PC-KCTR-HMAC: A High-Security Key Derivation Function Based on Convolutional Neural Networks

Abstract When the Source of Keying Material (SKM) for a Key Derivation Function (KDF) contains entropy but is non-uniformly distributed or non-pseudorandom, a fixed-length pseudorandom key (PRK) must first be generated using a randomness extractor (XTR) before being expanded into a variable-length key. This paper proposes PC-KCTR-HMAC, a KDF based on Convolutional Neural Networks (CNNs). First, the PC module efficiently generates high-entropy pseudorandom sequences that pass all 15 NIST SP 800 − 22 tests. Second, a counter-based HMAC mode is employed to extract and derive variable-length keys. Experimental results demonstrate that the security of the proposed scheme satisfies the requirements of a variable-output-length Pseudorandom Function (volPRF). Compared to existing methods, PC-KCTR-HMAC significantly enhances key randomness and security while maintaining high efficiency.