Wiretap Channels: Implications of the More Capable Condition and Cyclic Shift Symmetry

Characterization of the rate-equivocation region of a general wiretap channel involves two auxiliary random variables: U, for rate splitting and V, for channel prefixing. Evaluation of regions involving auxiliary random variables is generally difficult. In this paper, we explore specific classes of wiretap channels for which the expression and evaluation of the rate-equivocation region are simpler. In particular, we show that when the main channel is more capable than the eavesdropping channel, V=X is optimal and the boundary of the rate-equivocation region can be achieved by varying U alone. Conversely, we show under a mild condition that if the main receiver is not more capable, then V=X is strictly suboptimal. Next, we focus on the class of cyclic shift symmetric wiretap channels. We explicitly determine the optimal selections of rate splitting U and channel prefixing V that achieve the boundary of the rate-equivocation region. We show that optimal U and V are determined via cyclic shifts of the solution of an auxiliary optimization problem that involves only one auxiliary random variable. In addition, we provide a sufficient condition for cyclic shift symmetric wiretap channels to have U=\phi as an optimal selection. Finally, we apply our results to the binary-input cyclic shift symmetric wiretap channels. We solve the corresponding constrained optimization problem by inspecting each point of the I(X;Y)-I(X;Z) function. We thoroughly characterize the rate-equivocation regions of the BSC-BEC and BEC-BSC wiretap channels. In particular, we find that U=\phi is optimal and the boundary of the rate-equivocation region is achieved by varying V alone for the BSC-BEC wiretap channel.

💡 Research Summary

The paper investigates the rate‑equivocation region of discrete memoryless wiretap channels, focusing on simplifying the notoriously difficult optimization over the two auxiliary random variables introduced by Csiszár and Körner: the rate‑splitting variable U and the channel‑prefixing variable V. The authors identify two important classes of wiretap channels where the optimal choices of U and V can be characterized in closed form, thereby reducing the computational burden dramatically.

First, the authors consider the “more capable” condition, a partial order between the main channel p(y|x) and the eavesdropper channel p(z|x). A channel is more capable if for every input distribution P_X, the mutual information difference f(P_X)=I(X;Y)−I(X;Z) is non‑negative. Under this condition, Theorem 2 proves that channel prefixing is unnecessary: the optimal choice is V = X for the entire boundary of the rate‑equivocation region. Consequently, the region can be traced by varying only U, and the cardinality bound on U collapses from |X|+3 to |X|, a three‑fold reduction. Moreover, the converse is shown: if the channel is not more capable and f(P_X) attains its maximum at an interior point of the probability simplex, then V = X is strictly sub‑optimal, implying that a non‑trivial prefixing strategy can enlarge the achievable region. This result clarifies precisely when channel prefixing must be employed.

Second, the paper studies “cyclic shift symmetric” wiretap channels, where both the main and eavesdropper channels are invariant under cyclic permutations of the input distribution. Such channels include binary symmetric channels (BSC), binary erasure channels (BEC), and type‑writer channels. The key property is that the uniform input maximizes I(X;Y) (and likewise I(X;Z)). Theorem 3 reveals that the optimal auxiliary variables have a highly structured form: U takes |X| equiprobable values, while V can be represented as a pair (U, ĤV) with |ĤV|≤|X|. The distribution of ĤV and the conditional distribution p(X|ĤV) are obtained by solving a single‑auxiliary‑variable optimization problem (19), which involves only I(X;Y)−I(X;Z) and conditional mutual informations given ĤV. The remaining |X|−1 copies of V are generated by cyclically shifting p(X|ĤV). This construction reduces the dimensionality of the original problem from |U|·|V|≈|X|³ to at most |X|², and in many cases (when the uniform input also maximizes the difference I(X;Y)−I(X;Z), termed “dominantly cyclic shift symmetric”) the rate‑splitting variable can be omitted entirely (U=∅). Hence the entire boundary can be achieved by varying V alone.

The theoretical developments are applied to binary‑input cyclic shift symmetric wiretap channels. Two concrete examples are examined: (i) a BSC main channel followed by a BEC eavesdropper (BSC‑BEC) and (ii) a BEC main channel followed by a BSC eavesdropper (BEC‑BSC). By explicitly evaluating the function f(P_X) for binary inputs, the authors determine the regions of the crossover probability p and erasure probability ε where the more capable condition holds, where the dominant cyclic symmetry holds, and where rate‑splitting becomes necessary. For the BSC‑BEC case, the uniform input is optimal, U=∅, and the secrecy capacity C_s = max_{P_X}