BLISS Seminar: Semantic security versus active adversaries

Seminar | February 8 | 2-3 p.m. | 400 Cory Hall

 Ziv Goldfeld, Ben-Gurion University

 Electrical Engineering and Computer Sciences (EECS)

Physical Layer Security (PLS) guarantees protection against computationally-unlimited eavesdroppers without using a key. These guarantees come at the price of an unrealistic assumption that the eavesdropper's channel is fully known to the legitimate parties. Furthermore, typical PLS metrics are incompatible with the features of the data they are designed to protect. For these reasons, PLS has found limited use in practice despite its various benefits. By means of a novel and stronger version of Wyner's soft-covering lemma, we upgrade IT security proofs to the stronger and more practically viable semantic-security metric, while removing the ‘known eavesdropper channel’ assumption. As applications we derive the semantic-security capacity of the type constrained arbitrarily varying wiretap channel (WTC), and as its special case, solve the problem of the WTC of type II with a noisy main channel - a problem by Ozarow and Wyner that was open since 1984. The scenario where the state sequence is random (rather than arbitrary) is also considered. We construct a simple semantically-secure superposition code that strictly outperforms the best previously known achievable rates. The construction implicitly includes a key agreement phase (by means of the random and i.i.d. state sequence) that is crucial for the aforementioned improvement.