Secrecy Results for Compound Wiretap Channels

• With transmitter CSI, the secrecy capacity meets the general upper bound from, completing the coding theorem for this family.Source: §1 · p. 1
• A stronger secrecy criterion and maximum-error criterion are achieved with a universal decoder that ignores encoding randomisation while attaining the optimal secrecy rate across channels.Decoding the randomisation parameter may not work in this model.Source: §1 · p. 1
• Without CSI, the paper gives a non-computable multi-letter capacity expression and an example with positive CSI capacity but zero no-CSI capacity despite convex channel sets.Source: §1 · p. 2
• For partial CSI, the paper gives a multi-letter expression and a computable capacity description when eavesdropper channels are degraded versions of legitimate channels.The results are also extended to arbitrary sets of wiretap channels via standard approximation techniques.Source: §1 · p. 2

2.1. Definitions

• A compound wiretap channel pairs legitimate and eavesdropper channel families; codes use stochastic encoders and mutually disjoint decoding sets, with maximum error probability specified for the code.Source: §2.1 · p. 2
• With no CSI, neither sender nor receiver knows the channel state; with transmitter CSI, encoding may depend on the state while decoding sets remain universal, and the eavesdropper knows the state.Source: §2.1 · p. 2
• The section defines achievable secrecy rates through sequences of codes and denotes the resulting capacities by C_S(W) and C_S,CSI(W).Source: §2.1 · pp. 2–3

2.2. Hints on operational meaning of strong secrecy

• As the message count grows, strong secrecy makes the eavesdropper’s average and maximum decoding error tend to one, although a small fraction of messages may still be decoded correctly.Source: §2.2 · p. 3
• The identification interpretation treats each possible message as a separate hypothesis test, with the eavesdropper choosing an output subset for each message and channel state.Source: §2.2 · p. 4
• The paper leaves open which communication tasks beyond message transmission, and to what extent, are protected by the strong-secrecy condition.Source: §2.2 · p. 5

3.1. Preliminaries

• Preliminaries summarize typical-sequence properties and give a Chernoff-Hoeffding bound used alongside them in the subsequent arguments.The text identifies these tools as needed in the sequel and as a standard large-deviation result.Source: §3.1 · pp. 5–6

3.2. CSI at the transmitter

• The CSI secrecy capacity is min_t max_{U_t→X_t→(Y_t,Z_t)} [I(U_t;Y_t) − I(U_t;Z_t)], with the maximum over distributions satisfying the stated Markov condition.The theorem takes the minimum over channel states t; U_t has range A, and X_t is connected to Y_t and Z_t through W_t and V_t, respectively.Source: §3.2 · p. 7
• The CSI achievability proof uses a decoder that avoids decoding the randomization parameter, then expurgates to obtain maximum-error reliability and establishes strong secrecy uniformly across channel states.The randomization index is used in encoding, while the legitimate receiver decodes the message without attempting to recover that index.Source: §3.2 · pp. 8, 10, 11
• For a known transmitter state pair (t,s), the capacity expression maximizes I(U;Y_t) − I(U;Z_s) over U→X→(Y_t,Z_s).The passage states this case requires only a slight modification of the preceding argument.Source: §3.2 · p. 11

3.3. No CSI

• When neither terminal knows the channel state, the proof uses a stochastic encoder independent of the channel realization and chooses the randomization rate to accommodate the maximum across states.The construction uses the same encoder codebook and randomization index across channel states.Source: §3.3 · p. 12
• The no-CSI case has a multiletter secrecy-capacity expression, with an upper bound established by a converse.The section also establishes achievability and uniform strong secrecy across channels in the compound family.Source: §3.3 · pp. 13–15
• When every eavesdropper channel is degraded from every legitimate channel, the no-CSI secrecy capacity is given by a specialized expression.The stated degradation condition is that each Vs is a degraded version of every Wt.Source: §3.3 · pp. 15–16

3.4. Channel state to the legitimate receiver is known at the transmitter (CSIt)

• With transmitter knowledge of the legitimate-receiver channel state, the section gives a secrecy-capacity result when eavesdropper states remain unknown.It also gives a specialized capacity result when each eavesdropper channel is degraded from every legitimate-receiver channel.Source: §3.4 · pp. 16, 19

3.5. Compound wiretap channel with CS = CS,CSI

• Under the stated degradation ordering among legitimate and eavesdropper channels, compound-channel capacity equals capacity with transmitter CSI.The ordering requires one legitimate channel to be degraded from the others and the remaining eavesdropper channels to be degraded versions of a designated eavesdropper channel.Source: §3.5 · pp. 19–20
• For the considered compound wiretap channel, the lower bound on capacity without CSI matches the capacity with CSI.Source: §3.5 · p. 20

4.1. Example 1

• In the CSI example, reliable message transmission is achievable even though the decoder cannot identify both the message and randomization indices for all message-index pairs.The passage contrasts this with the single-channel case.Source: §4.1 · p. 21

4.2. Example 2

• With convex legitimate-receiver and eavesdropper channel families, the example has positive secrecy capacity with transmitter CSI but zero secrecy capacity without CSI.The stated result is CS,CSI(W) > 0 and CS(W) = 0.Source: §4.2 · p. 22
• For uniform input, I(p, W_t) − I(p, V_t) is positive for every t ∈ [0, 1].The example gives this strict information difference across the channel family.Source: §4.2 · p. 22