arXiv:2609.36162
From Classical to Quantum Channels: Achieving Positive Covert Rates
Hassan ZivariFard, Xiaodong Wang, Alexei Ashikhmin
Abstract
In this paper, we study the conditions under which covert communication at positive rates is feasible over discrete memoryless classical, classical-quantum, and quantum channels. For classical point-to-point channels, we show that if the dimension of the channel input probability simplex exceeds that of the channel output probability simplex, equivalently, if the input alphabet has larger cardinality than the output alphabet, then positive covert rates are achievable for certain classes of Discrete Memoryless Channels (DMCs). We further show that allowing the innocent symbol (i.e., the symbol transmitted in the no-communication mode) to be chosen appropriately can enlarge the class of DMCs for which positive covert rates are achievable. We also study covert communication over classical Multiple-Access Channels (MACs), where the additional transmitter effectively enlarges the set of available channel input pairs, and we show that the conditions required to satisfy the covertness constraint are less restrictive for MACs than for point-to-point channels. We extend these results to classical-quantum DMCs by showing that if the dimension of the input probability simplex exceeds the affine dimension of the set of output states that can be induced at the channel output, then positive covert rates are achievable for certain classes of classical-quantum DMCs. Finally, for quantum channels, we show that if the innocent state (i.e., the state transmitted in the no-communication mode) is mixed, then the covertness constraint can always be satisfied by a non-trivial input ensemble. Consequently, positive covert rates are achievable whenever the legitimate receiver can distinguish at least two states in a suitable such ensemble.
AI Summary
The paper studies when memoryless classical, classical–quantum, and quantum channels support positive-rate covert classical communication rather than square-root scaling. Covertness requires the warden’s block output to approach its no-communication product distribution or state in total variation or trace distance; shared secret keys and stochastic encoding are allowed. For classical channels, it states the capacity region $R\le I(U;Y)$, $R_K\ge[I(U;Z)-I(U;Y)]^+$, optimized over $U\to X\to(Y,Z)$ with $p_Z=q_0$. The geometric analysis asks whether the innocent output lies in the convex hull of non-innocent outputs. For binary-output channels, feasibility is equivalent to $\min_{i\ne j}\epsilon_i\le\epsilon_j\le\max_{i\ne j}\epsilon_i$; with at least three inputs, choosing a median-output symbol as innocent guarantees feasibility. Binary-input multiple-access examples distinguish independent from correlated inputs, with correlation permitting additional covert ensembles. Classical–quantum extensions replace probability vectors by density operators. For fully quantum inputs, every mixed innocent state admits a nontrivial ensemble reproducing its warden output, by convex decomposition and channel linearity. Positive rates additionally require distinguishable receiver outputs. For pure innocent states, feasibility instead requires another input state with the same warden output. Achievability uses random product-state coding, packing, and quantum channel resolvability; a classical converse uses information identities and continuity. Quantum results are inner bounds, not capacity characterizations. Larger input dimension alone does not guarantee feasibility beyond binary outputs, and satisfying covertness alone does not guarantee positive communication rates.
Error found: The independent-input MAC theorem falsely guarantees a non-trivial covert distribution after choosing the innocent pair freely: two BSC components with output probabilities 0.82 on diagonal input pairs and 0.18 off diagonal admit only deterministic independent inputs matching any innocent pair. The 'One BSC Component' example miscalculates d_11 and proposes inputs producing warden output probability 1/2 rather than the generally different innocent probability, invalidating its claimed universal feasibility guarantee.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 44 pages, 9 figures, single column, part of the results will be presented at the 2026 IEEE Information Theory Workshop