The eavesdropper channel, or "wiretap channel", models scenarios in
which an encoder is connected by a broadcast channel to a decoder and
an eavesdropper.  Performance can be characterized by a tradeoff
between the message rate communicated to the decoder as well as the
equivocation, i.e., the conditional uncertainty about the message, at
the eavesdropper.  Non-zero equivocation at the eavesdropper
corresponds to some amount "secrecy" of the transmission, and "perfect
secrecy" corresponds to the case in which the rate and equivocation
are equal.

In this work, we extend results for the eavesdropper channel in two
directions.  First, we develop a modern, and slightly more intuitive,
proof of the Csiszar-Korner result for the largest rate-equivocation
region for discrete memoryless broadcast channels without a common
message.  Second, using the intuition gained from the discrete
memoryless case, we extend the results to rate-equivocation tradeoffs
for general eavesdropper channels in the sense of Verdu and Han,
including a new variation on epsilon-capacity for such models.

Our work differs from other recent work on the eavesdropper channel in
two important ways.  First, we consider the entire tradeoff between
transmission and equivocation rates, instead of requiring the two to
be equal for perfect secrecy.  Second, the results for general channel
models apply to non-stationary or non-ergodic models that can be
important in some applications, e.g., wireless communications with
multipath fading and diversity transmission.