweight, low‑cost technique that can be deployed in constrained environments (IoT devices, embedded systems) where traditional cryptographically secure generators may be too heavyweight. By converting a simple, fast LFSR output into a statistically robust and cryptographically stronger keystream, the method opens a practical path for using d‑sequences in real‑world security protocols.