Universal Secure Multiplex Network Coding with Dependent and Non-Uniform Messages

We consider the random linear precoder at the source node as a secure network coding. We prove that it is strongly secure in the sense of Harada and Yamamoto and universal secure in the sense of Silva and Kschischang, while allowing arbitrary small but nonzero mutual information to the eavesdropper. Our security proof allows statistically dependent and non-uniform multiple secret messages, while all previous constructions of weakly or strongly secure network coding assumed independent and uniform messages, which are difficult to be ensured in practice.

💡 Research Summary

This paper addresses the problem of securing network coding in a single‑source multicast setting when multiple secret messages are statistically dependent and non‑uniform. Traditional secure network coding schemes assume that the secret messages are independent and uniformly distributed, an assumption that is difficult to guarantee in practice. The authors propose a universal secure multiplex network coding scheme that simultaneously satisfies the strong security notion of Harada and Yamamoto and the universal security notion of Silva and Kschischang, while allowing an arbitrarily small but non‑zero information leakage to an eavesdropper (Eve).

The network model consists of a source node with at least n outgoing links, each transmitting a packet of m symbols over a finite field 𝔽_q. Linear network coding is performed at intermediate nodes, and all legitimate receivers can recover the n·m transmitted symbols. Eve can eavesdrop on up to μ ≤ n links, obtaining μ·m symbols. The authors adopt a family of two‑universal hash functions to construct a random linear precoder at the source. This precoder mixes the secret messages S₁,…,S_T together with an additional random vector S_{T+1} (treated as dummy randomness) via a random linear transformation L drawn uniformly from the two‑universal family.

The core theoretical contribution is a strengthened privacy‑amplification theorem (Proposition 4). For random variables A₁ (the whole secret vector), A₂ (a part of it), and a two‑universal hash family F, the theorem bounds the moment‑generating function of the conditional mutual information: \