•  
  •  
 

Document Type

Article

Keywords

Cryptography, Chaotic, DQN, NIST, Randomness

Abstract

In contemporary digital environments, exponential cyber threat growth has made cryptographic key generation a critical security challenge. Traditional Pseudo-Random Number Generators (PRNGs) and existing chaos-based methods often exhibit insufficient entropy, limited randomness quality, and inadequate resistance to statistical attacks. Current implementations frequently produce suboptimal entropy values and fail to meet modern cryptographic security standards and rigorous randomness testing protocols. This paper aims to design and implement an advanced cryptographic key generation system that combines Deep Q-Networks (DQN) algorithms with multiple chaotic maps to produce cryptographically secure stream key bits with high randomness and strong resistance to cryptanalytic attacks. The proposed DRLKG-Chaotic (Deep Reinforcement Learning Key Generation with Chaotic maps) system implements six distinct experimental scenarios utilizing five chaotic maps: Tent, Ikeda, Chua's, Rössler, and Double Pendulum. The first five scenarios individually integrate each chaotic map with a DQN algorithm, whereas the sixth scenario implements a novel fusion approach that incorporates all five maps simultaneously. Each scenario generates key streams of three different lengths (128-bit, 192-bit, and 256-bit) to accommodate varying security requirements. A comprehensive evaluation using the National Institute of Standards and Technology (NIST) statistical test suite, brute-force attack resistance analysis, Auto Correlation (AC), Cross Correlation (CC), and Discrete Fourier Transform (DFT) analysis demonstrates the significant improvements over standard chaotic implementations. The results indicate that the DQN scenarios achieve entropy values ranging from 0.9097--0.9999, whereas the standard chaotic maps achieve values ranging from only 0.3627--0.5463. All NIST test P values consistently exceed 0.90 across all the parameters, indicating superior randomness characteristics. In addition, reliable results are obtained when various types of attacks, such as brute-force attacks, side‑channel attacks, and timing attacks, are applied.

Download

Share

COinS