Communicating under Adversarial Attacks: Models, Codes, and Fundamental Limits (10.1.2 Mobile Ad Hoc Networks)
Abstract
This project considered how coding can be used to achieve resilient communication in the face of adversarial attack. Prior work on this topic makes one of two rather limiting assumptions: first, that the adversary operates atthe physical layer and therefore injects errors that are akin in a certain sense to noise, and second, that source channel separation can be assumed, meaning that data compression can be performed separately from coding against adversarial errors. In reality, the adversary may be able to subjugate a node in the network and thereby inject errors at a much higher layer of abstraction. Also, prior work of the PI has shown that the performance penalty associated with enforcing source-channel separation can be arbitrarily large in this application.This project provides code constructions that jointly perform data compression and error correction to provide superior resiliency against adversarial attack, which are novel in several respects. In addition to jointly performing data compression and error correction and targeting errors at a higher layer in the network stack, the codes gracefully degrade with increasing strength of the adversary and rely on regular, instead of modulo, integer arithmetic.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 23, 2017
- Accession Number
- AD1058666
Entities
People
- Aaron B. Wagner
Organizations
- Cornell University